Transcriptomic predictors of outcomes in acute infection

Suspected acute infection is a major cause of hospital attendance in the UK. Acute infectious diseases that progress to sepsis, defined as life-threatening organ dysfunction due to a dysregulated host response to infection, lead to high patient morbidity and mortality. Conditions that mimic acute infections and sepsis also lead to delays in appropriate patient care and adverse outcomes given the low threshold for suspicion and initial treatment of infection in the emergency department (ED). Distinguishing between acute infection/sepsis and ‘sepsis mimics’ is difficult given the limitations of existing diagnostic tools which depend on clinical signs and symptoms and pathogen detection. Predictive gene expression signatures have demonstrated diagnostic potential in discriminating between bacterial and viral infections. However, this is an over-simplified dichotomisation of clinical reality, where the differential diagnosis for the patient presenting with suspected acute infection comprises multiple potential infectious and non-infectious aetiologies. Suspected acute infection in the ED is under-studied given the difficulties in obtaining both patient samples and consent. However the ED presents a unique opportunity to study the host response to infection before established sepsis, and provides a window for the development and application of improved diagnostics and early interventions. To address this question, the Bioresource in Adult Infectious Diseases (BioAID) study was established. BioAID is a registry and bioresource of unselected adult patients presenting to ED with suspected acute infection. The overall aim of this thesis was to use the BioAID cohort (n=1693) to investigate inter-individual variation in the host transcriptional response to suspected acute infection and leverage these differences to develop and validate a predictive gene expression model to aid clinical diagnostics in the ED. I developed an objective diagnostic algorithm using ICD-10 discharge codes and positive microbiology to assign patients presenting to ED with suspected acute infection to broad infection classes comprising respiratory infection, urinary infection, indeterminate infection, and no infection. Using differential gene expression and pathway analysis, weighted gene co-expression network analysis, and immune cell deconvolution techniques I examined the host transcriptomic response to infection in peripheral blood. I sought to understand how the presence of infection, site of infection, and type of pathogen affected gene expression and the implications of these for the underpinning biology. I subsequently applied a penalised logistic regression (least absolute shrinkage and selection operator, LASSO) approach to develop a multinomial predictive gene expression signature to distinguish between bacterial, viral, and no infection (but acutely unwell) states. Class imbalance and cost-sensitivity were addressed in the model given that not all classes were equally represented, and also to reflect the differing urgency to diagnose and institute therapy between the infection states as well as the availability of appropriate targeted treatments. This resulted in a 19-gene signature (Multinomial-19) which most effectively discriminated viral infections from bacterial and no infection states. Distinguishing between bacterial and no infection states remains a challenge, and likely reflects their shared immune response and pathways. Multinomial-19 was then validated in two independent BioAID cohorts and benchmarked against existing binomial signatures intended to discriminate between bacterial and viral infections. The immune dysregulation in sepsis can be characterised by the presence of maladaptive loops such as impaired innate immunity, disordered coagulation, and ‘functional neutropenia’. Evidence of these were observed in patients with acute infection who reached an adverse outcome, defined as intensive care unit admission or inpatient mortality. Transcriptional biomarkers based on the host response to infection can contribute to improved diagnostics of infection state. All three infection states can result in inflammatory organ dysfunction and failure through shared immune maladaptations, which may be mediated by common (bacterial-driven or sterile systemic inflammatory response syndrome, SIRS) or distinct (bacterial or viral SIRS/sepsis) pathways. Multinomial modelling, although challenging in methodology, can provide a more complete diagnostic assessment by offering a breadth of diagnostic categories. The rapid and accurate diagnosis of infection state provides a window to intervene with targeted and appropriate treatment, where pathogen and/or host can be targeted therapeutically, early in disease presentation. This has potential implications for improved patient care and overall reduced hospital length of stay.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2024

Authors dc:creator

• Shah, Dhupal
• Patel, Dhupal

Advisors dc:contributor.advisor

• Davenport, Emma
• Conway-Morris, Andrew

• Transcriptional biomarkers
• Predictive gene expression modelling
• Acute infection
• Emergency department

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/66dc96e9-1afe-4c4f-812d-1d610cdee82c/download
• http://purl.org/NET/rdflicense/allrightsreserved

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.119416

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/386010