Detecting bacterial lung infections: in vivo evaluation of in vitro volatile fingerprints

The strains used in this study were P. aeruginosa PAO1-UW, P. aeruginosa FRD1 and S. aureus RN450 (courtesy of Professor G L Archer, Virginia Commonwealth University). For in vivo test preparation, strains were incubated aerobically in tryptic soy broth (TSB; 16 h, 37 °C, 200 rpm; final OD₆₀₀ > 3.0). For in vitro test preparation, strains were incubated aerobically in TSB (16 h, 37 °C, 200 rpm) whereafter 50 µL was used to inoculate 50 mL TSB for 24 h (16 h, 37 °C, 200 rpm, OD₆₀₀ > 3.0).

Six- to eight-week-old male C57BL/6J mice were purchased from The Jackson Laboratories (Bar Harbor, ME). All mice were housed in the Association for Assessment and Accreditation of Laboratory Animal Care-accredited animal facility at the University of Vermont (Burlington, VT). The protocol for animal infection and respiratory physiology measurements was approved by the Institutional Animal Care and Use Committee, in accordance with Association for Assessment and Accreditation of Laboratory Animal Care guidelines.

An acute airway exposure model was applied in this study. Briefly, overnight cultures of PAO1, FRD1 and RN450 were measured for OD₆₀₀, centrifuged at 13 000 rpm for 1 min, washed twice with phosphate buffer solution (PBS), and resuspended to give the desired concentration of bacteria (5 × 10⁶ CFU for PAO1, 1 × 10⁷ CFU for FRD1 and 1 × 10⁸ CFU for RN450) in 40 µL PBS. Mice were briefly anesthetized (isoflurane by inhalation) and infected by oropharyngeal aspiration as described previously [37, 38]. Uninfected mice were exposed to 40 µL PBS as a negative control. Five mice per group were exposed and testing was conducted on two different days. After breath collection, the lungs were harvested and homogenized, and the lung bacterial cell counts were obtained by plating on selective media, yielding averages of 1 × 10⁵ CFU/lung for PAO1, 5 × 10⁵ CFU/lung for FRD1 and 2 × 10⁶ CFU/lung for RN450.

The mice were anesthetized with pentobarbital 24 h after infection and their tracheas were cannulated. The mice were placed on the ventilators (Flexivent, SCIREQ, Montreal, QC, Canada) and paralyzed with intraperitoneal pancuronium bromide (0.5 mg kg⁻¹), and an ECG was applied to monitor heart rate to ensure proper anesthesia. Breath coming out of the ventilator was collected in Tedlar bags (SKC, Eighty Four, PA) at 180 breaths min⁻¹ with a positive end-expiratory pressure (PEEP) of 3 cm H₂O for 1 h.

Breath volatiles analyses were performed using secondary electrospray ionization mass spectrometry (SESI-MS). The instrumental setup has been previously described [1, 35, 39]. Briefly, the original ionization source of an API 3000 mass spectrometer (SCIEX, Concord, ON, Canada) was replaced with a stainless steel SESI-MS reaction chamber equipped with an electrospray capillary and a gas transfer line through which the breath volatiles are introduced into the reaction chamber (for a detailed schematic of the SESI-MS system, please see reference [35]. Gas flow of 5 L min⁻¹ was driven by a mechanical pump that connected to the sampling gas outlet of the SESI-MS reaction chamber. The breath sample was introduced into the reaction chamber for 30 s at a flow rate of 3 L min⁻¹, supplemented with 2 L min⁻¹ CO₂ (99.99%) at ambient temperature. Formic acid (0.1% (v/v)) was used as the electrospray solution, delivered at a flow rate of 5 nL s⁻¹ through a non-conductive silica capillary (40 µm ID) with a sharpened needle tip. The operation voltage was ∼3.5 kV. Spectra were collected within 30 s as an accumulation of 10 scans in positive-ion mode. The system was flushed with CO₂ between samples until the spectrum returned to background levels.

Analyst 1.4.2 software (Applied Biosystems) was used for spectra collection and raw data processing. Mass spectra shown in each figure are the average spectra of all replicates in each group. Full scan spectra shown in the figures have been blank-subtracted (for in vitro tests, the blank spectrum is the spectrum generated by sterile media; for in vivo tests, the blank spectrum is humidified room air collected using the same procedure as for mice breath) and normalized to the peak of greatest intensity. To evaluate the reproducibility of breathprints in each group, we used the Spearman rank correlation. Then we applied the multivariate analysis method principal component analysis (PCA) to establish statistical evidence for the uniqueness of breathprints from different lung infections. PCA is a statistical tool used to compress complex information and is typically applied when the measurements have a large number of observed variables (e.g., m/z from the mass spectra). Peaks between 20 and 200 m z⁻¹ (mass-to-charge ratio) and greater than 1% relative intensity (after blank subtraction) were used as variables, with their absolute intensities used for the calculations in PCA. All experimental replicates were used as observations. SAS version 9.2 and JMP version 9 (SAS Institute Inc., Cary, NC, USA) were used to generate Spearman rank correlation coefficient, conduct PCA, as well as to determine the statistical significance of observed PCA score differences.

Breath was collected for 1 h from 15 mice infected with either P. aeruginosa PAO1, FRD1, or S. aureus RN450 (five per group), and from five uninfected controls. The breath volatile compounds were fingerprinted using SESI-MS (figure 1), yielding 32 peaks from mice infected with P. aeruginosa PAO1, 61 peaks from mice infected with P. aeruginosa FRD1 and 63 peaks from the S. aureus-infected mice. Qualitative analysis of these spectral breathprints shows that P. aeruginosa infections generate a different volatile fingerprint compared to S. aureus, and both lung infections have different breathprints compared to the uninfected control. In order to check the biological reproducibility of our data, Spearman rank correlation coefficients were calculated and the average values between the biological replicates of each infection group ranged from 0.60 to 0.84 (with standard errors less than 0.09). The first important question to address in diagnosing lung infections is whether or not healthy lungs can be distinguished from infected lungs, regardless of the infectious species. We applied PCA to assess group separation based on the spectra in figure 1. By using only the first principal component, the breathprints of the infected mice are statistically different from the uninfected mice (p <0.005). There are, additionally, important chemical nuances embedded in this analysis. Five measurable peaks (m/z = 54, 65, 70, 72 and 91; table 1) can only be measured in the breath of infected mice. In addition, there are 16 peaks (table 1) that are only observed in the breath of the uninfected controls. In combination, the presence and absence of these 21 peaks may serve as breathprint markers of bacterial infection, but breathprints from additional lung pathogens will be required to validate the universality of these putative markers.

Zoom In Zoom Out Reset image size

Next, we explored the possibility of identifying the infectious species with SESI-MS breathprinting. As seen from figure 1, the breathprint from mice infected with P. aeruginosa is different from those infected with S. aureus, including 18 peaks that are unique to the breath of mice infected with S. aureus, which are not present in the breath of uninfected mice or mice infected with P. aeruginosa (table 1). Both strains of P. aeruginosa are statistically separated from S. aureus using only the first principal component (p < 0.005). We also found that SESI-MS breathprinting provides enough information to differentiate between the two strains of P. aeruginosa used in this study: PAO1 (an acute infection isolate) [40] and FRD1 (a chronic infection isolate) [41]. PAO1 and FRD1 share 19 peaks in their breathprint, but there are 10 unique peaks that distinguish between PAO1 and FRD1 (one peak, m/z = 92, belongs to PAO1 and nine belong to FRD1), as seen in table 1. These 10 peaks are not produced by uninfected or S. aureus-infected mice. The PCA showed that the volatile breathprint from the three bacteria-infected mice groups and the uninfected control group can all be separated with two principal components (p < 0.05). Therefore, for the strains investigated in this study, SESI-MS analysis of the breath of mice can distinguish between infected and uninfected animals, and identify the infectious species to the strain level.

To our knowledge, this is the first report profiling the breath volatiles of mice infected with P. aeruginosa or S. aureus. However, there have been publications characterizing the volatiles produced by several strains of P. aeruginosa and S. aureus, and the SESI-MS peaks we report here are consistent with volatile biomarkers that have been previously reported for these species. A summary of previously published data on protonatable P. aeruginosa and S. aureus volatiles from both in vitro and in vivo studies is provided in table 1.

The majority of published studies that propose to use volatile organic compounds (VOCs) to diagnose infectious disease utilize in vitro data and assume that these data translate to the host-pathogen VOC fingerprint, and comparison of in vivo data to in vitro VOCs is rare [4, 22, 23]. We hypothesize that some VOCs produced in the headspace during in vitro experiments will be present in the breath of an infected host, but that there will also be VOCs that are unique to the host-pathogen interaction. To test this hypothesis, we grew P. aeruginosa PAO1, FRD1 and S. aureus RN450 in tryptic soy broth (TSB) and analyzed the in vitro headspace volatiles by SESI-MS (figure 2) for comparison to the in vivo data collected for mouse lung infections. When comparing the in vivo and in vitro data side by side (figure 1 versus figure 2), it is obvious that the SESI-MS fingerprints are strongly influenced by the growth conditions. For example, the three most abundant peaks from the breath of P. aeruginosa PAO1-infected mice are peaks m/z = 73, 53 and 147, while the dominant peaks of flask-grown PAO1 are m/z = 109, 118 and 82. However, we also observed some shared peaks between these two experimental conditions for each bacterial strain. To quantify the relatedness of the mouse breath and flask-grown volatile fingerprints, the in vivo and in vitro SESI-MS data were pooled together to generate a SESI-MS volatilome for the three bacterial strains in this study, which yielded a total of 52 peaks for PAO1, 78 peaks for FRD1 and 131 peaks for RN450 (table 2). For these three strains, only one quarter to one-third of the total metabolome is shared between the in vitro and in vivo conditions we tested. In addition, less than half of the total SESI-MS volatile metabolome for FRD1 can be captured in vitro.

Zoom In Zoom Out Reset image size

The high degree of variation we observe between in vitro volatile fingerprints and in vivo breathprints could be attributed to a combination of factors. First, bacterial metabolism will change in response to a new environment [42], particularly when infecting a new host [43, 44]. For example, FRD1, a chronic lung infection isolate of P. aeruginosa, has acquired several mutations to become better suited to the host environment. Among these FRD1 adaptations are the loss of catabolic repression control (i.e. looser metabolic regulation) [45], which may account for the large number of in vivo-specific peaks it produces. In addition, hallmark volatiles that are present in vitro may not be measurable in breath. For instance, our data show that S. aureus RN450 loses almost two-thirds of its in vitro SESI-MS fingerprint when compared to the murine lung infection breathprint (table 2). A second set of factors influencing the in vivo breathprint of infection are the volatiles the host produces in response to the pathogen, which are challenging to predict based on in vitro bacterial data alone.