Background
Community-acquired pneumonia (CAP) is a major cause of infection-associated death worldwide, with an incidence of 30–50% in adults [
1]. Due to the complexity and heterogeneity of the disease, diagnosis of CAP, especially for severe CAP, remains a clinical challenge [
2]. Failure to provide timely treatment may result in water, electrolyte and acid–base balance disorders, causing multiple organ dysfunction and even septic shock in critically ill patients [
3,
4]. Therefore, a timely diagnosis, assessment of the severity of CAP and initiation of appropriate treatment can improve patients’ outcomes.
Biomarkers can facilitate early severity assessment of diseases, as well as help predict treatment response and develop new insights into ongoing pathophysiological processes [
2,
5]. In recent years, the focus of the discovery of CAP biomarkers has been increasingly directed towards molecular expression profiles, including gene and protein expression biomarkers in body fluids, for the diagnosis and clinical management of pneumonia [
2,
6‐
10]. Further downstream in the biologic system, however, are small metabolites such as amino acids, carbohydrates or lipids, some of which play important roles in homeostasis and disease states, contributing to processes such as redox balance, oxidative stress, signalling, apoptosis and inflammation; these metabolites can therefore provide a more relevant and amplified signature in CAP [
11]. A few metabolomics studies have offered a powerful approach for biomarker discovery and for elucidating underlying mechanisms of pneumonia [
12‐
14], but none of the studies so far has been focused on changes in the metabolic profiles in CAP patients with different severity, and none had gone beyond the discovery phase.
In the current study, untargeted metabolomics research using liquid chromatography with tandem mass spectrometry (LC-MS/MS) was performed to identify CAP-related metabolic signatures. The identified metabolites were evaluated in the validation cohort by targeted assays. The relationship between the identified metabolites and the clinical severity of CAP as well as the determining performance for severe CAP were then investigated.
Methods
Study design
The study samples were obtained from the multi-centre CAP biological specimen bank of the respiratory and critical care medicine department of Peking University People’s Hospital (PKUPH). All of the obtained CAP samples were originally collected from patients hospitalized in the respiratory medicine department or intensive care unit (ICU) of six hospitals in China from January 2013 to February 2017. The study was approved by the Institutional Review Board of the PKUPH (No. 2011-83) and registered at
ClinicalTrials.gov. Written informed consent was obtained from each participant.
The diagnosis criteria for CAP included the following [
15]: symptom onset began in communities; presenting with clinical manifestations of pneumonia (recent appearance of cough, expectoration or exacerbated symptoms of the previous respiratory diseases, accompanied by or without chest pain, dyspnoea or haemoptysis; fever; signs of pulmonary consolidation and/or moist rales in auscultation; and peripheral white blood cell (WBC) count > 10 × 10
9/L or < 4 × 10
9/L); and a new pulmonary infiltrate on chest radiograph.
Severe CAP was diagnosed by the presence of at least one major criterion or at least three minor criteria outlined by the American Thoracic Society [
16]. The main criteria included: requirement for invasive mechanical ventilation; and occurrence of septic shock with the need for vasopressors. The minor criteria were as follows: respiratory rate ≥ 30 breaths/min; PaO
2/FiO
2 ratio ≤ 250; multi-lobar infiltrates; confusion/disorientation; uraemia (BUN level ≥ 20 mg/dl); leucopenia (WBC < 4 × 10
9/L) as a result of infection; thrombocytopenia (platelet count < 100 × 10
9/L); hypothermia (core temperature < 36 °C); and hypotension requiring aggressive fluid resuscitation.
Patients with evidence of nosocomial infection, active pulmonary tuberculosis, malignancy, severe immunosuppression, non-infectious interstitial lung disease, pulmonary embolism and pregnancy were excluded. During the same period, healthy volunteers or subjects who visited the outpatient department of PKUPH for a routine health examination and were without CAP were also enrolled as a control group. Baseline clinical parameters of CAP patients were obtained from their clinical records and then uploaded to an online case management system of Severe Acute Respiratory Infectious Disease in China by their physicians. The primary endpoint representative of severe CAP was defined as death within 30 days following inclusion. The secondary endpoint was invasive mechanical ventilation or ICU admission within 30 days following study inclusion. Outcomes were assessed at hospital discharge, and by structured telephone interviews at 30 days following inclusion.
Sample collection and preparation
All of the participants were in an overnight fasting state, and 5 ml of peripheral venous blood was taken in the morning within 48 h of hospital admission. The blood was then allowed to clot for 30 min at 4 °C, followed by centrifugation at 1500 × g for 15 min. The serum supernatant was then collected and stored at − 80 °C until further use.
Before testing, the serum samples were thawed at 4 °C on ice. Then 400 μl of pre-chilled methanol (containing l-tryptophan-d5, l-glutamine-13C5, l-glutamine-15N2, terfenadine and propranolol as internal standards) was added to 100 μl aliquots of the serum samples to precipitate the proteins. After vortex-blending for 15 s and incubation at − 80 °C for 1 h, the mixture was centrifuged at 13,400 × g for 20 min at 4 °C. The supernatant was then transferred to a new 1.5-ml Eppendorf tube and dried before storage in the − 80 °C freezer. A pooled quality control (QC) sample solution was prepared by combining equal volumes of serum from each sample and treated using the same procedure already described. Before experimental sample analyses, six QCs were injected to stabilize the instrument. The analysis sequence of the samples to be tested was randomized with a QC sample between every 10 experimental samples. QC samples were used to monitor the reliability of the whole experiment including sample preparation and LC-MS/MS sequence runs.
Untargeted LC-MS/MS analysis and metabolite identification
Details of liquid chromatography–mass spectrometry-based metabolomics analyses are described in Additional file
1. Samples were analysed using an Ultimate 3000 UHPLC (Dionex) system coupled to a Thermo Q-Exactive (Orbitrap) mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA).
Two levels of identification were performed simultaneously using TraceFinder (Thermo Fisher Scientific, San Jose, CA, USA). Metabolites were first potentially identified according to the endogenous MS database by accurate masses. At the same time, the metabolites that matched with the spectra in the fragment database were confirmed at the tandem mass spectrometry level. For precursor and fragment matching, 10 ppm and 15 ppm mass tolerance was applied. Moreover, a 0.25-min retention time shift was allowed for quantification.
Multivariate data analysis
Multivariate statistical analysis for the MS/MS data was performed using SIMCA 14.0 software (Umetrics AB, Umea, Sweden). Unsupervised principal component analysis (PCA) was employed to assess the quality, homogeneity, outlier identification and dominating trends of the group separation inherent in the dataset. A supervised orthogonal partial least squares discriminant analysis (OPLS-DA) was applied to distinguish between the classes and to identify the differentially expressed variables. Corresponding variable importance in the projection (VIP) values and S-plot were generated in the OPLS-DA model. The quality of the multivariate statistical analysis (MVA) models was evaluated by cross-validation analysis of variance (CV-ANOVA),
R2Y and
Q2 values. Student’s
t test was used to determine the significance of each metabolite between two groups, and the relevant false discovery rates (FDR) based on the
p values were estimated. Variables with covariance > 0.1 on the S-plot, VIP > 1 and FDR < 0.05 were considered the highest potential metabolites that could discriminate CAP from controls and for severity assessment. These metabolites were then identified through matching accurate mass and MS/MS spectra using in-house metabolite MS/MS databases. Pathway analysis of selected differential metabolites, a heatmap with a Euclidean distance measure of relative intensity of metabolites (logarithmic scale) and a Pearson’s correlation heatmap were generated using MetaboAnalyst 3.0 (
http://www.metaboanalyst.ca/; Wishart Research Group, University of Alberta, USA) [
17].
Validation and quantification of metabolites
Targeted metabolomics and absolute quantification was performed for differential metabolites identified between CAP patients and controls under the same experimental conditions and procedures in the validation cohort. Concentrations (μg/ml) of the metabolites were calculated by area response ratio (analyte peak area/isotopically labelled internal standard peak area) from independent calibration curves for each metabolite.
Statistical analysis
Values for categorical variables were described as percentages and those for continuous variables were expressed as median (interquartile range (IQR)). Comparisons between groups were performed using the Mann–Whitney U test or Kruskal–Wallis test for continuous variables, and the chi-square test or Fisher’s exact test for categorical variables, as appropriate. Pearson’s correlation coefficients (r) were calculated to assess the strength and direction of linear relationships between the identified metabolites, and clinical parameters included WBC, percentage of neutrophils (NE%), erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), procalcitonin (PCT), confusion, urea level, respiratory rate, blood pressure, and age > 65 years (CURB-65), pneumonia severity index (PSI) and Acute Physiology and Chronic Health Evaluation II (APACHE II) scores. Goodness of the diagnostic method was evaluated using a receiver operating characteristic (ROC) curve based on multivariate logistic regression data. Data were analysed using SPSS Statistics v19.0 (IBM, NY, USA) and MedCalc software version 15.8 (MedCalc Software, Ostend, Belgium). All statistics were two-tailed and the significance level was defined at p < 0.05.
Discussion
The current study describes a novel application of metabolomics in determining the metabolic profile and severity assessment of CAP. Untargeted metabolic analysis could clearly discriminate CAP patients from the age and sex-matched controls, suggesting that CAP causes a significant disruption in biochemical homeostasis. Fifteen metabolites showed key differences between CAP patients and controls in our discovery cohort. Pathway analysis revealed that these dysregulated metabolites were potentially related to the metabolic pathways of sphingolipid, arginine and proline, pyruvate and inositol phosphate. After analysing the relationships between the identified metabolites with clinical parameters, and calculating the diagnostic efficacy of these metabolites using ROC, we concluded that three metabolites—sphinganine, lactate and DHEA-S, related to the severity of CAP—might represent a panel of potential small molecule biomarkers for assessing CAP severity.
Some of the preceding metabolomics studies aiming to identify biomarkers associated with diseases adopted the criteria based on fold change (FC), VIP, FDR or
p value to select the differential metabolites [
18,
19]. The S-plot generated from OPLS-DA analysis visualizes both the covariance and correlation between the metabolites and the modelled class designation, and thereby it can help in identifying statistically significant and potentially biochemically significant metabolites, based both on contributions to the model and their reliability [
20]. Therefore, we strictly limited the criteria to select the differential metabolites in this study, based on the combination of VIP > 1, FDR < 0.05 and covariance > 0.1 on the S-plot, and finally identified 15 differential metabolites between CAP patients and the control group.
Among the identified potential metabolic biomarkers for CAP, the levels of sphinganine in the serum of CAP patients is lower than for controls, and it was observed positively correlated with NE%, ESR and CRP, so it might reflect the presence of an infection and inflammatory response. The serum concentration of sphinganine in severe CAP patients is higher than that in non-severe CAP patients, and lower than controls, but the AUC to distinguish between non-severe CAP and severe CAP was 0.821 and it correlated positively with CURB-65, PSI and APACHE II scores. We therefore considered that it could roughly be related to the severity of CAP. Sphinganine is a major component of sphingolipids, which are one of the active constituents of the mucous secreted by the alveolar epithelium, which protects the lung tissue from invading pathogens [
21].One of the important aspects of sphingolipids and their primary and intermediate metabolites is their interconvertible nature, which enables them to both integrate and regulate a plethora of cellular functions [
22,
23]. Previous studies have shown that
Mycoplasma pneumonia infection in the lungs results in the induction of autoantibody production against glycosphingolipids, suggesting the involvement of sphingolipids in promoting lung inflammation [
24,
25]. Moreover, compelling evidence indicates that certain pulmonary pathogens such as
Chlamydia cause the trafficking of sphingolipids from the trans-Golgi apparatus towards the inclusion membrane to ensure their intracellular survival, which contributes to the immune-evading mechanisms of bacteria [
26]. Therefore, we speculated that sphingolipids might be involved in pulmonary inflammation during infection.
We found that the DHEA-S concentration was lower in the CAP patients than in controls, and its levels were inversely correlated with disease severity. Dehydroepiandrosterone (DHEA) is the most abundant adrenal steroid hormone in humans, and DHEA-S, the sulphated ester of DHEA, is the hydrophilic storage form bound to albumin in the bloodstream [
27]. It has been shown that DHEA modulates the function of the immune system [
28]. A previous study reported that, upon activation by a variety of stimuli such as mitogens or antigens, CD4
+ T cells in healthy adults pre-treated with DHEA produced significantly greater amounts of interleukin (IL)-2 and mediated more potent cytotoxicity than CD4
+ T cells not pre-treated with DHEA [
29]. In another study, patients with tuberculosis showed decreased DHEA levels compared with healthy control subjects, and patients with the lowest DHEA activity showed the highest disease severity [
30]. Moreover, biological activity for DHEA-S was found to be able to enhance the activity of human neutrophils. Therefore, diminished levels of DHEA-S could have adverse effects, especially in relation to susceptibility to bacterial infection [
31]. Based on these findings, we speculated that suppression of DHEA-S levels in patients with severe CAP might further downregulate the immune response to foreign pathogens. This indicates that DHEA or DHEA-S shows potential for use in alternative therapies, as a supplement to currently used long-term antibiotic treatment regimens, or as a preventive strategy against disease recurrence.
During lung infection, glycolysis results in the production of pyruvate, which is converted by lactate dehydrogenase to lactate under anaerobic conditions [
32]. Serum lactate levels have been use for many years in the assessment of tissue hypoxia and perfusion status, and are often used clinically as an indicator of the severity of sepsis and of patient outcomes in sepsis/septic shock [
33‐
35]. A previous study showed that a lactate level in arterial blood gas of > 1.8 mmol/L at admission could predict a need for mechanical ventilation, vasopressors, ICU admission or hospital mortality in patients with CAP [
36]. In our study, serum lactate levels were slightly increased in the patients with non-severe CAP and highly increased in the patients with severe CAP. The increase might reflect the state of anaerobic glycolysis of glucose during severe pulmonary infections. Furthermore, ROC analysis suggested that high serum lactate levels could predict the occurrence of severe CAP, with an optimal cut-off of > 120.021 μg/ml (1.34 mmol/L). Lactate could therefore be considered as a potential metabolic biomarker for the assessment of risk of severe CAP.
It is important to note that the predictability for separation of severe CAP from the controls (Q2 = 0.854) in the OPLS-DA model is not better than the separation for non-severe CAP from controls (Q2 = 0.955), while we expect to observe more alteration in severe CAP than non-severe CAP and better separation from controls. This result possibly was related to the high heterogeneity of the metabolic status of the severe CAP group in our study, which might have an impact on the degree of distinction and the predictability for separation between severe CAP and the control group.
Our independent validation study adds tremendous validity to our initial findings; nevertheless, biomarker discovery is no small undertaking and typically requires years of validation testing before the application phase is reached [
11]. This study is just an initial step in CAP severity assessment using a metabolomics approach. It has certain limitations. First of all, due to the small sample size of non-survivors in our validation cohort, the prognostic predicative value of the identified potential metabolic biomarkers for CAP could not be determined precisely. Next, the proportion of ventilation in patients with severe CAP was significantly higher than that in non-severe CAP in this study. The ventilation condition might affect the metabolomics pictures, so additional studies would be needed to evaluate the serum metabolic status of patients with CAP before mechanical ventilation. Subsequently, the convalescence-phase serum samples from the CAP patients in our study were not obtained for further analyses, and therefore we were unable to identify changes in metabolite concentrations over time, which might be predictive of disease progression, therapeutic response or clinical outcome. Additionally, we could not specifically elucidate the roles of the identified metabolites in CAP pathogenesis. Elaborate studies consisting of a larger external cohort are needed to validate the utility of the identified potential biomarkers, and research aimed at achieving the long sought-after goal—integration of multi-omics data in CAP research—will certainly support the development of precision medicine.
Acknowledgements
The authors wish to thank staff members of the cooperating hospitals (Jichao Chen, MS, Department of Respiratory Medicine, Central Hospital of China Aerospace Corporation, Beijing, China; Guizhen Tian, MS, Department of Respiratory Medicine, 263th Hospital of Chinese PLA, Beijing, China; and Chunliang Yan, MS, Department of Respiratory and Critical Care Medicine, Beijing Aerospace General Hospital, Beijing, China) for assistance with samples and clinical data collection. They are also grateful to the Metabolomics Facility at the Technology Center for Protein Sciences, Tsinghua University for technical support.