Transcriptomic data from two primary cell models stimulating human monocytes suggest inhibition of oxidative phosphorylation and mitochondrial function by N. meningitidis which is partially up-regulated by IL-10

Sepsis, defined as life-threatening organ dysfunction caused by a dysregulated host response to infection [1], remains a serious cause of mortality in intensive care units worldwide [2, 3]. The past decade has seen a large number of studies aiming to investigate the pathophysiology of sepsis through systems biology approaches, with genome-wide expression profiling being a widely used method [4‐6]. This approach has in part been motivated by the failure of a large number of clinical trials directed towards the specific inhibition of endotoxin and cytokines [7, 8], and the need to increase our understanding of the complexity of the sepsis syndrome [9, 10].

Meningococcal sepsis is an overwhelming form of the sepsis syndrome, with previous studies identifying mortality within 12–24 h in previously healthy children and adults [11]. The intense and rapidly evolving pro-inflammatory response resulting from high levels of meningococci and lipopolysaccharides (LPS, endotoxin) in blood [12‐16] is suggested to be the main determinant of septic shock with multi-organ failure, coagulopathy and mortality. Our research group has over the past two decades used a primary cell model consisting of elutriated and cryopreserved human monocytes to elucidate cellular mechanisms activated by meningococcal sepsis plasma [14, 17‐20]. We have viewed human monocytes as a relevant cellular model for several reasons, including their pro-coagulant activity in meningococcal sepsis [21, 22], and their role as phagocytic cells in various tissues after differentiation to macrophages [23].

Our group has recently investigated the gene expression profile induced by N. meningitidis in human monocytes. The first study demonstrated that N. meningitidis and purified LPS can differentially regulate the expression of over 4600 genes [18]. Follow-up studies have investigated the differential expression of genes induced by plasma from patients with meningococcal sepsis, with a special focus on the biological significance of the anti-inflammatory cytokine IL-10. The highest concentrations of IL-10 are found in non-survivors of meningococcal septic shock [16, 17, 24]. We have studied the effect of IL-10 in two previous studies, one stimulating monocytes with wild-type N. meningitidis [25] and recombinant IL-10, and the other using plasma samples from patients with severe meningococcal sepsis or septic shock [26]. In both studies we showed that IL-10 regulates a group of genes that are induced by N. meningitidis, and that these genes are associated with a broad range of functional categories.

Over a decade ago, the main approach to gene expression analysis was to focus on genes representing the largest difference between two experimental conditions, namely those at the top and bottom of a list. However, as described by Subramanian et al. [27], this approach has had a number of limitations. First, an extremely large list may make selection of relevant genes daunting. Secondly, individual genes may not meet the threshold for statistical significance once correction for multiple testing is performed. Finally, focusing on individual genes may overlook important effects suggested by a set of genes that are coordinately expressed, but where the fold change (FC) level does not exceed the cut-off point. It is in response to these challenges that gene set enrichment analysis (GSEA), as initially described by Mootha et al. [28] and later developed by Subramanian et al. [27], was established.

Our previous gene expression studies have used the commercially available tool, Ingenuity Pathway Analysis (IPA) for data analysis [29]. The IPA software analyzes the gene expression profile against molecular relationships in the Ingenuity Knowledge Base, which is a repository of functional annotations, biological interactions and modelled relationships between proteins, genes, cells, tissues, drugs and diseases reported from primary literature sources, including peer-reviewed journal articles and review articles, and other databases (e.g., KEGG, EntrezGene, the Gene Ontology Project). IPA has become among the most widely used tools for gene expression analysis. Similar to GSEA, IPA is a form of enrichment analysis, which is an analytic strategy aiming to identify whether a group of genes is significantly associated with a particular biological process or signaling pathway [30]. GSEA and IPA differ with respect to the statistical approach taken to identifying the significantly enriched groups of genes [27, 31].

Over time, the increasing number of bioinformatics tools have made it easier for investigators to make biological interpretation of genes differentially expressed between different experimental conditions. However, the increasing number of tools have also made comparisons between gene expression studies, such as those for sepsis [5], more difficult, since biological interpretation may differ depending on the methods’ underlying assumptions and statistical tests. One approach to overcome these challenges is to expose the dataset to two different data analytical methods, and consider whether the methods yield similar biological interpretation. In this study we used IPA and GSEA to analyze previously generated DNA microarray data from studies of human monocyte host response to N. meningitidis, IL-10 and meningococcal sepsis plasma [25, 26]. The objectives were to compare whether these two methods resulted in similar biological interpretation of the datasets; and identify whether GSEA provided additional insight about the human monocyte host response to meningococcal activation.

The emergence of a large number of bioinformatics tools over the past decade have made it easier for researchers to interpret the vast amount of transcriptomic data generated from DNA microarrays, RNA sequencing and other gene expression methods. However, the diverse range of tools have also created a number of challenges. First, differences in assumptions underlying the methods used have made it more difficult to compare studies that have used genome-wide expression profiling to study similar pathophysiological phenomena, such as sepsis [5]. Secondly, depending on the methods used, the same gene expression dataset may result in different interpretations, thereby requiring close attention to the underlying statistical assumptions that have affected data interpretation.

The first aim of this study was to compare the biological interpretation resulting from two different bioinformatics tools – GSEA and IPA. Specifically, we investigated the pro-inflammatory response induced in two different model systems of meningococcal activation of human monocytes. Overall, both GSEA and IPA associated the differentially regulated genes to similar biological processes and functional groups, identifying that N. meningitidis and meningococcal sepsis plasma depleted for IL-10, respectively, induced genes associated with pro-inflammatory innate immune activation, including TNF-signaling, Toll-like receptor signaling, JAK-STAT-signaling, type I and type II interferon, as well as apoptosis and mitochondrial dysfunction. Some enriched gene sets in GSEA, such as KRAS signaling, allograft rejection, epithelial mesenchymal transition, and estrogen late response did not immediately appear relevant to the experimental condition and disease under study, and were, except for estrogen late response, not detected by IPA. In addition, when we focused specifically on genes differentially induced by the presence of IL-10, GSEA and IPA differed in the enrichment of biological processes and signaling pathways.

Overall, the rapid induction of pro-inflammatory genes by sepsis plasma is consistent with comparable previous studies of sepsis-induced gene expression at an early stage of the sepsis syndrome [36‐40]. It should be noted that all these studies investigated gene expression in whole-blood or leukocyte isolates collected within 24 h of admission, which differs from our dataset which is generated by inducing gene expression in human monocytes elutriated from heparinized whole blood from healthy blood donors combined with sepsis plasma samples [26]. These studies have therefore the advantage of reporting data generated from leukocytes isolated directly from patients during the clinical course of sepsis. An advantage of our model is that the onset of pathogenic activation of the monocytes as well as the bacterial load is known. In most clinical sepsis studies, the exact onset of the disease is unknown. Furthermore, the initial blood samples are drawn from various time-points during the first 24 h of admission, and the bacterial load in plasma or the microbial molecules driving the innate immune reaction often remain unknown [36‐40].

A few differences exist between GSEA and IPA. The first and foremost is that GSEA does not implement a filtering criteria prior to analysis, and therefore consider the expression of all the genes in the dataset. The main strength of GSEA is to identify the extent to which an a priori defined list of genes, which previous experiments have associated with biological processes or signaling pathways, is coordinately expressed in the dataset. This differs from the conventional approach to gene expression analysis in IPA, which normally implies setting a pre-defined cut-off based on a corrected or uncorrected P-value, and FC values, such as 1.5 and 2.0. A risk with this approach is that the investigator may unintentionally exclude relevant genes, and risks the loss of biologically relevant information which the bioinformatics tools are unable to detect [41]. In comparison, the underlying statistics of GSEA enables the detection of biological functions associated with coordinated, but subtle changes in gene expression levels.

In our case, the previous interpretation of gene expression profiling of human monocyte response to meningococcal infection, generated both in a model system [25] and in a patient plasma system [26], overlooked the large number of significantly regulated genes associated with mitochondrial dysfunction and oxidative phosphorylation. This was primarily due to the implemented filtering criteria consisting of a cut-off for statistical significance (P < 0.01) and FC (1.5). However, in the present study, using a filter of FDR 5% only, the IPA analysis suggested that N. meningitidis induces a strong down-regulation of a large number of genes associated with oxidative phosphorylation, consistent with what was identified by GSEA.

Another challenge to directly comparing GSEA with IPA is that the gene sets in GSEA and canonical pathways in IPA do not directly correspond to each other. The gene sets in GSEA are curated based on gene expression patterns identified in previously conducted experimental conditions. In many cases, the gene sets correspond to a broadly defined biological process (such as “inflammatory response”, “complement” or “coagulation”), or are named after the response to a particular stimulus (such as interferon-gamma or TNF). In comparison, the canonical pathways in IPA delineate more detailed biochemical steps known to result from specific exposures or receptor-binding, or describe mechanistic interactions occurring in specific diseases (refer to Panel 1). Furthermore, IPA does not limit itself to interactions observed in gene expression studies, which is the case for the hallmark gene sets from the Molecular Signatures Database [34] used for the present study. Another advantage is that IPA is also more frequently updated.

We used these two methods in a complementary manner. The Gene Set Enrichment Analysis, using the hallmark gene sets or other gene sets from the Molecular Signatures Database [34] when relevant, produced an unbiased, initial analysis of the extent to which there was coordinated expression of genes related to defined biological functions. Verification of these results was conducted in IPA, which enabled a more detailed examination of whether the differential regulation of genes was associated with mechanistic pathways, as well as locating the position and direction of change of specific genes in pathways.

The second aim was to investigate whether the biological processes associated with the gene expression changes in our model system of meningococcal sepsis was comparable to the results found in our patient plasma system. GSEA suggests that the simple model system, consisting of human monocytes stimulated in a normal pooled plasma with wild-type N. meningitidis and IL-10, induces transcriptomic changes associated with similar functional groups observed by the more complex plasma from patients with meningococcal sepsis. This finding has two biological interpretations. First, our simple human monocyte model system appears robust for the study of cellular mechanisms and biological processes associated with the host response to N. meningitidis. It might thus be applicable to other infectious agents. Second, the gene expression changes in the model system, where the transcriptomic response is initiated by the treatment of monocytes with N. meningitidis and IL-10, are reproduced in the patient plasma system before and after IL-10 immunodepletion. It can therefore be suggested that the dominant inducer of gene expression in meningococcal sepsis plasma is N. meningitidis, and that the dominant modulator of the Nm-induced gene expression pattern is IL-10.

Finally, this study wanted to identify whether GSEA and IPA together could generate additional insight about the human monocyte host response to meningococcal infection. The IPA analysis, supported by GSEA, suggested that N. meningitidis is a strong down-regulator of genes associated with oxidative phosphorylation and mitochondrial function. The reduced expression of genes associated with mitochondrial function and oxidative phosphorylation has also been identified in white blood cells isolated from sepsis patients [42, 43], and in large animal sepsis models [40]. Tang et al. have reported the inhibited expression of genes associated with mitochondrial dysfunction in a study isolating neutrophils from sepsis patients [42]. However, their study reported that gene expression was even lower in the non-septic controls. Wong et al. used IPA and identified that genes up-regulated in neutrophils from sepsis patients enriched pathways related to mitochondrial dysfunction [40]. Two more recent studies have more specifically studied the differential regulation of genes associated with oxidative phosphorylation in sepsis patients [44, 45]. One study by Weiss et al. [45] determined the differential expression of 296 nuclear-encoded mitochondrial genes in whole blood collected within 24 h of admission from pediatric patients with septic shock compared with non-septic controls. They identified that 118 of 296 genes were differentially regulated (48 up-regulated and 70 down-regulated) in septic shock patients. Between survivors and non-survivors of septic shock, the genes cytochrome C oxidase subunit VIIb and NADH dehydrogenase flavoprotein were differentially regulated. They also identified that a sub-group of patients with higher rate of multiple-organ failure and higher mortality rate had a greater repression of nuclear-encoded mitochondrial genes, compared with two other sub-groups. Another smaller study, reported in a letter to the editor by Raman et al. [44] describes gene expression in whole-blood in previously healthy children with meningococcal septicemia at 0, 4, 8, 12, 24, and 48 h from time of admission. They identified an immediate reduction of gene expression associated with oxidative phosphorylation processes, and it continued to decrease over the 48 h, leading the authors to suggest that mitochondrial dysfunction contributed to multi-organ failure. However, the letter does not report the clinical outcomes of the patients investigated. The findings from our and previous gene expression studies are consistent with a long held view that mitochondrial dysfunction and altered oxidative phosphorylation have implications in sepsis pathophysiology [46, 47]—a theme that has recently received renewed attention [44, 48, 49]. In the context of meningococcal sepsis, the intense down-regulation of genes associated with oxidative phosphorylation indicates that N. meningitidis as part of a pro-inflammatory response may induce dysfunction of oxidative phosphorylation and deranged energy metabolism [50, 51]—which has also been referred to as “cytopathic hypoxia” [52]—which contributes to rapidly evolving multiple organ failure. We identified furthermore that IL-10 partially reverses the down-regulation of genes associated with oxidative phosphorylation. To our knowledge, this has not been reported in previous studies of the effect of IL-10 on gene expression induced by LPS [53‐56]. We have in our previous studies shown that IL-10, in response to the meningococci, both up- and down-regulates genes associated with a broad range of cellular functions [25, 26]. The increased expression of genes associated with oxidative phosphorylation induced by IL-10 indicates that activation of IL-10-signaling may improve cellular respiration, and that this may part of IL-10’s well-known attenuation of the harmful effects of meningococcal activation.

Overall, these findings should motivate functional validation by research groups specialized in the study of mitochondrial function and the oxidative phosphorylation pathway in order to generate stronger evidence about the effect of meningococci on oxidative phosphorylation in human monocytes and possibly other immune cells, and the regulatory effect of IL-10.