Abstract
Sepsis remains a major global health issue in pediatric and adult populations, largely due to a lack of understanding of its complex pathophysiology. Despite its high mortality rate, there have been few advancements in sepsis-specific therapies over recent decades. The study aimed to investigate the potential benefits of a genome-wide transcriptomic approach to sepsis in pediatric and adult populations in reducing sepsis-related mortality and enhancing sepsis guidelines. The scoping review explored gene expression data pertinent to developing sepsis guidelines related to its definition, classification, disease severity, molecular biomarking, and benchmarking. A system-biology approach using transcriptomics was adopted to enhance the understanding of sepsis at the mRNA gene expression level. The study involved a search of the PubMed database for original research or systematic reviews that involved transcriptomic application in the context of clinical sepsis published over a ten-year period, from 2012-2022. Of the 14,048 studies retrieved, a full-text analysis was performed. Five main concepts emerged: case definition, classification, quantifying severity of sepsis, transcriptomic biomarkers, and benchmarking. Studies were categorized according to these five categories. The results showed evidence of a connection between the transcript and clinical sepsis, demonstrating that transcript-driven sepsis categorization is possible. Integrating transcriptomic data with clinical endpoints holds promise for more precise sepsis treatment. Although further exploration is needed, the methodology shows potential for disease modification.
Introduction
Sepsis continues to be a major contributor to in-hospital mortality globally and poses a significant public health burden, particularly in pediatric populations 1. Consensus guidelines have been developed for managing sepsis in both adult and pediatric patients, emphasizing early recognition and intervention, aiming to standardize sepsis management practices 2, 3. However, challenges persist in creating protocols due to gaps in clinical evidence, as a relevant framework to understand sepsis has not been finalised. A reason why adult sepsis definitions have undergone many iterations in the past decades 4, 5. The pediatric surviving sepsis guidelines exemplify the impact of an unclear definition for developing guidelines feature mostly weak recommendations due to low-quality evidence 6. These knowledge gaps have helped identify numerous research priorities and pathophysiological questions. The prevailing view of sepsis as a clinical phenomenon has potentially limited the comprehension of its molecular underpinnings, which may hinder the identification of diagnostic and therapeutic targets, as well as the establishment of a definitive sepsis definition applicable to all age groups7–9.
Despite potential therapeutic breakthroughs for sepsis, such as vitamin C10 and activated protein-C (APC), these have shown limited success in clinical trials or meta-analyses 11. While successful in cancer and viral infections, immunotherapy has not gained traction in bacterial sepsis due to its affect on a multitude of molecular pathways, resulting in immune dysfunction. The slow progress in sepsis treatment emphasizes gaps in scientific knowledge 12, 13 and the need for early recognition, diagnosis, and resuscitation. Current laboratory protein biomarkers like CRP and Pro-calcitonin have limited predictive capabilities 14, and the inability to stratify sepsis patients based on biochemical and immunological profiling further complicates matters 15,16.
Sepsis represents a complicated challenge due to its multifactorial heterogeneity resulting in a variable disease process 17. Further, the transition from infection to septic shock remains poorly understood due to factors such as innate and adaptive immune mechanisms, the severity of infection, host age, adequacy of treatment, and genetic variability/susceptibility 18. Moreover, Genetic determinants for sepsis are only partially understood, adding to the complexity. Additionally, the quality of clinical care provided to patients with severe infections influences the disease’s natural history, with serious consequences in children 19.
Omic approaches, including lipidomics, metabolomics, proteomics, and transcriptomics, have been employed to understand complex disease paradigms like sepsis. Transcriptomics focuses on analyzing all RNA transcripts in a biological system, allowing a study of gene expression patterns, detection of differentially expressed genes, and description of alternative splicing events. Hasson et al. (2022) highlighted the potential of transcriptomic analysis in understanding sepsis-associated acute kidney injury and uncovering underlying pathophysiological mechanisms 20. Transcriptomics offers a systems-based approach to understanding biological processes 21 and facilitates precision medicine strategies 22, 23. The enhancement of in silico techniques has facilitated system-wide gene expression analyses, contributing to a burgeoning body of knowledge. Consequently, the current study seeks to explore the application of transcriptomic analysis to deepen our understanding of sepsis. This progression is the next sequence in the genetic dogma, signifying a transition from DNA to mRNA, including not only the study of protein-encoding genes (from coding regions of the DNA) but also non-coding regions that generate microRNA, circular RNA, and long noncoding RNAs. One of the key advantages of RNA-based technologies over DNA studies is the ability to reveal real-time dynamic changes to develop a temporal understanding of sepsis.
Hence a scoping review was undertaken to understand whether findings from transcriptomic studies can be implemented into clinical practice, particularly into international sepsis guidelines. As a part of this gaps in the literature were also to be identified.
Therefore a principal aim of this scoping review was to scrutinize the peer-reviewed literature spanning a decade, from 2014 to 2023, with a primary focus on applying transcriptomic findings to clinical sepsis. In order to translate facets from the research literature and interdigitate these with gene expression studies, a framework is presented (Figure 1). An enhanced framework holds potential clinical advantage, such as paving the way for guideline development, standardizing care pathways and advancing the principles of precision medicine. The framework as illustrated uses key aspects crucial in establishing evidence-based sepsis guidelines. An important objective was to discern whether transcriptomic studies can lend support for an enhanced understanding sought by peer-lead sepsis committees. This work will have important implications for research as gaps shall be highlighted with implications for guidelines and clinical practice depending on the presence of relevant research literature.
Using transcriptomic information to support translating clinical sepsis research to the bedside. Key questions linking gene expression (GE) to the framework are shown (central boxes). These then support components of the research framework (Diagnosis/Definition, Disease Progression, Disease Severity, Biomarking, and Benchmarking). mRNA is thus vital for cellular function and consists of mRNA protein-coding and non-protein-coding RNA functions. The two facets allow mRNA to play a role in gene code translation for protein synthesis and a gene regulatory role. Essentially, mRNA is the genetic mediator guiding ribosomal protein synthesis based on information provided in the DNA. At this moment, transcriptomics aims to document gene activity by quantifying mRNA, analyzing gene expression patterns, and measuring gene levels in sepsis.
Gene-to-gene connections are shown, with genes illustrated as nodes. The interconnections between the genes, then, represent the regulatory relationship. Therefore the network interactions amongst the genes form a Gene Regulatory Network (GRN). Sepsis is a heterogeneous process impacted by host factors such as age, infection timing, and pathogen-associated factors.
Red and dark blue are clinical attributes; green pertains to a biological construct related to a cellular function or clinical end-point; yellow depicts a standard against which other parameters are compared. In light blue are target end-points which can be deduced from the data features, according to the generated (vector) of data-points for each patient.
1. The diagnosis and definition of sepsis are interrelated. Changes in cellular activity can be detected with respect to mRNA GE, providing patterns indicative of a pathophysiological response indicative of sepsis. Thereby resulting in GE patterns consistent with the diagnosis and definition of sepsis.
2. Sepsis progression is important because, if unabated, this can then progress to multi-organ dysfunction syndrome (MODS). Single-organ dysfunction in sepsis is rare, with the subsequent failure of each organ being associated with an increased risk of a poor outcome68. The Sepsis 3 task force concluded that the misleading model that sepsis follows a continuum from severe sepsis to shock was misleading17. Further, the task force concluded that the term severe sepsis was redundant.
3. Applying transcriptomic methods to sepsis is in predicting organ dysfunction. Scoring systems exist to help quantify the degree of organ dysfunction. The sequential Organ Failure Assessment (SOFA) score is used in adult sepsis and pediatric logistic organ dysfunction (PELOD) in children.
4. An essential aim of the transcriptomic analysis is to improve the application of clinical therapies in a more precise approach, mindful of host-pathogen complexity. The aim is for therapies to be tailored according to a specific profile or sepsis sub-type. Sepsis Subtyping may be undertaken from a gene function perspective, such as according to a distinct pathophysiological mechanism known as Endotypes.
5. Understanding how the transcript correlated to disease severity allows the linkage of mRNA GE, a proxy of cellular function, to clinical categorization.
Clinical categories of different severity levels include sepsis, severe sepsis, and septic shock. Relating the gene transcript to different levels of disease severity could allow insights into sepsis pathogenesis and provide an interpretation of the host-infection relationship.
6. The relationship of biomarkers to gene expression is of particular interest, especially from a temporal perspective allowing the tracking of sepsis, therefore, understanding disease trends when managing patients. This can be used in a predictive capacity and to pre-empt disease progression, thereby providing information to the clinician to make management choices.
7. The transcript may also have value in benchmarking sepsis, such as correlating to clinical variables, standards, and endpoints.
Transcriptomics provides the ability to enhance endpoint analysis, aiding in disease categorization/classification and with respect to prognostication.
One of the challenges in developing a clinical research framework for sepsis is that the components defining the framework may not be clear due to the lack of clarity in the original definition of sepsis. Therefore the likely overlap between components, for example, thought endotype, could allow the clustering of groups of patients; this approach may also have value as part of a biomarking strategy.
Materials and Methods
In accordance with the PRISMA-ScR guidelines 24, 25, a scoping review was conducted to investigate the use of systems-biology approaches in examining gene expression and its relation to clinical sepsis. The protocol was registered on the Open Science Framework (https://osf.io/3jbv2), with the associated project osf.io/5c2wr.
Identifying the research question
This study hypothesized that transcriptional research could support a clinical research framework based on components important for sepsis guideline design. The research question posed was: What gene expression studies can be utilized to capture relevant information for the development of sepsis guidelines concerning sepsis definition, classification, disease severity, molecular biomarkers, and benchmarking? This study incorporates gene expression investigations that span both coding and non-coding domains. The research encapsulates micro RNA studies (<200 NBP), along with the exploration of circular (Circ) RNAs and lncRNAs.
Study Selection
A systematic review of peer-reviewed literature was undertaken from PubMed-indexed journals utilizing PubMed’s online search tool, which incorporates MEDLINE, PMC, and BookShelf databases. Pertinent literature from a 10-year period leading up to the search date, 29th June 2023 was undertaken. The primary search title terms included ’gene expression’ and ‘Sepsis’ in conjunction with one of the thematic research terms (Endotype, Biomarker, Definitions, Diagnosis, Progression, Severity, and Benchmark) (Figures 2A-G). The derived search string is shown (Supplement Table 7). At the subsequent stage, articles were filtered to include human studies in English, whilst excluding review articles, drug and vaccine studies. As the key theme ’Benchmark’ yielded zero selections, the filtration step was expanded beyond the title for a full article search. This modification was also applied to the ’Endotype’ theme to secure a more comprehensive collection of studies.
Studies were first selected (10th July 2023) using the Pubmed web server according to selective keywords related to the framework headings (see methods). The identified studies were then screened, excluding review articles and including gene expression studies [Keyword Search Term: ‘(gene expression) not (review)’]. Human studies in the last ten years were then deemed eligible. Non-drug, vaccine, non-high throughput gene (HGT) studies and only research published in English were included in the final narrative analysis [Keyword Search term: not (drugs) not (vaccines) and (English)]. For the category ‘Sepsis Endotype’ the search strategy was changed as the title search only eluded to three studies; instead, a search through the text identified 64 studies, of which 17 were deemed eligible (2 were letters and comments to the editor and the third was a protein study). This left 14 included as HGT studies and for narrative review (Fig. 2A). In the category ‘Sepsis Biomarker’ 409 studies were identified after screening; this filtered the studies to 30, of which 23 were eligible, and six were included after exclusion (Fig. 2B). In the ‘Sepsis Definition’ category 125 studies were identified of which non were eligible after screening(Fig. 2C). For ‘Sepsis Diagnosis’ 19 papers were deemed eligible, which after the exclusion, led to 15 studies of which only 7 were HGT studies (Fig. 2D). For ‘Sepsis Progression, ’ 124 studies were identified, 17 after screening, of which only nine were eligible, and two were included for narrative analysis (Fig. 2E). For ‘Sepsis Severity,’ 19 papers were eligible; after exclusion, this eluded 2 HGT studies (Fig. 2F). For ‘Sepsis Benchmark’ the search strategy was changed to extend the search strategy through the body of the document (Fig. 2G).
This figure explores the relationship between changes at the cellular level (left) to various data parameters (right) and their relationship to the disciplines of statistics for biological interpretation and Machine learning for data modeling is shown.
A. Sepsis causes a multi-cellular response with associated activation of a complex immunological response. Various technologies are available to assess mRNA changes across the transcriptome (microarray, RNA-seq, sc-RNA-seq). This provides a holistic view of various processes and is advantageous over single-pathway approaches.
B. However, given the heterogeneity of sepsis, the complex situation is challenging to analyze and develop from an omic perspective. Correlating clinical data to an omic perspective, such as gene expression, is complicated by the non-linear relationship of data points.
C. Statistical method is required to adequately map biological changes at the cellular level, using gene expression data, to a bioinformatic interpretation. This has allowed the introduction of sepsis biomarkers. In an advancement from biomarking, the concept of endomarking allows the grouping of data according to functional biological characters. Statistical analysis requires a framework interpretation, which likely involves a degree of understanding of the data for simplification.
D. Machine learning is useful as a modeling technique as it does not require a deep understanding of the inherent data structure. It is possible to apply various algorithms to the gene expression information and use various methods to train the data to develop a certain model. The model can then be tested with retrospective data and then validated prospectively.
Study Inclusion and Exclusion Criteria
Articles were included in the review if they met three criteria: (1) focused on sepsis as the primary disease process and incorporated transcriptional (mRNA) analysis; (2) addressed one of five clinical areas of sepsis, including case definition, classification, Sepsis Severity Endotyping, Biomarkers and Benchmarking; and (3) involved human subjects.
Exclusions comprised conference abstracts and articles lacking full-text access or were not available in the English language. AR and JA independently screened the articles, resolving disagreements through discussion until consensus. Relevant information on applying transcript studies to sepsis was extracted and tabulated from the selected article using the Artificial Intelligence (AI) engine ‘Bing’ incorporated in the Microsoft Edge browser. The words “summarise the paper into header, methods, results and conclusions” were input into the AI engine, and the information generated was tabulated (Supplement: Tables 1 to 6). However, additional human scrutiny was applied to the tabulated studies with the discussion by the authors in the main text of the narrative review.
Data Charting
We developed a data extraction form using Microsoft Excel (AR). Two independent reviewers (AR and JA) extracted data from the full-text articles to ensure consistency. The extracted data included population and study characteristics (e.g., demographics, aim, transcript information, significant genes, study outcomes, and conclusions).
Data Collation and Result Reporting
The descriptive quantitative analysis undertaken in this study involved aggregating the number of articles based on key words in accordance to the aforementioned sepsis framework (Figure 1). The acquired studies were partitioned into two main framework categories. The first category encompassed concepts associated with cellular changes related to sepsis, such as endotypes and biomarkers. The second category addressed elements more directly related to the clinical presentation of sepsis, including its definition, diagnosis, progression, and severity. The results are subsequently presented according to this sequence, with a discussion of the relationship between the framework terms and gene expression.
Results
The sepsis framework search terms identified 2,333 articles (Figure 2) and are detailed below (see headings) according to the framework search terms. A duplicate occurred once with the ‘Endotype’ and ‘Diagnosis’ searches and was recorded in both these sections.
The Sepsis Endotype
"Endotype," derived from "endogenous phenotype," refers to disease subtypes defined by distinct biological mechanisms rather than just observable symptoms. Leveraging transcriptomic analysis can identify unique endotypes, transforming sepsis diagnosis and prognostics based on molecular characteristics and mechanisms.
Several studies have explored sepsis endotypes and mortality. Zhang et al. (2020) used deep learning to identify two sepsis classes - immunosuppressed Class 1 with higher mortality and relatively immunocompetent Class 2 with elevated mortality risk from hydrocortisone therapy 26. The VANISH trial found SRS2 endotype was associated with higher mortality when treated with corticosteroids 27. Pediatric sepsis had two endotypes (A and B), with Endotype A associated with higher 28-day mortality rates in patients with acute hypoxemic respiratory failure (AHRF) 28.
Transcriptomic analysis is valuable in understanding sepsis pathogenesis. Baghela et al. (2022) demonstrated gene expression signatures accurately predicting sepsis severity and identifying mechanistic endotypes in early sepsis 29. Got et al. (2020) linked sepsis-induced Epstein-Barr virus (EBV) reactivation to an immunosuppressed host transcriptomic endotype 30. Kwok et al. (2023) investigated neutrophils and emergency granulopoiesis in sepsis, finding altered gene expression in sepsis patients’ circulating hematopoietic stem and progenitor cells 31. Darden et al. (2021) used single-cell RNA sequencing to reveal the role of non-myeloid cells in chronic critical illness (CCI) and persistent inflammation, immunosuppression, and catabolism syndrome (PICS) after sepsis 32.
Combined with gene expression data, machine learning is a promising tool for identifying sepsis endotypes and improving prognosis. Sweeney et al. (2021) classified patients into Inflammopathic, Adaptive, and Coagulopathic endotypes, which are significantly associated with clinical outcomes and guiding personalized therapy 33. Banerjee et al. (2021) used Machine Learning to identify 20 differentially expressed genes predicting sepsis severity and outcome 34. Scicluna et al. (2017) identified four molecular endotypes (Mars1– 4) in sepsis patients, linked to severity scores, septic shock, and mortality 35.
Endotype-based research continues to evolve, supporting the potential for personalized sepsis management. Wong et al. (2012) developed ’PERSEVERE,’ a sepsis outcome prediction tool 36, while Lu et al. (2022) identified eight hub immune-related genes for sepsis diagnosis and prognosis 37. Baghela et al. (2023) utilized blood sepsis gene expression signatures to predict severity and endotypes in COVID-19 patients 38. They determined five endotypes reflecting distinct sepsis etiologies and therapeutic possibilities.
Efforts are underway to consolidate dysregulated gene sets linked to sepsis in the form of a library, such as ’SeptiSearch,’ a compendium of 103 unique gene sets developed by Baghela et al. (2023) 39. SeptiSearch includes a description of certain endotypes, thus included in this section. In summary, endotyping, along with machine learning techniques, holds significant promise for advancing sepsis management and delivering personalized care to patients.
Sepsis Biomarker
Biomarkers, especially those derived from gene expression, are becoming increasingly vital in diagnosing and monitoring sepsis, providing measurable indicators of disease presence or severity. Among them, Zheng et al. (2020) distinguished bacterial and fungal sepsis via specific gene sets, introducing the bacterial sepsis Gene Set Variation Analysis (GSVA) index that exhibits remarkable discriminatory power between bacterial sepsis and non-sepsis samples 40. Moreover, Zhang et al. (2022) spotlighted the potential of ARG1 as a biomarker for diagnosing and prognosticating sepsis, linking ARG1 expression to disease severity and treatment response41. MicroaRNAs also demonstrate their utility in this field, with Huang et al. (2014) identifying eight novel miRNAs associated with the early diagnosis of sepsis 42, and Li et al. (2022) showcasing the diagnostic and prognostic value of BCL2A1 as a novel biomarker for sepsis management 43.
In a pivotal study, De Almeida et al. (2023) identified genes that linked Non-Thyroidal Illness Syndrome (NTIS) and sepsis, providing a critical insight into shared molecular mechanisms. Importantly, certain mitochondrial genes (mitGenes) stood out as potential survival prediction biomarkers. These mitGenes could differentiate between sepsis survivors and non-survivors, underlining the significant role they could play in sepsis endotyping. Among these, ROMO1, SLIRP, and TIMM8B emerged as potential predictive biomarkers of mortality in pediatric sepsis patients 44.
Transcriptomic biomarker panels also offer promise in the management of sepsis. Bauer et al. (2016) developed such a panel, which effectively quantified systemic inflammation and immune dysfunction in sepsis while also differentiating infected patients from those without infection. Additionally, this panel associated a down-regulated component of the genomic score with mortality 45.
Sepsis Definition
The sepsis definition serves as a crucial basis for conveying research and clinical findings, with few studies taking on this challenge head-on. The Sepsis-3 committee acknowledged the complexity involved in matching the clinical physiological approach to the initial immunological changes in sepsis, noting that ambiguity in the definition could lead to inconsistent mortality reporting17.
The Sepsis-3 definition tackled this challenge by designating sepsis as a syndrome, recognizing the absence of a definitive diagnostic test. While the adult Sepsis-3 definition attempts to embody the intricacies of sepsis, it remains vague. It underscores the dysregulated immunological aspects of sepsis without fully being able to encompass the intricate and complex details. Transcriptomic perspectives could shed light on functional alterations in sepsis. For instance, Schaack et al. (2018) found distinct sepsis patient clusters exhibiting varying degrees of T-cell and monocyte functional loss, alongside dysregulated granulocytic neutrophil activation 46. Reyes et al. (2020), employing scRNA-seq analyses, identified 16 unique immune cell states, indicating that a transcriptomic functional interpretation of sepsis might aid in understanding its dysregulation. Nonetheless, a unifying immunological pattern that defines sepsis across various studies remains elusive47.
The pursuit of a comprehensive sepsis definition that encompasses age and pathogen type is a complex endeavor. Wynn et al. (2011) posited that age-related differences exist in septic shock, as neonates exhibit diminished gene expression in crucial immune-related pathways unlike other age groups48. This revelation prompts doubts about the viability of a universal, age-independent definition. Regarding the relationship of a sepsis definition according to pathogen type, research on SARS-CoV-2 has underscored the parallels between bacterial sepsis and COVID-19 dysregulated mechanisms. Karakike et al. (2021) reported that most ICU-hospitalized COVID-19 patients satisfied the Sepsis-3 criteria49.
Further, the emergence of SARS-CoV-2 has redirected research focus towards viral sepsis, unveiling shared features between bacterial sepsis and COVID-19 dysregulated mechanisms 50. Additionally, Sohn et al. (2020) suggested that the immune-related transcriptome profiles of COVID-19 patients mirrored those in bacterial sepsis, advocating for a pathogen-agnostic innate host response 51. Furthermore, Barh et al. (2020) showed that transcriptome studies of lung tissue post-SARS-CoV-2 infection revealed shared pathways with bacteria, parasites, and protozoa 52. These findings hint at the possibility of a pathogen-agnostic sepsis definition, even though its actualization remains riddled with hurdles 50. Innovative methods like transcriptomic approaches discussed in this review might be instrumental in bridging the gap in sepsis understanding, heading to an improved definition of sepsis.
Sepsis Diagnosis
Kalantar et al. (2022) found that host gene expression from whole blood and plasma from 221 ICU patients could accurately differentiate sepsis from non-sepsis53. They used machine learning to develop classifiers based on host gene expression and pathogen detection. Combining host and microbial features improved sepsis diagnosis and predicted sepsis in patients with negative or indeterminate microbiological testing. Lukaszewski et al. (2022) identified specific gene signatures predicting infection or sepsis three days before clinical presentation54. Machine learning techniques accurately distinguished infection from uncomplicated recovery and sepsis from other postoperative presentations. Also, Xu et al. (2022) showed that microRNAs combined with TLR4/TDAG8 mRNAs and proinflammatory cytokines had utility as sepsis diagnosis biomarkers for early sepsis diagnosis55. Zhou et al. (2021) developed a 10-core gene expression panel for diagnosing pediatric sepsis, with ROC showing an AUC of the 10 core genes for diagnosing pediatric sepsis above 0.9 56. Given the importance of the immune system in sepsis, Lu et al. (2022) focused on immune-related genes (IRGs) and their association with sepsis diagnosis and prognosis 37. Here, machine learning approaches identified hub IRGs from multiple datasets, establishing an IRG classifier based on 8 hub IRGs, which showed superior diagnostic efficacy and prognostic value compared to clinical characteristics alone (see the section on endotyping). The study also correlated the IRG classifier with immune-related characteristics, such as immune cell infiltration and cytokine expression. Sweeney et al (2018) validated a gene-expression test, the Sepsis Metacore (SMS), for sepsis in neonates. The SMS was accurate in three different cohorts of neonates with sepsis, and better than standard laboratory tests. Thereby suggesting that the SMS could help reduce unnecessary antibiotic use and improve outcomes for neonates with sepsis57.
Sepsis Progression
The capacity to anticipate sepsis complications early, based on gene expression profiles, could offer the prospect of disease modification, allowing for individualized and targeted therapies. Fiorino et al (2022) undertook a prospective observational cohort study of 277 patients with infection, sepsis, or septic shock58. They used RNA sequencing of whole blood to measure the host gene expression response to infection and to identify signatures that could predict sepsis progression and mortality. The researchers found no gene expression signature for sepsis progression defined by the Sepsis-3 category, but they found signatures for sepsis progression defined by new organ dysfunction or ICU admission/mortality. They also validated four previously published gene signatures for sepsis mortality. By comparing the gene expression patterns of patients who progressed to more severe forms of sepsis or died within 28 days with those who did not, the authors identified genes and pathways associated with sepsis progression. Thus showing the ability to label sepsis progression based on host gene expression as a biomarker of the host response to infection. The elicited genes and molecular pathways could reflect different mechanisms (endotypes) of sepsis progression. The authors also used predictive modeling to generate gene expression signatures that classify patients into risk groups according to sepsis progression and mortality. Here a molecular score was provided according to g to the gene expression signature elicited, complementing clinical parameters to guide personalized approach to clinical care. Glibetic (2022) used transcriptomic analysis to identify patient subgroups with altered biological responses to sepsis, investigating the ethnic basis for viral infection risk and sepsis progression in colorectal cancer (CRC) patients59. Their analysis revealed distinct sepsis gene signatures classified as early and late response sepsis genes in the Native Hawaiian cohort compared to Japanese patients. Furthermore, canonical pathway analysis showed significant up and downregulation of mechanisms related to viral exit from host cells and epithelial junction remodeling. These findings suggest that genetic background plays a crucial role in sepsis heterogeneity, which could enable personalized approaches for risk stratification and targeted therapies.
Sepsis Severity
De Jong et al. (2021) developed an innovative method to scrutinize disease-associated molecular changes using gene ensemble noise60. This measure, which represents the variance of gene groups, disrupts the conventional gene regulation model. The authors argue that cellular dynamics aren’t simply responsive to gene up- or down-regulation. Instead, they suggest the stochastic nature of gene expression impacts cellular responses. This approach allowed them to identify disturbances in sepsis-relevant pathways and protein complexes. They also noted its successful application to H1N1 infection and sepsis mortality. Their model predicted patient survival post-sepsis, and they incorporated WGCNA, emphasizing its value in understanding non-linear relationships. This approach predicted COVID-19 disease severity, revealing potential pharmaceutical targets.
In a parallel study, Baghela et al. (2022) sought gene expression signatures for sepsis severity and endotypes upon initial clinical presentation29. They proposed that sepsis is a syndrome comprising different endotypes, each representing a distinct group with unique severity and outcomes. Using whole blood RNA-Seq and machine learning, they analyzed gene expression profiles from ER and ICU patients with suspected sepsis. Their analysis identified five distinct endotypes with diverse underlying mechanisms. Two of them were associated with high severity and mortality, while one showed benign characteristics. The researchers developed a classification tool based on a multinomial regression model with LASSO regularization. This model, built on 40 genes, accurately predicted endotype status in sepsis patients and could be instrumental in early triage, potentially guiding personalized therapies. The study shed light on sepsis as a heterogeneous syndrome and provided valuable insights for predicting endotypes in sepsis patients.
Sepsis Benchmark
Benchmarking, the process of comparing a system’s or method’s performance against a recognized standard, plays a crucial role in clinical research, especially in sepsis studies 61. However, the core challenge is identifying the correct standard against which to benchmark. One solution involves classifying gene patterns based on disease conditions or processes, thereby creating a reference library of gene patterns. In this vein, Altman et al. (2021) developed a transcriptomic benchmarking framework called BloodGen3Module, designed to facilitate the analysis of gene expression data62.
In a notable study by Sweeney et al. (2017), three gene expression diagnostic classifiers, namely the 11-gene Sepsis MetaScore, FAIM3:PLAC8 ratio, and the Septicyte Lab, were tested on 39 publicly available sepsis datasets 63. The objective was to determine how well these classifiers could distinguish patients with infection from those with non-infectious inflammation. The three diagnostic classifiers performed similarly in separating non-infectious SIRS from sepsis, but the Septicyte Lab performed less well in separating infections from healthy controls. In a subsequent study, Sweeney et al. (2018) conducted a validation study of the Sepsis MetaScore for diagnosing sepsis in neonates, demonstrating its superior performance over standard laboratory measurements across three distinct cohorts 57.
Another significant contribution to sepsis benchmarking was made by Scicluna et al. (2020). The authors carried out a next-generation microarray analysis of leukocyte RNA from 156 patients with sepsis and 82 healthy subjects, eight of whom underwent a lipopolysaccharide challenge in a clinically controlled setting, a process known as human endotoxemia 64. The study found significant alterations in long non-coding RNA and, to a lesser extent, small non-coding RNA, in sepsis patients compared to healthy subjects. Moreover, their results highlighted the potential relevance of non-sensory olfactory receptor activity among other pathways, and suggested that long non-coding RNA profiles in sepsis could serve as a benchmark for future studies.
While sepsis benchmarking remains invaluable, it is not without challenges. The need for well-labeled transcriptomic data and the necessity of a universally accepted definition of sepsis across studies are significant hurdles. Nonetheless, sepsis benchmarking continues to be a vital tool in the evaluation of gene transcriptomics tools and methodologies, serving to guide researchers in their quest for effective prediction methods, biomarker discovery, and the development of novel therapeutics.
Discussion
A system-wide approach, using transcriptomic analysis, was undertaken connecting conventional sepsis themes as a part of ten-year literature scoping review. Hence, a transcriptomics-oriented approach was proposed using gene expression studies spanning various categories—diagnostic, organ dysfunction, sepsis severity, endotyping, classification, biomarking, and benchmarking. In the presented framework, the terms were broadly grouped into two categories. The first provides scientific insights based on changes at the molecular or cellular level (Biomarkers and Endotypes). The other category provided more high-level insights appropriate to a clinical understanding. The interplay between scientific (biological) or clinical terms to features of genomic data seemed intuitive.
Biomarkers and endotypes formed components of the scientific category. A biomarker refers to objectively measurable signs, such as molecules found in the blood or changes in body function, that indicate biological or pathogenic processes or responses to therapeutic intervention 65. Conversely, an endotype is employed in classification, highlighting the connection between a disease and a distinct pathophysiological mechanism66. Biomarkers have proven their worth in sepsis studies, providing a convenient and objective method for disease observation, tracking, and patient response monitoring67. Nonetheless, endotypes could offer a more profound bridge between the biological and clinical contexts. Endotypes, representing distinct underlying disease pathways and mechanisms, are typically identified using advanced diagnostic techniques like genetic analysis, molecular profiling, and biomarker identification. By discerning specific gene expression patterns or molecular signatures unique to a patient subgroup within a broader disease category, endotypes enable more personalized treatment strategies. Continuous research and technological advancements promise to refine our understanding of endotypes, heralding a future of increasingly personalized and effective medical care.
The disparity between scientific insights in sepsis and their clinical application remains a limitation across sepsis studies in general. Factors contributing to the difficulty in transferring insights across RNA-based studies, including dealing with the variability due to platform heterogeneity, sample collection timing, and host-pathogen differences, including demographic aspects. These factors make the linkage of different studies of sepsis problematic. Most importantly, a clear definition of sepsis is a major obstacle in creating an effective research framework. This may be the reason for a further limitation of this study due to the blurred boundaries between research terms, especially those related to clinical descriptors such as sepsis severity and progression. Moreover, although there is abundant research on the use of the chosen framework terms, the lack of a universally accepted structure to ensure their clinical relevance added to the complexity.
Future development should bridge the two domains, scientific and clinical, in forwarding sepsis research. This would facilitate the application of omic approaches to clinical practice. Then this could allow a more seamless application of omics-based methods, such as transcriptomics, to fill knowledge gaps. Hence the application of high-resolution metrics based on gene expression technologies for personalized diagnostics in sepsis should remain a future research goal. Moreover, given the likely highly non-linearity of gene expression data, future research could explore the application of artificial intelligence to further address the issue of ambiguity. Advanced algorithms and machine learning techniques could help interpret complex gene expression data and other omics information, offering further clarity and precision to disease classification and treatment strategies. These strategies may be instrumental in overcoming the inherent challenges and moving closer to more precise and personalized care for sepsis patients. Advancements in defining sepsis and identifying biomarkers can positively impact benchmarking, thereby extending applicability across various clinical sepsis realms. Moreover, the exciting development of technologies towards faster processing of high throughput gene expression data may allow the application of transcriptomics in acute sepsis. Importantly, gene expression methods, as detailed in this study, hold the future promise of validating a consensus-driven approach for sepsis management, bridging existing knowledge gaps in sepsis pathophysiology and allowing the refinement of clinical protocols using precision strategies.
Conclusion
This narrative review underscored the potential of a transcriptomics-oriented approach as a pivotal tool for bridging knowledge gaps between pathophysiological changes, and cellular modifications applied to the clinical context. The transcriptomics-oriented approach used gene expression studies spanning various sepsis categories—diagnostic, progress, severity, endotyping, classification, biomarking, and benchmarking. This revealed the application of transcriptomics across numerous aspects of sepsis, offering promising avenues for integration with other omic strategies and interpretive frameworks. Nonetheless, the inherent complexities and interpretive challenges of sepsis persistently echoed throughout the review. Future omic research should look at sealing the gaps between biological changes translated to the clinical context. By adopting a transcriptomic-driven approach, researchers and clinicians can collectively navigate the intricacies of sepsis, directing future progress and facilitating improved patient outcomes from sepsis.
Funding
No Funding requirements
Author Contributions
Conceived and designed the methodology: AR
Performed the experiments: AR
Analyzed the data: AR, PAC
Contributed analysis, methods, and tools: AR
Wrote the first draft of the paper: AR
Supervision: MGS, AA, AH
Revised critically for importance and intellectual content: AR, KW, HA, BH, PK, AS,CK, BSB, MT, ZH, HV, GB, ZM, RN, RM, SD, NK, RK, AS, PW, MGS, JS, SAZ, RP, MT, WZ, MAZ, HS, AA, AH
Data Availability
All cited literature has been used for this scoping review and is available as per the references provided according to journals in the public domain.
Details the search results according to the chosen sepsis framework search term ‘Endotype’.
* Literature Reference
** This paper appears in ‘Endotype’ and ‘Diagnosis’ searches.
Details the search results according to the chosen sepsis framework search term
‘Biomarker’.
* Literature Reference
Details the search results according to the chosen sepsis framework search term ‘Diagnosis’.
* Literature Reference
** This paper appears in ‘Endotype’ and ‘Diagnosis’ searches
Details the search results according to the chosen sepsis framework search term
‘Progression’.
* Literature Reference
Details the search results according to the chosen sepsis framework search term ‘Severity’.
* Literature Reference
Details the search results according to the chosen sepsis framework search term ‘Benchmark’.
* Literature Reference
Table shows the different level of filtering applied to the different search terms. For sepsis Endotype and Benchmark the search terms were expanded to include the full body of the text. This was in order to expand the search and ensure inclusion of all relevant inform
Acknowledgments
The authors thank the anonymous reviewers for their insightful comments and suggestions. Nuha Kidwai, North London Collegiate, Dubai for online search term testing. Hussain acknowledges the support of the UK Engineering and Physical Sciences Research Council (EPSRC) - Grants Ref. EP/M026981/1, EP/T021063/1, EP/T024917/1. For Dr. Binu George for keeping us on track with the study goals. To Professor Hector Wong, a pioneer in the field of transcriptomics of sepsis; to his enduring contribution to the field.
Footnotes
The search process has been updated with a more focused research question
References
