46 Chapter 3 patients with lower respiratory tract infections was superior in differentiating bacterial from viral etiology compared to the use of PCT [47]. Transcriptome analyses of blood leukocytes of adult ICU patients with influenza A pneumonia, bacterial pneumonia and non-infectious respiratory compromise delineated a 29-gene classifier of viral infection that discriminated viral from bacterial infection and noninfectious disease; this classifier could not distinguish bacterial infection and noninfectious disease [48]. A set of seven genes was derived from a multicohort analysis that could discriminate bacterial from viral infections [49]. This gene classifier was validated in 30 independent cohorts and integrated together with the 11-gene Sepsis MetaScore in an antibiotics decision model which a sensitivity and specificity for bacterial infections of 94.0 and 59.8%, respectively [49]. Several investigations examined the value of transcriptomics to build host response classifiers that can assist in the discrimination between viral and bacterial causes of acute infections in children [50, 51]. A 35-gene set was able to divide infections based on broad groups of causative pathogens in pediatric patients, with distinct gene signatures for infections caused by influenza A virus, Escherichia coli or Streptococcus pneumoniae [50]. Discrimination within viral pathogens (respiratory syncytial virus, human rhinovirus and influenza) was achieved with 95% accuracy using a 70-gene classifier in children [52]. In febrile children presenting to the hospital a 38-transcript signature distinguishing bacterial from viral infection was discovered, which was reduced to a 2-transcript signature (FAM89A and IFI44L) by removing highly correlated mRNA levels [53]. Transcriptomics has also been used to develop biomarkers that potentially can provide insight into the prognosis of patients with sepsis. In a prospective cohort study of ICU patients with sepsis due to CAP, transcriptomics of peripheral blood leukocytes defined two sepsis response signatures (named SRS1 and SRS2), wherein SRS1 had a immunosuppressive phenotype with a higher 14-day mortality rate when compared to the SRS2 [54]. These signatures not only provide insight into the main pathophysiology of individual patients, but could be the start of targeted therapy and better outcome prediction. In pediatric sepsis the Pediatric Sepsis Biomarker Risk Model (PERSEVERE) was developed to estimate baseline mortality risk in children with septic shock, consisting of serum proteins selected based on gene expression profiles [55, 56]. PERSEVERE has recently been adapted for adults with septic shock [57]. Notably, biomarkers associated with mortality were in part different between adult and pediatric patients; i.e., while in children with septicshock granzyme B, heat shock protein 70 kDa 1B, C-C chemokine ligand 3 (CCL3), IL-8 and matrix metalloproteinase 8 (MMP8) contributed to the predictive capacity of the model, in adult patients, granzyme B, heat shock protein 70 kDa 1B, CCL3, IL-8, IL-1α and CCL4 contributed to its prognostic capability [57]. Considering that the initial approach to select the PERSEVERE protein biomarkers left 68 genes unconsidered, in a subsequent study this group of investigators determined if these previously not incorporated genes could advance the performance of PERSEVERE [58]. A network containing 18 mortality risk assessment genes related to tumor protein 53 (TP53) was revealed and combined with PERSEVERE into the so-called PERSEVERE-XP, which was superior to PERSEVERE in differentiating between survivors and non-survivors [58]. It should be noted that all these studies on host gene expression have led to the publication of multiple gene expression data sets with only partially overlapping classifier genes (Table 1), underlining the necessity to critically determine the study
RkJQdWJsaXNoZXIy MTk4NDMw