Introduction
Community-acquired pneumonia (CAP) is of global importance to medicine with less than 5% mortality in outpatients and 5–10% and 25–30% mortality in hospitalized patients and intensive care unit (ICU) patients, respectively. Mortality can reach 50% among those who are in septic shock and require vasopressors [
1,
2]. CAP was reported as the seventh most common cause of mortality in the USA with 50,000 deaths in 1.5 million hospitalized patients each year [
3‐
5]. The morbidity of CAP is between 0.3 and 0.5% worldwide in adults [
6]. Management of CAP patients is often complicated due to poor quality evidence of clinical data such as radiographic findings, difficulty with accurate diagnosis, poor prognostic signs, and non-specific therapeutic strategies that make the prediction of patient outcomes uncertain [
7]. There is a global interest in predicting short-term (< 90 days) and long-term (1 year) CAP mortality [
2]. Current severity and mortality assessment methods such as scoring systems and biomarkers are not sensitive or specific enough to predict mortality accurately. APACHE II/III, SAPS, SOFA, and PSI are commonly used scoring systems for the prognosis and severity assessment of CAP that can enhance the prediction of prognosis in association with other current biomarkers such as procalcitonin (PCT) and C-reactive protein (CRP) [
1]. Several scoring systems, such as PSI, APACHE II, CURB-65, and SAPS [
8], have been used to categorize severity and predict short-term and long-term mortality of CAP in association with two putative biomarkers: PCT and CRP [
9]. Metabolomics, both non-targeted and targeted approaches, provide a powerful tool to identify and quantify low molecular weight compounds (metabolites) in biofluid samples that contribute to normal and pathological pathways as primary, intermediate, and/or end products of metabolism [
10,
11]. Metabolites, and their biopatterns, are being used as biomarkers for the diagnosis, prognosis, and prediction of mortality in critically ill patients [
12,
13]. In this retrospective observational matched cohort study, we aimed to examine the prediction or prognosis of 90-day mortality among patients with bacterial CAP and identify those who are at the highest risk of dying in hospital (in-hospital mortality) using a multi-platform metabolomic approach. We applied non-targeted proton nuclear magnetic resonance (
1H-NMR) spectroscopy, gas chromatography mass spectrometry (GC-MS), and targeted direct infusion tandem mass spectrometry (DI-MS/MS) to analyze metabolomic biopatterns of plasma samples for CAP patients collected within 24 h of admission to hospital for prognostication of mortality.
Discussion
The present study was designed to examine the role that metabolomics might play in the prognosis of mortality among patients with bacterial CAP. In a comprehensive analysis, the results of this study showed that lipid compounds offer important insights into the prognosis of 90-day mortality of bacterial CAP on the 1st day of admission to the hospital. Moreover, we also demonstrated that lipid profiling is capable to predict in-hospital mortality from survivors (> 90 days) using a plasma sample drawn on the 1st day of admission to the hospital. Our result provides important insight into the prediction of mortality in patients who are at the highest risk of dying in hospital. In a targeted approach using DI-MS/MS, lysoPCs and ACs were prognostic metabolites for the mortality of bacterial CAP when compared to survivors. Correspondingly, decreased lysoPCs, increased ACs, and decreased PCs significantly changed yielding a fingerprint to prognosticate in-hospital mortality. Besides the prognosis of mortality of bacterial CAP, we also showed that targeted lipid profiling using a DI-MS/MS platform could be used to predict the need for ICU admission and assessment of severity for the patients with bacterial CAP. Interestingly, decreased lysoPCs, increased ACs, and decreased PCs were associated with severity and ICU admission requirement.
Power analysis using multivariate data analysis (using R-based analysis) [
21] showed that using the most differentiating metabolites, 24 samples in each group could provide a significant difference between cohorts (FDR < 0.05) with a power
β = 0.8 for the study. Thus, the enrollment of 75 samples in each group indicates the study should have sufficient power to detect a difference between groups with 80% power.
To our knowledge, this is the first study to identify lipids and related metabolites as potential predictors of in-hospital mortality in CAP patients as well as prediction of ICU admission and assessing CAP severity. Metabolite biomarkers can be used independently or to supplement other approaches like genomics and proteomics biomarkers or even clinical scoring systems in precision medicine. Lipids are very diverse biomolecules comprising a range of different classes including ACs, glycerophospholipids, SMs, sterol lipids, and glycerolipids. Lipids can play important roles in many biological and physiological functions such as structural components of cell membranes, intermediates in signaling pathways, homeostasis, and immunity [
22]. Once these studies are externally validated, these findings have a promising capability of being translated into clinical practice. Primarily, DI-MS/MS can be considered as a shotgun technique to quantify limited number or hundreds of metabolites per sample in a very short period of time (hours). Using the quantitative and targeted approaches shown in this study will allow one to develop a new prognostic tool for bacterial CAP investigations and prognostication. In addition, this study showed that the semi-quantitative and untargeted GC-MS and NMR analytical platforms appear to be not efficient enough at this time to be prognostically helpful for bacterial CAP. Importantly, this study showed that lipids are important metabolites for the prognosis of 90-day mortality and in-hospital mortality while other metabolites quantified in this study such organic acids, amino acids, amines, sugars, and sugar alcohols were not as predictive of mortality in bacterial CAP. Specifically, lipid compounds including saturated and unsaturated fatty acids are in close relationship with mechanisms of inflammation, the central essential mechanisms of the host response to bacterial infections such as seen in bacterial CAP [
23]. They are active substances and important inflammatory mediators in both pro-inflammatory and anti-inflammatory mechanisms [
24,
25]. On the other hand, 90% of surfactant in the lung is formed by lipids and PCs make up more than 80% of these lipids. Pneumonia may cause surfactant changes leading to an alteration in lipid metabolism [
26]. The low concentration of lysoPCs in non-survivors may be caused by the consumption of lysoPCs in the early stages of disease or by conversion of lysoPCs by phospholipase A due to increased secretion of autotoxins [
27]. PCs are also major components of the lipid bilayers of cell membranes as well as lung surfactant and both are important in lung development [
28]. Alveolar type II cells are responsible for the synthesis and accumulation of PCs in the lung [
29]. It is assumed that damage to the cell membrane, alveolar cell integrity, and surfactant dysfunction due to inflammatory illness such as pneumonia is associated with increased PCs in the blood that could be correlated with the severity and mortality of CAP. Indeed, several studies have shown changes of lipid concentration in the blood after acute lung injury due to sepsis and bacterial and viral infections [
30,
31]. All of the former strongly suggests that lipids may be interesting targets as putative biomarkers for the prognosis of mortality of CAP in clinical practice.
The current findings are consistent with those reports showing the importance of fatty acids and lipids in the diagnosis and prognosis of CAP and other respiratory complications such as ARDS and septic shock. For example, low- and high-density lipoprotein cholesterol (LDL-C and HDL-C) were found to be independent predictors for bacterial CAP adverse outcomes [
22]. In addition, the alterations of some fatty acids such as docosahexaenoic acid, eicosapentaenoic acid, and oleic acid were associated with increased and decreased risk of CAP in women and men [
32,
33].
Several studies suggest that an increase of AC compounds occurs in patients with CAP. Overall, previous studies have shown increased short-, medium-, and long-chain ACs in CAP patients compared to non-CAP patients [
34] and in patients with other types of infections (intraabdominal infections, acute pyelonephritis, and primary gram-negative bacteremia) [
35,
36]. In terms of sepsis, ACs were high in bacteremic patients that did not survive sepsis [
37] and in patients with sepsis compared to non-infectious SIRS [
38]. These studies show that ACs could be specific biomarkers for CAP and might be associated with disease severity as these compounds increase in non-survivors. In the current study, we also showed increased ACs in CAP non-survivors vs. survivors.
Decreased PC concentrations in blood have been reported in some invasive bacterial infectious diseases such as sepsis, CAP, and bacteremia [
39]. It has been reported that there is a decreased level of PCs and phosphatidylinositol (PI) in BALF and a significant decrease in phosphatidylglycerol (PG) and SMs in severe pneumonia, and in ARDS associated with pneumonia compared to other less severe pneumonia patients and controls [
40]. There is also a report of decreased level of PCaaC34:3 in CAP patients compared to different types of infections [
41]. Glycerophospholipids such as lysoPCs, lysoPEs, and lysoPIs are other possible biomarkers for the diagnosis and prognosis of CAP reported in several studies, which are briefly discussed below. A decrease of lysoPEs and lysoPCs was found in CAP patients compared to non-CAP patients. LysoPEs were found decreased in fatal cases of CAP compared to non-fatal cases [
36] and also were found increased in survivors of CAP compared to other types of infection [
41]. Specifically, lysoPCs were significantly lower in CAP non-survivors than in survivors, introducing these as potential prognostic markers for CAP patients who require hospitalization [
35].
Although similar to ACs, lysoPCs have been shown to be increased in patients with CAP compared to patients with other types of infection [
41]. Nonetheless, the decreased lysoPCs (LPCs 16:0, 16:1, and 18:0) were associated with severe septic shock non-survivors at day 1 and day 7 in 28-day and 90-day mortality studies [
34]. At the clinical phenotypic level, lower concentrations of lysoPCs were in inverse correlation with PSI and CURB scores in CAP non-survivors on day 1 of admission to hospital and non-survivors admitted to the ICU with severe sepsis or septic shock in 28-day mortality studies [
35,
42]. The data showed glycerophospholipids, particularly lysoPCs, appear to be specific biomarkers for bacterial CAP and are probably associated with the severity of CAP as they decrease in non-survivors versus survivors. These results also support our current findings showing lower levels of lysoPCs in CAP non-survivors vs. survivors. Decreased lysoPCs are associated with the acute stage of CAP that can increase over time during treatment [
27]. The data highlight the role of glycerophospholipids (lysoPCs) in respiratory diseases as lysoPCs are major lung surfactant phospholipids which are potentially involved in cellular inflammation and proliferation mechanism association with atherosclerosis and inflammatory disorders [
38]. LysoPCs are the main degradation product of phospholipids when they are oxidized during apoptosis and lead to either harmless or highly toxic phospholipids [
43,
44].
Regardless of the role of lipid profiling for the diagnosis and prognosis of bacterial CAP, profiling of other metabolites has been widely applied to diagnose and differentiate respiratory disorders including CAP. Plasma-based metabolomics of 240 critically ill patients (SIRS, sepsis, sepsis-induced ARDS) revealed the application of metabolomic profiling for the prognosis of 28-day mortality using GC-MS and LC-MS techniques. Amino acids, carbohydrates, and nucleotides were among the most differentiating metabolites to predict the mortality [
45]. In a study of only 30 CAP patients from the GenIMS study, the same study population of the current study, UHPLC-MS/MS- and GC-MS-based metabolomics showed a contribution of 423 metabolites to separate survivor from non-survivor cohorts. Of these, 56 metabolites were selected based on their lower false discovery rate (
q < 0.1) which showed the biggest differences between the two cohorts [
46]. Also, increased phytosphingosine, sphinganine, creatine, lactate, and methoxyacetic acid and decreased 4-hydroxybenzensulfonic acid, dehydroepiandrosterone sulfate (DHEA-S), and
l-arginine were capable of differentiating patients with severe CAP from a non-severe cohort [
47]. The high sensitivity of metabolites to intrinsic stimuli in association with high throughput revealed 11 volatile organic compounds (VOCs) in exhaled breath samples which could discriminate pneumonia patients from controls (patients without pneumonia). Moreover, 52 VOCs were significantly lower in patients with positive cultures compared to those with negative culture [
48].
Here we show that metabolites, particularly lipids, could be more reflective biomarkers for prognosis of mortality of bacterial CAP rather than other historic biomarkers such as proteins and cytokines. Abnormally expressed plasma cytokines, chemokines, and PCT and CRP can be used for the diagnosis and outcome prediction of CAP; however, they may not be predictive enough for the prognosis of CAP outcomes especially early in the disease process [
6]. Also, lipid profiling particularly ACs and lysoPCs may be considered as further potential biomarker to assess patients who need ICU admission and to assess pneumonia severity. In addition, the most common severity scoring systems do not have high enough AUROCs to be capable of predicting 30-day mortality in CAP [
8].
PCs and lysoPCs are the most abundant glycerophospholipids and are major components of all cell membranes and pulmonary surfactant [
49], and moreover, PCs are the most abundant phospholipids contributing to ATP synthesis and multiple critical mitochondrial functions such as apoptosis, autophagy, and mitochondrial electron transport chain reaction [
50]. Importantly, the decrease of PCs and lysoPCs could reflect a loss of alveolar epithelial cells and their functions.
The strength of the current study is reflected by the comprehensive metabolomic approach that shows the potential application of lipid profiling for the prognosis of 90-day mortality and in-hospital mortality based on samples from the 1st day of admission to hospital using highly specific and significant predictive models. Although the prediction score (Q2 = 0.298) may not be high for the prognosis of 90-day mortality in this study, the intercorrelation of lipid compounds from the same subclass (i.e., lysoPCs, SMs, PCs, and ACs) and similarly changing trends of lipid metabolites enhances the predicting power of lipid profiling for the prognosis of mortality. Additionally, using both multivariate and univariate data analysis, two different approaches, showed the same changing trends among lipids, which strengthens the probability of the prediction value of lipids in bacterial CAP.
Limitations of this study include a relatively small sample size especially since there is considerable heterogeneity of the bacterial CAP cohorts due to comorbidities such as sepsis, CHF, and neurological disorders (Table
1) and the heterogeneity within complexity and severity of bacterial CAP can affect the prediction power. In addition, the prediction power for the prognosis of short-term outcome may not be as strong in retrospective case-control studies when compared to prospective studies. Adding to the variation of the study, sampling has been done in more than 24 centers (in this multicenter study population). Sample handling in multicenter studies could be one of the major sources of variation among samples impacting metabolomic profiling and therefore prediction accuracy. Nonetheless, we believe that the multicenter sampling, in particular, the geographical distribution of the current study population might be a potential strength for the validity of prognosis of mortality in which samples represent the different populations of north, northwest, and central USA. Validation of this study using lipid-based metabolomic determination is required in further analyses.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.