Characteristics of each of the patient groups are summarized in Table . For each of the bacterial pneumonia patients, the pathogen responsible for infection and the specimen from which the result was obtained is listed in Additional file 1
, Table S2. No difference in the severity of illness (as measured by APACHE II scores) was found for patients in the bacterial pneumonia compared with the H1N1 influenza A pneumonia group (P
= 0.82). The mean age of bacterial pneumonia patients was higher than that of the influenza A patients (P
= 0.00040). We therefore incorporated age as a covariate in the linear mixed-model analysis. All results reported henceforth have accounted for the difference in age between groups.
Characteristics of the individuals included in the study
The linear mixed-model analysis showed that changes in levels of gene expression were determined by patient phenotype (H1N1 influenza A, bacteria, or SIRS). Other variables, such as disease severity, day of ICU stay, and patient age, were not associated with any change in gene-expression levels. With the exception of Y-linked genes RPS4Y1, JARID1D, EIF1AY, UTY, and RPS4Y2, patient gender was not found to influence gene-expression levels. Each phenotype was associated with significant changes in gene expression in a large number of genes, as summarized in Table .
Number of genes up- and downregulated for each patient phenotype, compared with healthy controls
Venn diagrams reveal overlaps in the lists of upregulated and downregulated genes compared with healthy controls for the three patient phenotypes (Figure ). At 5% FDR, 1,350 genes were upregulated compared with healthy controls in all three patient phenotypes. Biological pathways overrepresented in these genes included apoptosis (p = 4.4E-8), immune system response (P = 4.3E-6), DNA-damage response (P = 1.4E-5), and inflammatory response (P = 6.8E-5).
Figure 1 Overlap of differentially expressed genes in H1N1 influenza A pneumonia, bacterial pneumonia, and noninfective systemic inflammatory response syndrome. Venn diagrams for genes upregulated (A) and genes downregulated (B) compared with healthy controls, (more ...)
A distinct gene-expression profile was found for the H1N1 influenza A group. This gene-expression profile is found predominantly in the upregulated genes (Figure ). Biological pathway analysis of the 1,416 genes uniquely upregulated in H1N1 influenza A infection revealed overrepresentation of pathways related to the cell cycle and its regulation (p = 4.2E-20), DNA-damage response (P = 4.2E-9), apoptosis (P = 1.3E-4), and protein degradation (P = 4.1E-4). Figure lists the top overrepresented biological pathways in the order of statistical significance.
The top-ranking biological pathways in genes upregulated in H1N1 influenza A infection, ordered by statistical significance (with cell cycle being the most significant among the top 10 pathways).
In contrast to influenza A infection, a gene-expression signature was not found in bacterial pneumonia. The genes uniquely upregulated in response to bacterial pneumonia (n = 253) were not overrepresented in any biological pathway or network ontology, implying a generic inflammatory and immune response, but no specific response to bacterial infection.
A larger number of genes were upregulated in SIRS (586 genes). Further analysis showed that they were overrepresented in multiple biological pathway and network ontologies, including inflammatory response (P = 6.3E-6), cell differentiation (P = 1.6E-5), angiogenesis (P = 1.1E-4), and immune system response (P = 2.6E-4). This is consistent with the known biology of SIRS, which is a nonspecific host response to a variety of stresses, including trauma, surgery, and infection.
A large number of genes were downregulated in H1N1 influenza A infection, bacterial infection, and SIRS groups (Figure ). Biological pathway analysis of the downregulated genes was performed for each of the three patient phenotypes (Figure ). In the H1N1 influenza A group (934 unique genes), many genes were overrepresented in inflammatory-response and immune system-response pathways. Further interrogation into the immune-response pathways showed that activation and signaling pathways of interleukins (IL-8, IL-2, IL-15, IL-6, IL-10, IL-7, IL-3, IL-13, IL-17, and IL-23) were heavily overrepresented in the downregulated gene list. This suggests a significant degree of immunosuppression in severe H1N1 influenza A infection. In contrast, the degree of downregulation in biological pathways was considerably less in both the bacterial-infection and the SIRS groups (Figure ).
Representation of biological pathway ontologies in the downregulated genes at 5% false discovery rate (FDR) for H1N1 influenza A, bacterial pneumonia, and systemic inflammatory response syndrome (SIRS), compared with healthy controls.
Pathway analysis of the direct comparison between the H1N1 influenza A and bacterial groups revealed a consistent picture, with 671 genes upregulated in H1N1 influenza A compared with bacterial (by using linear mixed model, 5% FDR) showing remarkable overrepresentation in the cell cycle and its regulation ontology (P = 2.9E-20). The DNA-damage response was also highly enriched in this list of genes (P = 6.9E-10). No such overrepresentation was seen for cell-cycle pathways in the 78 genes expressed at higher levels in the bacterial infection group (P = 0.35). The biological pathways overrepresented by these 78 genes include immune and inflammatory responses. However, these immune/inflammatory genes are also upregulated in SIRS and are therefore not specific to bacterial pneumonia.
The immune cell subsets that gave rise to most of the gene-expression signals outlined earlier are shown in Figure , as revealed by immune cell deconvolution. Far more neutrophil-tagging genes were upregulated in the bacterial group compared with the H1N1 influenza A pneumonia (P = 2.4E-17). Conversely, a greater representation of T-helper cell-tagging genes was found in the top 100 upregulated genes for H1N1 influenza A pneumonia (P = 2.1E-11). In addition, B-cell genes were significantly overrepresented in the H1N1 influenza A pneumonia group compared with the bacterial group (P = 0.0062). These findings are consistent with the known biology of infection, in which bacterial infection is driven by a neutrophil-dominant response, and viral infection is driven by a lymphocyte-dominant response. Across the 5 days of patient follow-up, the expression level of T-helper cell-tagging genes is consistently higher in H1N1 influenza A, whereas the expression level of the neutrophil-tagging genes is consistently higher in the bacterial group, as shown in Figure .
Figure 4 Immune cell deconvolution of the top 100 upregulated genes for bacterial pneumonia and H1N1 influenza A pneumonia, compared with healthy controls. Fisher Exact test two-tailed P values are given for cell types with significantly different proportions (more ...)
Figure 5 Expression of neutrophil and T-helper cell-specific genes across 5 days for H1N1 influenza A pneumonia and bacterial pneumonia patients. Intensity of red corresponds to level of upregulation, whereas intensity of green refers to level of downregulation. (more ...)
A group of genes well known to be associated with viral infection, referred to as interferon-stimulated genes, were highly represented in the H1N1 influenza A gene signature. With Gene Set Enrichment Analysis, the interferon-stimulated genes were shown to be significantly enriched in the genes overexpressed in H1N1 influenza A pneumonia, compared with healthy controls (FDR = 0.0010). In contrast, even at a 5% FDR, no significance was observed for interferon-stimulated genes among genes overexpressed in bacterial pneumonia, compared with healthy controls (FDR = 0.080). We repeated the analysis by directly comparing the bacterial and H1N1 influenza A groups. Again, a highly significant enrichment of the interferon-stimulated genes was noted in genes overexpressed in the H1N1 influenza A group (FDR = 0.0010) but not for genes overexpressed in the bacterial group (FDR = 0.97).
Because the H1N1 influenza A infection group displayed a gene-expression profile distinctively different from that of bacterial infection, we explored the potential of using the gene-expression profile to diagnose H1N1 influenza A infection. By using an SVM algorithm, we found a 29-gene class predictor to be highly accurate in discriminating H1N1 influenza A infection from bacterial pneumonia (Figure ). This ability to discriminate between bacterial and viral infection was consistent across the 5 days of patient follow-up (see Additional file 1
, Figures S1 and S2). When this class predictor was tested on two independent datasets, it was shown to provide clear separation of both H1N1 influenza A pneumonia patients from a healthy control cohort, and bacterial pneumonia patients from a cohort of patients containing both influenza A- and influenza B-infected individuals. These results support the robustness of the class predictor, as clear separation was observed in independent datasets generated by using different microarray platforms and normalization methods.
Figure 6 The Support Vector Machines (SVM) class-prediction integer in training and validation datasets. The x-axis corresponds to the threshold of 36.03, with all samples falling above the line predicted as belonging to an individual with influenza infection. (more ...)
Surprisingly low overlap is found when comparing the 29-gene class predictor presented in this study with the class predictors presented in previous studies by Zaas et al
] (30 genes), and Ramilo et al
] (35 genes). Only five genes are present in more than one of the three-gene signature lists: IFI44, LY6E, MX1, OAS1
, and IFI27
(see Additional file 1
, Table S3). Notably, each of these five genes is a well-established interferon-inducible gene.
Further analysis of the 29-gene signature showed overrepresentation in biological pathways related to the cell cycle and its regulation (P = 2.1E-4). Specific cell-cycle pathways overrepresented were transition and termination of DNA replication (P = 7.1E-4) and start of DNA replication in early S phase (P = 9.3E-4). No other pathway ontology was significantly overrepresented in the 29-gene signature. Immune cell deconvolution of the 29-gene signature revealed that 14 of the 29 genes were predominantly expressed in T-helper cells. This finding suggests that the 29-gene signature reflects the T-cell response during influenza infection.
The diagnostic performance of the 29-gene signature to identify viral infection remained high even for patients with concurrent bacterial coinfection. We performed an analysis on blood samples of three patients who had both H1N1 influenza A infection and superimposed bacterial infection. Figure shows the cluster analysis after these new samples were incorporated into our original dataset. With the 29-gene signature, all the H1N1 influenza A samples fell into the first cluster, whereas the bacterial or SIRS samples were grouped in a second cluster. Importantly, all three patients with viral and bacterial coinfection were in the H1N1 influenza A group. This suggests that the 29-gene viral signature is not affected by the presence of a bacterial coinfection. One of these three patients had an additional sample collected on day 13. At this point, the H1N1 influenza A pneumonia had been resolved; however, the bacterial infection remained. We note with interest that the day-13 sample was more similar to the bacterial infection cohort in its gene-expression profile. The repeated cluster analysis on day 13 showed that this patient had migrated to the bacterial and SIRS cluster (data not shown).
Figure 7 Dendrogram for clustering bacterial (CAP), H1N1 influenza A (H1N1), systemic inflammatory response syndrome (SIRS), and concurrent bacterial and H1N1 influenza A infection (CAP+H1N1) patients for the 29-gene signature by using Euclidean distance and average (more ...)