β-Pinene (99 %), cis-pinonic acid (98 %), camphoric acid (99 %), 4-hydroxybenzoic acid, naphthalene (≥ 99.7 %), 1,2-naphthoquinone (97%) and pimelic acid (98 %) were all obtained from Sigma Aldrich (Merck, Switzerland). Optima LC–MS grade water, methanol, acetonitrile, formic acid and acetic acid were obtained from Fisher Scientific (Switzerland). PM2.5 ambient samples were collected on 150 mm Pall Tissuquartz membrane filters (VWR, Switzerland). SOA samples were collected on 47 mm PTFE membrane filters with a 0.2 µm pore size (Whatman, Merck, Switzerland).

In this study laboratory-generated SOA and ambient samples were collected and characterized to cover a wide range of organic aerosol components.

Two precursors, β-pinene and naphthalene, representing natural and anthropogenic sources, were used to generate SOA particles via O3 and OH oxidation with a compact aerosol flow tube, the “organic coating unit” (OCU) (Keller et al., 2022). The detailed setup for SOA generation, concentrations and masses deposited onto the filters is presented in the Supplement (Fig. S1 and Table S1). Five filter samples were collected for each SOA type and storage condition to assess reproducibility. Prior to particle collection, each filter was cleaned in order to remove residual organic products from manufacture by rinsing with LC–MS-grade methanol and air-dried in the fume hood.

Ambient PM2.5 samples were collected with a Digitel DH-77 high-volume air sampler fitted with a PM2.5 inlet (Digitel, Switzerland). The urban sampling site was on the roof of a building at 20 m height above street level at Klingelbergstrasse 27, Basel, Switzerland. Prior to sampling, each quartz filter was baked out for 6 h at 550 ∘C in order to remove residual organics and to ensure reproducibility; cleaned filters were stored at -80 ∘C and were wrapped in aluminum foil and in an airtight plastic storage bag until use. High-volume ambient aerosol samples (HVASs) were collected at a flow rate of 500 L min-1 for 24 h. The exposure area of each filter was 169.7 cm2.

An overview of all samples collected and the time between collection and extraction and analysis is given in Table S2. All samples were stored in the dark and at temperatures of +20 ∘C (hereafter referred to as room temperature), -20 or -80 ∘C and were analyzed either immediately or after storage times of up to 44 d. Due to the large number of samples and LC–MS analyses it was not possible to analyze all samples after the exact same number of days.

The filter extraction of SOA and ambient samples differed due to the difference in the properties of the filter material, PTFE for SOA and quartz for ambient. The extraction procedure was partially adapted from the method described in Keller et al. (2022) and adapted for this study as described below for SOA samples. Deviations of the method for ambient samples are indicated in parentheses.

Each filter was cut into equal quarters (for ambient filter samples five 1 cm punches were used), placed into 2 mL Eppendorf safe-lock tubes (Eppendorf, Switzerland), and placed in a freezer (i.e., -20 or -80 ∘C) or extracted immediately. For extraction, 1.500 mL extraction solvent (1 : 5 water : acetonitrile (ACN) v/v) was added to the safe-lock tube using Eppendorf Research^® plus 200 and 1000 µL pipettes (Eppendorf, Switzerland), and then the samples were vortexed at maximum speed (2400 rpm) for 2 min each and placed on a Fisherbrand™ open-air rocker (Fisher Scientific, Switzerland) for 30 min (post-extraction, ambient samples were additionally put in a centrifuge for 10 min at 12 000 rpm to separate the quartz filter slurry from the liquid sample). A total of 1.500 mL of the sample extract was then pipetted into an empty Eppendorf tube to remove the filter material (for ambient samples 1.0 mL of the sample extract was transferred to an empty Eppendorf tube using a 5 mL gastight glass Hamilton syringe and a PTFE 0.45 µm pore size syringe filter (Agilent Technologies, Switzerland) in order to avoid larger particles from being transferred into the LC, a common source of blockages). The samples were then placed into a benchtop rotary evaporator (Eppendorf Basic Concentrator Plus, Eppendorf, Switzerland), and extracts were dried for 2 h at 45 ∘C in vacuum concentrator alcohol (V-AL) mode until complete dryness was reached; this process was conducted in batches where necessary. Samples were then reconstituted with 500 µL (ambient samples with 400 µL in order to further concentrate the samples) reconstitution solvent (1 : 10 ACN : water v/v) and vortexed again for 90 s before they were split into five aliquots of 100 µL (ambient samples: 80 µL) in amber LC–MS vials with 150 µL glass inserts. These were then either stored for the times stated in Table S2 or placed directly in the LC autosampler for analysis.

Liquid chromatography was conducted using the Thermo Vanquish Horizon UHPLC with a binary pump and split sampler (Thermo Fisher Scientific, Reinach, Switzerland). The Waters HSS T3 UPLC column (100 mm × 2.1 mm, 1.8 µm, Waters AG, Baden, Switzerland) was used at a temperature of 40 ∘C and a flow rate of 400 µL min-1. Water and 10 mM acetic acid (mobile phase A) and methanol (mobile phase B) were used as mobile phases at the following gradient in a 30 min method: 99.9 % A from 0–2 min, a linear ramp-up to 99.9 % B from 2–26 min; 99.9 % B was held until 28 min and was then switched to 99.9 % A for column re-equilibration from 28.1–30 min. To clean up between sample injections and to prevent carryover, a needle wash using 1 : 4 ACN : H2O (with 0.1 % acetic acid) was performed for 15 s prior to each sample injection. Additionally, a seal wash of 1 : 10 methanol : water (with 0.1 % formic acid) was used. To ensure system suitability, the stability of the signal intensities and retention times over multiple weeks of analyses, and batch correction where necessary, an HPLC gradient test mix injection consisting of phenol, uracil and a mixture of parabens (Sigma Aldrich, Merck, Switzerland) was run daily.

An Orbitrap Q Exactive Plus (Thermo Fisher Scientific, Switzerland) was used for mass spectrometric detection in negative electrospray mode. The following instrument parameters were used: spray voltage 3.5 kV, sheath gas flow 60 a.u., auxiliary gas flow 15, sweep gas flow 1, capillary temperature 275 ∘C and auxiliary gas heater temperature 150 ∘C. The scan parameters were set to full MS, a scan range of m/z 85 to 1000, a resolution of 70 000, an automated gain control (AGC) target of 3×106 and a maximum injection time of 25 ms. The mass spectrometer was calibrated daily using the Thermo Scientific Pierce Negative Ion Calibration Solution (Fisher Scientific, Switzerland). Additionally, a standard mix consisting of camphoric acid, cis-pinonic acid, 4-hydroxybenzoic acid, 1,2-naphthoquinone and pimelic acid was run at concentrations between 10 ng mL-1 and 0.01 mg mL-1 in order to obtain the calibration curves of compounds with atmospheric relevance and which were also used along with the HPLC gradient test mix in order to monitor the stability of the signal intensity and retention times (see Sect. S2 and Table S3). Cis-pinonic acid and 1,2-naphthoquinone were additionally used for annotation.

In total 810 (270 per sample type) LC–MS injections were run, including repeats and excluding blanks and conditioning runs. Raw data files were converted to mzML format using ProteoWizard (MSConvert, version 3) software (Chambers et al., 2012). LC–MS data analysis was performed in R 4.2.1 (R Core Team, Austria) in RStudio 2022.07.1 (Boston, MA, USA) using the XCMS package for untargeted peak detection (Smith et al., 2006; Tautenhahn et al., 2008; Benton et al., 2010) and the peakPantheR (Wolfer et al., 2021) package for targeted feature extraction. For the untargeted analyses, the XCMS centWave algorithm was used for peak detection on the centroided data in order to produce a table of m/z retention time (RT) pairs, henceforth referred to as features. All reported features are assumed to be the deprotonated (i.e., singly charged, [M–H]-) species unless otherwise indicated. Additional in-house scripts using R and Python were used for post-processing data analysis.

To observe variation and trends in the large datasets produced, principal component analysis was used, as this method easily illustrates the dominant sources of variation in multivariate data. Multivariate statistical analysis was performed with SIMCA^® 17 (Sartorius, Germany); model performance was evaluated using R2 values as a measure of proportion of variance explained by the model and using the Q2 value, which estimates the predictive power of the model through 7-fold cross-validation using randomly selected test/train subsets taken from the whole dataset. Hotelling's T2 statistic was used to estimate potential outlier samples in the principal component analysis (PCA) scores relative to the whole dataset using the multivariate probability distribution. The ggplot2 package (Wickham, 2016) in R was used to plot the PCA score plots from the SIMCA data. Python (Van Rossum and Drake, 2009) implemented in Spyder IDE 5.1.5 (Raybaut, 2009) with the Matplotlib (Hunter, 2007) and NumPy (Harris et al., 2020) packages was used for time series plots.

Error bars in the time series plots using the peak area represent the total relative uncertainty of ±20 %. This was calculated as the sum of the following individual uncertainties: the standard deviation of the UHPLC–MS injection repeats, which was 4 %; the standard deviation of the detected peak area for specific features of the filter sample repeats, which was 13 %; and the variation due to the filter extraction procedure, which was calculated from the immediately extracted samples and which was as high as 23 %.

β-Pinene was chosen as a representative biogenic precursor for SOA (Hallquist et al., 2009). In order to reduce the large number of total features detected and to remove potential interferences from non-informative noise and background peaks, a peak intensity filter was set to 7×105 (equivalent to 0.12 % of the highest peak intensity in the sample); hence only features with a peak intensity higher than this value were considered for further analysis. This led to 4735 features being detected for each of the 270 β-pinene SOA samples analyzed (excluding blanks); this figure is comparable with previous studies, with a similar number of features being detected in ambient PM2.5 samples using LC–MS characterization (Pereira et al., 2021).

The PCA score plot of principal components (PCs) 1 and 2 (Fig. 1) using non-normalized peak intensities shows that for samples stored as extracts and filters, the key parameter to ensuring stable sample composition over weeks was the storage temperature. The samples immediately extracted and analyzed on the day of collection represent the freshest samples available, and the tight clustering of these indicates the stability between different filters and the reproducibility of the aerosol generation and extraction. Both frozen sample types demonstrated little deviation in the multivariate space from the fresh samples, which confirmed the initial assumption that keeping both extracts and filters at cool temperatures best preserves the chemical profile for at least several weeks as represented by the peak intensity for SOA samples. For log⁡10(x) normalized peak intensities, the PCA score plot is presented in Fig. S3. The same trend can be seen with the exception of the extracts stored for 4 weeks, where the overall composition also starts to deviate significantly from the fresh samples.

In contrast, samples kept at room temperature drift away from the fresh samples in the PCA model, indicating a change in composition. Samples stored as filters or extracts at room temperature displayed a different behavior in PC1 and PC2 (PC1 giving the biggest variance for the filters and PC2 for the extracts). This suggests that there is a significant difference between samples that are extracted immediately and ones that are kept as filters at room temperature. For these room temperature samples, there is a clear temporal trend over the storage time of about 4 weeks: the longer the samples were kept at room temperature, the larger the deviation from the fresh samples (see also Fig. S2, displaying the storage time for each data point). Both the filters and the extracts stored at room temperature for 2 and 4 weeks surprisingly unveil signals outside of Hotelling's ellipse representing the 95 % limit of the multivariate probability distribution for the dataset, indicating that if the sample set was unknown, these samples might be qualified as outliers. The filters and extracts stored at room temperature seem to change their overall composition most significantly during the first days of storage because the biggest change per day seems to occur at the beginning of the storage time (Fig. S2).

The four most intense peaks in the base peak chromatogram (see Fig. S4) of the immediate extracts were chosen as representative of how the relative concentration of individual chromatographic peaks change over time under the different storage conditions. Figure 2 illustrates these temporal trends, sorted by retention time, where each point represents the average of two repeated analyses of each of the five filters collected (i.e., the average of 10 UHPLC–MS analyses). All four compounds or isomers of these have been identified in previous studies as carboxylic acids in SOA from gas-phase oxidation of α-pinene/β-pinene (Glasius et al., 2000; Yasmeen et al., 2010; Sato et al., 2016), and we tentatively confirmed the Mw 184 (detected as m/z 183.1027, C10H15O3) peak at 11.74 min to be cis-pinonic acid through comparison with an authentic standard.

The time series plots show a similar trend to the PCA results: the samples kept at -20 or -80 ∘C demonstrated the highest stability, where peak areas are also mostly within 25 % of the values detected in the freshly analyzed samples for Mw 172 (detected as m/z 171.0663, RT 6.73 min, C8H11O4), Mw 200 (detected as m/z 199.0976, RT 7.20 min, C10H15O4) and Mw 186 (detected as m/z 185.0819, RT 8.34 min, C9H13O4). This clearly indicates that storing the samples at -20 ∘C or below conserves samples sufficiently to prevent significant changes to these highest-intensity peaks.

In contrast, for features with Mw 172, 186 and 200, the extracts and filters at room temperature demonstrated noticeable increases over time (Fig. 2). This observation seems to contradict the hypothesis that compounds decay during storage. However, a possible explanation for this increase in these prominent features in the monomeric mass region might be a decomposition of oligomers (i.e., compounds with 11 or more carbon atoms). Since it is assumed there is limited oxidation chemistry occurring during storage, it is unlikely that the concentration of these compounds increased due to oxidation reactions, which is the dominant formation pathway of these compounds in the atmosphere. One class of oligomers frequently described in the literature is dimer esters (Hall and Johnston, 2012; Kenseth et al., 2018; Kristensen et al., 2016). The hydrolysis of dimer esters in samples stored in aqueous solution results in an increase in the intensity of the precursor monomers as decomposition products (i.e., compounds with 10 or fewer carbon atoms) (Zhao et al., 2018), which in our case would be the carboxylic acids discussed here. A time series analysis of the dimer ester Mw 388 (detected as m/z 387.0759, C18H28O9) (Kristensen et al., 2016) is given in Fig. S5. This compound showed a clear decrease over time for samples stored as extracts at room temperature, and it might therefore be one of the compounds decaying in the sample, causing the observed concentration increase in compounds presented in Fig. 2.

An exception to this trend is cis-pinonic acid (Mw 184, RT 11.74 min), which had little temperature dependency, but the signal dropped by about 75 % for the samples which were kept on the filters, whereas it remained relatively stable in immediately extracted samples. Previous studies have observed similar results, where cis-pinonic acid demonstrated different behavior in comparison with the rest of the dataset, i.e., with a desorptive loss upon purging spiked filters with clean air (Glasius et al., 2000) or a decrease in acetonitrile and an increase in water over time (Wong et al., 2021).

Overall, the results for β-pinene SOA demonstrate that samples, both extracts and filters, kept at temperatures of -20 ∘C or below exhibited good stability of signal intensity over time, emphasizing that for studies conducting detailed offline analysis of SOA, composition samples should immediately be frozen after collection until analysis. However, these results also indicate that at least some compounds change over time, even under these low-temperature storage conditions, and the impacts of these artifacts on quantitative and compositional analyses must be considered. For samples kept at room temperature, there were clear and significant temporal changes in the signal intensity for many features as illustrated in Figs. 1 and 2, and samples stored for a day or longer at room temperature before analysis should not be considered for detailed chemical characterization.

Naphthalene SOA (a representative anthropogenic aerosol; Eiguren-Fernandez et al., 2004) samples were analyzed analogously to the β-pinene samples. A total of 5640 peaks with an intensity higher than 7×105 (equivalent to 0.19 % of the highest intensity in the sample) were detected in 269 analyses. The PCA score plot for naphthalene SOA in Fig. 3 displayed similar overall trends using the non-normalized peak intensities as for β-pinene SOA (see Fig. 1). The generation of naphthalene SOA particles in the flow tube is slightly more unstable than for β-pinene; therefore the spread of fresh samples was higher across the five filter repeats as compared with the β-pinene samples. Similar to β-pinene SOA, the naphthalene samples kept frozen at -20 or at -80 ∘C exhibited closer profiles to the immediately analyzed samples and deviated little beyond the spread of the freshly extracted samples in the PCA model. For the room temperature samples there was a clear trend of variation associated with storage time for the extracts, which showed the largest variation in PC1, and for the filters, which showed the largest variation in PC2. This similarity with the β-pinene samples again indicates that the overall composition of the SOA samples stored for 2–4 weeks at room temperature deviated significantly from the immediately analyzed samples and that the influence of the extract and filter storage results in very different compositional changes. The samples kept at room temperature for 2–4 weeks fell outside Hotelling's T2 ellipse (see Fig. S6), again indicating that relative to the other samples, they have differing profiles and much larger variance across their features. The PCA score plot with log⁡10(x) normalized peak intensities is given in Fig. S7 and shows the same trends for all three temperatures as the score plot for the non-normalized peak intensities.

These trends were also visible for the four most intense peaks in the base peak chromatogram of naphthalene SOA samples as presented in Fig. 4. Again, the most stable storage conditions were freezing of the samples, and extracted samples indicated a slightly improved temporal stability over the samples stored on filters in the freezers. The most noticeable changes occurred for samples kept at room temperature. The most significant decay over time at room temperature was seen for Mw 158 (detected protonated anion of Mw 158: m/z 159.0451, C10H7O2) at 13.26 min, which was identified as 1,2-naphthoquinone through comparison with an authentic standard. For extracts, the signal intensity dropped to less than half in the first 24 h, before disappearing completely in the samples analyzed after 1–4 weeks, and it appeared stable only when stored at -80 ∘C as the extract. 1,2-Naphthoquinone is of increasing interest in the literature, as oxidized polycyclic aromatic hydrocarbons (PAHs) are known to cause oxidative stress in human lung cells and are thus a direct contributor to particle toxicity from anthropogenic sources (Kelly, 2003). It is evident from this data that particle extraction and storage conditions need to be carefully described and considered when these compounds are used for source apportionment or to infer particle health effects from laboratory-generated samples.

Mw 166 (detected as m/z 165.0192, RT 6.83 min, C8H5O4) has previously been found in naphthalene SOA samples and identified as phthalic acid (Kleindienst et al., 2012). The most stable conditions for this compound were again observed when samples were kept frozen, while in extracts stored at room temperature, this compound steadily increased to almost double the intensity after a month. A possible explanation for the increase in especially Mw 166 could again be the decay of oligomeric compounds, causing an increase in their monomeric counterparts.

The other isomers of Mw 210 (detected as m/z 209.0455, RT 5.72 min, C10H9O5) and Mw 150 (detected as m/z 149.0243, RT 7.50 min, C8H5O3) selected for Fig. 4 showed moderate changes in comparison with the previously discussed compounds. Both exhibited relatively little change over time in samples which were kept in the freezers. The largest time-related effect can be seen for the samples kept at room temperature, where there is either a decrease (Mw 210) or an increase (Mw 150) of around 40 % after 4 weeks.

These four most intense peaks contributed the most to the variance observed in the PCA score plots, thus driving the separation of samples by storage condition, and again reinforce the requirement to store organic aerosol samples in a freezer to best preserve their original composition.

To assess if the significant temporal trends and the differences in storage (i.e., filter vs. extracts) observed for laboratory-generated SOA samples were also visible in ambient samples, we collected five high-volume ambient aerosol samples in the city center of Basel and analyzed, extracted and stored them using the same methods as for the laboratory-generated SOA samples.

The PCA score plot with non-normalized peak intensities for the ambient samples is given in Fig. 5 (the score plots with log⁡10(x) normalized peak intensities are given in Figs. S9 and S11, showing the same trends). The colors represent the five different HVASs, and shapes correspond to storage temperatures. More detailed information on storage temperature and type is given in separate score plots in the Supplement (Figs. S8 and S10). During LC–MS analysis of the ambient samples, a different batch of UHPLC-grade water from the supplier was needed for the samples stored for 3–4 weeks, causing higher background signals and a reduced overall signal intensity for peaks with lower intensities. This difference in signal intensity could be adjusted in the time series analysis of the compounds previously detected in SOA samples through the intensity of our standard mix but was difficult to account for in the PCA. In order to solve this problem, the peak intensity parameter was increased from 7×105 (as used for the SOA samples discussed above) to 4×106 (equivalent to 4 % of the highest peak intensity in the sample) to reduce the number of total compounds detected from 2800 to around 400 because the higher intensity peaks were not significantly affected by this increased background. Additionally, a time series of the signal intensity of individual compounds was checked manually to exclude compounds which had a clear “step function”, leaving roughly 240 compounds to be included in the PCA. The non-corrected version of the score plot is given in Fig. S12, where the same general trend is still visible as in Figs. 5, S8 and S10.

In strong contrast to the laboratory-generated SOA samples, the PCA score plots for the ambient samples indicated little storage-dependent variation in the signals, as samples grouped together in the first two PCs independently of storage temperature or condition, indicating a much larger influence of individual samples on the variance than of the storage condition. The HVASs from days 3–5 showed similar scores, as they were all sampled in the same week or even on consecutive days. To ensure that there was no additional variation between the storage temperatures, we also looked at PCA score plots of the individual HVASs, which presented the same trends (data not shown).

We conclude that in ambient samples the concentration of organic components is overall more stable over time and is apparently less affected by storage conditions compared with laboratory-generated SOA samples. This could be due to several factors. Organic components in ambient particles originate not only from SOA sources but include many primary particle components from other sources such as biomass burning, fossil fuel combustion, industrial activities (e.g., solvents) and primary biological material (Seinfeld and Pankow, 2003). Components from these sources might be more stable than SOA components. In addition, in ambient samples, a significant fraction of the total particle mass is inorganic components (mainly ions like sulfate, nitrate and ammonium), resulting in a more diluted concentration of individual organic components (compared with pure laboratory-generated SOA samples), which might limit the availability of organic reaction partners, and thus increasing the stability of some organic components.

For a compound-specific comparison between SOA and ambient samples, we analyzed four compounds which were detected in all HVASs, and which were also among the four highest peaks in the SOA samples (see Figs. 2 and 4). The time series for these compounds in the HVASs is given in Fig. 6. HVAS 1 was excluded from this analysis because of the missing 2-week time point (Table S2). Overall, these compounds were more stable over time in the ambient samples compared with the pure SOA samples, as also indicated in the PCA analysis, supporting the hypothesis that the lower concentrations of individual organic compounds in ambient aerosol lead to less signal change over time. This increased stability might also be due to the lower oligomer content in ambient aerosol in comparison with laboratory-generated SOA (Kourtchev et al., 2016). Nevertheless, clear changes were observed for Mw 166 and 186 for samples stored at room temperature and as extracts, which showed similar patterns to ambient and pure SOA samples. Slight changes over time (especially after 4 weeks) were seen for the Mw 172 feature in the room temperature samples. The largest difference between ambient and laboratory SOA samples was observed for cis-pinonic acid (Mw 184), where there was no significant difference between filter and extract storage in the ambient samples, but a large decay occurred in the pure SOA samples stored on filters. Reasons for this very different behavior are unknown but could be related to the different filter material used for ambient and lab samples (quartz vs. PTFE). Another cause could be desorptive loss of cis-pinonic acid due to the large air masses in the HVAS as previously reported (Glasius et al., 2000).

Overall, the storage of ambient samples on filters demonstrates very good stability of the signal intensity and provides confidence that the concentration of organic components may not change significantly in ambient urban samples which are collected weeks before analysis and which are stored on filters.