APPI-MS Interface and Novel Bias Electrode Characterisation
Figure 1 shows a cross-section of the inner structure of the APPI interface. The UV lamp (8) and driver electronics were harnessed from a commercial APPI source (ThermoFisher). The grounded metal collar (7) surrounding the lamp (8) forms a gas tight seal with a bespoke 3D printed enclosure (6). A photopolymer resin material was used rather than the more common extruded (fused deposition) plastic type due to better outgassing characteristics of UV-cured resin when under UV irradiation (from the APPI lamp). The lamp is inserted such that the front face of the lamp aligns with the back edge of the 3D printed chamber. Here a thin (0.4 mm) metal electrode (4) is located to apply a positive potential bias, with respect to the inlet (1), to confer positive ions created in the ionisation chamber a drift velocity in the direction of the MS inlet. PTFE spacers (10) with holes drilled to enable gas entry and exit provide an ionisation volume directly in front of the MS inlet. A second thin electrode (3) provides the reference potential for the bias voltage and is electrically connected to the sampling cone of the mass spectrometer. Finally, a 3D printed piece (2) push-fits over the sampling cone, and threaded rods/nuts compress each element into a gas tight sampling chamber. It should be noted that the final 3D printed piece can be readily designed to fit onto any mass spectrometer atmospheric pressure interface (API) by changing the diameter of the exit orifice to suit. Rubber bungs with stainless steel and PTFE hoses provide gas entry (5) and exit (9) from the source via either pressurisation from the inlet region or evacuation by a small diaphragm pump at the exit.
An offline characterisation of the ion distribution generated by the UV lamp was performed to aid design of the APPI-MS interface. Figure 2 depicts the resultant application of a potential bias between UV lamp and segmented detector. A steady and substantial increase in total measured ion current (cumulative ion current hitting individual strips) as electric field intensity is increased from 0 V/mm to approximately 50 V/mm is observed, which subsequently plateaus for higher field strengths. The increasing force and therefore ion velocity due to the increase in applied field yields an order magnitude increase in ion transmission to the detector. This is likely, at least in part, due to limiting recombination processes between positive and negative ions in the irradiance chamber. Plotting the ion current measured on individual strips enables visualisation of the ion beam cross-sectional area. Figure 2 insert shows the ion distribution; the ion current is plotted for each strip with the applied voltage between the bias electrode and detector set at 1000 V, corresponding to an electric field intensity of 50 V/mm. The irradiance diameter at maximum ion transmission was approximately 14 mm, decreasing from 35 mm under no potential bias conditions. The diameter of the internal irradiance chamber was therefore constructed to be 20 mm to avoid charging of the insulating 3D printed plastic. Some surface charging of the internal wall structure is expected and we expect this will act to further confine the ion cloud diffusion50, thereby increasing the quantity of ions sampled by the MS inlet. Furthermore, utilising the smallest irradiance volume possible also reduces neutral analyte diffusion, thereby potentially improving ionisation efficiency due to increased analyte concentration.
Parameter Optimisation
To establish the optimal operating conditions, a series of experiments were performed to assess each of the tuneable parameters in the design for a range of compounds related to breath analysis. Ethanol, acetone, ethyl acetate, 2-butanone and eucalyptol were examined. These compounds were selected to encompass a range of breath related VOCs, with a mass range from 47 to 155 amu and boiling points ranging from 56°C to 176°C.
Cone Bias Voltage
Cone voltage and potential bias electrode are coupled parameters therefore they were optimised in tandem; Figure 3 shows the signal intensity heatmaps for each compound examined. Supplementary Figure S2 shows the average mass spectrum extracted from the maximised bias and cone voltage experiments for each analyte. Applying potential bias between lamp and inlet improves the signal intensity by a factor of ~10 for all analytes examined. Maximised signal intensity occurs at a bias voltage of 200 V for each compound. It would be of interest to examine higher mass analytes to establish if a broad mass dependency exists but this is out of scope for purposes of this present study. Water (m/z 37) and ethanol (m/z 47) both yielded a narrow band of cone voltages that gave relatively high signal intensities from 10 V to 20 V and 15 V to 25 V, respectively. Outside of these narrow ranges the signal intensity dropped off significantly. Acetone and 2-butanone also shared a signal intensity response but instead of bands, in the heat map depiction, they formed concentric circles of increasing signal intensity, peaking at 200 V and 35 V and 200 V and 30 V for bias and cone voltages, respectively. Acetone and 2-butanone showed a higher degree of tolerance towards unoptimised conditions than water or ethanol. Ethyl acetate and eucalyptol exhibited similar concentric circular profiles to acetone and 2-butanone but with a smaller tolerance for unoptimised parameters. Optimum values for bias and cone voltages for ethyl acetate and eucalyptol were both 200 V and 20 V, respectively. Thus, for the remainder of the study, a 200 V bias voltage and 20 V cone voltage were selected to give broadly optimal transmission (~10-fold increase compared to no bias electrode) over the mass range of interest.
Carrier Gas Flow Rates
Introduction of standards into the APPI chamber is conducted via dosing liquid analytes into an N2 carrier gas at precise flow rates using a syringe pump driver (SS Scientific). A 1/16” stainless steel capillary is concentrically inserted into a ¼” stainless steel tube and fixed using Swagelok compression fittings. The flow rate of the carrier gas is adjustable via a mechanical variable area flow meter (Brooks Instruments) within the range 0-5 Lmin-1. The liquid solution is dispensed to the end of the capillary where subsequent nebulisation and transportation to the APPI lamp is facilitated via the carrier gas. A tubular heating element is placed outside of the ¼” tubing to aid vaporisation of the analyte. No significant carryover is observed when the syringe driver is stopped; after a few seconds the analyte signal returns to background level.
A series of experiments were performed to determine the response of the system to sample and gas flow rate changes for a 20 ppm solution of eucalyptol in water. Figure 4 shows the signal intensity of the [M+H]+ protonated molecular ion for eucalyptol for each gas flow rate tested. The amount of eucalyptol introduced into the carrier gas stream was 0.5, 1, 2, 5 and 10 µLmin-1 corresponding to 15.4, 30.7 61.4 153.5 and 307.0 pg of analyte. The nitrogen gas flow rate was varied between 1 and 5 Lmin-1 in 1 Lmin-1 steps. For carrier gas flow rates above 3 Lmin-1, the signal response was linear across the range investigated (R2 = 0.996, 0.998 and 0.995 for 3, 4 and 5 Lmin-1, respectively). For 1 and 2 Lmin-1, a reduced upper limit of linearity was observed. The loss in dynamic range is attributed to the higher concentration of water vapour in the gas stream. It is well known that increased solvent concentration supresses analyte signal in APPI due to absorption of photons by the much higher concentration of solvent51,52. The choice of water here was intended to gauge the applicability of the system for future breath analysis which contains a relatively high moisture content. Increased signal response for lower carrier gas flow rates is due to a reduced dilution of the analyte in the carrier gas stream. A carrier gas flow rate of 5 Lmin-1 was used to produce calibration curves, whilst bacteria headspace sampling was carried out using a reduced carrier gas flow rate of 0.2 Lmin-1 to improve sensitivity.
Vaporisation Temperature
The final element investigated to determine optimal operation was the vaporising heater temperature. A series of experiments were performed by increasing the vaporising heater temperature from 30 °C to 190 °C in 20 °C steps. 20 ppm solutions of acetone, 2-butanone and eucalyptol in water were individually prepared for optimisation. Each analyte was fed into the capillary and the temperature allowed to stabilise before a measurement was initiated. Figure 5 (a) shows the signal intensity of m/z 59, 73 and 155 peaks relating to acetone, 2-butanone and eucalyptol, respectively, for each temperature set. The intensity values generally increase with increasing temperature for all analytes. Presumably this is due to more efficient vaporisation of the analyte which reduces any condensation losses onto the hoses and/or chamber structure. In the case of eucalyptol this increase was continuous as temperature increased, however acetone (b.p. 55.8°C) and 2-butanone (b.p. 79.5°C) peaked at ~130 °C and ~150 °C, respectively, before declining, possibly due to thermal degradation of these analytes. Eucalyptol has the highest boiling point of the three analytes assessed. Whilst increasing the temperature appears advantageous in terms of individual analyte sensitivity, Figure 5 (b) depicts the relative height of the peaks in respect to the total ion current. It can be observed that increasing the temperature reduces the signal-to-noise ratio of the peaks of interest, possibly due to the other system contaminants being thermally desorbed from the APPI chamber material and gas delivery system. Since the goal of the present study is to determine the metabolite profile of different bacterial samples, the temperature was fixed at 70 °C to avoid the emergence of spurious peaks at the expense of sacrificing some sensitivity.
Quantification and Limits of Detection
Calibration curves were produced for each compound diluted in ultra-pure water. Individually, each analyte (ethanol, acetone, 2-butanone, ethyl acetate and eucalyptol) produced a highly linear calibration curve with a coefficient of determination ≥ 0.99 (supplementary information, Table S1). The calibration curves can be found in supplementary information Figure S2; ethanol produced a linear response between 31 ppbv and 315 ppbv, acetone and 2-butanone gave linear responses between 2 ppbv and 25 ppbv, ethyl acetate and eucalyptol also gave linear responses between 3 ppbv and 36 ppbv.
Ethanol can be found in breath due to bacterial activity in the gut53,54 with expected concentrations in healthy breath between 10-1000 ppb. Ethanol’s presence can also give information about lifestyle such as recent consumption of alcohol. Acetone is another VOC naturally found in breath at approximately 1-1000 ppbv54 with elevated concentrations greater than 1800 ppbv corresponding to patients with diabetes mellitus.11,12 In isolation acetone detection is permissible with a limit of detection (LOD) of 6.8 ppbv and limit of quantification (LOQ) of 27.8 ppbv. The APPI-MS method is also suitable for the detection of 2-butanone with an LOD of 1.6 ppbv and LOQ of 6.5 ppbv, which is normally present in the breath of healthy people at approximately 20 ppbv5,55 and is found in the headspace of bacteria samples of Pseudomonas aeruginosa.41 Some have suggested ethyl acetate is a potential marker related to lung disease.56 When analysing the breath of these patients, ethyl acetate may be present in concentrations up to 100 ppb5 which is undetectable in the breath of a healthy person. Our method is suitable for ethyl acetate analysis with an LOD of 0.7 ppbv and an LOQ of 5.0 ppbv. Finally, eucalyptol was detectable at 0.9 ppbv and quantifiable at 4.8 ppbv. Eucalyptol was included for future reference, as it is not expected to be found in breath naturally, but can be found if an individual has recently consumed mint.
Bacterial Culture Classification
To assess the suitability of the apparatus for potentially determining the bacterial origin of lung infections in CF patients, PSA and SA cultures were prepared and sampled using 1 L Tedlar bags as described in the methods section. A further cautionary note on Tedlar bag suitability for direct analysis, such as ambient ionisation techniques, can be found in supplementary information. The resultant collected headspace was evacuated from the bag and passed through the APPI chamber by pumping via a small diaphragm pump. Samples were collected and analysed in batches of four over a three-week period. A new culture was initiated each Monday and samples collected on the subsequent Friday, in total 12 of each type of bacterial headspace was sampled. Escherichia coli (EC) which is unrelated to CF was included as a control and as a means to improve the robustness of the classification. Figure 6 shows centroided spectra for one PSA sample and one SA sample; only peaks with relative intensities above 20% were retained for display purposes. Immediately obvious are a number of distinct visual differences between samples with a significant number of peaks appearing in only one of the samples. This gives confidence that a classification model can successfully be applied to the dataset. The total ion chromatograms and time indexes of extracted spectra for all samples are shown in supplementary Figures S3-5.
After pre-processing spectral data as outlined in the methods section, 247 peaks were found across all samples and included in a principal component analysis (PCA) model. A data table containing 247 dependant variables and 36 observations was collated. PCA was compiled to visualise spectral differences and dimensionally reduce the dataset. The first three principal components accounted for 86.9 % of the total variance which is an excellent result. A PCA biplot can be seen in Figure 7 displaying excellent separation and grouping of sample classes with clear class boundaries evident for all 3 groups. Principal component (PC)1 is the discriminating component for EC and PSA whilst PC2 was responsible for separation of SA. The remaining PCs (not shown) were not found to differentiate between the sample classes.
Following PCA, a linear discriminant classification model was built using the 247 features. 10 times cross validation was used to avoid overfitting and improve robustness. 100% of samples were correctly classified by the model. This was a very pleasing and significant result, since the samples included multiple cultures, with sample collection compiled and analysed over several weeks. A further set of three SA and three PSA headspace samples was acquired a week later from a fresh batch of cultures. Class predictions were made by the model generated from the training data set in a blind study. 100% of the blind study samples were correctly classified. The results are comparable to that of current SESI-MS and SIFT-MS methods.40,43 The confusion matrix showing classification results is shown in supplementary Figure S6. Since our laboratory is not designated to handle Class 2 bacterial cultures, it was not possible to conduct online sampling/analysis. Therefore, non-ideal Tedlar bag sampling was used to collect the headspace (Figures S7 & S8). Presumably even better results could have been achieved with direct headspace sampling. Nevertheless, a low-resolution mass spectrometer and without requiring tandem mass analysis, this approach has correctly identified SA and PSA in real-time and directly from gaseous samples, without the need of reagents, such as dopants previously identified in APPI studies or reagents for SESI. This shows the excellent potential of this approach to be extended for online breath analysis in-clinic with a portable system; for example, to aid diagnosis or to monitor treatment effectiveness.