Online gas and particle phase measurements of organosulfates, organosulfonates and nitrooxyorganosulfates in Beijing utilizing a FIGAERO ToF-CIMS

A Time of Flight Chemical Ionisation Mass spectrometer (CIMS) utilizing the Filter Inlet for Gas and AEROsol (FIGAERO) was deployed at a regional site 40 km north west of Beijing and successfully identified and measured 17 sulfur containing organics (SCOs = organo/nitrooxyorgano sulfates and sulfonates) with biogenic and anthropogenic precursors. The SCOs were quantified using laboratory synthesized standards of lactic acid sulfate and nitrophenol organosulfate (NP OS). The variation in field observations was confirmed by comparison to offline 25 measurement techniques (orbitrap and High performance Liquid Chromaotography (HPLC)) using daily averages. The mean total (of the 17 identified by CIMS) SCO particle mass concentration was 210 + 110 ng m and had a maximum of 540 ng m, although contributed to only 2 + 1% of the organic aerosol (OA). The CIMS identified a persistent gas phase presence of SCOs in the ambient air, which was further supported by separate vapour pressure measurements of NP OS by a Knudsen Effusions Mass Spectrometer (KEMS). An increase in relative humidity (RH) promoted 30 partitioning of SCO to the particle phase whereas higher temperatures favored higher gas phase concentrations. Biogenic emissions contributed to only 19% of total SCOs measured in this study. Here C10H16NSO7, a monoterpene derived SCO, represented the highest fraction (10%) followed by an isoprene-derived SCO. The aAnthropogenic SCOs with polycyclic aromatic hydrocarbon (PAH) and aromatic precursors dominated the SCO mass loading (51%) with


Introduction
Atmospheric particulate matter (PM) is known to play a major role in affecting climate and air quality leading to severe 15 health issues, such as respiratory and cardiovascular degradation (Pope et al., 2002;Kim et al., 2015). Secondary organic aerosols (SOA), formed through reactions of volatile organic compounds (VOCs) yielding semi volatile products that partition into the aerosol phase, represents a significant fraction of PM (Hallquist et al., 2009) and remain the most poorly understood PM source (Foley et al., 2010) due to the complexity of their chemical nature, resulting in discrepancies between observations and models (Heald et al., 2005). Annual average PM1 (particulate matter of 20 diameter less than 1 micron) concentrations in Beijing reached 89.5 µg m -3 in 2013 and, although recently dropped to 80.6 µg m 3 , is still significantly above the Chinese National Ambient air quality Standard (CNAAQS, 35 µg m 3 annual average). The knowledge gap of PM primary emissions and secondary production limits scientifically based abatement strategies targeting effects of secondary pollution in highly polluted regions (Hallquist et al., 2016;. Therefore, Beijing is an ideal case study region for intense measurement campaigns to increase our understanding of 25 the sources and processes involved in atmospheric aerosol chemistry in megacities. A growing number of field studies in this region have been performed in recent years, specifically focused on the haze events investigating the composition of primary and secondary particle aerosols and their formation mechanisms (Guo et al., 2012(Guo et al., , 2014(Guo et al., , 2013Huang et al., 2010;Hu et al., 2016Hu et al., , 2017Li et al., 2017).
OSs have previously been measured in ambient aerosols at a number of geographical locations, from remote regions to highly populated urban environments (SurratttSurratt et al., 2007Kristensen et al., 2011;Stone et al., 2012;Zhang et al., 2012b;Worton et al., 2013;Shalamzari et al., 2013;Hansen et al., 2014) although their composition and contribution to organic mass can vary significantly (Huang et al., 2015). To date, several of their precursors are not known (e.g., Hansen et al., 2014). Mechanistic studies reveal multiple possible pathways for SCO formation, which depend on availability of reactants in the atmosphere (Hettiyadura et al., 2015), increasing the complexity of 5 understanding their occurrence and descriptions within models. Measurements of specific SCOs have shown they may individually contribute up to 1% of the total organics (Olson et al., 2011;Liao et al., 2015).
Isoprene SCOs are hypothesized to be the most abundant precursor in the ambient atmosphere (SurratttSurratt et al., 2007;Liao et al., 2015) and are often used as markers of isoprene-derived SOA in field campaigns (Zhang et al., 2012b).
Aromatic SCOs, considered to originate from anthropogenic sources, have been recently observed in Lahore, Pakistan 10 (Stone et al., 2012) and in urban sites in East Asia (Lin et al., 2012). Riva et al. (2015 and2016) have also previously probed the SCO formation potential from PAH and alkane oxidation in the presence of acidic sulfate aerosols. Glycolic acid sulfate (GAS) is considered another potentially important SCO due to its common abundance and possible sources (Olson et al., 2011;Liao et al., 2015). It is thought to form via a gas phase precursor reaction with acidic aerosol sulfate or from the particle phase reaction of methyl vinyl ketone with a sulfate particle, although both of these mechanisms 15 are yet to be proven (Liao et al., 2015). GAS is also the only SCO to date, which has been detected in the gas phase (Ehn et al., 2010), providing possible importance of gas to particle phase partitioning of some SCOs.
SCOs are thought to be good tracers for heterogeneous aerosol phase chemistry and SOA formation since the known formation mechanisms involve reactive uptake of gas phase organic species onto aerosol (SurratttSurratt et al., 2010).
Due to their hydrophilic nature, polarity and relatively low volatility, they may significantly help nanoparticle growth 20 and increase their potential to become cloud condensation nuclei (Smith et al., 2008). Therefore, it is imperative to improve our knowledge of SCO abundance, formation, distribution, precursors and fate to help develop our understanding of SOA formation.
Mass spectrometry coupled with electrospray ionization is a common method to detect SCOs (Iinuma et al., 2007;Reemtsma et al., 2006;SurratttSurratt et al., 2007;Gomez-Gonzalez et al., 2008). Liquid chromatography can 25 efficiently separate aromatic and monoterpene derived organic sulfates containing aromatic rings or long alkyl chains and is used for speciation of SCOs (Stone et al., 2012). Furthermore, hydrophilic interaction liquid chromatography has been utilized as a very selective technique due to its ability to allow the SCO to retain a carboxyl group, enabling detection of a larger suite of compounds (Gao et al., 2006). The methods above often rely on sampling filters in the field and therefore provide a relatively low measurement frequency. This can limit the ability to evaluate production 30 pathways when concentrations are often integrated over a period of hours or more. Further reactions on filters between the organics and sulfates have also been postulated to add a bias to the SCO concentration measured with respect to initial deposition onto the filter (Hettiyadura et al., 2017;Kristensen et al., 2016). Recently, a Particle Analysis Laser 4 Mass Spectrometer (PALMS) was utilized to measure a number of OSs over the United States highlighting the ability of time of flight mass spectrometers to measure several SCOs at high time frequencies (Liao et al., 2015).

This study utilizes a Filter Inlet for Gas and AEROsol (FIGAERO) Time of Flight Chemical Ionisation Mass
Spectrometer (ToF-CIMS) for the measurement of ambient SCOs at a semi-rural site 40 km from Beijing, China. This instrument enables measurements of either the gas-phase components or thermally desorbed particles by a high 5 resolution mass spectrometer via a multi-port inlet, as described in detail by Lopez-Hilfiker et al. (2014). The soft and selective ionization technique and high time resolution coupled with the FIGAERO enables the simultaneous detection and measurement of SCOs in the gas and particle phase at ng m -3 concentrations. This work aims to identify dominant SCOs in Beijing and their precursors. The high time resolution measurements are utilized to probe their abundance under different chemical and environmental regimes providing insight into their formation. 10

Site description
The data presented here was collected during the measurement campaign "Photochemical smog in China" with an initiative to enhance our understanding of SOA formation via photochemical smog in China (Hallquist et al., 2016). 15 The campaign was coordinated by Peking University (PKU) and the University of Gothenburg with focus on spring/summertime episodic pollution episodes in Northeast China through gas and particle phase measurements. The setup was situated at a semi-rural site 40 km North East of downtown Beijing close to Changping town (40.2207° N, 116.2312° E). All on-line instruments sampled from inlets on the 4 th floor laboratory (12 metres above ground) at Peking University Changping Campus from the 13 th May to 23 rd June 2016, while filter measurements took place on 20 the roof. The average temperature and relative humidity throughout the campaign were 23°C and 44% respectively.
The wind speed averaged at 2 ms -1 from the South-South West. A total of 4 pollution episodes were observed during the campaign period, which are classified as sustained periods of high aerosol loading reaching a maximum of 115 micrograms per cubic meter (μg m -3 ) for PM1. The episodes were dominated by organic and nitrate aerosols although episode 3 contained a high sulfate loading, equal to that of nitrate. The mass loading for the semi-rural site showed 25 good correlation with the PKU campus measurement site (30 km South-South-West of the Changping site and 12 km North West of downtown Beijing) throughout the campaign allowing for extrapolation of the semi-rural site results to inner city conditions. HYSPLIT back trajectory results showed the pollution episodes often correlated with air masses coming from the direction of Beijing (South-South-East). Clean air days were mostly with North Westerly winds with clean air coming from the rural mountain regions North West of Beijing and the measurement site. 30 A high resolution Time of Flight Aerosol Mass Spectrometer (ToF-AMS) was utilized to measure the mass concentrations and size distributions of non-refractory species in submicron aerosols, including organics, sulfate, ammonium and chloride (DeCarlo et al., 2006;Hu et al., 2013). The setup of this instrument has been previously described by Hu et al. (2016). An Ionicon Analytik high sensitivity PTR-MS (Proton TRansfer Mass Spectrometer) as described by de Gouw and Warneke et al. (2007) provided supporting precursor VOC measurements.

ToF-CIMS setup 5
Gas and particle phase species were measured using an iodide ToF-CIMS coupled to the FIGAERO inlet (Lopez-Hilfiker et al., 2014). The ToF-CIMS can be operated in either negative or positive ionization modes, and a variety of reagent ion sources can be used. In this work the ToF-CIMS was operated in single reflection mode. The negative Iodide ion (I -) was used as the reagent in all experiments. Dry ultra high purity N2 was passed over a permeation tube containing liquid CH3I (Alfa Aesar, 99%), and the flow was passing a Tofwerk X-Ray Ion Source type P (operated at 10 9.5 kV and 150 μA) to produce the ionization ions. The ionized gas was then directed to the Ion-Molecule Reaction (IMR) through an orifice (Ø = 1 µm). Reaction products (e.g., compound X) were identified by their corresponding cluster ions, XIor the deprotonated ion, allowing for the collection of whole-molecule data. The nominal reagent and sample flow rates into the IMR chamber of the instrument were 3.5 liters per minute (LPM) and 2 LPM respectively.
The IMR itself was temperature controlled at 40˚C and operated at a nominal pressure of 500 mbar. The ToF-CIMS 15 was configured to measure singularly charged ions with a mass to charge ratio (m/z) of 7 -620, a reduced mass range in order to compensate for the lower count rate emitted by the soft X-ray source with respect to the Polonium-235 radioactive source as commonly deployed. The tuning was optimized to increase sensitivity, which resulted in a spectral resolution of 3500. The mass range was at some instances during the campaign changed to a higher mass range (1000 m/z) to ensure no major contributing peaks were being unaccounted for. Perfluoropentanoic acid was utilised as a mass 20 calibrant up to m/z 527 through its dimer and trimer. This range of mass calibration peaks also limited accurate peak identification above m/z 620.

FIGAERO inlet
The FIGAERO inlet collected particles on a Zefluor® PTFE membrane filter. The aerosol sample line was composed 25 of 12 mm copper tubing, while 12 mm Teflon tubing was used for the gas sample line. The FIGAERO was operated in a cyclic pattern; 25 minute of gas phase sampling and simultaneous particle collection, followed by a 20 minute period during which the filter was shifted into positioned over the IMR inlet and the collected particle mass was desorbed.
Desorption was facilitated by a 2 LPM flow of heated UHP N2 over the filter. The temperature of the N2 was increased from 20 to 250˚C in 15 minutes (3.5˚C min -1 ), followed by a 5 minute temperature soak time to ensure that all remaining 30 mass that volatilizes at 250˚C was removed from the filter. The resulting desorption time series profiles allowed for a distinct separation of measured species as a function of their thermal properties. 6 2.5 SCO Measurement

Identification
Spectral analysis was performed using Towfware V2.5.11. the average peak shape for the tuning utilised for this campaign was used to calculate the mass resolution and optimization of the baseline fit. The mass spectrum was mass 5 calibrated (allowing for accurate centroid peak position to be estimated, improving on a Gaussian assumption) using 4 ions up to mass 527 (the dimer of perfluorpentanoic acid) and applying a custom peak shape to achieve accurately peak identification below 5 ppm error across the mass range (0-620 AMU). A time series of the mass calibrant error for the entire campaign (figure SI1) illustrates how the error deviated by only +/-1 throughout the measurement period. This provides confidence that variation in signal and peak positioning did not result in large errors of identification and 10 quantification of the analyzed peaks. A priori unknown peaks were added to resolve overlapping peaks on the spectra until the residual was less than 5%. Each unknown peak was assigned a chemical formula using the peaks exact mass maxima to 5 decimal places and also isotopic ratios of subsequent minor peaks. An accurate fitting was characterized by a ppm error of less than 5 and subsequent accurate fitting of isotopic peaks. An example of the spectra and peak fitting can be found in Figure 1, highlighting the mass spectral fit for GAS and C9H9SO5.. Although the structure cannot 15 be determined with CIMS, it is assumed that no fragmentation of larger SCO species contribute to the SCO identified due to the soft ionization technique employed. The SCOs were identified in the spectra as negative ions assumed to be formed by hydrogen removal. Here, we present 17 SCOs that were identified in the mass spectra, which are displayed in Table 1 with their respective exact mass, formula, literature nomenclature and possible precursors. The peak fittings for all 17 SCOs is presented in the supplementary (SI2). All 17 SCOs represented a significant signal in the average 20 desorption spectra from the particle phase analysis. It must be noted that gas phase spectra at times contained other ions at a similar mass to the SCOs that contributed to higher counts than the SCO. This may result in a variable error to the measurement, although this should be at a minimum due to the use of a custom peak shape and low general mass calibration error of the spectra. The SCOs detected ions ranged from 154.96 m/z (GAS) to 294.06 m/z (C10H11NSO7).
The number of oxygen in the SCO ranged from O3 (C7H7SO3) to O7 (C5H8SO7). It is acknowledged that the CIMS may 25 not detect all SCOs in the ambient air due to peak fitting resolution limitations and limits of detection, therefore enabling the possibility for misrepresentation of the dominant SCO and an underestimation of total abundance. However, no physical features of the SCO (structure, O:C ratio, mass etc.) should inhibit the CIMS identifying the major SCO in the Beijing ambient air. Consequently, we here, and to facilitate descriptions of the relationship between individual SCOs and the total SCO measure, assumed that the measured SCOs do represent a significant fraction. 30

Quantification of SCOs
The OS and nitroxy organo sulfates (NOS) calibrations normalized to formic acid calibrations (as described in Le Breton et al., 2012Breton et al., , 2013 to account for any drift in sensitivity throughout the campaign. This relative sensitivity technique has been previously utilized for N2O5 and ClNO2 and has been verified with laboratory experiments (Le Breton et al., 2014). As a result of low mass range of the SCOs, common functionality, relatively small change in 5 polarity and lack of available stable SCO standards, we calibrated for 2 SCOs (lactic acid sulfate (LAS) and NP OS) and applied an average sensitivity for all the SCOs detected in Beijing. The ToF-CIMS sensitivity utilizing iodide as a reagent ion is known to vary by up to 3 orders of magnitude; therefore, further work is necessary to develop SCO standards and assess possible variations in sensitivity. NP OS is available commercially from Sigma Aldrich and was utilised to calibrate for the NOSs. L(+) Lactic acid from Sigma Aldrich (95%) was utilised as the preliminary agent for 10 lactic acid sulfate synthesis and was produced using the same technique as used by Olson et al. (2011). Briefly, a solution of 76.1 mg, 1.29 mmol, lactic acid in 2 mL di-methyl-formamide (DMF) was added dropwise to sulphur trioxide pyridine (0.96 g, 7.75 mmol) in 2 mL DMF at 0 o C. The solution is then stirred for 1 hour at 0 o C and 40 minutes at room temperature, the solution is re-cooled to 0 o C and trimethylamine (0.23 mL, 1.66 mmol) was added for quenching and the mixture wad further stirred for 1 hour. The solvent is then evaporated under vacuum and NMR is 15 directly utilised to calculate the purity which was found to be 8.2%.
A known mass of the solid calibrant (NP OS and Lactic acid sulfate) was added to 3 different volumes of milliQ water to produce different concentration standards. A known volume of each solution was then placed onto the FIGAERO filter and a desorption cycle was performed. The total ion counts for the high resolution (HR) SCO peak relates directly to the sensitivity of the system with respect to total ion counts per molecule reaching the detector. Figure 2 shows a 3-20 point calibration curve for NP OS and the corresponding thermogram, mass spectra and peak fit. The sensitivity of LAS and NP OS calibrations was calculated to be 2.0 and 1.6 ion counts per ppt Hz -1 respectively. All SCO were calibrated using the LAS sensitivity and all NOS using the NP OS sensitivity.
During desorption of both SCOs, fragmentation of the organic core and sulphate group was observed resulting in a desorption profile at m/z 97 (the bisulphate ion) and the deprotonated organic mass, i.e., C3H5O3 for lactic acid. A 25 number of different temperature ramping rates was performed with the FIGAERO to further probe the fragmentation and it was found that an increase in ramp rate ( o C/minute) decreased the calculated sensitivity due to an increase in fragmentation. This not only serves to highlight how the calibration tests of a species must mimic the exact measurement conditions, but also suggests potential interferences from fragmentation on the organic m/z´s. The relatively low concentration of the organic precursor with respect to the SCO results in little error in quantification, although this ratio 30 may significantly change in different air masses and a number of products of organic oxidation may fragment resulting in a significant error. This fragmentation can also be observed within the high resolution thermograms of the FIGAERO as a double desorption and further highlights the necessity for detailed thermogram analysis to accurately deconvolve desorptions relevant only to particle loss from the filter and not fragmentation or ion chemistry in the IMR. The fragmentation is considered to be constant throughout the campaign. The error for the SCO measurements may vary for each individual SCO possibly due to structure, volatility and fragmentation. It is commonly accepted within the literature for compounds lacking calibration that a functional group sensitivity can be applied (e.g., Lee et al. (2016) for organic nitrates (ONs)). Here we calculate an average error of 52 % for the SCOs, calculated using the standard 5 deviation of the NP OS calibration time series data.
The limitation of FIGAERO temperature ramps to 250 ˚C may result in further error as some SCOs may not be fully desorbed from the filter due to their low vapour pressures. To evaluate the mass left on a filter, several double desorption cycles were performed where mass is collected and desorbed such as in standard use. This is performed by re-heating the same filter once cooled to attain a second thermogram of the same filter. The second thermogram exhibited an 10 average of 90% reduction of counts for the SCO, although the NOSs had an average decrease of 82% counts. This indicates that most, but not all mass, are removed from the filters when desorbing. For the interpretation of the results of the field campaign this effect will induce a small distortion on the time evolution of SCOs when comparing to other parameters, e.g., 9% of NOSs will remain on the filter and being subjected to the subsequent desorption cycle. 15

Offline and online measurement comparison of SCOs
Filter measurements, for orbitrap and HPLC MS analysis, were taken diurnally at the same sampling site, although from a different inlet and location in the building. The CIMS hourly desorption data was averaged over the corresponding collection period to attain a day and nighttime CIMS data point. The period between the 23 rd May and 1 st June was selected due to all instrument measurements being undisturbed during this period. It must be noted that 20 CIMS data is lacking one data point daily while the background filter measurements were taken. The CIMS, orbitrap and HPLC do not measure all of the same species. Here a comparison of total 5 ions is presented with the HPLC and 2 from the orbitrap, where a further extensive comparison is to be performed in an accompanying manuscript. Figure SI3 illustrates the time series of CIMS (hourly and diurnal) measurements of GAS and IEPOX sulfate alongside the HPLC measurements. The diurnal data agrees well with an R value of 0.78 and 0.82 for GAS and IEPOX sulfate respectively. 25 The sum of the time series we have multi instrument data for (GAS; IEPOX sulfate, LAS, C4H7SO7, C5H11SO7 and C5H7SO7) for HPLC and C9H11SO5 and C9H8SO5 for orbitrap is displayed in the top panel. In general, the time series agree well and also have a good correlation (R = 0.7 and 0.81 for HPLC and orbitrap respectively) illustrating the ability for CIMS to agree with the offline methods and measure the SCOs accurately at low and high time resolution.

9
The KEMS technique was utilized to measure the vapour pressure of SCOs observed in the gas phase measurements by the CIMS. The KEMS technique is able to measure vapour pressures from 10 -1 to 10 -8 Pascal (Pa) ranging from volatile organic compounds to extremely low volatility organic compounds. A full description of the technique can be found in Booth et al. (2009Booth et al. ( , 2010 and the measurements of a series of compounds over a large vapour pressure (VP) range, in a recent inter-comparison study from this instrument, can be found in Krieger et al. (2017). Briefly, the 5 instrument consists of a temperature controlled Knudsen effusion cell, suitable for controlled generation of a molecular beam of the sample organic compounds in a vacuum chamber, coupled to a quadrupole mass spectrometer. The cell has a chamfered effusing orifice with a size ≤1/10 the mean free path of the gas molecules in the cell. This ensures the orifice does not significantly disturb the thermodynamic equilibrium of the samples in the cell (Hilpert, 2001). The system is calibrated using the mass spectrometer signal from a sample of known vapour pressure, in this case malonic 10 acid (vapour pressure at 298K = 5.25×10 -4 Pa (Booth et al., 2012)). A load-lock allows the ioniser filament to be left on, then a new sample of unknown vapour pressure can be measured. Solid state vapour pressures measured in the KEMS can then be converted to sub-cooled liquid vapour pressures using the melting point, enthalpy and entropy of fusion, which are obtained by using a Differential Scanning Calorimeter (DSC) (TA instruments Q200).

SCO contribution to PM1 at Changping
The SCOs measured at the Changping site had a mean campaign concentration of 210+ 110 ng m -3 (Table 1). The highest concentration of total SCOs during the campaign was 540 ng m -3 and the lowest 40 ng m -3 , thus they are 20 omnipresent and have significant sources during most atmospheric conditions. These concentrations are consistent with Stone et al. (2012) reporting an average SCO concentration of 700 ng m -3 in a number of rural and urban sites in Asia.
A mean SCO contribution to organic aerosol (OA) in the work presented here was calculated to be 2.0  1% (Table 1), within the range of values calculated by Stone et al. (2012) (0.8% to 4.5%), further supporting evidence that the SCO contribution to PM1 mass is relatively low in Asia. The CIMS cannot claim to measure total SCO, rather than singularly 25 identify and measure SCOs contributing to the total mass loading. Therefore, the SCO contribution reported in this work should be considered as a lower limit. The Liao et al. (2015) study also supports the idea that the SCO contribution to PM1 mass in anthropogenically dominated regions is less significant than that from biogenically dominating air masses by observing a significantly higher contribution of IEPOX sulfate to PM1 mass on the East coast of the United States (1.4%) than the West Coast (0.2%). 30 The observation of higher relative contribution of SCOs to total organics in more remote regions compared to a densely populated urban area, supports the idea that SCOs provide a higher contribution to PM in aged air due to their secondary production pathways. Similar to Lahore (as studied by Stone et al., 2012), Beijing has many strong primary anthropogenic sources which will dominate the mass loading and therefore, initially, will contribute to a lower fraction of the total concentration from secondary production due to limited processing near the source. Throughout the campaign, a good correlation (R 2 = 0.66) was observed between an increase in SCO mass and PM1 mass, although the SCO contribution to PM1 decreased exponentially (Figure 3) indicating that the pollution episodes contain a lower 5 fraction of SCOs with respect to total PM1. This result suggests that SCO do not play as large a role as expected even though their precursors (organics and sulfate) are abundant within the episodes, indicating the conditions of their formation may be more vital than the absolute concentrations of precursors.

Gas to particle phase partitioning of SCOs 10
The FIGAERO ToF-CIMS data exhibited indication of SCOs in both the particle and gas phase. Previous studies have supported the existence of, e.g., gas phase GAS in ambient air (e.g., Ehn et al., 2010), although some work has attributed other measurement techniques detection of gas phase SCO to result from measurement artefacts (Hettiyadura et al., 2017;Kristensen et al., 2016). Once all HR peaks have been identified, the batch fitting and HR time series for the whole data set is processed and then separated into gas phase measurements and particle phase desorption profile time 15 series. The data is background corrected, i.e., subtraction of both the gas phase background periods and blank filter desorption's. Upon analysis of the resultant data, significant concentrations of gas phase SCOs were observed. Figure   4 depicts the overall sum SCO mass concentration time series in the gas and particle phase. The mean contribution from gas phase SCO to total SCO was found to be 11.6%, 23+8 ng m -3 . This suggests a significant amount of SCO is always present in the gas phase and factors that influence gas-to-particle partitioning influence the level of this contribution. 20 These changes in contribution also reduce the possibilities for memory effect, e.g., one possibility is the deposition of SCOs onto the IMR walls during the temperature ramp of the desorption which in time may de-gas and be observed in the gas phase. This would likely result in a constant ratio of particle to gas phase concentrations and would likely cause a hysteresis in the observed gas phase measurements with respect to the particle phase, which was not observed.

11
(2015) VPs; 6.2x10 -3 , 1.3x10 -3 , 1x10 -3 Pa. Using this agreement for well-known substances we notice the Tmax of SCO to be in the range where it can provide significant gas-phase concentration. Still, the observed presence of gas phase GAS and IEPOX-OS does not agree with previous studies of these compounds (Stone et al., 2012, Hettiyadura et al., 2017. Therefore, one needs to be cautious and deeper analysis into exact VPs and partitioning from the present work must be performed to assess whether their gas phase presence could be fully confirmed. So far we note that 5 fragmentation of organic species (oligomers) during desorption could lead to a potential artefact and a lower Tmax at a monomer peak (Stark et al., 2017 andLopez-Hilfiker et al., 2016). However, here we identify and expect no dimers or oligomers that could fragment to form the SCOs identified. Furthermore, the higher mass organics are likely to have a much higher VP than the lower mass SCO and provide a second Tmax which would produce a lower VP value due to the greater energy required to break the bonds. Analysis of Tmax throughout the campaign shows no double peak 10 thermograms and an acceptable stability of Tmax (SI4). Tmax varied by up to 14 degrees Celisius and appeared to correlate well with particulate loading, similar to that observed by Huang et al. (2018), who suggested that this is a result of diffusion limitations within the particle matrix. If the data is tentatively analyzed to assess the mechanism regarding their partitioning, aerosol liquid water content would affect the partitioning of gas phase compounds to aerosols (Zhang et al., 2007). Data point size coding the correlation of the gas and particle phase SCO concentrations indicates 15 partitioning towards the aerosol phase at lower relative humidities (Figure 4). Conversely, as the temperature increases (as indicated by red colour shading) the SCOs partition further towards the gas phase, as thermodynamically expected.
Further work is necessary to validate these findings and determine the mechanisms and importance of gas phase SCO abundance in ambient air. For example, the high contribution in the gas phase could be perturbed if equilibrium between condensation to particle phase and gas phase formation has not been established. It must be noted that the a correct 20 calibration of Tmax with VP would be necessary to extract such information, but qualitatively the relative VP compared to NP OS could be utilized as a reliable scale due its independent calibration by KEMS.

SCO sources at the Changping site
SCOs are known to have biogenic and anthropogenic sources and some have multiple sources from both, e.g., GAS (Hettiyadura et al., 2017;Hansen et al., 2014). Burning events are known to emit high levels of organics and nitrates and potentially sulfur, depending on the type of fuel used. This enables biomass burning to be a potential anthropogenic and biogenic source of SCOs. The site at Changping was influenced by both regional anthropogenic pollution from the 30 Beijing area and localized anthropogenic activity (industry, biomass burning and traffic) but also emissions from biogenic sources, as it is situated in a semi-rural area, with forest, vegetation and plantations. This was evident from the benzene and isoprene PTR-MS measurements which have mean campaign concentrations of 0.55 + 0.4 and 0.27 + 0.19 ppb respectively with maxima of 5 and 1.5 ppb respectively. Thus, as shown in Figure 5, PTR-MS measurements mean daily concentrations were utilized to evaluate if the ratio between benzene and isoprene can indicate a higher mass loading and contribution of aromatic and biogenic SCOs measured in this work. Data on days with incomplete time series have been removed to ensure the data presented represents a full mean of the day concentration. A good 5 correlation between the benzene:isoprene ratio and sum of SCOs is observed. It suggests an increase in relative anthropogenic emissions promotes an increase in total SCO loading. It should be noted that C6H10SO7 has no known precursor in the literature, although it contributes significantly to the SCO mass loading in this work (16%).

Biogenic and anthropogenic SCOs 10
Biogenic SCOs are known to be comprised of monoterpene, sesquiterpene and isoprene derived SCOs which have been identified in rural, sub-urban and urban areas around the world, and have been shown to be a major constituent of SOA Shalamzari et al., 2014, Liao et al., 2015. IEPOX sulfate is commonly found to be the most dominant SCO at many locations and was identified also at the Changping site. The IEPOX sulfate mean concentration represented 0.11% of the OA mass, agreeing well with concentrations found in Western USA (significant 15 anthropogenic emissions) and lower than the Eastern USA as expected due to higher biogenic and isoprene emissions (Liao et al., 2015). Although IEPOX sulfate is considered one of the most abundant individual organic molecules in aerosols (Chan et al., 2010), here our results show it only contributed to 2% of the SCO mass and was the 8 th most abundant SCO in the particle phase. Additionally, two other isoprene derived SCOs, C5H8SO7 and C4H8SO7, were measured by the CIMS with mean campaign concentrations of 2 and 3 ng m -3 respectively and a contribution of 0.02% 20 to OA mass. The highest contributing biogenic SCO to the ambient air was a NOS, C10H16NSO7, a known NOS derived from alpha-pinene oxidation. This NOS had a mean campaign concentration of 21 ng m -3 and a 0.2% contribution to OA mass.
Anthropogenic SCOs, including PAH derived SCOs have received more attention in recent studies due to their identification (Nozière et al., 2010;Hansen et al., 2015). Aromatic SCOs and sulfonates have only recently been 25 identified as atmospherically abundant SCOs (Riva et al., 2015). In this work we find that the PAH derived SCO C11H11SO7 is the most dominant SCO in Beijing with a mean concentration of 40 ng m -3 , contributing to 20% of the total SCO mass and 0.4% of the OA mass. This SCO has been identified in laboratory studies as an SCO forming from the photo-oxidation of 2-methyl napthalene, one of the most abundant gas phase PAHs and is thought to represent a missing source of urban SOA (Riva et al., 2015). This work presents the possible significance of PAH SCOs in Beijing 30 and further evidence that photo-oxidation of PAHs represents a greater SOA potential than currently recognized. A further 8 anthropogenic aromatic derived SCOs were identified as common components of the PM1 representing more than half of the total SCOs with C7H5SO4 contributing to 24 ng m -3 and 0.23% OA mass. The total anthropogenic related SCOs had a mean mass of 120 ng m -3 and contributed to 1.2% of the OA mass.

Biomass burning source of SCOs
NP (a product of benzene oxidation and nitration) has previously been detected in the gas and aerosol phase (Harrison 5 et al., 2005) and is an important component of brown carbon (Mohr et al., 2013). NP has primary sources, such as vehicle exhausts and biomass burning (Inomata et al., 2013 andMohr et al., 2013) and secondary sources via the photooxidation of aromatic hydrocarbons in the atmosphere (Harrison et al., 2005). High levels on anthropogenic activity, biomass burning and strong photochemistry in Beijing therefore enable this region to be a strong potential source of NP. Both NP and NP OS diurnal profiles exhibit an increase in the morning (6 am onwards) although NP OS appears 10 to increase in concentration more rapidly. The early morning biomass burning and anthropogenic activity are likely to contribute to production of both species, although the higher sulphate content of the biomass burning emissions may promote a faster production of NP OS and conversion of NP to NP OS. Both compounds continue to increase with a photochemical profile with one peak at midday but also a peak around 4 pm, likely to be a second source of the day from anthropogenic activity. The NP OS continues to increase until sunset, which could result from further 15 photochemical production from the NP emitted throughout the day whereas NP falls off after the 4 pm peak. The campaign time series for NP and NP OS can be seen in Figure 6. Unlike its precursor and most other pollutant markers measured in this work, including all other SCOs, NP OS exhibits higher concentrations between 17th to the 22nd May compared to the 28th May to 3rd June. The only compound with a similar campaign profile is acetonitrile (a marker for biomass burning), which has significantly enhanced concentrations between 6 and 8 am from the 17th to the 22nd May. 20 Back trajectories of these two time periods show the air mass during the first period comes from the west, a more rural region of China and known to be influenced heavily by biomass burning, whereas the second time period has wind directions mainly bringing in air masses that have gone through the Tianjin and Beijing area. It is therefore hypothesized that the NP OS, which peaks later in the day than the NP and acetonitrile, is a secondary product formed from the biomass burning and has aged after being emitted from air masses further away. Here NP can have more local sources 25 of biomass burning and traffic which then can contribute to NP OS production, but at a slower time scale, which in this data set, appears as lower production due to the limited oxidation of local air masses.

Precursor analysis 30
The availability of the organic precursors of SCOs is a limiting factor for the SCO production rate. The measurement of the precursors in the gas and particle phase by CIMS enables a more descriptive mechanism to be outlined as the partitioning of the precursor will vary the distribution between gas and particle production pathways and therefore rate of corresponding SCO production. Glycolic acid (GA) has on average 75% of its mass in the gas phase for the measurement period whereas GAS is dominantly in the particle phase (Figure 7). The GAS particle phase concentration is observed to increase as the SO4 2mass loading increases and the GA gas and particle concentrations increase, although the partitioning of the GA towards the gas phase restricts the SCO production. This can be seen in Figure 7 as for a 5 given SO4 2concentration, the data with warm colors (red), representing a high fraction of precursor GA in the particle phase, generally provides a higher concentration of particulate GAS.
The main formation mechanism of GAS is thought to be via the reaction of GA in the gas phase with an acidic aerosol sulfate (Liao et al., 2015), contrary to what is observed here. Although an increase in GAS is observed to correlate with the GA, it appears that a partitioning towards the particle phase promotes GAS production. An R 2 correlation of 0.68 is 10 observed between GAp and GASp whereas an R 2 of 0.4 is observed between GAg and GASp.
The sum of benzene SCOs exhibits a good correlation to the gas phase benzene time series (Figure 8), although their abundance should also rely on the availability of sulfur in the particle phase and the age of the air mass, if it is assumed that they are formed via secondary reactions of primary pollutants. In order to assess how the SCO production rates may vary due to these factors, two distinct high benzene SCO events with similar benzene concentrations were 15 scrutinized, i.e., the 29th May and the 1st June. The first period has lower SO4 2concentrations, higher H2SO4 levels and a higher total benzene SCO concentration. The exact age (or time for oxidation) of compounds in an air mass are without an extensive modelling study complicated to derive. However, as proxies to attain an approximation about oxidation state one may use some trace compounds. Monoterpene oxidation by the hydroxyl radical (OH) or O3 results in the formation of multifunctional organic acids such as pinonic acid which can then be further oxidised by OH to 20 form 3-methyl-1,2,3-butane-tricarboxylic acid (MBTCA), both of which are measured by CIMS. Therefore, in an air mass containing monoterpene emissions, as known here through the identification of their products such as C10H16NO7S, we can utilize the ratio of pinonic acid and MBTCA; as tracers of monoterpene SOA processing as detected in ambient aerosols in Europe, USA and the Amazon (Gao et al., 2006) as a relative photochemical clock.
During the high benzene event on the 1st June, according to the pinonic acid: MBTCA ratio, the air mass is less oxidized 25 relative to the air mass on the 29th May (Figure 8). This would allow less time for secondary production and explain the relatively lower concentration of SCOs, irrespective of higher SO4 2concentrations and similar benzene concentrations.
To further elaborate The Gothenburg Potential Aerosol Mass (Go:PAM) reactor (Watne et al, 2018) was tested and utilized to simulate aging of the air mass during periods of the campaign. Here the ratio between pinonic acid and 30 MBTCA was observed to increase by an average of 3 during aging within the Go:PAM which has been calculated to be the OH exposure equivalent of 2 days in the ambient atmosphere. As this ratio increased with aging, the SCO concentration also increased exponentially, further supporting the secondary production of SCO in photochemically aged air mass. Although limited data is available here for simultaneous Go:PAM and CIMS measurements, the results indicate the potential utilization of the chamber to probe secondary production processes.

Aerosol acidity
The molecular ion H3S2O7was identified in the mass spectra throughout the campaign, which has previously been 5 detected by Liao et al. (2015) using a Particle Analysis Laser Mass Spectrometer to measure SCOs. They attribute this mass to be a cluster of HSO4with sulfuric acid (H2SO4). Particles in the presence of H2SO4, and therefore high acidity, form this cluster whereas neutralized ions are likely to favour the unclustered HSO4form. Therefore, the ratio between the cluster and the bisulfate ion increases with increasing aerosol acidity (Murphy et al., 2007;Carn et al., 2011). Liao et al. (2015 validate the appropriateness of this cluster as a marker for aerosol acidity thorough comparisons to a 10 thermodynamic model with gas and aerosol phase measurement inputs. Acidity was also calculated utilizing the gas and particle phase H2SO4 and liquid H + ion concentration analysed using an offline technique, as described by Guo et al. (2010), from diurnal samples taken at the site. This method showed good agreement with the integrated diurnal counts of the H3S2O7ion. Therefore, we employ the HSO4 -.H2SO4cluster in this work as a qualitative scale for particle acidity utilizing similar assumptions. Figure 9 shows how total SCO mass concentration generally increased as total 15 organic mass from the AMS increased. The correlation indicates that higher acidity (darker colours) tends to promote formation of SCOs when in the presence of high levels of organics and SO4 2-(larger symbol sizes), supporting the growing consensus that aerosol acidity plays an important role in ambient SCO formation. This importance of acidity agrees well with both the acid-catalyzed epoxydiol ring opening formation mechanism (SurratttSurratt et al., 2010) and the sulfate radical initiated SCO formation because efficient formation of sulfate radicals also requires acidity 20 (Schindelka et al., 2013).

Conclusions
The FIGAERO ToF-CIMS was successfully utilized for the ambient detection of 17 SCOs in Beijing in the gas and aerosol phase with limits of detection in the ng m 3 range. Good agreement with offline filter measurements further supports the robustness of this method for high and low time resolution measurements of SCOs. Further calibrations and comparisons to total SCO measurements are required to evaluate its performance limitation with regard to 30 sensitivity application and peak identification. The SCOs measured by CIMS contributed to 2% of the OA at the semirural site, highlighting the relatively low contribution of SCOs in Beijing, an anthropogenically dominated environment. This calculation from CIMS may only be valid to infer each individual SCO contribution to total SOC mass as limitations in SCO identification and quantification limit the CIMS ability for total SCO measurements. Significance of their secondary production pathway prevailed, although still present in relatively fresh air masses. Contributions of SCO to total organics (2.0+1%), sulfate (15+19%) and PM (1.0+1.4%) indicate the concentrations observed in Beijing result from highly processed ambient air masses. 5 Gas phase SCOs were identified for all the SCOs measured at the site, contributing to in average to 12% of the total SCO mass. The possibility of gas phase SCOs in ambient air was supported by KEMS vapour pressure measurements of NP OS and derived Tmax values which suggest a vapour pressure in the semi volatile range. The partitioning towards the gas phase was more efficient at high atmospheric temperatures, while lower relative humidities promoted partitioning to the particle phase. 10 Biogenic SCOs contributed a small fraction of the total SCO mass at Changping and were dominated by an -pinene derived OS with 0.2% contribution to the OA mass. IEPOX sulfate was only the 8 th most abundant SCO measured, contrary to common reports that it is one of the most abundant SCOs. Anthropogenic precursors contributed to more than half of the SCO mass loading with a PAH derived SCO contributing to as much as 1.2% of the OA mass. Benzene derived SCOs correlated well with gas phase benzene levels and were heavily influenced by photochemical aging. The 15 contribution of each benzene derived SCO to total benzene derived SCO mass varied daily and throughout the campaign highlighting the complexity of the atmospheric processing and composition of SCO. Significant contributions from aromatic SCOs highlight the importance of anthropogenically emitted organics in the Beijing region and their contribution to the Beijing outflow and subsequent photochemistry. NP OS was attributed to biomass burning emissions due to co-occurrence with high levels of acetonitrile. This highlights the importance of anthropogenic emissions and 20 their contribution to SOA from the urban Beijing outflow.
A qualitative CIMS marker for aerosol acidity highlighted the increase in SCO production rate in acidic aerosols in the presence of high SO4 2and organics. The correlation of SCO production and RH becomes more complex for individual SCOs, which cannot be resolved within this studies framework.