Severe sepsis and septic shock are leading causes of death, representing 30-50% of hospital-reported mortality [1]. Sepsis treatment outcomes are disappointing, despite a long history of interventions, such as numerous antibiotics including penicillin, efforts to follow guidelines from the Surviving Sepsis Campaign (SSC), and development of supportive modalities for organ dysfunctions accompanying sepsis (e.g., dialysis, ventilators, extracorporeal membrane oxygenation). We have seen the rise and fall of recombinant human activated protein C (drotrecogin alfa) for the treatment of severe sepsis, while the disappointing results might be explained by statistical insignificance stemming from the relatively lower mortality rate (25%) in the Protein C Worldwide Evaluation in Severe Sepsis (PROWESS) study [2, 3]. In addition to activated protein C, treatments with agents such as toll-like receptor (TLR) 4-blocker (eritoran) and human recombinant lactoferrin (talactoferrin) are also viewed with skepticism [4, 5, 6]. Failure of these treatments in clinical trials might be predictable for several reasons. Sepsis is the result of a complex chain of events composed of innate and adaptive immune responses, including activation of the complement system, coagulation cascades, and the vascular endothelial system (Fig. 1). Such complexity makes it difficult for new drugs targeting a single immunological event to improve sepsis outcome. In addition, immune responses are based on individual patient factors including age, underlying diseases, nutritional state, and even genetic variability. For this reason, treatments, especially immunotherapy, have to be individualized. Furthermore, pathogen factors also vary by patient. Given that adjunctive therapy for sepsis has shown disappointing results, conventional management is of immediate importance in the real world. Practically, "bundled care" for sepsis, with early administration of appropriate antibiotics and supportive care based on SSC guidelines, improves outcome [7, 8]. This emphasizes the necessity for early and accurate detection of sepsis. However, a definite microbiological diagnosis cannot be made in approximately one-third of patients with clinical manifestations of sepsis [9, 10]. For this reason, good biomarkers can guide the early diagnosis and management of sepsis. Here, we discuss sepsis biomarkers and directions for future research.

Sepsis is the result of host response to infection by microbial pathogens, meaning that antimicrobial agents are insufficient for treatment of this infectious disease. In 1904, William Osler noted, "It appears that patients are dying not from their infections but rather their reaction to them." Sepsis has traditionally been considered as a result of uncontrolled inflammatory response, a "cytokine storm" that results in shock or organ dysfunction [11]. More than 30 clinical trials have focused on blocking these inflammatory cascades, such as steroids, tumor necrosis factor (TNF)-α antagonist, and anti-endotoxin. However, the paradigm of sepsis understanding and treatment has shifted toward its immunosuppressive effects [12]. For example, elderly patients with sepsis are lack of fever and other immune responses, that are associated with poor prognosis. Such immunosuppression is now considered a key pathogenesis associated with sepsis mortality. Immunosuppressed conditions lead to secondary infections due to nosocomial pathogens such as Acinetobacter, Enterococcus, Stenotrophomonas, or Candida species, which could worsen outcome. In addition, several clinical trials have shown that immune-enhancing therapies such as recombinant human interleukin (IL)-7 and granulocyte-macrophage colony-stimulating factor may have beneficial effects [13, 14]. Immunosuppression in sepsis has been also identified in post-mortem studies of patients who died of sepsis [11, 15]. There was a marked decrease in lipopolysaccharide (LPS)-stimulated cytokine secretion of mediators including TNF, interferon-γ, IL-6, and IL-10 in splenocytes from patients diagnosed with sepsis. In addition, immune effector cells, including clusters of differentiation (CD) 4, CD8 cells, and human leukocyte antigen-DR were significantly decreased in splenic tissue of patients dying of sepsis compared to control patients [15]. Early hyper-inflammatory and late compensatory anti-inflammatory response syndromes are included in current immunological models of sepsis (Fig. 2). However, this is not a simple biphasic model in many cases. The degree and duration of immune response differs from patient to patient according to age, underlying physical state, comorbidities, pathogen virulence, pathogen burden, and genetic factors. These cycles may repeat, with waxing and waning of clinical symptoms. During the course of sepsis, the duration and degree of immunosuppression could affect the outcome, which leads us to consider tailored immunomodulatory therapy.

The traditional sepsis model is the immune response activated when TLR expressed on the macrophage recognizes LPS in cell walls of gram-negative bacteria. This is an example of pattern recognition receptors (PRR) and pathogen-associated molecular patterns (PAMP). This recognition stimulates secretion of proinflammatory cytokines, such as TNF-α, IL-1β, and IL-6. Various inflammatory cytokines and LPS have therefore been studied as sepsis biomarkers.

TNF-α and IL-1β are released within minutes of exposure to LPS. In the late 1990s, investigators found that LPS-treated mice died after serum TNF-α and IL-1β returned to basal levels, suggesting that mediators other than TNF-α might contribute to death. There are two well-known inflammatory mediators, high-mobility group box 1 (HMGB1) protein and macrophage migration inhibitory factor (MIF), which are important in late phase of severe infections.

Tillet & Francis first discovered C-reactive protein (CRP) in a patient with lobar pneumonia in 1930. It was identified as a protein responsible for precipitating C polysaccharide during the acute phase of Streptococcus pneumonia infection [41]. CRP was also found in patients with endocarditis or rheumatic fever. Its response is stronger in acutely ill patients; levels decrease as patients recover. These characteristics make CRP a member of the class of acute-phase reactants. CRP is an old biomarker used most commonly in clinical settings. It is a nonspecific marker of inflammation that also increases after surgery, burns, myocardial infarctions, and rheumatic diseases [42]. The sensitivity and specificity of CRP as a marker for bacterial infections are 68-92% and 40-67%, respectively [43, 44, 45, 46]. Its low specificity and inability to differentiate bacterial infections from noninfectious causes of inflammation makes CRP of limited diagnostic value. However, CRP shows promise for evaluating sepsis severity and prognosis. CRP plasma levels have been shown to correlate with the severity of infection [47]. A rapid decrease in CRP levels has been reported to correlate with good response to initial antimicrobial therapy in septic patients [48]. CRP is a useful biomarker to monitor treatment response. However, hasty interpretation or antibiotic guidance within 1-2 days after starting empirical antibiotic treatment is problematic in many clinical situations. Clinicians cannot interpret changes in CRP levels without considering the kinetics of this marker.

Procalcitonin (PCT) is a precursor of calcitonin, a calcium regulatory hormone secreted from thyroid tissue in healthy individuals. In infectious conditions, PCT is released from nearly all tissues including lung, liver, kidney, pancreas, spleen, colon, and adipose tissues. In 1993, PCT was first described as a marker elevated in bacterial infections [49]. In 2008, PCT was proposed as an adjunctive diagnostic marker to differentiate acute bacterial infection from other inflammatory states by the American College of Critical Care Medicine and the Infectious Diseases Society of America [50]. In a systematic review and meta-analysis, PCT was found to be more specific (specificity 81% [95% CI: 67-90%]) than CRP (67% [95% CI: 56-77%]) for differentiating bacterial infection among hospitalized patients [46]. The cutoff median PCT value in this meta-analysis was 1.1 ng/mL (interquartile range: 0.5-2.0 ng/mL). PCT cutoffs for diagnosis of sepsis or guidance of antibiotic choice have not yet been fully determined; the sensitivity and specificity of this marker for diagnosis of sepsis are affected by different cutoff values. PCT values need to be further evaluated according to different sites of infection, hosts, and pathogens. Another recent meta-analysis showed that PCT is a useful marker for early diagnosis of sepsis in critically ill patients, with sensitivity and specificity of 77% (95% CI: 72-81%) and 79% (95% CI: 74-84%), respectively [51]. PCT levels are also elevated after surgery, cardiogenic shock, heat shock, acute graft-versus-host disease, and immunotherapy such as granulocyte transfusion, which could limit its usefulness as a sepsis biomarker [52, 53]. PCT has also drawn attention because it can be used for guidance of antibiotic stewardship to reduce inappropriate use of antibiotics [54]. However, many experts recommend that PCT-guided decision-making should be an adjunctive method based on consideration of the patient's clinical course.

Serum lactate levels can reflect tissue hypoperfusion and anaerobic metabolism in severe sepsis and septic shock. At a cellular level, energy production depends on glucose and oxygen metabolism. Glycolysis converts glucose to pyruvate and yields 2 adenosine triphosphates (ATPs). Pyruvate then enters the Krebs cycle, which produces more ATPs. However, in hypoxic circumstances, pyruvate is instead converted to lactate. Elevated lactate levels and lactate-to-pyruvate ratios result mostly from increased glycolysis and lactate production as well as limited tissue oxygenation. Elevated levels are also related to impaired hepatic lactate clearance and mitochondrial dysfunction [20]. Several studies have demonstrated that elevated lactate levels are related to mortality in patients with sepsis [55, 56, 57, 58]. In a retrospective study of critically ill patients, serum lactate levels greater than 2 mmol/L on admission were associated with a 1.94-10.89-fold increased mortality compared to levels below 2 mmol/L [59]. In a large study of 1,278 patients with infections, those with lactate levels above 4 mmol/L had higher in-hospital mortality rates than patients with lactate levels less than 2.5 mmol/L (28.4% vs. 4.9%) [57]. Another study has reported that sustained hyperlactatemia is predictive of in-hospital mortality [60]. In contrast, however, early lactate clearance was associated with improved outcomes in patients with severe sepsis and septic shock [61]. A recent systematic review further confirmed the utility of monitoring serial blood lactate and its value as a predictive marker of in-hospital mortality [58]. Recently, data from a retrospective study by the Vasopressin Septic Shock Trial and a single-center septic shock cohort (St. Paul's Hospital cohort) have suggested that even minimal increases in arterial lactate concentration within the reference range (1.4-2.3 mmol/L) may predict 28-day mortality (sensitivity and specificity of 86% and 27%, respectively). Furthermore, the data suggested that patients with lactate levels below 1.4 mmol/L might benefit from vasopressin infusion [56]. Therefore, lactate screening and monitoring may be a valuable tool for risk stratification and to predict sepsis outcome.

Like PCT, proadrenomedullin (proADM) is a kind of "hormokine" that encompasses the cytokine-like behavior of hormones during inflammation and infections. Adrenomedullin (ADM) is a 52-amino-acid peptide produced by the adrenal medulla. ADM is produced during physiological stress and has various functions including vasodilation and anti-inflammatory and antimicrobial effects [62]. Plasma ADM concentration and ADM gene expression increases in patients with sepsis [63]. However, ADM is rapidly cleared from the circulation, making measurements unreliable. Therefore, instead of ADM, serum quantification of the mid-regional fragment of proADM has been studied. Recent clinical data have shown that circulating mid-regional proADM levels are significantly higher in patients with sepsis than in patients with systemic inflammatory response syndrome (SIRS) [64]. A recent study of febrile patients with hematologic malignancies reported that proADM could predict localized bacterial infections and differentiate sepsis from SIRS [65]. In addition, proADM is responsible for hypotension associated with severe sepsis, which has been proposed as a good marker for risk assessment and predicting sepsis prognosis [64, 66]. If further data support these findings on the predictive value of proADM, it could be useful as both a prognostic marker and a diagnostic marker for early stages of localized infections.

Angiopoietin (Ang)-1 and -2 are endothelial-derived vascular growth factors that play opposing roles during sepsis. Ang-1 stabilizes the endothelium, whereas Ang-2 facilitates loss of endothelial integrity and vascular leakage. Ang-1 or Ang-2 activates the transmembrane endothelial tyrosine kinase Tie2, which mediates the quiescent, healthy state of blood vessels [88]. Ang-2 plays a crucial role in induction of inflammation [88, 89]. Elevated levels of circulating Ang-2 are associated with sepsis with multi-organ dysfunction, which is indicative of impaired vascular endothelial integrity. A cohort study revealed that elevated Ang-1 and lower Ang-2 levels were observed in sepsis survivors [90]. The endothelium and Ang-Tie2 receptor ligand system have been the recent focus of ongoing sepsis studies.

We have discussed several sepsis biomarkers. Numerous biomarkers have been evaluated for clinical use in sepsis, with moderate to good sensitivity and specificity for diagnosis and prognosis. However, the results of measuring a single biomarker are inconclusive in clinical settings. Owing to this limitation, combination approaches measuring multiple biomarkers have recently been introduced. "Scoring systems" have also been developed, which use both clinical and laboratory markers [28, 91, 92]. In 2003, the infection probability score (IPS) was introduced to assess the probability of infection in critically ill patients. The IPS ranges from 0 to 26 points, and includes patient body temperature (0-2 points), heart rate (0-1 points), respiratory rate (0-1 points), white blood cell counts (0-3 points), CRP (0-6 points), and sepsis-related organ failure assessment score (0-2 points). The AUC-ROC of IPS was 0.82 for predicting the probability of infection. Patients with <14 points have only a 10% risk of infection [93]. Several clinical examples of combinations of biomarkers and scoring for sepsis are shown in Table 2 [27, 31, 44, 82, 91, 92, 94, 95].

Combined biomarkers and inclusion of sepsis scoring systems showed better AUC-ROC values than single biomarkers. Theoretically, combining multiple markers can improve diagnostic and prognostic values, because sepsis is composed of multiple immune responses with various changes in cytokines and biomarkers. However, which and how many combinations of biomarkers are most informative have not yet been investigated for use as a high-throughput technology. Cost-effectiveness and comprehensive clinical interpretation must also be evaluated.