Clinical phenotyping and transcriptomics in young infants with suspected sepsis in Malawi : defining host-response endotypes

Severe infection remains a major cause of preventable death in young infants in low-resource settings. Clinical management is challenging because presentations are non-specific, microbiologic confirmation is frequently unavailable, and disease can be rapidly lethal. Clinicians must make high-stakes decisions soon after admission, with limited diagnostic support. The overarching goal of my thesis was to understand heterogeneity among infants with suspected sepsis in Malawi by integrating host-response molecular profiling with clinical phenotyping. I hypothesized that infants would demonstrate distinct host-response endotypes at presentation, and that these endotypes would correspond to heterogeneity in illness severity and in-hospital outcomes. This study is a prospective cohort of infants under three months with suspected sepsis, enrolled consecutively at Kamuzu Central Hospital in Lilongwe, Malawi. At presentation, infants underwent standardized clinical assessment, microbiology testing, blood sampling for transcriptomic profiling, and were followed until discharge or transfer. In total, 401 infants with suspected sepsis and 51 controls were enrolled, and 214 underwent whole-blood RNA-sequencing. In Chapter 2, I describe the cohort and the challenges of conducting prospective pediatric research among young infants in an LMIC referral setting. In Chapter 3, I show that bedside features associated with mortality clustered across physiologic domains, most prominently respiratory compromise, neurologic depression, and impaired perfusion, with additional contributions from feeding and abdominal signs. These findings support the value of structured admission assessment for identifying high-risk infants. In Chapter 4, analysis of a purposively selected, mortality-enriched RNA-sequencing subset identified three candidate endotypes associated with clinical presentation and outcome. These endotypes comprised (1) low innate activation with adaptive/interferon-leaning signals, (2) broad innate and antiviral suppression with metabolic-erythroid predominance, and (3) broad innate/interferon/interleukin activation with metabolic suppression. RNA-sequencing also distinguished microbiologically positive cases and infants with severe outcomes. Overall, my thesis shows that infants with suspected sepsis present with biologically distinct host-response states that map onto clinically distinct phenotypes. By linking clinical features with transcriptomic endotypes in a high-burden setting, this work provides a platform for understanding disease processes, informing future therapeutic hypotheses, and developing clinically feasible biomarkers of biological risk to support earlier recognition and improved management of young infants with suspected sepsis.