Sepsis is characterized by an acute and rapidly evolving disease process, associated with systemic instability. Assuming MSS as a model for septic shock, a dataset  was identified for WGCNA analysis. This is a temporal transcriptome from infants with MSS admitted to PICU. Herein all infants demonstrated a clinical phenotype consistent with severe septic shock and diffuse intravascular coagulopathy. To date, this dataset remains unique in the clinical literature, given the six mRNA sampling points, allowing the documentation of sepsis evolution through changes in gene-expression. On this basis, the application of WGCNA allowed the correlation of PELOD at 0, 24 and 48 hours, to microarray gene-expression data. Thereby resulting in the visualisation of a module-trait matrix, correlating gene expression to clinical parameters (Figure 2). We believe this is the first instance a pediatric clinical scoring system, PELOD, combined with transcriptomic data has been used to elucidate an enriched gene expression pathway associated with acute sepsis. Herein, WGCNA analysis suggested a dynamic enrichment pattern of gene function pathways, from that of nuclear, to cytoplasmic and finally to an extracellular, all in relation to the PELOD 0 hours, 24 hours and 48 hours. Gene expression consistent with nuclear activity in sepsis was also noted by Wong et al  in pediatric polymicrobial sepsis. Further, Walsh et al, using WGCNA, analyzed skeletal muscle from adult ICU patients, investigating gene modules in association with clinical muscle phenotypes. Over a longer time-frame (7 days to 6 months) Walsh et al commented on the analytical enhancement of WGCNA in eliciting gene-modular relationships. Specifically, Walsh et al demonstrated gene modular enrichment for skeletal muscle regeneration and deposition of extracellular matrix. Accordingly MSS enrichment box plots (Figure 2) illustrated the usefulness of time-series gene expression data, noting an increased innate response associated at PELOD 24 hours and PELOD 48 hours in comparison to PELOD 0 hours.
Multiple factors affect the interpretation of the genomic response to sepsis. For example, there is the inherent complication of unifying gene pathways according to polymicrobial sepsis studies. Further, age can affect the host’s genomic response to sepsis. For example, Wynn et al studied neonates, infant, toddler and school age within 24 hours of PICU admission in septic shock . Wynn et al demonstrated that in sepsis, developmental-age impacts the early whole blood transcriptomic response. Furthermore, age affects on the transcriptomic were explored by Raymond et al  showing infants and children being most similar, whereas neonates and adults were most dissimilar. Raymond et al showed that neonates had a reduction in gene pathways related to signaling and inflammatory recognition. Adults on the other hand demonstrated decreased inflammation, pathogen sensing and myeloid function compared to infants and children. In our study, we attempted mitigation of polymicrobial and age-based factors. Firstly a single organism approach, as suggested by Wong et al , was undertaken. Thereby studying Neisseria Meningitis associated MSS, aiming to simply the investigation of complex host-pathogen interactions. Secondly, with respect to age, patients recruited included infants with no previous co-morbidities, from a similar age-range.
The use of WGCNA as an improvement over standard statistical methods for differential gene expression has been previously studied. Here Langfelder investigated the use WCGNA for hub-gene selection finding WGCNA as an improvement over standard statistical approaches incorporating the p value . However Langfelder also found with regards to analytical repeatability using independent data sets, that standard statistical methods were an enhancement over WGCNA. Further, WGCNA methodology is advantaged by type 1 and type 2 statistical error minimization. Moreover, WGCNA applied to sepsis may show potential beyond traditional clinical biomarkers. For example, LONG et al combined WGCNA with a machine learning algorithm and applied this workflow to three publicly available sepsis datasets . Then by applying artificial intelligence concepts to WGCNA, Long et al presented a diagnostic classifier with the potential for early diagnostic benefit.
A drawback of our gene-expression study relates to the small number of patients recruited. However WGCNA sample-sets ideally should contain at least 15-20 samples . Thus from a mathematical perspective, there were sufficient data-points. A further drawback of this study relates to the labeling of patient’s according to the PICU admission. This allotment of time points is independent of the timing of onset of infection. The arbitrary labelling across a sepsis time trajectory, could affect the analysis of time-related changes in gene function. Further using non-continuous physiological data, such as PELOD scoring may, on the one hand, simplify the dataset. On the other hand, such an approach could impede genome to clinical trait matching.
The ability to match clinical parameters to gene modules in infants, using a gene modular approach is an advancement in application of sepsis transcriptomics. However the timing from infection to the development of symptoms differ across the recruited patients, which could affect analysis. As the methodology adopted encompassed all temporal datasets, this could help mitigate any disadvantages with respect to time constraints. Further, there is the challenge of inter-individual variation with respect to the timing of infection. This being related to such facts as symptom onset, rapidity of pathogenesis, ability to seek medical assistance etc. Countering this potential disparity in transcriptomic datasets was the fact that all patients received standard treatment, aiming for physiological stabilisation. Thereby it could be argued that the transcriptome studied reflects a uniform therapeutic impact upon the transcriptome. Finally, the temporal nature of this study takes advantage of an inherent benefit of time-related gene expression datasets. Namely that, for each patient, one dataset is related to the next, due to the evolving nature of sepsis. This connectivity may be advantageous to network methodology.
As the sepsis transcriptome can be affected by age, gender and ethnicity, future work requires larger datasets. Greater patient numbers would then allow subgroup analysis according to various host-related factors. Further, in this study, the relationship between clinical parameters and gene function, during acute sepsis, has been sort. Close matching of clinical physiological data to genomic modules was attempted. However further work in physio-genomics is required from the temporal perspective. Accordingly, by recording continuous clinical data-points, from the time of PICU admission, could allow an enhancement in the matching of bed-side physiological changes to gene function. In this way, a larger physiological dataset could provide improved prognostication using WGCNA a gene-modular approach.
In summary this study using time-related trajectory transcriptomic data, gene co-expression network analysis has been applied to understand sepsis evolution. In particular, the study showed the capability of using WGCNA in matching gene-expression to clinical trait information. The benefit of using network methods for the isolation of biologically significant gene modules, diagnosis, therapy and prognostication in sepsis requires further clinical study.