Direct analysis of volatile organic compounds in foods by headspace extraction atmospheric pressure chemical ionisation mass spectrometry

Rationale The rapid screening of volatile organic compounds (VOCs) by direct analysis has potential applications in the areas of food and flavour science. Currently, the technique of choice for VOC analysis is gas chromatography/mass spectrometry (GC/MS). However, the long chromatographic run times and elaborate sample preparation associated with this technique have led a movement towards direct analysis techniques, such as selected ion flow tube mass spectrometry (SIFT‐MS), proton transfer reaction mass spectrometry (PTR‐MS) and electronic noses. The work presented here describes the design and construction of a Venturi jet‐pump‐based modification for a compact mass spectrometer which enables the direct introduction of volatiles for qualitative and quantitative analysis. Methods Volatile organic compounds were extracted from the headspace of heated vials into the atmospheric pressure chemical ionization source of a quadrupole mass spectrometer using a Venturi pump. Samples were analysed directly with no prior sample preparation. Principal component analysis (PCA) was used to differentiate between different classes of samples. Results The interface is shown to be able to routinely detect problem analytes such as fatty acids and biogenic amines without the requirement of a derivatisation step, and is shown to be able to discriminate between four different varieties of cheese with good intra and inter‐day reproducibility using an unsupervised PCA model. Quantitative analysis is demonstrated using indole standards with limits of detection and quantification of 0.395 μg/mL and 1.316 μg/mL, respectively. Conclusions The described methodology can routinely detect highly reactive analytes such as volatile fatty acids and diamines without the need for a derivatisation step or lengthy chromatographic separations. The capability of the system was demonstrated by discriminating between different varieties of cheese and monitoring the spoilage of meats.

with pre-concentration techniques such as thermal desorption. 17 However, GC/MS has a number of drawbacks including the need for molecular derivatisation of reactive functional groups such as carboxylic acids and amines and potentially long chromatographic runs (30-60 min) for complex analyte mixtures. Real-time mass spectrometry techniques such as PTR-MS and SIFT-MS have also been used for this type of analysis, and their potential for monitoring volatiles in real-time has been proven. [18][19][20][21] In addition, the development of innovative methods using standard ambient ionisation sources such as atmospheric pressure chemical ionisation (APCI) and electrospray ionisation (ESI) has considerably increased in recent years. [22][23][24][25][26][27][28][29] These ionisation sources allow for the interfacing of volatile analysis with mass spectrometry without requiring specialised instrumentation limited to analysing VOCs. For example, APCI has been widely used for the analysis of volatiles. 26,27,[29][30][31] Extractive electrospray ionisation (EESI) has also been used to analyse volatiles in different matrices including beer, 32 breath, 23,24,29,33 an active pharmaceutical ingredient, 34 and fragrances. 35 This ionisation method has also enabled the detection of non-volatile compounds such as nicotine, 25 urea, 24 and creatinine. 36 The work presented here describes the design and construction of an interface for the direct introduction of VOCs into the APCI source of a compact quadrupole mass spectrometer. This modification accommodates the introduction of volatiles using a new jet-pump designed in-house. The pump operates on the basis of the Venturi effect, which enables the introduction of volatiles into the mass spectrometer via the APCI gas line. The suitability of the interface for a variety of qualitative and quantitative applications is demonstrated.
Volatile profiles for compounds such as fatty acids and biogenic amines, which are challenging to analyse by GC/MS without derivatisation, were obtained with minimal effort using our design.
The food industry has shown a particular interest in rapid analysis to differentiate the freshness of different foods by measuring the release of specific volatiles. 4,13,37 Currently, there is also a high demand for new methods to screen meat products from supermarket chains, which would enable discrimination between different types of meat and the identification of adulteration. Ambient ionisation methods such as liquid extraction surface analysis/mass spectrometry (LESA/MS) 38 and rapid evaporative ionisation mass spectrometry 39 have been implemented for this purpose, enabling rapid in situ detection, and have proven to be useful in authenticating meat products. In this application different cheese samples and meat samples were used to demonstrate the capabilities of the interface for fatty acid and biogenic amine profiling in the gas phase.

| Forensics
The characterisation of volatile profiles that identify human remains is of significant interest in forensic research. It is widely accepted that body decomposition starts shortly after death. 40 Part of the decomposition process is constituted by the self-digestion of cells, where proteins are released and broken down by bacteria into their building blocks. From these processes several compounds are released, amongst these are putrescine, cadaverine, phenol, indole, butanol, and octanal. 41 Pig carcasses have been proposed as human body analogues to serve as training aids for human remains detection dogs. 41 Therefore, pork has been selected for the decomposition studies presented here. This particular application focused on monitoring the production of indole in decomposing meat due to deamination of tryptophan.
Volatile fatty acids, indole, cadaverine, putrescine, formic acid, ammonium acetate, methanol, acetone, ethanol, and water were acquired from Sigma Aldrich Ltd (Gillingham, UK). Food samples were acquired from a local supermarket. Standards were used to tune the instrument, optimise the conditions for each experiment and confirm the identity of analytes.

| Compact mass spectrometer: direct VOC sampling interface
Experiments were performed using an in-house designed Venturi pump coupled to the APCI probe of an Expression single quadrupole compact mass spectrometer (Scheme 1) from Advion, Inc. (Ithaca, NY, USA). The gas line, sample inlet and jet pump were heated to 180°C using a commercially available heated transfer line (Quantitech Ltd, Milton Keynes, UK). The APCI nitrogen gas supply flow rate was adjusted to 4 L/min, which gave a measured suction from the sample inlet of over 500 mL/min. The instrument was operated in both negative and positive ionisation mode depending on the target analytes. Table 1 shows the mass spectrometer conditions used for the studies reported here. All the spectra were acquired within the range of m/z 30-300 and blank measurements were acquired for each case.

| Optimization of the interface
The principal criteria considered in the development of the volatile interface were that the condensation of volatiles onto cold surfaces should be prevented and that the interface can be readily coupled to the APCI source of the compact mass spectrometer. To address the first criterion the system was heated above 70°C using nitrogen gas heated by a commercially available heating jacket (Quantitech), see Scheme 1.
In order to connect the Venturi pump to the APCI source a bespoke connector was used. The collection of volatiles was achieved by using Technologies, Stockport, UK) was used to calibrate the system.

| Temperature studies
Temperature proved to be a critical variable in the development of the interface. The optimisation was carried out by setting the sample line at temperatures of 40°C, 70°C and 100°C. The latter temperature was shown to be the optimal temperature across a number of applications, since it prevented moisture from building up in the transfer line.

| Cheese study
Four different cheeses, Red Leicester, Wensleydale, Blue Stilton and Goats, were chosen for this study. The samples were cut into 1 × 1 cm sections. Approximately 2 g of each cheese was placed into a 20-mL headspace vial. The samples were heated to an optimal temperature of 70°C for 2 h. Sampling was achieved by connecting the headspace vial to the sample inlet as shown in Scheme 1. To promote formation of negative ions, 50:50 methanol/water +10 mM ammonium acetate was infused into the APCI source at a rate of 10 μL/min. Samples were analysed for 5 min; a blank vial was analysed for 5 min prior to each analysis to enable subtraction of the background from the sample mass spectra.

| Multivariate analysis
Spectra were obtained using the experimental conditions described above. Mass spectra were extracted from the Advion Data Express SCHEME 1 Schematic diagram of the compact mass spectrometer APCI source. The APCI gas line and sample inlet line were accommodated inside the heated transfer line and directed to the Venturi pump  observed for all of the VOCs measured directly from foods in this study and probably occurred because of variable rates of VOC release from the un-homogenised food samples. In addition using these conditions minimal sample carryover was observed between samples when the sample line assembly was maintained at 100°C even with the some of the larger and less volatile species (such as octadecanoic acid) analysed in these experiments. The system has been tested with difficult compounds and two different applications of this versatile design will be discussed in the following sections.

| Detection of fatty acids in cheese
Real-time screening of volatiles in cheese was possible with no sample pre-treatment. The cheese headspace was directly analysed from the vials using the volatiles interface. Figure 2 shows the full scan mass spectrum of the volatile profile obtained for the selected cheeses.
Common volatiles found in all the samples are shown in Table 2

| Multivariate statistical analysis
Following APCI-MS analysis the cheese varieties were compared against each other individually using a supervised partial least-squares discriminant analysis approach. This approach forces a separation between the two groups and enables the construction of s-plots which can be used to identify the variables which contribute most to the separation. This enabled the construction of a 339 variable model (Table S1, supporting information) that reflected the differences between each of the four cheese varieties.      detected at day 0 and observed to increase dramatically over the next 2 days. The highest response for these diamines was observed at day 2 and it began to decline after. Trimethylamine was also detected at m/z 60 (Figures 4 and 5). This amine is highly volatile compared with the other amines, which thus have lower vapour pressures at room temperature, and it is observed at low levels in all the spectra obtained after day 1. In some pork samples aged for a longer period ( Figure 6) a peak assigned to indole was also reliably detected after 6 days of decomposition. Figure 5 shows

| Quantitation of indole in decomposing pork
The volatiles interface developed in this study is completely open to the atmosphere. As a consequence, the humidity in the source can change depending on the laboratory environment, and other species  Figure S2 (supporting information) we can determine that indole is present at 14.1 μg/g pork at day 6 and at 51.8 μg/g pork at day 7. The level of indole present at day 5 is too low to be quantified by the system. Please note that these figures assume that 100% of the indole is extracted from the pork. This is unlikely with a complex matrix and the level of indole in the pork may be higher than these results; however, this does demonstrate the capability of the system to quantitatively monitor the level of indole over the 7-day period. The linear dynamic range of the system can potentially allow for quantitation of the volatiles.
However, care needs to be taken by operating the system under regulated conditions. Another problem that may arise using the external standardisation shown here is that, without any pre-separation of analytes passing through into the source, ion suppression effects may impact negatively on the quantitative performance. This would particularly affect low proton affinity analytes as other species present with higher proton affinity will sequester the charge in the ion source.
Indole has a relatively high proton affinity of 933.4 kJ/mol 44

| CONCLUSIONS
The direct analysis interface system is a simple and cost-effective alternative to traditional techniques for the introduction of volatiles into the mass spectrometer. The capability of the system for direct analysis of fatty acids and biogenic amines from decomposing meat samples was successfully demonstrated. These analytes, which are highly reactive and problematic to analyse directly with traditional methodologies such as GC/MS, were detected without the need for any derivatization reaction or chromatographic separation. The system is shown to be capable of discriminating between different varieties of cheese using an unsupervised principal component analysis model highlighting a potential application of the deployable instrumentation in food authenticity and provenance testing. Quantitative analysis was also achievable using external standardisation for indole and is shown to be capable of limits of detection in the ng/mL range, but only when the system was under controlled headspace conditions. Due to the absence of a separation method, competitive ionisation effects in the ion source will also limit the quantitative performance of this approach; however; this issue can be overcome in targeted studies by using an isotopically labelled internal standard.
The example applications demonstrated here display the versatility of this approach for the rapid and semi-quantitative analysis of VOCs from food samples. This study shows that the low-resolution Advion compact mass spectrometer possesses the requisite sensitivity and selectivity for food authenticity and safety testing when used in conjunction with the VOC sampling interface. The combination of this interface with a compact transportable mass spectrometer suggests this method has the potential for deployment in the field, enabling on-site testing and reducing the need to send samples to specialised laboratories for analysis by high-resolution methods, thus greatly improving the speed and throughput of analysis.

ACKNOWLEDGEMENT
The authors are grateful for financial support from the EPSRC, which was awarded as part of the impact acceleration account.