Molecular characterization of gaseous and particulate oxygenated compounds at a remote site in Cape Corsica in the western Mediterranean Basin

Abstract. The characterization of the molecular composition of organic carbon in both
gaseous and aerosol is key to understanding the processes involved in the
formation and aging of secondary organic aerosol. Therefore a technique
using active sampling on cartridges and filters and derivatization followed
by analysis using a thermal desorption–gas chromatography–mass spectrometer (TD–GC–MS) has been used. It is aimed at studying the molecular composition of
organic carbon in both gaseous and aerosol phases (PM2.5) during an
intensive field campaign which took place in Corsica (France) during the
summer of 2013: the ChArMEx (Chemistry and Aerosol Mediterranean Experiment)
SOP1b (Special Observation Period 1B) campaign. These measurements led to the identification of 51 oxygenated (carbonyl and
or hydroxyl) compounds in the gaseous phase with concentrations between 21 and 3900 ng m−3 and of 85 compounds in the
particulate phase with concentrations between 0.3 and 277 ng m−3. Comparisons of these measurements with collocated data using other
techniques have been conducted, showing fair agreement in general for most
species except for glyoxal in the gas phase and malonic, tartaric, malic and
succinic acids in the particle phase, with disagreements that can reach up to
a factor of 8 and 20 on average, respectively, for the latter two acids. Comparison between the sum of all compounds identified by TD–GC–MS in the
particle phase and the total organic matter (OM) mass reveals that on
average 18 % of the total OM mass can be explained by the compounds
measured by TD–GC–MS. This number increases to 24 % of the total water-soluble OM (WSOM) measured by coupling the Particle Into Liquid Sampler (PILS)-TOC (total organic carbon) if we consider only the sum of the
soluble compounds measured by TD–GC–MS. This highlights the important
fraction of the OM mass identified by these measurements but also the
relative important fraction of OM mass remaining unidentified during the
campaign and therefore the complexity of characterizing exhaustively the
organic aerosol (OA) molecular chemical composition. The fraction of OM measured by TD–GC–MS is largely dominated by
di-carboxylic acids, which represent 49 % of the PM2.5 content
detected and quantified by this technique. Other contributions to PM2.5
composition measured by TD–GC–MS are then represented by tri-carboxylic
acids (15 %), alcohols (13 %), aldehydes (10 %), di-hydroxy-carboxylic
acids (5 %), monocarboxylic acids and ketones (3 % each), and
hydroxyl-carboxylic acids (2 %). These results highlight the importance of
polyfunctionalized carboxylic acids for OM, while the chemical processes
responsible for their formation in both phases remain uncertain. While not
measured by the TD–GC–MS technique, humic-like substances (HULISs) represent the
most abundant identified species in the aerosol, contributing for 59 % of
the total OM mass on average during the campaign. A total of 14 compounds were detected and quantified in both phases, allowing the
calculation of experimental partitioning coefficients for these species. The
comparison of these experimental partitioning coefficients with theoretical
ones, estimated by three different models, reveals large discrepancies
varying from 2 to 7 orders of magnitude. These results suggest that the
supposed instantaneous equilibrium being established between gaseous and
particulate phases assuming a homogeneous non-viscous particle phase is
questionable.


key to understand the processes involved in the formation and aging of secondary organic aerosol. 23 Therefore a technique using active sampling on cartridges and filters and derivatization followed by 24 analysis using a Thermal Desorption-Gas Chromatography/mass spectrometer (TD-GC/MS) has been 25 used to study the molecular composition of organic carbon in both gaseous and aerosol phases during Comparisons of these measurements with collocated data using other techniques have been 1 conducted showing fair agreement in general for most species except for glyoxal in the gas phase and 2 malonic, tartaric, malic and succinic acids in the particle phase with disagreements that can reach up 3 to a factor of 8 and 20 on average, respectively for the latter two acids. 4 Comparison between the sum of all compounds identified by TD-GC/MS in particle phase with the total 5 Organic Matter (OM) mass reveal that 18% of the total OM mass can be explained by the compounds 6 measured by TD-GC/MS for the whole campaign. This number increase to 24% of the total Water 7 Soluble OM (WSOM) measured by PILS-TOC if we consider only the sum of the soluble compounds 8 measured by TD-GC/MS. This highlights the non-negligible fraction of the OM mass identified by these 9 measurements but also the relative important fraction of OM mass remaining unidentified during the 10 campaign and therefore the complexity of characterizing exhaustively the Organic Aerosol (OA) 11 molecular chemical composition. 12 The fraction of OM measured by TD-GC/MS is largely dominated by di-carboxylic acids which 13 represents 49% of the PM2.5 content detected and quantified by this technique. Other contributions to 14 PM2.5 composition measured by TD-GC/MS are then represented by tri-carboxylic acids (15%), alcohols 15 (13%), aldehydes (10%), di-hydroxy-carboxylic acids (5%), monocarboxylic acids and ketones (3% each) 16 and hydroxyl-carboxylic acids (2%). These results highlight the importance of poly functionalized 17 carboxylic acids for OM while the chemical processes responsible for their formation in both phases 18 remain uncertain. While not measured by TD-GC/MS technique, HUmic-LIke Substances (HULIS) 19 represent the most abundant identified species in the aerosol, contributing for 59% of the total 20 identified OM mass on average during the campaign. 21 14 compounds were detected and quantified in both phases allowing the calculation of experimental 22 partitioning coefficient for these species. The comparison of these experimental partitioning 23 coefficients with theoretical ones, estimated by three different models, reveals large discrepancies 24 varying from 2 to 7 orders of magnitude. These results suggest that the supposed instantaneous 25 equilibrium being established between gaseous and particulate phases assuming a homogeneous non-26 viscous particle phase is questionable. 27 28 1 Introduction 29 It is now recognized that aerosols have an impact on human health, climate and ecosystems. However at room temperature until the sampling. After sampling, cartridges are stored at room temperature 24 during 5 days, optimum for the derivatization step using PFBHA (Ho and Yu, 2002) Sampling are performed on Teflon quartz filters which are stored at -16°C after sampling waiting for 1 analysis. Derivatization is performed after sampling directly on filters. Filters are put in stainless steel 2 tubes cleaned following the same protocol than for carbonyl compounds. Tubes are then sealed and 3 maintained vertically with 10 µl of MTBSTFA put in the bottom cap for passive impregnation during 4 24h at room temperature. 5

Analytical system 6
The analytical system used in this study is composed by three successive modules: a thermal 7 desorption system, a gas chromatography unit and a mass spectrometer. 8 The thermal desorption allows the extraction of adsorbed compounds on sample support by increasing 9 the temperature without any preliminary solvent extraction and collecting them on a cold trap before 10 flash injection in GC/MS instrument. The thermal desorption system (Markes™, model unity 1) is 11 coupled with an automated system (Markes™, model Ultra 50:50). Thermal desorption parameters are 12 listed in Table 1

Internal calibration protocol 22
For a more efficient quantification, internal calibration has been set up for both family of compounds 23 (carbonyl and hydroxyl) and for both phases. This procedure aims at taking into account drift in MS 24 sensitivity and derivatization efficiency. Two types of internal standards are used: substitutes which 25 are deuterated compounds getting at least one derivatized function; and an internal standard which is 26 a compound with no derivatized function. 50 ng of Substitutes are added prior to the derivatization 27 step to take into account every steps of sample preparation as well as analysis steps. The list of 28 substitutes selected is given in Table 2. The internal standard selected is pentadecane and 50 ng is 29 added on cartridges grid just before the analysis. 30
Overall uncertainties have been determined taking into account precision, detection limit and 1 systematic errors (including uncertainties on standard concentrations, on calibration, on blank 2 determination and on sampling volume; following Gaussian error propagation). Overall uncertainties 3 have therefore been estimated to be 35% and 54% on averaged in gas phase for carbonyls and 4 hydroxyls and carboxylic acids respectively and to be 41% and 47% on averaged in particulate phase 5 for carbonyls and hydroxyls and carboxylic acids respectively. 6 sampling flow rate was set at 150 mL min -1 . The instrument was operated at a reactor pressure and a 19 temperature of 1.33 mbar and 40°C, respectively, leading to an E/N ratio of 135 Td. 20

Ancillary measurements
An automated zero procedure was performed every hour for 10 min. Humid zero air was generated by 21 passing ambient air through a catalytic converter to perform zeros at the same relative humidity than 22 ambient air. 23 Signals from protonated VOCs were normalized by the signals of H3O + and the first water cluster 24 H3O + (H2O) as proposed by de Gouw and Warneke (2007). Concentrations were calculated using Eq. (1): 25 During the campaign, NO and NO2 were measured by a commercial ozone chemiluminescence analyzer 4 (Cranox II; Eco Physics®) with a time resolution of 5 min. NO was measured directly, while NO2 was 5 converted into NO using a photolytic converter. O3 was measured using a commercial analyzer (TEI 49i; 6 Thermo Environmental Instruments Inc®) using UV absorption with a time resolution of 5 min. 7

Particulate ancillary measurements 8
Mass concentrations of PM10 and PM1 were measured during the campaign using two tapered element 9 oscillating microbalance (TEOM) equipped with a filter dynamic measurement system (FDMS) (Thermo 10 Scientific™). In addition, aerosol chemical composition was measured by online technique (aerosol 11 chemical speciation monitor -ACSM) and offline-method (Ion chromatography, GC/MS and HPLC) on 12 filters collected daily with 2 HiVol samplers (30 m 3 hr -1 ) equipped with PM1 and PM2,5 inlets. instrument with monodispersed (300 nm diameter) ammonium nitrate particles was performed 2 18 months before the campaign. Because ambient air was dried by a Nafion membrane and because 19 ammonium nitrate was low during the campaign, constant collection efficiency (CE) of 0.5 has been 20 kept. The Q-ACSM was operated continuously during the whole campaign at a time resolution of 30 21 min. 22

Ion Chromatography 23
Soluble anions and cations were analyzed by ionic chromatography (IC, ThermoFisher ICS3000) 24 following protocol similar to that described elsewhere (e.g. Jaffrezo et al., 1998). Briefly, 38 mm 25 diameter sub-samples from each filter were soaked for 20 min in 10 mL of Milli-Q water with orbital 26 shaking, and then filtered using 0,22 µm-porosity Acrodisc filters before analysis. ASA11-HC and CS16 27 columns were used for anions and cations analyses, respectively. (split ratio 50) at 280°C. The column temperature program was held at 65°C hold for 2 min, and ramped 8 at 6°C/min up to 300°C, followed by an isothermal hold at 300°C for 20 min. GC-MS response factors 9 were determined using authentic standards. Compounds, for which no authentic standard are 10 available, were quantified using the response factor of compounds with analogous chemical 11 structures. Field blank filters were also treated with the same procedure. 12

HPLC 13
The analysis of a large array of organic acids (including pinic and phthalic acids, and 3-MBTCA) was 14 conducted using the same water extracts as for IC and HPLC-PAD analyses. In brief, this was performed 15 OC fraction only and a VOC denuder was set upstream the collection to avoid semi-volatile VOC 7 contamination. Daily blanks were conducted every day for 1h by placing a total filter upstream of the 8 sampling system. 9

HULIS measurements 10
The water soluble HULIS fraction is analyzed according to a protocol described in detail in Baduel et al. fraction is subsequently collected manually and the carbon content is analyzed with a DOC analyser 20 (Shimadzu TOC-VCPH/CPN) by catalytic burning at 680°C in oxygen followed by non-dispersive infrared 21 detection of the evolved CO2. 22 23 3 Results and discussion 24  came from the south-west sector and 20% from the western sector (see Figure 1). Winds coming from 1 south-west sector are predominant during daytime and nighttime and correspond to wind speed 2 maxima. Winds from the west and north-east are also recorded, but during daytime only. Low NOx 3 concentrations were observed during the campaign (0.57 ppbv on average) with a few spikes above 1 4 ppbv corresponding to local influence from traffic especially when air masses came from the south 5 (e.g. 27 th July). 6

Particles and organic fraction 7
Mean, median, maximum and minimum of mass concentrations of PM10, PM1 and organic fraction in 8 NR-PM1 are summarized in Table 3 for the whole campaign. The averaged mass concentrations for 9  corresponding to a raise in temperatures. Overall, the organic fraction evolution follows the one of the 1 PM1 mass fraction. 2 identification, the substitute used to account for the derivatization efficiency, the external standard 10 used for their quantification, the fragment used for quantification and the averaged concentrations 11 measured in both phases. For the carbonyl compounds, the mono-functionalized compounds 12 identified contained from 3 (e.g. propanal) to 10 (e.g. decanal) carbon atoms and from 2 (e.g. glyoxal) 13 to 5 (e.g. 4-oxopentanal) carbon atoms for the bi-functionalized compounds. For the hydroxyl 14 compounds and the carboxylic acids, the mono-functionalized identified compounds contained from 15 3 (e.g. propanoic acid) to 18 (e.g. octadecanoic acid) carbon atoms. Several poly-functionalized 16 compounds have also been identified: hydroxy-acids and di-acids from 2 (e.g. glycolic acid) to 8 (e.g. 17 mandelic acid) carbon atoms; triols, di-hydroxy-acids, hydroxyl-di-acids, tri-acids from 3 (e.g. glycerol) 18 to 9 (e.g. 2-Hydroxy-4-isopropyl-hexanedioic acid) carbon atoms; and two tetra-functionalized 19 compounds (methyl-tetrols and citric acid). 20

Results from the TD-GC/MS analysis
It is worth noting that several compounds exhibited very close quantities in the air sample and in the 21 blank (designed as "blank" in the supplementary material 1). Therefore, the presence of these 22 compounds in the air sampled cannot be certain. For the compounds that have been quantified 23 successfully and present concentrations significantly above the quantification limit (3σ above averaged 24 blank measurements), higher levels are observed in the gas phase. The averaged concentrations 25 ranged from 21 ng m -3 (Mandelic acid) to 1600 ng m -3 (glycerol) for hydroxyl compounds in the gas 26 phase and from 0.3 (Pyruvic acid) to 277 (oxalic acid) ng m -3 in the particulate phase. For the carbonyl 27 compounds, the averaged concentrations ranged from 85 ng m -3 (hexanone) to 3900 ng m -3  Oxopentanal) in the gas phase and from 1 ng m -3 (e.g. methylpropanal or glyoxal) to 20 ng m -3 (4-29 methylpentanal) in the particulate phase. Figure 3 presents the distribution of all quantified 30 compounds along their saturation vapor pressure and their O/C ratio. The phases in which these 31 compounds were identified are also shown in Figure 3. While compounds only present in the gas or 32 aerosol phase exhibit high and low saturation vapor pressure, respectively, some exceptions are 33 noticeable. Indeed, some gaseous compounds have low vapor pressure (down to 10 -8.6 atm) such as 34 https://doi.org/10.5194/acp-2020-1051 Preprint. Discussion started: 29 October 2020 c Author(s) 2020. CC BY 4.0 License. long chain linear mono carboxylic acids (up to 15 carbon atoms) and some compounds only found in 1 the particle phase have high vapor pressure (up to 10 -0.8 atm), normally incompatible with their 2 presence in such phase, such as small mono carbonyls (e.g. methylpropanal, methylbutanone, 2-3 methylbutanal…). We also found compounds exhibiting high vapor pressure (up to 10 -0.4 atm) in both 4 phases, which is normally incompatible with their presence in aerosol phase, such as small carbonyls 5 (e.g. propanal, acrolein, methacrolein, MVK…). This latest point is discussed further in section 3.2.5. 6

Data intercomparison 7
A comparison of data measured by TD-GC/MS with other techniques available on site has been 8 performed, for both phases, to test the reliability of these measurements. the disagreement observed here is related to an underestimation of the concentrations measured by 1 DNPH cartridge analysis. Furthermore, recent studies on humidity dependence of the DNPH-HPLC-UV 2 method for some ketone compounds, revealed that the collection efficiency is inversely related to 3 relative humidity, with up to 35 %-80 % of the ketones being lost for RH values higher than 50 % at 4 22 °C (Ho et al., 2014). Furthermore, dimerization issues for MVK during analyses using DNPH method 5 has also been identified, during more recent measurements, that can cause strong underestimation of 6 this technique (>50%). Nevertheless, larger disagreements have been observed for some compounds (see Figure 8). An 17 overestimation of TD-GC/MS analysis compared to HPLC analysis of a factor of 8 and 20 on average, 18 respectively for malic acid ad succinic acid, is observed. For malic acid, the external standard used for 19 the estimation of the response factor (glycolic acid) is maybe not appropriate which may explain this 20 discrepancy. As a test, succinic acid and glutaric acid (two other di-acids) have been used as external 21 standard for malic acid quantification with no improvement in the agreement observed. For succinic 22 acid, the authentic standard has been used and such problem cannot explain the discrepancy 23 observed. No interference in the peak region is observed and this cannot neither explain the 24 differences observed. 25 On the whole, comparisons of TD-GC/MS with other techniques deployed during the campaign are 26 satisfactory for both phases with results at least in the same order of magnitude for the measured 27 absolute concentrations, except for some compounds. Therefore, these observations allow us to use 28 TD-GC/MS data both in gas and aerosol phase to study further the behavior of organic carbon at a 29 molecular level at cape Corsica during ChArMEx campaign, keeping however in mind the potential 30 biases revealed during this data comparison exercise.

Molecular characterization of particulate matter 12
A time series of total mass quantified by TD-GC/MS in PM2.5 is presented in Figure 9. This sum has been 13 calculated using the QL/2 (quantification limit/2) value when data were below the limit of This sum is also compared to the organic matter mass concentration in PM2.5 (see Figure 9). OM is 18 calculated using the organic carbon (OC) concentration measured by the SUNSET field instrument with 19 a ratio between OC and OM of 1.9 for Cape Corsica as proposed by Michoud et al. (2017). On average 20 18% of the total OM mass can be explained by the compounds measured by TD-GC/MS for the whole 21 campaign. From 12 to 29 July, oxygenated compounds measured by TD-GC/MS represent more than 22 20% on average of measured OM while they represented less than 10% between July 29 and August and even more probably in particulate phase to explain their formation (Hammes et al., 2019). These 25 missing processes in chemical mechanism included in models might contribute to their inability to 26 reproduce correctly the formation and aging of SOA. If HULIS are considered in this analysis, they 27 represent 59% of the total identified OM mass on average, ranging from 21% of contribution at the 28 beginning of the campaign to more than 80% at the end of the campaign (from 31 July to 3 August). 29

Partitioning of organic carbon between gaseous and particulate phases 30
Many of the compounds identified during the campaign are present in both the gas and aerosol phases. 31 The partitioning coefficient is therefore key to understand processes governing the equilibrium 32 between both phases. For the compounds present in both phases, an experimental partitioning 33 coefficient can be determined following eq. 2 relying on the Pankow equilibrium. Experimental (averaged over the campaign) and theoretical partitioning coefficients obtained for 16 compounds identified in both phases are presented in Table 4 and Figure 13 and are compared to 17 experimental coefficient obtained in a previous field study in Corsica and a chamber study in the 18 EUPHORE simulation chamber (Rossignol et al., 2016). For most of the compounds, experimental 19 partitioning coefficients obtained for the three campaigns are relatively close to each other, with some 20 differences that can however reach up to an order of magnitude (e.g. dimethylglyoxal or acrolein, even 21 two orders of magnitude for glyoxal). These observed differences are small compared to the 22 differences recorded between experimental and theoretical coefficients, with an observed 23 underestimation of theoretical coefficients varying from 1 to 7 orders of magnitude. It is worth noting 24 that the three models used for theoretical coefficients determination are in good agreement. Higher 25 differences between experimental and theoretical coefficients are observed for hydroxyl compounds 26 and carboxylic acids with a shift of the equilibrium toward the particulate phase for experimental 27 partitioning coefficients. It is worth noting that a denuder is used upstream the filter collection to avoid 28 overestimation of particulate organic matter due to adsorption of semi-volatile compounds onto the 29 https://doi.org/10.5194/acp-2020-1051 Preprint. Discussion started: 29 October 2020 c Author(s) 2020. CC BY 4.0 License. filter, therefore excluding potential positive artefact for concentrations of compounds in particulate 1 phase that could have led to overestimation of experimental partitioning coefficients. Furthermore, 2 underestimation of gaseous concentrations for these compounds in such high proportion is unlikely, 3 especially when we look at the comparisons performed for OVOCs with other measurement 4 techniques (see section 3.2.2.1). 5 The differences observed between experimental and theoretical partitioning coefficient may be 6 explained by the high humidity conditions encountered during the campaign (mean RH value of 70%, 7 see Table 3). Indeed, theoretical partitioning coefficient as described by the Pankow equilibrium does 8 not take into account the presence of an aqueous phase or a deliquescent aerosol, while, soluble 9 organic compounds can split between gaseous, aqueous and particulate phase. Concerning the 10 partitioning between the gaseous and aqueous phases, the Henry law's constant and the activity 11 coefficients are considered to calculate the thermodynamic equilibrium. 12 These differences could also be explained by the fact that the equilibrium between both phases is not 13 reached. This could be due to the viscosity of particles. Some studies showed that organic aerosol can . The equilibrium could therefore only concern an external layer of 19 the particle and the gaseous phase (Davies and Wilson, 2015); or on the contrary a semi-solid external 20 layer, caused by the aging of the particle, could prevent the equilibrium to settle between the 21 particulate bulk and the gaseous phase. Even if an analytical artifact cannot be ruled out, for example a fragmentation of oligomers to form 12 back the monomer compounds during the analysis, numerous evidences support the experimental 13 results presented here and suggest that the instantaneous equilibrium being established between 14 gaseous and particulate phases assuming a homogeneous non-viscous particle phase is not fully 15 representative of the real atmosphere. 16  Competing interests. 12 The authors declare that they have no conflict of interest. 13 14 Special issue statement. 15 This article is part of the special issue "CHemistry and AeRosols Mediterranean EXperiments (ChArMEx; 16 ACP/AMT inter-journal SI)". It does not belong to a conference. 17 18 Acknowledgements. 19 This study received financial support from the MISTRALS and ChArMEx programs, ADEME, the French     The phase in which they are detected is color-coded: blue for compounds only detected in the gas 4 phase, red for aerosol phase only and orange for compounds detected in both phases. Each dot 5 represents a single compound and the dot area is proportional to the sum of concentrations if detected 6 in both phases from 0.3 ng m -3 for the smallest dot to 3.9 µg m -3 for the biggest one. Name of some 7 noticeable compounds are also given.