Rapid Detection of Sepsis in Rats through Volatile Organic Compounds in Breath

BACKGROUND: Sepsis is one of the main causes of death in adult intensive care units. The major drawbacks of the different methods used for its diagnosis and monitoring are that they are unable to provide fast responses, being unsuitable for bedside use. In this study, performed using a rat sepsis model, we evaluate breath analysis with Ion Mobility Spectrometry (IMS) as a fast, portable and non-invasive strategy. METHODS: This study was carried out on 20 Sprague-Dawley rats. Ten rats were injected with lipopolysaccharide from Escherichia coli and ten rats were IP injected with regular saline. After a 24-h period, the rats were anaesthetized and their exhaled breaths were collected and measured with IMS and SPME-gas chromatography/mass spectrometry (SPME-GC/MS) and data were analyzed with multivariate data processing techniques. RESULTS: The SPME-GC/MS dataset processing showed 92% accuracy in the discrimination between the two groups, with a confidence interval of between 90.9% and 92.9%. Percentages for sensitivity and specificity were 98% (97.5%-98.5%) and 85% (84.6%-87.6%), respectively. The IMS database processing generated an accuracy of 99.8% (99.7%-99.9%), a specificity of 99.6% (99.5%-99.7%) and a sensitivity of 99.9% (99.8%-100%). CONCLUSIONS: IMS involving fast analysis times, minimum sample handling and portable instrumentation can be an alternative for continuous bedside monitoring. IMS spectra require data processing with proper statistical models for the technique to be used as an alternative to other methods. These animal model results suggest that exhaled breath can be used as a point-of-care tool for the diagnosis and monitoring of sepsis.


Introduction
Thanks to the pioneering work of Pauling L et al, it has been known since the 1970s that human breath is a complex mixture of hundreds of compounds (1). Gas Chromatography-Mass Spectrometry has made it possible to identify some of these compounds, revealing that exhaled breath included traces of many volatile organic compounds (VOC), small inorganic molecules and non-volatile substances such as isoprostanes, cytokines or leukotrienes (2,3). Accordingly, there is now a consensus about the diagnostic potential of breath, and there is considerable evidence available to support the use of breath analysis as a diagnostic tool for pulmonary diseases, liver diseases, gastric diseases, diabetes and inflammatory diseases such as sepsis (3,(4)(5)(6)(7).
The analysis of exhaled breath has a number of advantages compared with traditional diagnostic techniques: it is a non-invasive, painless procedure that does not require skilled medical staff (8,9).
Despite this undeniable interest, however, only a few breath tests, such as capnography and the urea breath test, are typically used in clinical routine. One reason for the continued reluctance to use breath analysis as a common diagnostic tool in clinical practice is lack of knowledge about the compounds' metabolic pathways of the compounds, although another factor is the lack of normalization and standardization methods (8,10). Furthermore, bedside systems are not always compatible with the sophistication now required of analytical instruments. Gas chromatography-mass spectrometry (GC/MS), for example, is the most widely used instrument in breath analysis (11)(12)(13) but, although it offers very good sensitivity, precision and resolution, the sampling procedures and subsequent interpretation of data can be demanding and time-consuming and require qualified personnel.
Alternative chemical sensing techniques, such as solid-state sensor arrays (e-noses) and ion mobility spectrometers (IMS), offer new approaches to this problem. E-noses have been applied for many preliminary studies (14) but their limited sensitivity and selectivity hinder their clinical application when the analytes of interest are at sub-ppm levels. Alternatively, IMS is a simple, portable and sensitive instrumental analytical technique that it is gradually expanding its range of applications, from security to food and environmental and clinical applications. IMS provides a spectra response of less than a second to the trace levels of volatile organic compounds, based upon the mobility of gas phase ions in weak electric fields (15). The mobility K of an ion depends on the ion's mass, charge, shape and size, but also on the measurement conditions, particularly pressure and temperature. An ion's mobility is usually reported by means of reduced mobility K0, where the effects of measurement conditions have been harmonized to standard conditions. Moreover, an IMS can offer two different responses: the mobility spectra for positive ions and for negative ions.
Promising results of breath analysis with IMS have been reported and its potential for application as a diagnostic instrument is huge. Lung cancer has been the main focus of attention (16,17) but also interesting findings have also emerged with respect to COPD, sarcoidosis and vaginitis (18)(19)(20). Other diseases such as sepsis have yet to be tested by IMS technology, however, even although the potential capability of breath test for the diagnosis of sepsis has been posited in some works (3,7).

Sepsis is a clinical condition characterized by systemic inflammation, aberrant
immune response, and microcirculation/coagulation disorders generated by the host in combating an infection due to bacterial toxins absorbed from infected wounds and passed into the bloodstream (21). It is a common cause of morbidity and mortality in elderly, immuno-compromised and critically ill patients, and it is the commonest cause of death in adult intensive care units (22). Several rodent models have been used for sepsis studies in various works (23)(24)(25) and, although the correlation between animal and human models is not perfect, rat models can make a positive contribution in many areas, such as the reduction and standardization of experimental variability and the simplification of the experimental setup. One of the common sepsis models used in murine is the induction of sepsis through an exogenous bacteria infection. It is well known that mimicking bacterial infection by using lipopolysaccharide (LPS), which is a structural component of gramnegative bacteria, has been used as a model to activate the immune system, inducing fever, sepsis and multi-organ injury (26). The main cytokines acting as endogenous pyrogens in response to LPS are interleukin (IL)-1β, IL-6 and tumor necrosis factor-α (TNF-α). Moreover, pulmonary dysfunction, including edema, is a well-recognized dysfunction in sepsis (27).
The analysis of exhaled breath in rats has been reported in some studies (28)(29)(30) and the analysis of rodent's breath with an IMS instrument has been reported in a recent feasibility study by Vautz et al (31).
This work explores the viability of the IMS instrumentation and chemometric techniques to generate a VOC discriminatory pattern of sepsis through breath sampling. This work has been performed in a rat model as a first step towards a possible future application in humans. Furthermore, although IMS is undoubtedly capable of providing fast VOC analysis, it also needs to be compared with a gold standard technique in VOC analysis. In order to fill this gap, this study includes GC/MS measurements of the rat's breath as a reference technique, while also providing analyte identification capabilities using proper MS libraries.

Animals
This study was carried out on 20 Sprague-Dawley male rats from Charles River [ Figure 1 about here]

Assessment of rat status
After expiratory gas sampling, the rats were sacrificed by aortic exsanguination. The rats' septic status as a result of the bacterial-LPS injection was investigated by assessing lung edema and systemic inflammation. To assess lung edema, the lungs of all the rats were excised, quickly weighed after removing the main airways, weighed (wet weight=W), dried at 70ºC for 48h and weighed again (dry weight=D). The ratio W/D was computed as a conventional index of lung edema.
The systemic inflammatory status of the rats injected with LPS was determined by measuring the plasma concentrations of two representative inflammatory cytokines: IL1-β and TNF-α. To this end, the peripheral blood was processed to isolate the plasma (centrifugation at 3,000 g using a vasculant rotor for 15 min at 4 ºC). Enzyme-linked immunosorbent assays (ELISA) for IL1-β and TNF-α were performed (Quantikine, R&D Systems, Minneapolis, MN, USA).

Instrumentation and measurements:
The IMS used in this study was the GDA2 (Airsense Analytics, Germany) based on a 100MBeq Ni 63 ionization source that works in both positive and negative modes. An electrostatic gate allows the ions to travel at atmospheric pressure into the drift tube where a constant electric field accelerates them (length 6cm). At the end of the drift tube ions become neutralized in the collector and an electric current is measured. In this manner, the time that the ions need to reach the collector is measured. The collector current is sampled at 33.3kHz. The IMS provides a different sample mobility spectrum 28ms in length every 3s. This spectrum corresponds to an average of 16 consecutive spectra for noise reduction. In this work, the measurements were made at 50% internal dilution of sampling, with a sampling flow of 200ml/min. All the samples were measured twice for up to 40s after 5min of stabilizing the system and IMS. Once these measurements with the IMS were finished, 1L of breath was collected in PFA bags (Jensen Inert Products, USA) for subsequent SPME-GC-MS analysis.
The GC/MS used in this study was FOCUS GC-DSQII (Thermo Scientific, USA).
Breath was collected in 1L PFA bags and analyzed with SPME/Gas Chromatography-Mass Spectrometry. A Carboxen/Polydimethylsiloxane (CAR/PDMS) 75-µm-thick fiber from Supelco was used for the pre-concentration of the analytes. The sorption conditions were 30min at room temperature. The desorption of volatiles from the fiber was undertaken at 250ºC for 5min at the GC/MS injection port. The GC/MS analyses were performed on a Focus GC-DSQ II with a split/splitless injector. A 60m x 0.32mm x 1.8µm capillary column DB-624 (Agilent Technologies) was chosen for the chromatographic separation. Helium was used as carrier gas, with a flow rate of 1ml/min. The MS analyses were carried out in a full scan (scan range 35-350amu) with ionization energy of 70eV. The oven program temperature was as follows: initially, 40ºC held for 5 min, then ramped 10 ºC min-1 to 180ºC; held for 1 min, then ramped 15ºC/min to 230ºC; and then held for 10min.
At the beginning and at the end of each session of measurements, the blanks of the sampling system and the air of the laboratory were measured with IMS in order to ensure the reproducibility of the measurements. In order to counteract the anesthetic drugs in the ion mobility spectra and in the chromatogram, these products were measured in a head-space mode by IMS and by SPME-GC-MS.

Signal Processing and statistical analysis
The signal processing strategies used in this work were based on multivariate signal processing and implemented in MATLAB 7.5 (Mathworks, USA) using the PLS Toolbox 5.8 (Eigenvector Research, USA). The signal processing applied to the IMS dataset was designed to find differences between healthy and diseased rats, and the signal processing applied to the GC/MS dataset focused on identifying compounds that could be potentially useful as sepsis biomarkers The IMS dataset pre-processing includes a base line correction by fitting a 4 th order polynomial to specific spectral intervals devoid of peaks and a cubic Savitzky-Golay filter (32) of length fifteen smoothing procedure. The drift time of the IMS Reactant Ion Peak (RIP) was used as a reference for the spectra alignment and an area normalization procedure was applied to each spectrum. The multivariate signal processing strategy

Pathophysiological rat status
As expected, pulmonary edema was found only in the LPS-treated rat group (W/D=6.88±0.58; mean±SEM) compared to control animals (W/D=5.40±0.28).
Moreover, concentrations of circulating inflammatory markers in plasma were significantly increased in LPS-infected mice compared to controls. Whereas in the control animals the concentration of IL1-β and TNF-α were 1.51±1.01 pg/mL and 1.43±0.14 pg/mL, respectively, in the LPS-injected animals these concentrations rose to 313.45±81.80 pg/mL and 5.99±0.30 pg/mL, respectively. All the differences observed between the controls and the animals with the bacterial endotoxin were statistically significant (t-test or Mann-Whitney rank sum test, as required): p=0.034, p=0.002 and p<0.001 for lung edema, IL1-β and TNF-α, respectively.

Ion Mobility Spectrometry dataset
The IMS dataset featured 10 spectra from 40 breath samples (10 healthy rats + 10 LPS treated rats and an additional replicate of each one). MCR-LASSO was used to decompose IMS raw spectra into their pure contributions: pure spectra components, S, and their related concentration time evolution, C, were extracted. As a result, fourteen relevant pure components were obtained from negative and positive spectra. In Figure 2 [ Figure 4 about here] Although not all the peaks of the samples can be identified, Table 1 lists nineteen compounds found and identified in breath samples from diseased and healthy rats.
Three compounds were identified as related to a fever induced by LPS and one compound was identified as linked to the anesthesia. All of these were discarded for the subsequent data evaluation study. In the end, fifteen compounds were selected as possible compounds associated with sepsis and the area under the peak was calculated for each one using MzMine2 (38).
[ Table 1 about here] The results of the application of PCA-LDA with rank products are shown in Table 1.
Five compounds with a p-value less than 0.001 were chosen by the algorithm as possible compounds related with sepsis. Although other techniques such as ELISA, ProCalcitonin Test (PCT) assays and DNA detection by Polymerase Chain Reaction (PCR) are faster, they need between 2 and 6 h to obtain a response and they are incapable of following the dramatic changes occurring in sepsis (39). In the face of a lack of a real-time monitoring system for sepsis, breath analysis with IMS must be considered a promising and prospective alternative.
The potential capability of breath tests for the diagnosis of sepsis has been indicated in some works (3,7) but, as far as we know, sepsis still remains untested by IMS technology. Other technologies such as GC/MS are also capable of offering a high performance in breath analysis but usually they cannot provide the portability and simplicity of the IMS measurements. IMS is more suited to the clinical trend of developing bedside patient systems but unfortunately it cannot identify easily unknown volatile compounds in a sample, so, in this respect, GC/MS measurements complement this lack of knowledge as a reference technique. This study includes, for the first time, the measurement with IMS technology of rats' breath infused with LPS from E. coli as a sepsis animal model. This represents a first step in the potential applicability of IMS for the diagnosis of sepsis in human patients.
Although is well known that the injection of live bacteria and the injection of only LPS in an animal has some different pathological effects, LPS is commonly used in sepsis models because the injected dose is completely controlled by the experimenter, while this is not the case when injecting live bacteria. Further, although it has been demonstrated that LPS-induced models of endotoxic shock in rodents do not exactly reproduce septic complications in humans (40), they have been used to investigate endotoxin-dependent mechanisms in vivo. Thus, the LPS from the E. coli model used in rats in our study is adequate to examine LPS-dependent aspects of septic shock. GC/MS measurements provided a list of compounds in the rat's breath. After the elimination of the compounds from the SPME-fiber and the anesthesia, fifteen compounds can be potentially used to separate healthy rats from treated rats. To obtain a subset of compounds related to sepsis, PCA-LDA and rank products were used as techniques that allow a maximum discrimination between classes and a ranking of compounds according to their discrimination importance. Moreover, this methodology allows us to obtain a significance level for selected compounds considered as a p-value (37). Thus, the p-value represents the probability of observing a compound at a certain rank, and compounds with the lowest rank are the most important in the separation. In this study we selected compounds with a p-value lower than 0.001. In the end, the first five compounds listed in Table 1 were selected as the most representative compounds in the discrimination between septic and healthy animals, and this could be considered a pattern correlated with sepsis. In this reduced space, a pattern recognition system provides promising rates of bootstrap validation: 85% of accuracy, 91% of specificity and 80% of sensitivity. These percentages must be understood in the light of the bootstrap validation procedure: they mean that, after 500 random selections of different sets of rats, overall 85% of the rats were well classified, and the same interpretation can be made for the specificity and sensitivity figures.
Despite the good figures achieved with GC/MS measurements, the time, cost and infrastructure needed for the sampling and measurement make it impossible to use of these instruments in a bedside setting. The IMS alternative, however, does allow for this possibility because the sampling and measurement time takes only few minutes. With respect to the IMS results, multivariate signal processing was able to detect the spectra of pure breath constituents. After a fine counteraction of external pollutants and anesthesia, and after applying pattern recognition procedures, a pattern of three components was found. Although it is not possible to identify these compounds, they can be separated into two classes, with good levels of accuracy (99.8%), specificity (99.6%) and sensitivity (99.9%) figures under bootstrap validation. It must be stressed that bootstrap validation is designed to avoid over-optimistic results. It is interesting to note that even better results are achieved by processing the full IMS spectra instead of selected molecules. In this respect, we believe that sepsis produces a general alteration In conclusion, breath analysis with IMS has been presented as an alternative for a rapid diagnosis of sepsis. The performance of this methodology in separating a healthy rat group from a diseased rat group is excellent and provides encouraging conceptual evidence at the experimental level. Therefore, the results obtained in the present animal study warrant further clinical studies in septic patients, in order to explore the routine capability of IMS as a non-invasive point-of-care diagnostic tool.