The optimization and comparison of two high-throughput faecal headspace sampling platforms: the microchamber/thermal extractor and hi-capacity sorptive extraction probes (HiSorb)

Disease detection and monitoring using volatile organic compounds (VOCs) is becoming increasingly popular. For a variety of (gastrointestinal) diseases the microbiome should be considered. As its output is to large extent volatile, faecal volatilomics carries great potential. One technical limitation is that current faecal headspace analysis requires specialized instrumentation which is costly and typically does not work in harmony with thermal desorption units often utilized in e.g. exhaled breath studies. This lack of harmonization hinders uptake of such analyses by the Volatilomics community. Therefore, this study optimized and compared two recently harmonized faecal headspace sampling platforms: High-capacity Sorptive extraction (HiSorb) probes and the Microchamber thermal extractor (Microchamber). Statistical design of experiment was applied to find optimal sampling conditions by maximizing reproducibility, the number of VOCs detected, and between subject variation. To foster general applicability those factors were defined using semi-targeted as well as untargeted metabolic profiles. HiSorb probes were found to result in a faster sampling procedure, higher number of detected VOCs, and higher stability. The headspace collection using the Microchamber resulted in a lower number of detected VOCs, longer sampling times and decreased stability despite a smaller number of interfering VOCs and no background signals. Based on the observed profiles, recommendations are provided on pre-processing and study design when using either one of both platforms. Both can be used to perform faecal headspace collection, but altogether HiSorb is recommended.


Introduction
The metabolic profiling of volatile organic compounds (VOCs) for the detection or monitoring of disease is a rapidly maturing omics sub-discipline sometimes referred to as Volatilomics.Its non-invasive nature, in addition to ease of collection and the mostly absent requirement for sample pre-treatment, mark its suitability for high-throughput applications, making it increasingly popular.Of crucial importance is the continuous effort towards the optimization and standardization of the various platforms used for VOC sampling and their subsequent analysis [1][2][3][4][5][6][7][8][9][10][11][12].To study gut-health, faecal samples provide the closest proxy towards the anatomical location of the gut while considering gut-microbiome characteristics.However, collecting and analysing faecal VOC profiles has not been not straightforward for many research-labs, because the gold standards for VOC sampling and analyses over different bio-matrices were not harmonized.This current work optimized and compared two faecal headspace sampling strategies that work in harmony with thermal desorption (TD) units often utilized in exhaled breath and, more recently, urinary volatilomics [1].
Several Omics strategies exist to extract information from faecal material.The most obvious choice would be to measure the microbiome directly [13].Here, two main strategies include amplicon sequencing and whole genome shotgun sequencing (WGS).Where amplicon sequencing targets a genomic region that varies in multiple microorganisms and quantifies their relative abundancies, WGS additionally identifies functional genes.Both strategies have proven extremely valuable, and have shown gut dysbiosis to underlie e.g.including irritable bowel syndrome, inflammatory bowel disease, colorectal cancer, and liver disease [14].However, the involved mechanisms for these diseases remain unclear.Functional assessment of the gut microbiome is typically challenging, because non-identical micro-organisms might metabolize identical products.Identification is therefore not informative on a functional level.And although WGS does provide indications on the overall genomic potential, relative expression rates remain unknown.
To assess the metabolic function of the gut microbiome, metabolomics can be more informative.Not only does it have a smaller dimensionality compared to microbiome datasets, but it directly considers metabolic end-products while taking host-and diet interactions into account.As the metabolic output of the gut microbiome is to a large extent volatile, Volatilomics can act as surrogate matrix for the metabolic activity of the gut microbiome.A typical faecal VOC profile consists of short chain fatty acids (SCFAs), branched chain fatty acids (BCFAs), alcohols, esters, aldehydes, sulphides, cresols, indoles, terpenoids, ketones, alkanes, and more [15].Oddly, however, faecal volatilomics uptake by the volatilomics community has lagged behind.For example, compared to exhaled breath it has noticeably received relatively little attention.A quick Pubmed search reports a roughly seven-fold higher number of studies for exhaled breath compared to faecal headspace (i.e. with 2058 papers found for exhaled breath and 312 for faecal headspace using the search keywords 'Breath Volatile Organic Compounds' and 'Faecal Volatile Organic Compounds' , respectively).As a result, despite a hundred-fold and ten-fold larger genetic-and cell load of the microbiome compared to the host [16], the number of VOCs reported in faecal headspace (i.e.443) is remarkably lower compared to exhaled breath (i.e.1488) [17].Of these, many may be unique to the faecal headspace: e.g.studies so far have only shown a roughly 23% overlap of faecal VOCs with exhaled breath VOCs.
The relatively low uptake of faecal headspace analysis by the volatilomics community may in part be explained by challenges in the instrumental setup and not by disinterest.Traditionally faecal headspace collection is performed using solid phase micro extraction (SPME) fibres [18].A variety of fibres exist through which a selection can be made to target specific volatiles.Procedures using such fibres are typically time consuming and do not typically work in harmony with typical TD systems generally used for Exhaled Breath analyses (i.e. for many groups the overarching platform).This hinders system-wide volatile analyses for many research groups.Although gut health can be traced via other matrices [19], faecal headspace may be the most informative for diagnosing purposes.Moreover, origin determination of VOCs asks for confirmation studies where faecal headspace can be used as direct proxy.
To tackle these disadvantages, recently Markes International® has started merchandising the Micro-Chamber/Thermal Extractor© (here referred to as Microchamber) [20] and High-capacity Sorptive extraction probes (here referred to as HiSorb) [21], see figure 1.The Microchamber consists of a series of four or six chambers that can be heated up and sealed while being purged under a constant flow of nitrogen.After placing a desired sample in such a chamber, TD tubes can be attached to their exhausts, trapping volatiles of the respective headspace.This is a dynamic headspace sampling method.HiSorb probes consist of metal bars with stationary phase attached to their tips.They can be placed above or immerged in (i.e. in case of liquids) samples to perform static sampling.Upon sampling they can be placed in empty sorbent-free TD-tubes, and thus can be used in line with TD systems.Both sampling approaches enable high-throughput sampling, and are more sensitive due to increased surface areas of their stationary phase compared to SPME-fibres.As a result equilibria are reached faster and less sample pre-treatment is needed to capture semi-volatiles.There are multiple stationary phases, the choice of which depends on the application.Both HiSorb probes as well as the Microchamber platform have been applied in previous studies [22].These studies mainly looked into emissions of consumer products [23], construction materials [24], and foods and beverages [25][26][27].Applications concerning bio-fluids have so far remained limited, but include HiSorb extractions of cell-line cultures [28] and urine analyses [21].
The current study aimed for the optimization and comparison of the Microchamber and the HiSorb probes for faecal headspace sampling.As the faecal volatilome remains largely unexplored, this optimization was generic and pointed towards the broadening of untargeted signals while emphasizing biological variation and minimizing sampling variation (i.e.making it broadly applicable).Our optimization strategy used statistical methodology from design of experiment (DOE) with response factors based on semi-targeted and untargeted definitions of biological and instrumental variation.In addition, upon optimization both set-ups were compared to one another.

Experimental design
Faecal samples are notoriously reactive.They are challenging to homogenize (i.e.especially coming from a −80 • C freezer) due to their varying physical texture and faecal water content.Moreover, in view of the fast and unstable onset of fermentation, the amount of non-analytical time they are defrosted should be kept at a minimum.As a result, although many researchers would use pooled samples for optimization studies, it is questionable how representative such a pooled sample would be while being challenging to fabricate.Therefore, the current study used an optimization strategy in which no pooled samples were created, and instead focussed on optimizing relevant sources of variation between different metabolic profiles, as described below.This way, the optimization process itself adhered fully to the standard operating procedure it is designed for.

Sample collection and preparation
Volunteers were asked to donate faecal samples for either the Microchamber (i.e.sampling event 1) or the HiSorb optimization (i.e.sampling event 2), or both.These volunteers consisted of selfreported healthy individuals from different households.Optimization of Microchamber sampling included usage of faeces coming from five volunteers (i.e. three male and two female, with an age range of 28-68), whereas for the HiSorb probes this included four volunteers (2 females and 2 males with age range of [28][29][30][31][32][33][34].Per sampling event, four faecal tubes (Sarstedt, Nümbrecht, Germany) were filled, summing up to roughly a total of 30-40 g of faecal material per subject.Upon collection samples were frozen overnight and, while frozen, transported to the analytical laboratory using cool transport containers (Sarstedt, Nümbrecht, Germany) within 24 h after collection.Upon arrival, samples were stored at −80 • C until subsequent analysis.Samples only underwent only one freeze-thaw cycle.

Microchamber sampling and optimization
Upon storage, at maximum one month after collection, samples were kept on ice during subsequent analytical procedures while the Microchamber was heated up to 40 • C. Next, 250 mg faecal matter aliquots were weighed in disposable sterilized glass vials and placed in the Microchamber.Thereupon, to allow for sample defrosting and minimize ambient air contamination, the headspace in the chambers was first purged with nitrogen without sample collection (referred to as equilibration time, EQT).Following, the headspace was trapped by attaching TD-tubes (i.e.biomonitoring Tenax/Carbograph 5TD, Markes International) to the chamber exhausts (here referred to as Sampling Time, ST).When finished, the glass vials were capped and disposed.Between sampling rounds the empty Microchamber was purged with nitrogen at least 10 min to reduce contamination or memory effects of previous samples.Upon sampling TD-tubes were purged with nitrogen at room temperature to reduce water content (40 ml min −1 ).
The optimization of the Microchamber methodology aimed at optimizing EQT and ST at a constant temperature of 40 • C and flow of 50 ml min −1 .As the chambers have a volume of 250 ml, EQT was set to a minimum of 5 min to allow for effective defrosting while minimizing room air contamination.The investigated levels of EQT and ST were 5, 10 and 15 min for both.To correct the chromatograms for different collected quantities due to sampling times, splits in the TD unit were adjusted to trap 6%, 4% and 2% of the collected headspace for STs of 5, 10 and 15 min, respectively.Purging settings were determined to be 15, 25 and 35 min for ST of 5, 10 and 15 min (i.e.determined using in-house experiments, data not shown).Five replicates per subject per setting were sampled and subsequently analysed.The experiment was fully randomized, including the five Microchamber sampling blanks per condition that were considered.

HiSorb sampling and optimization
For the static headspace analysis, 250 mg of faecal aliquots were we weighed and capped in previously sterilized glass vials that priorly had been purged with nitrogen (Markes).To allow for defrosting the samples were homogeneously heated for an EQT of 5 min using the Agitator (i.e. an agitation unit to be used in conjunction with HiSorb probes).Next, the HiSorb probes were passed through the ceiled cap to statically sample the faecal headspace while being heated.Upon static sampling the probes were removed, wiped using lint free tissues, and placed in a pre-conditioned empty TD tube.Tubes were not purged with nitrogen.
The optimization of static sampling using HiSorb probes considered ST and temperature.The levels investigated for ST and temperature were 5, 10, 15 and 30 min at temperatures of 40 • C and 80 • C. Four replicates were collected per setting per subject.For the HiSorb probes the TD split was adjusted to pass 25% of collected headspace sample to subsequent chemical analysis for all collection times.Four sampling blanks per condition were obtained.

Gas chromatography-mass spectrometry (GC-MS)
Upon headspace collection, samples were analysed by GC-MS (TRACE 1300, Thermo Fisher Scientific), as described previously [29].In short, under split (i.e.TD-100, Markers International) volatiles were transferred to the GC column (RTX-5 ms, 30 m × 0.25 mm 5% diphenyl, 95% dimethylsiloxane, film thickness 1 µm), where they were kept at 40 • C for 5 min and subsequently heated up 10 • C min −1 till 270 • C, which was maintained for 5 min.Next, the separated VOCs were analysed using time-of-flight mass spectrometry (Benchtof-dx, Almsco).During the chemical analysis additional instrumental GC-MS blanks (i.e.hollow glass tube), TD-blanks (i.e.tube blanks), hollow TD-blanks (i.e.HiSorb tubes without probes), HiSorb blanks (i.e.cleaned HiSorb tube with probes), and quality controls were analyzed to safeguard quality and monitor instrument performance, and to determine the compounds truly related to the examined biological samples.

Data processing
Upon the chemical analyses data was pre-processed as described previously [30].Summarized, chromatograms were corrected for noise using wavelets and for baseline drifts using p-splines.Heavy tailing can be corrected using dynamic background compensation.Thereafter samples were normalized using probabilistic quotient normalization and aligned using correlated optimized warping.For the untargeted analysis, the pre-processed chromatograms were peak picked and log transformed to obtain the functional data matrix.For the so-called semi-targeted approach a selection of 26 representative VOCs was extracted based on the relevant mass ions, see table 1.Here, the Nist-library was used for putative identification of the VOCs.Representation was based on the set of VOCs spanning the retention time window of the entire chromatogram as well as being detectable across all subjects and typically being of interest in previous studies [15].Upon data pre-processing, the data was examined for the presence of outliers or clustering based on analytical (i.e.non-biological and not designed) variation using principal component analysis (PCA) and unsupervised random forest (RF) [31].All calculations were performed in Matlab 2018a.

Statistical analysis
To perform the optimization described, a DOE approach was utilized [32].In contrast to a strategy where one variable is optimized at a time (one variable at a time (OVAT), comparable to a grid search), DOE allows for simultaneous optimization of multiple variables while considering second order interactions and interaction effects.It does so by modelling so called response factors (Y) using the factors under investigation (i.e.X, in the current study ST, EQT, and temperature), see formula (1): with b 0 , b 1 and b 2 being the constant and linear terms, and b 1,2 , b 1,1 and b 2,2 the interaction and quadratic terms, respectively.Following formula (1) the response surface can be visualized, which constitutes the dependency of response factor Y on the individual factors.By modelling response output the experiment can be designed using a smaller number of experiments then would generally be achieved using OVAT, while maintaining more information by considering the higher-order terms.Indeed, OVAT oftentimes does not find the true optimal conditions, whereas DOE does [32].One can imagine to optimize multiple response factors simultaneously.When these point towards identical optima the solution is straightforward, whereas when these point towards different optima pareto-front optimization can be applied to decide an optimal solution [33].
Defining proper response factors is the most critical part of DOE.To be broadly applicable, the current study aimed to optimize an untargeted set of volatiles, in contrast to optimizing specific potential biomarkers for specific diseases.As such, the response factors in the current study consisted of the number of VOCs detected, the relative variation in faecal profiles between subjects (i.e. or biological variation), and the relative amount of sampling variation (i.e.noise).Each response factor was calculated twice, using the targeted and semi-targeted considerations of the obtained data.Here, performing a semi-targeted as well as untargeted analysis increases robustness of while providing an extra layer of certainty.All DOE calculations were performed using CAT software by Leardi and Melzi [34].A schematic overview of the statistical analysis is provided in figure 2 and the response factors are further explained below.

Number of VOCs
When following an untargeted strategy to find biomarkers for disease, detecting an increased number of VOCs increases the probability of including potential diagnostic markers.Therefore, this factor should be maximized.For the untargeted analysis the number of VOCs detected was based on the non-background VOCs that were detected among at least 80% of replicates per subject per condition.Here, non-background VOCs were defined as those having mean quantities in the faecal headspace higher than the 99.7% confidence interval of the mean in blanks, see formula (2): where i refers to a specific volatile, c to a specific condition (determined by EQT, ST and temperature), and b corresponding to blanks.
For the semi-targeted analysis the number of VOCs detected was defined by its inverse: sparsity.Sparsity refers to the number of missing values, or zeros, in a dataset.Sparsity was sought to be minimized.

Intra-individual sampling variation
With Intra-Individual sampling variation (here onwards: intra-variation) we refer to the variation observed between replicates within a subject at a specific condition.Here we assume it is proportional to noise and should be minimized.For the untargeted setting intra-variation was assessed by first performing PCA on all faecal samples, and subsequently calculating the average distance of a cluster of samples (i.e. the five replicates defined by their subject and condition) to their respective cluster centre.
For the semi-targeted analyses intra-variation was assessed using the multivariate coefficient of variation (MCV).MCV is the multivariate extension of the univariate coefficient of variation (CV), which can be described as the amount of noise compared to the magnitude of the signal, being the inverse of signalto-noise ratio.Here, MCV was used as defined by Albert and Zhang [35].
To reduce effects of non-significant VOCs, both measures were additionally combined with feature selection using univariate ANOVA tests.Here Benjain-Hochberg correction was performed to manage the false-discovery rate of the individual VOCs.

Inter-individual variation
Inter-individual variation refers to the variation between subjects.We therefore assume it is proportional to biological variation: the subject specific metabolic profiles.Here RF accuracy was used as a measure of inter-variation.RFs make use of decision making trees, are resistant to outliers, and have an internal validation procedure using bootstrapping combined with testing out-of-bag sample sets [36].To this aim, RFs were built to distinguish between subjects at specific conditions (i.e.build a classification algorithm to recognize the correct subjects among their replicates and those of other subjects).The better the discriminating capacity, the more subject-specific marker information is retained, whereas if discriminating capacity drops, the opposite holds true.To reduce statistical power in order to prevent RF from always achieving 100% accuracy, the decision making trees were challenged by enforcing them to keep a minimum of six samples in their final nodes, while each subject contained only five replicates per condition.The RF accuracies were used on both the untargeted as well as on the semi-targeted analyses.

Microchamber chamber effect
(V)ASCA+, the most recent update of multivariate extension of ANOVA by Camacho et al [37], was applied to test the chamber effect related to the different chambers within the Microchamber platform.

Comparison HiSorb versus microchamber
Upon having optimized both methods, faecal VOCs profiles of both collection methods were compared to each other using the number of detected VOCs and their respective univariate CVs.Here, the analysis was based on faecal material belonging to one subject and using the determined most optimal conditions.For both strategies only non-background VOCs were considered.CVs were calculated based on untargeted as well as targeted pre-processing.The authors emphasize that the CVs taken from one faecal sample do not reflect the absolute stability of the system over whole, as CVs are relative and dependent on the magnitude of the signal (which can vary between faecal samples).However, when the same material is analysed between two platforms CV can be interpreted relatively.

Optimization of headspace sampling using the microchamber
QT and ST were optimized using the three semitargeted and untargeted response factors.During the untargeted analysis, a decreasing number of features, decreasing inter-variation, and increasing intra-variation was observed for increasing EQT and ST.EQT and ST both showed similar trends, although only ST reached statistical significance during the statistical procedure.Variable selection with subsequent Benjamin-Hochberg correction for the individual VOCs did not change these outcomes.The only second order interaction that showed significance was ST and no significant interaction effects were observed.Figure 3 below shows the obtained response surface plots and the significance of the coefficients.
The results of the semi-targeted analysis were in line with those of the untargeted analysis, despite being based on only 26 VOCs.Again, only the firstand second order STs reached statistical significance, as shown in figure 4. Taken together, the most optimal conditions were concluded to be a ST and EQT of both 5 min (i.e.defined as the minimal required setting).Using these optimal conditions all VOCs reached sufficient reproducibility in at least one of the faecal profiles (CV < 25%), except propanoic acid, isovaleric acid, and pentanoic acid.The median CV across all targeted VOCs 10%.VASCA+ showed no statistical effect relating the chambers used.

Optimization of headspace sampling using HiSorb
Temperature and ST were optimized according to the same response factors used in the Microchamber optimization (i.e.again consisting of targeted and semi-targeted consideration of the number of VOCs detected, sampling variation, and biological variation).The results of the targeted analysis mirrored the semi-targeted conclusion.Here, increasing temperature resulted in an increase of intra-variation, increasing sparsity, and a reduction in the number of volatiles observed.Increasing ST resulted in an increase in inter-variation, with the second order coefficient reaching statistical significance.Here, 15 min at 40 • C were observed to be the most optimal sampling conditions.Enhancing sampling times to 30 min did not show beneficial effects.
Figures 5 and 6 visualize the obtained response surface plots and the significance of coefficients of the untargeted and semi-targeted analyses, respectively.Overall, less variation was observed between samples than was during the Microchamber optimization, showing the relatively lower responsibility of the HiSorb probes towards the factors in the DOE.Using the optimal conditions only methane amine had unacceptable CV (>25%).The median CV across all targeted VOCs was 10%.

Comparison
Lastly, faecal headspace of aliquots of one faecal sample were trapped using both collection methods and subsequently analysed.The collected chromatograms are visualized in figures 7(a) and (b) for the Microchamber and HiSorb faecal samples and sampling blanks, respectively.
Visually a greater number of VOCs in the HiSorb faecal headspace compared to that of the Microchamber faecal headspace are observed.This apparent difference may be caused by the dynamic sampling of the Microchamber, drastically increasing the captured quantities of the already dominating volatiles in the chromatogram.To prevent detector saturation, the gas fraction to be passed to the GC-MS was lowered to 4%, potentially throwing out lower quantity VOCs.Moreover, due to the vast quantities of dominating VOCs, excessive tailing is observed, which was not reduced by gas splitting.In contrast, the static sampling of the HiSorb probes in combination with a reduced affinity of more polar substances to its stationary phase, creates a better balance between the dominating SCFA and BCFA and the lower quantities of other VOCs.As a result, the relative differences in quantities are smaller and allow for better visual inspection of relatively low concentration VOCs.We speculate this increases reliability of quantifications.
Another reason for the visual greater number of VOCs observed using HiSorb probes compared to the Microchamber is the presence of background VOCs, meaning not all the volatiles detected using the HiSorb probes truly originated from faecal headspace.Many of these VOCs were detected in the respective HiSorb blanks, pinpointing their possible origin towards either room air contamination or coming from the probes themselves.Microchamber Lastly, the stability of the faecal profiles collected using HiSorb probes were compared to the stability collected using Microchamber samples.The coefficient of variation was used as stability measure.The average CV for the untargeted faecal profiles collected using HiSorb probes and Microchamber were 29% and 56%, respectively.For the HiSorb probes the respective signal comes from mixed origins: background VOC levels as well as faecal headspace, and therefore the CV is spuriously stabilized.Accounting for spuriously induced stability resulting from background signal (i.e. by subtracting the average concentrations of the respective VOC in blanks), the average CV for HiSorb probes increased to 35%.Here, 48% of VOCs (i.e.58 VOCs) had acceptable CVs (i.e.<25%).
The Microchamber resulted in drastically increased CVs, with only 5% of untargeted volatiles showing acceptable CV.A comparison of the CV of five representative VOCs confirmed to relatively higher stability of HiSorb probes, as seen in table 2. Although these results are not representative of absolute stability of the sampling system, they do show relative trends.

Discussion
The current study optimized faecal headspace collection methods by the Microchamber and HiSorb probes using statistical DOE.By applying DOE first and second order (i.e.quadratic and interaction) terms can be modelled while using a smaller number of experiments than would in general be achieved using OVAT.Moreover, by focussing the optimization towards capturing relevant variation, we believe our approach points towards more optimal conditions compared to when only one pooled sample would be utilized for optimization.In the latter scenario stability can be maximized while neglecting capturing relevant variation between samples.Due to current uncertainty of relevant biomarkers for many diseases, this optimization aimed at a maximum of VOCs detected while minimizing sampling variation and maximizing biological variation.These response factors were based on the respective targeted as well as semi-targeted pre-processing strategies.The optimal sampling strategy for faecal headspace using the Microchamber was determined to be an EQT and ST of both 5 min combined with a subsequent purgetime of 15 min.Between sampling rounds a 10 min clean purging time for the Microchamber was used to reduce memory contamination.Together, an overall sampling rate of 2.75 min tube −1 was achieved.For the HiSorb probes the most optimal sampling strategy was an ST of 15 min at 40 • C. Here, the sampling rate was 1.25 min tube −1 , more than double the rate compared  to the Microchamber.These optima were confirmed using a targeted as well as untargeted strategy.The practical implications are beyond dispute.
Comparing both sampling strategies, HiSorb probes were shown to detect a higher number of VOCs and to be more stable compared to the Microchamber.Similarly, the Microchamber showed superior characteristics in terms of signal to background ratios and background interference.However, these should be interpreted with caution, as the chambers are more susceptible to pass on previous headspace profiles to a subsequent sample (i.e.referred here as memory effects) and additional purging is necessary.For Microchamber some dominating peaks were even further amplified due to the dynamic sampling, resulting in tailing peaks.Therefore, these effects should be considered during pre-processing, whereas pre-processing of HiSorb data should focus on distinguishing biological from background signals.Table 3 summarizes the advantages and disadvantages of both methodologies.
Recently, the Microchamber was optimized according to number of peaks, peak area and peak intensity [38].Here, increasing sampling time resulted in an increasing number of peaks, larger peak areas and higher peak intensities.In turn the optimal ST was concluded to be 20 min at 30 • C and 49 ml min −1 .The current study hypothesized this set of response factors might be further improved by considering sampling variation and biological variation.Indeed, we arrived at a different optimum compared to previous authors.Moreover, the previous study does not describe TD-tubes to be dry-purged prior to GC-MS analysis, a process that might influence not only storage stability, but also the overall VOC profile [39].Not dry-purging TD tubes can, on the long-term, potentially damage MS equipment due to water sensitivity.TD Dry-purging should therefore be taken into account during method optimization.
Table Analytical characteristics of faecal headspace sampling using HiSorb probes versus the microchamber.Analytical quantities are relative and were obtained by measuring the same faecal sample using both optimized sampling methodologies.The number of non-background VOCs detected are defined by formula (1).Stability refers to average CV across all VOCs, whereas sufficient signal refers to number of VOCs with CV below 25%.Pre-processing using the HiSorb probes should focus on origin of VOCs (background versus true headspace), whereas pre-processing of microchamber samples should focus on tailing effects.S/B refers to the signal to background ratio VOCs.Microchamber samples are prone to contamination effects by the previous headspace profile in the chamber if not sufficiently purged.Time refers to relative amount of time spent on sampling.Interference refers to signals coming from the platform used or background VOCs.Faecal purging refers to the necessity of subsequent dry-blowing of the tubes.When considering biological and instrumental variation while taking dry-purging into account, the most optimal headspace strategy includes a 5 min ST, significantly speeding up sampling procedures while retaining more biological information.

Platform
The relatively low stability of Microchamber faecal VOC profiles, and the related lower number of detected VOCs, might correspond to the dynamic sampling strategy.Here, a very large range in quantities was observed between VOCs, with SCFA and BCFA, typical VOCs of saccharolytic and proteolytic fermentation, dominating.To prevent instrument saturation, very large and possibly less stable TD-splits were applied during this study.Due to these saturation effects, combined with tailing effects and subsequent purging of tubes, possibly smaller (i.e. and potentially meaningful peaks) VOCs can get sub merged and overlooked, or even purged off the tube (i.e. via breakthrough effects).Upon increasing water content, these effects might be more severe.Potentially these phenomena could result in increasing missing values [40].Moreover, we observed these effects are disadvantageous even to the dominating VOCs, shown by their relative unstableness.The HiSorb probes are less sensitive to water and polar substances, and thereby naturally resolve these challenges, decreasing the dynamic range needed to analyse all VOCs, reducing tailing effects and omitting the need for subsequent purging.However, due to interfering VOCs and large background quantities of a selection of VOCs, a more careful background-orsample consideration is needed, comparable to challenges observed currently in the breath community.
The advantage of the currently optimized methodology is that it can be used in-line with the TD set-ups that are widely used in by the Volatilomics community: i.e. especially those who perform breath metabolomics.Moreover, HiSorb probe sampling has been automated using CENTRI, and Microchamber sampling has been extended to online real time analysis to omit imperfect trapping of (very-) volatiles [2].Although we wish not to condemn one of the methods as right or wrong, altogether we hypothesize to use the Microchamber to capture extremely low quantity volatiles in a targeted setting, or its extension when worries exist regarding imperfect trapping of volatiles.Due to limited options for automation we imagine those studies to be relatively small.In contrast, in all other scenarios we recommend using HiSorb probes, which can be automated.
A limitation of the current study is the limited variety in sorbent materials used.For HiSorb probes PDMS stationary phases were chosen, whereas for the TD-tubes Markes biomonitoring sorbent material were chosen.As the biomonitoring sorbent beds are sensitive to water, this could have negatively affected the comparison; other sorbent materials are arguably less sensitive to water and might result in more stable results.However, the choice for alternative sorbent beds would simultaneously decrease the untargeted range of VOCs to be captured.Therefore, the most straightforward options were compared in the current study.
A second limitation is that both platforms were optimized independently, and then compared using an identical sample in a two-stage process.Optimally, identical samples would have been used during the optimization of both platforms.However, doubling the required amount of faecal material would have put a heavy burden on the current subjects.Not only is this practically unfeasible for some subjects (i.e. they cannot donate the required amount), but the process itself could trigger clinical symptoms (i.e.such constipation).Therefore, the optimization of both platforms was performed independently, after which they were compared using an identical sample.This approach balances the need for thorough optimization with the practical and health considerations of our subjects.
Faecal headspace analyses offer matrix to measure VOCs and relate to microbial metabolic processes.Smolinska et al previously found the microbiome to correlate to breath volatiles in a Crohn's disease population [19].Yet, microbiome analyses are typically expensive and only partly provide metabolic mechanistic insights.Oftentimes the metabolic output of the microbiome is not considered.Highthroughput faecal headspace analyses might fill this gap by offering orthogonal information.Faecal analyses have few other advantages that are worth considering.First, they show higher signal-to-background ratios for the respective VOCs compared to exhaled breath.Second, faecal headspace collections are very minimally polluted by ambient air.And lastly, despite being relative unstable as a matrix when defrosted, stool samples can be stored over long periods, whereas breath samples have very limited capacity of being stored, which results in challenges to tackle instrumental variation.

Conclusion
The analysis of VOCs for the diagnosis and monitoring of gastrointestinal diseases has shown extremely promising.Here, faecal headspace analysis provides an enrichment to other orthogonal omics platforms that analyse the microbiome.The current paper optimized and compared two faecal headspace analysis strategies the work in harmony with typical TD units omni-present across volatilomics laboratories.Although HiSorb probes and the Microchamber can both be utilized for faecal headspace collection, we recommend the usage of HiSorb probes due to their higher stability and larger volatile coverage.Moreover, the usage of HiSorb probes can be fully automated.We further suggest future studies to include a large number of randomized sampling blanks to correctly distinguish volatiles truly originating from the desired headspace.

Figure 1 .
Figure 1.(A) Front view of the microchamber thermal extactor (microchamber).(B) Top view of the Microchamber.Both panel A and B show six chambers that can be heated and purged with gas.(C) Thermal Desorption tubes can be fixed to the chamber exhausts to collect headspace volatiles.(D) Empty thermal desorption tube, used with HiSorb probes, which can be placed in such an empty tube.(E) HiSorb probes.Just below the tip, the stationary phase is visible (white).(F) HiSorb probe in glass vial.HiSorb probes can be used to sample just the headspace above the sample, but they can also be emerged in liquids.(G) HiSorb agitator that can be used to shake and heat glass vials while collecting sample headspace.

Figure 2 .
Figure2.Showing the data analytical procedure to optimize equilibration and sampling times for the microchamber, and sampling times and temperature for the HiSorb probes.For the microchamber faecal material of five volunteers was used, whereas the HiSorb was optimized using faecal material of four volunteers.Upon pre-processing and outlier detection, targeted and untargeted definitions of biological variations and noise were used to optimize the sampling methods.DOE = design of experiment, PCA = principal component analysis, CV = coefficient of variation.

Table 1 .
26targeted VOCs that were extracted from the chromatograms of faecal headspace.As internal standard 5D-bromobenzene was used.

Table 2 .
Relative CVs of 5 targeted VOCs based on the faecal headspace collection using HiSorb probes and the microchamber based on identical faecal sample.