Preservation of Vapor Samples on Adsorbent Alumina Capillaries and Implications for Field Sampling

Abstract

Dynamic vapor microextraction (DVME) is a vapor preconcentration method that employs a capillary trap coated with an adsorbent, followed by solvent elution to recover the sample. DVME has been developed for applications in the laboratory, including highly precise vapor pressure measurements, and in the field. When vapor collection is conducted outside the laboratory, samples must almost always undergo some interval of storage representing the time between collection and analysis. This interval may be hours, days, or longer, depending on the situation. Regardless, in all situations there must be confidence that the integrity of the samples is maintained until processing and analysis. In this paper, we present results of two studies that tested the stability of a 50 % weathered gasoline headspace sample on alumina PLOT (porous layer open tubular) capillaries stored at room temperature for periods from 24 h up to 20 wk. We used principal component analysis (PCA) to reduce the dimensionality of the chromatographic and mass spectral data and elucidate trends in stability with respect to the complex sample’s range of hydrocarbon classes and molecular weights. Both analyses identified changes over storage periods of six weeks or more. The hydrocarbon class analysis, which used selected ion monitoring (SIM) data as input, proved more sensitive to changes over shorter storage periods. Sample integrity was preserved for at least 24 h, but losses, especially of high-volatility compounds, occurred by 168 h (7 d). Near total loss of sample occurred by 20 wk. These findings, which are specific to the sample, adsorbent, and storage conditions, will guide choices in experimental and instrumental design to ensure that data from future field studies is reliable.

1. Introduction

Vapor or headspace sampling is a frequently-employed tool in many disciplines of science, including environmental monitoring [1-3], industrial hygiene [4, 5], air quality [6, 7], forensic chemistry [8-10], animal health [11-13], and human health [14]. These applications are not based in the laboratory but rather involve sampling in a field situation, such as at a Superfund site [3] or a workplace [4, 5]. A common implication of field sampling is that the samples undergo a period of storage before they are analyzed equal to the amount of time it takes to return the samples to the laboratory. Sample backlogs may also contribute to the storage interval. While whole air or breath samples can be collected for transport to the laboratory (e.g., in canisters or flexible polymer bags [5, 15]), more often the organic compounds are captured by preconcentration onto an adsorbent to stabilize them for transport [5, 11]. If compounds are not transferred, diffusion through the walls and adsorption become a concern. Consequently, for exhaled breath samples in polymer bags, a maximum storage interval of 10 h has been proposed [16, 17].

Multiple variables influence the degree of stability or instability of a stored vapor sample, including the device and fittings, adsorbent phase, storage temperature, and characteristics of the sample itself. Tenax tubes are packed with a hydrophobic polymer adsorbent phase, and the stability of organic compounds in this device has been studied in the most depth, likely because of their broad applicability. The effect of temperature has been studied in an occupational setting [18] and as it applies to exhaled breath [19, 20]. Solid phase microextraction (SPME) fibers, another common preconcentration device, have also been evaluated for stability over periods of hours up to three days [21-23]. These studies tend to use single compounds or simple mixtures as the test sample. Because stability depends on the device used and the sample, broad conclusions cannot be drawn about vapor sampling in general; instead, investigation of each sampling scenario is required before one can have confidence in the integrity of a stored sample.

Dynamic vapor microextraction (DVME), previously known as PLOT-cryoadsorption in some applications, was developed at NIST and first published in 2009 [24]. DVME is a small-volume purge-and-trap sampling method that concentrates vapor-phase analytes onto a section of porous layer open tubular (PLOT) capillary column. This capillary vapor trap is cooled to subzero temperatures during sample collection to enhance adsorption efficiency and stabilize any reactive species in the sample. In the laboratory DVME apparatus, the sample is placed into a crimp-capped vial inside an oven and the headspace is purged with helium or another inert gas at flowrates up to 10 mL/min. The inlet of the capillary pierces the septum and approximately 20 cm of its 100 cm length is within the oven, while the remainder is coiled outside the oven in a chilled enclosure. Headspace analytes swept by the gas into the capillary are adsorbed, and the gas flow rate and total collection volume are measured at the outlet of the capillary. This method has been applied in previous work to explosive materials [25], fuels and fire debris [26-28], food spoilage [29], and gravesoil [30]. It has also been used quantitatively to measure the vapor pressure of cannabinoids [31] and, most recently, n-eicosane [32]. The stability of vapor samples collected by DVME has not previously been assessed.

We recently developed a portable vapor sampling device intended for field use [33, 34]. The portable device uses a multicapillary trap consisting of six individual sections of PLOT capillary bundled in parallel (to achieve flow rates up to 100 mL/min) and set into an epoxy wafer for ruggedness. While the laboratory DVME apparatus uses positive pressure to push headspace vapor into the capillary, the portable device uses negative pressure to pull vapor through the capillary trap. In a simulated field test, the device successfully detected gasoline, markers of explosive materials, and markers of protein decomposition with short sampling times and in a variety of ambient temperatures [10]. The field test took place on the NIST campus and adjacent to the laboratory, however, and no delay for sample transport was incorporated in the procedure. The experiments we present here are an effort to understand whether, and how, such a delay would impact the results of future field deployment at a more remote location.

We selected 50 % weathered gasoline to generate headspace vapors to load each capillary prior to storage. Our background in fire debris and fuels research gave us a good working knowledge of the sample, which is one reason for choosing it. In addition, gasoline is a highly complex mixture of hydrocarbons with a wide range of volatilities and moieties. Its complexity allows us to study the effects of storage on each component of the mixture. The archival stability of headspace samples from fire debris is a topic of study in forensic science and arson investigation [35]; however, the time scale relevant to evidence preservation is measured in years, much longer than the periods studied in the present work. The samples we generated simulate a field scenario in which soil or groundwater is being tested for the presence of a fuel spill.

In this study, we evaluated the effect of storage time up to 20 wk on vapor samples collected onto alumina PLOT capillaries using DVME. These experiments generated a rich set of chromatographic data and collected information about the carbon number distribution, hydrocarbon class distribution, and total signal retained as a function of storage interval. Unlike most previous studies of vapor sample storage stability, which examine a single compound or small set of compounds, data analysis of these samples could not be completed using simple percent recovery or recovered mass over time. Gasoline contains so many compounds that designing a GC program that achieves baseline resolution of individual peaks is impractical and sometimes impossible (e.g., with xylenes). We therefore applied principal component analysis (PCA) to reduce the dimensionality of the data set and to identify differences among the capillaries as a function of their storage interval. This approach was an effective way to easily visualize changes in sample composition on an individual sample basis.

2. Materials and Methods

2.1. Materials

All chemicals and materials were commercially obtained. Unoxygenated gasoline was weathered at room temperature under a chemical hood to 50 % by volume and stored in 40 mL screw top vials until use. The purity of the acetone was verified before use. Capillary traps were made from approximately 100 cm lengths of 0.32 mm ID alumina PLOT columns with a film thickness of 8 μm. The capillaries were sealed in Kapak/Ampak SealPAK evidence bags (4.5 mil or 0.5 mm thick) using a Kapak/Ampak heat sealer.

2.2. Generation of capillary samples

The laboratory DVME apparatus, which is based on a repurposed gas chromatograph (GC), was used to create nominally identical vapor samples following procedures described in previous publications [28]. The GC oven controlled the sample temperature and the electronic pressure controller (EPC) at the GC inlet controlled the flow of purge gas. The helium flowed through fused silica capillary into a clean transfer vial (to protect against contamination) and then into the sample vial or saturator, which contained 100 μL of weathered gasoline. Both vials were 2 mL autosampler vials with crimp septum caps. First the vortex tube was turned on, and we waited until the temperature inside the insulated enclosure was ≤ 0 °C before setting the oven temperature to 60 °C. We waited for thermal equilibration inside the saturator and observed this phenomenon by tracking the flow rate of the gas inside the saturator, which started at room temperature. The approximate expansion volume of the system at 60 °C was 0.2 mL; this volume was included in our calculation of the total collection volume. When the flow returned to 0 sccm (standard cubic centimeters per minute), we adjusted the pressure via the EPC until the target flow rate was achieved. The conditions we selected for the generation of capillary samples were headspace temperature = 60 °C (± 1 °C) and total collection volume = 1.5 scc (± 0.06 scc). The target flow rate was 0.5 sccm; however, due to limitations in the precision of the EPC, this value varied between 0.3 sccm and 0.7 sccm. The variability for individual capillaries was ± 0.05 sccm.

2.3. Study design

We conducted two related studies with different time scales. The “long-term” study included time points measured in weeks: 0 wk, 6 wk, 13 wk, and 20 wk. We analyzed 10 replicate capillary samples for each time point, requiring 40 samples to be generated under nominally identical conditions. We actually generated 70 samples, but ultimately did not analyze 30 samples (i.e., planned time points after 20 wk). These samples were generated in one batch during a single two-day period. They were kept on the benchtop for no more than 30 min before being sealed in the evidence bags. After they were sealed in bags, the samples were randomly assigned to a time point group and stored together at room temperature (20 °C). The non-archived samples were sealed in evidence bags and analyzed as a batch at the end of the two-day period. A set of 10 unused capillaries was sealed in evidence bags at the same time and stored under the same conditions. No compounds were observed in unused capillaries that were stored alongside the sample capillaries for 20 wk, the longest storage interval, indicating that the evidence bags were not contributing a measurable background signal.

The “short-term” study was designed after we had the results of the long-term study and incorporated some adjustments to enable investigation of storage intervals shorter than 48 h. (In the long-term study, non-archived samples were stored up to 48 h prior to elution.) The short-term study included the following time points: 0 h, 24 h, 168 h (7 d), and 336 h (14 d). The nonzero time point capillary samples were generated six at a time, and each time point measurement was started on a different day. On every day that nonzero time point samples were generated, six replicate non-archived samples were also generated and analyzed immediately. In the short-term study, the capillaries were sealed in evidence bags immediately after headspace collection was complete without spending any time in the open air on the bench. Non-archived samples were not sealed in evidence bags; elution occurred within seconds of their generation. One 168-hr sample was discarded because the capillary broke during elution.

2.4. GC-MS analysis

We eluted the capillaries with approximately 1 mL of acetone to generate a liquid sample. We performed a second 1 mL elution to recover any residual analyte in the capillary; we did not detect compounds in any of the second elutions. The GC-MS program was as follows: 5 % phenyl dimethylsiloxane column (0.25 mm ID, 0.25 μm film thickness); 1 μL injection volume; 20:1 split ratio; inlet temperature 250 °C; 120 kPa in constant pressure mode. The oven program started at 30 °C with a hold for 2 min, 1 °C/min to 36 °C, 5 °C/min to 90 °C, and 25 °C/min to 250 °C. After a 2 min solvent delay, the MS was operated in dual mode with scan over 29-300 m/z and ions 55, 57, 69, 71, 83, 85, 91, 99, 105, 117, 118, 119, 128, 131, 132, 142, 156 m/z. We ran an n-alkane ladder (C₆-C₂₀) to support carbon number analyses.

In the long-term study, liquid samples were analyzed weeks apart, immediately after elution of a given time point. In the short-term study, liquid samples were also analyzed immediately after elution to provide real-time information. Then the vials were recapped and stored at 4 °C until the study conclusion, at which time the full collection of samples were analyzed together. Samples were weighed before and after cold storage to verify that no significant mass loss had occurred. Every 10 GC runs, we injected two control samples: acetone and a neat gasoline sample. The first confirmed that carryover was not occurring, and the second confirmed that the instrument response was consistent. Samples from the short-term study had more signal, therefore we used a 200:1 split ratio to enable quantitation of the largest peaks, namely toluene. All other GC-MS parameters remained the same. While analyzing all samples together, as was done in the short-term study, is ideal to eliminate instrument variability, the analyses we applied to both studies, described in the following section, also normalized the data to eliminate contributions from instrument variability.

2.5. Data analysis

Principal component analysis (PCA) is an unsupervised approach to data analysis that takes a data set with many variables (high dimensionality) and reduces it for easier visualization of variability. The original variables are transformed into new, orthogonal variables called principal components (PCs), which are linear combinations of the original variables that maximize the variability within the data set. Data processing was executed using Python scripts. The Python libraries Scikit-learn and Matplotlib were used for PCA analysis and visualization. First, samples that were injected in triplicate were averaged. Next, we subtracted the baseline by subtracting a solvent (acetone) blank from each sample. We further processed this data to study stability by both (a) volatility using the total ion current (TIC) chromatogram and (b) hydrocarbon class using the selected ion monitoring (SIM) chromatograms.

To examine changes in the liquid sample by molecular weight and thus volatility, we segmented each TIC signal based on the retention times of n-alkanes (Fig. 1). We integrated each segment and then normalized by the total integrated signal (C₇-C₁₅) from each sample to obtain the percent of total signal in each carbon number range. Compounds eluting before n-heptane were not considered in this analysis. Because of the normalization we applied in the carbon number analysis and the carbon class analysis (to be described next), correcting for the mass of acetone used to elute the sample from the capillary was not required.

Figure 1.

TIC chromatogram from a non-archived sample. The n-alkanes eluted at the retention times indicated by vertical lines, creating groupings of peaks based on molecular weight and volatility. For example, the retention times of n-heptane (3.17 min) and n-octane (6.38 min) are used as the boundaries of the area of the chromatogram that we call the C₇-C₈ carbon range.

To examine changes in the liquid sample by carbon class, we summed SIM signals to calculate the relative contribution of five carbon classes in each sample: alkanes (57, 71, 85, 99), cycloalkanes (55, 69, 83), aromatics (91, 105, 119), indanes (117, 118, 131, 132), and polyaromatics (128, 142, 156) (Fig. 2). These ions are frequently monitored to identify gasoline extracted from fire debris [8]. The signal for each carbon class was normalized by the total signal from all ions to calculate the percent of total signal within each carbon class. This approach allows us to look for changes in carbon class abundance between time points without interference from complicating factors like small differences in solvent rinse volume or instrument drift over time. The drawback of the normalization process is that it discards any changes in the sample that take place proportionally across the variables; for example, any signal loss that occurs equally across the five carbon classes.

Figure 2.

SIM chromatograms from a non-archived sample. We monitored 17 ions as markers of five compound classes: alkanes (57, 71, 85, 99), cycloalkanes (55, 69, 83), aromatics (91,105, 119), indanes (117, 118, 131, 132), and polyaromatics (128, 142, 156). Ions with greater abundance in the sample (e.g., from aromatics) have a smoother baseline compared to less abundant ions (e.g., from polyaromatics) because each chromatogram was normalized by its maximum signal and offset for easier visualization; this kind of normalization was not part of our analysis.

PCA results are presented in the form of biplots, which plot the PC coefficients (scores) of the data as a scatterplot on the same coordinate plane as the loading vectors. The loading vectors indicate how much each original variable contributes to each PC and how variables are correlated with each other. The original variables for our analyses are the fraction of signal due to compounds within a specific carbon number range (derived from TIC chromatograms) or the fraction of signal due to compounds of a specific carbon class (derived from SIM chromatograms). There were eight input variables for carbon number analysis and five input variables for carbon class analysis.

3. Results and Discussion

3.1. Carbon number analysis

Fig. 3 shows the PCA results by carbon number for the long-term and short-term studies. The long-term study does not include 20-wk results, because the TICs of those samples did not provide enough signal; i.e., most of the sample had been lost by 20 wk (Fig. 3a). The non-archived samples (in red) are most tightly clustered together, while scatter increased monotonically up to 13 wk. As storage time increased, the composition of the samples shifted to less volatile, high carbon number compounds. No major carbon number trends were observed in the short-term study as a function of storage time between 0 h and 336 h (Fig. 3b). The samples did not cluster by time point as was the case in the long-term study; however, some increase in scatter within each time point is apparent as storage time increases. Specifically, the 24-h samples are more tightly clustered together than the 168-h or 336-h samples.

Figure 3.

PCA biplots for carbon number analysis. (a) Long-term study has an explained variance of 91.1 %. (b) Short-term study has an explained variance of 84.9 %.

The loading plots in Fig. 3a and 3b show similar trends. In the long-term study, C₇-C₈ and C₈-C₉ contributions dominate the non-archived samples, i.e., the loading vectors point in that direction. The 6-wk samples cluster together in the direction of C₉-C₁₀. The loading vectors for carbon numbers above 10 group together and trend in the direction of the last time point (Fig. 3a, 13 weeks). Although the time points do not differentiate themselves in Fig. 3b, the loading vectors have similar relationships in that the C₇-C₈ and C₈-C₉ contributions are separate from each other and from the loading vectors for carbon numbers above 9, which group together. This is an indication that the same process occurs in both studies, albeit on a different timescale.

3.2. Carbon class analysis

In the long-term study (Fig. 4a), the non-archived samples (in red) are again most tightly clustered. Scatter increased with storage time, indicating that the samples not only changed relative to non-archived samples but diverged from each other. SIM data was the input to the carbon class analysis, and there was enough signal in the 20-wk samples to include them, even though their TICs were not distinguishable from baseline. The grouping on this plot is similar to the carbon number plot. The late time points cluster together in the direction of the larger compound classes (indanes and polyaromatics). These loadings trended together in PC1 and PC2. Aromatics and alkanes were inversely correlated in PC2.

Figure 4.

PCA biplots for compound class analysis. (a) Long-term study has an explained variance of 87.9 %. (b) Short-term study has an explained variance of 93.8 %.

In the short-term study (Fig. 4b), the non-archived samples generated on different days cluster together, indicating that there were no meaningful differences in headspace collection. Furthermore, the 0-h and 24-h sample clusters are completely overlapping, indicating that the samples were stable with respect to carbon class for this storage interval. The least scatter occurred in the 0-h and 24-h samples. Changes began to occur after the 1-week storage interval, with 168-h and 336-h samples showing less contribution from alkane compounds and more in the aromatic category. 168-h samples shifted away from the two earlier time points and appear scattered in between those earlier points and the 336-h samples. While alkanes and cycloalkanes were more closely correlated in the short-term study, aromatics and alkanes were again inversely correlated.

3.3. Discussion

Carbon number analysis and carbon class analysis were equally successful at capturing the loss of high-volatility compounds that occurred after storage for 6 weeks or more. We note that analysis by carbon class (SIM data) was advantageous because the 20-wk samples could be included, whereas in the carbon number analysis based on TICs, there was not enough signal to meaningfully include these samples. Furthermore, carbon class analysis was more sensitive to changes that occurred after shorter storage intervals. While Fig. 3b did not reveal obvious differences between storage intervals, Fig. 4b clearly does. In fact, this analysis visualizes progressive changes over time, with 168-h samples in between earlier (24 h) and later (336 h) time points. The correlation of the original variables is also revealing. In Figure 4b, alkanes and cycloalkanes are correlated, indanes and polyaromatics are correlated, and aromatics are inversely correlated with both. Fig. 2 shows that the average volatility of compounds present in these classes is similarly correlated. This indicates that compound loss is driven by volatility; we observe no evidence of preferential retention by the alumina adsorbent.

Our work is not the first to apply multivariate analysis methods to gasoline-derived samples. Sandercock and Du Pasquier used PCA to distinguish GC-MS data from unevaporated and evaporated gasoline samples towards the goal of ‘fingerprinting’ gasoline recovered following arson with incendiary bombs or delayed-ignition devices [36, 37]. Similar to our results, two PCs captured more than 90% of the original variance. However, only a small number of compounds were captured and analyzed. Specifically, the researchers used solid phase extraction with an alumina adsorbent to isolate polyaromatic hydrocarbons (PAHs). The final model input included 11 integrated peaks obtained from ions derived from C₀-C₂ naphthalenes. The focus on a subset of compounds, present in all samples, allowed most of them to be baseline separated, despite the complexity of the original gasoline samples.

Breath samples are another highly complex mixture that have been analyzed by GC-MS or LC-MS data and interpreted with PCA [12, 13, 38]. Aksenov et al. examined differences in the exhaled breath condensate of managed and wild dolphin populations [12]. In that work, the researchers deconvoluted the chromatograms and preprocessed their data into compiled peak tables. Likewise, Hamblin and Almirall used peak areas of individual compounds detected in the breath of cigarette smokers and non-smokers [38]. Both groups used a 20 % exclusion criterion, meaning that any compounds not appearing in at least 20 % of samples were discarded from the data set. The difficulty of using peak tables as PCA inputs, which both groups acknowledged, is that because not all compounds are found in every sample, the data contains many zero values that can have an undesirably strong influence on PCA calculations. To prevent zero values from skewing their data, Aksenov et al. replaced zeroes with the minimum intensity value for each compound divided by three. Hamblin and Almirall did not apply additional handling to minimize the effect of zero values in their data sets. In contrast, our data preprocessing methods resulted in data sets with no zeroes to deal with. Additionally, the PCA in both previous studies provided lower explained variance compared to our results. For Hamblin and Almirall, the first four PCs explained only 61 % of the variance in the data (both smokers and non-smokers; additional variance breakdown in the PCs is not provided). In Aksenov et al., the first two PCs explained just over 30 % of the variance in the managed population subset of data. Of course, our samples started out nominally identical, in sharp contrast to the variability inherent in human or animal samples in breathomics research – an important distinction that may also account for the differences between our explained variance and theirs.

4. Conclusions

Vapor samples are often collected by preconcentration in the field and analyzed hours or days later, after their transport to the laboratory. Dynamic vapor microextraction (DVME), tested herein, is a portable vapor preconcentration method in which samples are adsorbed within a capillary trap. We studied the stability of samples collected from the headspace of weathered gasoline, simulating a fuel spill scenario, and stored for intervals up to 20 wk. To visualize differences and trends, we developed a system for evaluating sample stability that utilizes principal component analysis (PCA), and found it to be an effective way to reduce the dimensionality of the large and complex data set generated by GC-MS analysis of over 100 samples. We determined that the integrity of vapor samples trapped on alumina capillaries and sealed in evidence bags was maintained for 24 h but began to deteriorate by 168 h (7 d) when stored at room temperature. Hydrocarbon class analysis, which used SIM data as input, and carbon number analysis, which used TIC data, both revealed loss of higher-volatility compounds as the primary effect of storage for 6 wk. However, we were only able to distinguish storage intervals of 24 h, 168 h, and 336 h, with the more sensitive hydrocarbon class analysis. We envision several future studies based on this work, as there are many variables to be explored. First, temperature is known to influence the stability of vapor samples and cold storage should be investigated if storage intervals greater than 24 h are needed. While the ends of the capillaries in this study were not capped, leaving them vulnerable to sample loss, the simple addition of a silicone septum or other closing mechanism may be effective. Adsorbent phases other than alumina may prove to retain analytes better. Finally, but perhaps most importantly, we must understand the stability of non-hydrocarbon compounds and specifically compounds relevant to additional field sampling scenarios.