Circulating Mitochondrial DNA in Patients in the ICU as a Marker of Mortality: Derivation and Validation

1Multidimensional scoring systems based on commonly measured clinical and physiologic parameters help to improve the prognostication of patients admitted to intensive care units (ICUs). Despite the improvement in prognostication there is a need to develop biomarkers that can help clinicians to further improve risk prediction for patients in the ICU. While many biomarkers (>170) have been evaluated in various ICU subpopulations, few have been tested in large populations, or add substantially to the predictive value of multidimensional scoring systems. Only a small number of biomarkers reflecting the severity of critical illness are commonly measured in clinical practice (e.g., lactate and procalcitonin),. Accumulating evidence suggests that cellular mitochondrial DNA (mtDNA) levels are decreased, and circulating cell-free mtDNA levels are increased in response to stimuli such as trauma and microbial infection –. Moreover, mtDNA acts as a damage-associated molecular pattern that can drive molecular processes, leading to inflammatory responses and organ injuries,,,. Based on these findings we hypothesized that circulating cell-free mtDNA levels would be associated with mortality and would improve risk prediction in ICU patients. The objective of this study was to test whether measuring circulating cell-free mtDNA is useful as a biomarker in the ICU. To our knowledge this is the first study to provide replicated evidence that plasma mtDNA levels both are associated with mortality and improve risk prediction in the field of critical care illness.

2The protocols for patient recruitment and data collection for the two patient cohorts, the Brigham and Women's Hospital Registry of Critical Illness (BWH RoCI, primary patient cohort) and the Molecular Epidemiology of Acute Respiratory Distress Syndrome (ME ARDS, replication cohort), have been described previously – and are discussed in detail in Text S1 in Supporting Information S1. In brief, from 24 July 2008 to 14 June 2011, 200 patients were prospectively recruited into the BWH RoCI study from the Brigham and Women's Hospital medical intensive care unit (MICU), for whom clinical phenotype data, including mortality, and blood plasma samples were subsequently collected. Acute kidney injury was defined as a ≥0.5 mg/dl increase in serum creatinine. Presence of respiratory, heart, and liver failure were defined by admitting physician report. For replication, we selected 243 patients from the ME ARDS study, which recruited patients from ICUs at Massachusetts General Hospital from 10 November 2002 to 16 June 2010 and at Beth Israel Deaconess Medical Center from 1 February 2007 to 24 April 2010, for whom both 28-d mortality data and a blood plasma sample had been collected. In addition to baseline demographic data, assessments of sepsis and acute respiratory distress syndrome (ARDS) were performed, and Acute Physiology and Chronic Health Evaluation (APACHE) II scores were measured –. 28-d mortality (in both cohorts) and in-hospital mortality (in the BWH RoCI cohort) were ascertained from patient medical records, and long-term mortality was assessed in the BWH RoCI cohort by linking the cohort to the US Social Security Administration's Death Master File. Written informed consent and institutional review board approval were obtained.

3In the BWH RoCI study, blood samples were drawn and transferred into blood collection tubes within 24 h of study inclusion, and processed within 2 h after venipuncture. In the ME ARDS study, blood samples were drawn and transferred into blood collection tubes within 48 h after the establishment of a diagnosis.

4Plasma used in this study (from both RoCI and ME ARDS) was collected in EDTA-coated blood collection tubes. The plasma samples were stored at −80°C. DNA was isolated from plasma using the DNeasy Blood and Tissue Kit (#69504; Qiagen), according to the manufacture's manual. Samples were thawed on ice and were then mixed briefly by vortex. The 100 µl of plasma was mixed with 100 µl of PBS, followed by brief vortex. The diluted plasma was centrifuged at 700g at 4°C for 5 min, and the supernatant (190 µl) was carefully saved by avoiding touching any pellets and the bottom of the tubes with pipette tips. The obtained supernatant was further centrifuged at 18,000g at 4°C for 15 min, and the resulting supernatant (170 µl) was carefully saved. Contamination of cells, cell debris, or pellets into supernatant might lead to a significant change of the results. The obtained supernatant was processed for DNA isolation. In brief, we incubated plasma samples with lysis buffer (included in the kit) and proteinase K at 56°C for 15 min. At the final step of DNA isolation, DNA was eluted in 200 µl of elution buffer (included in the kit). For the quantitative real-time polymerase chain reaction (qPCR) assay, the DNA solution was further diluted 10 times with nuclease-free deionized, distilled H2O.

5DNA level in diluted samples was measured by SYBR Green dye-based qPCR assay using a PRISM 7300 sequence detection system (Applied Biosystems). The primer sequences were as follows: human NADH dehydrogenase 1 gene (mtDNA): forward 5′-ATACCCATGGCCAACCTCCT-3′, reverse 5′-GGGCCTTTGCGTAGTTGTAT-3′,; human β-globin (nuclear DNA): forward 5′-GTGCACCTGACTCCTGAGGAGA-3′, reverse 5′-CCTTGATACCAACCTGCCCAG-3′; bacterial 16S ribosomal RNA: forward 5′- CGTCAGCTCGTGTTGTGAAA-3′, reverse 5′-GGCAGTCTCCTTGAGTTCC-3′. Plasmid DNA with complementary DNA sequences for human mtDNA was obtained from ORIGENE (SC101172), and plasmid DNA with complementary DNA sequences for human nuclear DNA was obtained from Sino Biological. Concentrations were converted to copy number using the formula; mol/gram×molecules/mol = molecules/gram, via a DNA copy number calculator (http://cels.uri.edu/gsc/cndna.html; University of Rhode Island Genomics and Sequencing Center),. DNA solutions were diluted in 10-fold serial dilutions and used as standards.

6The thermal profile for detecting mtDNA was carried out as follows: an initiation step for 2 min at 50°C is followed by a first denaturation step for 10 min at 95°C and a further step consisting of 40 cycles for 15 s at 95°C and for 1 min at 60°C. A representative standard curve, dissociation curve, and amplification plot are shown in Figure 1.

7All samples were analyzed in duplicate, and a no-template control was included in every analysis. mtDNA levels in all of the plasma analyses were expressed in copies per microliter of plasma based on the following calculation:where c is the concentration of DNA in plasma (copies/microliter plasma); Q is the quantity (copies) of DNA determined by the sequence detector in a PCR; V DNA is the total volume of plasma DNA solution obtained after extraction, typically 200 µl per extraction; V PCR is the volume of plasma DNA solution used for PCR, typically 5 µl of ten times diluted plasma DNA solution; and V ext is the volume of plasma extracted, typically 50–100 µl.

8The specificity of mtDNA primers was examined using bacterial DNA (Text S2 in Supporting Information S1; Figure S1) and by BLAST search (Text S3 and Tables S1 and S2 in Supporting Information S1).

9Plasma level of procalcitonin was measured using the Procalcitonin Human ELISA (Enzyme-Linked Immunosorbent Assay) Kit (ab100630; Abcam).

10mtDNA level was assessed as both a continuous and a binary variable, where indicated. Because mtDNA levels were non-normally distributed, continuous levels from the BWH RoCI cohort were assessed after log10 transformation and dichotomized using visual inspection and iterative methods (using c-statistics) to select the level of mtDNA that maximized the area under the curve for the prediction of 28-d mortality (mtDNA ≥3,200 copies/µl versus <3,200 copies/µl). This level was then replicated in the ME ARDS cohort. Survival was assessed as both a binary and a time-dependent variable. Bivariate analyses were conducted with Fisher's exact tests (for categorical variables) and Wilcoxon rank-sum tests (for continuous variables). Both bivariate and multivariate logistic regression models were used in association analyses. Survival analyses were performed using log-rank and Cox-proportional hazards models.

11To assess the value of mtDNA level in risk prediction, we compared bivariate and multivariate models with and without mtDNA level using the net reclassification improvement approach (using risk categories in increments of 10% from 0% to 100%). In contrast to c-statistics, the net reclassification index (NRI) may have unique value in assessing the ability of a biomarker to improve risk prediction when groups can be split up into risk strata. All analyses were performed using Statistical Analysis Software version 9.2 (SAS Institute). Analyses comparing c-statistics and net reclassification were performed using rocplus. p<0.05 was considered statistically significant.

12Baseline characteristics of ICU patients included in the BWH RoCI (n = 200) and the ME ARDS (n = 243) cohorts are presented in Table 1. While the BWH RoCI cohort was entirely composed of MICU patients, the ME ARDS cohort included patients from different ICU types. Patients in both studies were more likely to be white, to be female, and to carry a diagnosis of sepsis; however, these covariates were even more prevalent among patients in the ME ARDS cohort. A diagnosis of ARDS was more common in ME ARDS patients than in BWH RoCI patients. Age, APACHE II scores, and mtDNA levels were similar in both studies. Of note, only one patient in either study (in ME ARDS) received activated protein C during their stay in the ICU.

13The additional analysis of patient characteristics is shown in Text S4 in Supporting Information S1.

14Of the 200 patients in the BWH RoCI study, 60 (30%) died within 28 d of MICU admission, and 51 (25%) died in the hospital. The median follow-up time for the group that died in the hospital was 5 d (interquartile range 1–54 d). In the ME ARDS study, 40 patients (16%) died within 28 d of ICU admission. In the ME ARDS study, the rate of death among MICU patients (i.e., medical or cardiac ICU patients, n = 31 deaths, 20%) differed from that among non-MICU patients (i.e., surgical, trauma, or neurological ICU patients, n = 9 deaths, 10%, p = 0.03 for the difference). A majority of plasma samples were drawn within 4 d of ICU admission in both the BWH RoCI (n = 186, 93%) and the ME ARDS (n = 194, 90%, data missing on 34 patients) studies.

15In the BWH RoCI cohort, mtDNA level was higher in ICU patients who died within 28 d of MICU admission than in those who did not (median 9,504 copies/µl versus 1,927 copies/µl, p = 2×10−8; Figure 2A), as well as in patients with diagnoses commonly associated with mortality in the ICU such as sepsis (Figure 2B) and ARDS (Figure 2C). For each log10-unit increase in mtDNA level, patients had an increased odds of dying within 28-d of MICU admission (odds ratio [OR] 1.9, 95% CI 1.5–2.4, p = 2×10−7). Patients with an mtDNA level ≥3,200 copies/µl had an increased odds of dying within 28 d of MICU admission (Table 2) and an increased odds of dying in the hospital (OR 8.4, 95% CI 3.7–19.2, p = 4×10−7). Of those who died within 28 d of MICU admission, 80% (n = 48 of 60) had an mtDNA level ≥3,200 copies/µl compared to 38% (n = 53 or 140) of those who did not die. In the BWH RoCI cohort, the odds of dying within 28 d of MICU admission was higher among those with an mtDNA level ≥3,200 copies/µl when the level was obtained <48 h after ICU admission (n = 52, OR 23.0, 95% CI 4.4–120.0, p = 2×10−4) compared to those with an mtDNA level ≥3,200 copies/µl that was obtained ≥48 h after ICU admission (n = 148, OR 4.4, 95% CI 1.9–9.8, p = 4×10−4); however, this difference was of borderline statistical significance (mtDNA by ICU status interaction p = 0.08). The additional analysis of nuclear DNA level and mtDNA level in different patient groups in the BWH RoCI study is shown in Texts S5 and S6 in Supporting Information S1.

16There was no evidence in the BWH RoCI cohort that the association between mtDNA level and MICU mortality was attenuated in models adjusted for clinical covariates, including APACHE II score, sepsis, or ARDS (Table 2). There was no statistical evidence that any of these variables significantly modified the effect of mtDNA level on mortality.

17While an elevated mtDNA level was associated with acute kidney injury (OR 3.2, 95% CI 1.7–6.3, p = 4×10−4), the need for mechanical ventilation (OR 2.7, 95% CI 1.5–4.8, p = 7×10−4), the use of vasopressors (OR 3.6, 95% CI 1.9–6.6, p = 6×10−5), and an underlying diagnosis of cancer (OR 1.8, 95% CI 1.0–3.1, p = 0.05), there was no significant association of elevated mtDNA with physician report of either liver failure (p = 0.65) or heart failure (p = 0.45). The association between an elevated mtDNA level and 28-d mortality remained significant in the BWH RoCI cohort even after adjustment for additional variables including use of vasopressors, need for mechanical ventilation, cancer, acute kidney injury, or admitting physician impression of either heart or liver failure (OR 9.5, 95% CI 3.5–25.6, p = 3×10−5; see Table S3 in Supporting Information S1).

18Binary classification analysis for elevated mtDNA levels and 28-d MICU mortality are shown in Table 3.

19Of the 200 patients from the BWH RoCI cohort, we were able to obtain simultaneous measurements of procalcitonin in 146 (73%) and lactate in 106 (53%) patients. To determine whether the association between mtDNA and 28-d mortality could be explained by measures of procalcitonin or lactate, we tested each of these biomarkers for association with mortality and then assessed whether adjusting for either procalcitonin or lactate attenuates the association between mtDNA and mortality. All three biomarkers were significantly associated with 28-d mortality (Table 4). mtDNA level remained significantly associated with 28-d mortality even after adjusting for either procalcitonin or lactate. In contrast, the associations between procalcitonin and lactate and mortality were attenuated in models additionally adjusting for mtDNA level (Table 4).

20Of the 200 BWH RoCI patients, 96 (48%) remained in the ICU for 7 d and had a repeat measurement of mtDNA. Mortality level was lowest among patients with mtDNA levels <3,200/µl copies at both their initial blood draw and day 7 (0% mortality), highest in those with levels ≥3,200/µl copies at both visits (39% mortality), and intermediate for those with an elevated mtDNA level at only one of the two time points (∼12% mortality; see Table 5).

21Follow-up data were available on the patients in the BWH RoCI study for periods ranging from 1 d to 3.5 y after admission to the ICU. 54% (n = 108) of the cohort died during follow-up. Patients who had an mtDNA level ≥3,200 copies/µl had an increased risk of death (hazard ratio 2.4, 95% CI 1.6–3.6, p<1×10−4; see Figure 3). There was no evidence that the effect of mtDNA level on the risk of death was attenuated in adjusted models (Text S7 in Supporting Information S1). mtDNA level was associated with an increased risk of death even in analyses limited to patients with sepsis (after adjusting for age, gender, race/ethnicity, and APACHE II score, an mtDNA level ≥3,200 copies/µl was associated with an increase in the risk of death, hazard ratio 3.0, 95% CI 1.5–6.3, p = 0.003) or ARDS (after adjusting for age, gender, race/ethnicity, and APACHE II score, an mtDNA level ≥3,200 copies/µl was associated with a 2.3-fold increase in the risk of death, hazard ratio 2.3, 95% CI 1.5–3.5, p = 2×10−4).

22Comparable to our findings in the BWH RoCI cohort, mtDNA levels were higher in ME ARDS patients who died within 28 d of ICU admission (median 7,457 copies/µl) than in patients who did not (median 2,846 copies/µl, p = 5×10−6; Figure 2A), and were higher in patients with sepsis (Figure 2B) or ARDS (Figure 2C) than in patients without these conditions. Also similarly to the BWH RoCI cohort, for each log10-unit increase in mtDNA level, patients had an increased odds of dying within 28-d of ICU admission (OR 1.6, 95% CI 1.2–2.0, p = 9×10−4), those with an mtDNA level ≥3,200 copies/µl had an increased odds of dying within 28 d of ICU admission (Table 2), and there was no evidence that the association between mtDNA level and ICU mortality was attenuated in models adjusted for covariates (Table 2).

23In contrast to the BWH RoCI study, the ME ARDS study included ICU patients from non-medical ICUs (i.e., surgical, trauma, and neurological ICUs). In stratified analyses, the association between an elevated mtDNA level and 28-d mortality was limited to MICU patients. Importantly, there was no evidence for an association between mtDNA level and 28-d mortality in non-MICU patients (mtDNA by ICU status interaction 0.03; see Table 2).

24To further assess the importance of mtDNA as an ICU biomarker, we tested the ability of mtDNA level to improve reclassification of patients in the BWH RoCI cohort. The inclusion of mtDNA level ≥3,200 copies/µl resulted in a 79% improvement in net reclassification when added to a clinical model (including age, gender, race/ethnicity, APACHE II score, and sepsis; see Table 4). The improvement in net reclassification remained significant, albeit attenuated, when mtDNA level was added to clinical models including known ICU biomarkers (e.g., procalcitonin or lactate; see Table 4). While measurement of lactate levels improved the reclassification of patients when added to a clinical model, measurement of procalcitonin did not (Table 4). In contrast, measurement of lactate did not improve reclassification when added to a clinical model that included mtDNA level (NRI 4%, standard error [SE] 14%, p = 0.80). Similarly, in the BWH RoCI cohort there was evidence that an mtDNA level of ≥3,200 copies/µl improved 28-d mortality prediction when added to a model with clinical variables using c-statistics (the c-statistic was 0.76, 95% CI 0.68–0.83, for a model including age, gender, race/ethnicity, APACHE II score, and sepsis, which improved to 0.83, 95% CI 0.77–0.90, with the inclusion of mtDNA level, p<0.001 for the comparison) (Figure 4A).

25Similar to our findings in the BWH RoCI cohort, the addition of an elevated mtDNA level improved net reclassification of ICU patients when added to a clinical model in the ME ARDS cohort (Table 4). The improvement in net reclassification in the ME ARDS cohort was due to the improvement in reclassifying MICU patients (Table 4), as no significant improvement was noted in non-MICU patients (NRI 4%, SE 16%, p = 0.82). Similarly, there was evidence that an mtDNA level of ≥3,200 copies/µl improved 28-d mortality prediction when added to a model with clinical variables using c-statistics (the c-statistic was 0.85, 95% CI 0.79–0.92, for a model including age, gender, and APACHE II score, which improved to 0.86, 95% CI 0.80–0.92, with the inclusion of mtDNA level, p = 0.02 for the comparison) in the ME ARDS cohort (Figure 4B), although improvement in prediction was limited to MICU patients (the c-statistic was 0.87, 95% CI 0.81–0.93, for a model including age, gender, and APACHE II score, which improved to 0.89, 95% CI 0.83–0.94, with the inclusion of mtDNA level, p = 0.008 for the comparison) (Figure 4C), as no significant improvement was noted in non-MICU patients (the c-statistic was 0.82 regardless of the inclusion of mtDNA level, p = 0.97 for the comparison).

26Additional analyses combining both cohorts are presented in Text S8 and Table S4 in Supporting Information S1.

27Elevated levels of circulating cell-free mtDNA are associated with ICU mortality, particularly among MICU patients, from multiple cohorts. We demonstrate that mtDNA can improve risk prediction beyond commonly measured biomarkers, such as lactate, and combined measures of baseline characteristics and underlying physiology (e.g., APACHE II score) among MICU patients. In the BWH RoCI cohort, based primarily on the strength of the association between elevated mtDNA level and 28-day mortality, elevated mtDNA levels predicted an increased risk of death that persisted over up to 3.5 y of follow-up.

28Based on the data we present in this study, it is not possible to determine whether the observed increased circulating cell-free mtDNA levels were increased as the result of the severity of illness alone, or whether mtDNA actively contributes to disease pathogenesis in MICU patients –. Recent observations show that mtDNA contributes to activation of inflammasome, a large multimolecular complex that controls the proteolytic enzyme caspase-1,. Inflammasome activation is exclusively responsible for caspase-1-mediated IL-1β and IL-18 maturation and secretion in immune cells,, and circulating IL-1β and IL-18 levels are known to be increased in critically ill patients with both sepsis and ARDS,,. These findings, in conjunction with the findings presented in here, suggest that circulating mtDNA could serve as a marker of death and, independently, may represent an important pathogenic determinant that contributes to an exaggerated systemic inflammatory response.

29Our data are consistent with prior studies demonstrating that circulating cell-free mtDNA levels are elevated in critically ill subpopulations including adults with community-acquired bacterial meningitis, trauma, or sepsis in the emergency room, and further implicates the mitochondria, both in terms of genotyping (mtDNA haplogroups) and now cell-free mtDNA plasma measurement, as an important factor predicting survival. In addition, our study demonstrates, to our knowledge for the first time, that elevated mtDNA levels may be particularly important in improving risk prediction in the MICU.

30In addition to the evidence that elevated mtDNA levels are associated with sepsis, ARDS, and mortality, an important and unique aspect of our study is that we provide replicated evidence that mtDNA level improves risk prediction in MICU patients. Metrics of association and prediction should not be confused, as only the latter suggest that a finding may have clinical applicability as a test for an individual patient. Although we present statistical evidence that mtDNA level improves our ability to predict death among MICU patients using both c-statistics and reclassification methods, it is important to note that some studies suggest that reclassification methods are more sensitive than c-statistics when comparing the magnitude of predictive improvement,.

31Our study has a number of important limitations. First, although we demonstrate that an elevated mtDNA measurement is associated with ICU mortality among MICU patients from multiple cohorts, there was no evidence for a similar association among non-medical (predominantly surgical) ICU patients. Although we cannot definitively determine the reasons behind these differences within the context of this study, we can't rule out that subtle differences in the timing of sample collection could have influenced the associations within these two groups. In addition, these findings suggest that mtDNA could be released from cells or tissue as the result of surgery, and in this situation an elevated mtDNA level may be unlikely to be associated with an increased risk of mortality. Of note, although some studies demonstrate that patients who meet criteria for systemic inflammatory response syndrome (SIRS) have worse outcomes, this does not appear to hold among surgical ICU patients, where the presence of systemic inflammatory response syndrome is very common. Second, some of the comparisons between biomarkers in the BWH RoCI study are limited because of missing data (e.g., lactate measurements). Third, although our assessments of repeated measures of mtDNA suggest that the trajectory of this measurement may be of clinical importance, we urge caution in interpreting these results, as our findings also demonstrate that the probability of being alive to have had mtDNA remeasured was also dependent on the initial mtDNA level. These findings support the idea that future studies of mtDNA in MICU patients could benefit from repeated measurements, including those prior to ICU presentation. Fourth, although fungemia was rare in both the BWH RoCI and ME ARDS cohorts (Text S4 in Supporting Information S1), it is worth noting that fungi have mtDNA that may cross-react with human mtDNA primers,. Finally, it is important to note that our analyses reflect that the mtDNA level adds to, rather than replaces, information gleaned from commonly measured physiologic parameters and biomarkers.

32In conclusion, we show that mtDNA level in plasma is associated with mortality in MICU patients. Elevated mtDNA level adds to mortality risk prediction in the ICU, and represents an advance over currently used ICU biomarkers and the APACHE score. Given the relative ease and rapidity (the entire process currently takes <6 h) of mtDNA measurement, circulating cell-free mtDNA could be a valuable addition to assessment of MICU patients, and points the way to the possibility of new diagnostic and/or therapeutic approaches for patients with critical illness.