Introduction
The mortality associated with acute respiratory distress syndrome (ARDS) remains above 40% despite many advances in intensive care [
1]. Many clinical trials have evaluated the efficacy of certain drugs in ARDS, but these have mostly failed to improve the clinical course [
2]. In clinical practice, it is important to screen patients at risk of developing ARDS and to modify risk factors as much as possible. The Lung Injury Prediction Score (LIPS) is a scoring system used to screen for patients at high risk of developing ARDS [
3,
4]. However, it has some limitations as a predictive scoring system, because it has a high sensitivity but a low specificity.
In addition, since ARDS is a heterogeneous disease in terms of its causes and clinical aspects, intensivists have been interested in classifying subtypes and applying different treatments to these. ARDS subphenotypes have typically been divided into direct (pulmonary) ARDS and indirect (extrapulmonary) ARDS, according to the etiology. In direct ARDS, alveolar collapse, fibrin deposition, and pulmonic edema are more common in terms of pathological findings [
5,
6]. Ground glass opacities and consolidations are relatively asymmetrical in the radiological findings in direct ARDS [
7]. On the other hand, indirect ARDS shows relatively bilateral ground glass opacities, rather than asymmetrical consolidations, and with compliance, it responds better to positive end-expiratory pressure than does direct ARDS [
6‐
8]. These clinical differences result from the differences in the main pathophysiology between direct ARDS (epithelial injury) and indirect ARDS (endothelial damage and systemic inflammation). Because these subphenotypes present different clinical courses and treatment responses, they may be helpful for distinguishing the ARDS phenotype. Although biological discrimination of the two groups is not easy due to overlapping etiologies related to lung damage, several recent studies using metabolomics have shown that there are differences in metabolic fingerprints between these two groups [
9,
10].
Metabolomics is a new, rapidly expanding field of systems biology with the ability to measure all small molecules, chemicals, and metabolites that can be identified in a given sample comprehensively [
11,
12]. Metabolomics is based on high-throughput approaches, which separate the population of unknown small molecules for subsequent quantification and identification, rather than quantifying a molecule of known identity [
13]. The resolution of compounds using gas chromatography (GC) and liquid chromatography (LC) has provided significant benefits as compared with two-dimensional gel electrophoresis [
13]. Mass spectrometry (MS) and nuclear magnetic resonance (NMR) spectroscopy have promoted accuracy and sensitivity for identifying unknown metabolites [
14]. Such an untargeted approach allows acquisition of the global metabolite profile in a biological compartment, without any prior hypothesis, facilitated by a blind comparison between cases and controls [
15]. Nevertheless, limitations in quantitative measurements and metabolite annotations remain problematic in untargeted metabolomics [
16]. On the other hand, a targeted metabolomics approach measures and analyzes metabolites in known or predicted metabolic pathways. Analytical methods for targeted metabolomics can be optimized and their quantitative results should be more reliable than those of untargeted metabolomics [
16].
The aim of this study was to find diagnostic metabolites that distinguish sepsis-induced ARDS patients from non-ARDS controls using this targeted metabolomics approach, and to identify metabolites and related pathways that can differentiate sepsis-induced direct and indirect ARDS. A targeted metabolomics strategy, focusing on energy metabolism, free fatty acids, amino acids, sphingolipids, and phospholipids were chosen because it has been reported that these metabolites might have some connection to metabolic alterations in ARDS [
17‐
22].
Discussion
In this study, we investigated the differences in the metabolome and related pathways between sepsis-induced ARDS and non-ARDS controls, and between sepsis patients with direct and with indirect ARDS as a subphenotype, using LC–MS/MS and GC–MS for targeted metabolomics. Among 186 metabolites detected, 102 metabolites could differentiate sepsis-induced ARDS patients from the non-ARDS controls, and 14 metabolites could discriminate between the direct and indirect ARDS subphenotypes. In PLS-DA, we found that sepsis-induced ARDS was metabolically distinct from the non-ARDS controls, and that sepsis-induced direct and indirect ARDS were also metabolically different subgroups.
We identified significant differences in sepsis-induced ARDS patients and non-ARDS controls in terms of metabolites related to ARDS biology. The main substances involved were PE plasmalogens, including C18(Plasm) lysoPE, C18(Plasm) 20:4 PE, and C18(Plasm) 22:6, which were decreased in patients as compared to controls. Plasmalogens are plasma-borne antioxidant phospholipid species that provide protection as cellular lipid components during cellular oxidative stress [
25]. According to a previous human study, plasmalogens were decreased in the bronchoalveolar lavage (BAL) fluid of an ARDS patient [
26]. In a recent study that evaluated the serum lipid profile of COVID-19 patients, it was confirmed that PE plasmalogens were significantly decreased as the P/F ratio decreased [
27]. In sepsis patients, plasmalogens also decreased, which suggested that these molecules may be a marker of oxidative stress [
28]. Recently, it was found that PE plasmalogens were decreased further in moderate to severe COVID-19 disease, and among the relevant metabolites, the decrease in PE (P-18:0/20:4) and PE (P-18:0/22:6) was consistent with our results [
29].
It had not previously been clear how PC levels change in ARDS. The concentrations of PC (33:6) and PC (32:0) were increased in sepsis-induced ARDS patients as compared to non-ARDS controls. However, according to a recent study, plasma PC concentrations, mainly in the PC (18:2) series, were significantly lower in ARDS patients than in normal controls, which was thought to be due to abnormal changes in PC synthesis in the liver of ARDS patients [
30]. However, in our results, PC (33:6) and PC (32:0) levels were higher in sepsis-induced ARDS patients than in the controls, suggesting that the direction of increase or decrease may be different for each PC series. While several studies have found an increase in PC levels in sepsis patients [
31‐
33], other studies have shown a decrease in these patients as compared to normal controls [
31,
32,
34,
35]. In one study, the PC (34:3) level was significantly lower in patients with pneumonia as a primary focus of sepsis than in patients with other primary foci, such as intra-abdominal, urinary tract, or blood stream infections. On the other hand, a high PC (34:1) level was shown to be a prognostic marker suggestive of septic shock in a pneumonia group [
31]. These results suggest that PC species containing long-chain fatty acids can be important metabolites in sepsis-induced lung injury or ARDS.
We confirmed that the metabolites that showed the most significant difference between sepsis-induced direct and indirect ARDS were lysoPCs, including lysoPC (17:6), lysoPC (18:0), and lysoPC (16:0). The concentrations of lysoPC in the sepsis-induced ARDS groups were significantly decreased as compared with the non-ARDS control group, while the direct ARDS group showed higher levels than did the indirect ARDS group. LysoPC is a lipid mediator derived from membrane PC, which has been suggested to regulate immune responses. PC is hydrolyzed by phospholipase A
2 (PLA
2), resulting in the production of lysoPC [
36]. LysoPC is known to contribute to inflammation by increasing chemokine production and activating endothelium, neutrophils, monocytes, macrophages, and lymphocytes [
37]. However, the role of lysoPCs in ARDS has not yet been clearly elucidated. Although there were differences in the change in lysoPC levels in each study, the enzymes involved in this process have been suggested to be biomarkers for ARDS or acute lung injury [
38]. In a preclinical study on the activity of type II PLA
2, the concentration of lysoPC in BAL fluid was higher than that of controls with a higher activity of type II PLA
2 [
39]. In clinical studies, PLA
2 activity in BAL fluid and plasma of patients were also increased as compared to controls [
40,
41]. Additionally, when comparing direct and indirect ARDS, PLA
2 activity was higher in direct ARDS cases [
40]. On the other hand, studies of sepsis patients showed the opposite results. In sepsis patients, serum lysoPC levels were lower than those in controls, and among them, lysoPC (16:0) and (18:0) were decreased [
42]. Lower concentrations of serum or plasma lysoPC predicted worse outcomes [
43], and the ratio of lysoPC/PC was lower in sepsis patients than in healthy controls [
42]. In our study, the decrease in serum lysoPCs was similar to that seen in sepsis. However, higher concentrations in direct than in indirect ARDS patients were consistent with previous studies showing increased results of lysoPC (16:0) and lysoPC (18:0) in lung diseases with epithelial damage [
44]. That is, our results appear to be primarily based on biological changes in sepsis because our study patients had ARDS based on sepsis. But, since they showed metabolic differences following direct or indirect lung injury, the corresponding lysoPCs could be important markers for differentiating between the two subphenotypes of ARDS.
We identified that sphingolipid metabolism is an important pathway for distinguishing between sepsis-induced direct and indirect ARDS patients, and confirmed that there was a difference in S1P. S1P is a naturally occurring bioactive sphingolipid generated by sphingomyelin metabolism [
45]. It is generated by the phosphorylation of sphingosine, catalyzed by sphingosine kinases (SphKs) 1 and 2, and is catabolized by lipid phosphate phosphatases, S1P phosphatases, and S1P lyase [
45]. S1P plays an important role in the vascular and immune systems [
46]. In acute lung injury, S1P has been recognized as a potent angiogenic factor enhancing lung endothelial integrity and inhibiting vascular permeability [
45,
47]. Previous studies showed different results, depending on whether it was a preclinical or clinical study [
48‐
52]. Preclinical studies showed upregulation of S1P in lung tissues, BAL fluid, and plasma in cases with acute lung injury [
48‐
50], whereas a clinical study showed lower serum S1P levels in ARDS patients [
52], which was consistent with our results. In addition, we observed a distinct pattern between direct and indirect ARDS at the level of sphingosine. In indirect ARDS patients, sphingosine was upregulated as compared to non-ARDS controls or direct ARDS patients. This result suggested that the following three mechanisms can be a clue to distinguishing between indirect and direct ARDS: 1) activation of S1P phosphatases, which are rate limiting enzymes for reversing S1P into sphingosine; 2) inactivation of SphKs, which convert sphingosine to S1P; or 3) upregulation of the S1P receptor. However, this will require further research, as we could not confirm the activity of these enzymes in each group.
Since there was a difference in the underlying disease between the ARDS patients and the non-ARDS controls, we analyzed whether the presence of co-morbidities such as chronic liver disease and solid tumor affected metabolic differences. Thus, we performed PLS-DA for ARDS patients and control individuals without chronic liver disease, and found that metabolites such as C18(Plasm) LPE, PC(33:6), C18(Plasm) 20:4 PE, and PC(32:0) were still included among the top differential metabolites between the ARDS patients and the control group. Also, when we ran the analysis again by excluding those with a solid tumor, the main 4 metabolites that distinguished ARDS from non-ARDS controls remained in the top metabolites. In the ARDS subphenotype analysis (direct vs. indirect), there was no significant difference in the co-morbidities between the groups, except for the higher BMI in the indirect group. However, we assume that BMI would not have significantly affected the distinguishing metabolites for two reasons. Firstly, the proportion of patients in the indirect ARDS group who were actually obese (BMI ≥ 30 kg/m2) was low; secondly, a correlation analysis between BMI and distinguishing metabolites indicated that the value of the correlation coefficient was very small, albeit a significant negative correlation between them. However, because the number of patients in our study was small, it is necessary to confirm this with more patients in the future.
Our study shares similarities with the Metwaly study [
9], as both compare ARDS and control groups using metabolomics, analyze direct and indirect ARDS, and present associated metabolic pathways and biomarkers. A significant commonality between the two studies is that the ARDS group in both cases was sourced from the sepsis network. Glycerophospholipid and tryptophan metabolism were consistently identified as metabolic pathways associated with ARDS in both studies. Additionally, the finding that the primary pathway of indirect ARDS is related to energy metabolism was consistent across both studies. However, there were several differences between the two studies. First, the metabolomics methodology varied; the Metwaly study employed an untargeted approach using
1H-NMR spectroscopy and GC–MS, while our study used a targeted approach with LC–MS/MS and GC–MS. Second, the blood sampling time differed. In our study, a single blood collection for the ARDS group occurred within 48 h after ICU admission, while the Metwaly study collected samples on ICU admission day 1 and tracked them temporally in some recovered patients. Third, the study controls were distinct; the Metwaly study used ICU-ventilated patients unrelated to lung diseases as controls, while our study's controls were subjects who visited medical institutions for health checkups without acute disease or chest radiography abnormalities. Lastly, the ARDS patient characteristics varied. Our study's patients were older, predominantly male, had higher severity at ICU admission, and experienced higher 28-day mortality than those in the Metwaly study. Despite these differences, both studies share key findings: ARDS is metabolically distinct compared to the control group, and ARDS subphenotypes exhibit clear differences.
Our results align with a recent study by Alipanah-Lechner et al. [
53], which differentiated ARDS into hyperinflammatory and hypoinflammatory subtypes. This study, like Metwaly's [
9], focused on sepsis patients with ARDS, making its patient group quite similar to ours. The hyperinflammatory and hypoinflammatory subtypes in their study closely resembled the indirect and direct ARDS subphenotypes in our research. Patients with hyperinflammatory ARDS exhibited significantly lower plasma lipid concentrations and increased levels of glycolytic metabolites, such as lactate and pyruvate, compared to those with hypoinflammatory ARDS. Similarly, in our study, patients with indirect ARDS had lower lipid levels (primarily very long-chain fatty acids, lysophosphatidylcholines, and some sphingomyelins) and higher levels of glycolytic metabolism. Our findings also revealed that the primary metabolic pathways involved in indirect ARDS were glycolysis/gluconeogenesis metabolism and pyruvate metabolism, which supports this observation. These overlaps between the results of the two studies suggest that the clinically different ARDS subphenotypes are consistent with metabolically distinct subtypes.
The present study had several limitations. First, we collected a sample only at one time point, at ICU admission. The serial change patterns of metabolites are related to the prognosis of ARDS [
9], but we could not confirm this. Second, although blood samples were taken as early as possible on the first day of diagnosis, there could be differences between the onset of disease and time of diagnosis. Thus, some metabolites may have been limited in terms of their presence in the serum. Third, since all of our study patients had infections, the interpretation of our results is limited as to whether these findings are specific to ARDS. Further validation of the identified metabolites and pathways is required, and additional research on ARDS caused by non-infectious etiologies is necessary.
In conclusion, despite these limitations, our study demonstrated a marked difference in the metabolic pattern between sepsis-induced ARDS and non-ARDS controls. We also identified that direct ARDS and indirect ARDS are metabolically distinct subphenotypes. In particular, lysoPC (17:6), lysoPC (16:0), and lysoPC (18:0) of glycerophospholipid metabolism and S1P of sphingomyelin metabolism demonstrated potential as important markers for subphenotype distinction. This study provides a basis for further research into the development of theranostics based on these metabolites.