Development and validation of a novel molecular biomarker diagnostic test for the early detection of sepsis

Systemic Inflammatory Response Syndrome (SIRS) is an overwhelming whole body reaction that may have an infectious or non-infectious aetiology and is described in association with critical conditions that include, but are not limited to, pancreatitis, ischemia, trauma and severe tissue injury, including that encountered in patients having major surgery. Major open surgery is a controlled form of physical insult that results in varying degrees of systemic inflammation. To date, it has been reported that the occurrence of SIRS following cardiac bypass surgery [1, 2], is common, as well as being a major cause of postoperative complications including death. In particular, Michalopoulos and colleagues observed that within the immediate 12 hours following Coronary Artery Bypass Surgery, 100% of their study population (n = 2,615) had clinical signs and symptoms consistent with SIRS. This SIRS response is also known to occur post-operatively following other types of major procedures [3, 4]; however, only in cardiac surgical patients has this uniformity of response been demonstrated in the intensive care setting.

If infection is suspected in addition to SIRS presentation, the term sepsis is applied [5, 6]. This condition is the leading cause of mortality in the adult intensive care unit (ICU), ranging from between 18 to 50% [7–9]. In developed countries, the incidence of sepsis is expected to rise due to aging populations, immune-compromised patients, increasing longevity of patients with chronic diseases, antimicrobial resistance, especially in younger people, as well as viral illnesses such as AIDS.

Recent research has suggested that, owing to the amount of cellular necrosis from major physical injury and trauma, mitochondrial DNA is released into the circulation, where it is capable of eliciting inflammatory signals, also referred to as damage (or danger)-associated molecular patterns (DAMPs) [10]. It is these DAMPS (which are also contributed to by nuclear DNA, purine metabolites, as well as other nuclear and cytosolic proteins), that stimulate an acute phase response by the innate immune system which is biologically concordant with pathogen-associated molecular patterns released during infection. Furthermore, the presence of DNA anywhere other than the nucleus or mitochondrial nucleoids elicits a response mediated by toll-like receptor-9 and DNA-dependent activator of interferon regulatory factors (DAI), resulting in the activation of the innate immune response, and can also be associated with the generation of autoantibodies and immune complexes [11]. Thus, following major physical trauma or other pathophysiological processes that result in cell lysis en masse, the immune system reacts in a similar manner to that which occurs when sterile tissues become infected. Moreover, the findings from Zhang et al. [10], demonstrate the difficulty in separating infectious from non-infectious SIRS when the innate immune system responds similarly to both conditions.

For many decades, the cornerstone of sepsis diagnosis and treatment has been identifying the causative circulating pathogen and quantitating single immune-related blood analytes -- medical determinants which are not necessarily specific to sepsis, but routinely evaluated to assess the patient's physiological response to the pathogen. In contrast, it has recently been proposed by Hotchkiss et al. [12], that the sepsis condition is multifactorial and inclusive of the both an early exuberant innate immune response (or "hyperinflammation") followed by a stage of protracted immunosuppression that is referred to as immunoparalysis [12–14]. While this is still conjecture, there would appear to be a combination of phases involved in septic episodes that are not necessarily assessable in terms of presentation, physiology, chemistry, or pathogen load. Given that the immune response to sepsis is complex and difficult to evaluate with single analytes, high throughput technologies such as multiplex PCR have substantive utility for the clinical development of a diagnostic test that is capable of evaluating perturbations in circulating gene expression profiles, and thus determining the status of the patient's immune system.

Thus, as the majority of patients admitted to the tertiary care ICU setting have undifferentiated SIRS, it is of great clinical importance that those patients who have a suspected infection or are at high risk of infection can be identified early, in order to initiate evidence-based and goal-orientated medical therapy. Hence, the primary objective was to validate a molecular biomarker signature identified a priori from pre-clinical research by determining performance outcomes, in a population of critical care patients that included post-operative surgical patients and blood culture-positive sepsis patient.

Pre-clinical research using a novel equine model of sepsis was designed to examine the innate and adaptive immune responses. In brief, endotoxemia and sepsis was induced in a cohort of foals (n = 12) after animals were administered a high carbohydrate diet for 48 hours causing damage to the lining of the intestine and allowing live and dead bacteria to invade the bloodstream, resulting in bacteremia and multiple organ dysfunction. Given recent findings reported by Wade and colleagues [22], where it was demonstrated that 53% of equine chromosomes show conserved synteny to a single human chromosome, this suggests that the horse genome has strong concordance with that of the human and, validates this model as an appropriate one for clinical research and development. Furthermore, based on these pre-clinical findings, a complete clinical research program was developed in order to establish the utility of the biomarkers for the early diagnosis of sepsis (Figure 1).

From these studies, 145 human white blood cell (WBC) genes associated with acute infection and inflammation were identified as being abnormally expressed both in the equine and human contexts. Following, a panel of 42 genes were identified a priori using pre-clinical research outcomes from the pre-clinical results. This panel of biomarkers has been primarily linked with innate and early adaptive immune function/activation pathways, including those governing cell cycling and growth, WBC differentiation, extracellular matrix remodelling, and immune modulation (Figure 2). Based on the physiological response to infection, that is, monocytosis, neutrophil activation, monocyte-macrophage differentiation, extravasation, as well as cell-mediated responses, these patented markers appear to characterise a large cross-section of early immune behaviour. We, therefore, developed a clinical research program to validate this novel biomarker panel to determine its efficacy as a diagnostic test for the separation of sepsis within an inflammatory milieu.

Of the surgical in-patients enrolled into this study, 19 had major open cardiac surgery that included CABG, pericardiectomy, valve replacement and atrial septal defect closure procedures. A further 14 participants had major thoracic procedures that included pleurodesis, lung wedge resection with and without lung volume reduction, lung lobectomy and pulmonary decortication. Lastly, five participants underwent major open vascular and neurosurgery that included AAA repair, aortic, iliac, femoral and femoral-popliteal endarterecetomy procedures, as well as stereotactic removal of a cerebral lesion, respectively. Based on routine clinical catheter tip culture monitoring, none of these participants had any evidence of infection during, or after their participation in this trial; however, all of these surgical in-patients were minimally on broad-spectrum prophylactic antibiotic therapy at the time of the study blood collection.

Of the 27 patients recruited into this trial who met the criteria for sepsis, and who had evidence of systemic infection, 14 had microbiological findings consistent with a gram-positive infection, 11 had findings consistent with a gram-negative infection, and 2 had findings of both gram-negative and gram-positive infections. As this was an observational clinical study, all sepsis participants were treated as per clinical guidelines regardless of SeptiCyte Lab results. On average, the sepsis participants' stayed in the ICU for five days.

In preliminary microarray investigations, HGU133 Plus 2.0 GeneChips demonstrated that signatures for HC, PS and sepsis were well separated when visualised using a Principal Component Analysis (Figure 3A). The area under the curve (AUC) for the ROC curve, using the HC, PS and sepsis data sets was greater than 95% for the detection of sepsis. To evaluate the robustness of this gene expression signature, GEO control samples were compared with gene expression data from the sepsis and HC cohorts. In Figure 3B a strong separation was observed between all 'controls' (GEO and HC) when compared with the sepsis cohort, thus generating a specificity of greater than 99% using ROC curve analyses.

The objective of the GEO data exploration was to investigate whether or not the diagnostic signature is robust to inter-laboratory variation. There is limited evidence that the signature is robust, but inter-laboratory variation in positive sepsis samples would have to be evaluated before this can be proven. Importantly, the high specificity for the GEO data must be interpreted with caution given that no sepsis data from the GEO database were included. Whilst the GEO database does contain gene expression information from sepsis trials, the clinical status of the participant cohorts appeared to be ambiguous based on Inclusion and Exclusion Criteria and for this reason, it was not included in these analyses. At this stage it may be the case that the very high specificity is adversely affecting sensitivity. However, it is encouraging that a classifier developed in one laboratory does not result in a large number of false positives when applied to samples from a large number of laboratories. This provides some evidence that the difference between sepsis and control patients from this trial is greater than the inter-laboratory variation in the gene expression data. Had the difference between control and sepsis patients from the current trial been small with respect to between laboratory variations in 'control' gene expression, it is likely to have resulted in the GEO data decreasing the sensitivity. In this sense, the GEO data provide preliminary evidence that the sensitivity of the test is robust to inter-laboratory variation.

It is routine practice in research and development that microarray studies be employed in preliminary investigations, particularly in molecular oncology, in order to identify statistically relevant panels in which to re-evaluate and further refine in the clinical setting [23–25]. Given that these analyses were not restricted to inflammatory biomarkers, these results provided a strong rationale for further clinical research.

When the a priori set of 145 biomarkers was compared with all gene expression change from the Affymetrix Genechip data they were strongly representative of widespread change and, had the greatest capacity to differentiate within the MI cohort. Thus, based on preliminary outcomes, a further subset of 42 biomarkers including three control/normalisation genes were tested in a larger cohort of critical care patients, to ascertain the clinical utility of this sepsis test using the more efficient quantitative real time PCR platform.

Following MT-PCR data extrapolation, comparisons were first made between HC and generalised systemic inflammation and wounding (MI). Findings from these analyses indicate, based on area under the curve (AUC) for ROC curve, that the ability to accurately detect severe systemic inflammation was approximately 92% using the full gene panel (Table 2). Based on further posterior probability analyses that were conducted using algorithms for 42 and 7 gene expression biomarkers (Figure 4A, C), this result was determined to be statistically significant (P < 0.002) in a permutation test. Furthermore, to demonstrate that there was no underlying bias in the patient population, the HC and MI cohorts had classifiers removed and were randomly mixed during posterior probability analyses, indicating that diagnostic performance (by chance) was approximately 0.5 (Figure 4B, D).

Comparison	Biomarker Set	Mean	SD‡	P-value*
MI Vs HC	42	0.921	0.0568	< 0.002
MI Vs HC	7	0.862	0.0640	< 0.002
Sepsis Vs HC	42	0.938	0.0557	< 0.002
Sepsis Vs HC	7	0.910	0.0479	< 0.002
PS Vs HC	42	0.891	0.0687	< 0.002
PS Vs HC	7	0.833	0.0661	< 0.002
Sepsis Vs PS	42	0.921	0.0568	< 0.002
Sepsis Vs PS	7	0.862	0.0640	< 0.002

‡ Standard Deviation; *Statistical significance is assumed when P < 0.05. Based on 500 random permutations the minimum possible P-value is < 0.002

§ HC -- Healthy Control; MI -- Mixed Inflammation; PS-Post Surgical

Secondly, the mixed inflammatory cohort was split into PS and sepsis, where AUC ROC permutation results for the full gene set (n = 42) and a sub-set (n = 7) demonstrate the ability to detect sepsis was between 86 and 92% (P < 0.002), as displayed in Figure 5A, C. Again, when participant classifiers were removed for the purpose of posterior probability analyses, random diagnostic performance was on average 0.5 (Figure 5B, D).

Interestingly, when small sets of gene expression markers are used, the performance of the algorithm to accurately detect sepsis remained high. In fact, several sets of seven biomarkers were evaluated, and performance remained high for each set, suggesting that there is substantial redundancy in the 42-biomarker set. Despite this redundancy, the SeptiCyte Lab will continue to include all 42 biomarkers for future studies designed to evaluate sepsis status in patient populations from different settings such as: in neonatology and paediatrics settings, patients who are immune-compromised following chemotherapy, as well as in mixed ethnic populations. The larger biomarker set provides a higher chance of success for these more diverse populations.

Quantitative real time PCR, also referred to as the MT-PCR method in this paper, is a fast and efficient platform on which to conduct gene expression studies, and an ideal platform for use in a tertiary care clinical pathology unit. Given the current level of automation, and the requirement for as little as 10 ng of RNA (< 1 PAXgene or EDTA blood sample), test results can be reported back to the referring clinical team within three hours. Hence, there is the capacity to substantively improve early clinical management plans when considering it may take between 48 and 72 hours for blood culture results to become available, a limitation that is further exacerbated by the failure to grow an organism in a high percentage of blood cultures taken from patients who have been deemed to be extremely likely to have sepsis on clinical grounds.

To date, the application of such a highly multiplexed gene expression test for the early detection of sepsis is unique; however, there are similar diagnostic products on the market for other immune-related conditions in the field of heart-lung transplantation [26]. Importantly, multi-marker diagnostics may provide guidance where there is nonlinearity of data, a situation commonly associated with complex disease states.

In terms of limitations, this test currently has the capacity to distinguish between normal individuals (healthy controls) versus those with systemic inflammation of unknown aetiology (MI), as well as PS inflammation (surrogate SIRS) and sepsis (confirmed bacteraemia). However, as indicated there are a number of 'high risk' populations that develop infection due to immunomodulating medications used following transplantation, chemotherapy regimens and management of autoimmune diseases. To use this test efficaciously in such complex cases would require baseline gene signatures for these populations. Further studies are underway to derive these.

In summary, these findings demonstrate that this novel gene expression biomarker test (SeptiCyte Lab) has the capability, based on diagnostic performance outcomes, to accurately detect early evidence of sepsis well before the availability of microbiology results. Current clinical research will further investigate the efficacy of this assay to provide guidance on the host immune response via the monitoring of severe sepsis patients, as well as to provide diagnostic information on other 'at risk' groups such as post-operatively following major surgery and patients with chronic immune incompetence.