In the initial stage, the sampling phases were subjected to a preliminary analysis following the conditioning procedures outlined in the “Conditioning step” section. The aim of this analysis was to assess the analytical cleanliness of the various sampling phases under consideration, i.e., their VOC content before any sampling or spiking was carried out. The resulting chromatograms for the five sampling phases analyzed by TD-GC×GC/ToFMS after the implementation of a conditioning step are displayed in Fig. 2. A blank chromatogram of the analytical system depicted in Fig. 2f was obtained by analyzing an empty stainless-steel thermodesorption tube under the conditions described in the “Thermodesorption” and “GC×GC/ToFMS” sections. In Fig. 2, each spot represents a compound. Consequently, a clean sampling phase is defined as a chromatogram with a minimal number of spots. This is crucial for preventing confusion during real sampling procedures, where compounds originating from the body volatolome and compounds already present in blanks may be misidentified. As illustrated in Fig. 2, it was almost impossible to obtain a chromatogram devoid of spots, as the analytical system itself produced some VOC emissions: mainly silylated emissions linked to column bleeding or fluorinated emissions linked to the various seals in the system.

TD-GC×GC/ToFMS contour plot chromatograms of blanks for a raw gauze, b Getxent® microtube conditioned for 4 h at 100 °C, c glass beads, d PowerSorb®, and e PSP conditioned for 4 h at 240 °C; f contour plot chromatogram of a TD-GC×GC/ToFMS system blank. All chromatograms are at the same intensity scale (from 7 × 106 to 1 × 108)

Another important point is the repeatability of the observed emissions, so that they can be monitored and, if necessary, removed when processing real samples. Firstly, the TD-GC×GC/ToFMS system itself contained an average of 160 spots with a relative standard deviation (RSD) of 52% (n = 7). That high variability was mostly linked to the spots observed at the beginning of the chromatogram in Fig. 1f between 0 and 500 s for the first-dimension retention time (1D-RT) and 0.5 and 1.5 s for the second-dimension retention time (2D-RT). That chromatographic area was consistently populated by numerous spots, even after the implementation of cleaning procedures on the system (blanks and bake-out procedure on the TD). Consequently, this area was rarely exploited for real samples. If needed, the area ratios can be employed to ascertain that the compound associated with the observed spot was predominant in the sample compared to the blank. For the sampling phases, the following number of peaks were detected (RSD given in parentheses, three replicates per sampling phase): 240 for glass beads (23%), 210 for PSP (2%), 150 for PowerSorb® (24%), 625 for gauze (8%), and 43 for Getxent® (48%). Although Getxent® and gauze exhibited the lowest and highest number of peaks, respectively, it is important to note that these two phases could only be desorbed at 100 °C due to their poor ability to resist elevated temperatures. Indeed, prolonged exposure of the gauze to 100 °C resulted in browning, while the Getxent® became sticky. This thermodesorption temperature, 120 °C lower compared to the other phases, had two consequences: (i) it detracted from the apparent good blank quality of the Getxent® and amplified the contaminating ability of the gauze, and (ii) it limited the capacity of the Getxent® to release semi-volatile compounds. Also, owing to the lack of resistance to temperature, the gauze could not be conditioned like the other phases, and Getxent® was only conditioned at 100 °C. Taking this point into consideration, the best blank quality was obtained for the PowerSorb®, whose more intense spots (Fig. 2d, ~ 500 s, 1000 s, and 1400 s in 1D-RT) were assigned to silylated compounds, not possibly originated from body odor and therefore easy to remove from real samples.

Secondly, Fig. 3 illustrates the impact of the conditioning step specifically for a ready-to-use phase such as PowerSorb®. As anticipated, the high sensitivity obtained by using TD-GC×GC/ToFMS meant a higher required efficacy level of conditioning, which could not be ensured by the manufacturer’s conditioning. The use of the method described in the “Conditioning step” section was therefore validated for conditioning phases such as the PowerSorb®. A total of 710 peaks were detected for analysis upon receipt for the self-described ready-to-use stick and 150 after the conditioning step. For the sum of areas, they measured from 5 × 1010 to 9 × 109.

TD-GC×GC/ToFMS contour plot chromatograms of a PowerSorb upon receipt, and b PowerSorb after conditioning step (4 h at 240 °C on TC- 20, see part 2.4). Chromatograms are at the same intensity scale (from 7 × 106 to 1 × 108)

Following an assessment of the blank quality of the sampling phases, it was necessary to evaluate their capacity to trap VOCs and subsequently release them during thermodesorption. This was investigated by spiking the sampling phases using the synthetic mix 57 with the method described in the section “Synthetic body odor mix and spiking.” However, as the blanks had already been evaluated, it was deemed important to ascertain the number of the 57 compounds that were already detectable in the phase’s blanks, to have unbiased data. The obtained results are summarized in the bar chart of Fig. 4.

Bar chart showing the number of compounds of the mix 57 recovered after TD-GC×GC/ToFMS analysis of the five spiked sampling phases. Hatched bars show the number of compounds of the mix 57 already detected in sampling phase blanks. Error bars show the standard deviation values obtained for the three replicates

Figure 4 illustrates the importance of considering the blank. The most crucial and meaningful information is the differential between the spiked phase and its blank. Indeed, a sampling phase in which the blank was already highly crowded could lead to bias during spiking or sampling, as the origin of the compounds could be misinterpreted. This was the case here for the gauze which, while exhibiting the lowest number of recovered compounds (36 compounds), also contained a considerable number of the molecules considered in the mix 57 in its blanks (29 compounds). Because of its poor and biased performance, the gauze was not subject to further consideration. Conversely, it was possible to conclude that PowerSorb® and PSP successfully trapped and released the 57 compounds of the mix for each replicate. This was evidenced by the fact that even the compounds detected in the blanks exhibited higher areas in the spiked samples. Nevertheless, PSP already had 22 compounds in its blanks, whereas PowerSorb® had seven. For the Getxent® microtube, compounds having higher boiling points like docosane, p-cresol, lilial, or naphthalene were not observed. This lower number of compounds (44 compounds) recovered for the Getxent® microtube can be attributed to its thermodesorption temperature not exceeding 100 °C, while other phases were desorbed at 220 °C. Given this limitation, Getxent® microtubes were not further considered. For the glass beads, a satisfactory number of compounds (49 compounds) were recovered. As previously highlighted in the literature [16, 25], the missing compounds were mainly volatile and apolar compounds (alpha-pinene, camphene, beta-pinene, or 3-carene).

To provide a semi-quantitative understanding of the differences between the phases’ affinities with regard to the polarity and volatility of the molecules, Fig. 5 depicts in greater detail the areas obtained for the 57 compounds of the mix, ranked in ascending order of elution in the first dimension (i.e., the more the compound is toward the right, the less volatile it is), for the PowerSorb®, PSP, and glass beads. This bar chart demonstrates that the areas obtained for glass beads were either negligible or null for compounds ranging from hexane to limonene. Compared to PowerSorb® and PSP, the variability in peak areas was higher for glass beads across most compounds, including those trapped and released in larger quantities, such as eicosane, heneicosane, and docosane. These findings confirm the greater affinity of glass beads for less volatile and polar compounds, highlighting their limited versatility for the intended application: sampling the entire body volatolome. Additionally, Fig. 4 illustrates that the quantities recovered for PSP and PowerSorb® were equivalent, with higher areas obtained with the PSP for volatile compounds (mostly hexane to beta-pinene). Regarding the variability in those areas, a mean variability of 70% was obtained for glass beads in comparison to 11% for PowerSorb® and 5% for PSP. Both of these results demonstrate very satisfactory repeatability. The recovery yields were calculated for PSP and PowerSorb® using the method described in the section “Synthetic body odor mix and spiking,” resulting in an average of 71% for PowerSorb® and 95% for PSP. For each sampling phase, the minimum recovery yield was 28% with the PowerSorb® for hexane and 68% with the PSP for lilial. The global recovery yield variability was 22% for PowerSorb and 10% for PSP. However, it is important to note that the masses of the phases were not similar (166 mg for PSP and 77 mg for PowerSorb®), despite the close dimensions of the sticks. Furthermore, while Tenax TA of the PSP involves adsorption phenomena, the physicochemical processes of the PowerSorb® are less obvious, as the supplier did not disclose the exact nature of the polymer.

Bar chart of the average areas obtained for each compound of the mix 57 for spiked glass beads, PowerSorb®, and the PSP

At this juncture, the outcomes were encouraging for both PowerSorb® and the PSP, exhibiting promising extraction yields, good repeatability, and good versatility. However, as PSP is not commercially available, it cannot be considered a viable option for larger-scale applications at this stage. Nevertheless, its development remains a significant area of interest. Consequently, PowerSorb® was selected as the optimal sampling phase for the remainder of the spiking studies.

The final objective being the sampling of VOCs coming from the body volatolome, it was interesting to assess trends regarding the sampling time and the maximum trapping capacity of the PowerSorb®. The assessment of the sampling time was performed in simulated conditions using the mix 57 for the PowerSorb®, as follows. A fixed quantity of VOCs was introduced in a closed environment at t0, and then the time allowed for diffusion was varied. After the sum of peak areas for the 57 compounds of the mix was calculated and normalized using TD8 area, these values were plotted against the sampling time, as displayed in Fig. 6. It can be seen that a plateau was reached after 30 min of exposure of the PowerSorb® to the mix 57, meaning that 30 min was needed to reach equilibrium. A 24-h sampling test was also carried out and showed an increase in the total area by less than a factor of 2. Therefore, to minimize the sampling time and to obtain a qualitative and representative sample (i.e., easily observable peaks and stable values over a certain time range), a sampling time of between 30 and 60 min appeared to be reasonable. With regard to repeatability, satisfactory results were obtained for 30 and 60 min, with RSD ranging between 7% and 15%. Given that for real sampling, the total quantity of compounds was not introduced at t0 but gradually emitted by the sampled body area, a sampling time of 60 min was chosen to ensure the most representative sample possible.

Sum of peak areas for the 57 compounds of the mix depending on the sampling time studied: 5, 15, 30, 45, 60, and 1440 min. Areas were normalized using the TD8 area. An axis break is present between the 60-min and 1440-min durations

Then, to better understand the PowerSorb® sampling phase and estimate its trapping capacity, different volumes of mix were added in a restricted environment, with a fixed diffusion time. The objective was to approximate the maximum quantity of VOCs that could be trapped by the PowerSorb®. Once again, the sum of peak areas for the 57 compounds of the mix was plotted, this time against the introduced mass of VOCs. As can be seen from Fig. 7, a plateau was reached when around 22 μg of VOCs had been introduced. This observed plateau meant that one PowerSorb®, weighing on average 77 mg, was theoretically able to trap 22 μg of VOCs. This maximum trapping capacity was evaluated with an equilibrium time of 30 min. In addition, this assessment was performed using one PowerSorb®; based on the study of Bicchi et al. [12], greater amounts of compounds can be obtained by increasing the available trapping surface. This was easily accomplished here by adding more PowerSorb® sticks. In their study, Bicchi et al. [12] showed that the recovered quantity was tripled when the surface area of PDMS was doubled. To investigate this point, the same experiment was repeated with two PowerSorb® units. As can be seen from the obtained results displayed in Fig. 7, the maximum mass of VOCs that could be trapped with two PowerSorb® units increased and seemed to approach 40 μg. Even if, by looking at the directional coefficient of the linear part, it seems that equilibrium was reached more slowly, increasing the maximum quantity of VOCs trapped was a desired advantage to ensure that saturation was not reached during actual sampling. Nevertheless, while these results were intended to provide an initial understanding of how the phase works, it should be noted that further experiments would be needed to fully understand its properties related to adsorption/absorption kinetics, such as displacement effects and individual or simultaneous stick thermodesorption. Such experiments would however be beyond the scope of the present study.

Sum of peak areas for the 57 compounds of the mix found on one and two PowerSorb® units depending on the introduced mass for 30 min of exposure. Areas were normalized using the TD8 area

To conclude this study on real sample applications, real sampling was conducted by sampling from the armpits of one individual for 1 h, with PSP on one side and PowerSorb® on the other. Figure 8 shows the obtained chromatograms, after software subtraction of sampling device blank chromatograms (Figure S2) for armpit samples taken with PowerSorb® and PSP. For the three repetitions, the mean number of detected peaks for PSP was 750 with an RSD of 17%, and for the PowerSorb®, 485 with an RSD of 10%. These results were consistent with the results described in the section “Spiking,” where the recovery yields for the PSP and the PowerSorb® were 95% and 71%, respectively. Figure 7 also shows a quite similar pattern for the two chromatograms of body odor collected on the PSP and PowerSorb®, but with higher intensities for the PSP. As the separation process was conducted using a normal-phase configuration, meaning having an apolar column in the first dimension and a medium polar column in the second dimension, molecules were segregated into strata based on their polarity. This meant that when tracing along the second-dimension axis, compound families were found in the following order: hydrocarbons, aromatics, esters, ketones, aldehydes, alcohols, and acids, as shown in the visual aid in Fig. 8c. Thus, here all the chemical families previously mentioned were found, justifying the need for versatile sampling phases such as the PSP and PowerSorb®.

TD-GC×GC/ToFMS contour plot chromatograms of armpit body odor samples taken with the sampling phase a PSP and b PowerSorb®. c Copy of chromatogram (a) with a visual aid to help elucidate the results based on the stratification into chemical families along the y axis. Dashed lines trace the expected 2D retention time pattern of different classes of chemicals, and the vertical dashed double arrow shows where alcohol and acid compounds can be found. Chromatograms are at the same intensity scale (from 5 × 106 to 1 × 108)

A comparison of the chromatograms obtained with the PSP and PowerSorb® revealed the presence of 120 peaks in both sampling phases. However, 188 peaks were observed exclusively in the body odor chromatogram obtained with the PSP; these peaks are reported in Tables S2 and S3 with their tentative identification. These results were obtained by manually examining the presence of each peak in sampling system blanks (Figure S2). After focusing on a specific retention time window, mass spectra were compared to ensure that the same compound was considered, and then peak areas were checked. For each peak it was verified that it was not present in the blank or that its area was twice as large in the real sample. The results demonstrated that the molecules exclusively present in PSP samples were primarily compounds whose peaks were also faintly visible in the PowerSorb® chromatogram. However, these peaks were not detected or validated, as their intensity was below the noise or blank emission threshold. Less than 30 peaks were observed in the body odor sample obtained with PowerSorb®, the majority of which were unidentified and therefore not reported here.

Ultimately, the findings substantiated the feasibility of sampling authentic body odor through the utilization of PSP or PowerSorb® as the sampling phase. While greater recovery yields could be achieved with PSP, satisfactory results were also obtained with PowerSorb®. Although PowerSorb® seemed slightly less sensitive, it had the essential characteristic of not discriminating between different chemical families. To enhance these outcomes and boost the sensitivity of PowerSorb®, it is essential to consider reducing the emissions from the sampling system to improve the quality of the background signal. In conclusion, the total area of the body odor sample obtained with PowerSorb® was approximately 177, after normalization using the TD8 area. Based on the findings described in the previous section, it can be concluded that the two PowerSorb® units captured approximately 30 μg of VOCs during a 1-h real sampling event, indicating that saturation was not reached but that two PowerSorb® units were needed.