The derivatization reagents O-(2,3,4,5,6-pentafluorobenzyl) hydroxylamine (PFBHA) (99 %), N,O-bis(trimethylsilyl)trifluoroacetamide (BSTFA), N-trimethylsilyl-N-methyl trifluoroacetamide (MSTFA), and trimethylchlorosilane (TMCs) as catalysts purchased from Sigma-Aldrich (Madrid, Spain) were used directly in this study without any further purification. Methylglyoxal at 30 % in water (MGLY), glyoxal (GLY), glutaraldehyde, methyl vinyl ketone (MVK), methacrolein, glycoladehyde, hydroxyacetone, succinic acid, and benzaldehyde were supplied by Fluka (Madrid, Spain). Other α-dicarbonyl compounds, such as 4-oxo-2-pentenal and E-butenedial, were synthetized (provided by Leeds University, Organic Chemistry Department).

This methodology was carried out in four steps. The first one was the chemical modification of the fibre with PFBHA (carbonyl reagent). The second proposed step was the sampling on a cell of multi-oxygenated compounds for “on-fibre derivatization” of carbonyl compounds. The third step was the chemical modification of the fibre with MSTFA plus TMCs (hydroxyl or carboxylic compounds reagent). Finally, derivatized OVOCs on SPME were injected on GC for thermal desorption and their analysis by mass spectrometry. These steps were carried out by an automated procedure achieved with a Combi Pal autosampler (CTC Analytics, Zwingen, Switzerland) controlled by Cycle Composer software and equipped with SPME sampling adaptors, sample trays, and a temperature-controlled agitator tray. Figure 1 shows a scheme of multi-step derivatization protocol.

For the development of our proposed methodology, a sampling system with an inert sampling line and a cell was designed. The sampling line was of sulfinert^® material, and it was heated at 80 ∘C to avoid losses of sticky OVOCs compounds. The sampling cell of 5cm×4cm×4cm was made of stainless steel. The air passed through, in a turbulent regime at 10 Lmin-1, which corresponds to 50 cms-1, to guarantee air velocity higher than 10 cms-1, defined by some authors as a certain critical value (Gómez-Alvarez et al., 2012; Augusto et al., 2001). Moreover, the sampling cell was installed just below the simulator chamber but at laboratory conditions, i.e. without the influence of solar radiation and under controlled temperature at 20 ∘C, critical parameters for SPME sampling (Gomez-Alvarez, 2007).

The SPME selected fibre was polydimethylsiloxane/divinylbenzene (PDMS/DVB) coating, a stable flex fibre, with a 23 gauge needle size and coating thickness of 65 µm (Sigma Aldrich, USA). This type of fibre was chosen because of its high affinity to PFBHA (Zhu et al., 2015). A robotic injection system (PAL Auto Sampler, Agilent Technologies, USA) was used. The protocol started with the conditioning of SPME fibre in the GC injector port (30 min at 200 ∘C). By doing so, every analysis also constitutes pre-conditioning for the following sampling step, avoiding fibre contamination. Secondly, a PFBHA derivatization solution (87 mg of PFBHA in 2 mL pure) was prepared and located in a reagent tray position 1. Thirdly, PFBHA headspace vapour was generated in an incubation cell under optimized conditions of temperature, agitation time and speed, and adsorption time. Fourthly, SPME doped with PFBHA was located in the sampling cell where air is passed through during a selected sampling time. Fifthly, pure MSTFA plus catalyst (40 µL plus 10 µL of TMCs) was prepared and located in a reagent tray position 2. Silylant reagent head space vapour was also generated in an incubation cell under temperature, agitation time and speed, and adsorption time optimized conditions. Finally, the sample was analysed by GC–MS. The chemical reactions involved in these derivatization steps are shown in Fig. S1 in the Supplement.

The sample was thermally desorbed in a GC port of an Agilent GC–MS (Santa Clara, USA) equipped with a HP-5MS column of 30m×0.25mm I.D ×0.25 mm film (Agilent, Santa Clara, USA). The chromatograph was programmed at 80 ∘C for 2 min, then ramped at a rate of 12 ∘Cmin-1 to 240 ∘C and 100 ∘Cmin-1 to 280 ∘C. The injection port was held at 250 ∘C, and the transfer line from GC to MS was held at 300 ∘C. Samples were injected in splitless mode, using helium as carrier at a flow of 1 mLmin-1. The EI voltage was 70 eV, the ion source temperature was set at 200 ∘C, and the quadrupole temperature was set at 100 ∘C. Full scan mode was used (m/z 50–650) to identify the most abundant ions of multi-oxygenated compounds. The selected ion chromatograms of the most abundant ions were used to quantify them.

The on-line SPME–GC–MS optimization was carried out in the high-volume outdoor smog chamber EUPHORE (EUropean PHOtoREactor) (Valencia, Spain). These chambers consist of two half-spherical fluoropolymeric bags, each one of 200 m3, with integrated measuring systems for monitoring pressure, humidity, temperature, precursor species, and reaction products. Pressure, relative humidity, and temperature were measured using a pressure sensor (Air-DB-VOC, Sirsa, Madrid, Spain) and a dew point hydrometer (TS-2, Walz, Effeltrich, Germany). A Serenius 50 ozone monitor (Ecotech, Melbourne, Australia) was also used (Borrás and Tortajada-Genaro, 2012).

The OVOCs were determined by two techniques. First, samples for the on-line SPME–GC–MS technique were taken over 6 h at a sampling frequency of 20 min (10 Lmin-1, sampling time: 5 min). Second, a White-type mirror system (path length of 553.5 m) coupled to a FTIR with a MCT detector (Nicolet 670, Thermo Scientific, USA) was used. Spectra were collected at 1 cm-1 resolution by averaging 300 scans (sampling time: 5 min). The quantification was based on aldehydic C–H stretching for methylglyoxal, within the spectral region of 2750–3000 cm-1 and C-C and CH2 bands for isoprene, methyl vinyl ketone, and methacrolein in 900–1047 cm-1 by using ANIR software (Ródenas, 2008).

For the optimization of the on-line SPME–GC–MS methodology, methylglyoxal, glyoxal, glutaraldehyde, methyl vinyl ketone, methacrolein, benzaldehyde, glycoldehyde, hydroxyacetone, succinic acid, 4-oxo-2-pentenal, and E-butenedial – selected carbonyl, α-dicarbonyl, hydroxyl-carbonyl compounds, and carboxylic acids – were injected in the smog chamber using an impinger. A stream of hot air (>150 ∘C) produced by a hot gun enabled their volatilization and the transference inside the chamber at a flow of 10 Lmin-1. Moreover, our technique was applied in an intercomparison campaign and in the monitoring of isoprene ozonolysis. Reactants and products are diluted during experiments, and to determine the correct concentration values, they must be corrected. The dilution rate in the chamber was calculated from the decay of SF6 by FTIR areas in the IR range of 762–956 cm-1. The specific dilution process was determined by FTIR adding 120 µgm-3 of SF6 as a non-reactive tracer (value of 1.1×10-5 s-1) to the reaction mixtures at the start of the experiments.

Milli-Q water was added by a sprayer system (died, cleaned air at 2 bar) into the EUPHORE chamber. The dew point system was used to confirm relative humidity values.

A large intercomparison campaign of oxygenated organic compound measurements was held at EUPHORE from mid-May to 1 June 2018 (Muñoz et al., 2019; Ródenas et al., 2021). Taking advantage of it, our proposed SPME–GC–MS plus derivatization technique was compared to a number of instruments: four on-line instruments together with one off-line analytical method. The on-line techniques were on-line SPME–GC–MS plus derivatization, FTIR, two proton-transfer time-of-flight mass spectrometers (PTR-ToF-MS), and a selected-ion flow-tube mass spectrometer (SIFT-MS). The off-line techniques were DNPH cartridges analysed by liquid chromatography–mass spectrometry (LC–MS). These techniques and methodologies were operated by the University of York, University of Leeds, Forschungszentrum Jülich, and Fundación CEAM. A detailed summary of techniques and institutions is in Table S1 in the Supplement.

Measurement of OVOCs at unknown concentrations were done under different relative humidity conditions (0 %–50 % RH). After overnight cleaning, all instruments sampled background air for 1 h. After that, OVOCs were added into the simulation chamber. I.e, methacrolein, acetone, 2-butanone, hydroxyacetone, glycolaldehyde, formaldehyde, benzaldehyde, acetaldehyde, methyl vinyl ketone, glyoxal and methylglyoxal were added in a range of 40–60 ppbV. Samples were taken over 1 h followed by the addition of water to increase the relative humidity up to 50 %. Subsequently samples were taken during a 30 min period, and the gas mixture was diluted within 1 h down to 50 % of the initial concentration. This protocol was repeated twice.

Ozonolysis of isoprene. The ozonolysis experiment consisted in adding isoprene: 220 ppbV and 160 µL were fed to the photoreactor via heated air stream. O3 was added at 23 ppbVmin-1 from an O2 purity of 5.0 using a UV lamp, until reaching 215 ppbV. CO was added as an OH radical scavenger (230 ppm of CO, 5000 ppmV at 10 Lmin-1). Later, all the reactants were mixed for 10 min, and all the instrumentation and techniques took samples over almost 6 h throughout the experiment.

The first challenge was an effective on-fibre derivatization of the carbonyl compounds to oxime products based on a substitution reaction by PFBHA. For assay development, methylglyoxal was selected as the model compound, and a synthetic air mixture was generated at the EUPHORE smog chamber at 50 ppbV (confirmed by FTIR). Complete factorial designs were performed to study the variable effect (Table 1).

The generation of PFBHA reagent vapour in a headspace mode was studied, varying reagent amount, temperature, agitation time, and speed. The selection criteria were (1) maximum oxime signal, (2) minimal artefacts for the reagent depletion, and (3) minimal standard deviation between replicates. Results showed that the signal saturation was achieved at PFBHA solution of 87 mgL-1. In fact, higher PFBHA concentrations reduced the derivatization capacity and increased the interfering peaks as was previously described by Yu et al. (2017). Regarding the incubation temperature, the oxime formation slightly improved when incubation temperature cell increased. The best values were reported at 50 ∘C. The modification of agitation time and speed factors was also tested, leading an improvement of reaction yield and reproducibility. Once an effective generation of the derivatizing reagent in the headspace was achieved, the fibre exposition was evaluated. The chosen adsorption time was 4 min, in accordance with previous studies (Gómez-Alvarez et al., 2012).

The next experiments were aimed at the on-fibre derivatization of carbonyl compounds. A sampling flow of 10 Lmin-1 over 5 min and no incubation time produced maximum capture of analyte at a fast sampling rate. Finally, derivatized compounds were desorbed and directly transferred to the GC–MS. Injector desorption conditions (10 min at 250 ∘C) assured total desorption, calculated from peak areas of methylglyoxal derivatives (13.9, 14.2, 14.4, and 14.5 min), with the presence of a underivatized chromatographic peak of methylglyoxal also being negligible (retention time 7.65 min). Thus, the on-line SPME PFBHA derivatization was completed for up to 95 % methylglyoxal in a reduced time (total time 18 min) and low reagent consumption (0.17 mg per assay). Also, memory effect on the sampling cell was evaluated, including a blank derivatized sample in the sampling sequence. No compounds were observed, confirming the absence of memory effect.

The reusability of PDMS/DVB fibres was tested, performing replicate experiments (>50 assays). Since reproducible responses were registered, on-fibre derivatization is a cost-effective approach for the detection of OVOCs in air samples.

The multiplexed automated on-fibre derivatization was approached, studying the methodology for a mixture of eight carbonyl compounds, including aldehydes, ketones, aromatic aldehydes, and α-dicarbonyls. Then, the air sample contained the selected compounds: methylglyoxal, glyoxal, glutaraldehyde, methyl vinyl ketone, methacrolein, benzaldehyde, 4-oxo-2-pentenal, and E-butenedial.

The chromatograms obtained from air mixtures confirmed the correct on-fibre derivatization and later chromatographic separation process. The chromatographic oxime peaks of each carbonyl compounds, and its corresponding mass spectra are reported in Figs. S2 and S3. The identification of derivatives was based on the molecular ion in the EI mass spectra. The m/z fragments were 265 for both methacrolein and methyl vinyl ketone: 301, 448, 462, 490, 488, and 476, for benzaldehyde, glyoxal, methylglyoxal, glutaraldehyde, 4-oxo-2-pentenal, and E-butenedial, respectively. Also, several fragments were examined, such as ions with m/z= M-30, M-181, and M-211, resulting from loss of NO, C6F5CH2, and C6F5-CH2NO, respectively. Table 2 summarizes the main chromatographic features for the specific oximes formed.

The number of underivatized compounds was negligible (<1.5 %), and the number of PFBHA interference residue peaks was reduced and perfectly resolved (resolution >1.5). The calculated recoveries ranged from 91 % (glutaraldehyde) to 99.7 % (methylglyoxal) as shown in Table S2. Therefore, the suitability of the developed method was demonstrated for a wide range of compounds, considering organic functionalities, molecular sizes, boiling temperatures, and reactivities. Therefore, our method avoided the main limitation of SPME-based methods for multiplexed purposes. Often, the simultaneous determination of several analyte families led to a wide range of recoveries, because these methods are highly sensitive to interferences and experimental conditions, such as sampling time, analyte concentrations, passive or active systems, reagent concentration, temperature, operator handling, etc. (Koziel and Novak, 2002).

The effect of humidity was examined because several techniques such as proton-transfer-reaction mass spectrometry – time of flight (PTRMS-TOF) or cavity-enhanced absorption spectroscopy (CEAS) showed an erroneous determination for air samples with high water content, depending on the applied data evaluation routine (Thalman et al., 2015). Air mixtures of the model organic compounds were prepared and mixed with water vapour. A statistical test demonstrated that the humidity effect was negligible (t test, p values <0.005).

OVOC air mixtures at different concentrations were analysed – ranging from 5 pptV to 100 ppbV. As a selective separation was achieved, appropriate calibration curves were obtained. The regression coefficients were 0.990–0.998.

The evaluation of sensitivity was performed analysing blank samples and serial dilutions of standard air mixtures (dilution factor 1:10). Detection limits (LOD) were calculated as 3 times the standard deviation of blank samples at the retention time corresponding to each compound (Table 3). The estimated detection limits ranged from 6 pptV (glyoxal) to 100 pptV (methacrolein), equivalent to 14 to 237 ngm-3. Limit of quantification (LOQs) were calculated as 10 times the standard deviation of blank samples at the retention time corresponding to each compound. The limits were from 20 pptV (glyoxal) to 300 pptV (methacrolein). The LOD values were compared to those reported in previously published papers. For glyoxal, the mist chamber (Cofer scrubber) obtained a LOD of 2.7 pptV (Spaulding et al., 2002), the DOAS system obtained 2 pptV, and CEAS systems obtained 19 pptV (Pang et al., 2013) or 75 pptV in Pang et al. (2014). For methylglyoxal, other studies obtained a LOD of 170 pptV using CEAS systems (Pang et al., 2013), 185 pptV in Pang et al. (2014), or 89 pptV using PTR-ToF-MS (Michoud et al., 2018). Our detection limit was 97 pptV. The detection limit of our technique is comparable to most spectrometric methods and improves the values for small molecules such as glyoxal and methylgyoxal. This result is particularly relevant because these α-dicarbonyl compounds play a key role in SOA formation in the atmosphere. Although this research was performed in a high-volume simulation chamber (200 m3), the detection limits were below to typical concentrations of these compounds observed in ambient air.

Method precision was estimated from replicate experiments at 1, 10, 25, 50, and 100 ppbV (n=5). The percentage of relative standard deviation (RSD) was 0.2 %–7 %. Other techniques reported higher errors. In Pang et al. (2013), the error was 22 % using microfluidic lab-on-a-chip derivatization methodology. On the other hand, in Pang et al. (2014) the reproducibility was 6.6 % for glyoxal and 7.5 % for methylglyoxal using the same methodology. Therefore, we can conclude that our method presented comparable, even better, results.

The on-line derivatization of hydroxyl or carboxylic groups, directly or after the described process for carbonyl groups, was studied based on a silylation reaction. For that, MSTFA was chosen as the silylant. In addition to previously selected ketones and aldehydes, hydroxyacetone, glycolaldehyde, and succinic acid were selected as hydroxyl carbonyl and carboxylic acid model compounds, due to their atmospheric relevance (Pospisilova et al., 2020; Mellouki et al., 2015). The derivatization sequence studied implied adsorption of PFBHA, sample loading, and MSTFA plus catalyst adsorption. Thus, the oximes (-C=O groups) were generated before the formation of silanes (-OH and -COOH groups). Alternatively, BSTFA was tested as a silylant reagent, but the chromatograms showed more residual peaks and lower response for product peaks. These data suggested that short chain or branched compounds were hindered by steric impediment, as was observed in solution derivatization (Borrás and Tortajada-Genaro, 2012).

The main experimental variables of the MSTFA-based method were optimized from the corresponding chromatographic peaks (Table S3). Together with the molecular ions and PFBHA-associated fragments, the main ions were m/z= M-15 and M-73 resulting from loss of CH3 and Si(CH3)3. The results indicated that the direct loading of vaporized reagent and catalyst (TMCS) was an effective way to transform PFBHA products formed on the PDMS/DVB fibre. The absence of memory column effects and low number of artefacts were observed. Also, a successful thermal desorption of PFBHA MSTFA products was achieved given the quantitative recoveries (>95 %).

Regarding the analytical performances, the LOD was 0.08–0.15 ppbV, the linear range was 5–150 ppbV and reproducibility, expressed as standard deviation, was 2–4 % (see Table 4). In conclusion, double derivatization treatment allowed the proper determination of OVOCs, independently of the functionalized group (-C=O, -OH and/or -COOH), even carbonyl compounds with α-hydrogen.

An intercomparison campaign for the measurement of small multi-oxygenated compounds was carried out at the EUPHORE atmospheric simulator. The study included the variation in OVOC concentration and the impact that the presence of potential interferants, such as high humidity, and dilution steps can have on the methodologies evaluated in this work (see Table S1).

From the different OVOCs, we selected methylglyoxal since it was previously used as the OVOC model (see Sect. 3.1). Figure 2 shows the average concentration in each step. Some techniques (spectroscopic and off-line) presented large interferences. The methodologies provided consistent results of methylglyoxal quantification. As can be observed, the results from SPME–GC–MS plus the derivatization technique were in great agreement with the theoretical values – a known quantity of compound was introduced into EUPHORE chamber – and with the results obtained by other techniques, both optical and mass spectroscopic methodologies. In fact, we can affirm that it does not present interferences with relative humidity lower than 60 %. For that, a t test for paired samples was performed with a result of α<0.5 for our proposed methodology, confirming that high humidity did not significantly affect our measurements.

The intercomparison included OVOCs identified as main degradation products of biogenic pollutants (Aschmann and Atkinson, 1994; Iannone et al., 2010). Figure 3 shows the determined concentration levels of methacrolein and methyl vinyl ketone in a mixture containing both compounds. As previously observed, the results from SPME–GC–MS plus the derivatization technique agreed with the theoretical values from chamber dilutions. Regarding other techniques, the concentrations were comparable to SIFT-MS, FTIR, and DNPH cartridges analysed by LC–MS. However, PTR-ToF-MS cannot discriminate between the structural isomers of both carbonyl compounds. In this case, the sum of MVK and methacrolein (MACR) is measured due to PTR-MS methods not being selective. Both compounds have a different sensitivity factor, imposing an additional inaccuracy on the data; for more details see Ródenas et al. (2021). On the contrary, the on-line SPME–GC–MS approach can be used for a reliable monitoring of both MVK and MACR in atmospheric reactions.

The ozonolysis of isoprene, one of the main biogenic compounds emitted to the atmosphere, was studied at the EUPHORE atmospheric simulator. On-line SPME–GC–MS plus PFBHA derivatization and MSTFA derivatization was applied for tracing the formation of the multi-oxygenated compounds. Figure 4a shows the decay of isoprene and ozone and the major degradation product (formaldehyde) formation. The degradation rate of isoprene and ozone was fitted to first-order decay as previously described in Karl et al. (2004). Regarding minor products, the OVOCs determined were 2-butanone, methacrolein, methyl vinyl ketone, glycoladehyde, hydroxyacetone, glyoxal, and methylglyoxal. Some of them are plotted in Fig. 4b. The results were fitted to a standard growth for degradation products. In the case of 2-butanone, the formation was fast, and, after 1 h, a decay by chemical degradation was registered. The maximum concentrations were (23.6±1.2), (25.4±1.3), (4.2±0.2), (4.1±0.2), (3.8±0.2), (2.1±0.1), and (2.2±0.1) ppbV, for 2-butanone, methacrolein, methyl vinyl ketone, glycoladehyde, hydroxyacetone, glyoxal, and methylglyoxal, respectively. Other OVOCs such as methyl vinyl ketone were also detected. A total of seven OVOCs were identified and quantified in the isoprene ozonolysis, in good accordance with previous studies (Karl et al., 2004; Wennberg et al., 2018).

Therefore, our proposed on-line SPME–GC–MS plus derivatization enabled the accurate sensitive atmospheric monitoring of secondary pollutants. This technique will support their relevance since the ozonolysis of biogenic VOCs such as isoprene gives multifunctional oxygenated organic compounds that participate in the formation of aerosols. SOA formation during the atmospheric oxidation of biogenic organic compounds is estimated at 20–380 Tg r-1 globally, influencing human health and climate (Mellouki et al., 2015).