2.1 Instrument description

The GC-TOF-MS system consists of three main components: (1) a thermal desorption preconcentrator (TDPC) for sample collection, (2) a gas chromatograph (GC) for sample separation, and (3) high resolution time-of-flight mass spectrometers (HR-TOF-MS) for sample detection. Each of these components is described in the following sections. While the in situ GC can be operated with either the Vocus PTR-TOF-MS or EI-TOF-MS as individual detectors, coupling the GC with both detectors creates a technique that produces three complementary data sets: (1) real-time Vocus PTR-TOF-MS, (2) GC-Vocus PTR-TOF-MS, and (3) GC-EI-TOF-MS. Hereafter, these three techniques will be referred to as RT-Vocus, GC-Vocus, and GC-EI-TOF, respectively. It should be noted that the instrument described in this work and deployed for the ATHLETIC campaign was a prototype system used to demonstrate this technique. The instrument has been undergoing continued development to improve sensitivities and chromatographic performance and to extend the volatility range of resolved compounds since this campaign.

2.2 Thermal desorption preconcentration

Ambient VOCs are typically present at low mixing ratios (sub-ppb), and thus to increase GC-MS sensitivity a preconcentration method is required. For this study, the samples were collected using a simplified version of a thermal desorption preconcentrator (TDPC) (Aerodyne Research, Inc.). Briefly, the TDPC employed for this study relied upon a single-stage adsorbent trap for preconcentration of analytes. The design is based upon that of Tanner et al. (2006), and uses a commercial cold-plate Peltier thermoelectric cooler (CP-110, TE Technology) to allow for precise ambient to sub-ambient temperature regulation. Results from this TDPC have been described previously (Anderson et al., 2019). The system was simplified by not using water trapping or oxidant scrubbing before sample collection due to the expected low humidity and oxidant mixing ratios in this study. The sample trap was a commercial glass sorbent tube (TO-15/TO-17 cold trap, Markes International) operated at 20 ^∘C during sample collection to avoid potential water condensation. The chosen sample trap was a multi-bed adsorbent trap equipped with three stages of adsorbents (Tenax, Carbopack X, Carboxen 1003; Markes International, personal communication, 2020) to expand the volatility range of compounds that can be trapped and desorbed for analysis. The combination of adsorbents in the TO-15/TO-17 trap allows for the analysis of a wide range of VOCs (including oxygenates) in the C₂–C₃₂n-alkane volatility range. However, for the system deployed for this work, the instrument was optimized for VOCs in the C₅–C₁₂ volatility range. Details of operational parameters (e.g., temperatures, flows) are described in Sect. 2.8.

2.3 Gas chromatograph

To separate analytes before detection with TOF-MS, a compact GC from Aerodyne Research Inc. (hereafter referred to as ARI GC) was used. The ARI GC is designed to be an in situ, field-deployable system. It fits into a 55 cm × 55 cm × 30 cm rack, weighs 24 kg, consumes 300 W of power during typical operation, and contains all hardware for GC sample collection and control of TDPC and GC flows and temperatures, including a make-up flow needed for GC-Vocus measurements (described in Sect. 2.8). Here, the flow path contained three two-position chromatography valves with Nitronic 60 valve bodies (VICI Instruments): one 10-port and two 6-port valves (Fig. 1b) to direct flows during the GC cycle. The chromatography valves and transfer lines (Sulfinert-treated 304-SS, 1.6 mm OD, 0.76 mm ID, Restek) are housed in a heated enclosure held at 150 ^∘C. The carrier gas (UHP helium; Matheson) was controlled by a mass flow controller (MKS Technology) with variable set-point capability in the range of 0.1–10 cm³ min⁻¹. The GC column is housed in a custom interlocking aluminum spindle (12 cm × 3 cm) with surface-mounted flexible resistive heaters, as described by Lerner et al. (2017). For this study, the ARI GC was configured as a one-channel system (single-column separation), with a 30 m Rxi-624 analytical column (Restek, 0.25 mm ID, 1.4 µm film thickness) installed in the spindle. This column resolves non- to mid-polarity VOCs including hydrocarbons, oxygenates, and nitrogen- and sulfur-containing compounds, with the exception of high-polarity compounds like carboxylic acids. The volatility range that the GC can resolve is a function of both the chosen GC column and the TDPC adsorbent trap. With the combination of column and adsorbent trap used for this study, the ARI GC was optimized for C₅–C₁₂ hydrocarbons, along with oxygen-, nitrogen-, halogen-, and sulfur-containing VOCs.

Figure 1Instrument schematics of (a) dual-detector GC-TOF-MS instrument configuration with valving shown for GC detector selection (EI-TOF or Vocus) and Vocus inlet source (room or supply air or GC effluent). (b) GC flow path and valve positions to incorporate a single-stage thermal desorption preconcentrator (TDPC), single column separation, and dual TOF-MS detection.

Download

2.4 HR-TOF-MS detection

2.5 Sample acquisition, separation, and detection

At the start of the GC cycle (22 min), the sample was collected onto the adsorbent trap held at 20 ± 1 ^∘C for 6 min at 30 cm³ min⁻¹. The adsorbent trap was then backflushed for 1 min with 2 cm³ min⁻¹ of UHP helium (Matheson) to remove oxygen and water from the trap. Next, the carrier flow was increased to 5 cm³ min⁻¹ and the sample was thermally desorbed onto the GC column by flash heating the adsorbent trap to 225 ^∘C for 20 s at 10.5 ^∘C s⁻¹. During this sample transfer, the GC column was held at 40 ^∘C. To gauge the desorption efficiency, we would run a sample and then an instrument blank, with no sample flow through the trap during the collection period, to measure the residual sample remaining in the trap. The result of the instrument blank was <1 % of the signal measured for all reported compounds in the sample, indicating highly efficient transfer of sample, and this was deemed acceptable.

The method of chromatographic separation was as follows: after sample transfer the column temperature was held at 40 ^∘C for 40 s, then ramped from 40 to 100 ^∘C at 40 ^∘C min⁻¹, increased to 150 ^∘C at 15 ^∘C min⁻¹, and finally increased to 225 ^∘C at 30 ^∘C min⁻¹ and held at 225 ^∘C for 100 s for a total chromatogram time of 9 min. Following the separation, the GC column was backflushed with UHP helium for 140 s, while being held at 235 ^∘C and the sample trap was switched out of the path and backflushed with 2 cm³ min⁻¹ UHP He while heated to 225 ^∘C for 20 s to prepare the adsorbent trap for the next sample collection. After this cleaning period, the column was cooled and held at 40 ^∘C over 575 s with continuous backflushing until the next sample transfer. Column flow rate was 2 cm³ min⁻¹ of UHP helium controlled by a mass flow controller during sample transfer and chromatography, and by a critical orifice during backflushing.

After the chromatographic separation, the column effluent was automatically directed to either the EI-TOF or the Vocus for detection by switching the 3rd chromatography valve (Fig. 1). The ARI GC was coupled to the EI-TOF by a fused-silica capillary line (Siltek guard column, 0.25 mm ID, Restek) which passed through a heated capillary feed-through, kept at 230 ^∘C, so that the GC effluent was directed into the ionization region of the EI-TOF. For GC-Vocus measurements, the ARI GC was coupled to the Vocus by a passivated stainless steel (Sulfinert-treated, 1.6 mm OD, 0.76 mm ID, Restek) transfer line heated to 150 ^∘C. For GC-Vocus measurements, the total flow provided from the GC must be adequate for the Vocus ambient pressure inlet, which is fixed at 100 cm³ min⁻¹ via a PEEK capillary tube and pressure controller. Since the GC column effluent flow rate is 2 cm³ min⁻¹, an additional make-up flow of 100 cm³ min⁻¹ of zero-grade air (Airgas) was introduced upstream of the GC-Vocus transfer line, downstream of the GC column (Fig. 1b).

2.6 Normalization of instrument response and calibration

Normalization of the EI-TOF data is required to account for changes in instrument sensitivity due to changes in detector response. For this study a normalization method that utilizes the detection of long-lived halocarbons in the atmosphere was used, as described previously by Lerner et al. (2017). Specifically, ambient carbon tetrachloride (CCl₄) was used here to normalize the EI-TOF data, as no sources were expected in the weight room, consistent with a lack of significant variations in its time series. A normalization factor (NF) was calculated for each sample by dividing the GC peak area of CCl₄ by the average CCl₄ peak area for the entire campaign. The EI-TOF data was then corrected by dividing the raw data by the NF.

The Vocus signal is dependent on the concentration of the analyte, time spent in the ion-molecule reactor, the rate of reaction, and the concentration of the reagent ion (Yuan et al., 2017). Variability of analyte signal is reduced by normalizing to a constant reagent ion signal of 1×10⁶ counts per second (cps). During the ATHLETIC campaign, the largest signal observed in the Vocus was (H₂O)₃H⁺; actual ion concentrations in the reactor were dominated by H₃O⁺ ions, but these are poorly measured due to mass discrimination in the quadrupole ion guide between the reactor and time-of-flight analyzer. Analyte signal was divided by (H₂O)₃H⁺ and multiplied by 1×10⁶ cps to obtain normalized signal. This method of normalization depends on the E∕N value used, which for this study was 150 Td. Monoterpene calibration of the RT-Vocus signal was accomplished immediately following the campaign, where a 6-point calibration of a gravimetric standard of limonene was diluted in zero air, resulting in a calibration factor of 139 ncps (normalized counts per second)/ppbv for C₁₀H${}_{\mathrm{17}}^{+}$. The validity of using a single monoterpene to generate a general sensitivity for the RT-Vocus to represent the mixture of monoterpenes present in ambient air will be discussed in Sect. 3.3.1. Backgrounds for the Vocus were obtained using a zero-air generator for 30 s every 15 min and subtracted with linear interpolation between background collection periods (Krechmer et al., 2018).

At the end of the campaign both the GC-EI-TOF and GC-Vocus were calibrated for aromatic compounds and monoterpenes. This was accomplished by performing a 3-point calibration curve with duplicates for each detector using a 3 L Tedlar bag (Restek) containing 2 ppb each of BTEX (benzene, toluene, ethyl benzene, m-, p-, and o-xylene) and five monoterpenes. The calibration sample was made by diluting a standard solution of BTEX (Restek) with methanol (HPLC grade, Sigma-Aldrich) and creating a solution of five standard monoterpenes (α-pinene (99 %), β-pinene (99 %), camphene (96 %), carene (97 %), and limonene (99 %) (Sigma-Aldrich)) by diluting in hexanes (HPLC grade, Sigma-Aldrich). Using a glass micro-syringe, 5 µL of each of these solutions was then injected into a stream of UPH N₂ (Airgas) flowing into the Tedlar bag at nominally 500 cm³ min⁻¹. This mixture was attached to the sample inlet of the GC for calibrations. For the GC calibrations the sampling rate of the GC was kept constant (30 cm³ min⁻¹), but the collection time was varied (1, 3, and 6 min of collection) to generate the calibration curves. This created a curve that gave instrument response (normalized counts; ncts) versus sample volume. By dividing the response at each sample volume by the compound concentration in the Tedlar bag, the instrument sensitivity (ncts ppb⁻¹) was calculated. The difference in units between the RT-Vocus and the GC data should be noted; the RT-Vocus data is reported as normalized counts per second (ncps), while the GC data is given as normalized counts (ncts), the integration of detector response across the peak elution time. Figure S1 in the Supplement shows calibration curves of GC-TOF instrument sensitivity versus sample volume of selected monoterpenes. The limits of detection (LOD) were calculated as 3 times the standard deviation of the baseline multiplied by the full width at half maximum (FWHM) of the chromatographic peak and then divided by the sensitivity. Calibration results for both the GC-EI-TOF and GC-Vocus, including ions used for quantification, instrument sensitivity, LODs, and correlation coefficients (R²), are given in Table 1. From our calibration data, we estimate typical 1-σ uncertainties to be 12 % and 5 % for the GC-EI-TOF and GC-Vocus configurations, respectively, with typical precisions of 5 % and 1 %. The individual uncertainties for each calibrated compound reported from the GC are listed in Table S1. Available RT-Vocus sensitivities measured for this campaign are also reported in Table 1 for comparison with the GC sensitivities. The ambient GC data was converted to mixing ratio by dividing the normalized signal with the sensitivity. GC-Vocus chromatograms of both calibration and ambient indoor air are shown in Fig. S2 to demonstrate the chromatographic separation of this system.

Table 1Quantification ions, measured sensitivities (normalized counts per ppb; ncts/ppb), limits of detection (ppt), and linearity (R²) for compounds used for calibration of both the GC-EI-TOF and GC-Vocus. RT-Vocus sensitivities (normalized counts per second per ppb; ncps/ppb) for a subset of the compounds are included for comparison.

^a The ncps and ncts for both RT- and GC-Vocus measurements are for a Vocus operation with E∕N = 150 Td. ^b Sensitivity and LOD for a sample volume of 180 cm³, the volume used for GC-TOF ambient sampling during the ATHLETIC campaign. ^c LOD calculated as 3 times the standard deviation of the baseline multiplied by the full width at half maximum (FWHM) of the chromatographic peak divided by the sensitivity.

Download Print Version | Download XLSX

3.1 Benefits of dual-detector system and instrument performance

The novelty of this system is the ability to produce three complementary data sets: (1) RT-Vocus, (2) GC-Vocus, and (3) GC-EI-TOF during routine operation. As shown in Fig. 1a, this setup allowed the Vocus to sample in real time from the weight room, the supply air, or the GC effluent. With this instrument configuration, the Vocus can make real-time measurements with fast time resolution (<1 Hz) or automatically switch to GC detection for molecular speciation. When the Vocus was not sampling from the GC, the column effluent was instead sent to the EI-TOF for detection (shown in Fig. 1a), which allowed continuous coverage of the GC identification measurements.

Another benefit of this system is that the two detectors use different ionization methods: proton transfer reaction (PTR) versus electron ionization (EI). By alternating between the detection methods, two sets of chromatograms were created with different types of information about each molecule (which had the same GC retention time regardless of the detector). For GC-Vocus chromatograms, the analytes are ideally detected as their protonated molecular ion [MH⁺]. The extent to which this is true depends on instrument operating parameters such as the reduced electric field (E∕N), where greater ratios tend to induce fragmentation due to increased collisions while also limiting the formation of cluster ions (Yuan et al., 2017). For GC-EI-TOF chromatograms, the analytes are detected by their ion fragments and identified with their EI fragmentation pattern. Benefits of EI detection, compared to PTR, is compound identification through fragmentation pattern matching (e.g., NIST/EPA/NIH mass spectral library, Linstrom and Mallard, 2020, https://webbook.nist.gov/chemistry/, last access: 23 December 2020) and the ability to detect compounds such as saturated hydrocarbons, which have a proton affinity that is too low for detection by PTR-MS (Yuan et al., 2017). However, since the Vocus generally detects an intact molecular ion, this can lead to a simpler analysis, where each compound is ideally detected as a single ion (giving the molecular formula of the protonated parent) rather than a series of fragments as with EI detection.

Calibration results (measured sensitivities, LODs, and R² values) from both the GC-EI-TOF and GC-Vocus are reported in Table 1. A quantification ion was chosen for each method: for GC-Vocus this was the protonated molecular ion [MH⁺], and for GC-EI-TOF the most abundant fragment ion present in the mass spectrum was used. These ions were chosen because they generally result in the highest sensitivity due to their abundance; however, in select circumstances (e.g., interference such as co-elution) another ion may be chosen. The comparison of GC-EI-TOF and GC-Vocus sensitivities and LODs is a comparison of detector response since GC operation was identical for calibration of each detector. EI-TOF detection was on average 4.3 times more sensitive than detection by the Vocus. This increase in sensitivity implies higher ion counts at the detector for the EI-MS versus PTR-MS; this may be attributed to several parameters including ionization efficiency (Cappellin et al., 2012; Sekimoto et al., 2017; Harland and Vallance, 1997).

Available RT-Vocus sensitivities are also reported in Table 1 for comparison with those measured by the GC-Vocus. For the limited overlap of calibration compounds presented in this work, the GC-Vocus was on average 50 times more sensitive than the RT-Vocus. The gained sensitivity with the addition of the GC is due to differences in sample volume. When the Vocus is operating in the RT-Vocus mode, it samples ambient air at 100 cm³ min⁻¹ with 1 s resolution, and thus each data point is representative of 1.7 cm³ of ambient air. However, sensitivity is gained with the addition of the GC due to preconcentration of sample, where 180 cm³ of ambient air is preconcentrated over the 6 min sample collection period before injection into the Vocus with typical peak widths here of ≤ 2 s (FWHM). The measured 50-fold increase in sensitivity agrees very well with the expected increase of a factor of 53, which is calculated by dividing the 180 cm³ sample preconcentrated by the GC and injected into the Vocus over a 2 s wide chromatographic peak (90 cm³ s⁻¹) by the 1.7 cm³ s⁻¹ analyzed by the RT-Vocus. The factor of 50 increase in sensitivity is due to the sampling schemes employed for this study. However, RT-Vocus LODs can be improved by averaging the signal to reduce noise. Figure S3 shows the Allan variance plot for the RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal (protonated molecular ion of monoterpenes) during a relatively unperturbed, low-concentration sampling period of room air. The Allan variance plot shows a broad minimum around 250 s, indicating the maximum period of effective sample averaging; assuming Poisson statistics for the data, this averaging window reduces RT-Vocus noise by a factor of ∼16. This analysis shows that averaging the RT-Vocus data beyond 250 s would not decrease the noise further, as the RT-Vocus signal no longer follows Poisson statistics for these larger time periods, likely due to environmental factors such as changes in instrument temperature.

The GC LODs reported in Table 1 are a function of the standard deviation of the baseline surrounding the chromatographic peak, chromatographic peak width, and instrument sensitivity. Across these compounds, the LODs for each instrument are very similar, each averaging 1.6 ppt for a 180 cm³ sample (GC-EI-TOF LOD${}_{\mathrm{avg}}=\mathrm{1.6}\pm \mathrm{1.1}$ ppt; GC-Vocus LOD_avg = 1.7 ± 1.2 ppt). As discussed in Sect. 2.8, the GC-Vocus measurements include a 100 cm³ min⁻¹ make-up flow for instrument operation. For this study, ultra zero air was used for the make-up flow, as a gas with low purity will create elevated baselines and negatively impact the instrument LODs.

For each detection method, the differences in sensitivity and LODs between isomers (e.g., α-pinene versus limonene) are primarily a function of the extent to which the compound fragments. The greatest sensitivity would occur if the ionization of an analyte resulted in zero fragmentation, so that all analyte signal was associated with a single ion. Instead, if a compound fragments, the signal associated with that analyte is spread out across multiple ions, taking signal away from the ion used for quantification and decreasing the sensitivity (and increasing the LOD). In the α-pinene versus limonene example given above, α-pinene fragments less than limonene, resulting in a greater sensitivity (and lower LOD) to the α-pinene monoterpene isomer.

3.2 Molecular identification of VOCs in an indoor environment

The GC data set was extensive and included detection of hydrocarbons, oxygenates and halocarbons in the volatility range of C₅–C₁₂ n-alkanes. Table 2 reports the EI characteristic ion and the ion(s) detected by the Vocus (typically a combination of the molecular ion [MH⁺], along with water clusters and/or fragments) for a subset of the chromatographic peaks that were identified through the GC analysis. These identified species include alcohols, aldehydes, ketones, ethers, nitrogen-containing compounds, halocarbons, siloxanes, alkanes, alkenes, and aromatics. It should be noted that not every compound listed in Table 2 can be reported quantitatively from the GC system due to breakthrough in the thermal desorption trap or other losses in the system. However, even for these species that are difficult to quantify, the GC is an excellent tool for compound identification. Specific results from the GC-TOF-MS system are reported in detail in Sect. 3.3 for monoterpenes and in Sect. 3.4 for dimethylsilanediol.

Table 2Compounds identified during the ATHLETIC campaign using the GC analysis with both EI-TOF-MS and Vocus PTR-TOF-MS detection.

Download XLSX

Table 2 presents three types of information about each molecule resolved by the GC: the retention time (which is a function of its vapor pressure, polarity, and functionality) and mass spectrometric response from both the EI-TOF and Vocus. One way to identify compounds is to analyze standards, where each standard compound is injected into the GC to directly measure the analyte retention time and detector response. However, doing this analysis for each compound present in ambient air is time-consuming and may not be feasible as some compounds are not available for purchase as analytical standards. Alternatively, retention time indices along with the mass spectrometric data can be used to confidently identify compounds without authentic standards. For example, a compound that eluted from the GC with retention time 369 s was detected in the Vocus as both C₇H₆OH⁺ and C₇H₈O₂H⁺, which are formulas that could correspond to the protonated molecular ion of either benzaldehyde or methyl-benzenediols, respectively. However, the EI fragmentation pattern of this compound showed large signal at C₇H₆O⁺ and C₆H${}_{\mathrm{5}}^{+}$, which matches the pattern expected for benzaldehyde (C₇H₆O⁺, molecular ion; C₆H${}_{\mathrm{5}}^{+}$, due to loss of the aldehyde group). This information is enough to infer that the peak is benzaldehyde and that the ions detected by the Vocus were the protonated molecular ion (C₇H₆OH⁺) and its water cluster (C₇H₆O[H₂O]H⁺). However, this assignment can be made unambiguous by also comparing the retention time and retention index to other compounds present in the chromatogram (Van den Dool and Kratz, 1963; Linstrom and Mallard, 2020, https://webbook.nist.gov/chemistry/, last access: 23 December 2020). The compound retention time of 369 s is between those of o-xylene (311 s) and limonene (374 s), which have retention indices of 890 and 1030, respectively (Linstrom and Mallard, 2020, https://webbook.nist.gov/chemistry/, last access: 23 December 2020). Because benzaldehyde has a retention index of 960, we can predict it would elute between o-xylene and limonene, whereas methyl-benzenediols have retention indices of ≈ 1200, meaning these would elute after limonene. With this method, we can use the combination of mass spectrometric data from the Vocus and the EI-TOF, along with the chromatographic retention time, to definitively identify compounds. Most species were positively identified using the workflow demonstrated for benzaldehyde; however, some compounds were only detected by a single detector. Unsaturated alkanes (e.g., n-hexane) and halocarbons (e.g., carbon tetrachloride, bromoform) were only detected by the EI-TOF, which was expected as these classes of compounds have proton affinities that are too low for detection by PTR-MS methods (Yuan et al., 2017).

Aside from using the GC to aid in compound identification, the GC data can also be used to characterize the ion chemistry occurring in the Vocus. From the information reported in Table 2, it can be seen that classes of compounds showed similar responses in the Vocus (e.g., forming single or double water clusters (M[H₂O]H⁺ or M[H₂O]₂H⁺) or undergoing fragmentation). Both results are non-ideal as they complicate the mass spectrum and make quantification and interpretation of the PTR-MS signals more difficult. Alcohols, aldehydes, ketones, siloxanes, and an unknown nitrogen-containing compound all formed significant water clusters within the Vocus ion-molecule reactor, while alkenes and aromatics demonstrated fragmentation. Classes of compounds that were exclusively observed as their protonated molecular ion in the Vocus include ethers, large aromatics (e.g., C₉ aromatics), and some nitrogen-containing compounds. The extent to which species form clusters or undergo fragmentation in the PTR-MS is a function of instrument operational parameters like the E∕N ratio (operated here at 150 Td). Although these interferences complicate the RT-Vocus interpretation, with the addition of GC separation the molecular identification and identification of fragment and cluster signals it is possible to perform for complex ambient samples. Furthermore, by using GC separation to quantify the ratio of fragmentation or cluster formation to the protonated molecular ion, the RT-Vocus measurements can be constrained and more easily interpreted.

Figure 2RT-Vocus detection of C₁₀H${}_{\mathrm{17}}^{+}$ with occasional speciation with GC-Vocus analysis. The RT-Vocus trace only contains room air measurements for simplicity. Pie charts show the GC-Vocus speciation of the RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal into six resolved monoterpene species during each of the highlighted (grey) sampling periods. The pie chart fractions represent the contribution (in concentration, ppb) of each monoterpene isomer to the total monoterpene concentration measured by GC-Vocus. Some specific events are numbered and highlighted in light blue with the following distinctions: (1) exercise session, (2) in flow of outdoor air, (3) football game, and (4) post-game activities.

Download

3.3 Monoterpenes in an indoor environment

3.3.1 Quantitative assessment of monoterpene detection

A subset of the C₁₀H${}_{\mathrm{17}}^{+}$ time series, the monoterpene protonated molecular ion measured by RT-Vocus, is shown in Fig. 2. During this period, GC-Vocus and GC-EI-TOF chromatograms were acquired and used to resolve individual monoterpene isomers. The separation of the C₁₀H${}_{\mathrm{17}}^{+}$ RT-Vocus signal into six different monoterpenes by the GC-Vocus is shown as pie charts in Fig. 2, where the pie chart fractions represent the contribution of each isomer (by concentration, ppb) to the summed concentration of all monoterpene isomers resolved by the GC-Vocus. As mentioned above, when the GC effluent was not being sent to the Vocus it was directed to the EI-TOF. Figure S4 shows the GC-EI-TOF time series of monoterpenes during the same sampling period shown in Fig. 2. During these sampling periods, limonene accounted for 47 %–80 % of the measured monoterpene composition due to activities occurring in and near the weight room. Details of the temporal behavior are discussed in Sect. 3.3.2.

The RT-Vocus, GC-Vocus, and GC-EI-TOF monoterpene data sets are shown together in Fig. 3. The RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal, along with the speciated monoterpenes (and their sum) resolved by GC-Vocus and GC-EI-TOF, are overlaid in Fig. 3a and b, respectively. The GC-EI-TOF monoterpene sum (Fig. 3b) agrees within a factor of 1.2 on average with the RT-Vocus signal across the entire measurement period. The GC-Vocus monoterpene sum also agrees within a factor of 1.2 with the RT-Vocus signal on 17 November; however, the agreement was not as good (only within a factor of 3) on 16 November. The better agreement on 17 November is due to the monoterpene composition being 70 % limonene on average, which was the isomer used to calibrate the RT-Vocus monoterpene signal.

Previous studies have found that comparing GC speciation with online PTR-MS measurements may result in discrepancies (de Gouw et al., 2003; de Gouw and Warneke, 2007; Kari et al., 2018; Kim et al., 2009). For example, Kari et al. (2018) demonstrated errors of 26 % in PTR-MS ambient measurements when the terpene composition was not accounted for, and instead a calibration factor determined from one isomer was used to represent the mixture. To avoid these errors, they urge PTR-MS users to use complementary methods (e.g., GC) to identify the speciated terpene composition and calibrate the total signal from PTR-MS using this speciation. The findings from Kari et al. (2018) are supported by the results shown here, where the best agreement between the speciated (GC) and online (RT) measurements are when the monoterpene composition is dominated by the isomer that was used to calibrate the RT-Vocus. When the mixture includes significant fractions of multiple isomers, the bias in the RT-Vocus measurement increases. However, if RT-Vocus calibration factors for monoterpenes other than limonene were measured, the RT-Vocus signal could be weighted according to the continuous GC-EI-TOF speciation measurements and corrected for the mixed monoterpene composition.

Figure 3Quantitative speciation of RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal (grey trace) into resolved monoterpene isomers by (a) GC-Vocus (b) GC-EI-TOF. Panels (c) and (d) show the same GC-Vocus and GC-EI-TOF time series as in (a) and (b), respectively, but focused on isomers detected at lower concentrations. Absent data points in (d) are for chromatograms where the chromatographic peak area was below the limit of detection. RT-Vocus trace in (a) and (b) only shows RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ detected in the room (not the supply air) for a direct comparison with the GC samples.

Download

As discussed above, fragmentation and cluster formation can complicate the interpretation of real-time measurements, where now there are multiple ions associated with one compound. An example of this behavior is shown in Figure S5, where Fig. S5a is a GC-Vocus chromatogram of the monoterpene protonated molecular ion, C₁₀H${}_{\mathrm{17}}^{+}$, and Fig. S5b is the chromatogram of C₆H${}_{\mathrm{9}}^{+}$, a common monoterpene fragment ion formed in proton-transfer reactions. Relative signals for the fragment ion versus the protonated molecular ions for each monoterpene (labeled 1–6 in Fig. S5a) are reported in Table 3. For all monoterpenes, detection of the fragment ion C₆H${}_{\mathrm{9}}^{+}$ occurred in approximately a 1:1 ratio to the intact parent ion, C₁₀H${}_{\mathrm{17}}^{+}$.

Specifically, the ratios measured for the isomers ranged from 0.78 to 1.90, with an average of 1.15 ± 0.40; these results are comparable to monoterpene fragmentation ratios reported elsewhere (Steeghs et al., 2007). While the GC can be used to select PTR-MS parameters that optimize the formation of the protonated molecular ion, these other pathways are unavoidable across the range of compound classes observed, and routine GC measurements allow the user to account for them.

3.3.2 Monoterpene temporal behavior during ATHLETIC

During ATHLETIC, several monoterpene enhancement events took place while the system operated in multiple modes. On 16 November, from 09:00 to 12:00 LT (local time), the RT-Vocus measured an increase in the C₁₀H${}_{\mathrm{17}}^{+}$ signal in the room air versus the supply air (Fig. 2). This increase correlated with the presence of people in the weight room, who presumably acted as the source of the elevated levels of monoterpenes (likely from personal care products). The GC measurements between 13:00 to 16:30 LT on 16 November were taken while people were present in (and then left) the weight room. During this time the RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal and the limonene measured by GC-Vocus decreased while the other monoterpene isomers stayed relatively constant (Fig. 3a, c). Interestingly, at about 17:00 LT on 16 November, the sum of monoterpenes measured by the GC-EI-TOF showed an increase that was also observed in the RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal (Fig. 3b), and the GC speciation showed that the concentration of all the monoterpene isomers other than limonene increased (Fig. 3d). The increase was first observed by the RT-Vocus in the supply air (Fig. S6A), which, when combined with the GC-EI-TOF identification of non-limonene monoterpenes, allows us to attribute this event to an inflow of outdoor air not influenced by game-related activities.

Table 3Relative signals of ions detected in the Vocus, fragmentation ion (C₆H${}_{\mathrm{9}}^{+})$ versus protonated molecular ion (C₁₀H${}_{\mathrm{17}}^{+})$, for monoterpene isomers.

Download Print Version | Download XLSX

An increase on 17 November, between 09:00 and 15:00 LT, occurred during a tailgating event and football game that took place adjacent to the Dal Ward Athletic Center. During this event, the room air and the supply air both showed increased C₁₀H${}_{\mathrm{17}}^{+}$ signals (Fig. S6B) indicating a source in close proximity to the facility. Since the supply air for the weight room enters the building from an intake on the north side of the football stadium, VOCs emitted during the sporting event could be subsequently transported through the HVAC system and into the weight room. The large spike in C₁₀H${}_{\mathrm{17}}^{+}$ observed that day at 15:00 LT, however, was predominantly measured inside the weight room rather than in the supply air (Fig. S6C), suggesting an emission source in the room associated with people and possibly cleaning activities in the Dal Ward center after the game. Evaluating the 17 November enhancement events with the GC measurements, it is clear that the increase in the RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal during the football game and the spike that occurred at 15:00 LT were due to an increase in limonene (Fig. 3a, b), with no corresponding significant increases in the other monoterpene isomers (Fig. 3c, d). The large monoterpene enhancement observed during and after the football game, which was dominated by limonene, was likely due to VCPs from personal care products used by the athletes and spectators and/or cleaning supplies.

3.4 Detection of dimethylsilanediol

In the GC-EI-TOF chromatograms, a well-resolved peak was observed at retention time 230 s (Fig. S7A) with a mass spectrum base peak at m∕z 77 (Fig. S8A). Using TERN analysis software, the unit mass resolution (UMR) electron ionization fragmentation pattern for this chromatographic peak (Fig. S8) was compared against the NIST mass spectral database (Linstrom and Mallard, 2020, https://webbook.nist.gov/chemistry/, last access: 23 December 2020) resulting (90.8 % probability) in an identification of that compound as dimethylsilanediol (DMSD). Analyzing the same chromatographic peak with the high-resolution EI-TOF data, the mass spectral peak at UMR m∕z 77 was assigned as molecular formula CH₅O₂Si⁺ (Fig. S9A). This characteristic ion is formed from DMSD through the loss of a methyl (CH₃) group. The protonated molecular ion of DMSD (C₂H₈O₂SiH⁺) and its protonated water cluster (C₂H₈O₂Si[H₂O]H⁺) were identified at the same retention time in the GC-Vocus chromatogram (Fig. S7B). The tailing observed only in the GC-Vocus chromatographic peak was likely from inefficient transport through the transfer line due to non-uniform heating. The high-resolution mass spectral fits for these peaks from the GC-Vocus are shown in Fig. S9B and C, respectively. From the high-resolution identification of the molecular ion (GC-Vocus) and the fragmentation pattern with characteristic ions (GC-EI-TOF), this compound was conclusively identified as DMSD.

It should be recognized that siloxanes have historically been a class of compounds that are difficult to measure analytically due to artifacts created throughout sampling systems (Rücker and Kümmerer, 2015). For example, siloxanes have been associated with artifacts for GC analyses that utilize septa in the sample path (de Zeeuw, 2005; Wang, 2006). For the instrument used in this study, no septa are present, and with our in situ GC sampling we avoid many artifacts that can be introduced due to sample collection and storage for offline analysis. The sample trap used for preconcentration is also of concern for artifact generation. To determine if DMSD is being produced as an artifact from the adsorbent trap, we conducted system zeros using dry (RH < 2 %) and humidified (RH ≈ 50 %) UHP N₂ to cover humidity conditions relevant for this study and saw no evidence of DMSD. From these results, we conclude that the DMSD observed during this study was not an artifact of the instrumentation but was instead present in the air sampled from the weight room.

Figure 4 shows the time series of the integrated chromatographic peak areas detected from the GC-EI-TOF. The time series shows dynamic behavior where DMSD builds up in the room starting at midnight on 17 November until a sharp decay after the football game, as seen with the RT-Vocus C₁₀H${}_{\mathrm{17}}^{+}$ signal (Fig. 2) and the GC-EI-TOF limonene signal (Fig. S4a).

Figure 4Time series of DMSD (black trace) and D₅ siloxane (grey trace) detected by GC-EI-TOF (normalized counts, ncts), weight room relative humidity (pink trace), and the fraction of outside air in the room supply air (blue trace). The area highlighted in yellow represents the time during which pre-game activities and a football game were occurring adjacent to the athletic center.

Download

This elevated concentration of DMSD on 17 November is also observed in the more limited GC-Vocus data set where only 10 chromatograms were obtained. As shown in Fig. S9, DMSD was detected as both the protonated molecular ion (C₂H₈O₂SiH⁺) and as the protonated water cluster of the molecular ion (C₂H₈O₂Si[H₂O]H⁺). After the decay on 17 November, the DMSD stabilizes to a background concentration until another increase in concentration starting in early afternoon on 18 November. The temporal behavior follows that of the RH in the room (Fig. 4), where DMSD begins to increase following an increase in RH and then declines sharply as the RH begins to decrease. Figure 4 shows that during the first DMSD enhancement, the fraction of outside air in the supply air was tapering down from 40 % to 15 %; however, during the enhancement on 18 November, outside air was increasing from near 0% to about 20 %. The lack of correlation between DMSD and the fraction of outdoor air in the supply air indicates that this compound was not being transported from outdoors but instead had a source inside the room or building. The sharp decrease in DMSD, which correlated with the observed decay for limonene on 17 November, appears to be from reduced production of DMSD in the room, along with a fast loss from ventilation. The observed behavior of DMSD does not follow that of decamethylcyclopentasiloxane (D₅), a siloxane commonly used in personal care products like deodorants (Coggon et al., 2018; Tang et al., 2015). The time trace for D₅ is also shown in Fig. 4. There is an enhancement in the D₅ signal during the football game and associated activities on 17 November (due to people inside and adjacent to the weight room), but the D₅ time series does not track the initial buildup of DMSD on 17 November and shows no enhancement during the DMSD pulse on 18 November. Therefore, it is apparent that there are different sources for these two organosiloxane compounds.

DMSD has been shown to be an environmental degradation product of both cyclic (cVMS) and linear (PDMS) siloxanes (Rücker and Kümmerer, 2015; Tuazon et al., 2000). Both classes of organosiloxanes degrade to DMSD through gas-phase oxidation by hydroxyl (OH) radicals (Tuazon et al., 2000) and through condensed-phase hydrolysis reactions (Xu et al., 1998; Lehmann et al., 1994a,b; Lehmann et al., 1995; Carpenter et al., 1995). While there is uncertainty surrounding the importance of OH radical chemistry indoors, previous work has estimated that typical indoor OH radical concentrations are on the order of 10⁵ molec. cm⁻³ due to low lighting conditions that reduce conventional photolysis reactions that produce OH radicals (Weschler and Carslaw, 2018; Abbatt and Wang, 2020; Pagonis et al., 2019; Gligorovski and Weschler, 2013; Young et al., 2019). In the weight room during the DMSD enhancement events there was no natural light (i.e., no windows in the room) and little artificial light as the space was often unoccupied. In addition, as discussed above and shown in Fig. 4, DMSD enhancement events do not track the observed behavior of D₅ siloxane. For these reasons, we conclude that the DMSD observed in this study was not formed inside through gas-phase oxidation of cVMSs (e.g., D₅ siloxane) by OH radicals, and instead we hypothesize that the production is through condensed-phase reactions followed by volatilization.

The mechanism by which PDMS (the less volatile class of organosiloxanes) decompose to form DMSD has primarily been studied with regards to degradation in soils. There, PDMS undergoes moisture-dependent hydrolysis to form DMSD, a process shown to be catalyzed by minerals present in clay (Xu et al., 1998). Figure 5 provides a proposed mechanism for the formation of DMSD from PDMS, though note that the PDMS co-product of DMSD can also undergo hydrolysis to produce more DMSD.

Figure 5Proposed mechanism for the formation of DMSD from the hydrolysis of PDMS.

Download

Figure 6DMSD (normalized counts, ncts) measured by the GC-EI-TOF versus relative humidity (%). The data was fit with a sigmoidal curve that shows the greatest rate of change in DMSD production at ≈ 20 % RH.

Download

Once formed, the DMSD then either volatilizes or is retained in the condensed phase to undergo subsequent reactions. Depending on the moisture conditions and mineral composition, hydrolysis occurs on timescales of minutes to days and efficiently in the presence of calcified minerals like Ca-kaolinite, Ca-beidellite, and Ca-nontronite (Xu et al., 1998). The hydrolysis of PDMS to DMSD has been shown to only occur in the presence of moderate amounts of water (Rücker and Kümmerer, 2015; Lehmann et al., 1998). Some water is required for the hydrolysis reaction, but when moisture levels are too high the reaction shuts off, presumably due to saturation of catalytic mineral sites by adsorbed water.

While the exact source of DMSD observed during this study cannot be determined from our measurements, we can suggest at least two possible sources for the DMSD production that would account for our observations. As shown in Fig. 6, the DMSD signal has a sigmoidal relationship with RH, where the greatest rate of change in DMSD production is around 20 % RH. The steepness of the curve suggests the DMSD is being produced from a discrete process, likely due to some material that combines siloxanes with a mineral catalyst taking up water around 20 % RH providing the needed water for the hydrolysis reaction.

Indoors, sources of PDMS are numerous: paints, coatings, sealants, textiles, and electronics (Andriot et al., 2007). Particularly relevant is their use in paints, where PDMS are used as the binder (30 % w∕w–100 % w∕w of the mixture) and as additives (0.1 % w∕w–5 % w∕w of the mixture), depending on the product and manufacturer (Andriot et al., 2007). Another significant component of paints are inorganic minerals, used as pigments and extenders, constituting 20 %–50 % of the paint mixture (Karakaş et al., 2010). While TiO₂ is commonly added for its optical properties, minerals like calcite, Ca-kaolin, talc, and dolomite are used as fillers (Karakaş et al., 2011). The abundance of painted surfaces indoors and the correlation between room humidity and enhancement of DMSD (Fig. 4) provide a possible explanation for the observation of DMSD during this study: mineral-catalyzed hydrolysis of PDMS on painted surfaces analogous to what occurs in soils. While the presence of PDMS in paints and coatings makes this chemistry plausible, the partitioning of lower-volatility cVMS could also act as a source of condensed-phase siloxanes in indoor environments. Using the D₅ siloxane vapor pressure measured by Lei et al. (2010), the saturation vapor concentration (C^*) is estimated to be ≈ 10⁶ µg m⁻³. According to Pagonis et al. (2019) and Algrim et al. (2020), for compounds with $C^{*}<{\mathrm{10}}^{\mathrm{8}}$ µg m⁻³ deposition to surfaces can compete with ventilation as a removal process. Thus, some of the cVMS emitted in the weight room may partition to surfaces where they could react to produce DMSD as discussed above for PDMS.

The observation of DMSD as an atmospheric reaction product of cVMS, as a soil degradation product of PDMS, and now in an indoor environment where its formation is proposed to occur by surface hydrolysis of siloxanes, demonstrates the need to quantify its environmental impact. A review by Rücker and Kümmerer (2015) refers to DMSD as the most important intermediate of organosiloxanes degradation, and yet very little is known about its physical properties, even boiling point or vapor pressure. This lack of knowledge has been due in part to an absence of analytical methods that can measure these compounds. Here, we have demonstrated an in situ, field-deployable technique capable of measuring DMSD that can be used in a wide variety of environments to understand the sources and fates of this compound. These measurements are of particular interest considering that the eventual fate of DMSD in the atmosphere is to form CO₂ and SiO₂, where SiO₂ could contribute to new particle formation and growth in the atmosphere (Bzdek et al., 2014).