Measurement of alkyl and multifunctional organic nitrates by proton-transfer-reaction mass spectrometry

A commercial PTR-TOF-MS has been optimized in order to allow the measurement of individual organic nitrates in the atmosphere. This has been accomplished by shifting the distribution between different ionizing analytes, H3O/H3O(H2O)n or NO+/NO+2 . The proposed approach has been proven to be appropriate for the online detection of individual alkyl nitrates and functionalized nitrates. It has been shown that hydroxyl and ketonitrates have a high affinity towards NO, leading to the formation of an adduct that allows the easy identification of the organic nitrate (R) from the R–NO ion signal. The recorded sensitivities for both ionization modes correspond to detection limits of tens of ppt min−1 in the case of hydroxyand ketonitrates. Alkyl nitrates exhibit a moderate affinity towards NO ionization leading to detection units of few hundreds of ppt and the highest sensitivity in H3O mode was obtained for the water adducts signals. However, this method exhibits much lower capabilities for the detection of peroxyacetyl nitrates with detection limits in the ppb range.


Introduction
Organic nitrates are an important species of the reactive nitrogen (NO y ) budget in the troposphere.They are formed in NO x rich air by the degradation of hydrocarbons initiated by OH (daytime) and NO 3 (nighttime) radicals.Since organic nitrates have lifetimes of several days or weeks (Perring et al., 2010), they can act as reservoirs for reactive nitrogen by undergoing long-range transport in the free troposphere before decomposing and releasing NO x .They therefore play a key role in the ozone formation as they sequester reactive nitrogen in rich NO x regions and release it in regions where production of ozone may be NO x limited.The significant impact of organic nitrate chemistry on the ozone budget has been confirmed by recent modeling studies (Curci et al., 2009;Horowitz et al., 2007).
In addition, several field studies have shown that both polluted and remote atmospheres contain a large variety of organic nitrates, which significantly affects the NO y budget (for example, 35-40 % as reported by Buhr et al., 1990, and up to 70 % according to a modeling study performed by Madronich and Calvert, 1990).They are not only monofunctional alkyl nitrates and peroxyacetyl nitrates (PANs) but also multifunctional alkyl nitrates (Browne et al., 2013;Fischer et al., 2000;Kastler and Ballschmiter, 1999;Muthuramu et al., 1993;O'Brien et al., 1995).The latter include (i) hydroxynitrates formed by the OH oxidation of alkenes followed by isomerization processes of alkoxy radicals (Arey et al., 2001), (ii) carbonyl nitrates which are produced by the NO 3 oxidation of alkenes (Berndt and Böge, 1995) and also as second-generation oxidation products of hydrocarbons and (iii) dinitrates which are also expected to be secondgeneration oxidation products (Atkinson, 2000;Barnes et al., 1990;Roberts, 1990).
However, measurements of organic nitrates during field campaigns remain rare, thus preventing a precise evaluation of their impact on the NO y budget in a wide variety of environments.The main problems in analyzing organic nitrates in the atmosphere are, first, the great complexity of the mixtures due to the huge number of precursors and formation pathways and, second, the fact that organic nitrates are ex-plosive compounds and only few of them are commercially available.So standards have to be synthesized.These analytical difficulties also affect our capability to study the chemical processes in which organic nitrates are involved during lab experiments (e.g., in simulation chambers).
Table 1 summarizes, without the intent of being exhaustive, some keystone articles concerning the organic nitrates analysis in field and laboratory studies.
Two approaches have been applied for the analysis of complex mixtures of organic nitrates during field and lab experiments.The first is analysis at the molecular scale, which is a powerful approach to elucidate mechanisms but has the drawback to be often limited to a low number of species, followed by functional group analysis, which is very useful to assess global budgets but brings little information about the understanding of processes.The second approach allows us to shortcut the complexity of the organic nitrates chemistry governing their production, transformation and removal processes and assess directly the sum of peroxynitrates AN and alkyl nitrates AN (Day et al., 2002;Perring et al., 2010).The thermal dissociation (TD) properties of different classes of nitrates were used as an analytical tool to probe the global chemistry of RONO 2 , utilizing laser-induced florescence (LIF), cavity ring down spectroscopy (CRDS) (Paul et al., 2009) or cavity attenuated phase shift (CAPS) (Sadanaga et al., 2016) to quantify the NO 2 evolved from the organic nitrates decomposition.In a similar way, the infrared spectroscopy (IR) was used to monitor the time-dependent loss of organic nitrates and measure the rate of homolytic O−N bond cleavage (Francisco and Krylowski, 2005) by employing specific absorption bands in IR (1638 cm −1 ).Although notoriously powerful analytical techniques, the LIF, CRDS, CAPS and IR exhibit poor capabilities in the individual quantification of organic nitrates mixtures due to their intrinsic conceptual operation mode.
Although the chromatographic separation represents an effective analytical tool, the organic nitrate identification relies mainly on the retention times, while the dedicated detectors are only able to confirm, in the best case scenario, the presence of the nitrate functional group in the molecule.Illustrative examples are given by the studies of Luxenhofer et al. (1994) and Kastler and Ballschmiter (1999), who succeed to analyze complex mixtures of alkyl and multifunctional organic nitrates by combining separation with liquid and gas chromatography and detection using the intense mass-tocharge (m/z) 46 fragment ion that corresponds to NO + 2 .The same study highlights possible interferences with dinitrophenols, nitro-and dinitrocresols, pentachloronitrobenzene and to a smaller extent nitrophenols.
Besides the poor temporal resolution, another major drawback of this method is represented by the recovery factor decline with longer times in the chromatographic columns.According to Roberts et al. (2002), the sensitivity of this method for PANs is characterized by a diminishing response factor proportional to the compounds retention time through the column.Measurement of functionalized (oxygenated) nitrates appears to be a greater challenge as the lower vapor pressures and stronger surface interactions of these molecules make sampling and chromatographic techniques less appropriate.Additionally, the detection of functionalized nitrates by electron impact mass spectrometry has proven to be difficult, mainly due to the instability and thus the fragmentation of the molecular ion formed (Mills et al., 2016;Roberts, 1990;Rollins et al., 2010).
Lately, the newly developed capabilities in atmospheric pressure/chemical ionization mass spectrometry (AP-CIMS) (Huey, 2007;Perraud et al., 2010;Slusher et al., 2004;Teng et al., 2015) prove to be potentially powerful tools in organic nitrates analysis.Several types of CIMS have been highlighted by the literature, and the technique is currently one of the most promising analytic tools.
The AP-CIMS uses methanol as a proton source in order to generate RH + peaks of the parent ions.Therefore the nitrogen-containing ions are characterized by even m/z ratios, while the analytes containing only C, H and O appear at odd m/z.Several PANs were identified as gas-phase products of the α-pinene + NO 3 reaction with this technique (Perraud et al., 2010).The protonated molecular ions were identified and the most intense fragments in the MS/MS scan corresponds to losses of NO 2 , HNO 3 and to a smaller extent HOONO 2 .Hydroxynitrates and ketonitrates have also been detected with this technique.For hydroxynitrates, RH + peaks as well as fragments corresponding to the losses of H 2 O and NO 2 + H 2 O in the MS/MS mode were detected (Perraud et al., 2010;Schoon et al., 2007;Tuazon et al., 1999).
The thermal dissociation-chemical ionization mass spectrometry (TD-CIMS) technique has been used for measurement of PANs and other multifunctional organic nitrates by the means of I − reaction (Lee et al., 2014;Slusher et al., 2004;Xiong et al., 2015).The obtained carboxylate ion is unique for each parent species, and the only significant interference that has been identified is at ppb levels of NO  (Slusher et al., 2004).The CF 3 O − was equally successfully tested as ionizing source in a CIMS approach in order to identify hydroxynitrates formed during the OH oxidation of alkenes in the presence of O 2 and NO (Bates et al., 2014;Teng et al., 2015).The quantification of the nitrates formed was assured by a complementary TD-LIF technique after subsequent GC separation.Other polyfunctional organic nitrates were scarcely detected using this technique.
The proton-transfer-reaction mass spectrometry (PTR-MS) can be positioned as a subset of CI.Its use in atmospheric research has expanded rapidly these last years but few studies have tested this technique for the detection of organic nitrates (Aoki et al., 2007;Hansel and Wisthaler, 2000;Inomata et al., 2013).
The recent study of Müller et al. (2012) Hansel and Wisthaler (2000) have measured several PANs (PAN, MPAN and PPN) with PTR-MS and have reported a low detection limit (70 pptv, 15s integration time) and an overall accuracy of 15 %.It is worth noting that the reported analytical sensitivities (15-25 counts ppb −1 ) have not been calculated for the molecular ion signal but for typical molecular fragments in the case of the alkyl nitrates or signals of speculative subsequent reaction products of the protonated molecular ions for PANs.These results are illustrative of the organic nitrates overall affinity towards fragmentation due to chemical ionization in a low pressure drift tube under high voltage.To our knowledge, polyfunctional organic nitrates have never been detected using this technique.
So, it appears useful to better explore the possibility of detecting organic nitrates in real time with PTR-MS.In the present work, the detection of alkyl and multifunctional organic nitrates (PANs, ketonitrates, hydroxynitrates) by this technique was investigated.For this purpose, different organic nitrates were synthesized and mixtures at the ppb-ppm level were generated in a smog chamber.A commercial PTR-TOF-MS instrument was used but the operating mode as well as the chemical ionization reagent have been modified and optimized for each type of organic nitrate in order to gain sensitivity in the detection and to reduce fragmentation of ions.Mass spectra have been carefully interpreted and detection limits have also been determined.

Instrumentation
A commercially available, high-mass-resolution PTR-TOF-MS instrument (Kore Technology "Series II" High Performance PTR-TOF-MS) was used in the present study.The mass resolution of this instrument is > 5000 M/δM.Mass calibration of the recorded spectra was systematically performed.The procedure allowed us to accurately assign the mass spectra peaks.
The PTR-MS is equipped with the newly designed radio frequency (RF) ion funnel which is used to focus ions in the drift tube and hence to increase the detection sensitivity (Barber et al., 2012).This ion funnel uses a series of electrodes with progressively reducing aperture sizes.In addition to the standard direct current (DC) electrical field, an alternative current (AC) electric field is provided at a radio frequency that creates a strongly repulsive effective potential near the surface of the electrodes, which combined with the reducing aperture sizes serves to focus the ions radially.This system has been shown to increase the sensitivity by 1 or 2 orders of magnitude (Barber et al., 2012).It is worth noting that operating in the RF mode modifies the dynamic range over which the drift tube is operated and the contribution of the radio frequency to the global effective E/N ratio remains difficult to estimate (Barber et al., 2012).
The PTR-MS sampling line was designed in order to assure the highest transfer efficiency of the compounds.This line is made of a 1.5 m long Silcosteel ® coated stainless steel tubing (2.1 mm inner diameter), which has been shown to be appropriated for the transfer of low-level polar organic compounds (Smith, 2003).This line was heated at 40 • C as a compromise between the thermolability of the organic nitrates and the condensation processes.With a sampling flow rate of ∼ 3 cm 3 s −1 , the residence time in the line has been estimated to be lower than 2 s.The PTR-MS inlet system is equally temperature controlled and made with Silcosteel ® coated stainless steel tubing.Running the system at ambient temperature-induced PTRMS signal instability, while subsequent heating of the sampling line revealed desorption processes (peaks of analytes following the heating steps).

Experimental strategy
In PTR-MS, proton transfer reactions with hydronium ions (H 3 O + ) are typically used to ionize compounds having a proton affinity (PA) higher than that of water (691 kJ mol −1 ; Hunter and Lias, 1998): In this case, the proton transfer reaction is exothermic and proceeds at a reaction rate close to the collision rate (10 −9 cm 3 molecule −1 s −1 ; Flocke et al., 1991).In order to avoid a massive fragmentation of the analyte and consequently a decrease of the intensity and the specificity of the analytical signal, the energy transferred during the reaction should be lower than the bond energies of the compound of interest.
Only few available data can be found concerning the PA of organic nitrates, which were calculated using ab initio quantum mechanical methods exclusively.The PA of organic nitrates has been estimated to be 740 kJ mol −1 for methyl nitrate (Lee and Rice, 1992), 753 kJ mol −1 for ethyl nitrate (Kriemler and Buttrill, 1970), 748 kJ mol −1 for methyl peroxynitrate (Ravelo and Francisco, 2007) and 759-773 kJ mol −1 for peroxyacetyl nitrate (Tureček, 2000).These studies show that the PA of organic nitrates is higher than that of water, so that proton transfer reaction may occur.Furthermore, the functionalized nitrates concerned by the current study are equally expected to easily protonate as it was shown that the presence of an additional oxygen atom in the molecule enhances the overall PA (Ravelo and Francisco, 2007).
It is well known that other chemical processes can also occur in the PTR reactor (Blake et al., 2009;de Gouw and Warneke, 2007).The H 3 O + and R + can cluster with water molecules in the sampled air.The water cluster ions formation in the PTR cell is generally considered as problematic since their presence complicates the mass spectra treatment.Typically, their formation is limited by high kinetic energy of the ions into the drift tube at E/N ratios superior to 100 Td.However, the proton transfer reaction of H 3 O + (H 2 O) n clusters is more selective than the H 3 O + due to a higher PA (808 kJ mol −1 ; Goebbert and Wentold, 2004).This property may be valuable in our case as it allows a softer ionization of analytes.Recently Jacobs et al. (2014) have used H 3 O + (H 2 O) n ions (distribution centered around n = 4) in PTR-CIMS technique with the objective of monitoring isoprene-derived hydroxynitrates and their oxidation products.In order to evaluate the efficiency of the protonation mode by H 3 O + (H 2 O) n for the detection of other organic nitrates, a wide range of E/N ratios will be tested.
The NO + chemical ionization has also been tested for the detection of organic nitrates during our experiments.It has been shown to be a complementary, sensitive and reliable method for the detection of some organic compounds such as ketones, aldehydes and alkenes by CIMS and SIFT-MS techniques (Knighton et al., 2009;Mochalski et al., 2014;Perraud et al., 2010;Wang et al., 2004).NO + ions are produced from the ionization of dry air within the hollow cathode discharge ion source.This process produces large amounts of NO + with only minor amounts of NO + 2 , O + 2 and H 3 O + (H 2 O) n impurities (Knighton et al., 2009).The intensities of O + 2 and H 3 O + (H 2 O) n are quite low -less than 1 % of the primary NO + signal.In addition, it has been shown that the intensity of the NO + 2 impurity is controllable and dependant on how the hollow cathode ion source is operated (Knighton et al., 2009).The ion source extraction voltage and the hollow cathode discharge current are the two most important parameters.
Previous studies interrelated the PA of water to the NO + chemical ionization binding energies (BEs), creating an absolute NO + affinity scale (Cacace et al., 1997).The study used the H 2 O-NO + BE as a reference anchor between the two conjugated parameters and concerned various classes of ligands (alkyl halides, alkyl nitrates, alcohols, nitro-alkanes, nitriles, aldehydes, ketones, aromatic and heterocyclic compounds).With the exception of the aromatic compounds, for which the NO + σ -type complex structures are replaced by the π complexes, the overall data exhibit a highly correlated linear dependence (BE NO + = 0.367 PA − 174.5 kJ mol −1 ).When compared with protonation energies, the BEs charac-terizing the NO + chemical ionization are significantly lower (Cacace et al., 1997).
This characteristic offers the opportunity to generate soft ionizations with little fragmentation of the product ions, which is of high interest for the present study.Reactions which may occur are As a function of the chemical affinity, NO + chemical ionization may follow several pathways.The charge transfer Reaction (2a) seems to be the most common process for compounds having ionization energies slightly lower or close to that of NO (IE = 9.26 eV) (e.g., isoprene (Karl et al., 2012) butadiene, benzene (Knighton et al., 2009), monoterpenes, terpenoids (Amadei and Ross, 2011;Rimetz-Planchon et al., 2011) and phenols (Wang et al., 2004)).This hypothesis is confirmed by Smith et al. (2003), who reported for a series of ketones (C 3 -C 11 ) that the yields of parent radical cation increase as the IE decrease.
The adduct formation pathway (Reaction 2b) has been previously reported for the detection of C 3 -C 4 alkanes, alkenes and terpenes (Diskin et al., 2002;Španěl and Smith, 1998), ketones (Smith et al., 2003;Wang et al., 2004) and flavoring esters (Iachetta et al., 2010) by selected ion flow tube mass spectrometry.The speculated mechanism which has been proposed to explain the formation of the adduct involves the formation of an excited intermediate complex (R • NO + ) * , which is then stabilized by collision with the carrier gas (Smith et al., 2003).Evidently the process is enhanced at upper PTR reactor pressure involving higher collision rates.
The O + 2 ionization can generate artifacts, particularly when the PTR-MS is operated in dry air mode, as it proceeds via a dissociative charge transfer reaction, producing parent cations M + as well as fragment ions.As a consequence, this process is of little interest for chemical ionization of large molecules.
To summarize, several parameters will be tested in order to establish the optimal conditions for the detection of each class of organic nitrates.
A carefully thought-out procedure was performed in order to establish the optimal measurement conditions for each class of organic nitrates: -First, the instrument has been used in a dual mode, employing alternatively H 3 O + and NO + as a reagent ion www.atmos-meas-tech.net/10/1445/2017/by shifting the glow discharge (GD) source gas from water vapor to dry air.Dry air was supplied to the hollow cathode ion source using the alternative GD ionization line.A third operation mode which consists of suppressing the GD source gas once the plasma is established has also been tested.This last operational mode implies hence the retro-diffusion of the sampling line dry air into the GD cavity and will be further referred to as the retro-diffusion mode (RDiff).
-The ion intensity and distribution are sensitive to the extraction voltage used to inject the reagent ions into the drift tube and to the hollow cathode discharge current.The PTR-MS instrument was thus operated in a wide range of E/N (electric field to number density of the gas) ratios (30-180 Td), hence shifting the distribution between the different ionizing analytes in the protonation mode ( or into the dry air mode (NO + / NO + 2 ).-The sampling line and the inlet system along with the drift tube was slightly heated (40 • C) and the reactor pressure ranged from 0.7 to 1.6 mbar.The sampling flow rate (2.9 to 3.5 cm 3 s −1 ) affects through the mass of the sampled analyte or through the third body collision processes the instrument response.An enhanced influence of this parameter may be noticed while operating in the RF mode.
-The influence of radio frequency ion funnel mode on the detection of the various organic nitrates has been tested by performing experiments with the RF mode on and off.
Ideally, little or no fragmentation occurs in the case of a soft ionization process.The large differences in ionization energy of the colliding species enhance the fragmentation of the organic molecule.Therefore the ionizing gas species should correspond, especially for labile molecules, to the analyte of interest in order to achieve the high yield ionizations with low excess energy.
In addition, the soft ionization processes demand the usage of uncommon settings of the PTR-MS device.The various effects of the altered PTR-MS tuning are termed in detail elsewhere (Hewitt et al., 2003;Knighton et al., 2009).Supplementary information relating the ion source discharge current and the extraction voltage influence is discussed in the Sect.3.
All reported signal intensities took systematically into account the background values, representing in most of the cases negligible low values of a few counts per minute.Background measurements are achieved before every single experiment by sampling the dry synthetic air of the reaction chamber.

Normalizing analytical signal for H 3 O + and NO + ionization
In this study, calibrated response factors of PTR-MS are determined for several types of organic nitrates in both H 3 O + and NO + ionization modes.For that purpose, known amounts of organic nitrates are introduced in LISA simulation chamber (see following section).To quantify the PTR-TOF-MS response factors (S; ncpm ppbv −1 ), the ion signal (I R + ; ncpm) is divided by the known concentration of organic nitrate (ppbv −1 ), as measured by the FTIR in situ technique.
Corrections of the product ion signal (I R + ) due to changes of the operational conditions have to be taken into account, as shown by previous studies (Knighton et al., 2009): The intensity of the product ion raw signal (i R + ) expressed in counts per minute is adjusted to the variabilities of the ionizing analyte (i IA + ) normalized to its averaged values ( i IA + ), (f as it has been shown that both signals (i R + and i IA + ) are related.The drift tube temperature (T ) and pressure (P ) are also included in this expression to account for the small changes in the reaction time and gas number density that can occur during measurements.In our experiments, the temperature was very stable (±1 K) but the PTR pressure was affected by the pressure in the simulation chamber, which usually slowly decreases because of the sampling from the constant volume of the rigid chamber.
Due to the high abundance of the ionizing species, the direct measurement of H 3 O + and NO + reagents was not possible as the ion counting signals (m/z 19 and 30) were regularly saturated.The natural isotopic abundance properties of 17 O (0.038 % of O), 18 O (0.200 % of O) and 15 N (0.368 % of N) are used to overcome this drawback and evaluate the primary H 3 O + and NO + ion intensity using the m/z 21 (H 18 3 O) and 31 respectively ( 15 NO + and N 17 O + ) (De Laeter et al., 2003).The I H 3 O + = I m/z21 ×500 and I NO + = I m/z31 ×247 formulas were applied as a result of isotopic abundance probability calculations.

LISA atmospheric simulation chamber
Gaseous mixtures of organic nitrates at the ppb-ppm level were generated in the simulation chamber at LISA.This chamber comprises a Pyrex reactor of 977 L equipped with a multiple reflection optical system interfaced to a FTIR spectrometer (Vertex 80 from Bruker).Details of this smog chamber are given elsewhere (Doussin et al., 2003).
All experiments were conducted in the dark at 298 ± 2 K and atmospheric pressure.Mixtures of organic nitrates were generated in synthetic air (N 2 80 % + O 2 20 %) by introducing a known amount of the organic nitrate into the chamber and cross-monitored by long-path in situ FTIR and PTR-MS techniques.Concentrations of organic nitrates were checked from their infrared spectral absorption bands.Integrated band intensities (IBIs; expressed in cm molecule −1 and calculated in decimal logarithm) and the spectral integration areas used to quantify PAN-like compounds and other nitrates are 9.50 × 10 −18 for PANs (765-812 cm −1 ), 8.30 × 10 −18 for alkyl nitrates (820-900 cm −1 ), 1.17 × 10 −17 for hydroxynitrates (816-882 cm −1 ) and 1.08 × 10 −17 for ketonitrates (820-870 cm −1 ).The IBIs are given by Allen et al. (2005) and Spittler (2001) for the PANs and the hydroxynitrates, respectively, whereas in the current study for the alkyl nitrates and ketonitrates the IBIs were calculated from repetitive injections of known amounts of analyte.It is worth noting the weak disparity of cross sections concerning the characteristic −ONO 2 absorption band.Reference gas-phase FTIR absorption spectra of the synthesized organic nitrates (KnC3, KnC5, 1OH3C3) are plotted in Fig. S1 in the Supplement and compared with the recorded spectra of a commercially available alkyl nitrate (AlkC3).
Before and after every single experiment, a cleaning procedure was applied in order to avoid memory effects from an experiment to the next one.It consists of vacuum cleanup down to 10 −2 mbar for at least 10 h and UV irradiation (340 and 360 nm fluorescent tubes) in order to heat the reactor and improve desorption of semi-and low-volatile compounds from the walls.
The alkyl nitrates considered in the present study (n-propyl nitrate -97 % Janssen Chimica; AlkC3 and isobutyl nitrate -96 % Sigma Aldrich; AlkiC4) were used as commercially available, without further purification.
Ketonitrates were synthesized using Kames' method, (Kames et al., 1993): a liquid/gas-phase reaction in which the corresponding hydroxy ketone is reacted with NO 3 radicals released from the dissociation of N 2 O 5 , at ice bath temperature, under dry condition.The carbonyl nitrates and nitric acid were separated by multiple vacuum headspace procedure.The carbonyl nitrates' structure and purity were verified by FTIR.In addition, it was checked by GC-MS that the organic reactant used for the synthesis, here the hydroxy ketone, was fully consumed, preventing any interference in IR and mass spectra.The presence of HNO 3 is regularly noticed since it is injected as byproduct of the carbonyl nitrates synthesis.1-hydroxy-2-propanone (95 % Alfa Aesar) and 3-hydroxy-3-methyl-2-butanone (97 % Sigma Aldrich) were used for the synthesis of 1-nitroxy-2-propanone and 3nitroxy-2-propanone respectively.
The hydroxynitrate was synthesized starting from the commercially available 3-bromo-1-propanol (97 % Sigma Aldrich).Its conversion to the analog iodide was performed by a Finkelstein reaction with sodium iodide (Baughman et al., 2004) and the subsequent mild conversion with AgNO 3 (Castedo et al., 1992) leading to the formation of 1-hydroxy-3-nitroxy-propane (1OH3C3).The total consumption of the bromo-alcohol was checked by GC-MS.
The PAN-type compounds were generated in the simulation chamber from the gas-phase oxidation of corresponding aldehydes by NO 3 radicals (Hanst, 1971).In our experiments, PAN was formed from the oxidation of acetaldehyde (> 99.5 % Sigma Aldrich).This reaction has been shown to produce mainly PAN with a yield close to 70 % (Doussin et al., 2003).
The presence of impurities as synthesis byproducts may add supplementary signals in the PTRMS mass spectra.These signals are, however, systematically at lower masses than those in the main ionization processes of the organic nitrates and therefore will not interfere in the main frame of the discussion.

Influence of the operating conditions on the ionizing analytes signals
As already termed, intensities and mixing ratios of the ionizing species are dependent on the E/N ratio.For a standard drift tube, E/N is a well-defined quantity but, when the instrument is run in RF mode, the influence of the additional AC electric field on the E/N ratio has to be taken into account and this is not obvious.Barber et al. (2012) have attempted to empirically estimate an effective E/N by running the instrument with the RF mode on and off and by seeking operating conditions for the RF on mode that match the performance obtained when the RF mode is off.
The criterion used to estimate the performance is the ratio During our experiments, the additional AC electric field was fixed.Thus, variations of E/N ratio result only from variations of the DC component of the drift tube.
Typically recorded distributions over a relevant range of E/N are illustrated in Fig. 1 for both ionization modes, with and without the above described RF mode.When the RF mode is not employed, the water cluster ion distribution obtained in this study (Fig. 1) is in reasonable agreement with those calculated and measured by de Gouw and Warneke (2007), asserting that (i) the H 3 O + (H 2 O) 2 cluster is abundant for E/N ratios < 60 Td, (ii) the H 3 O + (H 2 O) signal is maximum around 80 Td and (iii) the H 3 O + signal is predominant for values larger than 100 Td.In the RF mode, it has been observed that the signal of H 3 O + ion is much higher than those of clusters for the entire range of E/N ratios.The H 3 O + (H 2 O) is 1 order of magnitude lower than the H 3 O + and 1 order of magnitude more abundant than the H 3 O + (H 2 O) 2 .As the distributions of ions obtained with and without RF mode are quite different, it was not possible to estimate an effective E/N as proposed by Barber et al. (2012).So E/N ratios have been calculated without taking into account the RF contribution.In the case the RF mode is on, E/N ratios are indicated with * (E/N * ).
The NO + ionization mode has been found more promising in the RF mode, since the abundance of the ionizing species is largely superior.Furthermore, the NO + / NO + 2 ratio exhibits the highest values around 40 Td in the RF mode, increasing the probability to form the analyte-NO + adduct.Therefore, all further results presented in the current work in the NO + mode have been recorded with the RF mode and reported E/N * was calculated without taking into account the AC radio frequency contribution.

H 3 O + ionization mode
A series of tests has been conducted in order to seek the optimum operating conditions for the measurement of alkyl nitrates with H 3 O + ionization mode.In particular, the E/N ratio was spanned in the 50-140 Td range.The intensity of the main signals obtained for n-propyl nitrate (AlkC3) with the RF mode off has been plotted as a function of E/N in Fig. 2. The recorded mass spectra of AlkC3 are character- ized by a high degree of fragmentation, even for the lowest E/N (50 Td).The AlkC3 mass spectra recorded at the lowest extent of fragmentation of the protonated molecular ion is illustrated by Fig. 3a.The same result has been observed for AlkiC4 as depicted by the Fig. S2 in the Supplement.Consequently, the sensitivity of the molecular ions formed is low (see Table 2).The protonated alkyl nitrates are expected to adopt the ion-dipole complex conformation (ROH • NO + 2 ; R1a) and not a covalently bound RONO 2 H + one (Cacace and de Petris, 2000).
The bound energy of these complexes was calculated by ab initio methods to be around 82 kJ mol −1 in the case of methyl nitrate (Lee and Rice, 1992).
The only other study (Aoki et al., 2007) describing a tentative of alkyl nitrate detection with PTR-MS used E/N ratio spanning between 96 and 147 Td.The low BE characterizing the formed complex enables its decomposition in ROH and NO + 2 by collision-induced dissociations.The recorded signals of NO + 2 (m/z 46) and RO + (m/z 59 for AlkC3 and 73 for AlkiC4) are probably formed by the mechanisms: This statement is supported by the decrease of the protonated signal in conjunction with the increase of the NO + 2 with increasing E/N .
The presence of the m/z 43 (AlkC3) and 57 (AlkiC4) signal was clearly identified as the alkyl fragment (R + ) of ni- trate by the means of high-resolution mass spectrometer allowing us to differentiate between oxygen containing analytes and alkyl fragments.At higher abundances of the water clusters, expected to occur in our study due to the usage of low E/N ratios, the R5 mechanism is considered more likely, as proposed by Aoki et al. (2007).The formation of R-OH • H + ion can also be explained by Reaction (R6) as proposed by Spanel and Smith (1997).This ion is then expected to form the alkyl fragment (R + ) by Reaction (R7): A third option is to consider the direct formation of the alkyl fragment from the collision-induced dissociation of the protonated alkyl nitrates formed in Reaction (R1a) into R + and HNO 3 .Further collisions of the alkyl fragment (R + = C 3 H + 7 ) might generate the loss of two hydrogens and explain the weak signals at m/z 41 in the case of AlkC3.
It should be noticed that the analog process for the m/z 55 in the case of AlkiC4 cannot be observed since it falls near a relatively abundant water cluster signal.
Results obtained with the RF mode are equally shown in Fig. 2 for the AlkC3 for comparison.The ionization pattern of the analyte is slightly different from the one obtained without RF mode and enables, adjacent to the formation of the protonated alkyl nitrate, the identification of other specific signals like the adduct AlkC3 • H 3 O + formation.Characterized by a high degree of fragmentation, the spectra contain the protonated ion-dipole complex ROH • NO + 2 at m/z = 106 (Reaction R1a) with the highest sensitivity above 60 Td.The E/N ratio is calculated without taking into account the RF contribution.The m/z 124 signal can be assigned to the AlkC3-H 3 O + adduct formation.An analog process is described in the literature for long alkanes (> C6) providing M • H 3 O + signals (Španěl and Smith, 1998), in contrast with the alkenes which are readily protonated due to their higher PA.
The alkyl (R + ) intermediary parent ion, described in R5 and R6 (R-OH • H + ), is equally recorded, corresponding in the case of AlkC3 to the m/z 61 signal.The increase of the E/N ratio over 60 Td contributes to a higher degree of decay of this intermediate ion in the particular conditions of the RF mode.The RO + ion at m/z 59 is equally present in the case of AlkC3, confirming the existence of the Reaction (R4) mechanism leading to HONO formation.
To summarize, the use of H 3 O + ionization is not really suitable for the detection of alkyl nitrates as it leads to high fragmentation, even at low E/N ratios, hence generating intense signals of common organic analytes, unsuitable for a reliable identification of these compounds.In addition, if considering the more characteristic signals of the protonated ion-dipole complex ROH• NO + 2 , the H 3 O + ionization mode exhibits a poor detection limit: 5 ppb min −1 for both AlkC3 (m/z 106) and AlkiC4 (m/z 120).This detection limit was estimated for a mean signal to noise ratio of 3 at the chosen m/z.
In Table 2, the characteristic signals of each organic nitrate studied here as well as their detection limits are listed.

NO + ionization mode
The performances of the NO + ionization were also evaluated.The most promising results, enabling the identification of characteristic signals of the studied alkyl nitrates, were obtained in the RF mode.The scan of a large range of E/N * ratios confirmed that the highest sensitivity towards NO + -induced ionization and identification is obtained around 40 Td as illustrated in Fig. 2 for the AlkC3 case.As mentioned previously, the calculated E/N ratio does not take into account the RF contribution to the ion funnel field.
The recorded mass spectra of the analyte of interest in terms of relative peak intensities as a function of the m/z are represented by the thick black lines in Fig. 3b.It should be kept in mind that the NO + and NO + 2 signals illustrated above are the sum of the corresponding nitrate fragments as well as the ions formed into the dry air plasma GD.The presence of O + 2 (m/z = 32) and its interfering isotopic abundance signals at m/z = 33 and 34 has been observed and could result in interfering signals.
For both alkyl nitrates, the adduct formation (Reaction R2b) appears to be the main ionization mechanism under these given conditions, leading to intense peaks at m/z = 135 (10 5+ 30) and m/z = 149 (119 + 30) for AlkC3 and AlkiC4 respectively (see Figs. 3b and S2b).The hydride abstraction (R(-H)) + was also detected as a minor pathway (Reaction 2c) at m/z = 104 and m/z = 118 for AlkC3 and AlkiC4 respectively.Despite a weak intensity, this signal can be used as an interrelated identification signal for alkyl nitrates.The fraction of the abstraction channel is at no time higher than 10 % of the intensity of the adduct formation channel.
An intense peak corresponding to the alkyl fragment (R + ) was observed in spectra at m/z 43 (AlkC3) and 57 (AlkiC4).The alkyl fragment signal is followed by a downward series of signals at m/z 41 and 39 in the case of AlkC3 and at m/z 55 and 53 for the AlkiC4.Considering the collisions of the alkyl fragment mechanism generating the loss of two hydrogens (Reaction R8), as proposed by Aoki et al. (2007) for the protonation mode, could explain the recorded signals.
In conclusion, the NO + ionization mode appears to be more suitable for the detection of alkyl nitrates than the H 3 O + ionization mode as it produces two characteristic signals, corresponding to the adduct formation (M • NO + ) and the hydride abstraction (M(-H) + ).The detection limit for the M • NO + signal was estimated to be 205 ppt min −1 for AlkC3 (m/z 135) and 126 ppt min −1 in the case of AlkiC4 (m/z 149) (see Table 2).Although superior to the protonation mode, the NO + ionization exhibits a poor sensitivity.For measurements in real atmosphere, where alkyl nitrates mixing ratios are usually in the ppt level, accumulations for several hours will be necessary.

The H 3 O + ionization
As for alkyl nitrates, the measurement of hydroxynitrates by PTR-MS has been tested with H 3 O + ionization mode, with and without RF mode.For this purpose, 1hydroxy-3-nitroxy-propane (1OH3C3) was synthesized and used as a reference.In the classical mode (RF off), the main  loss of H 2 O after protonation of the hydroxyl nitrate, as suggested by several studies for alcohols (Schoon et al., 2007;Smith et al., 2012;Tuazon et al., 1999).
When the RF mode is on, the same signals (43,104,122,139) were observed but with different relative abundance.The influence of E/N on the intensity of these signals has been equally plotted in Fig. 4. Above 40 Td, the signal characterizing the water adduct formation (m/z = 139) is more abundant in the RF mode with a maximum around 45Td.
The water clusters are probably abundant in the PTR reactor due to the low E/N ratios and dissociate only into the RF region after the ionization occurs.
The mass spectrum illustrated by Fig. 5a was recorded in absence of the RF funnel and is dominated by fragments like m/z 59, 60, 73, while the specific signal of the protonated molecule m/z 122 is present close to the noise limit.
The above described water adduct formation (m/z = 139) is competed by a H 3 O + adduct (m/z = 140), equally present into the spectra.The ionization is probably induced by the expected high levels of water cluster, following an analogous process with the one described in R5.This involves the release of HNO 3 and arise the signal at m/z 77 (C 3 H 6 OH)-OH • H + and subsequently at m/z 59 with the loss of supplementary water (Reaction R7).Further collisions of the abovementioned ions may well explain the signals at m/z 75 and 57, generated by the loss of two hydrogens (Reaction R8).
In RF mode, the mass spectrum obtained for 1OH3C3 at E/N * = 45 Td (corresponding to the operational condition for which the signal at m/z 139 is the most intense) was illustrated in Figs. 5 and S3.It is worth noticing that all the other competitive ionization processes are strongly diminished in comparison to the spectrum obtained in classical mode.Only the specific signal at m/z = 104 exhibits a stronger signature compared to the case above.The fragmentation pattern is strongly diminished in this case.
To conclude, it has been observed that protonation of the hydroxynitrate is a minor process in comparison to fragmentation and to adduct formation.The use of the RF mode significantly reduces the fragmentation for the benefit of the M • H 2 O + adduct formation.The lowest DL (80 ppb min −1 ) was obtained at 45 Td * in the RF mode for the signal corresponding to the water adduct formation at m/z 139.The same signal in absence of the RF effect is at least twice weaker in terms of sensitivity.Sensitivities of other specific signals are equally proposed for comparison in Table 2.

The NO + ionization
The detection of hydroxynitrates in NO + ionization mode has been evaluated for the first time by varying the E/N ratio (39-44 Td) and by studying the influence of the RF mode.As stated before the highest sensitivity towards the NO + adduct formation is obtained in the RF mode, most likely due to the higher abundance of the ionizing species, as already seen in Fig. 1.
www.atmos-meas-tech.net/10/1445/2017/Atmos.Meas.Tech., 10, 1445-1463, 2017 Results obtained with RF mode are shown in Fig. 4. The main signals that have been observed are 151 and 167, which have been attributed to M • NO + and M • NO + 2 adduct formation, and 43 (C 2 H 3 O + ), which is characteristic of fragmentation.Although weak, the hydride abstraction (Reaction R2c) leading to a signal at m/z 120 has also been observed in the recorded spectra.This process has already been observed in previous studies for various saturated and unsaturated alcohols (C 1-10 ) and phenol (Karl et al., 2012;Schoon et al., 2007;Spanel and Smith, 1997) using SIFT and PTR-MS techniques.The hydride ion transfer of these compounds generates the corresponding carboxy ion (and HNO), while the hydroxide ion transfer gives the corresponding hydrocarbon ion (and HNO 2 ).Assuming the corresponding ionization of hydroxynitrate in the case of the 1OH3C3 would involve the formation of the (O 2 NO-C 3 H 6 )-O + at m/z 120 and (O 2 NO-C 3 H 6 ) + at m/z 104, but their spectral signature is marginal into the spectra.
As expected, the intensity of the signal 43 increases with increasing E/N ratio.The opposite tendency was observed for adducts.The mass spectrum obtained at E/N * = 41, which corresponds to the highest sensitivity for the m/z 151 signal, is shown in Fig. 5b.
Besides the abovementioned characteristic signals, other specific mechanisms could be associated to the spectral signature of 1OH3C3.The mechanism is reviewed and suggested by Harrison (1999) for the particular case of NO + ionization of primary alcohols.In this study, an additional product (R(-2H) + NO + ) has been observed which would correspond in our case to an m/z = 149, hypothetically formed by the oxidation of the alcohol to the corresponding aldehyde and its subsequent ionization.Although relatively weak, the signal is systematically recorded in the mass spectra of the hydroxynitrate.The 45 and 67 ions are equally present in mass spectra and can be attributed to fragmentation.
The detection limit obtained in NO + ionization mode at E/N ratio of 41 Td * for the M • NO + signal at m/z 151 was quite low (37 ppt min −1 ), showing that this ionization mode is more suitable than H 3 O + mode for the measurement of hydroxynitrates.As shown in Table 2, the hydrogen abstraction signal (m/z 120) is at least 1 order of magnitude less sensitive.
In order to identify the optimal conditions for the detection of ketonitrates in H 3 O + ionization mode, tests were performed by varying E/N ratios into a large domain, from 45 to 170 Td (Fig. 6).With the data available in literature we expect that for polar compounds the ionization process may follow two similar pathways: by proton transfer (Reaction R9a) and by ligandswitching reaction (Reaction R9b) (de Gouw et al., 2003).
The bound energy of the cluster ions formed in the reaction above (Reaction R9b) is weaker than the water cluster bond.
In the given conditions, there is a high probability that the drift tube dissociative effect can split the formed cluster ions and lead to the formation of MH + and MH + • (H 2 O).Since the cluster ion distribution may be governed by the water vapor drift tube loads, the detection efficiency can be reliant on humidity in this particular case (de Gouw et al., 2003).
Redistribution processes among the various precursor ions formed into the glow discharge can equally occur, leading to the formation of auxiliary hydrated ions such as NO + (H 2 O) n , able to produce ligand-switching reactions similar to the ones presented above (Reaction R9b).
For intermediate E/N ratios, where H 3 O + (H 2 O) water cluster is the key analyte, we notice that the signal corresponding to M kn • H + (m/z = 120) is maximum.In addition, two other signals corresponding to M kn • NO + (m/z = 149) and M kn • NO + 2 (m/z = 165) adduct formation have also been observed and rise up to a maximum for these intermediate E/N ratios.These adducts can be explained by the formation of NO + and NO + 2 analytes in the GD, which increases with the increasing E/N ratios.
For high E/N ratios, the m/z 43 (C 2 H 3 O + ) signal, which is a common fragment of organic compounds, strongly increases.This reveals that mainly fragmentation occurs, making this region unsuitable for the detection of ketonitrates.
From these results, the intermediate E/N ratios (70-80 Td) appear to be the most suitable for the detection of ketonitrates in H 3 O + ionization mode.Mass spectra obtained at E/N = 75 Td, which corresponds to the highest sensitivity of the M kn • H + signal, are shown in Fig. 7 for KnC3 and in Fig. S4 for KnC5 respectively.
For KnC3, the most intense signal corresponds to M kn • H + (m/z = 120).As discussed above, additional processes occur which are responsible for other signals: M kn • NO + adduct formation at m/z 149 and fragmentation at m/z 43 (C 2 H 3 O + ).These characteristic signals are tailed by their corresponding isotopic abundance signals mainly due to 13 C isotope at m/z = 121 and 150.
The already discussed Reactions (R5) and (R6) could explain the intense signal of m/z 75 since the resulting (C 3 H 5 O)-OH • H + ion seems the accurate match of this signal.
It is worth noting that due to the low E/N ratios, imposed by the breakability of organic nitrates, in the above illustrated mass spectra examples, the intensities of the signals m/z 19 (H 3 O + ) and 37 (H 3 O + (H 2 O)) are of the same order of magnitude while the m/z 55 (H 3 O + (H 2 O) 2 ) is 1 or 2 orders of magnitude lower.
In the case of KnC5, the M kn • H + (m/z = 148) signal is not the most intense one, suggesting that fragmentation is more favored that for KnC3 under similar conditions.The characteristic fragmentation pattern of this analyte exhibits a characteristic m/z 85 signal corresponding to the C 5 H 9 O + group after the cleavage of the NO 3 fragment.An analogous process is described by Aoki et al. (2007), asserting the alkyl group as the main signal in the mass spectra of alkyl nitrates.The abundant m/z 59 fragment (C 3 H 7 O + or C 2 H 3 O + 2 ) is most probably a result of subsequent fragmentation/recombination processes.
The intense m/z = 103 signal could be ascribed to the already mentioned Reactions (R5) and/or (R6), both conducting to the formation of a (C 5 H 9 O)-OH • H + ion with release of HNO 3 and water.It is worth noting that the elimination of supplementary H 2 O from the resulting complex could also contribute to the m/z 85 signal (C 5 H 9 O + ).
In the RF mode, the water cluster distribution is dominated by the H 3 O + ion, for the entire range of E/N * ratios.Its sig- nal is the most abundant, while all the other H 3 O + (H 2 O) n species are at least 1 order of magnitude lower.The redistribution of water cluster in the RF mode modifies obviously the ketonitrates' fragmentation pattern and the burden of each ionization channel in a similar way as in the case of hydroxynitrates.No straightforward correlation can be made in between the two modes, showing the complexity of the ion funnel influence.
The most noteworthy difference in the RF mode resides in the presence of an intense signal corresponding to a H 2 O adduct formation (m/z 137 and 165 respectively), similar to the hydroxynitrate, with a maximum intensity around 48 Td (Table 2) as calculated without taking into account the RF contribution.
With the same PTR entry voltage, the RF mode activation induces the enhancement of the fragmentation due to higher input energies and the maximum of the protonated ketonitrate signal glides towards 48 Td.Making the assumption that the highest sensitivity is related in the two modes to an analogous E/N state, a coarse assessment of the RF mode contribution is estimated (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) as the difference between the two modes maximum sensitivity for the protonated molecular ion signal.
To conclude, it has been observed that protonation of ketonitrates is the main ionization process in the absence of a RF field.The use of the RF mode modifies the fragmen-

The NO + ionization
For the first time, the detection of ketonitrates using NO + ionization was tested.In a similar way as for the protonation process, mass spectra of KnC3 and KnC5 were recorded as a function of E/N ratios.The previously proven affinity of NO + coupling -C(O) functional group in ketones to form adducts (Smith et al., 2003) was for the first time tested in a PTR reactor.The adduct formation was successfully achieved in the presence of NO + for both studied ketonitrates, for which very similar IE are expected.
In the given ionization mode and considering the higher IE of ketonitrates, the adduct formation is expected to prevail over the charge transfer, the literature stating that the yield of parent radical cation formation seems to be anticorrelated with the IE of the ketones, the lowest IE analytes, presenting the highest probability for the charge transfer reaction (Smith et al., 2003).
Following the same approach as in the previous case, the E/N * region was set in order to provide the highest NO + / NO + 2 ratio.Figures 6 and S5 present the influence of the E/N ratio variation in the RF mode over the analyte signal of interest.We notice that the quick enhancement of the NO + / NO + 2 ratio is accompanied by the adduct signal increase.We plot as an example, the KnC5 adduct signal at m/z = 177 (147 + 30) as a function of two fragments (m/z = 43 and 85) with potentially, as explained above, different formation pathways.A decrease of the fragmentation yield in favor of the adduct formation in the PTR reactor is observed over the 35 Td shoulder.The high sensitivity of the instrumental setup for the identification of ketonitrates is quantified in Table 2.
Due to the rising incidence of the NO + and NO + 2 ions, simultaneously formed in the GD, the The KnC3 spectrum was recorded at E/N * = 40 Td as shown in Fig. 7b.We notice a low fragmentation pattern of the analyte with a major adduct signal contribution at m/z = 149.The existence of a secondary (water-controlled) competitive process might be revealed by the existence of the minor protonated signal at m/z 120 backed by the characteristic fragment at m/z = 75 as described by Reactions (R5)-(R7).
In an analog way, in Fig. S4, the mass spectrum of KnC5 is plotted for the instrumental setup corresponding at the lowest fragmentation (E/N * = 36 Td).We notice the distinct signal of the formed adduct (m/z = 177) as well as, to a lesser extent, the characteristic signals described above (m/z = 59, 85, 103), which are most likely formed following the mechanisms described in the previous paragraph, since the water is ubiquitously present in the system.
The best results of the NO + ionization are related typically to the highest NO + / NO + 2 ratios in the RF mode as seen in Table 2.The NO + high affinity towards adduct formation is confirmed by the low DL achieved for these highly characteristic signals: 30 and 40 ppt min −1 for the adducts formed at m/z 149 and 177, respectively, at an E/N * ratio < 45 Td.

PANs
The detection of PANs with PTR-MS has been tested by generating the PAN from the well-known NO 3 oxidation of acetaldehyde in the simulation chamber.The disadvantage of this procedure despite the 70 % PAN formation yield (Doussin et al., 2003) is that it generates several bi-products (nitric acid, formaldehyde, etc.) leading to more complex mass spectra.The PAN family analytes were all obtained by in situ generation using the corresponding aldehydes.In order to overcome any judgment error, the present study will only illustrate the PAN mass spectra while for the other analogous formed PAN-like products (P2MCrN and P3MCrN) we will only discuss the mass spectra signals which are assigned to them.

The H 3 O + ionization
The optimal conditions for the detection of PANs in H 3 O + ionization mode were explored by varying the E/N ratios from 55 to 120 Td.The most promising results were obtained in the case of PAN around 85 Td, where the protonated signal of PAN was recorded as we can depict from Fig. S6.
In a manner analogous to the Reaction (R1a), the protonated peroxynitrates are equally expected to form low energy gas-phase ion-dipole complexes (ROOH • NO + 2 ) with bound energies of 92 kJ mol −1 in the case of methyl-peroxynitrate (Ravelo and Francisco, 2007).Previous studies (Hansel and Wisthaler, 2000) have proposed a speculative decomposition of the protonated PAN: The same mechanism is equally indicated by later studies (Aoki et al., 2007), giving the mass spectral signal at m/z = 77, as recorded in the case of the PAN (Fig. S6).Another likely mechanism which, unexpectedly, may lead to the same analytical signal is reviewed by Roberts (1990), proposing the unimolecular decomposition of PANs to form the corresponding C n−1 alkyl nitrate.In our particular case the reaction may be written as Assuming the presence of both processes, the resulting alkyl nitrate may interfere with the abovementioned m/z = 117 signal in the case of a charge transfer reaction.More likely, in the case of a protonation process, the obtained m/z = 118 signal derived from the alkyl nitrate will enhance the isobaric signal due to the 13 C isotopic abundance of the analyte obtained by PAN decomposition.The signal of m/z 118 seems indicative of the presence of a double peak which might confirm the occurrence of both processes.This fact is equally confirmed by the I 118 / I 117 measured ratio (0.19-0.23 for the P2MCrN and 0.52-0.57for the P3MCrN), which is significantly different from the expected 13 C isotopic abundance for a RC(O)OOHH + -type compound (0.06).This difference suggests the presence of this kind of analytical interference, more pronounced in the case of P3MCrN.However the PAN decomposition path is considered to be several hundred times slower than the bond homolysis channel: In order to certify the mono-nitrogen-containing analytes for several ion signals in the spectrum, an analysis similar to the one performed by Inomata et al. (2013), which consists of subtraction of the isotopic effect of 13 C by calculating I even / I odd ratios of an ion signal at an even m/z [M] to the adjacent ion signal at an odd m/z [M + 1], was performed.
As described by Table 2 the normalized sensitivity for the protonated PANs is weak, spanning few counts of ppb −1 min −1 for all three considered PANs.

The NO + ionization
Unlike the ketonitrates, the PANs present a low sensitivity towards detection in NO + ionization mode and generally a complex chemical equilibrium in classical PTR analysis.
In the NO + ionization mode, the signals corresponding to adduct formation are barely noticeable for the entire range of E/N considered.However, the presence of weak signals at m/z = 191 and 207 for both P2MCrN and P3MCrN is indicative of the presence of PAN NO + and PAN NO + 2 adducts respectively.
The thermolability of the PANs was equally considered for the weak response factors of the instrumental setup in both ionization modes, since the sampling requires slightly heated inlet lines.We considered the 40 • C heating as a compromise between the thermolability of the organic nitrates and the risk of condensation in the sampling line.However, the total residence in the sampling line and the instrument inlet is very short, around 3s.At 40 • C, the lifetime of PAN (due to thermal decomposition) is around 300 s.So decomposition of PANs is expected to be negligible here.

Conclusions
Organic nitrates play a key role in atmospheric chemistry as they act as reactive nitrogen reservoir species.The use of PTR-MS for the measurement of volatile organic compounds has expanded a lot in atmospheric research these last years but few studies have investigated the performances of this instrument for the detection of organic nitrates.These studies have shown that this technique exhibits poor performances (high fragmentation and poor sensitivity) when it is run in classical mode.
In the present work, the detection of alkyl and multifunctional organic nitrates (PANs, ketonitrates, hydroxynitrates) by this technique has been studied by operating the instrument in the classical mode (H 3 O + as ionizing species) but also by testing an NO + -induced, soft ionization.The versatility of the instrument allows an easy change of the chemical ionization source from H 3 O + to NO + by simply replacing water vapor by dry air in the glow discharge.This is very useful for a double identification of the organic nitrates.
This study has shown that two complementary ionization modes can be used for the detection of organic nitrates: -The NO + ionization mode in presence of RF mode with an E/N ratio around 45Td is our first choice.This mode favors the M-NO + adduct formation and has been shown to be more efficient for the detection of all types of organic nitrates except PANs (i.e., alkyl, hydroxynitrates and carbonyl nitrates), as it minimizes fragmentation, favoring the identification of molecular composition.In addition, this mode is much more sensitive to the detection of hydroxy-and ketonitrates than the pro-tonation one.Thus, we recommend the use of this mode for alkyl and multifunctional organic nitrates.
-The protonation mode in absence of the RF mode, for an E/N ratio around 70Td which is recommended for the detection of PAN-type compounds is our second choice.This ionization mode can also be used for other types of organic nitrates but with lower sensitivity than the NO + mode.
Although slight variations appear in the optimization of the operational conditions for each type of compounds, it has been observed that a few unit shifts of the E/N ratios do not significantly affect the analytical sensitivity of the method.The NO + adduct ionization mode appears to be the most sensitive for the detection of hydroxy-and ketonitrates with detection limits in the range of tens of ppt/min.This high sensitivity is suitable for the detection of organic nitrates in lab studies, in particular in simulation chamber experiments.It also suggests that this method is very promising for the detection of these compounds in ambient air.The detection of alkyl nitrates and PANs with PTR-MS is less sensitive, with detection limits in the range of hundreds ppt of min −1 , whatever the ionization source used.
From this study, we now better understand how to run PTR-MS to allow the detection of individual organic nitrates during lab studies and this will be very useful for a number of research groups working on mechanisms studies which are equipped with this instrument.In particular, the operational mode of the instrument has been optimized for the detection of different types of organic nitrates and the ionization pattern (fragmentation, adduct formation, charge transfer, etc.) of these species is much better understood now.
A perspective of this work was to test the detection of organic nitrates in ambient air with this method.From the detection limits observed in this study, we can expect that accumulations over longer periods will be necessary to decrease the detection limits.
A crucial aspect to be taken into account in further studies for lab and field measurements using this method is the effect of the humidity of the sampled air.In addition, longer acquisition time, elimination of interfering ionization paths by selective ionization sources and softer ionization sources could improve the technique's performance.

Figure 1 .
Figure 1.Typical ionizing species distribution as a function of the E/N ratio, in the two ionization modes with RF mode on and off.When the RF mode is on, E/N * ratios were calculated taking into account only the contribution of the DC electric field while the additional input of the AC electric field remains difficult to estimate.

Figure 2 .
Figure 2. The E/N ratio influence over the AlkC3 identification for several representative signals in both ionization modes (H 3 O + , left; NO + , right) as recorded in absence (up)/presence (down) of a RF funnel.The common signal at m/z = 43 is equally plotted as a fragmentation mark (right axis).

Figure 3 .
Figure 3. Recorded mass spectrum of AlkC3 (black bars) at the lowest extent of fragmentation of the molecular ion at m/z = 106 (E/N = 78 Td) in the H 3 O + ionization mode in absence of the RF (a) and in the NO + ionization mode at m/z = 135 in presence of the RF (E/N * = 34 Td) (b).The thin green line represents the expected molecular ion of the analyte.The intense signals depicted by the red thin bars represent the ionizing and matrix analytes at m/z = 30 (NO + ) and 46 (NO + 2 ) and their isotopic abundance signals at m/z = 31 and 47, 48 respectively.The signals corresponding to the water clusters (H 3 O + (H 2 O) n at m/z 37 and 55 are systematically erased for simplification.
signals which have been attributed to 1OH3C3 are m/z 43 (C 2 H 3 O + ), m/z 104 (M 1OH3C3 (-OH) + ), m/z 122 (M 1OH3C3 • H + ) and m/z 139 (M 1OH3C3 • H 2 O) + ).Intensities of these signals have been plotted as a function of E/N in Fig. 4. As expected, the intensity of the signal m/z 43 increases with high E/N ratio while protonation and water adduct formation are favored by low E/N .A possible explanation for the formation of the ion M 1OH3C3 (-OH) + is a

Figure 4 .
Figure 4.The E/N ratio influence over the 1OH3C3 identification for several representative signals in both ionization modes (H 3 O + , left; NO + , right) as recorded in absence (up) / presence (down) of a RF funnel.The common signal at m/z = 43 is equally plotted as a fragmentation mark (right axis).

Figure 5 .
Figure 5. Recorded mass spectrum of 1OH3C3 (black bars) at the lowest extent of fragmentation of the molecular ion (E/N = 62 Td) in the H 3 O + ionization mode in absence of the RF device (a) and in the NO + ionization mode in presence of the RF (E/N * = 41 Td) (b).The thin green line represents the expected molecular ion of the analyte.The intense signals depicted by the red thin bars represent the ionizing and matrix analytes at m/z = 30 (NO + ) and 46 (NO + 2 ) and their isotopic abundance signals at m/z = 31 and 47, 48 respectively.The signals corresponding to the water clusters (H 3 O + (H 2 O) n at m/z 37 and 55 are systematically erased for simplification.

Figure 6 .
Figure 6.The E/N ratio influence over the KnC3 identification for several representative signals in both ionization modes (H 3 O + , left; NO + , right) as recorded in absence (up)/presence (down) of a RF funnel.The common signal at m/z = 43 is equally plotted as a fragmentation mark (right axis).

Figure 7 .
Figure 7. Recorded mass spectrum of protonated KnC3 (black bars) for E/N = 75 Td, corresponding to the highest sensitivity for the protonated analyte signal detection (m/z 120) in absence of the RF device (a) and in the NO + ionization mode in presence of the RF (E/N * = 40 Td) (b).The thin green line represents the expected molecular ion of the analyte.The intense signals depicted by the red thin bars represent the ionizing and matrix analytes at m/z = 30 (NO + ) and 46 (NO + 2 ) and their isotopic abundance signals at m/z = 31 and 47, 48 respectively.The signals corresponding to the water clusters (H 3 O + (H 2 O) n at m/z 37 and 55 are systematically erased for simplification.
signals corresponding to the M kn • NO + and M kn • NO + 2 adduct formation are equally strengthened (m/z = 149 and 165 for the KnC3 and m/z = 177 and 193 for the KnC5; see Figs. 7b and S4 respectively).All these characteristic signals are tailed by their corresponding isotopic abundance signals due mainly to 13 C isotope at m/z =121, 150, 166, 178 and 194.

Table 1 .
Summarized analytical approaches into the organic nitrate analysis.

Table 2 .
Organic nitrates normalized sensitivity as a function of the ionization mode.
* Without taking into account the RF contribution.
www.atmos-meas-tech.net/10/1445/2017/Atmos.Meas.Tech., 10, 1445-1463, 2017 M. Duncianu et al.: Measurement of alkyl and multifunctional organic nitrates tation pattern and enhances the mechanism, leading to the M • H 2 O + adduct formation.The recorded sensitivities listed in Table 2 for the H 3 O + ionization mode are corresponding to a DL of 70-80 ppt min −1 for the protonated signal of KnC3 (m/z 120).A similar DL is reached by the water adduct signal at m/z 137 under the influence of the RF ion funnel.The highest sensitivity for the KnC5 identification is attained by the water adduct signal at m/z 165 recorded at 48 Td in the RF mode with a DL of 20 ppt min −1 .Sensitivities of other specific signals are equally proposed for comparison in Table 2.
According to their statements the other protonated PAN parent compound generated in our study (C 4 H 7 -C(O)OONO 2 H + , m/z = 162) should lead to representative ionic signals of type C 4 H 7 -C(O)OOHH + , m/z = 117, which are indeed present in the deconvoluted mass spectra of our analytes.Moreover, the fact that this signal is absent during the NO + ionization mode (characterized by inferior H 2 O levels) may support this hypothetic decomposition of the PANs.