Samples (punch of 46 mm) were spiked with 5 µL of the internal standard (1S)-(+)-camphor-10-sulfonic acid (Sigma Aldrich, 98 %) at 20 µg mL⁻¹ in 50:50 (v/v) acetonitrile/ ultrapure water and cut into smaller pieces. Pieces were transferred to amber vials and extracted with 4 mL of acetonitrile (ULC/MS-CC/SFC grade, Biosolve, 99.99 %) using a Mini Shaker (15 mm Orbital, VWR) at 1000 rev min⁻¹ for 30 min. The extracts were individually filtered using a glass syringe coupled to a syringe filter (PTFE membrane, 13 mm ×0.2 µm, VWR). The filtered solutions were then evaporated to dryness using a 12 positions N-Evap (Organomation) under a gentle stream of nitrogen at 99 % purity fed by a CALYPSO 35 L min⁻¹ generator (F-DGSi, 2023). Samples were dissolved with 200 µL of 50:50 (v/v) acetonitrile/ ultrapure water, transferred to 250 µL vial inserts and stored for up to 24 h at -18 °C prior to analysis.

Samples were analyzed by means of an UPLC/ESI-IMS-QTOFMS system consisting of an ACQUITY™ UPLC I-Class system and a Vion™ ion mobility hybrid (IMS) QTOF mass analyzer (Waters™). A UPLC BEH C18 column (1.7 µm, 2.1×100 mm, Waters) was used as stationary phase. Mobile phases were ultrapure water with 0.1 % formic acid (v/v) (A) and acetonitrile with 0.1 % formic acid (v/v) (B) with an elution gradient of: 2 min 5 % B, 2–32 min from 5 % to 60 % B, 32–35 min from 60 to 95 % B, 35–38 min hold at 95 % B, 38–40 min from 95 % to 5 % B, and finally stabilization at 5 % B for 5 min. Separation was performed at a 40 °C column temperature and a flow rate of 0.4 mL min⁻¹. 2 µL of sample were injected in triplicates. Solvent blanks (50:50 (v/v) acetonitrile/ ultrapure water) were injected between replicates to check for any carry-over. Between sequences, a source cleaning step was performed to increase the instrument sensitivity (as detailed in Sect. 2.5).

Analysis was performed in negative ionization mode with the following ESI parameters: 120 °C source temperature, 600 °C of desolvation temperature and cone gas flow of 150 L h⁻¹. Mass spectra were recorded in full scan mode in the m/z range of 50–1000, where the m/z corresponds to the mass of the deprotonated molecules. Further data processing was performed only for a m/z range between 50–350. Compounds were then identified by means of their m/z, retention time (Rt) and collision cross section (CCS) with m/z error ≤5 ppm, CCS error ≤2% and Rt error ≤0.1 min. Quantification was done on the full scan spectra, in extracted ion current, with an external calibration in the range 10 to 200 µg mL⁻¹ with (1S)-(+)-camphor-10-sulfonic acid as internal standard. Calibrations were performed using the same extraction procedure as for the real samples. Further details of the calibration procedure are discussed in Sect. 3.1.5.

Samples (punch of 46 mm) were spiked with 5 µL of a heptanoic acid (Sigma Aldrich, 99 %) solution at 40 µg mL⁻¹ as internal standard and extracted in acetonitrile (HPLC grade, VWR, 99.95 %), filtered and evaporated to dryness similarly to the extraction procedure described in Sect. 2.3.1. Samples were then reconstituted by adding 50 µL of acetonitrile. 200 µL of N,O-Bis(trimethylsilyl)trifluoroacetamide (BSTFA) containing 1 % trimethylchlorosilane (TMCS) (Sigma Aldrich, purity 99 %) was added to each solution and then heated at 60 °C for 30 min to allow derivatization. This derivatization process, including the volumes, was selected following Albinet et al. (2019). Final extracts were stored at -18 °C and analyzed as quickly as possible after 24 h.

Furthermore, derivatization was used as a support for product identification. In the derivatization process with BSTFA, the labile hydrogens of the alcohols and acid functions of the compounds are replaced by trimethylsilyl-Si(CH3)3 groups as follows: R1 BSTFA derivatization gives specific ion fragments in mass spectra at m/z 73 [Si(CH3)3]⁺ and 117 [COO=Si(CH3)3]⁺. For compounds bearing labile H atoms, they exhibit a m/z 147 peak corresponding to [(CH3)2Si=OSi(CH3)3]⁺. For identification of individual compounds using authentic standards and evaluation of their response, a 50:50 (v/v) mixture of BSTFA and the standard solution was left reacting overnight at room temperature. BSTFA was selected as one of the most common derivatization reagents for compounds with labile hydrogens (Claeys and Maenhaut, 2021; Cochran et al., 2012; Chiappini et al., 2006) due to the predominance of acidic groups of the target markers (Table 1). However, other derivatization reagents could be used to expand the detection to other functionalities, for example, O-(2,3,4,5,6-pentafluorobenzyl)hydroxylaminehydrochloride for carbonyl compounds (Nozière et al., 2015; Orata, 2012).

Analysis was performed using a GC-MS made of a gas chromatograph (Clarus 650, Perkin Elmer) and a mass spectrometer (MS SQ8C, Perkin Elmer). Separation was achieved using an analytical column RXi-5Sil MS (30 m, 0.25 mm ID, 0.25 µm Restek) and the following GC oven temperature gradient: an initial temperature of 60 °C held for 15 min, followed by a ramp of 5 °C min⁻¹ from 60 to 280 °C, and 7 min held at 280 °C. Helium was used as carrier gas at a flow rate of 1 mL min⁻¹. An injection volume of 5 µL with a high precision syringe (84301 CR700-20 1–20 µL, Hamilton) was used. Ionization was performed using an electron impact source. Mass spectrometry analysis was performed in a mass range between 50 and 500 m/z with a mass scan time of 0.3 s between 6 to 66 min. We used the full scan total ion current (TIC) mode and when additional compound verification was required, the selected ion recording (SIR) mode for 6 channels. Compounds were identified by monitoring the Rt and major or specific ions derived from fragmentation using a mass spectra database built from the individual standards injection. Some of them were also verified using the NIST (National Institute of Standards and Technology) library. Quantification was performed by ion mass extraction in the TIC mode for characteristic fragments, subtracting the signal of the blanks and using external calibration with heptanoic acid as internal standard. Calibration standards underwent the same extraction and derivatization procedure as the real samples. Further details are provided in Sect. 3.2.3.

Different methods were tested to separate and identify 14 compounds using the UPLC/ESI-IMS-QTOFMS: cis-pinonic acid, pinic acid, norpinic acid, terebic acid, MBTCA, 4-nitrocatechol, syringaldehyde, 4-methyl phthalic acid, phthalic acid, 3-acetylbenzoic acid, 4-nitrophenol, 2-methyl-4-nitrophenol, azelaic acid and (1S)-(+)-camphor-10-sulfonic acid. Initial methods tested on individual standard solutions at 10 µg mL⁻¹ in 100 % methanol used elution gradients with varying mobile phases, slopes and lengths of 17 and 60 min. Mobile phases consisted of 0.1 % formic acid (v/v) added in both solvents, ultrapure water (A) and methanol (B). Addition of formic acid is a common practice used to stabilize the solution pH, reducing unwanted adducts formation and leading to the improvement in detection of the analyte signals, ionization, peak shape and separation efficiency (Kaufmann et al., 2024; Liigand et al., 2014).

For the 17 min elution method, the following gradient was used: 2 min isocratic at 5 % B, 2–10 min linear gradient from 5 % to 50 % B, 10–11 min linear gradient from 50 % to 99 % B followed by an isocratic at 99 % B for 2 min, 13–15 min linear gradient from 99 to 5 % B and equilibration at 5 % B for two min. For the 60 min method, the elution conditions were: 3 min isocratic at 5 % B, 3–25 min linear gradient from 5 % to 50 % B, 25–43 min linear gradient from 50 % to 90 % B, 43–48 min linear gradient from 90 to 5 % B and equilibration at 5 % B for 3 min. The 17 min method showed the overlap and irregular peak shapes for MBTCA, terebic acid, and phthalic acid with coelution of pinic acid and 4-nitrophenol, and of norpinic acid and camphor sulfonic acid (Fig. S1 in the Supplement). The 60 min method provided better performance associated with a decrease in the number of compounds coeluted (Fig. S2), however, given that most of the target compounds elute in the first 20 min, such a long method would be unnecessarily time and resource consuming.

An intermediate method of 45 min with mobile phases with 0.1 % formic acid (v/v) in ultrapure water (A) and 0.1 % formic acid in organic solvent (B) was selected. For this method, acetonitrile and methanol were considered for the evaluation and the following gradient was tested: 2 min isocratic at 5 % B, 2–32 min linear gradient from 5 % to 60 % B, 32–35 min linear gradient from 60 % to 95 % B, 35–38 min isocratic at 95 % B, 38–40 min linear gradient from 95 % to 5 % B, and equilibration for 5 min at 5 % B. Example chromatograms using the 45 min method with methanol and with acetonitrile as mobile phase are shown in Fig. S3. As observed, most of the standards were properly identified after chromatographic separation with good peak shapes and compounds eluting in the first 18 min for both organic solvents. To assess whether carry over would affect the measurements with a reduced method compared with the 60 min one, blanks were injected in between samples. No carry over was observed therefore, the 45 min elution method was selected for the analysis.

The use of methanol and acetonitrile as organic solvents for the chromatographic separation was evaluated by comparing the compound responses obtained using both solvents as shown in Fig. 1. For these tests, standard solutions were prepared in 50/50 ultrapure water/organic solvent mixture, matching the same organic solvent used for mobile phase B in order to minimize any possible artifact that could affect the peak shape even though the injection volume was only 2 µL. Among the 14 target compounds, six (pinic acid, 3-acetyl benzoic acid, 4-methyl phthalic acid, cis pinonic acid, syringaldehyde and 2-methyl-4-nitrophenol) showed a similar response with both solvents. In the case of terebic acid, camphor sulfonic acid, MBTCA, norpinic acid, and phthalic acid, higher responses were observed with acetonitrile as organic solvent and the opposite effect was observed for azelaic acid and 4-nitrocatechol. Azelaic acid was the compound most negatively affected using acetonitrile as an eluent. However, this compound was initially considered in our tests to identify possible sample contamination as it was identified by the software at multiple Rt. After its identification, we discarded azelaic acid presence. Acetonitrile was therefore selected due to the higher elution power observed for the analysis of the target compounds, better suitability for the column (operated at 40 °C) used in this work, and overall better response for most of the target analytes.

Samples in this work were collected on quartz fiber filters (Sect. 2.2), therefore, to mimic the sample conditions, we evaluate the response of the compounds after performing the sample extraction procedure (Sect. 2.3.1). Two mixtures of compounds were prepared: biogenic (cis-pinonic acid, pinic acid, norpinic acid, terebic acid and MBTCA) and anthropogenic (4-nitrocatechol, syringaldehyde, 4-methyl phthalic acid, phthalic acid, 3-acetylbenzoic acid, 4-nitrophenol, 2-methyl-4-nitrophenol) mixtures at 400 µg mL⁻¹. Then, blank filters were spiked with 5 µL of a (1S)-(+)-camphor-10-sulfonic acid and 5 µL of each 400 µg mL⁻¹ mixture containing the compounds under analysis. As shown in Fig. 2, the instrument variability between three randomly injected replicates of the mixture solution without filter extraction was less than 21 % for all target compounds. After filter extraction, a higher replicate variability for phthalic acid (27 %), 4-methyl phthalic acid (27 %), MBTCA (53 %) and camphor sulfonic acid (35 %) was observed. The response variability of the target compounds can result from the extraction procedure. Given the different compound polarities and volatilities, their extraction will be influenced by the dissolution, filtration and solvent evaporation steps.

Individual standards and liquid samples injected without extraction were conserved in actinic glass vials of 1.5 mL with a solution volume of 0.5 mL. In the case of samples extracted from filters, a lower volume was used (200 µL) to increase compound concentrations and detection probability, requiring the use of vial inserts. In this study, we chose a conical bottom shaped insert with plastic spring. To understand possible artifacts from the experimental setup, we evaluated the needle position, effect of inserts and storage time influence on the sample stability on a mixture containing the target anthropogenic standards (Table 1). The needle position test refers to variations in the height from the bottom of the vials.

We first focused on the analysis of anthropogenic standards in solutions containing mixtures at 0.5, 1.0 and 2.5 µg mL⁻¹. As shown in Fig. S4, the variability between replicates for samples directly injected from the vials is high (>14%). The compound-dependent injection variability from the vials was lower than 20 % for syringaldehyde while for 2,5-dihydroxybenzoic acid it was much higher (87 %). The same solutions on the vials were also injected in nonconsecutive triplicates during two different dates, labeled as new (14 March 2023) and old (3 March 2023) experiments (Fig. S5). The compound responses between triplicates were stable, with nitrophenol compounds showing the highest variability. When comparing the old and the new experiments, a compound-dependent variability in the response was observed. The responses were similar for acidic and aldehyde functionalities, but not for nitrophenol compounds, as higher responses were observed in the new experiment, especially at lower concentrations (0.5 µg mL⁻¹). This suggest that time influence the compound response (see also Sect. 3.1.4).

Signals were also higher than for samples using inserts, for which the replicate variability is lower. These differences between injections without and with inserts can be influenced by sample degradation and signal loss over time as samples in vials (without inserts) were analyzed first in the sequence. Although the differences between using and not using of inserts, their presence is needed to maximize the sample intake into the system due to the low sample volumes after extraction. Variations in sample response when using inserts could be influenced by the amount of sample entering the system. Bubble formation in the bottom of the conical shaped inserts during sample transfer into the insert can limit the sample intake. This hypothesis was discarded as changing the distance of the needle from the bottom of the vial between 5 and 10 mm did not affect the signals. Further explanations for the response variation are explored in Sect. 3.1.4. As no influence was observed at varying the needle position, we selected 10 mm distance from the bottom for future experiments as a default parameter suggested by the manufacturer.

To better understand the signal decrease of the target compounds, the response from the mixture solutions (Sect. 3.1.3) of the anthropogenic markers in the 50/50 solvent of ultrapure water/ acetonitrile directly injected (vial) is shown in Fig. S6. Stability tests were performed using mixture solutions at 1.5 and 2.5 µg mL⁻¹. A decrease between the first and the third injection was observed for all the compounds, which could be either associated to the stability of the solution or a decrease in the instrumental response. The target markers showed to be stable in solution as a variability <20 % was observed at consecutively injecting them (Table S2, Sect. 3.1.3). Therefore, the stability of the compounds in solution was discarded to have such an influence in the decrease of the response.

A control solution containing a mixture of acetaminophen, leucine enkephalin, sulfadimethoxide, sulfaguanidine and Val-Tyr-Val (Waters), referred to as SST solution hereafter, was systematically used as quality control, injected in 5 replicates at the beginning of each experiment, to assess the quality of the instrument calibration. The same SST solution was injected in 5 replicates before and after a sequence of 34 sample injections (duration approx. 25 h) to monitor the evolution of signal intensity over time. A signal decrease by more than 70 % was observed after the analyzed sequence (Fig. S7), which was associated with an instrumental signal loss since the SST solution is normally stable for up to several months and is thus unlikely degraded by 70 % over a 25h period.

To follow the signal decrease more precisely over time, we additionally used the LockMass solution (Waters) which is infused continuously at 15 µL min⁻¹ to correct for minor mass deviations in real time throughout whole analysis sequences. The LockMass solution consists of 200 pg µL⁻¹ Leucine-Enkephalin in 50/50 acetonitrile/ultrapure water with 0.1 % formic acid; this solution is normally stable for weeks at ambient temperature. For simple injection-to-injection comparison, the response of the LockMass was extracted by summing the signal (TIC) over the 45 min of each individual infusion (Fig. S8) and we observed a linear decrease in the solution response. Such variation observed at testing the signal evolution over time for the SST and LockMass solutions could only derive from the instrumental signal loss over time. Therefore, calibrations over time were performed during analysis to account for the signal stability (as detailed in Sect. 3.1.5).

Additionally, the cone gas flow was tested at three levels to increase the signal stability: 50, 100 and 150 L h⁻¹ using a fresh mixture at 2.5 µg mL⁻¹ containing the target anthropogenic compounds. As shown in Table S3, similar responses were observed at different levels of cone gas flow for acids. However, at 150 L h⁻¹ a slightly higher response for nitrophenol compounds was observed. Therefore, 150 L h⁻¹ was selected as the cone gas flow for further analysis.

To evaluate the compound response and account for instrumental signal loss, calibrations were performed in sequences on non-consecutive triplicates, using increasing concentrations of mixtures containing the target compounds. Standard solutions containing the biogenic and anthropogenic mixtures of target compounds were prepared in 50/50 ultrapure water/acetonitrile at concentrations of 4, 10, 20, 40, 100, 150 and 200 µg mL⁻¹. For each standard solution, 5 µL of each standard mixture were individually added to clean (baked) quartz filters together with 5 µL of a solution of (1S)-(+)-camphor-10-sulfonic acid as internal standard. The filters were extracted following the procedure of Sect. 2.3.1. Calibration curves were performed by considering the mass of the compound deposited on the filter before extraction.

The variability of the response of the internal standard is represented in Fig. S9. A 60 % decrease in the signal of camphor-10-sulfonic acid was observed between the first and second replicate, with a higher variability (30 %) compared to the replicates 2 and 3 which had a response difference of 5 %. A signal decrease from the first replicate was also observed for most of the calibration standards (Fig. S10). As an instrumental signal decrease was observed, the variations in the compound response between replicates can be attributed to this loss. For the nitrophenol compounds (2-methyl-4-nitrophenol, 4-nitrocatechol and 4-nitrophenol), an apparent increase in the signal response for intermediate concentration was observed. This variability can be influenced by the stability of the solution, possible signal interferences for nitro compounds and matrix effects which can affect individual compound responses, as this behavior was not observed for acids. We consider that a similar effect would occur also on real samples. Therefore, the variation observed for the calibration standards is representative of phenomena affecting the atmospheric sample responses. Carry-over effect on the column was rejected as an explanation as these nitro compounds were not detected in the blanks that were analyzed between replicates.

At normalizing the compound response for the internal standard response, the variability between the replicates decreased for compounds such as 4-methyl phthalic acid, but not for the nitro compounds. Although (1S)-(+)-camphor-10-sulfonic acid (open ring-acid) may not represent the more suitable internal standard for aromatic compounds, this acid shows variations <30% being representative of most of the target compounds. Therefore, we use the normalization to its response to reduce the uncertainties associated with the system set-up and extraction procedure for the samples analyzed here.

Linear calibrations with high determination coefficients were derived for all target compounds, including those with lower responses. We used sequences of a maximum of 54 injections (approx. 40 h). Because the quantification is replicate-dependent, for each set of experiments, we performed calibrations and sample injection in a consecutive way for each replicate. Calibrations over time in the sequence account for instrumental signal degradation and allows to perform the quantification with the closest replicate. The calibration for each replicate is reported in Table 2, together with the m/z, Rt and CCS values used to identify each compound. For compounds for which the mass derived from the normalized signal present a high variability between the triplicates, only the closest two replicates were considered in the quantification. VXO values for individual calibration replicates showed the increase in the method variability between replicate for compounds such as 2,5-dihydroxy benzoic acid (from 25 % to 40 %), which also showed the lowest recovery. The rest of the compounds, with exception of MBTCA, were associated to VXO values <30%, showing the good performance of the method at considering linear calibrations. The final individual compound concentrations are reported as the arithmetic mean, and the error associated to their concentrations consider the standard deviation between triplicates and the volume deviation during the sampling.

Compound response was evaluated for a mixture of 19 biogenic and anthropogenic markers (Table S1) with and without filter extraction. First, after an over-night derivatization as described in Sect. 2.4.1, a mixture of the target compounds at 6 µg mL⁻¹ was directly injected in the system on the following day. All the compounds added to the mixture were detected (Fig. 3). Therefore, 10 µL of a standard mixture of the target compounds at 50 µg mL⁻¹ (more concentrated to account for dilution) was spiked on a blank filter together with the internal standard and extracted as described in Sect. 2.4. A blank filter containing only the internal standard was simultaneously extracted to identify possible interferences.

As shown in Fig. 3, there was an increase in the background signal represented by a blank filter (blue chromatogram) compared to the standard solution directly injected without any extraction step, but with a derivatization step. Some of the peaks observed in the blank were also present in the filter containing the standards mixture. Those were assigned following the NIST library and most of them were identified as Si-groups such as siloxane and silanol compounds, which can come from impurities present in the quartz filters that appear during the extraction. In contrast to direct injection, not all the compounds (malonic acid, terebic acid) present in the solution were identified, showing possible issues during the extraction procedure and/or derivatization time. It is worth to consider that after extraction, the response of compounds of higher volatility can be influenced by the solvent evaporation step. Another additional peaks can arise from experimental manipulation (e.g., palmitic acid, lactic acid) or impurities present in the quartz filters used.

As a significant blank contribution was observed, we tested a mixture of acetonitrile HPLC grade (VWR chemical, 99.95 % purity) and the derivatization reagent BSTFA with the different steps of the method. First, the mixture was directly injected into the system and the noise level of Fig. 3 or possible impurities were not observed (grey plot, Fig. S11). Consecutively, the solution containing both components (solvent and BSTFA) was heated following the derivatization protocol, and the peaks previously observed in the blank in Fig. 3 were also present (blue plot, Fig. S11). As this blank contribution could originate from possible impurities derived from the solvent-filter interaction, we evaluated the response of acetonitrile ULC/MS-CC/SFC grade (Biosolve, 99.99 % purity) as it has a higher purity. No improvements in the blank signals were observed by switching the solvent (black plot, Fig. S11). The derivatization procedure is required to allow the decrease in polarity of some target compounds and therefore their identification and quantification. Although blank contributions to the signal were observed (Fig. 3), most of the target compounds could be identified and quantified. Therefore, a blank filter was simultaneously analyzed with each batch of real samples and its contribution was subtracted.

As deuterated standards were not commercially available, we selected heptanoic acid as internal standard. In online SFE-GC-MS, tridecane and/or o-toluic acid have been used as internal standards (Chiappini et al., 2006, 2019; Lamkaddam et al., 2020). Here, heptanoic acid was selected over other compounds such as tridecane and octanal due to its higher solubility in acetonitrile and presence of a labile proton, making it more representative of the target compounds (mainly carboxylic acids). The variability between replicates for the heptanoic acid was evaluated with and without filter extraction as shown in Fig. S12. A minimum variability was observed for the peak area without filter extraction (<3%), while it reached 36 % for solutions extracted from spiked filters. A similar variability between samples and replicates was observed for octanal (30 %), and attributed to the extraction procedure as no influence of the derivatization was found.

From a mixture of biogenic and anthropogenic standards at 50 µg mL⁻¹, volumes of 6, 8, 12 and 15 µL were individually added to quartz filters with a constant volume of 5 µL of a heptanoic acid solution. After filter extraction and analysis, the variability between triplicates for each injection volume was evaluated (Fig. 4). Replicates were injected on different days, within 3 and 4 d from the first injection. The time between injections was selected to best represent typical storage time of real samples, assuring instrumental response and non-significant solution degradation after storage at -18 °C. Longer storage times were not evaluated and are not discussed here.

The proposed method seems suitable for most of the target compounds: cis-pinonic acid, pinic acid, pinanediol, α-methylglyceric acid, 2-methylerythritol, 4-nitrocatechol, phthalic acid, DHOPA, succinic acid, glycolic acid, p-toluic acid and 2-methyl-4-nitrophenol. However, for pinanediol and 2-hydroxy-3-hydrobenzaldehyde (Fig. 4), the variability for the injections at 12 and 15 µL was high and the compounds were not detected in some replicates. The lack of detection of those compounds could derive from an incomplete derivatization after 30 min or enhanced matrix effects at higher concentrations. For other compounds, such as DHOPA and 2-methyl-4-nitrophenol, higher variability was observed at the lowest measured concentrations, close to the LOD.

Quadratic calibrations with R2>0.90 were used for the target compounds. The calibration information together with Rt, molecular weight and ions monitored after fragmentation are summarized in Table 3. While quadratic calibrations were more suitable for the GC-MS measurements, the VXO of the method showed higher values than those of the UPLC/ESI-IMS-QTOFMS. The lower performance of this method can be associated with a lower sensitivity. Comparison between both methods is discussed further in Sect. 3.3. Coefficients of variation (CV) are higher for more substituted compounds such as DHOPA and α-methylglyceric acid and compounds with aldehyde functions (2-hydroxy-3-methylbenzaldehyde), showing a lower performance for those compounds. Contrastingly, lower values of CV were observed for nopinone (no derivatization) and acids of lower molecular weight such as glycolic acid and succinic acid. For the rest of the compounds, CV was ∼ 40 %. The final individual compound concentrations are reported with their experimental error obtained from the quadratic fit for the compound mass and the volume deviation during the sampling.

Between UPLC/ESI-IMS-QTOFMS and GC-MS, 23 biogenic and anthropogenic organic markers were quantified, with 5 species being detected by both methods (Tables 2 and 3). Together, those methods allowed for the analysis of a substantial list of aromatic and non-aromatic compounds containing acids, alcohols and aldehyde functions. UPLC/ESI-IMS-QTOFMS offered the advantage of detecting phenol compounds at higher sensitivity. For example, UPLC/ESI-IMS-QTOFMS analysis showed lower values of the LODs (<5 ng) and higher recoveries (92 %–174 %) for 4-nitrophenol and 2-methyl-4-nitrophenol. This was not the case with the other phenol compound (nitrocatechol), which showed lower extraction recoveries (47 %–114 %). For the rest of the organic acids and aldehydes, LOD between 23 and 140 ng were obtained (Table 2). Lower extraction recoveries were additionally observed for more substituted markers such as MBTCA, terebic acid, syringaldehyde and 2,5-dihydroxy benzoic acid. Acceptable values of recoveries are suggested between 70 % and 130 % (Golubović et al., 2019). The higher ranges of recoveries observed for nitrophenol compounds shows the possible influence of the matrix.

As shown in Table 3, LODs for GC-MS were higher than for UPLC/ESI-IMS-QTOFMS, due to its lower sensitivity. Despite the differences in the sensitivity, the extra derivatization step on GC-MS offered the advantage of highly functionalized compounds detection, especially those of higher polarity, enhancing the number of markers that can be quantified. Recovery rates between 82 %–133 % were observed for the biogenic and anthropogenic markers with exception of α-methylglyceric acid, 2-methylerythritol and DHOPA for which recovery rates were lower. For 2-methylerythritol, a polyol with four OH groups susceptible to derivatization with labile protons, the lower recovery rate is attributed to incomplete derivatization, affecting the extraction efficiency. Similarly, α-methylglyceric acid and DHOPA functionalities can derive into three substitutions. Despite the lower LOD values observed by UPLC/ESI-IMS-QTOFMS, a higher recovery rate was observed for 4-nitrocatechol using GC-MS.

Table S4 summarizes LODs values observed in this work compared to those of the literature for some of the target species. Values for pinonic and pinic acids of this work (>44 ng) are higher than those reported by Chiappini et al. (2006). This variability can be attributed to differences in the extraction and derivatization steps performed during GC-MS analysis. Chiappini et al. (2006) performed online SFE, which allows the solvent removal from the separation step, while in this work the presence of the solvent and derivatization reagent mixture contributes to the background signal, influencing the LOD. When comparing with Albinet et al. (2019), LODs were compound dependent as similar values were observed for 2-methylerytritol, but not for pinic and cis-pinonic acids, both using GC-MS but different calibration methodologies. Variations were also observed between LC-MS techniques. For example, King et al. (2019) and Amarandei et al. (2023) provided LOD <5.7 ng mL⁻¹ for terebic acid, lower than the one obtained here. Similar LODs for 4-nitrophenol and 2-methyl-4-nitrophenol were observed in this work (17 and 22 ng mL⁻¹) compared with Hoffmann et al. (2007), but higher than those reported by Ikemori et al. (2019). For syringaldehyde, the LOD was one order of magnitude higher than Hoffmann et al.'s (2007). Such differences among the validation parameters between the different studies can result from instruments sensitivity, sample preparation protocols and calibration types.

Samples collected in the Rambouillet forest were analyzed by both UPLC/ESI-IMS-QTOFMS and GC-MS. Some of the biogenic and anthropogenic markers identified are summarized in Fig. 5.

Terebic acid was detected in four of the forest samples while MBTCA was detected in most of the samples using UPLC/ESI-IMS-QTOFMS. For GC-MS analysis anthropogenic and biogenic acids such as succinic and α-methylglyceric acid were detected. 4-nitrocatechol, which we considered here as a biomass burning marker, was quantified only during 19 July, where a fire event was reported (Menut et al., 2023). Five common compounds could be detected: cis-pinonic acid, pinic acid, 4-nitrophenol, 2-methyl-4-nitrophenol, and 4-nitrocatechol. Because the concentrations of the nitro-compounds were below the LOD for GC-MS, hereafter we focus on the comparison of cis-pinonic acid and pinic acid. As observed in Fig. 6, the comparison of the concentration values obtained for cis-pinonic acid and pinic acid showed good determination coefficients (R2>0.8). For cis-pinonic acid, most of the concentration's values obtained by the two methods are similar, except for three samples for which the concentrations obtained by UPLC/ESI-IMS-QTOFMS were about twice or three times higher than those measured by GC-MS. A similar behavior was observed for pinic acid for the highest concentrations. While for the remaining samples, at lower concentrations, values observed by GC-MS were higher than those measured by UPLC/ESI-IMS-QTOFMS.

As shown in the Bland-Altman plots (Fig. 6), measurements between GC-MS and UPLC/ESI-IMS-QTOFMS are comparable as they fall inside the limits of agreement (red lines). However, at comparing the concentration values, a mean difference (reported as mean values in Fig. 6) between both techniques of 11 ng m⁻³ for cis-pinonic and 10 ng m⁻³ for pinic acid were observed. Both compounds follow a similar behavior of standard deviation variation on the upper interval, with exception of three measurements, which were systematically closer to the lower limit. Differences between the concentrations observed for both techniques can be influenced by their sensitivity, extraction procedure and sample aging. As discussed in Sect. 2.4.1, the derivatization in GC-MS lowers the polarity of the compounds and influences their detection. Additionally, we assumed that filter samples used here have a homogenous distribution between different pieces used for this analysis, this together with a time of 8 months between UPLC/ESI-IMS-QTOFMS and GC-MS analysis can also introduce discrepancies between the techniques.