Breath markers for therapeutic radiation

This study was approved by the South East Scotland Research Ethics Committee 01 (16/SS/0059) and all clinical staff were trained, and proficiency tested for breath analysis prior to the start of patient recruitment. The study included a total of 62 participants receiving radiotherapy for: prostate cancer (n = 40, mean age = 68.63 yr, mean radiotherapy dosage = 6000 cGy in 20 fractions); breast cancer (n = 18, mean age = 56.88 yr, mean radiotherapy dosage = 4005 cGy in 15 fractions + 1000 cGy in 4 fractions (boost to tumour bed)); and, lung cancer (n = 4, mean age = 58.25 yr, mean radiotherapy dosage = 5500 cGy in 25 fractions). Each of 62 participants were asked to provide (if possible) a breath sample before radiotherapy, and then at 1, 3, and 6 h post-exposure (240 breath samples with 58 environmental and 55 air-supply samples were acquired). Figures S1 and S2 (available online at stacks.iop.org/JBR/15/016004/mmedia) summarise participant numbers and the subsequent numbers of samples analysed, data-processed and modelled within the study. Breath samples were collected with a respiration collector for in vitro analysis (ReCIVA^TM) device (Owlstone Medical, Cambridge, UK) supplied with clean air provided from room-air purified with an activated-carbon scrubber and a HEPA filter, at an average flow of 35 dm³ min⁻¹. The air-supply unit was built and tested by the Centre for Analytical Science, Loughborough, UK. End-tidal breath was sampled for 900 s, or for a total sample volume of 1000 cm³ whichever condition was achieved first, onto a Tenax®/Carbotrap 1TD hydrophobic adsorbent tube (Markes International Ltd, Llantrisant, UK). All materials were vacuum and temperature conditioned before use to reduce exogenous VOC artefacts [25]. Training, proficiency tests and protocol checklists were used throughout to ensure compliance with the study protocol. Environmental and air-supply samples were collected with each set of breath samples (if possible). Samples were sealed and chilled to ca. 4 °C immediately after collection, and transported to Loughborough Centre for Analytical Science within 48 h. On receipt all samples were dry-purged with a 120 cm³ of purified nitrogen at a flow rate of 60 cm³ min⁻¹. During the dry-purge a six-port valve was used to load Toluene-D8 (69 pg) and trichloromethane-d (280 pg) internal standards onto thesample tube. All samples were then sealed and stored at 4 °C prior to analysis.

Samples were analysed with thermal desorption (Unity-2, Markes International) interfaced to a GC (Agilent, 7890 A) coupled to a quadrupole mass spectrometer (Agilent, MS 5977 A), see table S1 for operating details.

A GC-IMS (Breathspec, GAS) was used to evaluate the operational feasibility of monitoring thio-VOC breath biomarkers at point-of-care in the clinic, see table S2 for operating details.

Calibration/verification standards of 3-methylthiophene (CAS:616-44-4), and 2-thiophenecarbaldehyde (CAS:98-03-3) were obtained (Avocado Research Chemicals Ltd, Lancaster, UK).

Continuous instrument condition monitoring was carried out by analysing 0.2 µl of a reference mixture containing 20 standards (table S3) daily before analysis. Multivariate (Principal Component Analysis (PCA) and univariate ($Z$-scores) statistical process control approaches were used to evaluate instrument performance with respect to peak area, height, width, symmetry and signal-to-noise ratio for the 20 standards in the reference mixture. Analysis was undertaken when instruments were within $Z$= ± 3 for more than 80% of the 100 quality control parameters [26]. Responses from the gaseous internal standards were also monitored continuously to track the combined stability of the TD-GC-MS analysis and dry purging system, see figures S3 and S4.

210 TD-GC-MS data sets obtained from 62 participants including associated environmental and control samples were deconvolved (AnalyzerPro Spectral Works, UK) and 250 to 400 VOC features per sample were recorded. These features were retention index aligned and sorted and grouped using the VOCCluster algorithm [27]. Each feature was assigned an identifier of (BRI - $m/{z_1}$ $m/{z_2}{\text{ }}m/{z_3}$ ... $m/{z_n}$). BRI indicated the retention index for the VOC breath feature and $m/{z_1}$ ... $m/{z_n}$ are the nominal masses of the ion-fragments in decreasing order of abundance needed to uniquely define the feature. A preliminary untargeted metabolomic study (n = 31 participants) was used to isolate and identify candidate VOC, and a validation pilot study (n = 31 participants) tested the findings from the preliminary study.

Candidate markers were isolated and identified with an eight-step multi-variate analysis data-workflow based on the use of SIMCA-P + software with integrated 7-fold cross validation groups to protect against overfitting (Version 16.1, Umetrics, UK [28]). The work-flow comprised of: removal of rare data; identification and removal of potential environmental contaminants; data-scaling; removal of non-discriminatory features; identification of discriminatory features, compound identification; statistical characterization of candidate markers; and, scoring and classification modelling, see figure S5.

2.5.1. Removal of rare data and potential environmental contaminants for untargeted metabolomics survey

2.5.2. Scaling, and Removal of non-discriminatory features

2.5.3. Identification of discriminatory features

2.5.4. Validation pilot study

Four compounds 1-(methylsulfanyl)propane, 1-(methylsulfanyl)-1-propene isomers, 3-methylthiophene and 2,4-dimethyl-1-heptene were identified as potential markers of radiotherapy. Their abundances were aggregated into a score ${R_i}$; the ratio of the aggregated peak areas of the three thio-VOC ($\mathop \sum \nolimits^{\left( {{A_{{\text{thio - VOC}}}}} \right)}$ to the peak area for 2,4-dimethyl-1-heptene (${A_{2,4 - {\text{dimethyl}} - 1 - {\text{heptene}}}}$) (equation (1)). ${R_i}$; was used in a logistic regression model ('fitglm') with a 5-fold cross-validation. to determine a binary classification between radiated (post 6 h) and non-radiated data (MATLAB, MathWorks, USA) (equation (2)). Untargeted metabolomics survey data was used for the training set and the candidate marker data from the validation pilot study used for the testing set.

Equation (1).

$$\begin{equation}{R_i} = \frac{{\mathop \sum \nolimits^{ \left( {{A_{{\text{S - VOC}}}}} \right)}}}{{{A_{2,4 - {\text{dimethyl}} - 1 - {\text{heptene}}}}}}\end{equation} \tag{ 1 }$$

$$\begin{equation}{P_R} = \frac{1}{{1 + {e^{ - t}}}},{\text{ }}t = {\beta _0} + {\beta _1}{R_i}\end{equation} \tag{ 2 }$$

Generalized linear regression model = fitglm(X, Y, 'linear','Distribution', 'binomial') [30]

Estimated coefficients:

	Estimate	Standard Error	tStat	pValue
${\beta _0}$	2.2939	0.7237	3.1696	0.0015
${\beta _1}$	−0.3505	0.1136	−3.0848	0.0020

Here ${P_R}$ is the probability of radiation exposure and ${\beta _0}$ and ${\beta _1}$ are the logistic regression parameters. The accuracy of the class boundary was assessed using the area under a receiver operator characteristic curve (AUROC).

The Metabolomics Standards Initiative was adopted [31] and 3-methylthiophene and 2-thiophenecarboxaldehyde were identified by obtaining the retention indices and mass spectra of pure reference standards along with their calibration data (Level 1 identification, Supplementary document Figure S6). Other putative Level 2 identities were based on retention-index and mass-spectral library matches.

A total of 240 breath, 58 environmental and 55 air supply (control) samples were obtained from 62 participants. For the first 31 participants (pilot set) the number of VOC isolated and placed in the breath matrices for 1, 3, and 6 h post-radiation exposure were 230, 278 and 207 respectively (Not all of the participants were able to provide breath samples at all the time points, figures S1 and S2 in supplementary data document). Table 1 lists the discriminating VOC that were identified at Level 2 or Level 1 from the prostate multi-variate PCA analysis (figure 1 shows the PCA analysis for 3 and 6 h post radiotherapy, and figure S5 shows the OPLS-DA supervised modelling analysis and data modelling workflow). There were 188 other compounds that were difficult to identify at Level 2 and the study identifiers are listed in table S4. which is accompanied by a spreadsheet containing all of the breath matrices for the 1, 3 and 6 h post-radiation therapy.

Zoom In Zoom Out Reset image size

The exhaled profiles changed over the 6 h of the study with different VOC panels discriminating between the pre- and post-radiation states at 1, 3 and 6 h. This was not surprising for the exhaled VOC was envisioned to reflect: the development of a clinical exposome; hunger and ketosis; the consumption of drinks and snacks; the washout of radiolysis products; the cascade of VOC associated with the progression of a radiation injury from the initial targeted site; and, systemic catabolism of radiation injury products. The data from the 6 h samples was selected for more detailed study as they showed the greatest differentiation from pre-radiation samples and it was also considered that the 6 h time point was most likely to be representative of a fully developed radiation VOC panel for all the participants.

Exhaled VOC from radiated participants may be classified with non-supervised PCA, and like previous studies, methylated alkanes appear to be associated with radiotherapy exposure. Methylated hydrocarbons have been associated with reactive oxidative stress (ROS), and ionising radiation increases free-radical ROS production causing oxidative damage that may spread to neighbouring cells activating protective metabolic processes [33, 34]. The radiation induced in vivo formation kinetics of methylated hydrocarbons, and their subsequent washout, has yet to be described, and such a study would be confounded by exogenous sources of hydrocarbons. Other challenges associated with the adoption of methylated hydrocarbons as radiation biomarkers derive from the difficulty in establishing an unambiguous identity, particularly in the presence of straight chain isomers (c.f. the mass spectra and retention indexes of octane and 2,4-dimethyl heptane [32]). An additional factor in selecting a target molecule as a radiation exposure marker is the ease with which it may be detected, at low concentrations, using portable instruments; alkanes are problematic in this regard.

Seven other compounds identified at the 6 h end point as potentially discriminating were excluded from further study as it was possible that they had arisen from exogenous sources, or as a consequence of the study design. Cyclohexanone is an industrial solvent (it is also a systemic inflammatory marker [35]). Methyl furan is a fragrance/flavour ingredient, and acetic acid is a food ingredient as likely to come from a packet of potato chips in this study as an endogenous process. 2-propanone is ubiquitous and was attributed to hunger and appetite/ketosis; this is likely identifying the 6 h duration of the study. 2(5H)-furanone is a flavour component of a wide range of foodstuffs including confectionary and baked goods. 2,2-dimethyl-3-methylenebicyclo[2.2.1]heptane (camphene), (1R,5R)-2,6,6-trimethylbicyclo[3.1.1]hept-2-ene (pinene) and/or 6,6-dimethyl-2-methylenebicyclo [3.1.1] heptane (beta-pinene) have extensive exogenous sources in a clinical/healthcare facility.

Surprisingly, five exhaled thio-VOC were observed at elevated levels that could not be attributed to an exogenous source; and thio-VOC have been reported previously to be associated with radiation [36]. It should also be noted that the instrumentation and components were not selected for sulfur inertness. Consequently, some surface adsorption artefacts were evident at the lowest concentrations used in the calibrations, and as such adsorption artefacts may be assumed to have supressed thio-VOC signals for the lowest intensity features.

Dimethylsulfone, and 2-thiophenecarbaldehyde were observed after 6 h. 2-thiophenecarboxaldehyde has not been previously identified in human breath [37, 38], however, 2-thiophenecarbaldehyde was excluded from further study as it was observed in less than 20% of breath samples. Dimethylsulfone was observed in 36 participants however, the differences in abundance at pre and post exposure were not statistically significant at 6 h post exposure and therefore dimethylsulfone was excluded from further study.

42 (out of the 48 participants that provided samples at 6 h, figure S1) had detectable levels of 1-(methylsulfanyl)propane in their breath, and 32 out of these 42 participants had an increase in abundance following radiation treatment. A total of 12 participants had detectable levels of 1-(methylsulfanyl)propane only after irradiation had occurred After 6 h the average change across the cohort was a 3-fold increase with a t-value statistic of 4.20 with a one-tailed critical t-value of 1.67, p = 4.08 × 10⁻⁵ with 66 degrees of freedom.

(1Z)-1-(methylsulfanyl)-1-propene and (E)-1-(methylsulfanyl)-1-propene were also present in the 6 h post radiotherapy exhaled breath at elevated levels. The retention index for (1Z)-1-(methylsulfanyl)-1-propene was found to be 753, while the retention index of (1E)-1-(methylsulfanyl)-1-propene was found to be 754. The mass spectra obtained for the two isomers are essentially the same on the quadrupole instrument used. In the absence of further verification of correct isomer classification, the abundances for these two isomers were combined and recorded as a single thio-VOC entity.

46 participants had detectable levels of exhaled 1-(methylsulfanyl)-1-propene isomers that increased in abundance in 32 of them following exposure to radiation. 15 participants had detectable levels of 1-(methylsulfanyl)-1-propene isomers only after radiation had occurred After 6 h the average change in exhaled abundance across the cohort was 2.4-fold with a t-value statistic of 3.28 with a one-tailed critical t-value of 1.66, p = 0.0007 with 89 degrees of freedom.

3-methylthiophene was detected in the breath of 44 participants and 22 of the participants had elevated levels of 3-methylthiophene following exposure to radiation. 11 participants had detectable levels of 1-(methylsulfanyl)propane only after radiation had occurred. The 3-methylthiophene was significantly elevated at 3 h, however, after 6 h the average fold-change across the cohort was a 1.2-fold increase, and this difference was not statistically significant across the whole cohort.

Pure standards for 3-methylthiophene and 2-thiophenecarbaldehyde were commercially available and these were used to verify their identification and calibrate the instruments enabling the limits of detection to be estimated (following European Medicines Agency International Conference on Harmonization Tripartite Guidelines [39]) as 120 pg and 110 pg on-column masses respectively (supplementary data, figure S6).

Their concentration ranges observed in breath samples were between 80 ng m⁻³ and 790 ng m⁻³ for 3-methylthiophene and 380 ng m⁻³ and 3.85 μg m⁻³ for 2-thiophenecarbaldehyde, the responses obtained for 1-(methylsulfanyl)propane and 1-(methylsulfanyl)-1-propene isomers, were comparable; indicative of similar concentrations. Figure 2 is an overlaid set of chromatograms from a participant that highlight the nature and scale of the radiation associated changes in the VOC profiles observed over 6 h following irradiation.

Zoom In Zoom Out Reset image size

2,4-dimethyl-1-heptene (Level 2 identification [31]) was found to differentiate post radiation breath at 1 h, 3 h and 6 h from pre-radiation breath. 2,4-dimethyl-1-heptene has been reported to increase in concentration in cancer cell lines, and has an association with cancer cell metabolism [40]. A total of 48 out of the 62 participants were able to provide a 6 h post-exposure breath sample and the mean reduction in exhaled abundance was 72%, with a t-value statistic of 5.6 with a one -tailed critical t-value of 1.68, p = 2.19 × 10⁻⁷, n = 48 with 67 degrees of freedom. In seven participants, the level of 2,4-dimethyl-1-heptene fell below detectable levels. The question of whether this is a general radiation exposure/ROS response, or a marker of radiation damage to a tumour is one to be addressed in future work.

The reduction in the abundance of 2,4-dimethyl-1-heptene following irradiation was marked. This could be due to tumour damage and as such a marker of treatment efficacy, or it could be a generalised marker for radiation exposure. The incorporation of the thio-VOC and 2,4-dimethyl-1-heptene ratios provides a radiation exposure classification approach for this cohort and by implication people with cancer receiving radiotherapy. The average calculated ${R_i}$-value of all the data (both discovery and validation pilots) was 3.71 at 0 h and 54.37 at 6 h with a t-value statistic 3.21 with a one -tailed critical t-value of 1.68, p = 0.001 n = 48 with 47 degrees of freedom). The area under the receiver operator characteristic (AUROC) value for the training (discovery) and testing (validation) sets were 0.91 and 0.75, respectively (figure 3).

Zoom In Zoom Out Reset image size

Further studies into the verification, development and biokinetics of thio-VOC production after radiotherapy will require point-of-care monitoring and GC-IMS is a candidate technology for this role. Details about GC-IMS and potential applications in different clinical settings are available in the book 'Ion Mobility Spectrometry' [23] and a recent review [41]. A single breath GC-IMS approach has recently been benchmarked following the 'Peppermint Protocol' [42] and this methodology was assessed for at-patient exhaled thio-VOC determination. The limit of the GC-IMS detector's response was estimated to be 210 fg s⁻¹. TD-GC-MS generates a 10⁴ enrichment from a 1 dm³ breath sample while the GC-IMS does not use an enrichment stage with an injected volume of 0.5 cm³ sample of end-tidal breath. Assuming a Gaussian elution profile with a 6 s peak width concentrations of 10 μg m⁻³ may be detected currently; with some evidence that sulfur-surface interactions inhibit detection at lower levels with the existing instrument configuration and method [39], figure 4.

Zoom In Zoom Out Reset image size

Further refinement and optimisation of sample volume and sampling materials/handling along with chromatography parameters are likely to be sufficient to deliver the analytical performance needed. Using a single breath sampling approach is likely to be tolerable for most radiotherapy patients during an extended breath sampling study to detect and monitor thio-VOC profiles during treatment.