By Anais Cottage-stone · 25 August 2026
On Wednesday, 26th November 2025, NeoNET AFRICA and the KlebNET Genomic Surveillance Platform held a joint seminar on Klebsiella pneumoniae genomics to support neonatal sepsis. Chaired by Professor Andrew Whitelaw (Stellenbosch University, South Africa), the session brought together four speakers on a single question: what genomic data can do for neonatal sepsis, and what has to be in place before it can do it.
Watch the full seminar recording here.
Dr Erkison Odih (London School of Hygiene and Tropical Medicine, UK), a postdoctoral fellow in Professor Kathryn Holt's group, opened with the contribution of nosocomial transmission to the burden of K. pneumoniae neonatal sepsis. The question matters because it determines where prevention effort should go: infections acquired in the neonatal unit are, in principle, preventable by infection prevention and control, while those acquired elsewhere are not.
Professor Kathryn Holt (London School of Hygiene and Tropical Medicine, UK) took up the vaccine question. Modelling work suggests a maternal K. pneumoniae vaccine could avert around 80,000 deaths each year globally, roughly 3.4% of all neonatal deaths and up to 6% in some countries. Whether that potential can be realised depends on whether a manageable number of antigens can cover the strains actually causing disease.
To find out, her team assembled a meta-analysis of 1,930 K. pneumoniae genomes from neonatal sepsis across 35 sites in 13 countries in Africa and South Asia, a collaboration spanning more than a dozen research networks. Since serological typing is not widely available, capsule and O antigen types were predicted directly from the genomes using Kaptive.
The picture is one of real diversity. More than 130 capsule locus types exist, and the five most common accounted for just under half of neonatal sepsis cases. Around 20 capsule types would be needed to cover roughly 73% of cases across regions. Better coverage with fewer antigens would mean either regionally targeted capsule vaccines or moving to O antigen vaccines, where effectiveness is not yet clear, or some combination of the two.
Two further findings stood out. The strains causing neonatal sepsis are highly resistant but not hypervirulent, which is a different problem from the one hypervirulent Klebsiella poses elsewhere. And roughly half of cases appear to be acquired through transmission within the neonatal unit, which puts infection prevention and control on equal footing with any future vaccine. The underlying data are explorable at klebsiella.shinyapps.io/neonatal, and the analysis is available as Stanton and Keegan et al, 2025, on MedRxiv.
Dr Dominique Anderson (South African National Bioinformatics Institute, University of the Western Cape, South Africa) addressed the gap that makes analyses like Professor Holt's so laborious. Collecting bacterial genomes alone is insufficient, because a genome without accurate demographic, clinical and epidemiological context cannot answer most of the questions worth asking of it.
Her team built a neonatal sepsis specific contextual metadata standard, developed in partnership with PHA4GE, on three foundations: a literature search to identify which fields are commonly reported and which existing standards could form the backbone, engagement with expert stakeholders to capture domain knowledge and ground the schema in real surveillance needs, and an ontology-based approach that formalises each concept and so allows the schema to interoperate with other standards.
The result is a data dictionary in which every field carries an ontology identifier, a definition and practical guidance, paired with a spreadsheet template. The fields are recognisably neonatal rather than generic: alongside sample collection and geography, the schema captures facility characteristics such as the number of neonatal beds, whether onsite neonatal surgical facilities exist, piped water availability and what healthcare-associated infection surveillance is in place, and host characteristics such as gestational age at birth, birth weight, delivery location and procedure, days in hospital before specimen collection, and health outcome at day three.
Practical design matters as much as the ontology work. The template uses dependent dropdowns so that answering one question narrows the next, autopopulates fields where it can, and allows free text under an "Other" option where the picklist does not fit. Dr Anderson closed with a call for community input, noting that user acceptance testing in real-world settings is the next step in refining the standard.
Dr Chante Brand (Division of Medical Microbiology, Stellenbosch University, South Africa) presented survey findings on laboratory capacity and sequencing practice across the continent, part of KlebNET work package 2 on protocols and data standards.
The survey ran for three months and drew 58 responses spanning research laboratories, clinical diagnostic laboratories and national public health laboratories, with the largest share from East Africa. Encouragingly, 46 respondents (79%) perform whole-genome sequencing in some form.
Beneath that headline the picture is more textured. Blood culture practice is split between manual and automated methods and between commercial and in-house media, with most laboratories processing between 11 and 50 blood cultures a month. Illumina and Oxford Nanopore are both widely used, whether internally or through external providers, and column-based kits dominate DNA extraction. Nearly a third of respondents reported difficulties with DNA quality or yield.
Bioinformatics is where the constraints bite hardest. Among those doing their own analysis, the tools are familiar and standard, including FastQC and Trimmomatic for quality control, SPAdes for assembly, Prokka for annotation, and ResFinder and MLST tools for resistance and typing. What limits them is infrastructure rather than expertise: power outages and inconsistent electricity, unreliable or slow internet, limited data storage, and limited access to high-performance computing.
Taken together, the four talks describe a sequence rather than four separate topics. Genomic data can tell us which strains cause neonatal sepsis, how much of it is transmitted within our own units, and what a vaccine would need to cover. Getting there depends on laboratories being able to generate and analyse the data, and on the metadata standards that make one site's genomes comparable with another's.
