Reproducible breath metabolite changes in children with SARS-CoV-2 infection

SARS-CoV-2 infection is diagnosed through detection of specific viral nucleic acid or antigens from respiratory samples. These techniques are relatively expensive, slow, and susceptible to false-negative results. A rapid non-invasive method to detect infection would be highly advantageous. Compelling evidence from canine biosensors and studies of adults with COVID-19 suggests that infection reproducibly alters human volatile organic compounds (VOCs) profiles. To determine whether pediatric infection is associated with VOC changes, we enrolled SARS-CoV-2-infected and -uninfected children admitted to a major pediatric academic medical center. Breath samples were collected from children and analyzed through state-of-the-art GCxGC-ToFMS. Isolated features included 84 targeted VOCs. Candidate biomarkers that were correlated with infection status were subsequently validated in a second, independent cohort of children. We thus find that six volatile organic compounds are significantly and reproducibly increased in the breath of SARS-CoV-2-infected children. Three aldehydes (octanal, nonanal, and heptanal) drew special attention, as aldehydes are also elevated in the breath of adults with COVID-19. Together, these biomarkers demonstrate high accuracy for distinguishing pediatric SARS-CoV-2 infection and support the ongoing development of novel breath-based diagnostics.


Thermal desorption and GCxGC parameters
Prior to analysis, sorbent tubes were brought to room temperature and loaded into autosampler (Utra-xr, Markes International, UK). A gaseous standard mixture (1.01 ppm Bromochloromethane, 1.04 ppm 1,4-Difluorobenzene, 1.04 ppm Chlorobenzene-D5, 0.96 ppm 4-bromofluorobenzene) was immediately added to each tube, followed by a purge predesorption step consisting of 10 min with He at 50 mL*min1, to remove water content in breath samples. Tubes were thermally desorbed for 10 min at 270°C (Unity-xr, Markes International, UK) and transferred to a "Universal" cold trap which matched the sorbent of the sample tube, held at 10°C and subsequently heated to 300°C, to minimize band broadening. The split flow after the cold trap was 15 mL*min-1.
Analysis by two-dimensional gas chromatography was conducted using an Agilent 7890B GC system, fitted with a flow modulator and a three-way splitter plate coupled to a flame ionization detector and a time-of-flight mass spectrometer with electron ionization (SepSolve, UK).
Chromatographic analysis was performed using a Stabilwax (30 m × 250 μm ID × 0.25 μm df) as the first dimension (1D)-GC column and a Rtx-200 MS (5 m × 250 μm ID × 0.1 μm df) as second dimension (2D)-GC column, both purchased from Restek (Bellefonte, PA, US). The following GC oven temperature program was used: initial temperature 40°C and held for 1 min, ramped to 260°C at 3°C*min-1. The final temperature of 260°C was held for 1 min. The total run time for the analysis was 75 min. Helium carrier gas was flowed at a rate of 1.2 mL*min-1. The flow modulator (Insight, SepSolve Analytical, UK) had a loop with dimensions 0.53 mm i.d. x 110 mm length (loop volume: 25 uL), and the modulation time was 2 s total.

TOF-MS Conditions
The GCxGC was interfaced with a BenchTOF-select time-of-flight mass spectrometer (SepSolve Analytical, UK). The acquisition speed was 50 Hz and mass range was 30-400 m/z.

Chemical standards and solutions
Nonanal, octanal, heptanal, tridecane, and 2-pentylfuran and isoprene were purchased from Sigma-Aldrich (St. Louis, MO, US). Dodecane was purchased from Merck (Darmstadt, Germany). To spike the compound of interest into a sorbent tube, a 10 ppm solution was prepared in HPLC grade methanol. Using a solution loading rig (Markes International Limited, UK), 1 μL of the solution was spiked into a sorbent tube. The sorbent tube was flushed for 3 min with nitrogen at a flow of 100 mL.min-1. All the stock solutions were stored in glass vials and kept at 4 °C. Sorbent tubes containing standards were analyzed by GCxGC BenchTOF-MS following the same protocols as described below for breath samples.

Quality control
Breath concentration of the canonical human volatile isoprene was performed to quality control for correct breath sampling, as a small or missing isoprene peak indicates an error in the sample collection and/or analysis, resulting in data being excluded. To check for changes in instrument sensitivity over time, a mixture of external standards was analyzed with the GCxGC BenchTOF-MS alongside the breath samples as described previously (1)

Data processing and statistical analyses
Data was acquired and processed using ChromSpace (SepSolve Analytical, UK). All statistical analyses were performed using RStudio v1.3.1073 (PBC, Boston, MA) and GraphPad Prism V.8.4.3 (GraphPad Software, San Diego, CA). The workflow for data processing and statistical analysis is shown in Supplementary Figure 2. Background from the raw BenchTOF data file was removed using ChromSpace, and the Dynamic Background Compensate (DBC) of 0.2s peak width and noise factor 6.9 for typical GCxGC data was applied. DBC files were then integrated using the following parameters: peak detection deconvolution algorithm with a minimum ion count of 2000, absolute minimum peak area was set at 15,000 counts, absolute minimum peak high was set at 10,000 counts, and no relative threshold was set for either mass height or absolute area. Compounds were given annotations using the NIST v.17 reference library.
Deconvoluted peaks were exported into .xls format file. The data were then processed using RStudio to generate integrated signal for every isolated feature. Isolated features included 84 targeted volatiles, as described below.
We targeted volatiles that have been previously associated with respiratory viruses from cell culture, from analysis of in vitro airway cells infected with human rhinovirus, in vivo breath profile in swine during Influenza A infection, previously associated breath markers of COVID-19 in adults, known human body volatiles (2)(3)(4)(5)(6)(7)(8), and authors' own unpublished breath VOC library (Supplementary Table 2). The data included three internal standards (Supplementary Table 2).
Chromatographic data was first normalized using internal standard (1,4-difluorobenzene), and a volatile was retained if it was present in more than 50% of the samples in either group (i.e. infected or uninfected). In total, 50 VOCs were retained and used for further statistical analysis.
Unpaired Student's t-test was used to identify metabolites that were significantly different between control groups and SARS-CoV-2 groups, with a p-value of 0.05 established as the threshold for statistical significance. Of note, multiple comparison corrections of metabolomic data can increase type II errors, because metabolites are typically highly correlated and not independent features(9, 10).