BIOMARKERS FOR PREDICTION OF DEVELOPMENT OF HYPOXEMIA DUE TO ACUTE LUNG INJURY

Abstract

The present invention relates to a set of biomarkers, which can be used to determine the risk of developing hypoxemia for subjects undergoing open heart surgery with the use of cardiopulmonary bypass (CPB).

Claims

1. A method for comprising: providing a biological sample from a subject, said subject having had open cardiac surgery with the use of cardiopulmonary bypass (CPB), wherein said sample has been provided from said subject between start of surgery and 72 hours after detachment from the CPB circuit; determining the level of at least one biomarker selected from the group consisting of carnitine, isobutyrylglycine, N-Acetyl-Glucosamine, arachidonic and eicosapentanoic acid, isobutyric acid and citrate; and/or determining the ratio between levels of biomarkers of at least one ratio selected from the group of ratios selected from citrate/phenylalanine, polyunsaturated fatty acids (PUFA)/phenylalanine, alanine/phenylalanine, citrate/trimethylamine-N-oxide (TMAO), carnitine/TMAO, isobutyric acid/phenylalanine, and isobutyrylglycine/arginine; comparing said levels of one or more biomarkers or ratios to one or more reference levels; determining that said subject is at risk of developing hypoxemia if said level of one or more biomarkers are above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said one or more levels are equal to or below the one or more reference levels; and/or determining that said subject is at risk of developing hypoxemia if said one or more ratios between levels of biomarkers are above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said level of one or more ratios between levels of biomarkers are equal to or below the one or more reference levels; and for a subject considered at risk of developing hypoxemia, initiating a preventive hypoxemia treatment.

2. The method according to claim 1, wherein said sample has been provided from said subject within 48 hours after detachment from the CPB circuit.

3. The method according to claim 1, wherein said sample has been provided from said subject before detachment from the CPB circuit until 2 hours after detachment from the CPB circuit.

4. The method according to claim 1, wherein at least the level of at least two biomarkers and/or at least one ratio between biomarkers are determined.

5. The method according to claim 1, wherein at least the levels of carnitine and citrate are determined.

6. The method according to claim 1, wherein the sample is a blood sample, such as whole blood, serum and/or plasma.

7. The method according to claim 1, wherein said level of biomarkers are determined by a method selected from the group consisting of mass spectrometry (GC-MS, LC-MS), HPLC, Raman, NIR, and NMR spectroscopy.

8. The method according to claim 1, wherein hypoxemia is a PaO.sub.2/FiO.sub.2 below 40 kPa (300 mmHg).

9. The method according to claim 1, wherein said open heart surgery is selected from the group consisting of coronary artery bypass grafting and valve surgery.

10. The method according to claim 1, comprising applying NMR spectroscopy on said biological samples so as to arrive at a resulting spectrum covering a spectral range representative of said biomarkers, and applying a computer-based algorithm in response, so as to determine an output indicative of said subject being at risk of developing hypoxemia.

11. The method according to claim 1, wherein said biological sample has been provided from said subject between a time during surgery, such as from weaning from the CPB circuit until 8 hours after weaning from the CPB circuit.

12. The method according to claim 1, the method further comprising, for said subject considered at risk of early developing hypoxemia, providing the preventive hypoxemia treatment to said subject before traditional symptoms of hypoxemia are observed.

13. The method according to claim 1, wherein the preventive hypoxemia treatment is selected from the group consisting of anti-oxidant treatment, anti-inflammatory treatment, glucose administration, insulin treatment, oxygen supplementation, respiratory support and fluid treatment.

14. (canceled)

15. A method comprising: completing the method according to claim 1; and for said subject considered at risk of developing hypoxemia, evaluating whether the initiated preventive hypoxemia treatment results in said subject not developing hypoxemia within one week after weaning from the CPB circuit.

16. The method of claim 1, wherein the preventive hypoxemia treatment results in said subject not experiencing hypoxemia within one week after weaning from the CPB circuit.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0039] FIG. 1 shows serum metabolic fingerprints: Representative 600 MHz one-dimensional Carr-Purcell-Meiboom-Gill (CPMG) 1H-nuclear magnetic resonance (NMR) spectral profiles of a patient presenting no sign of pulmonary dysfunction nor hypoxemia (partial pressure of oxygen in arterial blood, PaO.sub.2=10.7 kPa or 80.3 mmHg) (black) and a patient diagnosed with acute lung injury defined by hypoxemia (PaO.sub.2=4.9 kPa or 36.8 mmHg) (grey) 72 hours after weaning from cardiopulmonary bypass (CPB). Spectrum difference (unaffectedhypoxemia) reveals metabolite fluctuations between patients. Higher levels of signals from lipoproteins, fatty acids and lower levels of some smaller metabolites are observed in the hypoxemic compared to unaffected patient.

[0040] FIG. 2 shows the Receiver Operating Characteristic (ROC) curve of biomarkers of early progression into acute lung injury after cardiac surgery. Biomarker levels (either concentrations of single metabolites or ratios between two metabolites) were measured on the first morning after weaning from CPB.

[0041] FIG. 3 shows differences in metabolite levels between patient groups expressed as fold-changes. Metabolic fold changes were calculated as mean percent change between mild vs. unaffected patients (open rectangles) and severe vs. unaffected patients (filled rectangles) ((AU)/U)*100), where A=affected (mild or severe) and U=unaffected). Metabolites labelled with an asterisk were found to significantly vary between groups by using one-way ANOVA or its nonparametric analogue Kruskal-Wallis test (*0.09>p0.01, **10.sup.2>p10.sup.3, ***10.sup.4>p10.sup.5, ****p<10.sup.5).

[0042] FIG. 4 shows the distribution of metabolite concentrations in the groups of unaffected (U), mildly (M) and severely (S) affected patients by hypoxemia measured on the first morning after weaning from CPB.

[0043] FIG. 5 shows the distribution of the most significant metabolic ratios found differentiating the three patient groups (unaffected (U), mildly (M) and severely (S)).

[0044] FIG. 6

[0045] Workflow of the computer-aided method described in the invention. The left box represents standard processing techniques applied for NMR and metabonomics studies that are imbedded in packages connected to the NMR system. The right box represents the present workflow and it contains a script/test that performs: 1) automatic preprocessing of spectral data (generalized log transformation, normalization, scaling), 2) feature elimination of spectral regions not correlating to PaO.sub.2 values used in diagnosis three days postoperatively, 3) multivariate statistical data analysis with the use of the machine learning technique PLS regression to predict the oxygenation values (PaO.sub.2 or PaO.sub.2/FiO.sub.2 in kPa), 4) calibration and validation of PLS regression results, 5) visualization of the results including the predicted details of a patient based on the models used as references, 6) multivariate statistical data analysis with the use of the machine learning technique PLS-DA to screen for disease, to assesses in diagnosis (hypoxemia or none), and to determine the risk of ARDS, 7) calibration and validation of PLS-DA results, 8) visualization of the results including the predicted details of a patient based on the models used as reference. Because the script/program contains 500 spectra as references run on serum samples collected at different time points (after thoracotomy (skin incision) but before coupling to the CPB, just after weaning from CPB (0 hours), 2 hours, 4 hours and 8 hours post-CPB)), it can also be applicable to monitor the effect of different treatment and therapeutic interventions and to determine if a patient condition is in flare or remission after a treatment has been initiated.

[0046] FIG. 7

[0047] Representative example of regions of interest found correlating to the degree of hypoxemia determined three days postoperatively by the standard clinical test. The numbered regions corresponds to 1) Histidine, 2) Phenylalanine, 3) PUFA, 4) Carnitine, 5) TMAO, 6) Citrate, 7) N-AcGlc, and 8) Alanine.

[0048] FIG. 8 shows the relative concentration of metabolites represented as histogram plots with the mean and standard deviation indicated by the 95% confidence intervals.

[0049] A) Trigonelline; B) Hypoxanthine; C) Dopamine; D) Dihydroxyphenylalanine (Dopa); and E) Inosine. CPB=after sternotomy but before CPB; +CPB=after weaning from CPB (0 hours); 2 h=2 hours after weaning from CPB; 4 h=4 hours after weaning from CPB; 8 h=8 hours after weaning from CPB; 20 h=20 hours after weaning from CPB. None=patients not presenting hypoxemia (unaffected; Mild=patients affected mildly by hypoxemia; and Severe=patients affected severely by hypoxemia.

[0050] The present invention will now be described in more detail in the following sections.

DETAILED DESCRIPTION OF THE INVENTION

Definitions

[0051] Prior to discussing the present invention in detail, the following terms and conventions will first be defined:

Hypoxemia

[0052] Hypoxemia is a dangerous condition where oxygen content in arterial blood drops below normal levels. A decrease in oxygen delivery will affect lungs, brain, liver, and other organs, which in the most severe condition can lead to organ failure and even death. The most severe hypoxemia is seen in patients suffering acute respiratory distress syndrome (ARDS), which has a mortality rate of 20-40%. In the present context, hypoxemia is defined as the ratio between the partial pressure of arterial oxygen (PaO.sub.2) and the fraction of inspired oxygen (FiO.sub.2). Patients presenting PaO.sub.2/FiO.sub.2 values below 40 kPa (300 mmHg) are considered to have acute lung injury. The lower the value, the more severe the condition.

Acute Lung Injury

[0053] Acute lung injury (ALI) is a heterogeneous disease, which is defined by the acute onset of hypoxemic respiratory failure, with pulmonary infiltrates on chest x-ray due to non-cardiogenic pulmonary edema. The PaO.sub.2/FiO.sub.2 values in patients with ALI are below 40 kPa (300 mmHg); where PaO.sub.2 is the partial pressure of oxygen in arterial blood, and FiO.sub.2 is the fraction of inspired O.sub.2.

Cardiopulmonary Bypass (CPB)

[0054] Cardiopulmonary bypass (CPB) is a technique that temporarily takes over the function of the heart and lungs during cardiac surgery, maintaining the circulation of blood and the gas exchange. The CPB itself is often referred to as a heart-lung machine or the pump.

Reference Level

[0055] In the context of the present invention, the term reference level relates to a standard in relation to a quantity, which other values or characteristics can be compared to.

[0056] In one embodiment of the present invention, it is possible to determine a reference level by investigating the abundance of one or more of the biomarkers according to the invention in (blood) samples from subjects did not develop hypoxemia under the same surgical condition. By applying different statistical means, such as multivariate analysis, one or more reference levels can be calculated.

[0057] Based on these results a cut-off may be obtained that shows the relationship between the level(s) detected and patients at risk. The cut-off can thereby be used to determine the amount of the one or more biomarkers, which corresponds to for instance an increased risk of developing hypoxemia.

Risk Assessment

[0058] The present inventors have successfully developed a new method to predict the risk for developing hypoxemia for a patient several days in advance. The results presented in the examples show that the described biomarkers (alone or in combination) appear to be efficient markers for determining whether a patient has an increased risk of developing postoperative hypoxemia.

[0059] To determine whether a patient has an increased risk of developing hypoxemia due to related pulmonary complications such as acute lung injury, a cut-off must be established. This cut-off may be established by the laboratory, the physician or on a case-by-case basis for each patient.

[0060] The cut-off level could be established using a number of methods, including: multivariate statistical tests (such as partial least squares discriminant analysis (PLS-DA), random forest, support vector machine, etc.), percentiles, mean plus or minus standard deviation(s); median value; fold changes.

[0061] The multivariate discriminant analysis and other risk assessments can be performed on the free or commercially available computer statistical packages (SAS, SPSS, Matlab, R, etc.) or other statistical software packages or screening software known to those skilled in the art.

[0062] As obvious to one skilled in the art, in any of the embodiments discussed above, changing the risk cut-off level could change the results of the discriminant analysis for each patient.

[0063] Statistics enables evaluation of the significance of each metabolite level. Commonly used statistical tests applied to a data set include t-test, f-test or even more advanced tests and methods of comparing data. Using such a test or method enables the determination of whether two or more samples are significantly different or not.

[0064] The significance may be determined by the standard statistical methodology known by the person skilled in the art.

[0065] The chosen reference level may be changed depending on the mammal/subject for which the test is applied.

[0066] Preferably the subject according to the invention is a human subject.

[0067] The chosen reference level may be changed if desired to give a different specificity or sensitivity as known in the art. Sensitivity and specificity are widely used statistics to describe and quantify how good and reliable a biomarker or a diagnostic test is. Sensitivity evaluates how good a biomarker or a diagnostic test is at detecting a disease, while specificity estimates how likely an individual (i.e. control, patient without disease) can be correctly identified as not sick. Several terms are used along with the description of sensitivity and specificity; true positives (TP), true negatives (TN), false negatives (FN) and false positives (FP). If a disease is proven to be present in a sick patient, the result of the diagnostic test is considered to be TP. If a disease is not present in an individual (i.e. control, patient without disease), and the diagnostic test confirms the absence of disease, the test result is TN. If the diagnostic test indicates the presence of disease in an individual with no such disease, the test result is FP. Finally, if the diagnostic test indicates no presence of disease in a patient with disease, the test result is FN.

Sensitivity

[0068]
Sensitivity=TP/(TP+FN)=number of true positive assessments/number of all samples from patients with disease.

[0069] As used herein the sensitivity refers to the measures of the proportion of actual positives which are correctly identified as suchin analogy with a diagnostic test, i.e. the percentage of people having PaO.sub.2 below normal who are identified as having PaO.sub.2 below normal.

Specificity

[0070]
Specificity=TN/(TN+FP)=number of true negative assessments/number of all samples from controls.

[0071] As used herein the specificity refers to measures of the proportion of negatives which are correctly identifiedi.e. the percentage of mammals or people having PaO.sub.2 at a normal level who are identified as having PaO.sub.2 at a normal level. The relationship between both sensitivity and specificity can be assessed by the ROC curve. This graphical representation helps to decide the optimal model through determining the best threshold- or cut-off for a diagnostic test or a biomarker candidate.

[0072] As will be generally understood by those skilled in the art, methods for screening are processes of decision-making and therefore the chosen specificity and sensitivity depend on what is considered to be the optimal outcome by a given institution/clinical personnel.

[0073] It would be obvious for a person skilled in the art that it may be advantageous to select a higher sensitivity at the expense of lower specificity in most cases, to identify as many patients with disease as possible.

[0074] In a preferred embodiment, the invention relates to a method with a high specificity, such as at least 70%, such as at least 80%, such as at least 90%, such as at least 95%, such as 100%.

[0075] In another preferred embodiment, the invention relates to a method with a high sensitivity, such as at least 80%, such as at least 90%, such as 100%.

[0076] Another part of the invention relates to a method wherein the ratio between at least two markers is used to predict whether or not a subject is at risk of developing hypoxemia.

Method for Determining the Risk of Developing Hypoxemia

[0077] As described above, the present inventors have identified a novel set of biomarkers, which can be used to determine the risk of developing hypoxemia for a subject having had open cardiac surgery with the use of cardiopulmonary bypass (CPB). Thus, the first aspect of the invention relates to a method for determining the risk of developing hypoxemia, the method comprising [0078] providing a biological sample (said sample has been provided) from a subject, said subject having had open cardiac surgery with the use of cardiopulmonary bypass (CPB), wherein said sample has been provided from said subject between start of surgery (defined as skin incision) and 72 hours after detachment from the CPB circuit; [0079] determining the level of at least one biomarker selected from the group consisting of carnitine, citrate, isobutyrylglycine, N-Acetyl-Glucosamine, arachidonic acid, and isobutyric acid; [0080] and/or [0081] determining the ratio between levels of markers of at least one ratio selected from the group of ratios selected from citrate/phenylalanine, PUFA/phenylalanine, alanine/phenylalanine, citrate/Trimethylamine-N-oxide (TMAO), carnitine/TMAO, isobutyric acid/phenylalanine, and isobutyrylglycine/arginine; [0082] comparing said levels of one or more biomarkers or ratios to one or more reference levels; [0083] determining that said subject is at risk of developing hypoxemia if said level of one or more biomarkers are above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said one or more levels are equal to or below the one or more reference levels; [0084] and/or [0085] determining that said subject is at risk of developing hypoxemia if said one or more ratios between levels of biomarkers are above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said level of one or more ratios between levels of biomarkers are equal to or below the one or more reference levels.

[0086] A second aspect of the invention relates to a method for determining the risk of developing hypoxemia, the method comprising [0087] providing a biological sample, such as serum, plasma, or full blod (said sample has been provided) from a subject, said subject having had open cardiac surgery with the use of cardiopulmonary bypass (CPB), wherein said sample has been provided from said subject between the time of weaning from the CPB (0 hours) (or start of surgery (defined as time of skin incision)) and 20 hours after detachment/weaning from the CPB circuit; [0088] determining the level of at least one biomarker selected from the group consisting of Trigonelline; Hypoxanthine; Dopamine; Dihydroxyphenylalanine (Dopa); and Inosine; [0089] comparing said levels of one or more biomarkers to one or more reference (metabolic) levels; [0090] determining that said subject is at risk of developing hypoxemia if said level of at least one of the metabolites Trigonelline, Hypoxanthine, Dopamine, and Inosine are below the one or more reference levels, and/or determining that said subject is at risk of developing hypoxemia, if said level of Dihydroxyphenylalanine (Dopa) is above the one or more reference levels; [0091] or [0092] determining that said subject is not risk of developing hypoxemia if said level of at least one of the metabolites Trigonelline, Hypoxanthine, Dopamine, and Inosine are above or equal to the one or more reference levels, and/or determining that said subject is not at risk of developing hypoxemia, if said level of Dihydroxyphenylalanine (Dopa) is below or equal to the one or more reference levels.

[0093] For said second aspect, the levels of the biomarkers are preferably determined within 20 hours from detachment from the CPB circuit, more preferably within 4 hours, and even more preferably within few minutes after detachment/weaning from the CPB circuit, such as with in 10 minutes, such as within 5 minutes or such as within 2 minutes.

[0094] Because hypoxemia normally develops within 3-7 days following an initial insult such as CPB, an early prediction of the condition is desirable. To achieve an early prediction, it may be advantageous to obtain/provide (or have provided) the (blood) sample at several time points soon after surgery, within an interval of 2 days postoperatively. Thus, in an embodiment, said sample is provided (or has been provided) from said subject within 48 hours after detachment from the CPB circuit, such as within 30 hours, such as within 20 hours, such as within 16 hours, such as within 10 hours, such as within 8 hours, such as within 5 hours, such as within 4 hours, such as within 2 hours, such as within 1 hour, or just after/weaning from the CPB circuit. In another embodiment, said sample is provided (or has been provided) from said subject (e.g. 2 hours or 90 minutes, or 60 minutes) before detachment from the CPB circuit until 2 hours after detachment from the CPB circuit.

[0095] In example 2, samples were obtained first morning (16 hours) after weaning from the CPB. In example 5, samples were obtained at different time points after detachment from the CPB (t=0 until t=8 hours).

[0096] In example 6 it is shown that samples were obtained at different time points after sternotomy but before CPB and after weaning from the CPB (t=CPB, t=+CPB (0 hours) until t=20 hours). All examples provide strong predictive values in relation to the risk of developing hypoxemia.

[0097] It is of course to be understood that the levels in the indicated ratio could also be mathematically reversed, resulting in the indicated risk be the opposite. This is considered to also form part of the present invention.

[0098] The predicted hypoxemia may be predicted to occur within different time periods. Thus, in another embodiment, hypoxemia is predicted to occur within a week from initiation of cardiac surgery, such as within 72 hours, such as within 60 hours, or such as within 48 hours, preferably within 72 hours.

[0099] In an embodiment said sample has been provided from said subject after sternotomy but before detachment from the CPB circuit until 2 hours after weaning from the CPB circuit

[0100] When the sample is obtained, the subject will normally have normal oxygen levels (PaO.sub.2/FiO.sub.2>40 kPa (>300 mmHg)) and thus not suffer from hypoxemia. Thus, in a further embodiment said subject has normal blood oxygenation levels at the time the sample was obtained, such as having a PaO.sub.2/FiO.sub.2 equal to or above 40 kPa (300 mmHg). To increase the predictive value of the method, preferably more than one biomarker is determined by the method of the invention. Thus, in an embodiment, the levels of at least two biomarkers are determined. In another embodiment, the levels of at least three biomarkers, such as at least four biomarkers, such as at least five biomarkers, such as at least six biomarkers are determined.

[0101] In another embodiment, at least the ratios between citrate/phenylalanine, PUFA/phenylalanine and alanine/phenylalanine are determined.

[0102] In another embodiment, at least the levels of carnitine and citrate are determined. In yet another embodiment at least the levels of carnitine and citrate, and the citrate/U (unidentified metabolite occurring at 5.10-5.08 ppm on our NMR spectrum) ratio are determined. In yet another embodiment, at least the levels of carnitine, citrate, isobutyrylglycine, N-acetyl-glucosamine, arachidonic acid, and isobutyric acid are determined.

[0103] The type of sample material may vary. Thus, in an embodiment, the sample is a blood sample, such as whole blood, serum and/or plasma. The blood sample may have been provided from different locations from the subject. In an embodiment, the blood sample has been obtained from the left atrium (LA), the pulmonary artery (PA), venous blood and/or arterial blood.

[0104] The determination of the presence and levels of these metabolites in a sample can be performed by many different methods. Thus, in an embodiment, said level of biomarkers are determined by a method selected from the group consisting of mass spectrometry (GC-MS, LC-MS), HPLC, Raman, NIR and NMR spectroscopy. The skilled person may identify other methods, which may be used.

[0105] The severity of hypoxemia may vary. In an embodiment, said subject is determined at risk of having mild hypoxemia. In the present context, patients with mild hypoxemia have a PaO.sub.2/FiO.sub.2 in the range between 4026.6 kPa (300200 mmHg).

[0106] In another embodiment, said subject is determined to be at risk of having severe hypoxemia. In the present context, severe hypoxemia is defined as patients having a PaO.sub.2/FiO.sub.2<26.6 kPa (200 mmHg).

[0107] The type of open cardiac surgery may vary. Thus, in an embodiment, said open heart surgery is selected from the group consisting of coronary artery bypass grafting and valve surgery.

[0108] The reference levels may be determined in different ways. In an embodiment, said one or more reference levels are determined from levels of said biomarkers in blood samples from subjects having had open heart surgery with the use of CPB, who did not develop hypoxemia postoperatively as diagnosed by standard diagnostic tests at least 3 days postoperatively.

[0109] The predictive value of the method may be further increased by including even further biomarkers in the assay. Table 4 lists some of the biomarkers identified as also having a predictive value, albeit with a lower predictive power. Thus, in an embodiment said method further comprises determining the level of at least one marker selected from the group listed in Table 4.

[0110] In that case determination according to the method of the invention would be: [0111] determining that said subject is at risk of developing hypoxemia if said one or more levels of biomarkers are below/above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said one or more levels of biomarkers are equal to, above/below the one or more reference levels.

[0112] From table 4, the skilled person will be able to adapt the method accordingly.

Initiation of Treatment Based on Prediction

[0113] If a subject is considered at risk of developing hypoxemia it may be advantageous to initiate a preventive treatment to avoid hypoxemia occurring. Thus, another embodiment of the invention, the method further comprises, for a subject considered at risk of early developing hypoxemia, [0114] providing to said subject an earlier treatment protocol for the prevention of hypoxemia, such as anti-oxidant treatment, anti-inflammatory treatment, glucose administration, insulin treatment, oxygen supplementation, respiratory support and/or fluid treatment; [0115] and/or; [0116] having said subject under observation for the development of hypoxemia.
Method for Evaluating the Efficacy of Treatment to Prevent Hypoxemia By having identified biomarkers allowing early prediction of hypoxemia in a subject it also becomes possible to test/evaluate possible treatment protocols which may prevent the development of hypoxemia. Thus, another aspect of the invention relates to a method for evaluating the efficacy of treatment to prevent hypoxemia, the method comprising [0117] completing the method according to the first aspect of the invention; and [0118] for a subject considered at risk of developing hypoxemia, evaluating whether a treatment protocol for the prevention of hypoxemia, results in that said subject not developing hypoxemia within one week after weaning from the CPB circuit.

[0119] In embodiments, following Example 5 below, the method comprises applying NMR spectroscopy on said biological sample, e.g. 1D NMR, so as to arrive at a resulting spectrum covering a spectral range representative of said biomarkers, and applying a computer-based algorithm in response, so as to determine an output indicative of said subject being at risk of developing hypoxemia. Especially, said biological sample, such as a blood sample, has been provided from said subject between start of surgery (defined as skin incision) until 24 hours after weaning from the CPB circuit (preferably until 8 hours). Preferably the sample is a serum sample.

[0120] More specifically, said biological sample has been provided from said subject between a time during surgery, such as from weaning from the CPB circuit until 8 hours after weaning from the CPB circuit. In the present context, end of surgery is defined as time of skin closure (such as by stitching), which will often be 1-2 hours after weaning from the CPB. As shown in table 5 and 6, strong predictive values for the risk of developing hypoxemia are obtainable from just after weaning from the CPB circuit.

[0121] The method may comprise selecting from the resulting spectrum a plurality of spectral intervals (or regions), wherein a combination of spectral intervals are selected to cover at least spectral intervals indicative of at least one of: arginine, phenylalanine, PUFA, alanine, TMAO, carnitine, isobutyric acid, isobutyrylglycine and citrate. Especially, the combination of spectral intervals may be selected to cover all of: arginine, phenylalanine, PUFA, alanine, TMAO, carnitine, isobutyric acid, isobutyrylglycine and citrate. These compunds are part of the ratios considered the most relevant according to the present invention. In a further embodiment, the combination of spectral intervals are selected to cover at least spectral intervals indicative of at least one of: carnitine, isobutyrylglycine, N-Acetyl-Glucosamine, arachidonic and eicosapentanoic acid, isobutyric acid and citrate. In yet a further embodiment, the combination of spectral intervals may be selected to cover all of: carnitine, isobutyrylglycine, N-Acetyl-Glucosamine, arachidonic and eicosapentanoic acid, isobutyric acid, and citrate.

[0122] The combination of spectral intervals may be selected to cover at least spectral intervals indicative of at least 14 metabolites, such as 14-19 metabolites. The method may comprise performing a computer-based statistical analysis of said combination of spectral intervals, such as a computer-based statistical analysis involving applying a machine learning algorithm. The method may comprise presenting a visual output indicative of said subject being at risk of developing hypoxemia, e.g. a presentation of data allowing a medical person to determine if said subject is or is not at risk of developing hypoxemia.

[0123] Following Example 5, an aspect of the invention may be defined as a method for determining the risk of developing hypoxemia, the method comprising [0124] providing a biological sample from a subject, said subject having had open cardiac surgery with the use of cardiopulmonary bypass (CPB), wherein said sample has been provided from said subject between start of surgery (defined as skin incision) and 72 hours after weaning from the CPB circuit; [0125] applying NMR spectroscopy on said biological sample, such as performing a 1D CPMG NMR analysis, so as to arrive at a resulting spectrum representative of the level of at least one biomarker selected from the group consisting of carnitine, isobutyrylglycine, N-Acetyl-Glucosamine, arachidonic and eicosapentanoic acid, isobutyric acid and citrate;
and/or [0126] said resulting spectrum being representative of the ratio between levels of biomarkers of at least one ratio selected from the group of ratios selected from citrate/phenylalanine, PUFA/phenylalanine, alanine/phenylalanine, citrate/TMAO, carnitine/TMAO, isobutyric acid/phenylalanine, and isobutyrylglycine/arginine;
and [0127] applying a computer-based algorithm on said resulting spectrum involving a comparison with reference data, so as to determine if said subject is at risk of developing hypoxemia and/or so as to determine if said subject is not at risk of developing hypoxemia.

[0128] Further, following Example 5, an aspect of the invention may be defined as a computer program product having instructions which, when executed on a processor cause the processor to [0129] receive data representing a resulting spectrum from an NMR spectroscope representative of the level of at least one biomarker selected from the group consisting of carnitine, isobutyrylglycine, N-Acetyl-Glucosamine, arachidonic and eicosapentanoic acid, isobutyric acid and citrate;
and/or [0130] said resulting spectrum being representative of the ratio between levels of biomarkers of at least one ratio selected from the group of ratios selected from citrate/phenylalanine, PUFA/phenylalanine, alanine/phenylalanine, citrate/TMAO, carnitine/TMAO, isobutyric acid/phenylalanine, and isobutyrylglycine/arginine;
and [0131] to apply a processing algorithm on said resulting spectrum involving a comparison with reference data, so as to determine if said subject is at risk of developing hypoxemia and/or so as to determine if said subject is not at risk of developing hypoxemia.

[0132] Such computer program product may be designed for execution on a portable computer, a server, partly or fully integrated with software dedicated for an NMR spectrometer etc. The resulting spectrum determined by the NMR spectrometer may be transferred in wire or wireless form to a separate processing device arranged to execute the computer program product. Especially, the computer program product may be present on a computer readable medium.

Other Aspects of the Invention

[0133] The method according to the invention is believed also to find use in other medical conditions being triggers of acute lung injury. These conditions are related to the present medical condition (open heart surgery with the use of CPB), by also relating to acute lung injury as described previously.

[0134] These conditions may be sepsis, pneumonia, severe trauma, inhalation of toxic agents, aspiration of gastric contents, major surgery, and transfusions.

[0135] Thus, in an additional aspect, the invention relates to a method for determining the risk of developing hypoxemia, the method comprising [0136] providing a biological sample from a subject considered at risk of developing hypoxemia; [0137] determining the level of at least one biomarker selected from the group consisting of carnitine, citrate, isobutyrylglycine, N-acetyl-glucosamine, arachidonic acid, and isobutyric acid; [0138] and/or [0139] determining the ratio between levels of markers of at least one ratio selected from the group of ratios selected from citrate/phenylalanine, PUFA/phenylalanine, alanine/phenylalanine, citrate/TMAO, carnitine/TMAO, isobutyric acid/phenylalanine, and isobutyrylglycine/arginine; [0140] comparing said levels of one or more biomarkers or ratios to one or more reference levels; [0141] determining that said subject is at risk of developing hypoxemia if said level of one or more biomarkers are above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said one or more levels are equal to or below the one or more reference levels; [0142] and/or [0143] determining that said subject is at risk of developing hypoxemia if said one or more ratios between levels of biomarkers are above the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said level of one or more ratios between levels of biomarkers are equal to or below the one or more reference levels.

[0144] In a specific embodiment, said subject experiences sepsis, pneumonia, severe trauma such as cardiac, aortic, thoracic, and spinal surgery, traumatic brain injury, lung transplantation, multiple traumas and transfusions, or has inhaled toxic agents, or aspired gastric contents.

[0145] It should be noted that embodiments and features described in the context of one of the aspects of the present invention also apply to the other aspects of the invention.

[0146] All patent and non-patent references cited in the present application, are hereby incorporated by reference in their entirety.

[0147] The invention will now be described in further detail in the following non-limiting examples.

EXAMPLES

Example 1

Methods Employed for Identifying Biomarkers, which are Early Biomarkers for the Development of Hypoxemia

Patient Group and Clinical Outcome

[0148] Serum samples were obtained from forty-seven (n=47) patients undergoing coronary artery bypass grafting (CABG). Three days postoperatively, 15 patients showed no signs of hypoxemia; while 32 developed hypoxemia with PaO.sub.2 below normal values.

Samples Preparation

[0149] To avoid preanalytical bias due to sample collection, all blood samples were collected and prepared by the same person. Blood samples were obtained from both the left atrium (LA) and pulmonary artery (PA) precisely 16 h after weaning from the CPB circuit. To obtain serum samples, blood was allowed to clot at room temperature for 30 minutes, and was subsequently centrifuged at 3000 rpm for 10 minutes. Aliquots of LA and PA serum were immediately stored at 80 C. until assayed.

Sample Preparation for NMR

[0150] Prior to NMR measurements, samples were thawed for 2 h at 4 C., vortexed, and centrifuged for 5 minutes at 4 C. and 14000 rpm to remove cells and other precipitated materials. Aliquots of 400 L of supernatant were mixed with 200 L 0.2 M phosphate buffer (pH 7.4, uncorrected meter reading) in .sup.2H.sub.2O (99% .sup.2H) to minimize variations in pH and to reduce serum viscosity. Throughout the whole process, the samples were kept on ice. The mixture was pipetted into a 5-mm NMR tube and NMR analysis was performed.

NMR Measurements

[0151] Spectra were acquired on a Bruker DRX-600 NMR spectrometer (Bruker BioSpin, Germany and Switzerland) equipped with a TXI (hydrogen, carbon, nitrogen) probe (Bruker BioSpin, Switzerland) operating at 600.13 MHz for .sup.1H. The experiments were acquired at a constant temperature of 310.1 K (37 C.). For the analysis, a T.sub.2 relaxation-edited Carr-Purcell-Meiboom-Gill (CPMG) experiment was used. This experiment attenuates broad signals from slowly tumbling proteins and lipoproteins. 128 free induction decays (FIDs) were collected with 32768 complex data points over a spectral width of 11.97 ppm and an acquisition time of 2.28 s. A relaxation delay of 2 s was applied between each FID during which weak continuous wave irradiation (B.sub.1/2=26.6 Hz) was applied at the water frequency (presaturation). The total spin-echo relaxation delay was 67.4 ms and consisted of (--) elements, where r is a delay of 0.4 ms and n is a 180 pulse of 22 s length. Spectral processing was carried out in TopSpin version 2.1 (Bruker BioSpin, Germany). Prior to Fourier transformation, exponential multiplication corresponding to a line broadening of 0.3 Hz was applied to the FIDs, which were further zero-filled by a factor of 2 to double the number of points. Spectra were manually phase- and baseline-corrected, and the methyl signal of lactate was used as chemical shift reference at 1.33 ppm.

[0152] For metabolite identification, three types of two-dimensional (2D) experiments were acquired. 2D J-resolved .sup.1H-NMR experiments, with water pre-saturation during a 2 s relaxation delay were acquired with 8 FIDs for each of the 80 increments. Spectral width was set to 11.6 ppm and 54.7 Hz in F2 and F1 direction, respectively. Prior to 2D-Fourier transformation, both dimensions were multiplied by an unshifted sine bell function and the number of points was doubled by zero-filling. Further, spectra were tilted by 45 and symmetrized along the F1 axis. After processing, the skyline projections of 2D J-resolved spectra were manually baseline corrected and calibrated using the center of the methyl signal of lactate. 2D homonuclear .sup.1H,.sup.1H-TOCSY (Total Correlation Spectroscopy) and .sup.1H,.sup.13C-HSQC (Heteronuclear Single Quantum Coherence) spectra with presaturation were run on representative samples, with different number of FIDs, increments, spectral widths and mixing times in order to focus on different spectral regions. Information obtained from these spectra was used to find matching metabolites in the Human Metabolome Database. Further metabolite assignments were performed using AMIX (v. 3.9.10, Bruker BioSpin), BRUKER bbiorefcode databases (2.7.0-2.7.3), and literature.

Quantification of NMR Data

[0153] This was performed using the AMIX multi integration tool. Peaks of interest were integrated using the line shape analysis option with a fixed noise factor of 3.5 and a line shape threshold of 0.01. Peaks were integrated using the sum of all points in the region as the integration mode, and normalized to the glucose metabolite region (3.52 ppm (H-5 of -glucose) and 5.24 ppm (H-1 of -glucose)). A 36:64 equilibrium distribution between - and -glucose was used for the calculations. To calculate the concentration of a given metabolite, we used the following formula:

[00001]$C_X\left[\mathrm{mM}\right]=\frac{I_X}{N_X}.Math.C_{\mathrm{Glucose}}.Math.0.64.Math.\frac{N_{.Math.\text{-}.Math.\mathrm{Glucose}}}{I_{.Math.\text{-}.Math.\mathrm{Gluxose}}}$

where C.sub.X is metabolite concentration in mM, I.sub.X is the integral of the metabolite .sup.1H peak, N.sub.X is the number of protons contributing to the metabolite .sup.1H peak, C.sub.Glucose is the chemically determined glucose concentration, 0.64N.sub.-Glucose is the number of protons contributing to the -glucose signal (at 3.52 ppm) used as reference multiplied by 0.64 (the mole fraction of -glucose), and I.sub.-Glucose is the integral of the -glucose signal at 3.52 ppm.

NMR Data Preprocessing

[0154] CPMG spectra were converted into an n-by-m matrix (n=94 LA and PA samples from 47 patients, m=8700 equal buckets of 0.001 ppm width) in AMIX (Analysis of MIXtures software package, version 3.9.10, Bruker BioSpin, Germany) using the region between 9 to 0 ppm, and excluding the water signal region between 4.80-4.5 ppm. Further processing and multivariate modelling were carried out in MATLAB R2011b.sup.56 coupled with PLS-Toolbox 6.5 (Eigenvector Research, Wenatchee, Wash.). Binned data was iCoshifted.sup.57, normalized to lactate concentration measured 16 h after surgery, log transformed, and autoscaled.

Multivariate Modelling

[0155] Several mathematical and statistical modelling approaches were employed to relate the pre-processed NMR data to disease phenotypes. Partial least-squares discriminant analysis (PLS-DA) was applied to evaluate the diagnostic possibilities of NMR and to discover biomarkers. The most important variables obtained by the method were identified and quantified.

Statistics

[0156] The Statistical Package and Services Solutions (SPSS) software v22.0 was used for the statistical analysis. Descriptive statistics were computed for each group and the quantitative data was summarized using mean, median, and standard deviation.

[0157] The Shapiro-Wilk normality test was applied followed by either parametric or nonparametric methods. When comparing metabolite differences between unaffected and hypoxemic patients (Pa02<8.4 kPa, PaO.sub.2/FiO2<40 kPa (<300 mmHg)), either an independent t-test or Mann-Whitney U test was used. When comparing differences between the three groups (unaffected, mild, severe), either one way analysis of variance (ANOVA) with Tukey HSD or Dunnett T3 (depending on homogeneity of variance) multiple comparison post-hoc tests, or the Kruskal-Wallis test with Dunn's post-hoc test were applied. When comparing data between the left atrium and pulmonary artery, a paired t-test or the nonparametric analogue Wilcoxon-signed rank sum test was used. Significance was set at p<0.05.

[0158] In order to visualize global changes between groups, the means or medians (depending on data distribution) of each metabolite in the corresponding group were calculated. As such, metabolite fold changes were calculated as the ratio between mild and unaffected and severe and unaffected patients with the formula:

[00002]${\mathrm{FC}}_{\frac{M.Math..Math.\left(\mathrm{or}.Math..Math.S\right)}{U}}\left[\%\right]=\frac{M.Math..Math.\left(\mathrm{or}.Math..Math.S\right)-U}{U}.Math.100$

[0159] Where FC=fold change in percentage, M=mild, U=unaffected, S=severe. Because the ratio between two metabolite concentrations may carry more information than the two metabolites alone, the freely available web-based tool ROCCET (ROC Curve Explorer & Tester).sup.27 was used to compute all possible pairs of metabolite combinations that could be related to later outcomes. Metabolite ratios exhibiting statistical significance (p<0.05) were further used as biomarker candidates.

[0160] The association between metabolites and PaO.sub.2 values obtained 3 days postoperatively were assessed by Pearson's correlation analysis. ROC curves were established to determine the prognostic value of each biomarker in future diagnosis. The area under the curve (AUC) and corresponding 95% confidence intervals (CIs) are provided.

Example 2

Metabolome Screening Reveals Early Signs of Disease:

[0161] The systemic and pulmonary phenotypes were monitored by .sup.1H Nuclear Magnetic Resonance (NMR) spectroscopy. A typical one-dimensional (1D) serum NMR spectrum is characterized by broad resonances from lipids and glycoproteins, and narrow resonances from glucose, lactate, and citrate, among others. Spectra of two samples collected on the first postoperative day (exactly 16 h after weaning from CPB), one from a patient showing no signs of hypoxemia (PaO.sub.2=10.7 kPa or 80.2 mmHg), and one from a patient developing hypoxemia (PaO.sub.2=4.9 kPa or 36.7 mmHg), reveal differences in several signals, of which lipids are the most significant (FIG. 1).

[0162] Since the metabolome mirrors environmental changes, we hypothesized that the disease could be reflected at the metabolic levels on the first day postoperatively. Therefore, we screened for possible associations between the metabolome and the hypoxemic scores (PaO.sub.2) used in diagnosis. We have divided the 32 patients diagnosed with acute lung injury into patients developing mild and severe hypoxemia, to provide better understanding of underlying mechanisms. Subsequently, the differences between the three groups (unaffected, mild, and severe hypoxemia) were assessed by partial least-squares discriminant analysis (PLS-DA). The models displayed >95% accuracy, indicating the remarkable performance of our screening test.

Example 3

[0163] The metabolic fingerprints found in the screening test were investigated further. 64 different metabolites were analysed, of which one could not be identified (U (5.10-5.08 ppm)). Perturbations in the levels of metabolites involved in normal cellular functioning (amino acids, carbohydrates, ketones), cellular signalling (1,2-diacylglycerol), inflammation (arachidonic and eicosapentanoic acid), cell membrane and alveolar surfactant components (fatty acids, cholesterols, phospholipids) were found crucial in the development of injury. Carnitine, arachidonic and eicosapentanoic acid, glycoprotein, citrate, and phenylalanine, among others, showed the highest fold changes, indicating their key roles in later outcomes. Most metabolites showed consistent trends from none-to-mild-to-severe acute lung injury (FIG. 4), indicating their correlation to the degree of later pulmonary dysfunction and their possible function as predictive biomarkers.

[0164] The list of the most relevant biomarkers is listed in the tables below.

TABLE-US-00001 TABLE 4 Most significant metabolites found correlating to the Partial pressure of O.sub.2 (PaO.sub.2) measured on the day of diagnosis (72 h after weaning from CPB). Pearson Correlation (Metabolites vs. PaO.sub.2 (72 h)) Metabolites Correlation Sig. (2-tailed) 1 Citrate/U(5.10-5.08 ppm) 0.645** 1.77E06 2 Isobutyrylglycine/Phenylalanine 0.609** 8.86E06 3 Isobutyrylglycine/Arginine 0.602** 1.20E05 4 Citrate/Phenylalanine 0.572** 4.02E05 5 PUFA/Phenylalanine 0.562** 5.83E05 6 Alanine/Phenylalanine 0.556** 7.37E05 7 Isobutyrate/Phenylalanine 0.546** 1.03E04 8 PUFA/Arginine 0.546** 1.06E04 9 Arachidonic (Ara) and 0.509** 3.62E04 Eicosapentanoic acid (Epa) 10 Isobutyrylglycine 0.455** 1.70E03 11 Citrate/TMAO 0.452** 1.85E03 12 Carnitine 0.421** 4.00E03 13 Citrate 0.388** 8.50E03 14 Carnitine/TMAO 0.383** 9.37E03 15 NAc-Glc 0.330* 2.69E02 16 Phenylalanine 0.317* 3.38E02 17 Alanine 0.293 5.10E02

[0165] Abbreviations: Sig.=significance level; U=unknown metabolite, TMAO: trimethylamine-N-oxide, 1,2-DAG=1,2-diacylglucerol, GPE=glycerophosphoethanolamine, GPC=glycerophosphocholine, 3-HBA=3-hydoxybutyric acid, VLDL=very low density lipoproteins, LDL=low density lipoproteins, HDL=high density lipoproteins, DAGPL=diacylglycerolphospholipid, UFA=unsaturated fatty acids, Free FA=free fatty acids, NAcGal=N-acetyl-galactose amine, NAc-Glc=N-acetyl-glucosamine, Lyso-PC=lysophosphatidylcholine. In the present context PUFA refers to polyunsaturated fatty acids (PUFAs). PUFAs are categorized according to the number and position of double bonds in the fatty acids according to an accepted nomenclature that is well known to those of ordinary skill in the art. Polyunsaturated omega-3/6 fatty acids (with 3 double bonds); e.g. eicosopentaenoic acid[EPA, 20:5 (n3)], docosahexaenoic acid [DHA, 22:6(n3)], linoleic (18:2 n6), gamma-linolenic acid (18:3 n6), alpha-linolenic (18:3 n3) and stearidonic (18:4 n3) acid.

TABLE-US-00002 TABLE 2A Most significant metabolites found predicting later outcomes by applying t-test; Hypoxemia vs. Unaffected Patients. Hypoxemia vs. Unaffected Patients Metabolites AUC 95% C.I. AUC Sig. 1 Carnitine 0.90 0.79-0.979 6.96E07 2 Citrate/TMAO 0.89 0.749-0.993 3.59E06 3 Citrate 0.88 0.764-0.961 2.43E05 4 Carnitine/TMAO 0.86 0.714-0.971 5.21E05 5 NAcGlc 0.86 0.748-0.951 2.14E04 6 Citrate/Phenylalanine 0.85 0.722-0.942 1.31E04 7 Citrate/U (5.10-5.08 ppm) 0.85 0.705-0.948 2.60E05 8 Isobutyrate/Phenylalanine 0.84 0.676-0.935 5.64E04 9 Arachidonic and 0.82 0.688-0.953 5.95E05 Eicosapentanoic acid 10 Isobutyric acid 0.79 0.63-0.899 2.31E03 11 Alanine/Phenylalanine 0.78 0.633-0.908 2.94E03 12 Isobutyrylglycine/Arginine 0.78 0.635-0.911 1.32E03 13 Isobutyrylglycine/Phenylalanine 0.77 0.622-0.917 2.52E03 14 PUFA/Phenylalanine 0.76 0.601-0.882 4.83E03 15 PUFA/Arginine 0.76 0.594-0.894 2.87E03 16 Isobutyrylglycine 0.75 0.603-0.879 3.65E03

TABLE-US-00003 TABLE 2B Metabolites found significant in predicting later outcomes by applying t-test; Severe vs. Unaffected Patients. Severe vs. Unaffected Patients Metabolites AUC 95% C.I. AUC Sig. Citrate/Phenylalanine 1.00 1.00-1.00 1.06E06 Citrate/U (5.10-5.08 ppm) 1.00 1.00-1.00 2.71E05 Isobutyrate/Phenylalanine 0.98 0.912-1.00 5.88E05 PUFA/Phenylalanine 0.98 0.905-1.00 2.14E05 PUFA/Arginine 0.97 0.881-1.00 3.85E05 Alanine/Phenylalanine 0.97 0.865-1.00 1.45E05 Isobutyrylglycine/Arginine 0.95 0.857-1.00 7.73E05 Isobutyrylglycine/Phenylalanine 0.94 0.833-1.00 5.44E05 Arachidonic and 0.93 0.802-1.00 5.46E04 Eicosapentanoic acid Citrate 0.93 0.782-1.00 6.45E04 Carnitine 0.90 0.762-1.00 8.75E04 Citrate/TMAO 0.92 0.77-1.00 8.20E05 Carnitine/TMAO 0.88 0.714-1.00 1.59E03 NAcGlc 0.88 0.714-1.00 1.70E02

[0166] Abbreviations: AUC=area under the curve, C.I.=confidence level, U=unknown metabolite, TMAO=trimethylamine-N-oxide, NAcGlc=N-acetyl-glucosamine

Conclusion

[0167] A set of different biomarkers having predictive values for determining the risk of developing hypoxemia in patients undergoing open heart surgery with the use of CPB have been identified.

Example 4

Early Markers Predict Risk for Hypoxemia

[0168] Several serum metabolites measured 16 h postoperatively were found to significantly correlate to PaO.sub.2 measured 72 h after weaning from CPB (FIG. 4/Tables 1-2). Also, several metabolites showed strong predictive power for mild and severe hypoxemia with high area under the curves (AUC>0.8, p<0.0001) (tables 2A and 2B). This is the first study presenting predictive biomarkers of postoperative acute lung injury 56 h before hypoxemia was diagnosed.

TABLE-US-00004 TABLE 3 Metabolites considered most relevant (single marker or ratios). Regulation considered indication of Metabolites the risk of developing hypoxemia Carnitine Higher levels in hypoxemia Citrate/TMAO Higher ratio in hypoxemia Citrate Higher levels in hypoxemia Carnitine/TMAO Higher ratio in hypoxemia O-NAc-Glc Higher levels in hypoxemia Citrate/Phenylalanine Higher ratio in hypoxemia Isobutyric acid/Phenylalanine Higher ratio in hypoxemia Arachidonic acid (Ara, 20:4 -6) and Higher levels in hypoxemia Eicosapentanoic acid (Epa, 20:5 -3) Isobutyric acid Higher levels in hypoxemia Alanine/Phenylalanine Higher ratio in hypoxemia Isobutyrylglycine/Arginine Higher ratio in hypoxemia Isobutyrylglycine/Phenylalanine Higher ratio in hypoxemia PUFA/Phenylalanine Higher ratio in hypoxemia PUFA/Arginine Higher ratio in hypoxemia Isobutyrylglycine Higher levels in hypoxemia

[0169] Thus, the predictive markers presented in Table 3 were considered the most relevant metabolites identified. It is indicated in the table that higher levels are considered an indication of the risk of developing hypoxemia for the indicated biomarkers. Similarly, higher ratios between the indicated biomarkers are considered an indication of the risk of developing hypoxemia for the indicated biomarker ratios.

[0170] Table 4 contains a list of biomarkers also determined as being significantly different in patients at risk of developing hypoxemia, though with lower predictive values. One or more of these biomarkers may also be included in a method according to the invention to further increase the predictive value. The skilled person can adapt the method accordingly in relation to reference values.

TABLE-US-00005 TABLE 4 Metabolites considered relevant (single marker). Metabolites Phosphoglyceric acid (PGA) Lower levels in hypoxemia N-acetyl-galactosamine (NAcGal) Higher levels in hypoxemia Glycine Higher levels in hypoxemia Trimethylamine-N-oxide (TMAO) Lower levels in hypoxemia Dimethyl amine (DMA) Higher levels in hypoxemia Pyruvate Higher levels in hypoxemia Alanine Higher levels in hypoxemia Acetate Higher levels in hypoxemia Lysine Higher levels in hypoxemia Leucine, Isoleucine, Valine Higher levels in hypoxemia Malonic acid Higher levels in hypoxemia 3-Hydroxybutyric acid (3-HBA) Higher levels in hypoxemia Acetoacetate Higher levels in hypoxemia Formate Higher levels in hypoxemia Ethanol Higher levels in hypoxemia Isobutyrate Higher levels in hypoxemia 2-Ketobutyrate Higher levels in hypoxemia Carnitine Higher levels in hypoxemia Unsaturated fatty acids (UFA) Higher levels in hypoxemia Free fatty acids (FFA) + Adipate Higher levels in hypoxemia Arachidonic acid (Ara, 20:4 -6) and Higher levels in hypoxemia Eicosapentanoic acid (Epa, 20:5 -3) PUFA (Lipoic acid and related Higher levels in hypoxemia metabolites) Free Cholesterol Higher levels in hypoxemia Est. Cholesterol Higher levels in hypoxemia Lipoproteins (VLDL, LDL, HDL) Higher levels in hypoxemia Glycerophosphocholine (GPC) Higher levels in hypoxemia Lysophosphocholine (Lyso-PC) Higher levels in hypoxemia Choline Higher levels in hypoxemia Heparin Lower levels in hypoxemia Sphingomyelin Higher levels in hypoxemia Glycerol Lower levels in hypoxemia 1,2-Diacylglycerol (DAG) Higher levels in hypoxemia Diacylglycerophospholipid (DAGPL) Higher levels in hypoxemia Triacylglycerol (TAG) Higher levels in hypoxemia Plasmenyl Higher levels in hypoxemia Phosphoethanolamine (PE) Higher levels in hypoxemia Glycerophosphoethanolamine (GPE) Higher levels in hypoxemia Glutamate Higher levels in hypoxemia Glutamine Lower levels in hypoxemia Proline Higher levels in hypoxemia Creatine Lower levels in hypoxemia Phenylalanine Lower levels in hypoxemia Tyrosine Lower levels in hypoxemia Dopa Lower levels in hypoxemia Dopamine Lower levels in hypoxemia Histidine Lower levels in hypoxemia 1-Metylhistidine Higher levels in hypoxemia 3-Metylhistidine Lower levels in hypoxemia Urocanate Lower levels in hypoxemia Plasmalogen Lower levels in hypoxemia Xanthosine Lower levels in hypoxemia Hypoxanthine Lower levels in hypoxemia Arginine Lower levels in hypoxemia Citrulline Lower levels in hypoxemia U(5.10-5.08 ppm) Lower levels in hypoxemia Phosphoenolpyruvate (PEP) Higher levels in hypoxemia

[0171] In an embodiment of the invention, the methods of the invention further includes determining the level of one or more of the biomarkers listed in Table 4. As can be seen in the right column, for some of the markers lower levels are indicative of the development of hypoxemia. Thus, for these specific markers the method will instead include [0172] determining that said subject is at risk of developing hypoxemia if said level of one or more biomarkers are below the one or more reference levels, or determining that said subject is not at risk of developing hypoxemia, if said one or more levels are equal to or above the one or more reference levels.

[0173] An example of such a biomarker is Phosphoglyceric acid (PGA) listed in table 4 above.

Example 5

Early Indication of the Risk of Developing Hypoxemia Using NMR

Patient Cohort and Diagnosis

[0174] Patients scheduled for elective coronary artery bypass surgery (CABG) with use of cardiopulmonary bypass (CPB) were consecutively included in the study after informed consent was obtained (n=50).

[0175] The clinical diagnosis was based on the ratio between partial pressure of oxygen and fraction of inspired oxygen (PaO.sub.2/FiO.sub.2) calculated from PaO.sub.2 measured in arterial blood samples collected from the radial artery 72 hours after weaning from CPB and in order to standardize the measurements, arterial blood samples were taken while patients were breathing atmospheric air for at least 10 minutes.

Sample Collection

[0176] Blood was collected after the thorax was open but before patients were attached to the CPB circuit, just after CPB was removed (0 hours), 2 hours, 4 hours, and 8 hours after weaning from CPB.

Sample Preparation

[0177] Serum samples were thaw for 30 min. at 4 C., vortexed and subsequently centrifuged for 5 min at 12.100 g and 4 C. A total of 400 L of the clear supernatant was mixed with 200 L 0.2M phosphate buffer (pH 7.4 uncorrected meter reading, 99% .sup.2H.sub.2O) in a 5 mm NMR tube.

NMR Experiment

[0178] Following, NMR spectra of the samples were recorded on a Avance-III 600 MHz NMR spectrometer (BrukerBioSpin, Rheinstetten, Germany) equipped with a cryogenically cooled, triple-resonance, TCI probe, using a sample temperature of 298.1 K (25 C.). Spectra acquisition was controlled using TopSpin 3.1 software (Bruker Biospin). A T.sub.2 filter using the Carr-Purcell-Meiboom-Gill (CPMG).sup.21 pulse sequence with water presaturation was applied for each measurement. Each CPMG spectrum was acquired as 64 k data points over a spectral width of 20 ppm, 256 scans, a fixed receiver gain (RG) of 203, and a relaxation delay (d1) of 4 s.

Spectra Processing

[0179] Spectral processing was carried out in TopSpin 3.1. Data was exponentially multiplied corresponding to a line broadening of 0.3 Hz (CPMG), Fourier transformed, phase- and baseline corrected, and calibrated to the chemical shift of the methyl signal of L-alanine at 1.48 ppm. Spectra were reduced to regions of equal buckets (0.001 ppm) and the water region between 4.65 to 4.95 ppm was excluded in AMIX (Analysis of MIXtures software, v.3.9.10, Bruker BioSpin, Germany).

The Script

[0180] Subsequently, data was exported in Matlab R2011b (The MathWorks, Inc., MA) programming environment. A script based on complex pattern recognition was written containing (1) spectral post-processing (generalized log transformation (gLog).sup.22 to enhance small signals in the spectrum, normalization, scaling and centering), (2) feature extraction by selecting the significant spectral regions correlating to later outcomes, (3) multivariate statistical analysis based on the machine learning algorithms PLS regression and PLS-DA, (4) validation, and (5) visualization of results.

[0181] For the feature extraction procedure, selecting the regions of interest (ROI) was achieved by dividing the data into intervals of 50 buckets and correlating each interval to the PaO.sub.2 measured on the diagnostic day. The correlation was performed by the multivariate PLS regression analysis. Intervals with low correlation were discarded while significant intervals were further used in modeling. Extracting these distinctive features from serum spectra at each time point has shown to be the key characteristic of the success of our computer-aided diagnostic test. From a total of 10500 buckets, 7000-9500 buckets (depending on the time point analyzed) were removed as they did not show any correlation with the diagnostic PaO.sub.2 values. Even the number of remaining features/buckets was higher than the number of samples (1000-3500 buckets vs. 100 samples per model); we considered that the data was not over-fitted, as these variables correspond to the information imbedded in 14-19 metabolites. These regions were consistently picked at each of the measured time points with minor variations.

[0182] Taking into account that for each model we had 90-100 samples and the information from 14-19 metabolites, the models are considered not to be over-fitted. To assess the sensitivity and specificity of these selected regions, multivariate PLS-DA modelling was performed.

[0183] For the validation purpose the Venetian-Blinds cross-validation with 10 segmental splits was applied. Here 10 consecutive samples (5 patients in duplicate) were removed and a model was created. Following, the 10 samples were predicted. This procedure was repeated until all samples were removed once. To avoid overoptimistic interpretation and misleading results, we only refer to the cross-validated (CV) results of our modelling, however, the estimated (or calibrated) results are also provided. This validation procedure gave us an indicator of how well the model might work in predicting future samples.

[0184] A second level of model validation was performed by using permutation testing. Here, the models were tested for randomness, to show that no other model performed equally well or better than the main prediction model. After scrambling the PaO.sub.2 values measured three days postoperatively (for PLS regression modeling) and the group labels (hypoxemia vs. unaffected; for the PLSDA modeling) 500 times and performing multivariate modelling, we compared the true optimal PLS and PLSDA models with the permuted models. The true model performance were then statistically compared to the distribution of the permuted models, and a p-value was calculated by means of Wilcoxon's test. A p-value <0.001 was considered significant, meaning that the predictive power of our approach was significantly associated with PaO.sub.2 and the diagnostic outcomes, and was not a false-positive association resulting from random prediction.

Results

[0185] In an embodiment of the invention the method is meant to be applied as follow; when a patient undergoes cardiac surgery, its blood sample is collected at one or several time points, and serum is extracted. Serum is mixed with phosphate buffer (see below for Sample preparation), and the sample is then run on NMR (see below for NMR experiment). Following this, the achieved spectrum is preprocessed (see below for Spectra processing), and the data has to be scanned through the script/test provided. As such, an embodiment includes the following steps: [0186] 1. Obtaining the blood sample [0187] 2. Preparing the sample [0188] 3. Running a 1D CPMG NMR [0189] 4. Applying standardized spectral processing to reduce artifacts (e.g. water suppression, pH shifts, baseline noise) (FIG. 1, black box) [0190] 5. Fragmenting/binning the spectra to further minimize peak shifts [0191] 6. Exporting the data into the Matlab environment [0192] 7. Running the computed-aided diagnostic script/test (FIG. 1, grey box) [0193] 8. Reading or interpreting the diagnosis and/or the effect of treatment

[0194] The script/test (FIG. 6, gray box) is a computer-based diagnostic and/or monitoring algorithm based on machine learning. The script/test is optimized for 5 different time points, each time point containing 100 spectra run on serum samples collected before coupling to CPB, just after weaning from CPB but still during surgery, and 2, 4, and 8 hours postoperatively. First, the script post-processes the data (transformation, normalization, scaling, centering), then it selects the key ROI to be used for further diagnosis, it applies multivariate modelling (PLS or PLSDA) depending on the diagnostic interest (PaO.sub.2 or hypoxemia vs. none), and it returns a certain score (for PLS) or a threshold level (for PLSDA) for each sample in the analysis. Finally, based on these scores or thresholds, a regression and a scores plot are provided and each sample's position is highlighted. Based on the position of the samples on the plots the doctors can read/interpret the diagnosis. If one is interested in monitoring and determining the effect of the treatments started in the patients, the script/test can be used for this purpose as well. Here the same procedure is performed as previously mentioned, but the samples are collected after the end of surgery (e.g. 2, 4, and/or 8 hours postoperatively). Depending on the effect of the initiated treatments, a patient's position in the plots will eventually move towards recovery (higher scores levels).

[0195] To show the predictive power of our approach, the PLS regression modelling was applied on preprocessed spectral data with the purpose of predicting the diagnostic PaO.sub.2 values measured three days postoperatively. Already before patients are coupled to the CPB, specific ROIs on NMR spectra (FIG. 7) show 0.70 cross-validated correlation (R.sup.2=0.70) with the PaO.sub.2 values measured 3 days postoperatively, with a root mean-square error of cross-validation (RMSECV) of 0.75 kPa (5.62 mmHg) (Table 5). After weaning from CPB, a 0.91 correlation and a 0.40 kPa (2.62 mmHg) prediction error can be detected by NMR (Table 5). Models conducted on samples collected 2 hours, 4 hours and 8 hours postoperatively have R.sup.2>0.93 and RMSECV<0.36 kPa (Table 5).

TABLE-US-00006 TABLE 5 For the validation purpose the Venetian-Blinds cross-validation with 10 segments was applied. Here 10 consecutive samples (5 patients in duplicate) were removed and a model was created. Following, the 10 samples were predicted. This procedure was repeated until all samples were removed once. Due to the limitation of not having completely new samples to test the models, the values achieved in this work are relative and represent the average results for removing and predicting 10 of the 100 samples. Already after incision/thoracotomy, our approach can predict the diagnostic outcomes. The cross-validated correlation between the metabolome before CPB and the measured PaO.sub.2 on day 3 is 0.7. Increased correlation is observed when the metabolome is measured just after weaning from CPB (0 hours), 2 hours, 4 hours, and 8 hours postoperatively. R.sup.2 RMSE (PaO.sub.2 [kPa]) Slope Bias Cal CV Cal CV Cal CV After thoracotomy, 0.94 0.70 0.332 0.749 0 0.021 before CPB 0 hours (just after CPB) 0.98 0.91 0.146 0.402 3.5*10.sup.15 0.009 2 hours (post-CPB) 0.99 0.93 0.131 0.362 2.6*10.sup.15 0.001 4 hours (post-CPB) 0.99 0.95 0.123 0.302 2.6*10.sup.15 0.011 8 hours (post-CPB) 0.99 0.94 0.127 0.340 1.7*10.sup.15 0.013

[0196] The binary classifier PLS-DA was performed to detect the sensitivity and specificity of our method in diagnosing hypoxemia and in determining the risk of ARDS while patients were still undergoing the surgery. After thoracotomy, NMR with machine learning show a 77.8% CV sensitivity and 84.2% CV specificity towards differentiating between hypoxemic and unaffected patients (Table 6). Waiting until weaning from CPB, resulted in a significant increase in the method's accuracy, with 88.2% CV sensitivity and 92.2% CV specificity (Table 6). Also, applying the modeling on samples collected 2 hours, 4 hours and 8 hours postoperatively shows good agreement with aforementioned results (Table 6), indicating the great value of our approach in both diagnosis and monitoring of initiated therapeutical treatments.

TABLE-US-00007 TABLE 6 PLSDA applied on the selected ROIs and the diagnosis labels (hypoxemia/none) determine the diagnostic outcome while patients are still undergoing surgery. The 8 hours following surgery still reveal good sensitivity and specificity, indicating their value in diagnosis and monitoring of new treatment options. The results are shown for both the calibrated (Cal) and Venetian-Blinds cross-validated (CV) data. 500x Permutation testing Time Sensitivity Specificity Class error RMSE (p-val) point Cal CV Cal CV Cal CV Cal CV CV Pre- 0.833 0.778 0.952 0.842 0.107 0.190 0.292 0.392 <0.001 CPB 0 h 1.00 0.882 0.984 0.922 0.008 0.098 0.184 0.337 <0.001 2 h 1.00 0.912 0.984 0.938 0.007 0.075 0.179 0.075 <0.001 4 h 0.971 0.882 0.917 0.883 0.056 0.117 0.321 0.405 <0.001 8 h 0.971 0.943 0.967 0.917 0.031 0.070 0.226 0.362 <0.001

Conclusion

[0197] The present example provides a method for screening and early diagnosis of hypoxemia during cardiac surgery, allowing for direct support of surgeons in the operation theatre. Because the method may lead to diagnostic results within 15-20 minutes, the doctors can make critical and vital decisions during surgery. Also, because the method may provide up to 500 spectra references run on serum collected at different time points, it also allows doctors and nurses to monitor the effect of treatments in process several hours postoperatively.

[0198] Newly diagnosed patient samples can also be added to the provided database together with the known outcome, thus leading to a continuously growing and improving database for more accurate prediction.

Example 6

Aim

[0199] Test if Inosine; Dopamine; Dihydroxyphenylalanine (Dopa); Hypoxanthine; and Trigonelline are biomarkers in patients with an increased risk of developing hypoxemia after having had open cardiac surgery with the use of cardiopulmonary bypass (CPB).

Materials and Methods

Patient Group and Clinical Outcome

[0200] Serum samples were obtained from fifty (n=50) patients undergoing coronary artery bypass grafting (CABG). Three days postoperatively, 18 patients showed no signs of hypoxemia; while 32 developed hypoxemia with PaO.sub.2 below normal values, of which 9 developed severe hypoxemia.

Samples Preparation

[0201] To avoid preanalytical bias due to sample collection, all blood samples were collected and prepared by the same person. Blood samples were obtained from both the left atrium (LA) and pulmonary artery (PA) after sternotomy but before the attachment to the CPB, precisely after weaning from the CPB circuit (0 hours), and 2, 4, 8, and 20 hours ater weaning from the CPB circuit. To obtain serum samples, blood was allowed to clot at room temperature for 30 minutes, and was subsequently centrifuged at 3000 rpm for 10 minutes. Aliquots of LA and PA serum were immediately stored at 80 C. until assayed.

Sample Preparation for NMR

[0202] Prior to NMR measurements, samples were thawed for 2 h at 4 C., vortexed, and centrifuged for 5 minutes at 4 C. and 14000 rpm to remove cells and other precipitated materials. Aliquots of 400 L of supernatant were mixed with 200 L 0.2 M phosphate buffer (pH 7.4, uncorrected meter reading) in .sup.2H.sub.2O (99% .sup.2H) to minimize variations in pH and to reduce serum viscosity. Throughout the whole process, the samples were kept on ice. The mixture was pipetted into a 5-mm NMR tube and NMR analysis was performed.

NMR Measurements

[0203] Spectra were acquired on a Bruker DRX-600 NMR spectrometer (Bruker BioSpin, Germany and Switzerland) equipped with a TXI (hydrogen, carbon, nitrogen) probe (Bruker BioSpin, Switzerland) operating at 600.13 MHz for .sup.1H. The experiments were acquired at a constant temperature of 310.1 K (37 C.). For the analysis, a T.sub.2 relaxation-edited Carr-Purcell-Meiboom-Gill (CPMG) experiment was used. This experiment attenuates broad signals from slowly tumbling proteins and lipoproteins. 128 free induction decays (FIDs) were collected with 32768 complex data points over a spectral width of 11.97 ppm and an acquisition time of 2.28 s. A relaxation delay of 2 s was applied between each FID during which weak continuous wave irradiation (B.sub.1/2=26.6 Hz) was applied at the water frequency (presaturation). The total spin-echo relaxation delay was 67.4 ms and consisted of (--) elements, where r is a delay of 0.4 ms and n is a 180 pulse of 22 s length. Spectral processing was carried out in TopSpin version 2.1 (Bruker BioSpin, Germany). Prior to Fourier transformation, exponential multiplication corresponding to a line broadening of 0.3 Hz was applied to the FIDs, which were further zero-filled by a factor of 2 to double the number of points. Spectra were manually phase- and baseline-corrected, and the methyl signal of alanine was used as chemical shift reference at 1.48 ppm.

[0204] For metabolite identification, three types of two-dimensional (2D) experiments were acquired. 2D J-resolved .sup.1H-NMR experiments, with water pre-saturation during a 2 s relaxation delay were acquired with 8 FIDs for each of the 80 increments. Spectral width was set to 11.6 ppm and 54.7 Hz in F2 and F1 direction, respectively. Prior to 2D-Fourier transformation, both dimensions were multiplied by an unshifted sine bell function and the number of points was doubled by zero-filling. Further, spectra were tilted by 45 and symmetrized along the F1 axis. After processing, the skyline projections of 2D J-resolved spectra were manually baseline corrected and calibrated using the center of the methyl signal of lactate. 2D homonuclear .sup.1H,.sup.1H-TOCSY (Total Correlation Spectroscopy) and .sup.1H,.sup.13C-HSQC (Heteronuclear Single Quantum Coherence) spectra with presaturation were run on representative samples, with different number of FIDs, increments, spectral widths and mixing times in order to focus on different spectral regions. Information obtained from these spectra was used to find matching metabolites in the Human Metabolome Database. Further metabolite assignments were performed using AMIX (v. 3.9.10, Bruker BioSpin), BRUKER bbiorefcode databases (2.7.0-2.7.3), and literature.

Quantification of NMR Data

[0205] This was performed using the AMIX multi integration tool. Peaks of interest were integrated using the line shape analysis option with a fixed noise factor of 3.5 and a line shape threshold of 0.01. Peaks were integrated using the sum of all points in the region as the integration mode, and these integrals were used as relative concentrations as arbitrary units (a.u.).

NMR Data Preprocessing

[0206] CPMG spectra were converted into an n-by-m matrix (n=590 LA and PA samples from 50 patients, m=8700 equal buckets of 0.001 ppm width) in AMIX (Analysis of MIXtures software package, version 3.9.10, Bruker BioSpin, Germany) using the region between 9 to 0 ppm, and excluding the water signal region between 4.80-4.5 ppm. Further processing and multivariate modelling were carried out in MATLAB R2011b.sup.56 coupled with PLS-Toolbox 6.5 (Eigenvector Research, Wenatchee, Wash.). Binned data was iCoshifted.sup.57, normalized to total intensity, log transformed, and mean centered.

Multivariate Modelling

[0207] Partial least-squares discriminant analysis (PLS-DA) was applied to identify variables important in sample discrimination and the most significant variables obtained by the method were identified.

Statistics

[0208] The Statistical Package and Services Solutions (SPSS) software v23.0 was used for the statistical analysis. 2-way ANOVA with Tukey HSD multiple comparison post-hoc tests was used to analyse differences within and between groups. The groups were compoused by the time points before CPB (CPB), 0 hours after CPB (+CPB), 2, 4, 8, and 20 hours after CPB, and the classes unaffected (none), mildly affected by hypoxemia (mild), and severely affected by hypoxemia (severe). Significance was set at p<0.05.

[0209] In order to visualize changes between the time points and the groups, the means and standard deviation with the corresponding 95% confidence intervals (CIs) were calculated for each of the significant metabolites (e.g. Trigonelline, Hypoxanthine, Dopamine, Dihydroxyphenylalanine (Dopa) and Inosine).

Results:

[0210] Inosine, Dopamine, Dihydroxyphenylalanine (Dopa), Hypoxanthine, and Trigonelline measured after sternotomy but before CPB, after weaning from the CPB (0 h), and 2, 4, 8, and 20 h postoperatively were found significantly changing with time and with the risk of developing postoperative hypoxemia (FIGS. 8 A-E/Table 7) at an early stage after weaning from the CPB. The metabolites show consistent trends from none-to-mild-to-severe acute lung injury (FIGS. 8 A-E), indicating their correlation to the degree of later pulmonary dysfunction and their possible function as predictive biomarkers. The metabolic regulation is shown in FIGS. 8 A-E and explained in Table 7.

[0211] This is the first study presenting predictive biomarkers of postoperative acute lung injury just after weaning from the CPB or 72 h before hypoxemia was diagnosed.

[0212] The list of the relevant biomarkers found being significantly different within the groups is listed in the tables below.

TABLE-US-00008 TABLE 7 Metabolites found significantly changing with time according to the later diagnosis. The timepoins were consisting of samples collected after sternotomy but before CPB, right after weaning from CPB (0 h), 2 h, 4 h, 8 h, and 20 h after CPB; while the class labels were consisting of unaffected patients (none), mildly affected by hypoxemia (mild), and severely affected by hypoxemia (severe) Time Class Regulation considered Sig. Sig. indication of the risk of 2-way ANOVA p-value p-value developing hypoxemia Trigonelline 1.7E05 1.8E05 Lower levels in hypoxemia Hypoxanthine 1.5E11 2.7E04 Lower levels in hypoxemia Inosine 8.6E25 0.003 Lower levels in hypoxemia Dopamine 2.6E87 0.05 Lower levels in hypoxemia Dihydroxy- 3.9E14 2.04E09 Higher levels in phenylalanine (Dopa) hypoxemia

Conclusion

[0213] Preliminary data indicates that Inosine; Dopamine; Hypoxanthine; and Trigonelline are down-regulated while Dihydroxyphenylalanine (Dopa) are up-regulated in patients at risk of developing hypoxemia after having had open cardiac surgery with the use of cardiopulmonary bypass (CPB) at an early stage after end surgery.

Discussion of Results

[0214] It has been investigated whether the metabolome in patients progressing into postoperative hypoxemia, changes so early and dramatically that this complication could be predicted at least 2-3 days before the clinical signs. NMR spectra (FIG. 1) and multivariate modelling accurately related perturbations in the metabolome recorded during and after weaning from CPB with later diagnostic scores. Considering the challenges involved in predicting acute lung injury due to the lack of specific biomarkers, the presented method may therefore enable early disease detection.

[0215] From serum samples collected first morning postoperatively, several metabolites were identified varying between patients (FIGS. 3, 4, and 8). Of these, several markers significantly correlated with PaO.sub.2 (p0.05) and showed outstanding diagnostic power, with AUCs>0.8 (see tables above). These are markers of antioxidant status and peroxidation (histidine, PUFA); inducible nitric oxide (iNOS) production (arginine); ATP depletion and reactive oxygen species (ROS) formation (citrate); inflammation (arachidonic and eicosapentanoic acid, membrane injury (phospholipids); and mitochondrial dysfunction (carnitine). The AUC and detection rates for acute lung injury seen in this study are the first of their kind; no other similar studies have reported the use of metabonomics in predicting postoperative outcomes.

[0216] This suggests suggest an impaired oxidative phosphorylation. Carnitine, normally found in mitochondria and required for transport of FA across the membrane for -oxidation, was significantly increased in patients with hypoxemia, which may also indicate impaired oxidative phosphorylation and a possible leakage into the bloodstream. Supporting this, isobutyrylglycine, a marker of impaired mitochondrial FA -oxidation, was also associated with severity of hypoxemia. Mitochondrial impairments enhance oxidative stress, causing alveolar cell death. Under homeostatic conditions, however, various antioxidants are capable of alleviating the damaging effects of ROS from the system, but an insufficient antioxidant barrier cannot counteract this damage.

[0217] Membrane bound phospholipids and PUFA are susceptible to peroxidation leading to the formation of highly reactive hydroperoxides. These react with many biochemical substances having enormous impacts on normal cellular functioning, including endothelial activation and surfactant phospholipid disruption. PUFA and adipic acid (a byproduct of peroxidation) levels correlated with later pulmonary dysfunction, whereas plasmalogen (implicated in protection against ROS and peroxinitrite formation) levels were decreased, especially in the severe hypoxemic group.

[0218] Choline, phospholipids, 1,2-DAG, PUFA, and cholesterol are essential for structural integrity and cell membrane signalling. An increase in their levels reflects an activation of the protective mechanism and a possible structural derangement in patients progressing into disease. Lipids are highly interconnected signalling molecules that regulate metabolic, innate immune and inflammatory processes, and alteration in one lipid will automatically trigger major deregulation in several signalling pathways causing profound physiological responses. The levels of arachidonic and eicosapentanoic acids were higher in patients with acute lung injury, indicating an increased inflammatory environment. These PUFA may be released by activation of phospholipase A.sub.2, hydrolyzing membrane glycerophospholipids, which may also indicate cell membrane detachment. This is known to be one of the characteristics of acute lung injury. Because one of the properties of this disease is dysfunction of alveolar surfactants, the increased levels of some phospholipids, in samples of hypoxemic patients, may also be partly explained by surfactant leakage, which could indicate a damaged endothelial-alveolar barrier.

[0219] Decreased arginine levels were observed in the group developing severe hypoxemia, Blood arginine depletion has previously been linked to acute lung injury, sepsis and cystic fibrosis, and supplementation of arginine was observed to reduce inflammation. This deficiency may have triggered production of superoxides, which further interact with NO producing peroxynitrites, ammonia accumulation, and increased iNOS expression, causing pulmonary microvasculature disruption and tissue damage.

[0220] N-Acetylated glucosamine was also found positively correlating with disease. This carbohydrate is-part of pentose metabolism, which is known to be more active under stress, especially under hypotonic stress conditions. This is the first study of its kind, looking at metabolic changes on the first postoperative day following CABG in patients at-risk of developing ALI/ARDS due to the use of CPB. As described above, our findings suggest defects in the mitochondrial respiratory system affecting ROS generation, impaired antioxidant state, increased peroxidation and oxidative stress, disruption of unprotected cell membranes, impaired surfactant production, in patients developing hypoxemia. TCA, arginine, and PUFA take part in a wide range of biological reactions participating in both beneficial and detrimental results, thus therapeutic interventions targeting their paths may open new strategies to prevent postoperative acute lung injury.

[0221] We have identified metabolite hallmarks of hypoxemia, which may bridge the gap between pathogenesis and full-blown disease. Markers such as arachidonic and eicosapentanoic acid, citrate, carnitine, glycine, phenylalanine, arginine, and histidine serve as central nodes in their metabolism and therefore have a high impact in predicting disease progression. No single metabolite captured the complexity of injury alone, as different pathways simultaneously showed imbalances. Therefore, to prevent early progression into hypoxemia, therapeutic targeting several pathways would be more effective.