Exhaled breath condensate methods adapted from human studies using longitudinal metabolomics for predicting early health alterations in dolphins


Monitoring health conditions is essential to detect early asymptomatic stages of a disease. To achieve this, blood, urine and breath samples are commonly used as a routine clinical diagnostic. These samples offer the opportunity to detect specific metabolites related to diseases and provide a better understanding of their development. Although blood samples are commonly used routinely to monitor health, the implementation of a relatively noninvasive technique, such as exhaled breath condensate (EBC) analysis, may further benefit the well-being of both humans and other animals. EBC analysis can be used to track possible physical or biochemical alterations caused by common diseases of the bottlenose dolphin (Tursiops truncatus), such as infections or inflammatory-mediated processes. We have used an untargeted metabolomic method with liquid chromatography–mass spectrometry analysis of EBC samples to determine biomarkers related to disease development. In this study, five dolphins under human care were followed up for 1 year. We collected paired blood, physical examination information, and EBC samples. We then statistically correlated this information to predict specific health alterations. Three dolphins provided promising case study information about biomarkers related to cutaneous infections, respiratory infections, dental disease, or hormonal changes (pregnancy). The use of complementary liquid chromatography platforms, with hydrophilic interaction chromatography and reverse-phased columns, allowed us to detect a wide spectrum of EBC biomarker compounds that could be related to these health alterations. Moreover, these two analytical techniques not only provided complementary metabolite information but in both cases they also provided promising diagnostic information for these health conditions.

Collection of the exhaled condensed breath from a bottlenose dolphin from U.S. Navy Marine Mammal Program (MMP)


Early diagnosis of human diseases enhances the possibility to cure and extend the survival and quality of life for individuals. These diagnoses should ideally be performed before the patient shows symptoms of disease, when a diagnosis could make the most difference to guard against a possible upcoming health event. The recent improvements in metabolomic techniques allow the detection of specific metabolites associated with certain human diseases [1, 2]. Similarly to humans, bottlenose dolphins (Tursiops truncatus) can develop health conditions that benefit from early detection and medical care. Cetaceans can develop microbial infections and metabolic and nutrition-related conditions that may be difficult to detect during early stages of disease [3]. Because of the natural anatomy and physiology of dolphins, they can be particularly susceptible to respiratory infections caused by bacterial, fungal, or viral pathogens that may lead to alterations in a complex and unique respiratory microbial environment [4, 5]. These processes can produce specific compounds or biomarkers that define a unique metabolic fingerprint for a certain disease or disorder [6]. These processes define not only the totality of all body changes but also the entire “exposome” of the animal [7]. The exposome is the sum of all metabolic changes during the lifetime of an animal, which includes interactions with the environment, including exposure to chemical contaminants, changes in prey, or environmental triggers of physiologic stress [8,9,10,11,12,13]. This total metabolic snapshot can provide important health information.

Improved detection of health changes in wild, free-ranging cetaceans may help inform policymakers and population managers about how best to protect wild whales, dolphins, and porpoises; however, validation of novel health-monitoring techniques in wild, free-ranging cetaceans presents many challenges. Thus, an ideal approach is to develop and test novel diagnostic techniques in cetaceans that are under human care. These populations have a high level of standard veterinary care, including vigilant preventive medicine programs and application of advanced medicine treatments. Health indices are routinely tracked and recorded to maintain their optimum health, and health conditions are closely monitored and tracked [14]. The U.S. Navy Marine Mammal Program (MMP) has been dedicated to training and caring for bottlenose dolphins for many years [1, 15]. This program allows the determination of baselines for healthy dolphins by using physiologic and clinical data adjusted by species, age, sex, or reproductive status. Routine health monitoring of these animals includes traditional procedures to collect biological samples, including blood, urine, and feces, as well as daily visual observations by trainers, routine physical examinations by veterinarians, and digital imaging (e.g., ultrasound examinations) [14, 16,17,18]. An additional sampling technique that is relatively noninvasive is the collection of the exhaled breath or “blow”. Breath analysis techniques are relatively simple for humans, and with use of sampling devices adapted for dolphins (based on a mouthpiece), a separation valve, and a cooled condenser tube [10] are also well tolerated by dolphins under human care. This technique may serve as a useful diagnostic tool to assess the health condition of an individual, reflecting the physiology of the airway. Moreover, the explosive breathing and the separate respiratory and gastrointestinal systems of dolphins minimize cross-contamination and allows a fast and simple gas exchange. These advantages and the limited genetic variability between dolphins and humans allow a simpler way to discover several exhaled compounds that can be directly used for the design of human breath studies [19].

Exhaled breath condensate (EBC) is a water-based sample enriched in endogenous metabolite compounds. The substances from EBC are both gaseous and liquid, forming aerosol/droplets that contain particles that can be easily condensed through a cooling system [19,20,21]. Although condensed water vapor and volatile molecules (such as nitric oxide, carbon monoxide, and hydrocarbons) are present in EBC samples, nonvolatile compounds have shown the important diagnostic potential of EBC in several studies [9, 19, 20]. Nonvolatile molecules contain carbonyls, small amines, amino acids, peptides, lipids, urea, prostaglandins, carbohydrates, carbonic acids, and pharmaceuticals [22]. This diversity of compounds allows the detection of relevant breath-based metabolites that can be sufficient to discriminate human or other animal health conditions, as previously described in a preliminary study showing differences between fasting and nonfasting dolphins [20]. The primary objective of this study was to identify changes in health of dolphins over time and assess potential changes in breath-based biomarkers that were associated with these health changes. This information could be useful in the future in breath analysis for disease detection, to understand the metabolic pathways involved, and to find a way to prevent or treat these diseases at an early stage. In this work, a longitudinal study was conducted by our measuring breath metabolites from the same group of dolphins for more than 1 year by our collecting EBC samples together with relevant information about their physical status, blood-based health indices, lifestyle, medication, and external conditions. The EBC samples were collected, and nonvolatiles were analyzed by ultrahigh-performance liquid chromatography (LC) coupled with mass spectrometry (MS) to define the animal metabolite profile and detect the most representative features or compounds related to the health status. Clear disease case versus control discriminations were obtained for two of the dolphins studied, providing a list of significant biomarkers for each specific case condition.

Animals, materials, and methods

Animal care and welfare

The MMP houses and cares for a population of dolphins in San Diego Bay (CA, USA). The MMP is Association for Assessment and Accreditation of Laboratory Animal Care accredited, and adheres to the national standards of the US Public Health Service Policy on the Humane Care and Use of Laboratory Animals and the Animal Welfare Act. The MMP’s animal care and use program is routinely reviewed by an institutional animal care and use committee and the U.S. Navy Bureau of Medicine and Surgery. The animal use and care protocol for MMP dolphins in support of this Office of Naval Research funded study (grant number N000141310580) was approved by the MMP’s Institutional Animal Care and Use Committee and the U.S. Navy Bureau of Medicine and Surgery (approval number 106-2013 approved on September 24, 2013, BUMED NRD-964).

Health diagnostics

Five dolphins were monitored for 13 months (May 2014 to June 2015) by marine mammal veterinary technicians and research assistants at the National Marine Mammal Foundation (San Diego, CA, USA). Dolphin samples were collected for a suite of diagnostics (Table 1). Physical examinations, including physical observations and morphologic measurements, were conducted as part of routine health assessment as previously described [16]. Breath assessment was conducted once a week, and blood samples were collected monthly by venipuncture of either the periarterial venous rete in the caudal peduncle or a fluke blade [17]. Blood samples were sent to the Naval Medical Center San Diego for routine clinical chemistry tests and hematology, and to the US Department of Agriculture for cytokine analyses. Table 1 shows the blood values obtained, together with the clinical examination data used to evaluate the health condition of the dolphins.

Table 1 Description of the health diagnostic procedures and conditions for the dolphins studied

EBC sample collection from the dolphins

EBC samples were collected periodically from the dolphins as well as other additional information, such as the number of breaths and sampling time, breathing pattern, estimated lung volume based on animal weight, and atmospheric conditions. In total, 165 samples were collected, ranging from 34 to 48 samples per dolphin, except for dolphin 5, for which there were only five samples.

The EBC collection procedure was described in our previous work [10]. Briefly, in this study, the dolphins were managed animals under human care, and the EBC collection was performed with a sampling device that was placed and held over the dolphin’s blowhole during the sampling time. The EBC sampling was limited to 30 breaths and not time or expired volume controlled; the animals breathed at their comfort rate. Dolphins 2 and 3 had a habit to breathe at a fast rate, completing 30 inhalation-exhalation breath cycles in 90 s, but the other dolphins required three split sessions (20–30 s apart), completing 10 breath cycles per session, to produce a large enough EBC sample to analyze. Most of the dolphins did all the breath maneuvers in less than 5 min, except the oldest dolphin (dolphin 1), which needed a longer interval in between the breathing series. Immediately after collection, all EBC samples were retrieved from the device and stored at -80 °C inside closed glass borosilicate vials as described previously [20].

In addition to the EBC samples, blank samples were collected to consider the presence of possible contaminants that could be introduced during sampling. Seawater blanks were taken from the surrounding area, sealed in glass vials, and stored at -80 °C. The air blanks were collected by our charging a clean EBC sampling device with dry ice pellets and leaving it open to the ambient air for 5 min (air blanks can also be collected by an actively controlled flow rate, as an alternative). The condensed moisture from air was retrieved from the condenser tube with a plunger, sealed in a clean vial, and stored at -80 °C. All the stored samples and blanks were transported to the analytical laboratory at the University of California, Davis with the EBC samples for a further untargeted LC-MS analysis. The workflow of the whole procedure is described in Fig. 1.

Fig. 1

Untargeted metabolomic workflow for the dolphin longitudinal study. Blood samples and additional physical information were collected to determine the dolphins’ health condition. Breath metabolic content was investigated from the collected exhaled breath condensate (EBC) samples by liquid chromatography (LC)–mass spectrometry (MS; a), followed by suitable data analysis procedures (b). HCA hierarchical cluster analysis, PCA principal component analysis, PLS-DA partial least squares discriminant analysis, RP reversed phase, TIC total ion chromatogram, TOF time of flight, VIP variable importance in the projection

Metabolomic content analysis

To obtain a wide coverage of the nonvolatile compounds contained in the EBC samples, we used two complementary analytical platforms based on LC-MS. A reversed-phase (RP) column is commonly applied to retain most of the nonpolar molecules, and a hydrophilic interaction chromatography (HILIC) column is applied to determine more hydrophilic and polar molecules. The analytical procedure for the sample treatment and instrumental data acquisition is shown (Fig. 1, part a). The sample treatment protocol was designed to minimize sample transfer so as to reduce contamination and alteration of the sample’s chemical composition. Approximately 500 μL of the collected EBC sample was directly lyophilized inside the glass vials, and the dried extract obtained was then reconstituted with sonication in 100 μL of mobile phase containing 90% acetonitrile in water (both high-performance LC grade). This volume was split into two capped LC vials with inserts and kept at -20 °C for further LC-MS analysis with both platforms.

Instrumental data were acquired with an Agilent 1290 series ultrahigh-performance LC system coupled with an Agilent 6230 time-of-flight (TOF) mass spectrometer (Agilent Technologies, Santa Clara, CA, USA). The conditions were described in previous work [20]. Vials were kept at 4 °C in an autosampler before the injection of 20 μL of sample. An electrospray ionization (ESI) source with an Agilent Jet Stream nebulizer was used in positive mode, with acquisition in a mass range between 50 and 1700 Th (m/z). To supervise the data quality and system stability, the appropriate quality controls were analyzed with each sample batch containing sample blanks (seawater and air) and method blanks prepared with water, acetonitrile, and the reconstitution solvent. HILIC-aimed samples were analyzed with an Acquity UPLC bridged ethyl hybrid amide column (130 Å, 1.7 μm, 2.1 mm × 100 mm; Waters, Milford, MA, USA) maintained at 30 °C and with use of a mobile phase gradient composed by water (solvent A) and 90% acetonitrile in water (solvent B), both at pH 5 with an ammonium acetate and acetic acid buffer. A HILIC quality control (QC) column (product number 1806006963; Waters, Milford, MA, USA) and a custom-made QC column were selected as HILIC quality controls. For the RP samples, a Poroshell 120 EC-C18 column (2.7 μm, 3.0 mm × 50 mm; Agilent Technologies, Wilmington, DE, USA) held at 30 °C was used. The mobile phases consisted of 60% acetonitrile in water (solvent A) and 10% acetonitrile in 2-propanol (solvent B), both containing 10 mM ammonium formate and formic acid. The RP QC column used was an 6963 RP QC column (Waters, Milford, MA, USA). Subsequently, the remaining samples were pooled and analyzed by MS/MS for final chemical identification. MS/MS data was acquired with a model 6545 quadrupole TOF mass spectrometer (Agilent Technologies, Santa Clara, CA, USA) and the same high-performance LC conditions described previously.

Statistical data analysis

The metabolomic data analysis procedure is shown in Fig. 1, part b, and the LC-MS raw data were first properly processed to apply final statistical tools to correlate the biological parameters obtained and health conditions. Raw-data-containing LC-MS total ion chromatograms were first examined and preprocessed by Agilent Mass Hunter Qualitative Analysis B.05.00SP1. Peaks corresponding to coeluted species were deconvoluted with the “Find by Molecular Feature” algorithm within a mass range from 50 to 1700 Da and considering isotopic and adduct ions obtained by ESI in positive mode. The resulting extracted compound chromatograms of the corresponding ions provide the final abundances for each molecular feature or entity. These results were exported to .cef format, which allowed the feature alignment by Mass Profiler Professional 12.1. Peak alignment between samples was performed by our allowing a retention time shift of 1.5 min and a mass window set at 40 ppm and 25 mDa. The detected time-aligned molecular features, expressed as “mass@time,” were compiled in a peak table containing retention time in minutes, molecular mass, and the intensity values (peak area) for each sample.

The resulting peak tables were imported into MATLAB version 8.3 (The MathWorks, Natick, MA, USA). To reduce the impact of zeros and missing values in the final statistical analysis, these values were replaced by the minimum intensity of each feature divided by 10 [23]. Only peaks above an abundance threshold of 3000 arbitary units were considered as molecular features, and the sample peaks that were present with ten times higher abundance in any of the blanks were also removed from the peak table. The bias obtained during the sample collection and treatment was reduced by normalization per sample by application of the probabilistic quotient normalization with the median value. Spurious peaks (molecular features) were filtered by our considering only molecular features present over 70% in each sample group (disease versus control).

Preliminary data exploration was performed with principal component analysis (PCA) to detect outliers, visualize the data, and identify trends and patterns [24]. In this case, only detected molecular features with statistically significant differences between groups were used. For this, folder exchange, the Student t test, and ANOVA were previously applied with fold change values lower than 2 and p < 0.05 [25].

Finally, a partial least-squares discriminant analysis (PLS-DA) model was built to classify samples with similar characteristics [26]. Cross-validation was applied to select the optimum model dimensionality and avoid overfitting. The key features that explain each health condition for each dolphin were determined with a variable importance in the projection (VIP) threshold of 1.0. The VIP method evaluates the variables that significantly carry the information related to each class. For the class “case,” VIP values will determine if certain molecular features are upregulated or downregulated, with “upregulated” meaning that the signal in the “case” period stage is different from and has higher values (increasing) than that in the control condition periods, and “downregulated” meaning that the “case” signal has lower values (decreasing).

The upregulated or downregulated molecular features were tentatively identified (molecular formula) and annotated (compound name) in two steps. First, compounds were identified by our comparing tandem mass spectra (fragmentation patterns) with the METLIN library data. When MS/MS fragmentation was not available, the exact masses, isotope ratio, and isotopic distribution obtained by single MS (TOF) were compared with the METLIN data with a mass error of less than 10 ppm and a minimum database score of 65%, as well as with data from previous studies [6, 20] and complementary databases (e.g., Human Metabolome Database or LIPID MAPS Structure Database). In these cases, a mass window of 30 ppm and a mass tolerance of 25 mDa were used for protonated adducts, as well as sodium and ammonium species. The final potential biomarkers obtained were manually reviewed by our considering final scores and compound feasibility.

To find correlations between the EBC metabolites and blood parameters, a regression analysis with partial least squares was performed. Only selected highly correlated variables, or biomarkers, were used to fit the linear models. All calculations and statistical analysis were performed with Mass Profiler Professional and PLS_Toolbox version 8.1.1 (Eigenvector Research, Manson, WA, USA) in a MATLAB environment.

Animal health condition monitoring

Any given dolphin’s health condition was defined through the information collected from physical, hematologic, and clinical chemistry evaluations. The values obtained were compared with the age-specific and the sex-specific reference intervals previously established for bottlenose dolphin populations [14, 16,17,18]. When a dolphin became potentially ill or had a change in reproduction status (i.e., reproductively active male in rut with hormonal changes or a pregnant female), it was considered a “case” dolphin. Dolphins were considered to be a case on the basis of the clinical experience of the veterinarians and when hematologic and serum parameters were below or above the reference threshold (see Table S1).

Normal or control status was defined by our considering the baseline ranges of blood values, mainly based on two key indicators of inflammation in dolphins, the white blood cell count and the 60-min erythrocyte sedimentation rate. Longitudinal variations of the main blood parameters are represented in Fig. 2.

Fig. 2

Longitudinal variation by each dolphin through time (x-axis) for the two main blood parameters studied: white blood cell count (WBC; a) and 60-min erythrocyte sedimentation rate (ESR; b). The different health conditions are represented with different colors.

These different health case conditions are summarized in Table 2, where time tracking of the study for the dolphins is provided. The sex of the dolphins was controlled, and the ages ranged from 10 to 44 years. Of the five dolphins, three had case conditions that lacked control periods. Dolphin 1 (male, 44 years old) remained in rut throughout the study, with a hormonal alteration causing reproductive-related behaviors, and dolphin 5 (female, 26 years old) was monitored opportunistically because of the presence of an abnormal white blood cell count and erythrocyte sedimentation rate with no specific disease detected. Whereas dolphins 1 and 5 had a single condition, dolphin 3 (male, 36 years old) went through three different diseases, all of them based on infection or inflammation. First, dolphin 3 had an acute respiratory infection that was relieved by antibiotics (case 3a), then pneumonia was diagnosed (case 3b), and finally it had a dental inflammatory condition (case 3c). Lack of control reference data for health conditions in these three dolphins greatly limited our ability to identify which biomarkers predicted a health condition. However, differently from dolphins 1 and 5 with a single abnormal condition without a control period, dolphin 3 was studied to find differences between the three case conditions. Opportunely, samples corresponding to a normal health condition could be obtained from the remaining two dolphins: a perigenital swelling that lasted 3 months associated with an active reproductive stage was diagnosed in dolphin 2, and dolphin 4 (female, 13 years old) became pregnant during the course of the study.

LC-MS metabolites

To obtain a wider metabolite coverage, the LC-TOF-MS method was used on two analytical platforms. RP chromatography and HILIC are complementary separation techniques aimed at detecting different compounds on the basis primarily of polarity. RP chromatography with C18 columns provides good separation of nonpolar and weakly polar compounds, but to retain highly hydrophilic, ionic, and polar molecules, a HILIC column is required [27]. In this study, similar sample discriminations based on heath conditions were obtained for both techniques, but these differences were defined by different molecular features depending on the separation technique used.

Data pretreatment was optimized for both platforms, by application of a molecular feature extraction algorithm to the acquired datasets containing 157 and 165 samples for RP chromatography and HILIC respectively. The number of EBC samples collected during each time period is given in Table 2. The molecular feature extraction algorithm combines ion species generated by a single compound detected with ESI in positive mode. More than 30,000 molecular features were extracted with each technique. Of these, we found 13,852 (RP chromatography) and 12,563 (HILIC) were aligned across the samples from the study. The dimensionality of the datasets was first reduced by our filtering peaks found in blanks to 6222 (RP chromatography) and 6803 (HILIC) molecular features.

Table 2 Dolphins’ health condition through the longitudinal study (time in months). Numbers inside the colored bars (in parenthesis) correspond to the number of EBC samples from that condition. Animal code, gender, age and total number of EBC samples are also described

Results and discussion

Differences between dolphins

After completion of the PCA and application of the 70% filter described in “Statistical data analysis,” datasets were reduced to 251 and 269 peaks for RP chromatography and HILIC respectively, and no separation by dolphin or any other pattern was observed (data not shown). The next step was to filter the variables on the basis of their p values obtained from an ANOVA and fold change analysis. This reduced the number of molecular features from 251 to 13 for RP chromatography and from 269 to 38 for HILIC. All the data were transformed logarithmically and autoscaled before application of each model. No samples were removed as outliers by our considering Hotelling’s T 2 values and Q residuals above the 95% confidence limits. PCA score plots of the two first principal components (Fig. 3) showed a clear separation of dolphin 1 (red) from the rest of the dolphins with RP chromatography (Fig. 3, plot a) and HILIC (Fig. 3, plot b). However, slight differences can also be observed for dolphin 3 (blue) in RP chromatography and dolphin 5 (pink) in HILIC. The reason these dolphins are differentiated could be due to the absence of control or normal samples, emphasizing the singularity of their health status compared with the other dolphins. Dolphin 1 had a reproductive alteration (dolphin in rut) and is also the oldest dolphin (male, 44 years), and differs more from the other dolphins, which mainly had infection or inflammatory diseases (except for dolphin 4, pregnancy). No differentiation between dolphin 2 (green) and dolphin 4 (light blue) was observed, as well as any other data patterns by health. To detect the case. versus control differences, we worked by individual dolphins. In the case of dolphins 1 and 5, no control samples were obtained, so these dolphins were not considered in the next part of our study analysis.

Fig. 3

Principal component (PC) analysis scores plot of liquid chromatography–mass spectrometry (LC-MS) analysis of exhaled breath condensate by reversed-phase (RP) chromatography (a) and hydrophilic interaction chromatography (HILIC; b). Colored according to the dolphin

Differences within dolphins: health status

Differences within dolphin 2

Dolphin 2 was a male and the youngest animal in the study (10 years old). It started the study as a healthy individual but subsequently developed a perigenital swelling, which lasted 3 months. However, fewer EBC samples were collected during the case status (n = 5) compared with the samples collected as a control status (n = 31), and this unbalanced number of samples per class could lead to nonrepresentative models with skewed results. For this reason, the control group was reduced to the ten samples closer to the transition time that dolphin 2 converted to a case animal status, resulting in a final dataset of 15 total samples. These data were first filtered by our selecting the top 70% of the variables or molecular features present in each group or class, and obtaining the 214 (RP chromatography) and 216 (HILIC) molecular features used to build the PLS-DA models. This technique maximized the separation between dolphin 2 health conditions, allowing the identification of putative marker metabolites with VIP plots.

The PLS-DA results obtained by leave-one-out cross-validation for dolphin 2 are presented in Fig. 4, panel a, where score plots corresponding to RP chromatography and HILIC are shown. Clear discrimination between case and control samples was shown with both techniques, specially with the HILIC column (Fig. 4, panel a, bottom). This separation allowed the identification of the molecular features considered significant to discriminate control from case samples defined by the animal inflammation by our selecting VIP scores with the highest values greater than 1. The 34 (RP chromatography) and 27 (HILIC) possible molecular features that were selected are summarized in Table S2. Each feature was correlatively described by its accurate mass and retention time (in minutes) and its signal regulation (upregulation/downregulation). These regulations are exemplified by time versus abundance plots and box-and-whisker plots for the most important metabolites (highest VIPs) in Fig. 4, panel b. For instance, the abundances for molecular features 903.6857@8.00 (RP chromatography) and 601.3670@2.38 (HILIC) were reduced when the dolphin had inflammation (Fig. 4, panel b, left), in contrast to molecular features 858.5982@9.66 (RP chromatography) and 142.1110@4.27 (HILIC), which enhanced their abundance in the case condition. The formulas or elemental composition, the score values, and tentative annotation are also listed (Table S2). This is the case of low-polar compounds such as palmitoleoyl ethanolamide defined as an anti-inflammatory lipid, and phlegmarine, cerebroside D, avobenzone, maharimbine, isotachysterol 3, acylcarnitines, or retinol phosphate, all of which are possible biomarkers of dolphin 2 inflammation by the RP technique. More polar metabolites such as amino acids and peptides (Lys-Lys-Ser, Thr-Phe-Lys or leupeptin) and compounds such as C17 sphinganine and hydroxysphingosines were detected by HILIC. Although these results look promising, several potential biomarkers related to the animal disease are still unidentified, and further investigations involving comparison with reference standards should be performed.

Fig. 4

Partial least squares discriminant analysis (PLS-DA) results for liquid chromatography–mass spectrometry (LC-MS) analysis of dolphin 2 exhaled breath condensate samples colored per disease (green for control, and red for case). PLS-DA scores plot corresponding to the first two latent variables (a). Examples of the main molecular features or biomarkers with downregulation (left) and upregulation (right) (b). Each marker is represented by the abundance (y-axis) versus longitudinal time (x-axis) in months. Box-and-whisker plots by control (green) and by case (red) showing the first and third quartiles (box) divided by the mean, and the error bars to the minimum and maximum (whiskers). CNTL control, HILIC hydrophilic interaction chromatography, LV latent variable, RP reversed phase, RT retention time

The next step was to correlate the blood parameters with the selected EBC biomarkers. First, an initial case-control PLS-DA involving all the blood variables was performed. This allowed the discrimination of 12 blood variables that were highly influenced by dolphin 2 health status. From these, eight parameters presented correlations with HILIC metabolites and two parameters with RP metabolites. The correlation results are presented in Table 3, listing the main blood parameters and describing positive or negative correlation depending on the case status. For example, positive correlation variables had higher values when the dolphin was in a case condition. The regression coefficient (R 2) and root-mean-square errors of cross-validation assessed the quality of the correlation.

Table 3 Exhaled breath condensate (EBC) metabolite features that correlate hematologic and serum parameters in dolphins (biomarker annotations listed in Tables S2, S3 and S4). Positive and negative correlation is described by the case condition of each dolphin. Regression coefficients (R 2) and root-mean-square errors (RMSE) are reported by cross-validation (CV)

In the case of dolphin 2, RP biomarkers had a high correlation with cholesterol values, as well as HDL, red blood cell (RBC), and creatine phosphokinase (CPK) values. Lower levels of cholesterol and HDL, and higher RBC and CPK values, are related to biomarkers presents when the dolphin had inflammation (case). Other parameters that were correlated to the EBC metabolites for this dolphin were platelet count, alanine aminotransferase and aspartate aminotransferase values, packed cell volume, and mean corpuscular hemoglobin value.

Differences within dolphin 4

Dolphin 4 was a 26-year-old female that became pregnant during the course of the study. Pregnancy involves different hormonal and physical alterations that can be studied through a longitudinal monitoring of the animal. Within this population, pregnancy is diagnosed at approximately the second trimester, so dolphin 4 was tracked for 5 months as a control dolphin, changing to case (or pregnant) dolphin toward the end of the study. In this case, we again needed to balance the number of EBC samples that were obtained and analyzed by RP chromatography and HILIC between control (n = 18) and case (n = 22) status. As with the dolphin 2, data were first filtered by the “70% rule,” reducing the number of molecular features to 348 (RP chromatography) and 289 (HILIC). These datasets were used to build cross-validated PLS-DA models by random subsets (6 splits and 20 iterations). The results for dolphin 4 are presented in Fig. 5a.

Fig. 5

Partial least square discriminant analysis (PLS-DA) results for liquid chromatography–mass spectrometry (LC-MS) analysis of dolphin 4 exhaled breath condensate samples colored per disease (green for control, and red for case). PLS-DA scores plot corresponding to the first two latent variables (a). Examples of the main molecular features or biomarkers with downregulation (left) and upregulation (right) (b). Each marker is represented by the abundance (y-axis) versus longitudinal time (x-axis) in months. Box-and-whisker plots by control (green) and by case (red) showing the first and third quartiles (box) divided by the mean, and the error bars to the minimum and maximum (whiskers). CNTL control, HILIC hydrophilic interaction chromatography, LV latent variable, RP reversed phase, RT retention time

A clear discrimination between pregnant (case) and control samples was also obtained with both techniques, but in this case, the RP column achieved better separation (Figure 5, panel a, top). Table S3 summarizes the selected molecular features by VIP scores: 37 (RP chromatography) and 33 (HILIC) possible metabolites were related to the pregnancy. From those, compounds such as 559.5437@8.65 (RP chromatography) and 364.2703@2.56 (HILIC) were downregulated (reducing abundances with case condition) and compounds such as 425.3658@11.29 (RP chromatography) and 362.2470@1.13.27 (HILIC) were upregulated (enhancing abundances), as presented in Fig. 5, panel b. The most probable elemental composition is described in Table S3), along with possible annotations. However, biomarker elucidation was more complicated to define this dolphin’s case condition with few metabolite coincidences. Some related compounds are alkenes such as octane/methylheptene, isophenoxazine, or some ceramides detected by RP chromatography, and sphingosines or pinacidil detected by HILIC.

These EBC metabolites were also correlated to the blood parameters that were highly influenced by the pregnancy status (Table 3). In this case, lower correlations were obtained, with insulin total values being the ones highly related only to the RP biomarkers. This suggests that pregnant dolphins have lower levels of total insulin and these can be correlated to certain EBC metabolites detected in RP mode. Other parameters, such as packed cell volume, hematocrit, RBC count, and hemoglobin level were similarly correlated to RP and HILIC metabolites.

Differences within dolphin 3

Dolphin 3 was an adult, 36-year-old male. This dolphin had medical conditions throughout the study: an initial infection that was not pneumonia (case 3a, n = 12), followed by pneumonia (case 3b, n = 19), and ultimately an inflammatory dental disease, or osteomyelitis (case 3c, n = 17). However, considering that no control samples were available, this dolphin was also studied to find possible biomarkers that distinguish the different case status. The results obtained in this part of the study may be less reliable since we cannot compare any healthy reference values, but interesting information can be extracted from the possible related metabolites. Considering the three groups or classes, the 70% rule per group was also applied, and the datasets were reduced to 244 (RP chromatography) and 114 (HILIC) molecular features. Random subset (6 splits and 20 iterations) cross-validation was applied to build the PLS-DA model of the three classes (Fig. 6).

Fig. 6

Partial least square discriminant analysis scores plot corresponding to the first two latent variables for liquid chromatography–mass spectrometry (LC-MS) analysis of dolphin 3 exhaled breath condensate samples colored per disease (pink for case 3a, light blue for case 3b, and dark blue for case 3c). HILIC hydrophilic interaction chromatography, LV latent variable, RP reversed phase

Case 3a samples (Fig. 6, pink circles) showed a clear discrimination from the other conditions explained by the first component (latent variable 1), specially with the RP technique. Cases 3b (Fig. 6, light blue circles) and 3c (Fig. 6, dark blue circles) were weakly differentiated by the second principal component. This may indicate that the first infection of the dolphin had some specific molecular features that were distinguishable from those of the two secondary diseases. These possible compounds are summarized in Table S4, containing 16 (RP chromatography) and 37 (per HILIC) possible metabolites. This table also describes which case explains the PLS-DA discrimination per molecular feature, as well as if it was upregulated or downregulated (data not shown). For instance, molecular features 199.1571@5.91 and 420.2334@1.2 were upregulated and downregulated respectively for the initial infection (case 3a). In the same manner, 460.2650@2.14 and 415.2605@2.47 were molecular features for pneumonia (case 3b), and 243@2560@1.8 and 589.3678@2.47 were molecular features for osteomyelitis. However, it has to be taken into account that the observed changes in EBC metabolomics content justify this dolphin’s physiologic conditions, and further studies should be done considering control samples to describe each specific disease. The elemental composition and possible compound annotation is given in Table S4, with tentative identification of isophenoxizane, phenolic steroids, ceramides and glycerophospholipids by RP chromatography, and sphingosines and pinacidil by HILIC. Sphingolipid derivatives and different amino acids and peptides were identified in all the dolphins studied. The presence of these compound seems to be related to biochemical alterations of the dolphins, and may hold future diagnostic potential.

Although the dolphin had three different health conditions during the study, the selected biomarkers were also correlated to the blood analysis results, achieving low correlations (Table 3). However, high levels of hematologic parameters such as RBC count, packed cell volume, hemoglobin level, and eosinophil level are slightly related to the presence of the first infection (case 3a). In contrast, high mean corpuscular volumes are correlated to the final condition of dental disease (case 3c).


The ability to predict and diagnose diseases with exhaled breath metabolites is still at an early stage of development. Metabolomic studies with exhaled breath samples often aim to detect biomarkers unique to certain health diseases, correlating detected signals or metabolites with associated physical or biochemical alterations. In the future, breath-based biomarker detection may become an essential tool for predicting health changes at early stages. There are important aspects that should be considered when metabolomic data are studied with single case-study comparisons.

In this study, different dolphins were monitored weekly for more than 1 year to find putative biomarkers that explain specific diseases or health changes. In addition to use of blood-based biochemical analyses and observations of physical conditions, EBC samples were collected and analyzed by RP and HILIC LC-MS. Although these were managed dolphins under human care, it remains difficult to define an experimental design that can predate the onset of a specific health change. Thus, five dolphins were selected for the longitudinal study. From these, two were distinguished from the others by LC-MS data. Individual studies of the remaining three dolphins, however, showed a clear discrimination within each dolphin when case conditions were present. A list of specific biomarkers has been suggested (Tables S2, S3, and S4), and a set of metabolites is highlighted that present a significant separation capability between dolphins with a specific disease. Different blood parameters were also studied to assess the identification of the dolphin conditions:

  • Dolphin 2 biomarkers that explain the disease causing inflammation were highly correlated with cholesterol, HDL, RBC, and CPK values.

  • Dolphin 4 biomarkers that explain the animal pregnancy condition were mainly correlated with total insulin value.

  • Dolphin 3 biomarkers intended to explain three different health conditions, RBC count, packed cell volume, hemoglobin level, and eosinophil level, were slightly correlated to the first infection (respiratory), and mean corpuscular volume was correlated to the final-stage disease (dental inflammation).

Bottlenose dolphin RBCs have a relatively short half-life (16.5 days), and anemia is common in dolphins with inflammation and infections [28]. The association between inflammation and anemia in dolphins is likely similar to that between anemia and inflammation in humans; namely, raised levels of circulating cytokines and redistribution of iron from blood to the liver during systemic inflammation results in increased fragility of RBCs and decreased erythropoiesis [29]. This type of anemia resolves with resolution of inflammation.

Although these results appear promising, extension of a future longitudinal study could include more representative samples for the different conditions and would provide a good complement to test the robustness of our current results. Moreover, several molecular features related to animal disease are still unidentified, and further investigations involving MS/MS and comparison with reference standards could confirm those metabolite identities.

The noninvasive nature of exhaled breath analysis brings a unique angle to this study. Our noninvasive techniques hold the potential to be diagnostically important for veterinary and human medicine [30], and may provide a large advance over invasive measures for marine mammals. EBC comprises both volatile and nonvolatile compounds, providing a wide representation of metabolites that can correlate with human and animal health conditions. For the first time, we have shown how traditional health measures compare with these exhaled breath biomarkers across time in multiple dolphins with a variety of health conditions.


  1. 1.

    Nagana Gowda GA, Zhang S, Gu H, Asiago V, Shanaiah N, Raftery D. Metabolomics-based methods for early disease diagnostics: a review. Expert Rev Mol Diagn. 2008;8(5):617–33. https://doi.org/10.1586/14737159.8.5.617.

  2. 2.

    Zhang A, Sun H, Yan G, Wang P, Wang X. Metabolomics for biomarker discovery: moving to the clinic. Biomed Res Int. 2015;354671. https://doi.org/10.1155/2015/354671.

  3. 3.

    Miller DL, Ewing RY, Bossart GD. Emerging and resurging diseases. In: Dierauf L, Gulland FMD, editors. CRC handbook of marine mammal medicine. Boca Raton: CRC Press; 2001. p. 15–30.

    Chapter  Google Scholar 

  4. 4.

    Johnson WR, Torralba M, Fair PA, Bossart GD, Nelson KE, Morris PJ. Novel diversity of bacterial communities associated with bottlenose dolphin upper respiratory tracts. Environ Microbiol Rep. 2009;1(6):555–62. https://doi.org/10.1111/j.1758-2229.2009.00080.x.

  5. 5.

    Bik EM, Costello EK, Switzer AD, Callahan BJ, Holmes SP, Wells RS, et al. Marine mammals harbor unique microbiotas shaped by and yet distinct from the sea. Nat Commun. 2016;7:10516. https://doi.org/10.1038/ncomms10516.

    CAS  Article  Google Scholar 

  6. 6.

    Pasamontes A, Aksenov AA, Schivo M, Rowles T, Smith CR, Schwacke LH, et al. Noninvasive respiratory metabolite analysis associated with clinical disease in cetaceans: a Deepwater Horizon oil spill study. Environ Sci Technol. 2017;51(10):5737–46. https://doi.org/10.1021/acs.est.6b06482.

    CAS  Article  Google Scholar 

  7. 7.

    Wild CP. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomark Prev. 2005;14(8):1847–50. https://doi.org/10.1158/1055-9965.epi-05-0456.

    CAS  Article  Google Scholar 

  8. 8.

    King D, Aldridge B, Kennedy-Stoskopf S, Stott JL. Immunology. In: Dierauf L, Gulland FMD, editors. CRC handbook of marine mammal medicine. Boca Raton: CRC Press; 2001. p. 237–52.

    Chapter  Google Scholar 

  9. 9.

    Lima N, Rogers T, Acevedo-Whitehouse K, Brown MV. Temporal stability and species specificity in bacteria associated with the bottlenose dolphins respiratory system. Environ Microbiol Rep. 2012;4(1):89–96. https://doi.org/10.1111/j.1758-2229.2011.00306.x.

    Article  Google Scholar 

  10. 10.

    Zamuruyev K, Aksenov A, Baird M, Pasamontes A, Parry C, Foutouhi S, et al. Enhanced non-invasive respiratory sampling from bottlenose dolphins for breath metabolomics measurements. J Breath Res. 2016;10(4):046005.

    Article  Google Scholar 

  11. 11.

    Cumeras R, Cheung W, Gulland F, Goley D, Davis C. Chemical analysis of whale breath volatiles: a case study for non-invasive field health diagnostics of marine mammals. Metabolites. 2014;4(3):790.

    Article  Google Scholar 

  12. 12.

    Hunt KE, Moore MJ, Rolland RM, Kellar NM, Hall AJ, Kershaw J, et al. Overcoming the challenges of studying conservation physiology in large whales: a review of available methods. Conserv Physiol. 2013;1(1):cot006-cot. https://doi.org/10.1093/conphys/cot006.

    Article  Google Scholar 

  13. 13.

    Richard JT, Robeck TR, Osborn SD, Naples L, McDermott A, LaForge R, et al. Testosterone and progesterone concentrations in blow samples are biologically relevant in belugas (Delphinapterus leucas). Gen Comp Endocrinol. 2017;246:183–93. https://doi.org/10.1016/j.ygcen.2016.12.006.

    CAS  Article  Google Scholar 

  14. 14.

    Wilson B, Arnold H, Bearzi G, Fortuna CM, Gaspar R, Ingram S, et al. Epidermal diseases in bottlenose dolphins: impacts of natural and anthropogenic factors. Proc R Soc B Biol Sci. 1999;266(1423):1077–83.

    CAS  Article  Google Scholar 

  15. 15.

    Venn-Watson SK, Jensen ED, Ridgway SH. Evaluation of population health among bottlenose dolphins (Tursiops truncatus) at the United States Navy Marine Mammal Program. J Am Vet Med Assoc. 2011;238(3):356–60. https://doi.org/10.2460/javma.238.3.356.

    Article  Google Scholar 

  16. 16.

    Schwacke LH, Smith CR, Townsend FI, Wells RS, Hart LB, Balmer BC, et al. Health of common bottlenose dolphins (Tursiops truncatus) in Barataria Bay, Louisiana, following the Deepwater Horizon oil spill. Environ Sci Technol. 2014;48(1):93–103. https://doi.org/10.1021/es403610f.

    CAS  Article  Google Scholar 

  17. 17.

    Mazzaro LM, Johnson SP, Fair PA, Bossart G, Carlin KP, Jensen ED, et al. Iron indices in bottlenose dolphins (Tursiops truncatus). Comp Med. 2012;62(6):508–15.

    CAS  Google Scholar 

  18. 18.

    Goldstein JD, Schaefer AM, McCulloch SD, Fair PA, Bossart GD, Reif JS. Clinicopathologic findings from Atlantic bottlenose dolphins (Tursiops Truncatus) with cytologic evidence of gastric inflammation. J Zoo Wildl Med. 2012;43(4):730–8. https://doi.org/10.1638/2011-0054R.1.

    Article  Google Scholar 

  19. 19.

    Schivo M, Aksenov A, Yeates L, Pasamontes A, Davis C. Diabetes and the metabolic syndrome: possibilities of a new breath test in a dolphin model. Front Endocrinol. 2013;4:163. https://doi.org/10.3389/fendo.2013.00163.

    Article  Google Scholar 

  20. 20.

    Aksenov AA, Yeates L, Pasamontes A, Siebe C, Zrodnikov Y, Simmons J, et al. Metabolite content profiling of bottlenose dolphin exhaled breath. Anal Chem. 2014;86(21):10616–24. https://doi.org/10.1021/ac5024217.

    CAS  Article  Google Scholar 

  21. 21.

    Raverty SA, Rhodes LD, Zabek E, Eshghi A, Cameron CE, Hanson MB, et al. Respiratory microbiome of endangered southern resident killer whales and microbiota of surrounding sea surface microlayer in the eastern North Pacific. Sci Rep. 2017;7(1):394. https://doi.org/10.1038/s41598-017-00457-5.

    Article  Google Scholar 

  22. 22.

    Kubáň P, Foret F. Exhaled breath condensate: Determination of non-volatile compounds and their potential for clinical diagnosis and monitoring. A review. Anal Chim Acta. 2013;805:1–18. https://doi.org/10.1016/j.aca.2013.07.049.

    Article  Google Scholar 

  23. 23.

    Wehrens R, Hageman JA, van Eeuwijk F, Kooke R, Flood PJ, Wijnker E, et al. Improved batch correction in untargeted MS-based metabolomics. Metabolomics. 2016;12:88. https://doi.org/10.1007/s11306-016-1015-8.

    Article  Google Scholar 

  24. 24.

    Jackson JE. A user's guide to principal components. Hoboken: Wiley; 2004.

    Google Scholar 

  25. 25.

    Bartel J, Krumsiek J, Theis FJ. Statistical methods for the analysis of high-throughput metabolomics data. Comput Struct Biotechnol J. 2013;4(5):e201301009. https://doi.org/10.5936/csbj.201301009.

    Article  Google Scholar 

  26. 26.

    Barker M, Rayens W. Partial least squares for discrimination. J Chemom. 2003;17(3):166–73. https://doi.org/10.1002/cem.785.

    CAS  Article  Google Scholar 

  27. 27.

    Gika HG, Theodoridis GA, Plumb RS, Wilson ID. Current practice of liquid chromatography–mass spectrometry in metabolomics and metabonomics. J Pharm Biomed Anal. 2014;87:12–25. https://doi.org/10.1016/j.jpba.2013.06.032.

    CAS  Article  Google Scholar 

  28. 28.

    Sweeney JC, Ridgway SH. Common diseases of small cetaceans. J Am Vet Med Assoc. 1975;167(7):533–40.

    CAS  Google Scholar 

  29. 29.

    Nemeth E, Ganz T. Anemia of inflammation. Hematol Oncol Clin North Am. 2014;28(4):671–681, vi. https://doi.org/10.1016/j.hoc.2014.04.005.

    Article  Google Scholar 

  30. 30.

    Horváth I, Barnes PJ, Loukides S, Sterk PJ, Högman M, Olin A-C et al. A European Respiratory Society technical standard: exhaled biomarkers in lung disease. Eur Respir J. 2017;49(4). https://doi.org/10.1183/13993003.00965-2016.

Download references


Support for these investigations was provided by the Office of Naval Research grant N-00014-13-1-0580 (CED, SV-W, BCW), National Institutes of Health (NIH) grants 1U01EB022003-01, UL1RR024146-06, 1P30ES023513-01A1, and UG3-OD023365 (CED) and T32-HL007013 and T32-ES007059 (MS), and University of California, Davis School of Medicine and NIH 8KL2TR000134-07 K12 mentored training award and NIH grant 1K23HL127185-01A1 (MS). Student support was provided by NIH award T32 HL07013 (KOZ) and NIH award P42ES004699 (KOZ).

Author information



Corresponding author

Correspondence to Cristina E. Davis.

Ethics declarations

This work was performed under an animal care and use protocol reviewed and approved by the US Navy Marine Mammal Program Institutional Animal Care and Use Committee and the U.S. Navy Bureau of Medicine and Surgery. The US Navy has elected to file patent application 14/859,612 on the dolphin breath sampler used in this work. The patent rights are assigned to the government, and the authors declare they have no financial conflicts of interest.

Electronic supplementary material


(PDF 65 kb)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Borras, E., Aksenov, A.A., Baird, M. et al. Exhaled breath condensate methods adapted from human studies using longitudinal metabolomics for predicting early health alterations in dolphins. Anal Bioanal Chem 409, 6523–6536 (2017). https://doi.org/10.1007/s00216-017-0581-6

Download citation


  • Exhaled breath condensate
  • Breath analysis
  • Metabolomics
  • Longitudinal study
  • Liquid chromatography–mass spectrometry