Excess circulating angiopoietin-2 is a strong predictor of mortality in critically ill medical patients

From the ICU at the Internal Medicine Department at Hannover Medical School, Germany, a tertiary care university hospital, 43 patients were enrolled at the time of ICU admission and studied prospectively. Patients were subdivided into the following groups: severe sepsis (n = 12), septic shock (n = 17) and critically ill patients (n = 14) with no evidence or suspicion of bacterial infection or sepsis (SCCM/ESICM/ACCP/ATS/SIS definitions [28]). Enrollment was performed in a consecutive fashion after obtaining written informed consents from the patients or their legal representatives. If the patient was recovering and able to communicate, he/she was informed of the study purpose and consent was required to further maintain status as study participant. The study was performed in accordance with the declaration of Helsinki and approved by the institutional review board. There were no co-morbidities that led to exclusion, except for age younger than 18 years or older than 75 years, being pregnant and having a malignant neoplasm.

Subjects were ventilated in accordance with the ARDSNet-derived protocol [29]. In 29 patients, invasive haemodynamic monitoring was performed by the Pulse contour Continous Cardiac Output (PiCCO) system (Pulsion Medical Systems, Munich, Germany) in addition to standard techniques. This device enables invasive on-line monitoring of several haemodynamic parameters, such as mean arterial pressure (MAP), heart rate (HR), cardiac index (CI), systemic vascular resistance index (SVRI), intrathoracic blood volume index (ITBVI) and extravascular lung water index (EVLWI), based on a transpulmonary thermodilution technique [30, 31]. All relevant laboratory and medical data, including APACHE II [32] and SOFA scores [33], were obtained at the time of enrollment. Detailed patients' characteristics, including demographic, clinical and laboratory parameters, are shown in Table 1.

Twenty-nine age- and gender-matched healthy volunteers from the Hannover Medical School staff served as controls (16 males, 13 females; age 58 (25 to 73 years)).

Serum samples for quantification of Ang-1, Ang-2 and VEGF were obtained at the time of enrollment, immediately placed on ice, centrifuged and stored at -80°C. All measurements were performed in a blinded fashion by the same investigator.

Ang-1 and Ang-2 were measured by in-house Immuno Radiometric Sandwich Assay (IRMA) and ELISA, respectively as previously described [27, 34]. Polyclonal, anti-human Ang-1 affinity purified goat immunoglobulin (Ig) G and a monoclonal anti-human Ang-1 mouse antibody were obtained from R&D Systems (R&D, Oxford, UK). Recombinant human Ang-1 was purchased from Sigma-Aldrich (Sigma-Aldrich, Munich, Germany). Recombinant human Ang-2 monoclonal Ang-2 antibody and anti-Ang-2 antibody were purchased from R&D Systems (R&D, Oxford, UK).

Serum VEGF was measured using a sandwich ELISA kit according to the manufacturer's instructions (R&D Systems, Minneapolis, USA). This assay measures biologically active VEGF₁₂₁ and VEGF₁₆₅.

Differences between patients and healthy controls were evaluated using a non-parametric Kruskal-Wallis test. The Mann-Whitney rank sum test was used for comparison between individual groups. Correlations between variables were assessed by the Spearman rank correlation coefficient. Pearson's correlation coefficient and linear regression analysis was performed after logarithmic transformation of Ang-2 values (logAng-2). The primary outcome studied was 30-day survival and was calculated from the day of ICU admission to death. Patients who survived the follow-up period were censored at day 30. Parameters independently associated with survival were identified by univariate and multivariate Cox proportional hazards models.

Variables found to be statistically significant at a 10% level in the univariate analysis were included in the multivariate model using backward elimination. Different models were established, incorporating either Ang-2, logAng-2 or the Ang-2/Ang-1 ratio, respectively. Two-sided p-values < 0.05 were considered statistically significant for all statistical procedures used. The distribution of the time-to-event variables were estimated using the Kaplan-Meier method with log-rank testing. Receiver operator characteristics (ROC) procedures were used to identify optimal cut-off values. Data are displayed as median and range (minimum to maximum) unless otherwise stated. All statistical analyses were performed with the SPSS package (SPSS Inc., Chicago, IL, USA) and the GraphPad Prism software (GraphPad Prism Software Inc. San Diego, California, USA).

Ang-1 concentrations in critically ill non-septic patients (0.8 ng/ml, 0.5 to 11.7 ng/ml), patients with severe sepsis (0.5 ng/ml, 0.3 to 18.8 ng/ml) and patients with septic shock (0.9 ng/ml, 0.3 to 5.5 ng/ml were markedly decreased compared with healthy controls (56.4 ng/ml, 34.5 to 71.3 ng/ml, p < 0.0001) (Figure 1a). Ang-1 concentrations were no different between severe sepsis, septic shock and non-septic patients.

In contrast, median serum Ang-2 concentrations were consistently increased in critically ill non-septic patients (2.8 ng/ml, 1.0 to 9.0 ng/ml), in patients with severe sepsis (16.45 ng/ml, 2.7 to 39.7 ng/ml) and patients with septic shock (28.1 ng/ml, 3.7 to 72.6 ng/ml), compared with healthy controls (0.9 ng/ml, 0.3 to 2.6 ng/ml; all p < 0.0001 versus controls) (Figure 1b). Ang-2 was higher in patients with sepsis compared with non-septic patients (both p < 0.0001). Ang-2 concentrations were not different between patients with severe sepsis and septic shock (p = 0.12). Ang-1 and Ang-2 concentrations were neither linked to gender (Mann-Whitney test: p = 0.42 and p = 0.51) nor age (Spearman correlation: p = 0.83 and p = 0.24).

VEGF concentrations were markedly lower in critically ill non-septic patients (43.5 pg/ml, 4.1 to 200.0 pg/ml), patients with severe sepsis (112.7 pg/ml, 34.9 to 569.1 pg/ml) and patients with septic shock (70.5 pg/ml, 3.7 to 179.9 pg/ml compared with healthy controls (515.5 pg/ml, 280.6 to 1294.0 pg/ml, all p < 0.0001) (Figure 1c). VEGF concentrations were no different between patients with severe sepsis, patients with septic shock and non-septic controls. VEGF concentrations were not linked to gender (p = 0.67) and did not correlate with age (p = 0.33).

Linear regression analysis detected a strong association of logAng-2 concentration with APACHE II scores (r² = 0.28, p = 0.0003) and SOFA scores (r² = 0.62, p < 0.0001) (Figures 2a,b; n = 43). Hypoxia has been shown to induce the release of Ang-2 from endothelial cells in preclinical models [35, 36]. Of note, a strong correlation between Ang-2 concentrations and lactate levels as a surrogate marker for tissue hypoperfusion and microcirculatory tissue hypoxia was detected (r² = 0.25, p = 0.0007). Neither Ang-1 nor VEGF correlated with APACHE II scores, SOFA scores or C-reactive protein (CRP) levels.

Pre-clinical models have impressively demonstrated that the intact Ang-1/Tie2 signalling protects from ARDS in experimental sepsis [36‐38]. We therefore examined the association between several parameters of haemodynamic and pulmonary function with circulating Ang-1, Ang-2 and VEGF levels. Of those, only Ang-2 showed an inverse correlation with partial pressure of oxygen in arterial blood (PaO₂)/fraction of inspired oxygen (FiO₂) (r² = -0.31; p = 0.046), and PaO₂ (r² = -0.35; p = 0.023) as surrogate markers for ventilator support and pulmonary function. No association was seen for peak airway pressure (p = 0.6) or positive end expiratory pressure levels (p = 0.45). In addition to routine invasive haemodynamic monitoring (n = 43), 29 ventilated patients without atrial fibrillation qualified for detailed haemodynamic assessment by transpulmonary thermodilution technique (PiCCO system). Surprisingly, none of the measured angiogenic factors correlated with any of the haemodynamic parameters (MAP, CI, EVLWI, ITBVI, SVRI, vasopressor dose or central venous pressure; data not shown). The same results were obtained for invasive routine monitoring in all 43 patients (data not shown).

To determine the relation between Ang-2 levels at admission and mortality, we initially performed univariate Cox proportional hazards analyses. In our whole cohort of critically ill medical patients, age, gender or the presence of sepsis did not show prognostic significance for survival (Table 2). The same was true for MAP, HR, CVP, urine output, noradrenaline dose, FiO₂, PaO₂/FiO₂, thrombocytes, bilirubin, CRP and VEGF (Table 2). Among the tested variables, lactate (p = 0.006), APACHE II score (p = 0.013), SOFA score (p = 0.038) and the amount of circulating Ang-2 (p = 0.001) displayed prognostic significance (Table 2).

Subsequently, the following variables were found to be statistically significant at a 10% level in the univariate analysis and subjected to multivariate Cox regression analysis: lactate, APACHE II score, SOFA score and circulating Ang-2 (Table 2). Except for Ang-2 (p = 0.002), all other variables did not remain significant in the multivariate setting (lactate (p = 0.111), APACHE II score (p = 0.154), SOFA score (p = 0.167)). The same results were obtained when either logAng-2 (p = 0.003) or the Ang-2/Ang-1 ratio (p = 0.036) were tested instead of Ang-2 (Table 2). Thus, circulating Ang-2 was identified as a strong, independent prognostic factor for 30-day survival in our cohort of critically ill medical patients. Given the context-dependent synergistic effects of Ang-2 and VEGF, we analysed various ratios incorporating Ang-1, Ang-2 and VEGF (data not shown). Except for the Ang-2/Ang-1 ratio, none of these models reached statistical significance (Table 2).

Ang-2 yielded an area under the ROC curve (AUC) value of 0.79 (standard error of the mean (SEM) = 0.07; 95% confidence interval = 0.65 to 0.93; p = 0.001). For comparison, the APACHE II score yielded an AUC value of 0.75 (SEM = 0.08; 95% confidence interval = 0.59 to 0.91; p = 0.005). A median circulating Ang-2 of more than 11.08 ng/ml predicted death with a specificity of 74% (95% confidence interval = 57 to 86) and a sensitivity of 67% (95% confidence interval = 54 to 77). The odds ratio for 30-day mortality was 5.6 (95% confidence interval = 1.5 to 20.5), positive and negative predictive values were 76% (95% confidence interval = 61 to 88) and 64% (95% confidence interval = 49 to 75), respectively.

Figure 3 illustrates the Kaplan-Meier curves of 30-day survival stratified to Ang-2 (less versus higher than median (11.08 ng/ml)). Log rank test confirmed statistical significance for Ang-2 (p = 0.009). Accordingly, the hazard for Ang-2 (> median) in our cohort was three-fold in the high Ang-2 (> 11.08 ng/ml) group compared with the low Ang-2 group (≤ 11.08 ng/ml). Of note, the 30-day survival of patients among the low Ang-2 group was 57%, while it was 20% in the group of patients with high Ang-2 levels.