METHODS AND ASSOCIATED USES, KITS AND SYSTEM FOR ASSESSING SEPSIS

Abstract

The invention relates to protein biomarkers representing protein biomarker signatures to assess a patient who may develop sepsis, or who may have developed sepsis. The invention relates in particular to methods for assessment or monitoring with respect to diagnosis, prediction or progression of sepsis in a patient, as well as the responsiveness to, or selection of suitable agents for, the treatment of sepsis. The invention also relates to the use of protein biomarkers representing protein biomarker signatures for sepsis, and associated kits and system.

Claims

1. A method for analyzing a biological sample, obtained from a patient, to assess whether the patient may develop sepsis or to diagnose the patient as having sepsis, the method comprising the steps of: a. determining in the biological sample individual levels of protein biomarkers representing a protein biomarker signature; and, b. using the individual levels of the protein biomarkers collectively to assess whether the patient may develop sepsis or to diagnose a patient as having sepsis, wherein the protein biomarkers of the protein biomarker signature comprise at least four of CCL-16, CD28, CD244, FGF21, GALNT3, GT, IL-18BP, JAM-A, LDL-R, LILRB5, LTBR, MCP-2, MMP-1, NUCB2, SIGLEC10, TNF-R1, TNF-R2, TNFRSF10A, TNFRSF11A, TNFRSF14, TRAILR2 and UPAR.

2. The method according to claim 1, wherein the protein biomarker signature comprises CCL-16 and MCP-2.

3. The method according to claim 2, wherein the protein biomarker signature consists of LTBR, CCL-16, CD28, FGF21 and MCP-2.

4. The method according to claim 2, wherein the protein biomarker signature further comprises GALNT3, GT, LDL-R, LILRB5 and MMP-1.

5. The method according to claim 4, wherein the protein biomarker signature further comprises FGF21.

6. The method according to claim 5, wherein the protein biomarker signature consists of CCL16, CD28, FGF21, GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1 and TNFRSF11A.

7. The method according to claim 5, wherein the protein biomarker signature consists of CCL16, CD244, FGF21, GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1 and TNF-R1.

8. The method according to claim 5, wherein protein biomarker signature consists of CCL16, CD28, FGF21, GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1 and TNF-R1.

9. The method according to claim 5, wherein the protein biomarker signature consists of CCL16, CD28, FGF21, GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1 and one of U-PAR or TRAIL-R2.

10. (canceled)

11. The method according to claim 1, wherein the protein biomarker signature further comprises at least one additional biomarker from a list consisting of PCT, lactate, CRP, D-Dimer and PSP.

12. A method for analyzing biological samples, obtained from a patient at risk of, or having developed, sepsis, to monitor the patient, the method comprising the steps of: a. determining in the biological samples, obtained from the patient at a plurality of time points, individual levels of protein biomarkers representing a protein biomarker signature, wherein the protein biomarker signature comprises the biomarkers selected according to the method of claim 1; and b. using changes in the individual levels of the protein biomarkers collectively, across the plurality of time points, to monitor the patient and to predict whether the patient may develop sepsis, or to monitor the progression of sepsis in the patient.

13. A method for analyzing biological samples, obtained from a patient predicted or diagnosed as having sepsis, to monitor the responsiveness of the patient to treatment with an antimicrobial agent and/or immunosuppressive agent, the method comprising the steps of: a. determining in a sample, obtained from the patient at a plurality of time points, individual levels of biomarkers representing a protein biomarker signature, wherein the protein biomarker signature comprises the biomarkers selected according to the method of claim 1; and b. using changes in the individual levels of the biomarkers collectively, across the plurality of time points, to monitor the responsiveness of a patient to treatment with the antimicrobial agent and/or immunosuppressive agent.

14. A method for selecting a therapeutic agent and/or immunosuppressive agent for administration to a patient predicted or diagnosed as having sepsis, the method comprising the steps of: a. determining in a sample, obtained from the patient at a time point or plurality of time points, individual levels of biomarkers representing a protein biomarker signature, wherein the protein biomarker signature comprises the biomarkers selected according to the method of claim 1; and b. using the individual levels of the biomarkers, or the changes in the individual levels of the biomarkers collectively across the plurality of time points, to select the therapeutic agent and/or immunosuppressive agent.

15. (canceled)

16. A kit for implementing at least step a) of the method of claim 1, wherein the kit comprises a labelled reagent or a plurality of labelled reagents for detecting individual levels of each protein biomarker in a protein biomarker signature, in at least one sample taken from the patient, wherein the labelled reagent or reagents is/are capable of binding specifically to each protein biomarker selected according to the method of claim 1.

17. The kit according to claim 16, wherein the labelled reagents are antibody-based.

18. The kit according to claim 16, further comprising a test element to which the labelled reagents are, or are capable of being, incorporated or applied.

19. The kit according to claim 18, wherein the test element is a lateral flow device.

20. The kit according to claim 18, wherein the test element is a protein array.

21. The kit according to claim 16, wherein the kit further comprises an anticoagulant.

22. A system comprising: i. the kit of claim 16; j. a detector for monitoring, measuring or detecting the individual levels of the protein biomarkers; and k. a computer processor configured to analyse data produced by the detector.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0069] FIG. 1 is an illustration depicting the rationale for sample selection, including the selection of control samples, and the matching with sepsis patient samples;

[0070] FIG. 2 is a graph of the proportion of samples within the sepsis group outside the 90% quantiles of the SIRS and comparator groups for each of 718 protein analytes across time points; and

[0071] FIG. 3 is a dendrogram of the relatedness of protein analytes by cluster analysis.

DETAILED DESCRIPTION

[0072] The invention provides a method for analysing a biological sample, obtained from a patient, to assess the patient for sepsis, the method comprising the steps of: [0073] a. determining in the biological sample individual levels of biomarkers representing a protein biomarker signature; and [0074] b. using the individual levels of the biomarkers collectively to assess the patient by predicting or diagnosing sepsis, wherein the biomarkers of the protein biomarker signature comprises at least four biomarkers from a list consisting of CCL-16, CD28, CD244, FGF21, GALNT3, GT, IL-18BP, JAM-A, LDL-R, LILRB5, LTBR, MCP-2, MMP-1, NUCB2, SIGLEC10, TNF-R1, TNF-R2, TNFRSF10A, TNFRSF11A, TNFRSF14, TRAILR2 and UPAR.

[0075] Various investigations have been carried out, as described below, to determine the predicative accuracy of a series of protein biomarker signatures to predict sepsis, according to the Sepsis-3 definition, at time points that include the day prior to (Day −1) and day of (Day 0) sepsis diagnosis.

[0076] Methods

[0077] Study Inclusion Criteria

[0078] The study recruited 4385 elective surgery patients. Patients were admitted to the study if they gave informed consent, were between 18 and 80 years of age and undergoing a procedure that, in the clinician's opinion, had a risk of causing infection and ultimately sepsis. Typically, these were abdominal and thoracic surgeries. However, other surgical procedures were permitted and included, such as an extensive maxillofacial procedure that resulted in sepsis in one case. Patients were excluded if they were either pregnant, infected with a known pathogen (HIV, Hepatitis A, B or C), immunosuppressed or withdrew consent to take part in the study at any time during their stay. All patients received the normal standard of care once enrolled.

[0079] Acquisition and Storage of Patient Samples

[0080] Blood samples were collected according to an ethically-approved protocol. Briefly, a 4 ml aliquot of patient blood was separately collected into a sterile serum separation tube. Following centrifugation, the serum was pipetted into an appropriately sized vial. All samples were then stored at −20° C. and eventually transported on dry ice. Blood collection occurred once between 1 and 7 days before surgery and then once daily on each day post-surgery. Post-operative blood collection was stopped after the patient was discharged from hospital, or after 7 days post-surgery, or once the clinician had confirmed sepsis. Additional patient information (e.g. daily patient metrics, type of surgery and microbiology results) was captured using a bespoke database provided by ItemTracker, UK. All samples collected from patients were stored at Dstl in suitably alarmed freezers that were monitored daily.

[0081] Clinical Adjudication

[0082] A Clinical Advisory Panel (CAP), comprising experts from across the UK and Germany, was tasked to provide a definitive judgement on whether a patient had developed sepsis according to the Sepsis-2 criteria. Using a blinded elicitation approach, all relevant patient data was presented to them and a silent vote was conducted. The results of this process were captured by a facilitator whose role was to ensure that no conferring had occurred and record the clinical opinion. If a consensus of opinion for a sepsis patient was achieved, then the clinicians were asked to indicate the day of sepsis diagnosis (without conferring). If consensus was again achieved then the facilitator moved to the next patient. If no consensus was reached, either for patient outcome or on day of sepsis diagnosis, then clinicians were allowed to discuss the reasons for their mixed opinions. Following a discussion, the clinicians were asked to re-vote. Key points from the discussion as well as subsequent voting that led to a consensus of opinion or a majority opinion was recorded by the facilitator for both patient classification as well as day of diagnosis for sepsis patients. It should be noted that the order of voting was sometimes randomized to mitigate the effect of peer pressure by key clinicians. Voting data was analysed using Kappa statistics to quantify the level of agreement achieved by the CAP. It was anticipated that a high level of agreement by a panel of clinical experts would give high confidence in patient classification and subsequent biomarker selection. For this study the level of agreement reached for patient classification was high.

[0083] Further analysis was undertaken to understand what proportion of the sepsis patient cohort chosen by the CAP using Sepsis-2 criteria conformed to the new Sepsis-3 definition. The former relies on the presence of SIRS caused by a microbial agent. The latter relies on organ dysfunction, as measured by changes in the SOFA score (>2), to indicate a “bad infection” that is associated with organ dysfunction.

[0084] Following clinical adjudication, 155 elective surgery patients were judged to have developed sepsis, defined according to the Sepsis-2 definition, during the study. The incidence of sepsis in the patient cohort was therefore 3.53%. Of this Sepsis-2 cohort, 98 patients were judged to have fulfilled the Sepsis-3 criteria. For all Sepsis-2 (only) and Sepsis-3 patients, age/sex/procedure-matched comparators from the cohort of patients that either developed SIRS or who had an unremarkable recovery were selected.

[0085] The rationale for comparator selection is illustrated in FIG. 1, along with which patient samples were analysed and how the timeframes for patient samples taken at different days post-surgery were standardized. The time course of the development of sepsis in a patient is indicated by the Sepsis patient #1 bar. From the large number of patients who did not go on to develop sepsis following surgery, a suitable age/sex/procedure-matched control is identified and used as a comparator. In this example, the day of diagnosis of sepsis is day 7 post-infection. Therefore, the 3 days before sepsis diagnosis are days 4, 5 and 6 post-surgery. In terms of pre-symptomatic diagnosis, this may also be noted as Days −3, −2 and −1. In order to provide a robust and relevant post-operative comparison for each of the 3 days before sepsis diagnosis, the equivalent post-operative blood sample from the age/gender/procedure-matched comparator was used. In this case, the blood samples taken from days 4, 5 and 6 post surgery were used for comparison, acting as Day −3, −2 and −1 controls. The process of matching the pre-symptomatic blood samples of patients who went on to develop sepsis with their most appropriate post-operative comparators was then repeated for all sepsis patients. Table 2 summarises a series of top-level characteristics for patients involved in the study.

TABLE-US-00002 TABLE 2 Summary of patient ages, gender, delay for sepsis and types of surgery Sepsis Controls SIRS n = 50 n = 50 n = 49 Median 65 64.5 66 age (IQR) (58.5-74) (56.75-73) (54.5-73) Gender 7/43 5/45 6/43 (female/male) Median day of 3 n/a n/a sepsis diagnosis (2-4) (IQR) Median SOFA 4 0 0 score on day of (0-7) (0-0) (0-0) sepsis diagnosis (or equivalent day post-surgery)

[0086] O-Link Analysis

[0087] Analysis of patient proteome was conducted on samples from 50 patients who went on to develop sepsis using the O-link array platform (O-link Proteomics, Uppsala, Sweden). Additional samples from age/gender/procedure-matched control patients (n=50) and patients who developed SIRS (n=49) were also used. Analysis was performed in accordance with manufacturer's instructions. The panel chips used included: CARDIOMETABOLIC (v.3603), CARDIOVASCULAR II (v.5004), CARDIOVASCULAR III (v.6112), CELL REGULATION (v.3701), IMMUNE RESPONSE (v.3203), INFLAMMATION (v.3012), METABOLISM (v.3402) and ORGAN DAMAGE (v.3311).

[0088] Data Analysis

[0089] Graphs were generated using the software Graphpad PRISM V8.0. Statistical analysis was performed using IBM SPSS V26.0. NPX data from the three panels were collated into a single file. Where proteins had been investigated in more than one panel, the mean value was taken. Some missing data was replaced using a regression based with random effect method of imputation. These missing values principally consisted of one sample in certain analytes. All data NPX was used regardless of whether the values were within the limits of quantification or whether all quality controls were passed.

[0090] Data was organised into groups (‘SIRS’; ‘Sepsis’; and ‘Control’) and time prior to diagnosis of condition (‘Baseline’; ‘Time of Conventional Diagnosis’; ‘1 Day Prior to Diagnosis’; 2 Days Prior to Diagnosis’; and ‘3 Days Prior to diagnosis’). The 95.sup.th and 5.sup.th percentile were estimated for each protein analyte, for each time point, for the control and SIRS group combined. The proportion of proteins that were outside these percentages were calculated for each time point.

[0091] The top 40 protein analytes at time of diagnosis and 1 day prior were selected (i.e. the protein analytes outside the 90% quantiles of the SIRS and control samples in the most ‘sepsis’ samples. These protein analytes were subjected to stepwise cluster using Pearson's correlations. A dendrogram was then used to select protein analytes that were most unrelated. The “left-most” members of each cluster at different levels of similarity were selected because these represented the least related protein to the next cluster. The ability of different groups of protein analytes to predict sepsis was assessed using multilayer perceptron neutral networks. (Other algorithms that can manage heterogeneity, such as random forests are also suitable. Conversely, linear discriminant analysis would be less useful for the same reason). The SPSS adaptive algorithm was used to fine-tune the methodology of each analysis. The neural nets were trained ten times using 70% of the data at both time of diagnosis and 1 day prior. The same 70% of individuals was used at both time points. For each of the iterations, a random selection program was generated that ensured that the same 70% was used at both time points. The other 30% and other time points were used to predict efficacy. Efficacy was estimated and compared by Receiver Operator Characteristics (ROC) analysis of the membership estimates and the AUC of the ROC curve.

[0092] Results

[0093] Assay Reliability

[0094] In order to consider the general reliability of the O-link assay system, single analyte measurements were plotted onto scatter plots. These plots included 20 analytes that had been measured twice and two analytes that had been measured three times. A very high level of correlation was found in these data sets. Level of correlation typically corresponded to where the range of values was greatest. This analysis also allowed an estimation of the likely rate with which outliers occur. A total of 18 obvious outliers was observed in 28,106 readings indicating failure rate of 0.064% (0.041%, 0.101% 95% confidence intervals using the Wilson-Brown method).

[0095] Down-Selection of Protein Analytes Based on Likely Usability

[0096] The O-link output generated data for 718 protein analytes. A metric was needed for rapid down-selection of target protein analytes where the greatest proportion of readings in the sepsis group were outside the normal range of the two control data sets (comparator controls and SIRS). The strategy devised included first calculating the 5.sup.th and 95.sup.th percentiles of the two control groups at each time point and then using logic functions to numerate the number of sepsis readings at the same time point that were outside this range. The greatest number of sepsis samples with specific proteins outside this 90% range were considered most likely to be useful in sepsis diagnosis. The frequency of protein analytes meeting this metric was found to increase at time points closer to diagnosis (FIG. 2; data expressed as a Turkey plot, where protein analytes outside the 75% quartile+1.5× the IQR are expressed as symbols). This is consistent with expectations, as the individual's biology becomes more dysregulated by the sepsis.

[0097] Various down-selections based on this devised metric were made, identifying 40 protein analytes with the greatest values for these metrics at day of diagnosis and day prior to diagnosis. There was significant overlap in these protein analyte sets, providing 54 unique protein analytes.

[0098] Selection of Protein Analytes to Provide the Best Complementary Benefit

[0099] Further down-selection of the 54 candidate protein analytes was based on reasoning that the best approach would be to consider the protein analytes whose expression correlated least well to each other. To this end, cluster analysis of the protein analytes was performed using Pearson's correlations. This analysis generated a dendrogram of comparative relatedness (FIG. 3). Using a relatedness threshold of about between 18 and 19 Average Linkage provided ten clusters. Representatives that are “far right” were selected, as this will be the least related to the next cluster. Two of these clusters contain multiple closely related protein analytes that might be used as representatives.

[0100] Protein biomarker signatures containing between four and ten proteins showed evidence for predictive power when visualised individually. Importantly, the proteins within a biomarker signature correlated with each other very poorly. In this respect, it was reasonable to assume that these protein analytes will complement each other well in a multiple protein analyte diagnostic. Tumour Necrosis Factor Receptor 1 (TNF-R1) was part of a large cluster. In this respect, alternative protein analytes might be used with little effect and the fact that these alternatives are found in similar concentrations can also be visualised. Similarly, CD28 is similarly expressed to CD244.

[0101] Evaluation of Possible Predictive Power of Protein Analyte Panels

[0102] In order to evaluate the predictive power of these panels of proteins, multilayer perceptron neuronal networks were used. Given that sepsis is a blanket term for a variety of infections with pathologies, it was postulated that the best tool for diagnosis would be non-linear. The SPSS adaptive algorithm was used to fine-tune the numbers of nodes and methods. Training sets (70%) were selected randomly using Microsoft Excel random number generator and bespoke generated work sheets that allowed consistency across time points. The same 10 training/test sets were run for each iteration of analysis.

[0103] It was found that representatives (n=22) of the ten clusters at ˜20 similarity gave good ROC curves at day of and day prior to diagnosis. Table 3 describes the predictive efficacy, described in terms of AUC, of a series of biomarker subsets produced from the list of 22 proteins down-selected for the biomarker signature (SD: standard deviation).

TABLE-US-00003 TABLE 3 Predictive efficacy for sepsis of a series of biomarker signatures derived from 22 down-selected proteins Day −1 Day 0 Ref Biomarkers Mean Median SD 25.sup.th Q 75.sup.th Q Mean Median SD 25.sup.th Q 75.sup.th Q A CCL16, CD28, FGF21, 0.80 0.82 0.07 0.75 0.86 0.86 0.86 0.07 0.85 0.91 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TNF-R1 B CCL16, CD28, GALNT3, 0.77 0.79 0.03 0.75 0.80 0.83 0.80 0.04 0.80 0.85 GT, LDL-R, LILRB5, MCP-2, MMP-1, TNF-R1 C CCL16, CD28, GALNT3, 0.75 0.75 0.03 0.74 0.77 0.82 0.78 0.04 0.79 0.86 GT, LDL-R, LILRB5, MMP-1, TNF-R1 D CCL16, GALNT3, GT, 0.74 0.76 0.06 0.75 0.78 0.76 0.76 0.06 0.75 0.79 LDL-R, LILRB5, MMP-1, TNF-R1 E CCL16, GALNT3, LDL-R, 0.75 0.75 0.04 0.71 0.78 0.79 0.77 0.05 0.76 0.84 LILRB5, MMP-1, TNF-R1 F CCL16, GALNT3, LILRB5, 0.73 0.73 0.03 0.71 0.75 0.78 0.72 0.05 0.73 0.80 MMP-1, TNF-R1 G CCL16, LILRB5, MMP-1, 0.76 0.76 0.03 0.75 0.78 0.80 0.78 0.03 0.80 0.82 TNF-R1 H LILRB5, MMP-1, TNF-R1 0.73 0.73 0.02 0.71 0.74 0.74 0.74 0.02 0.73 0.76 I MMP-1, TNF-R1 0.70 0.70 0.02 0.69 0.70 0.73 0.71 0.02 0.71 0.74 J CCL16, CD28, FGF21, 0.76 0.77 0.03 0.74 0.79 0.87 0.86 0.04 0.85 0.90 MCP-2, LTBR K CCL16, CD244, FGF21, 0.81 0.82 0.04 0.78 0.83 0.87 0.86 0.02 0.86 0.89 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TNF-R1 L CCL16, CD28, FGF21, 0.78 0.79 0.07 0.75 0.82 0.84 0.86 0.09 0.83 0.90 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TNF-R2 M CCL16, CD28, FGF21, 0.78 0.80 0.06 0.74 0.82 0.85 0.84 0.04 0.83 0.87 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TRAIL-R2 N CCL16, CD28, FGF21, 0.86 0.88 0.06 0.86 0.91 0.86 0.87 0.06 0.86 0.90 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TNFRSF11A O CCL16, CD28, FGF21, 0.77 0.76 0.05 0.73 0.77 0.85 0.82 0.05 0.82 0.88 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TNFRSF10A P CCL16, CD28, FGF21, 0.79 0.78 0.08 0.71 0.84 0.85 0.84 0.05 0.81 0.88 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, LTBR Q CCL16, CD28, FGF21, 0.76 0.76 0.06 0.71 0.81 0.82 0.78 0.09 0.76 0.89 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, IL-18BP R CCL16, CD28, FGF21, 0.76 0.76 0.06 0.71 0.81 0.82 0.78 0.09 0.76 0.89 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, NUCB2 S CCL16, CD28, FGF21, 0.76 0.74 0.07 0.70 0.81 0.84 0.78 0.06 0.80 0.87 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, SIGLEC10 T CCL16, CD28, FGF21, 0.72 0.70 0.07 0.67 0.76 0.81 0.78 0.06 0.77 0.85 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, JAM-A U CCL16, CD28, FGF21, 0.76 0.77 0.07 0.71 0.82 0.84 0.83 0.05 0.83 0.87 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, TNFRSF14 V CCL16, CD28, FGF21, 0.80 0.81 0.06 0.75 0.84 0.86 0.85 0.05 0.80 0.90 GALNT3, GT, LDL-R, LILRB5, MCP-2, MMP-1, U-PAR

[0104] It will be understood that the present invention has been described above purely by way of example, and modification of detail can be made within the scope of the invention. For example, alternative approaches to a ‘sandwich’ reaction may be considered in a method, kit or system e.g. competitive assay, providing that such approaches enable quantitative measurement of the individual biomarkers of the biomarker signature in the sample being analysed. The labelled reagent(s) may include element(s) capable of specifically binding to at least one of the proteins in a protein biomarker signature according to the present invention, wherein the element(s) may be capable of being linked or associated with a labelling means during application of the method, kit or system that allows for identification of the presence of the protein. Each feature disclosed in the description and (where appropriate) the claims may be provided independently or in any appropriate combination.

[0105] Moreover, the invention has been described with specific reference to methods and associated uses, kits and systems relating to assessing sepsis defined according to the Sepsis-3 definition. Additional applications of the invention will occur to the skilled person.