The acute respiratory distress syndrome (ARDS) is a lung inflammatory process that develops primarily as a response to respiratory or systemic-induced sepsis, which causes a disproportionate mortality burden in the adult intensive care unit (ICU) and has disabling consequences for years in surviving patients (1–3). ARDS occurs in 7 cases per 100,000 people per year, although the estimate varies widely among studies since the clinical diagnostic criteria are nonspecific. Its overall mortality rate remains high in most series, around 30%–40% (2–4). ARDS still has no effective and efficient treatment despite multiple studies that have focused on identifying the pathophysiology and improving the prognosis of these patients since it was first described. To date, lung-protective mechanical ventilation (MV) remains the main standard supportive ARDS treatment (5, 6) and there is no specific pharmacological therapy for it. Thus, identifying specific biomarkers will help to develop early therapeutic and preventive therapies, while assisting in predicting the prognosis of individual ARDS patients (3, 7, 8).

Mitochondria are a bioenergetic and biosynthetic cell organelle and a signaling hub that controls several important cellular functions, including cell survival and differentiation, as well as functioning of inflammatory responses (9, 10). The multiorgan and cellular dysfunction underlying sepsis and leading to ARDS could trigger mitochondrial dysfunction, which is characterized by fragmentation of the mitochondria and loss of integrity of mitochondrial DNA (mtDNA) (11). Previous studies have shown that, while the cellular mtDNA levels decrease, the cell-free mtDNA levels increase in response to a stimulus due to major trauma or a microbial infection (12–16). Based on evidence from animal models (17) and patient studies (18, 19), circulating mtDNA levels have been proposed as a potential biomarker for the systemic inflammatory response and lung injury after major trauma.

Cell-free mtDNA is considered a molecular pattern associated with damage (DAMPs) and could act as a critical activator of the innate immune system and inflammation (7, 9, 20, 21). Circulating cell-free mtDNA, measured by quantitative PCR (qPCR), has been associated with the overall 28-day mortality in ICU patients (22). Among ARDS patients, plasma mtDNA levels measured by qPCR on day 7 after diagnosis were significantly higher among non-surviving patients (23). Similarly, the plasma mtDNA levels among sepsis patients admitted to the emergency room were significantly higher among those who did not survive, and a score combining their levels with plasma lactate concentration considerably improved the 28-day mortality prediction (24). In fact, molecular patterns associated with mtDNA damage in transfusion products significantly contribute to the incidence of ARDS after massive transfusions (25).

Based on this evidence, and following a pragmatic approach, we tested the association of array-based measures of whole-blood mtDNA (wb-mtDNA) copies within 24 h of sepsis diagnosis with 28-day patient survival. We hypothesized that early wb-mtDNA measurements could be associated with mortality in patients with sepsis and ARDS.

Demographic and clinical features of all sepsis patients and of those with or without ARDS are shown in Table 1 . As expected, there were large and significant differences between ARDS and non-ARDS patients for their severity scores, hospital and ICU length of stay, mortality rate, and physiological variables such as the partial pressure of oxygen (PaO₂). There were differences in the organ dysfunction among ARDS and non-ARDS patients (ANOVA, p < 0.001). However, as expected, they were due to the lung affectation. On the other hand, for the pathogens, we did not observe significant differences between ARDS and non-ARDS patients (ANOVA, p = 0.127).

We first found that the association between wb-mtDNA copies within 24 h of sepsis diagnosis and 28-day survival among all patients with sepsis from GEN-SEP was significant ( Table 2 ). We then tested the same model stratifying the sepsis patients by those who did not develop ARDS and those who developed ARDS. In all models, the proportional risk assumption was held. Although we did not observe any association between wb-mtDNA copies and 28-day survival in septic non-ARDS patients, we found a strong association in septic patients who developed ARDS (hazard ratio [HR] = 3.65, 95% confidence interval [CI] 1.39–9.59, p = 0.009). Taken together, this indicates that the significant association between the wb-mtDNA copies within 24 h of sepsis diagnosis and 28-day survival observed among all GEN-SEP patients could be explained by those developing ARDS. A Kaplan–Meier analysis of the 28-day mortality among the ARDS patients reinforced this observation (log-rank test p = 0.037) ( Supplementary Figure S1 ).

Given that there were demographic and clinical differences between ARDS patients that survived and that did not survive ( Table 3 ), sensitivity analyses were conducted to ensure that those differences did not explain the association with wb-mtDNA. Note that there were no overall differences among ARDS patients in the organ dysfunction (ANOVA, p = 0.159) or in the pathogens (ANOVA, p = 0.985), therefore not affecting the sensitivity analyses. We found that the association was robust to model adjustments by variables that were not significantly different between survivor and non-survivor ARDS patients (e.g., gender, comorbidities, and the first five PCs of genetic heterogeneity). Likewise, the results of the univariate model with wb-mtDNA levels were similar to those with independent adjustment by age, SAPS, APACHE II score, and organ dysfunction, where the assumption of proportionality risk was held ( Table 4 ). The only two adjusted models that did not hold for the proportionality risk assumption were those including as covariates the length of stay in ICU or in the hospital, for which conclusions should be taken with caution. Based on these findings, we tested multivariate models in the sensitivity analyses except those with a high proportion of missing data (i.e., SAPS) or were variable adjustments that turned into violated assumptions of the proportionality of risks in the sensitivity analyses (i.e., length of stay in ICU or in the hospital) ( Table 4 ). We found that when including age and APACHE II score, the association of wb-mtDNA with survival was similar (HR = 4.51, 95%CI = 1.62–12.47, p = 0.004) and the proportionality of risk was held. The results were similar, and the proportionality of risk was also held, when the model adjusting for age and APACHE II also included the organ dysfunction categories that reached nominal significance in the differences between ARDS patients that survived and that did not survive (HR = 4.40, 95%CI = 1.55–12.53, p = 0.006).

Among the ARDS patients, the AUC of the wb-mtDNA copy number for the 28-day survival was 0.612 (95% CI = 0.541–0.683) ( Supplementary Figure S2 ). However, this predictive value was similar to that provided by other clinical scores routinely used in clinical settings. As an example, for the same patients, the prognostic ability of the APACHE II score reached an AUC of 0.634 (95% CI = 0.565–0.704) ( Supplementary Figure S2 ). Combining both the wb-mtDNA copy number and the APACHE II score in the models, the AUC slightly improved to 0.676 (95% CI = 0.609–0.742) although the AUC of the two curves (wb-mtDNA alone and wb-mtDNA plus APACHE II together) were not significantly different (p = 0.062) based on DeLong’s test for two correlated ROC curves.

Given the multifactorial risks involved in the prognostic trajectories of ICU patients (33), the identification of an ideal biomarker for predicting outcomes is a difficult task. One of the hallmarks of ARDS is the presence of inflammatory, protein-enriched pulmonary edema (1), caused by an increase in the permeability of the lung tissue (34, 35) and elevating the risk of death (2). It has been reported that DAMPs, including mtDNA, increase endothelial permeability through neutrophil-dependent and independent pathways (36). In a sufficiently powered cohort of Spanish patients recruited from a nationwide network of postsurgical ICUs, we describe evidence supporting that mtDNA copies measured in peripheral blood within 24 h of sepsis diagnosis were associated with 28-day survival in patients developing ARDS. Given that the association was absent among non-ARDS patients from the same series, our findings might suggest an ARDS-specific effect.

In agreement with our findings, Nakahira and colleagues observed an association between circulating cell-free mtDNA with overall patient 28-day mortality in the ICUs and if they combined mtDNA levels with other clinical parameters, the prediction of the ICU patients improved (22). Other studies have shown that high mtDNA plasma levels could be associated with sepsis and ARDS (19, 23, 24, 37). In one of these studies, high mtDNA plasma levels and a strong association with 28-day survival were observed in patients with sepsis and septic shock (24). Supporting our results, Huang and colleagues also observed a positive association between higher mtDNA plasma levels and 28-day mortality among patients with all-cause ARDS, although their findings revealed an association only with mtDNA measures of day 7 after diagnosis (23). Likewise, mtDNA plasma levels were also analyzed in other critically ill patients, such as the patients affected by the coronavirus disease 2019 (COVID-19) (38) or trauma patients (19), where also high mtDNA levels were associated with poor prognostics or outcomes of these diseases.

Among the strengths of this study, we recognize that it was based on a well-phenotyped and clinically characterized cohort of sepsis and ARDS patients. In addition, the wb-mtDNA estimates were obtained with a method that is not subject to the sample conservation problems linked to the qPCR (39). Related to this, given that we relied on an SNP array platform, we were able to perform model adjustments by the genetic heterogeneity, which is inherent in any heterogeneous patient population. Nevertheless, our study has some major limitations as well. The main weakness is that the models lacked adjustments by platelet count or the different cell populations, precluding determination of the contribution of the different types of white blood cells and platelets to the overall wb-mtDNA copy number estimation (13, 40). Another important limitation is that we were unable to adjust the models for other relevant clinical data that can be prognostic of ARDS such as creatinine, mean arterial pressure, Glasgow coma scale, and urine, among others. The analyses also lacked longitudinal measures that could have provided dynamical insights into mtDNA levels and outcomes (23). Besides, we have considered all the mtDNA content from the peripheral blood limiting the comparisons with other studies that have focused on the circulating cell-free fraction of mtDNA (22, 41). A final important limitation is that we were not able to infer causality.

Wb-mtDNA copies measured within 24 h of sepsis diagnosis are significantly associated with 28-day survival in ARDS patients. Further studies should disentangle whether this association is independent of sepsis and whether the causality of wb-mtDNA elevation is involved in the pathogenesis of ARDS.

The raw intensity data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the Ethics Committee for Drug Research from the Hospital Universitario de Canarias (Code: CHUNSC_2018-16). The patients/participants provided their written informed consent to participate in this study.

TH-B: data analysis, interpretation, and manuscript drafting. BG-G, HR-P, IM-R, JL-S, AC: performed the experiments, data analysis, and revision of the manuscript. MP-G, AR-P, DC, JB, AA, EG-H, NC, MG-G, EE, AM, DD, AG, JA, MS, JB, JG: performed the experiments, sample and clinical data collection, and data analysis. JV: analysis, interpretation, critical revision of the manuscript. CF: study conception and design, data analysis, interpretation, critical revision of the manuscript and conception of the project. All authors contributed to the article and approved the submitted version.

This study was funded by the Instituto de Salud Carlos III, Madrid, Spain (CB06/06/1088, PI14/00844, PI16/00049, PI17/00610, PI20/00876, FI17/00177, FI18/00230, and CD19/00231), and by the European Regional Development Funds (ERDF) “A way of making Europe” from the European Union; by the agreement OA17/008 with Instituto Tecnológico y de Energías Renovables (ITER), Tenerife, Spain, to strengthen scientific and technological education, training, research, development, and innovation in Genomics, Personalized Medicine, and Biotechnology. The genotyping service was performed at CEGEN-PRB3-ISCIII, supported by grant PT17/0019 of the PE I+D+i 2013-2016, funded by the Instituto de Salud Carlos III and ERDF.

MG-G was employed by SPECTRUM, LLC.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.