WO2018146162A1 - Molecular biomarker for prognosis of sepsis patients - Google Patents

Abstract

The invention relates to a method for determining the survival prognosis of a patient admitted to an intensive care unit (ICU) or a sepsis patient, comprising steps of determining the expression level of at least BPGM and/or TAP2 in a biological sample of said patient, and compare said expression level(s) with a control and, determine said survival prognosis based on said comparison.

Description

MOLECULAR BIOMARKER FOR PROGNOSIS OF SEPSIS PATIENTS

TECHNICAL FIELD

The present invention relates in general to the use of the expression level of particular genes and their potential to predict the mortality of sepsis. More specifically, the invention relates to the expression level of the genes BPGM and TAP2 as a prognostic biomarker for a high risk of mortality sepsis endotype.

BACKGROUND

Sepsis remains a remarkable adversary to medicine, characterized by poor prognosis and high mortality rates ^1,2. Despite the burden on patients, their families and the health care system, treatment remains mainly supportive \ Unrecognized population substructures and the heterogeneity in the host response complicate the identification of high-risk patients who would benefit from specific adjuvant therapy ³. Blood transcriptional profiling has provided substantial advances in the context of sepsis ⁴. Although promising new diagnostic biomarkers have emerged from the application of blood genomics to sepsis ^5"7, patient selection for interventional trials and prognostication in sepsis continue to be driven by clinical criteria. While supervised analysis of sepsis patients discordant for survival have identified candidate protein and gene expression prognostic markers^8,9, substantial heterogeneity remained unexplained. Therefore, there remains a need in the art to identify subgroups (endotypes) of sepsis patients based on biomarkers present in blood. Moreover, there is an overwhelming need for biomarkers that can identify poor prognosis sepsis patients, which may in turn provide an invaluable tool to monitor patients' responses to critical care and treatment.

SUMMARY OF THE INVENTION

The present invention is based on the following findings.

To explore the molecular fingerprints that underlie the heterogeneity of survival prognosis between sepsis patients, we assessed genome-wide blood gene expression in blood of adult patients admitted to the ICUs of two hospitals in the Netherlands (discovery cohort, n=306; first validation cohort, n=216) and of patients with sepsis caused by CAP admitted to 29 ICUs in the U K (n=265; second validation cohort) ¹⁷ (Table 1.1). Gene expression data from the discovery cohort were analyzed using a previously developed unsupervised consensus clustering method (Figure 2.1) ¹⁹'²⁰·²⁸. Considering cluster (endotype) quality and stability ' , we reached a consensus in partitioning at four molecular endotypes (Figure 1.1A, B and Figure 2.2), designated Marsl-4. Patients who had positive silhouette widths ²¹ (81 Marsl, 94 Mars2, 63 Mars3 and 29 Mars4), indicative of their high intra-endotype similarity ^1ΒΆ2& _ι were subsequently used. In so doing, we identified 140 genes that clearly stratified Marsl to Mars4 endotypes (Figure 1.1C and Table 2.1).

Sepsis endotypes did not show an association with comorbidities and clinical scores including APACHE IV scores (Table 2.1); however septic shock prevalence and Sequential Organ Failure Assessment (SOFA) scores ²⁹ showed significant dependencies to sepsis molecular endotype classification (Figure 1.1D and E). The estimated effect size (Cramer's V) of septic shock was 0.23, indicating moderate dependency. Kaplan-Meier analysis revealed an association with mortality (Figure 1.1F). The worst outcome was found for those patients classified to Marsl, with 39% mortality at 28-days. Marsl classified patients had the worst outcome until 1 year of patient follow-up (logrank p = 0.023; Figure 2.3A). The hazard ratio for death within 28-days of patients classified to Marsl equated to 1.86 (95% CI: 1.21-2.86, p=0.0045). To test whether the combination of a molecular and clinical scoring system may be of benefit to patient risk stratification, we evaluated the net reclassification improvement and integrated discrimination improvement ^26,30 using a combined APACHE IV score (clinical) and sepsis endotype classification (molecular) model. This clinico-molecular model significantly improved 28-day mortality risk prediction (net reclassification improvement (continuous) [95% CI] : 0.33 [0.09 - 0.58], p=0.008; integrated discrimination improvement [95% CI] : 0.015 [0.0002 - 0.03], p=0.047) when compared to 28-day mortality risk prediction by APACHE IV scores alone (Figure 2.3B). Altogether, these results provide for a molecular classification of sepsis patients that possesses prognostic value either in isolation or in combination with established clinical scores.

In order to ascertain robustness of our findings we evaluated two validation cohorts (Table 1.1). Applying the 140 gene classifier clearly identified four sepsis endotypes in the first validation cohort consisting of all cause sepsis (Figure 1.2A and Figure 2.4A). Consistent with the discovery cohort, SOFA scores and septic shock were significantly associated to sepsis endotype classification (Figure 1.2B and C), with moderate dependency to septic shock (Cramer's V = 0.29). In concordance with the discovery cohort APACHE IV scores did not associate with endotype classification (Table 2.2), while Kaplan Meier analysis showed a significant association to mortality (Figure 1.2D and Figure 2.4B). Again the highest mortality rate was found for patients classified as Marsl, with 32% mortality at 28 days. The association with mortality was also evident until 1 year of patient follow-up, with Marsl classified patients having the worst outcome (logrank p = 0.0031; Figure 2.4B). The hazard ratio for death within 28 days of patients classified as Marsl equated to 1.97 (95% CI: 1.11-3.54, p=0.024). Combining clinical (APACHE IV) and molecular (sepsis endotype classification) data significantly improved 28-day mortality risk prediction (net reclassification improvement (continuous) [95% CI] : 0.38 [0.01 - 0.66], p-value = 0.008; integrated discrimination improvement [95% CI] : 0.028 [0.0018 - 0.055], p=0.036) as compared to APACH E IV scores alone.

We also detected four endotypes (Figure 1.2E) with favorable endotype stability (Figure 2.5A) in patients admitted to ICUs in the UK with sepsis caused by CAP. ¹⁷ Evaluation of the association to SOFA scores, septic shock and mortality revealed statistically significant associations; Marsl classified patients had relatively worst prognosis (Figure 1.2F-H). The hazard ratio for death within 28-days of ICU admission for those patients classified as Marsl in the second validation cohort equated to 2.02 (95% CI : 1.07-3.82, p=0.031). Considering that the second validation cohort only consisted of CAP patients, we combined the two all-cause sepsis cohorts (discovery plus first validation cohort) to show that the four subtypes were present irrespective of the primary site of infection; for this we separately analyzed the two main causes of sepsis, i.e., pneumonia (n=215) and peritonitis (n=123) (Figure 2.5B-E). In a cohort of pediatric sepsis patients (GSE13904, n=81) ¹⁸ (Table 2.3) we found three endotypes (Marsl, Mars2 and Mars4) with favorable stability (Figure 2.6A and B); in this cohort Mars3 was not reliably detected (Figure 2.6B) and evaluation of 28-day mortality as well as pediatric risk of mortality scores (PRISM) ³¹ revealed no significant dependencies on sepsis endotype classification (Figure 2.6C and D).

Collectively, these data provide robustness to the classification of adult sepsis patients as blood molecular endotypes and suggest it is only partially applicable to children with sepsis.

In order to provide a means to utilize our findings in a clinical context we sought to derive sepsis endotype scores. To this end, we assessed 77,840 combinations of genes in the 140 gene classifier for classification of the four molecular endotypes and identified 8 genes that in specific combinations reliably stratified patients from the discovery cohort as sepsis molecular endotypes (Figure 1.4A). Gene expression combinations of BPGM (bisphosphoglycerate mutase):L4P2 (transporter 2, ATP binding cassette subfamily B member), GADD45A (growth arrest and DNA damage inducible alpha):PCGF5 (polycomb group ring finger 5), AHNAK (AHNAK nucleoprotein):PDCD10 (programmed cell death 10) and IFIT5 (interferon induced protein with tetratricopeptide repeats S):GLTSCR2 (glioma tumor suppressor candidate region gene 2) were used to classify patients as Marsl, Mars2, Mars3 and Mars4 endotypes, respectively (Figure 1.4B and Table 2.4). These endotype biomarkers also accurately classified patients in the two validation cohorts (Figure 1.4C-F and Figure 2.8). Collectively, these findings provide candidate molecular biomarkers for the identification of sepsis molecular endotypes at ICU admission.

We identified four endotypes in three heterogeneous sepsis cohorts based on blood leukocyte genome- wide expression profiles on ICU admission. These sepsis endotypes (Marsl-4) had pathophysiologic and prognostic implications, and were not easily discernable by clinical characteristics. Both common and distinct biological signatures characterized the four sepsis endotypes. The poor prognosis Marsl sepsis endotype was associated with a dramatic decrease in expression of genes involved in innate and adaptive immune functions; whereas the relatively low risk Mars3 endotype had elevated expression of adaptive immune/T cell functions. Finally, 8 genes were derived as candidate biomarkers for the identification of sepsis endotypes at ICU admission, with BPGM and TAP2 transcripts delineating the poor prognosis Marsl sepsis endotype.

An overarching observation of patient endotypes (or subtypes in the oncogenomics field) was their association to varying degrees of disease severity and mortality. Our findings showed that the consistent association to mortality across the cohorts we tested was an important feature of the four sepsis endotypes identified from whole blood leukocyte transcriptional profiling.

Our analysis of the GAinS CAP cohort, applying an ensemble of methods for rigorously measuring quality and stability of sample partitioning, as well as classification by machine learning, showed that a four sepsis endotype model was favorable in this adult CAP cohort. Of note, the low-risk SRS2 endotype(17) was highly correlated to the low-risk Mars3 sepsis endotype, both characterized by heightened expression of genes predominantly involved in adaptive immune (mainly T cell) functions.

The poor prognosis sepsis endotype, Marsl, was characterized by decreased expression in genes that function in both innate and adaptive immune mechanisms concomitant to high expression of specific cellular metabolic pathways including heme biosynthesis. Glycine accumulation, biosynthesized by serine derived from the glycolysis pathway intermediate 3-phosphoglycerate, fuels heme biosynthesis and in turn modulates ATP synthesis via oxidative phosphorylation in mitochondria(35).We therefor believe that Marsl classified patients may represent an "immunoparalyzed" endotype with poor prognosis. The Mars2 and Mars4 endotypes were characterized by high expression of genes involved in proinflammatory and innate immune reactions, for example NF-kB signaling and interferon signaling, respectively. Thus, Mars2 and Mars4 classified patients may represent distinct hyper-inflammatory endotypes. Genes with elevated expression in the relatively lowest risk Mars3 endotype were over- represented for adaptive immune/T cell pathways. Clinical trials for sepsis seeking to modify the host response have thus far yielded no beneficial effect on outcome(38). A growing body argues for the re-assessment of clinical trial designs(39) to include biomarkers reflecting the status of the host response(38,40). Our study shows that sepsis patients can be classified to four endotypes based on host leukocyte response signatures with distinct pathophysiologic and prognostic features. We afford molecular biomarkers that reflect sepsis endotype with high accuracy in different sepsis populations and that add prognostic value to existing clinical severity scores. We envisage that endotype classification may provide more homogeneity to the notoriously heterogeneous sepsis population, can aid in identifying patient subgroups who would most likely benefit from precision therapeutics, and can be used to monitor effects of targeted therapies.

Based on the above findings, the invention provides a method for determining the survival prognosis of a patient suffering from sepsis, said method comprising steps of: (a) determining the expression level(s) of TAP2 and/or BPGM in a biological sample of said patient, (b) determining said expression level(s), (c) determining the risk of suffering from sepsis based on said expression level(s). In another aspect, the invention provides a method for determining the survival prognosis of a patient suffering from sepsis, said method comprising steps of: (a) determining the expression level(s) of GADD45A and/or PCGF5 in a biological sample of said patient, (b) determining said expression level(s), (c) determining the risk of suffering from sepsis based on said expression level(s). In another aspect, the invention provides a method for determining the survival prognosis of a patient suffering from sepsis, said method comprising steps of: (a) determining the expression level(s) of AHNAK and/or PDCD10 in a biological sample of said patient, (b) determining said expression level(s), (c) determining the risk of suffering from sepsis based on said expression level(s). In another aspect, the invention provides a method for determining the survival prognosis of a patient suffering from sepsis, said method comprising steps of: (a) determining the expression level(s) of IFIT5 and/or GLTSCR2 in a biological sample of said patient, (b) determining said expression level(s), (c) determining the risk of suffering from sepsis based on said expression level(s).

Preferably, said step of determining the expression level comprises the use of at least one techniques selected from the group consisting of:

i. PCR-based methods, preferably RT-PCR, Quantitative RT-PCR, nucleic acid microarray analysis, isothermal DNA/RNA amplification techniques, and/or in situ hybridization, ii. immunological methods, preferably immunohistochemistry, ELISA, binding assays, and/or Western Blot, iii. and/or spectroscopical methods, preferably Raman spectroscopy, and/or Mass spectroscopy.

In a preferred embodiment, said method comprises a step of comparing said expression level(s) with a reference control or reference value.

Preferably, said step b) comprises the step of checking whether or not the expression level of TAP2 and/or BPGM is higher or lower than a predetermined threshold level.

Preferably, said expression levels are normalized.

In a preferred embodiment, said method comprises the step of determining the ratio of the expression levels of the of TAP2 and BPGM and determining the risk of suffering from sepsis based on said ratio. Preferably, wherein the ratio of the expression levels of the of TAP2 and BPGM is compared

arithmetically with a reference control or reference value and the risk of suffering from sepsis is determined if said ratio is higher than said reference control or reference value.

Preferably, a TAP2:BPGM gene expression ratio of 1.15 or higher is associated with a poor survival prognosis.

In a preferred embodiment, said biological sample is a blood sample.

Preferably, said patient is a human.

BRIEF DESCRIPTION OF THE FIGURES

Figure 1.1 Unsupervised classification of sepsis patients and the association to clinical characteristics and outcome (discovery cohort). (A) Consensus clustering of the discovery cohort (n=306) shows optimal partitioning to four clusters, labeled as Marsl, Mars2, Mars3 and Mars4. (B) Silhouette width analysis illustrating stable partitioning to four molecular endotypes. Percent of patients assigned to each subtype is denoted. (C) Gene expression heatmap illustrating the 140 gene classifier derived to categorize the patients to endotypes. Red, over-expression; turquoise, under-expression (heatmap rows). Endotypes are color coded and labeled (top bar). (D-F) Endotypes were evaluated for their association to clinical severity indices, (D) septic shock, (E) Sequential Organ Failure Assessment (SOFA) scores, (F) 28-day mortality by Kaplan-Meier survival analysis. X2, chi-squared.

Figure 1.2 Assessment of sepsis molecular endotypes in the validation cohorts. (A) Random forest prediction of Marsl, Mars2, Mars3 and Mars4 endotypes in the first validation cohort (n=261). Heatmap illustrating the 140 gene expression classifier. Red, over-expression; turquoise, under- expression (rows). (B-D) Stratification of first validation cohort as Mars sepsis endotypes was evaluated for the association against (B) total Sequential Organ Failure Assessment (SOFA) scores, (C) septic shock, and (D) 28-day mortality by Kaplan Meier survival analysis. (E) Random forest prediction of sepsis endotypes and heatmap representation of the second validation cohort (United Kingdom, E-MTAB-4421, n=265)17. (F-H) Stratification of the second validation cohort as Mars sepsis endotypes was evaluated for the association against (F) total Sequential Organ Failure Assessment (SOFA) scores, (G) septic shock and (H) 28-day mortality by Kaplan Meier survival analysis.

Figure 1.3 Biological interpretation of sepsis molecular endotypes. (A) Volcano plot representation of differential gene expression in the discovery cohort categorized to Marsl, Mars2, Mars3 and Mars4 endotypes each compared to healthy subjects (n=42). Plots integrate gene expression (log2 fold expression of subtype versus healthy subjects, x-axis) and multiple-comparison adjusted P values (y-axis). Within plots, pie charts illustrate the extent of gene expression changes with red slices denoting the number of significantly over-expressed genes (adjusted P value < 0.05 and fold expression > 1.5), blue slices denoting significant under-expression (adjusted P value < 0.05 and fold expression < -1.5) and grey slices illustrating significantly differential gene expression (adjusted P value < 0.05) but outside of the fold expression < -1.5 and > 1.5 cutoff. (B) Correlograms illustrating the relationship of the gene expression changes in sepsis patients classified as Marsl-4 endotypes relative to healthy subjects. Rho, Spearman's correlation estimate. All correlations were significant (p < 1x10-10). (C) Venn-Euler diagram illustrating the relationship in gene expression changes across the four sepsis molecular endotypes relative to healthy subjects. Red arrows denote significant gene over-expression, blue arrows denote significant gene under-expression. The high-risk Marsl endotype is labeled to illustrate a portion of the gene expression response that is unique. (D) Ingenuity pathway analysis of unique canonical signaling gene sets per endotype (columns in heatmaps). Canonical signaling pathways were grouped into super pathways. Heatmaps represent over-representation Fisher's test probabilities (considering multiple comparison adjusted P < 0.01). Red spectrum, significantly over-expressed canonical pathways; turquoise spectrum, significantly under-expressed canonical pathways (rows).

Figure 1.4 Derivation, validation of candidate sepsis molecular endotype biomarkers. (A)

Receiver-operator characteristics (ROC) area-under-the-curve (AUC) analyses of gene expression sepsis endotype classifier scores for Marsl (BPGM:TAP2), Mars2 (GADD45A:PCGF5), Mars3 (AHNAK:PDCD10) and Mars4 (IFIT5:GLTSCR2) in the discovery cohort. 95% CI, bootstrap resampled 95% confidence interval. (B) Stripcharts illustrating the significant differences in sepsis endotype scores. Black horizontal line denotes median. Red horizontal line denotes the threshold defined by the best coordinate on the ROC curve. (C) Stripchart illustrating the Marsl sepsis endotype score in the first validation cohort from the Netherlands. (D) ROC AUC analysis of the Marsl sepsis endotype score in the first validation cohort. 95% CI, bootstrap resampled 95% confidence interval (E) Stripchart illustrating the Marsl sepsis endotype score in the second validation cohort from the United Kingdom. (F) ROC AUC analysis of the Marsl sepsis endotype score in the second validation cohort. 95% CI, bootstrap resampled 95% confidence interval

Figure 2.1. Methodological steps employed for the unsupervised consensus clustering and gene expression classifier construction using the discovery set and subsequent validation in independent validation datasets.

Figure 2.2. Unsupervised consensus cluster identification and stability in the discovery cohort (n = 306). (a) consensus cumulative distribution function (CDF) with increasing number of clusters (k2 to kl2). (b) Relative change in area-under-CDF curve with increasing number of clusters (k). (c) cophenetic correlation coefficients at each cluster (k). (d) patient tracking plot that follows patients with increasing numbers of clusters (k).

Figure 2.3. Mortality and net reclassification improvement (NRI) in the discovery cohort (n = 306) considering clinical and four molecular subtype models, (a) Patient follow-up as at 14, 90 days and 1 year with sepsis molecular endotype stratification, (b) Continuous NRI calibration plots of the combined Acute Physiology and Chronic Health Evaluation IV (APACHE IV) score and sepsis molecular endotypes considering 28-day mortality in discovery cohort. Plot illustrate the Hosmer-Lemeshow test probabilities, which denote optimal calibration for a continuous net reclassification assessment (p>0.05).

Figure 2.4. Molecular endotype prediction in first validation cohort from the Netherlands (n = 216). (a) Silhouette widths of sepsis endotype classification, (b) Kaplan-Meier analysis of follow-up data with sepsis endotype stratification as at 14, 90 days and 1 year, (c) Continuous net reclassification improvement calibration plots of the combined Acute Physiology and Chronic Health Evaluation IV (APACH E IV) score and sepsis molecular endotypes considering 28-day mortality in discovery cohort. Plot illustrate the Hosmer-Lemeshow test probabilities, which denote optimal calibration for a continuous net reclassification assessment (p>0.05).

Figure 2.5. Sepsis molecular endotypes in pneumonia and abdominal sepsis (A) Silhouette widths of community-acquired pneumonia patients from the UK cohort (E-MTAB-4421, n=265) classified to Mars endotypes. (B, C) Silhouette widths of the two Dutch cohorts combined (discovery and first validation) diagnosed as (B) pneumonia (n=215) and (C) peritonitis (n=123); patients with both pneumonia and peritonitis were excluded from this analysis. (D, E) Heatmap representation of 140 gene expression indices (rows) and patient samples stratified according to molecular endotype membership (columns) diagnosed at ICU admission with (D) pneumonia, or (E) peritonitis. Figure 2.6: (a) Random forest prediction of sepsis endotypes in the pediatric sepsis cohort, USA (GSE13904; n=81). Heatmap depicts the 140 gene classifier (rows) (b) Silhouette widths of pediatric sepsis patient samples classified to Mars endotypes. (c, d) Endotypes were evaluated for their association to (c) 28-day mortality (binary handled), and (d) pediatric risk of mortality (PRISM) scores.

Figure 2.7. (A) A molecular endotype model for the risk stratification of critically ill patients with sepsis. The relatively high-risk Marsl sepsis endotype was defined by elevated expression of heme biosynthesis genes concomitant with pronounced under-expression of pattern recognition receptor, cytokine signaling, lymphocyte signaling and antigen presentation pathways. PRR, pattern recognition receptor. (B) Stratification of community-acquired pneumonia patients from the U K GAinS cohort by sepsis endotype (Marsl-4) was evaluated for the association to sepsis response signatures 1 and 2 (SRS1 and 2). X2 p, chi-square test probability.

Figure 2.8. Candidate sepsis endotype biomarker assessment in two validation cohorts. (A) Dot plots of Mars2 (GADD45A:PCG F5), Mars3 (AH NAK:PDCD10) and Mars4 (I FIT5:G LTSCR2) scores in the first validation cohort from the Netherlands, and (B) in the second validation cohort from the U K. GADD45A, growth arrest and DNA damage inducible alpha; PCGF5, polycomb group ring finger 5; AH NAK, AH NAK nucleoprotein; PDCD10, programmed cell death 10; I FIT5, interferon induced protein with tetratricopeptide repeats 5; GLTSCR2, glioma tumor suppressor candidate region gene 2. Horizontal black line denotes median. DETAILED DESCRIPTION

The term "survival prognosis" as used herein, refers to the prediction of the likelihood of sepsis- attributable death.

The term "sepsis" as used herein is defined as a Systemic Inflammatory Response Syndrome to an infective process in which severe derangement of the host immune system fails to prevent extensive 'spill over' of inflammatory mediators from a local infection focus into the systemic circulation. Preferably, said patient of an infection is assessed as probable or definite using Center for Disease Control and Prevention(14) and International Sepsis Forum consensus definitions(15), as described in detail (12). In a preferred embodiment, "sepsis" is defined as the presence of infection with a probable or definite likelihood, accompanied by at least one additional parameter as described in the 2001 International Sepsis Definitions Conference (16). In a preferred embodiment, said sepsis patient of the invention suffers from community acquired pneumonia. The term "community acquired pneumonia" (sepsis) as used herein is equivalent and has been used interchangeably with the term "community acquired bacterial pneumonia."

The term "BPGM" as used herein refers to a gene encoding the bisphosphoglycerate mutase (RefSeq accession NC_000007.1).

The term "TAP2" as used herein refers to a gene encoding transporter 2, ATP binding cassette subfamily B member (RefSeq accession NC_000006.12).

The term "gene biomarker" as used herein refers to any or multiple of the genes of the invention.

The term "biological sample" as used herein refers to any sample from a patient for diagnostic, prognostic, or personalized medicinal uses and may be obtained from surgical samples, such as biopsies or fine needle aspirates, from paraffin-embedded tissues, from frozen tissue samples, from fresh tissue samples, from a fresh or frozen body fluid. Most preferably the sample contains white blood cells. However, any other suitable biological samples (e.g. bodily fluids such as blood, stool, etc..) in which the gene expression level of a gene of interest can be determined are included within the scope of the invention.

The term "determining the expression level" as used herein, refers to the process of determining whether a gene is expressed and if this is the case assessing to which extend it is expressed. These assessments are usually carried out in parallel, but of course they can also be carried out after each other. Therefore, the process of determining gene expression may include all necessary preparatory steps know in the art such as protein, mRNA, RNA, DNA and/or cDNA preparation; measurement using techniques such as real time PCR, immunohistochemistry or microarray; basic arithmetic operations such as determining a mean value, if gene expression level for one biological sample is determined using more than one probe since the average of the probes can then be calculated in order to increase the accuracy of the inventive method; etc.

The term "control" as used herein refers to a biological sample or samples of a patient suffering from sepsis for determining control expression levels; and/or a predetermined expression level or ratio for each of two of biomarker expression levels and/or a predetermined cut-off level. Preferably, a control refers to control of BPGM and TAP2 expression levels (or in certain embodiments to expression levels of the other biomarkers genes associated with the mars 2, 3 or 4 endotypes as described herein) ; and/or a predetermined expression level or ratio for each of two of BPGM and TAP2 expression levels levels (or in certain embodiments to expression levels of the other biomarkers genes associated with the mars 2, 3 or 4 endotypes as described herein) and/or a predetermined cut-off level. The control can for example be a reference profile to which test sample expression levels are compared, and/or a predetermined level or levels expressed for example as a numerical value and/or range (e.g. control range) corresponding to the biomarker levels in such sample or samples. For example, as demonstrated herein, control samples with a known outcome can be used to determine a cut-off above which subjects are predicted to have an outcome (e.g. poor outcome) and below which subjects are predicted to have a different outcome (e.g. good outcome). Test samples are then compared to the predetermined value determined using control samples. The control can be an average, median, or calculated cut-off value (e.g. threshold) for each of 2 of BPGM and TAP2 levels (or in certain embodiments for the other biomarkers genes associated with the mars 2, 3 or 4 endotypes as described herein) and/or a composite thereof (e.g. sum) above or below which value a subject can be classified with an outcome class— e.g. good outcome or poor outcome. In embodiments calculating a signature score match (SSM) for example, the control is a selected value above which corresponds with an outcome and below which corresponds with another outcome. In certain methods, for example wherein the method of determining expression involves a Nanostring type assay, a relative or normalized expression is determined to one or more internal normalization genes (e.g. internal to the test sample) which are known and/or are determined to be suitable e.g. not vary significantly due to BC and/or from patient to patient. Control samples can be used to establish a fold increase relative to the normalization gene or genes. Accordingly, the control can be, for each biomarker, a ratio of the biomarker gene expression level and the level of one or more internal standardization markers in a control sample. The control ratio is compared to a corresponding ratio determined for the sample. For example, if the ratio of the biomarker gene and internal standardization marker in a control sample is 1, a ratio of 1.5, 2, 2.5 or more is indicative of increased expression and a ratio of 0.8, 0.5, 0.3 or less is indicative of decreased expression. The ratios can also be used to determine a cut off or threshold level or used in a SSM calculation. In such cases the control is a selected value above which is determined to predict one outcome and below which is determined to predict a different outcome.

The cut-off, threshold or control signature score can for example be a median level or value, or composite signature score comprising the median expression level or levels, for example the weighted expression levels, in a population of subjects. Following a larger clinical study, a cut-off or threshold can be determined to optimize the trade-off between false negative and false positive discoveries, for example by optimizing the area under the ROC curve. It may also be desirable to define multiple thresholds, for example to assign patients to high, medium, and low risk groups. The threshold(s) may be at any percentile of risk scores in the study sample, for example corresponding to the lowest 90%, 80%, 70%, 60%, 50%, 40%, 30%, 20% or 10% of risk scores calculated form histologically normal margins in a population of subjects. A person skilled in the art would understand that "control" as herein defined is distinct from for example a PCR control, no template PCR control or internal control, which is used for example with quantitative PCR. For example an internal control is a non-biomarker gene that is expected to be expressed at relatively the same level in different samples that is used to quantify the relative amount of biomarker transcript for comparison purposes.

The term "determining an expression level" or "determining an expression profile" as used in reference to a biomarker means the application of a biomarker specific reagent such as a probe, primer or antibody and/or a method to a sample, for example a sample of the subject and/or a control sample, for ascertaining or measuring quantitatively, semi-quantitatively or qualitatively the amount of a biomarker or biomarkers, for example the amount of biomarker polypeptide or mRNA. For example, a level of a biomarker can be determined by a number of methods including for example immunoassays including for example immunohistochemistry, ELISA, Western blot, immunoprecipitation and the like, where a biomarker detection agent such as an antibody for example, a labeled antibody, specifically binds the biomarker and permits for example relative or absolute ascertaining of the amount of polypeptide biomarker, hybridization and PCR protocols where a probe or primer or primer set are used to ascertain the amount of nucleic acid biomarker, including for example probe based and amplification based methods including for example microarray analysis, RT-PCR such as quantitative RT-PCR, serial analysis of gene expression (SAGE), Northern Blot, digital molecular barcoding technology, for example Nanostring:nCounter™ Analysis, and TaqMan quantitative PCR assays. Other methods of mRNA detection and quantification can be applied, such as mRNA in situ hybridization in formalin-fixed, paraffin-embedded (FFPE) tissue samples or cells. This technology is currently offered by the QuantiGene^®ViewRNA (Affymetrix), which uses probe sets for each mRNA that bind specifically to an amplification system to amplify the hybridization signals; these amplified signals can be visualized using a standard fluorescence microscope or imaging system. This system for example can detect and measure transcript levels in heterogeneous samples; for example, if a sample has normal and tumor cells present in the same tissue section. As mentioned, TaqMan probe-based gene expression analysis (PCR-based) can also be used for measuring gene expression levels in tissue samples, and for example for measuring mRNA levels in FFPE samples. In brief, TaqMan probe-based assays utilize a probe that hybridizes specifically to the mRNA target. This probe contains a quencher dye and a reporter dye (fluorescent molecule) attached to each end, and fluorescence is emitted only when specific hybridization to the mRNA target occurs. During the amplification step, the exonuclease activity of the polymerase enzyme causes the quencher and the reporter dyes to be detached from the probe, and fluorescence emission can occur. This fluorescence emission is recorded and signals are measured by a detection system; these signal intensities are used to calculate the abundance of a given transcript (gene expression) in a sample.

The term "difference in the level" as used herein in comparison to a control refers to a measurable difference in the level or quantity of a biomarker or biomarkers associated in a test sample, compared to the control that is of sufficient magnitude to allow assessment of predicted outcome, for example a significant difference or a statistically significant difference. The magnitude of the difference is sufficient for example to determine that the subject falls within a class of subjects likely to have poor survival prognosis or good survival prognosis. For example, a difference in a level of biomarker level is detected if a ratio of the level in a test sample as compared with a control is greater than 1.15 for example, a ratio of greater than 1.5, 1.7, 2, 3, 3, 5, 10, 12, 15, or more and/or a ratio less than 0.7, for example a ratio less than 0.6, 0.5, 0.4, 0.2, 0.1, 0.05 or more.

In a further embodiment of the invention said steps of determining the expression level of one or more genes comprise the use of at least one techniques selected from the group consisting of:

(a) PCR-based methods, preferably RT-PCR, Quantitative real Time PCR, nucleic acid microarray analysis, isothermal DNA/RNA amplification techniques, and/or in situ hybridization,

(b) immunological methods, preferably immunohistochemistry, ELISA, binding assays, and/or Western Blot,

(c) and/or spectroscpical methods, preferably Raman spectroscopy, and/or Mass spectroscopy.

The methods of the invention involves a step of measuring the expression level of certain genes.

Various techniques for determining the expression level of a gene are known in the art and can be used in conjunction with the present invention.

The expression level of a gene biomarker can be measured by the biomarker's mRNA level, protein level, activity level, or other quantity reflected in or derivable from the biomarker's gene or protein expression data. The expression products of each of the gene biomarkers of the invention include both RNA and protein. RNA products of the genes of the invention are transcriptional products of the genes of the invention and include populations of hnRNA, mRNA, and one or more spliced variants of mRNA. Protein products of the genes of the invention may also be measured. The protein products of the genes of the invention include, for example, proteins, protein variants arising from spliced mRNA variants, and post translationally modified proteins.

Any suitable means of measuring the expression of the RNA products of the genes of the invention can be used in accordance with the methods described herein. For example, the methods may utilize a variety of polynucleotides that specifically hybridize to one or more of the RNA products of the genes of the invention including, for example, oligonucleotides, cDNA, DNA, RNA, PCR products, synthetic DNA, synthetic RNA, or other combinations of naturally occurring of modified nucleotides which specifically hybridize to one or more of the RNA products of the genes of the invention. Such polynucleotides may be used in combination with the methods to measure RNA expression including, for example, array hybridization, RT-PCR, nuclease protection and northern blots.

In one embodiment, the expression level of the genes of the invention may be determined using array hybridization to evaluate the level of RNA expression. Array hybridization utilizes nucleic acid members stably associated with a support that can hybridize with genes of the invention expression products. The length of a nucleic acid member attached to the array can range from 8 to 1000 nucleotides in length and are chosen so as to be specific for the RNA products of the genes of the invention. The array may comprise, for example, one or more nucleic acid members that are specific for the RNA products of the genes of the invention, or variants thereof (e.g., splice variants). The nucleic acid members may be RNA or DNA, single or double stranded, and/or may be oligonucleotides or PCR fragments amplified from cDNA. Preferably oligonucleotides are approximately 10-100, 10-50, 20-50, or 20-30 nucleotides in length. Portions of the expressed regions of the genes of the invention can be utilized as probes on the array. More particularly oligonucleotides complementary to the genes of the invention genes and or cDNAs derived from the genes of the invention genes are useful. For oligonucleotide based arrays, the selection of oligonucleotides corresponding to the gene of interest, which are useful as probes is well understood in the art. More particularly it is important to choose regions which will permit hybridization to the target nucleic acids. Factors such as the Tm of the oligonucleotide, the percent GC content, the degree of secondary structure and the length of nucleic acid are important factors. See for example U.S. Pat. No. 6,551,784.

Arrays may be constructed, custom ordered, or purchased from a commercial vendor. Various methods for constructing arrays are well known in the art.

In certain embodiments, the level of the expression of the RNA products of the genes of the inventions can be measured by amplifying the RNA products of the biomarkers from a sample using reverse transcription (RT) in combination with the polymerase chain reaction (PCR). In certain embodiments, the RT can be quantitative as would be understood to a person skilled in the art.

Total RNA, or mRNA from a sample may be used as a template and a primer specific to the transcribed portion of a genes of the inventions is used to initiate reverse transcription. Methods of reverse transcribing RNA into cDNA are well known and are described, for example, in Sambrook et al., 1989, supra. Primer design can be accomplished utilizing commercially available software (e.g., Primer Designer 1.0, Scientific Software etc.) or methods that are standard and well known in the art. Primer Software programs can be used to aid in the design and selection of primers include, for example, The Primer Quest software which is available through the following web site link: biotools.idtdna.com/primerquest/. Additionally, the following website links are useful when searching and updating sequence information from the Human Genome Database for use in biomarker primer design:

1) the NCBI LocusLink Homepage: world wide web at ncbi.nlm.nih.gov/LocusLink/, and 2) Ensemble Human Genome Browser: world wide web at ensembl.org/Homo_sapiens, preferably using pertinent biomarker information such as Gene or Sequence Description, Accession or Sequence ID, Gene Symbol, RefSeq #, and/or UniGene #.

General guidelines for designing primers that may be used in accordance with the methods described herein include the following: the product or amplicon length may be ~100-150 bases, the optimum Tm may be ~60° C, or about 58-62° C, and the GC content may be ~50%, or about 45-55%. Additionally, it may be desirable to avoid certain sequences such as one or more of the following: (i) strings of three or more bases at the 3'-end of each primer that are complementary to another part of the same primer or to another primer in order to reduce primer-dimer formation, (ii) sequences within a primer that are complementary to another primer sequence, (iii) runs of 3 or more G's or C's at the 3'- end, (iv) single base repeats greater than 3 bases, (v) unbalanced distributions of G/C- and A/T rich domains, and/or (vi) a T at the 3'-end.

The product of the reverse transcription is subsequently used as a template for PCR. PCR provides a method for rapidly amplifying a particular nucleic acid sequence by using multiple cycles of DNA replication catalyzed by a thermostable, DNA-dependent DNA polymerase to amplify the target sequence of interest. PCR requires the presence of a nucleic acid to be amplified, two single-stranded oligonucleotide primers flanking the sequence to be amplified, a DNA polymerase, deoxyribonucleoside triphosphates, a buffer and salts. The method of PCR is well known in the art. PCR, is performed as described in Mullis and Faloona, 1987, Methods Enzymol., 155: 335.

QRT-PCR, which is quantitative in nature, can also be performed to provide a quantitative measure of genes of the invention gene expression levels. In QRT-PCR reverse transcription and PCR can be performed in two steps, or reverse transcription combined with PCR can be performed concurrently. One of these techniques, for which there are commercially available kits such as Taqman (Perkin Elmer, Foster City, Calif.), is performed with a transcript-specific antisense probe. This probe is specific for the PCR product (e.g. a nucleic acid fragment derived from a gene) and is prepared with a quencher and fluorescent reporter probe complexed to the 5' end of the oligonucleotide. Different fluorescent markers are attached to different reporters, allowing for measurement of two products in one reaction. When Taq DNA polymerase is activated, it cleaves off the quencher of the probe bound to the template by virtue of its 5'-to-3' exonuclease activity. In the absence of the quenchers, the reporters now fluoresce. The color change in the reporters is proportional to the amount of each specific product and is measured by a fluorometer; therefore, the amount of each color is measured and the PCR product is quantified. The PCR reactions are performed in 96 well plates so that samples derived from many individuals are processed and measured simultaneously. The Taqman system has the additional advantage of not requiring gel electrophoresis and allows for quantification when used with a standard curve.

A second technique useful for detecting PCR products quantitatively is to use an intercalating dye such as the commercially available QuantiTect SYBR Green PCR (Qiagen, Valencia Calif.). RT-PCR is performed using SYBR green as a fluorescent label which is incorporated into the PCR product during the PCR stage and produces a fluorescence proportional to the amount of PCR product. Additionally, other systems to quantitatively measure mRNA expression products are known including Molecular Beacons™. Additional techniques to quantitatively measure RNA expression include, but are not limited to, polymerase chain reaction, ligase chain reaction, Qbeta replicase (see, e.g., International Application No. PCT/US87/00880), isothermal amplification method (see, e.g., Walker et al. (1992) PNAS 89:382-396), strand displacement amplification (SDA), repair chain reaction, Asymmetric Quantitative PCR (see, e.g., U.S. Publication No. US200330134307A1) and the multiplex microsphere bead assay described in Fuja et al., 2004, Journal of Biotechnology 108:193-205.

The level of gene expression can be measured by amplifying RNA from a sample using transcription based amplification systems (TAS), including nucleic acid sequence amplification (NASBA) and 3SR. See, e.g., Kwoh et al (1989) PNAS USA 86:1173; International Publication No. WO 88/10315; and U.S. Pat. No. 6,329,179. In NASBA, the nucleic acids may be prepared for amplification using conventional phenol/chloroform extraction, heat denaturation, treatment with lysis buffer and minispin columns for isolation of DNA and RNA or guanidinium chloride extraction of RNA. These amplification techniques involve annealing a primer that has target specific sequences. Following polymerization, DNA/RNA hybrids are digested with RNase H while double stranded DNA molecules are heat denatured again. In either case the single stranded DNA is made fully double stranded by addition of second target specific primer, followed by polymerization. The double-stranded DNA molecules are then multiply transcribed by a polymerase such as T7 or SP6. In an isothermal cyclic reaction, the RNA's are reverse transcribed into double stranded DNA, and transcribed once with a polymerase such as T7 or SP6. The resulting products, whether truncated or complete, indicate target specific sequences.

Alternatively, visualization may be achieved indirectly. Following separation of amplification products, a labeled, nucleic acid probe is brought into contact with the amplified nucleic acid sequence of interest. The probe may be conjugated to a chromophore, radiolabeled, or conjugated to a binding partner, such as an antibody or biotin, where the other member of the binding pair carries a detectable moiety.

Additionally, detection may be carried our using Southern blotting and hybridization with a labeled probe. The techniques involved in Southern blotting are well known to those of skill in the art and may be found in many standard books on molecular protocols.

In certain embodiments, Nuclease protection assays (including both ribonuclease protection assays and SI nuclease assays) can be used to detect and quantitate RNA products of the genes of the inventions. In nuclease protection assays, an antisense probe (e.g., radiolabeled or nonisotopic labeled) hybridizes in solution to an RNA sample. Following hybridization, single-stranded, unhybridized probe and RNA are degraded by nucleases. An acrylamide gel is used to separate the remaining protected fragments. Typically, solution hybridization can accommodate up to ~100 μg of sample RNA whereas blot hybridizations may only be able to accommodate ~20-30 μg of RNA sample.

The ribonuclease protection assay, which is the most common type of nuclease protection assay, requires the use of RNA probes. Oligonucleotides and other single-stranded DNA probes can only be used in assays containing SI nuclease. The single-stranded, antisense probe must typically be completely homologous to target RNA to prevent cleavage of the probe:target hybrid by nuclease.

A standard Northern blot assay can also be used to ascertain an RNA transcript size, identify alternatively spliced RNA transcripts, and the relative amounts of RNA products of the genes of the inventions, in accordance with conventional Northern hybridization techniques known to those persons of ordinary skill in the art. In Northern blots, RNA samples are first separated by size via electrophoresis in an agarose gel under denaturing conditions. The RNA is then transferred to a membrane, crosslinked and hybridized with a labeled probe. Nonisotopic or high specific activity radiolabeled probes can be used including random-primed, nick-translated, or PCR-generated DNA probes, in vitro transcribed RNA probes, and oligonucleotides. Additionally, sequences with only partial homology (e.g., cDNA from a different species or genomic DNA fragments that might contain an exon) may be used as probes. The labeled probe, e.g., a radiolabeled cDNA, either containing the full-length, single stranded DNA or a fragment of that DNA sequence may be any length up to at least 20, at least 30, at least 50, or at least 100 consecutive nucleotides in length. The probe can be labeled by any of the many different methods known to those skilled in this art. The labels most commonly employed for these studies are radioactive elements, enzymes, chemicals that fluoresce when exposed to ultraviolet light, and others. A number of fluorescent materials are known and can be utilized as labels. These include, but are not limited to, fluorescein, rhodamine, auramine, Texas Red, AMCA blue and Lucifer Yellow. A particular detecting material is anti-rabbit antibody prepared in goats and conjugated with fluorescein through an isothiocyanate. Non-limiting examples of isotopes include 3H, 14C, 32P, 35S, 36CI, 51Cr, 57Co, 58Co, 59Fe, 90Y, 1251, 1311, and 186Re. Enzyme labels are likewise useful, and can be detected by any of the presently utilized colorimetric, spectrophotometric, fluorospectrophotometric, amperometric or gasometric techniques. The enzyme may be conjugated to the selected probe by reaction with bridging molecules such as carbodiimides, diisocyanates, glutaraldehyde and the like. Any enzymes known to one of skill in the art can be utilized, including, for example, peroxidase, beta-D-galactosidase, urease, glucose oxidase plus peroxidase and alkaline phosphatase. U.S. Pat. Nos. 3,654,090, 3,850,752, and 4,016,043 are referred to by way of example for their disclosure of alternate labeling material and methods.

The expression level of a genes of the invention may also be measured by the biomarker's protein level using any art-known method. Traditional methodologies for protein quantification include 2-D gel electrophoresis, mass spectrometry and antibody binding. Preferred methods for assaying biomarker protein levels in a biological sample include antibody-based techniques, such as immunoblotting (western blotting), immunohistological assay, enzyme linked immunosorbent assay (ELISA), radioimmunoassay (RIA), or protein chips. For example, a biomarker-specific monoclonal antibodies can be used both as an immunoadsorbent and as an enzyme-labeled probe to detect and quantify the biomarker. The amount of biomarker present in the sample can be calculated by reference to the amount present in a standard preparation using a linear regression computer algorithm. In another embodiment, genes of the inventions may be immunoprecipitated from a biological sample (e.g., directly from urine or serum or from a lysate of cells, etc.) using an antibody specific for said biomarker. The isolated proteins may then be run on an SDS-PAGE gel and blotted (e.g., to nitrocellulose or other suitable material) using standard procedures. The blot may then be probed with an anti- biomarker specific antibody to determine the expression level of the genes of the inventions.

Gel electrophoresis, immunoprecipitation and mass spectrometry may be carried out using standard techniques, for example, such as those described in Molecular Cloning A Laboratory Manual, 2nd Ed., ed. by Sambrook, Fritsch and Maniatis (Cold Spring Harbor Laboratory Press: 1989), Harlow and Lane, Antibodies: A Laboratory Manual (1988 Cold Spring Harbor Laboratory), G. Suizdak, Mass Spectrometry for Biotechnology (Academic Press 1996).

As used herein, the term "antibody" (Ab) or "monoclonal antibody" (mAb) is meant to include intact molecules as well as antibody portions (such as, for example, Fab, Fab', F(ab')2, Fv, single chain Fv, or Fd) which are capable of specifically binding to a genes of the invention.

In some embodiments, expression levels of genes of the inventions in a biological sample of interest (e.g., a blood sample) are compared to the expression level of said genes in an expression level reference sample. The expression level reference sample may be a biological sample derived from one or more patients determined to be suffering from sepsis. In other words, the expression level reference sample serves as a standard with which to compare expression level values for each genes of the invention in a test sample. An increase of the expression level of BPGM compared to the expression level values in a reference sample indicates that the patient has an increased risk of mortality from sepsis. An decrease of the expression level of TAP2 compared to the expression level values in a reference sample indicates that the patient has an increased risk of mortality from sepsis. In concordance with microarray discovery data, BPGM gene expression was significantly higher in sepsis patients; whereas, TAP2 gene expression was significantly lower in sepsis patients.

In some embodiments, genes of the invention threshold expression level values are optionally set based on one or more statistical criteria for deviation from the genes of the invention expression level values in an expression level reference sample, e.g., two or more SDs away from the value for a reference sample genes of the invention expression level.

In some embodiments, the expression level reference sample is a "negative" reference sample, i.e., a sample of a healthy individual.

In some embodiments, the expression level reference sample is a "positive" reference sample, i.e., a sample from a sepsis patient which has died from sepsis.

In some embodiments, genes of the invention expression profiles are compared to those in both positive and negative reference samples.

Methods for obtaining RNA from biological samples (e.g., tissues or cells) including linear RNA amplification from single cells include, e.g., Luzzi et al. (2005), Methods Mol. Biol., 293:187-207. Further, diverse kits for high quality RNA purification are available commercially, e.g., from Qiagen (Valencia, Calif.), Invitrogen (Carlsbad, Calif.), Clontech (Palo Alto, Calif.), and Stratagene (La Jolla, Calif.).

In a further embodiment the method further comprises the step of normalizing at least one of the determined gene expression levels. As used herein the term "normalization of the determined expression levels" refers to the process of removing error from measured data. Normalization can be carried out against an endogenous unregulated reference gene transcript or against total cellular DNA or RNA content (molecules/g total DNA/RNA and concentrations/g total DNA/RNA). For example when using quantitative real time PCR to determine gene expression level genes, which are largely unregulated are usually assessed in parallel with the target genes. These unregulated genes are termed housekeeping genes.

As used herein, the term "housekeeping gene" (sometimes also called "reference gene") refers to genes that usually code for proteins whose activities are essential for the maintenance of cell function. They are thus ubiquitous genes expressed in most organ, tissue and/or cell types of an organism in a mainly unregulated or only weakly regulated fashion, or regulated to a constant gene expression rate. Housekeeping genes include, without limitation, glyceraldehyde-3 -phosphate dehydrogenase (GAPDH), Cypl, albumin, actins, e . g . β-actin, tubulins, cyclophilin, hypoxantine phsophoribosyltransferase 1 (HPRTl), Ribosomal protein L32, 28S, 18S, large ribosomal protein (RPLPO), Transferrin receptor (TFRC) and beta-glucuronidase (GUS). In a highly preferred embodiment, HPRTl is used, as it was the most stable. Other "housekeeper genes" were found less useful, as they were significantly altered due to sepsis.

In embodiments using array techniques of determining said expression level, other corrections and/or data processing may suitably be applied, including but not limited to background signal intensity correction, for example robust multi-average, normalization, for example using quantiles, and summarization, for example median polish and log transformation. Such corrections are especially useful when using high-throughput gene expression microarrays, Advantages are that these corrections/processing enable for a drastic reduction in systematic technical noise that would otherwise impinge on proper differential gene expression.

A further preferred embodiment comprises the step of checking whether or not the expression level of said gene is higher than a predetermined threshold level. As used herein, the term "threshold level" refers to a level of gene expression above a certain point as determined by, for example, the receiver operator characteristic curve (employed in the derivation and validation tests herein described)or calibrator samples in a qPCR, preferably in a point-of-care qPCR test. The "threshold value" is subject to the technology utilized to run the test, hence it will need to be derived afresh when the TAP2:BPGM gene expression biomarker is applied to other point-of-care testing devices.

In a further embodiment the method comprises the step of comparing, arithmetically, the ratio of the expression levels of TAP2 and BPGM with a reference control or reference value and determining the risk dying from sepsis based on said ratio. Using the said ratio also has the advantage of relativity thereby there is no need for normalization, for example, to a "housekeeper gene". The ratio of the expression levels of TAP2 and BPGM had threshold-independent ROC AUC of 0.845 (95% CI: 0.764- 0.917). A numerical threshold for the ratio of the expression levels of TAP2 and BPGM is preferably defined at 1.15.

The above disclosure generally describes the present invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as those commonly understood by one of ordinary skill in the art to which this invention belongs. A more complete understanding can be obtained by reference to the following specific examples which are provided herein for purposes of illustration only, and are not intended to limit the scope of the invention.

EXAMPLE

Patients

The study was performed within the context of the Molecular Diagnosis and Risk Stratification of Sepsis (MARS) project, a prospective observational study in the mixed ICUs of two tertiary teaching hospitals (Academic Medical Center in Amsterdam and University Medical Center in Utrecht) in the

Netherlands.⁵'¹²'¹³ All patients above 18 years of age admitted to the two ICUs between January 2011 and July 2012 with an expected length of stay longer than 24 hours were included via an opt-out method approved by the medical ethical committees of the participating hospitals. ^{5 12 13} For every admitted patient the plausibility of an infection was assessed in retrospect using a four point scale (ascending from none, possible, probable to definite) using Center for Disease Control and Prevention¹⁴ and International Sepsis Forum consensus definitions¹⁵, as described in detail.¹² The current study comprised consecutive patients admitted to the ICU with sepsis defined as the presence of infection with a probable or definite likelihood, accompanied by at least one additional parameter as described in the 2001 International

Sepsis Definitions Conference.¹⁶ Patients admitted in Amsterdam were used as discovery cohort; those admitted in Utrecht as first validation cohort. 42 healthy subjects (age 35 (30-63) years, median with interquartile ranges; 57% male) were also enrolled after providing written informed consent. The second validation cohort was from the United Kingdom (UK) Genomic Advances in Sepsis (GAinS) study of adult patients with sepsis due to community-acquired pneumonia (sepsis)¹⁷. Pediatric sepsis patients derived from a prospective observational study of children < 10 years old admitted to multiple pediatric ICUs in the United States were used as a comparative cohort.¹⁸ Blood RNA and microarrays

Of patients enrolled in MARS blood was collected in PAXgene blood RNA tubes (Becton-Dickinson, Breda, The Netherlands) within 24 hours of ICU admission. Gene expression profiles were analyzed using Human Genome U219 96-array plates and the GeneTitanR instrument (Affymetrix) as described.⁵'¹³ MARS gene expression data are available in the Gene Expression Omnibus under accession number GSE65682. Gene expression data for UK GAinS (ArrayExpress accession number E-MTAB-4421) were generated using lllumina Human-HT-12 version 4 Expression BeadChips¹⁷. Gene expression data of pediatric sepsis patients (GSE13904) were generated using the Affymetrix Human Genome U133 Plus 2.0 Array¹⁸.

Unsupervised clustering and classifier derivation

For endotype discovery (Figure 2.1), probes were ranked by median absolute deviation across 306 patient samples (discovery cohort). The top 5000 ranked probes were selected and analyzed by means of the consensus clustering method^19,20. We selected the agglomerative hierarchical clustering algorithm on 1-Pearson correlation distances, 99% item (sample) resampling, 1000 iterations and cluster range k = 2 to k = 12. To estimate k (number of endotypes) we combined cumulative distribution functions,^19,20 silhouette width analysis²¹ available in the cluster package²² and cophenetic distance correlation analysis to evaluate clustering stability.²³ To construct the k endotype classifier we selected patient samples with positive silhouette widths, representing core patients per endotype.^19,21 The 5000 probes were subsequently ranked by non-parametric (kruskal-wallis rank sum test) significance. 2994 unique gene probes were filtered by selecting for highest significance. Using a random forest classifier²⁴ (supervised classification with high dimensional data methods)²⁵, we evaluated sepsis endotype classification with 10-fold cross-validation of step-wise increments in gene numbers. We settled on the number of genes that yielded a cross-validation misclassification error rate < 10%. The sepsis endotype classifier gene set was then used to perform random forest prediction of endotypes in the validation cohorts.

Derivation of the molecular endotype biomarkers

Endotype biomarkers were assessed using previously described methods ^{5 6}.

Differential gene expression and Ingenuity pathway analysis Statistics

Statistical analysis was performed using the R statistical computing environment (version 3.1.2; R Foundation for Statistical Computing, Vienna, Austria; http://www.R-proiect.org/). The Cramer's V measure of effect size was used for a chi-square goodness of fit test. Correlation analysis of continuous data was performed using Spearman's method. Survival analysis was performed by Kaplan-Meier estimation (log-rank test) and Cox proportional hazards regression implemented in the survival method (version 2.37). For the latter we focused on the high-risk Marsl endotype with reference to the other endotypes (Mars2, Mars3 and Mars4 endotypes were recoded to a single group). Net reclassification improvement was assessed by means of a continuous model using the predictABEL method (version 1.2- 2)²⁶. One model encompassed only Acute Physiology and Chronic Health Evaluation (APACHE) IV scores²⁷ (clinical), while a second model encompassed both APACHE IV scores and sepsis endotype stratification (clinical + molecular). Unless otherwise stated, significance was demarcated at p < 0.05.

Derivation of the molecular endotype biomarkers

Sepsis molecular endotype biomarkers were derived by using previously described methods.1,2 The 140 gene expression indices that encompassed the endotype classifier were assessed for the best combination that classified the discovery cohort. In this context, a combination was a two-gene expression ratio (score):

Score = gene, / gene_j

Receiver operator characteristics (ROC) were assessed by using the pROC R package (version 1.5.4), with bootstrap resampled 95% confidence intervals.3 Thresholds were derived by using the coordinates function in the pROC method, specifying the "best" coordinate along the ROC curve. Considering the multi-combinatorial strategy (140 x 139 [gene ratios] x 4 [endotypes] = 77,840 tests) significance was defined by Benjamini-Hochberg4 adjusted p < 0.05 (adjusting for the 77,840 tests).

Differential gene expression and Ingenuity pathway analysis

Differential gene expression analysis was firstly performed by comparing patients stratified into each of four molecular endotypes to healthy subjects, and secondly by comparing each endotype to the other endotypes. For example the latter, Marsl patient gene expression data were compared to "others", where Mars2, Mars3 and Mars4 endotypes were recoded to a single group (others). These supervised analyses were done by means of moderated t tests implemented in the limma method (version

3.14.4)1,5-7. Throughout Benjamini-Hochberg multiple comparison adjusted probabilities4 (adjusted p <

0.05. defined significance. Ingenuity Pathway Analysis (Ingenuity Systems IPA, www.ingenuity.com) was used to identify enrichment of genes that pertain to distinct canonical signaling pathways. The Ingenuity gene knowledgebase was selected as reference and human species specified. All other parameters were default. Significance was evaluated by Fisher's exact test adjusted p-values (adjusted p < 0.01).

REFERENCES

1. Angus DC, van der Poll T. Severe Sepsis and Septic Shock. New England Journal of Medicine.

2013;369(9):840-851.

2. Liu V, Escobar GJ, Greene J D, et al. Hospital deaths in patients with sepsis from 2 independent cohorts.

Jama. 2014;312(l) :90-92.

3. Cohen J, Vincent J L, Adhikari N K, et al. Sepsis: a roadmap for future research. The Lancet. Infectious

diseases. 2015;15(5):581-614.

4. Maslove DM, Wong HR. Gene expression profiling in sepsis: timing, tissue, and translational

considerations. Trends in molecular medicine. 2014;20(4):204-213.

5. Scicluna BP, Klein Klouwenberg PM, van Vught LA, et al. A molecular biomarker to diagnose community- acquired pneumonia on intensive care unit admission. Am J Respir Crit Care Med. 2015;192(7) :826-835.

6. McHugh L, Seldon TA, Brandon RA, et al. A Molecular Host Response Assay to Discriminate Between Sepsis and Infection-Negative Systemic Inflammation in Critically III Patients: Discovery and Validation in Independent Cohorts. PLoS Med. 2015;12(12) :el001916.

7. Sweeney TE, Shidham A, Wong HR, Khatri P. A comprehensive time-course-based multicohort analysis of sepsis and sterile inflammation reveals a robust diagnostic gene set. Science translational medicine. 2015;7(287):287ra271.

8. Parlato M, Cavaillon J M. Host response biomarkers in the diagnosis of sepsis: a general overview. Methods in molecular biology. 2015;1237:149-211.

9. Sweeney TE, Wong HR. Risk Stratification and Prognosis in Sepsis: What Have We Learned from

Microarrays? Clinics in chest medicine. 2016;37(2) :209-218.

10. Curtis C, Shah SP, Chin SF, et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature. 2012;486(7403) :346-352.

11. Guinney J, Dienstmann R, Wang X, et al. The consensus molecular subtypes of colorectal cancer. Nature medicine. 2015;21(11) :1350-1356.

12. Klein Klouwenberg PM, Ong DSY, Bos LDJ, et al. Interobserver Agreement of Centers for Disease Control and Prevention Criteria for Classifying Infections in Critically III Patients. Critical Care Medicine.

2013;41(10):2373-2378.

13. van Vught LA, Klein Klouwenberg PM, Spitoni C, et al. Incidence, Risk Factors, and Attributable Mortality of Secondary Infections in the Intensive Care Unit After Admission for Sepsis. Jama. 2016;315(14) :1469-1479.

14. Garner JS, Jarvis WR, Emori TG, Horan TC, Hughes J M. Cdc Definitions for Nosocomial Infections, 1988.

American Journal of Infection Control. 1988;16(3):128-140.

15. Calandra T, Cohen J . The international sepsis forum consensus conference on definitions of infection in the intensive care unit. Critical Care Medicine. 2005;33(7):1538-1548.

16. Levy M M, Fink M P, Marshall JC, et al. 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference. Crit Care Med. 2003;31(4):1250-1256.

17. Davenport EE, Burnham KL, Radhakrishnan J, et al. Genomic landscape of the individual host response and outcomes in sepsis: a prospective cohort study. The Lancet. Respiratory medicine. 2016;4(4) :259-271. 18. Wong HR, Cvijanovich N, Allen GL, et al. Genomic expression profiling across the pediatric systemic inflammatory response syndrome, sepsis, and septic shock spectrum. Crit Care Med. 2009;37(5):1558- 1566.

19. Verhaak RG, Hoadley KA, Purdom E, et al. Integrated genomic analysis identifies clinically relevant

subtypes of glioblastoma characterized by abnormalities in PDGFRA, I DH1, EGFR, and N F1. Cancer Cell.

2010;17(1) :98-110.

20. Wilkerson M D, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics. 2010;26(12):1572-1573.

21. Rousseeuw PJ. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics. 1987;20:53-65.

22. Maechler M, Rousseeuw, P., Struyf, A., Hubert, M., Hornik, K. cluster: Cluster Analysis Basics and

Extensions. . R package version 1.15.3. 2014.

23. Brunet J P, Tamayo P, Golub TR, Mesirov J P. Metagenes and molecular pattern discovery using matrix factorization. Proc Natl Acad Sci U S A. 2004;101(12) :4164-4169.

24. Breiman L Random forests. Machine Learning. 2001;45:5-32.

25. Slawski M, Daumer M, Boulesteix AL. CMA: a comprehensive Bioconductor package for supervised

classification with high dimensional data. BMC Bioinformatics. 2008;9:439.

26. Kundu S, Aulchenko YS, van Duijn CM, Janssens AC. PredictABEL: an R package for the assessment of risk prediction models. European journal of epidemiology. 2011;26(4) :261-264.

27. Zimmerman J E, Kramer AA, McNair DS, Malila FM. Acute Physiology and Chronic Health Evaluation (APACHE) IV: hospital mortality assessment for today's critically ill patients. Crit Care Med.

2006;34(5) :1297-1310.

28. De Sousa EM F, Wang X, Jansen M, et al. Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions. Nature medicine. 2013;19(5) :614-618.

29. Vincent J L, Moreno R, Takala J, et al. The SOFA (sepsis-related organ failure assessment) score to describe organ dysfunction/failure. Intensive Care Medicine. 1996;22(7):707-710.

30. Pencina MJ, D'Agostino RB, Sr., D'Agostino RB, Jr., Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Statistics in medicine.

2008;27(2) :157-172; discussion 207-112.

31. Pollack M M, Ruttimann U E, Getson PR. Pediatric risk of mortality (PRISM) score. Crit Care Med.

1988;16(11):1110-1116.

32. Maslove DM, Tang BM, McLean AS. Identification of sepsis subtypes in critically ill adults using gene expression profiling. Crit Care. 2012;16(5) :R183.

33. Wheeler DS, Wong HR, Zingarelli B. Pediatric Sepsis - Part I: "Children are not small adults!". The open inflammation journal. 2011;4:4-15.

34. Prabhu Das M, Adkins B, Gans H, et al. Challenges in infant immunity: implications for responses to

infection and vaccines. Nature immunology. 2011;12(3) :189-194.

35. Amelio I, Cutruzzola F, Antonov A, Agostini M, Melino G. Serine and glycine metabolism in cancer. Trends in biochemical sciences. 2014;39(4) :191-198.

36. Cheng SC, Scicluna BP, Arts RJ, et al. Broad defects in the energy metabolism of leukocytes underlie immunoparalysis in sepsis. Nature immunology. 2016;17(4):406-413.

37. Hotchkiss RS, Monneret G, Payen D. Sepsis-induced immunosuppression : from cellular dysfunctions to immunotherapy. Nature Reviews Immunology. 2013;13(12) :862-874.

38. Marshall JC. Why have clinical trials in sepsis failed? Trends in molecular medicine. 2014;20(4):195-203. 39. Angus DC. The search for effective therapy for sepsis: back to the drawing board? Jama.

2011;306(23) :2614-2615.

40. van der Poll T. Future of sepsis therapies. Crit Care. 2016;20(1):106. Table 1.1. Baseline characteristics and mortality of patients in the discovery and validation cohorts

Clinical characteristics of patients included in the discovery and validation cohorts. Abbreviations: Q1 -Q3, st quartile- 3rd quartile. APACHE, Acute Physiology and Chronic Health Evaluation. SOFA, sequential organ failure assessment. COPD, chronic obstructive pulmonary disease. * Other includes: bone joint infection, endocarditis, mediastinitis, myocarditis, ear infection, oral infection, pharyngitis, post-operative wound infection and lung abscess.† APACHE IV score. $ APACHE II score. Table 2.1 : Characteristics of the discovery cohort classified as four molecular endotypes

Clinical characteristics of discovery cohort stratified by sepsis molecular endotypes. Abbreviations: Q1 -Q3, 1 st quartile-3rd quartile. APACHE IV, Acute Physiology and Chronic Health Evaluation IV. SOFA, sequential organ failure assessment. COPD, chronic obstructive pulmonary disease. ^* Other includes: bone joint infection, endocarditis, mediastinitis, myocarditis, ear infection, oral infection, pharyngitis, post-operative wound infection and lung abscess.

Table 2.2: Characteristics of the first validation cohort classified to four molecular endotypes

Clinical characteristics of first validation cohort stratified by sepsis molecular endotypes. Abbreviations: Q1 -Q3, 1 st quartile-3rd quartile. APACHE IV, Acute Physiology and Chronic Health Evaluation IV. SOFA, sequential organ failure assessment. COPD, chronic obstructive pulmonary disease. * other includes: bone joint infection, endocarditis, mediastinitis, myocarditis, ear infection, oral infection, pharyngitis, post-operative wound infection and lung abscess.

Table 2.3: Pediatric sepsis cohort characteristics

Clinical characteristics of the pediatric sepsis cohort, USA. Abbreviations: Q1 -Q3, 1 st quartile-3rd quartile. PRISM, pediatric risk of mortality.

Table 2.4: Performance characteristics of the candidate sepsis endotype biomarkers in the discovery cohort

5

*threshold-dependent scores (95% confidence intervals), ppv, positive predictive value, npv, negative predictive value. LR+, positive likelihood ratio. LR-, negative likelihood ratio. BPG M, bisphosphoglycerate mutase. TAP2, transporter 2, ATP binding cassette subfamily B member. GADD45A, growth arrest and DNA damage inducible alpha. PCG F5, polycomb group ring finger 5. AHNAK, AH NAK nucleoprotein. PDCDIO, programmed cell death 10. I FIT5, interferon induced protein with tetratricopeptide repeats 5. GLTSCR2, glioma tumor suppressor candidate region gene 2.

references

1. Scicluna BP, Klein Klouwenberg PM, van Vught LA, et al. A molecular biomarker to diagnose community-acquired pneumonia on intensive care unit admission. Am J Respir Crit Care Med. 2015;192(7):826-835.

2. McHugh L, Seldon TA, Brandon RA, et al. A Molecular Host Response Assay to Discriminate

Between Sepsis and Infection-Negative Systemic Inflammation in Critically III Patients: Discovery and Validation in Independent Cohorts. PLoS Med. 2015;12(12):el001916.

3. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and S plus to analyze and compare ROC curves. BMC Bioinformatics. 2011;12.

4. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate - A Practical and Powerful

Approach to Multiple Testing. Journal of the Royal Statistical Society Series B-Methodological. 1995;57(l):289-300.

5. Smyth GK. Limma: linear models for microarray data. In: R G, VJ C, H W, RA I, S D, eds.

Bioinformatics and Computational Biology Solutions using R. : Springer; 2005:397-420.

6. van Lieshout MH, Scicluna BP, Florquin S, van der Poll T. NLRP3 and ASC Differentially Affect the Lung Transcriptome during Pneumococcal Pneumonia. Am J Respir Cell Mol Biol. 2014;4(50):699- 712.

7. Scicluna BP, van Lieshout MH, Blok DC, Florquin S, van der Poll T. Modular Transcriptional

Networks of the Host Pulmonary Response during Early and Late Pneumococcal Pneumonia. Molecular medicine. 2015;21:430-441.

Claims

1. Method for determining the survival prognosis of a patient admitted to an intensive care unit (ICU) or a sepsis patient, comprising steps of determining the expression level of at least BPGM and/or TAP2 in a biological sample of said patient, and compare said expression level(s) with a control and, determine said survival prognosis based on said comparison, wherein said biological sample contains white blood cells.

2. Method according to claim 1, wherein said step of determining the expression level of BPGM and/or TAP2 comprises the use of at least one techniques selected from the group consisting of: i. PCR-based methods, preferably

RT-PCR,

Quantitative RT-PCR,

nucleic acid microarray analysis,

isothermal DNA/RNA amplification techniques, and/or

in situ hybridization,

ii. immunological methods, preferably

immunohistochemistry.

ELISA, binding assays, and/or

Western Blot,

iii. and/or spectroscopical methods, preferably

Raman spectroscopy, and/or

Mass spectroscopy.

3. Method according to claim 1 or 2, comprising comparing said expression level(s) with an

expression level of a reference control or reference value.

4. Method according to any one of claims 1-3, comprising the step of checking whether or not the expression level of BPGM and/or TAP2 is higher or lower than a predetermined threshold level.

5. Method according to any one of claims 1-4, wherein said expression levels are normalized, preferably based on HPRT1 expression.

6. Method according to any one of claims 1-5, comprising determining the ratio of the expression levels of the of BPGM and TAP2 and determining survival prognosis based on said ratio.

7. Method according to claim 6, wherein the ratio of the expression levels of BPGM and TAP2 is compared arithmetically with the ratio of the expression levels of BPGM and TAP2 in a reference control or reference value and wherein the survival prognosis is determined if said ratio is higher than said reference control ratio or reference value.

8. Method according to any one of claims 1-7, wherein a BPGM:TAP2 gene expression ratio is 1.15 or higher is associated with a poor survival prognosis.

9. Method according to any one of claims 1-8, wherein said biological sample is a blood sample.

10. Method according to any one of claims 1-9, wherein said patient is a human.

11. Method according to any one of claims 1-10, wherein the survival prognosis is further based on a parameter associated with hospital mortality.

12. Method according to claim 11, wherein the parameter associated with hospital mortality is the APACHE IV score or Sequential Organ Failure Assessment (SOFA) score.

13. Method according to any one of claims 1-12, wherein said patient is an adult.

14. Method according to any one of claims 1-13, wherein the expression level of at least one

biomarker gene is determined selected from the group consisting of: GADD45A, PCGF5, AH NAK, PDCD10, IFIT5 and GLTSCR2.

15. Method according to any one of claims 1-14, wherein the ratio of the expression levels is

determined of at least one of the following biomarker gene pairs:

i. GADD45A and PCGF5,

ii. AHNAK and PDCD10, and

iii. IFIT5 and GLTSCR2.