BGPT: Paper Review: Colibactin genes are highly prevalent in the developing infant gut microbiome

Fuel Your Discoveries

Quick Explanation Copied

Core finding: Using shotgun metagenomics in two longitudinal infant cohorts, the paper reports that pks+ (colibactin-associated) carriage is common in early life—peaking in the first 1–2 years—and is higher in NICU infants, with associations to early antibiotic exposure.

Key uncertainty: the study measures gene presence/abundance, not confirmed active colibactin production or human mutagenesis.

Long Explanation

Paper Review (Skeptical, Evidence-First): “Colibactin genes are highly prevalent in the developing infant gut microbiome”

DOI: 10.1080/19490976.2025.2604874

1) What the paper actually did (methods, observable quantities)

Design: serial stool sampling over the first 2 years from two cohorts: full-term (55 infants; 275 samples) and NICU-heavy (128 infants; 877 samples). The paper evaluates (i) pks+ sample prevalence and (ii) colibactin gene abundance in metagenomic data.
Sequencing / data processing (high-level): Illumina NovaSeq 2×150 bp shotgun metagenomics; trimming with fastp; human read removal; for environmental samples decontamination with Ganon2; assembly with metaSPAdes; gene prediction with Prodigal; read alignment with bbmap to quantify reads per million (RPM) across 19 colibactin genes; then mapping predicted genes back to a colibactin reference set and taxonomically annotating colibactin-bearing reads.
Operational definition of pks+: the paper uses a previously established criterion: a sample is “pks+” if it has at least one read in >8 colibactin genes.
Analyses: age-windowed prevalence calculations using repeated-sample adjustments; PERMANOVA for composition differences; Spearman correlations among genes; and generalized estimating equations (GEE) for longitudinal associations.

2) Main quantitative results (what’s “high prevalence”?)

The paper reports 31/55 (56%) full-term infants and 85/128 (66%) NICU infants having at least one pks+ sample during the first 2 years.

Age-pattern described by the authors

Full-term: prevalence “peaked at 6–12 months” with 38% of infants being positive, then declined.
NICU: prevalence was higher than full-term at 2 months (reported OR = 3.02; P = 0.047) and the paper describes a higher early-life baseline with a later trend.

3) Gene abundance and consistency among colibactin genes

The paper reports colibactin genes detected in 491/1152 (43%) samples, with a reported mean abundance of 342 RPM across samples.
Individual colibactin genes are described as being highly correlated (Spearman > 0.8), motivating use of a summed colibactin-gene measure.

4) Where are the genes coming from? Hosts and putative reservoirs

The paper reports Escherichia coli as the dominant colibactin host in both cohorts. It also reports that, in NICU infants who received antibiotics in the first 30 days, odds of colibactin being contained in Klebsiella pneumoniae were higher (OR = 3.49; P = 0.016).

Environmental reservoir check (NICU environment)

The paper reports that only 3/86 (3%) environmental samples exceeded the pks+ threshold (as defined by >8 genes with RPM criterion), making a large environmental reservoir unlikely within their sampled environment.

This plot intentionally encodes qualitative host findings the paper reported, without over-interpreting missing numeric host-abundance distributions.

5) Associations with antibiotics and maternal medication (correlation vs causation)

NICU: carriage in the first 6 months is positively associated with the total number of antibiotic courses infants received within the NICU (P = 0.001, plus a figure caption referenced).
NICU: days since receiving gentamicin is negatively correlated with colibactin abundance (P < 0.001).
Full-term: after FDR correction, the paper reports no significant associations with delivery mode, breast milk, or antibiotic use in that cohort (interpreted with the caveat that full-term infants likely had minimal antibiotic exposure).
Maternal H2 receptor antagonists (NICU): colibactin carriage/abundance is reported to be higher in infants whose mothers used H2 receptor antagonists during pregnancy (P = 0.014), but the paper emphasizes there is no direct evidence for a mechanistic pathway.

The paper also cites broader systematic reviews/meta-analyses that antibiotic exposure is linked with colorectal tumor risk in some contexts.

6) Critical interpretation: what follows vs what does not

6.1 What the data support (high confidence)

Gene-level prevalence is high in early infant life in these cohorts, using an explicit gene-read threshold for a pks+ call.
NICU context is associated with earlier/greater gene carriage and with antibiotic exposure measures, within their observational framework.
Host inference is consistent with E. coli dominance and an additional signal involving K. pneumoniae under early antibiotic exposure in NICU infants.

6.2 What the data do not establish (and why it matters)

Presence ≠ activity. The study uses metagenomic reads mapped to colibactin genes; it does not directly demonstrate colibactin protein expression, active metabolite production, or DNA adduct formation in infant tissue. The downstream “mutagenesis imprinting” claim is therefore mechanistically plausible but not empirically demonstrated in this dataset.
Operational “pks+” depends on mapping and coverage thresholds. The pks+ definition is based on having reads in >8 colibactin genes. Mapping-based thresholds can be sensitive to read depth, gene similarity, and assembly/annotation choices; the excerpt does not provide sensitivity analyses (e.g., how varying thresholds changes calls).
Transmission inference is underdetermined. The paper examines limited parental samples and NICU environmental swabs; it reports low high-threshold environmental positivity and low parental pks+ in their NICU parental subsample, but explicitly concludes the source is unknown. A definitive transmission pathway would require more systematic parental/environment sampling.

6.3 Integrating with prior mechanistic literature (context for plausibility)

The paper’s conceptual link between colibactin-associated pks+ bacteria and colorectal carcinogenesis is grounded in prior studies that identify colibactin/pks-driven mutational signatures and associations with CRC.
Its geographic/age variation hypothesis is contextualized by a Nature 2025 report associating geographic and age variations in mutational processes in colorectal cancer.

7) Blind spots & counterpoints (skeptical checklist)

Generalizability across populations: cohorts are from the United States (full-term: Inova Health System; NICU: Inova Fairfax NICU), and the paper itself suggests prevalence may differ in non-industrialized settings, where CRC/early-onset CRC rates differ.
Mechanism still missing: gene prevalence does not specify whether colibactin is expressed under infant gut conditions. In mechanistic colibactin research, specific enzymes/intermediates (e.g., ClbL amidase) and activation steps are required for genotoxicity, highlighting that genetic presence is only the first link in a multi-step biochemical pathway.
Antibiotic association could reflect confounding by severity of illness in NICU: NICU exposures (antibiotics, hospitalization practices, feeding, device use) are correlated with illness and microbiome maturation differences. The paper notes NICU infants are often preterm and have delayed microbiome maturation, which could help explain higher early prevalence.
Definition sensitivity: if reads mapping to colibactin genes can occur via partial/variant loci, thresholding could inflate pks+ calls relative to functional toxin potential. The excerpt does not show robustness tests for alternative thresholds.

8) Reproducibility & data access (what you can verify)

The paper reports public data access: full-term cohort reads at PRJNA988496, NICU cohort reads at PRJNA1280936, and environmental samples at PRJNA417283. It also states it reports no original code.

Item	What’s reported	Verification impact
Raw data	SRA projects PRJNA988496, PRJNA1280936, PRJNA417283	High: enables independent reprocessing of reads and replication of gene mapping results (with careful pipeline matching)
Original code	Not reported	Moderate: replication depends on rebuilding methods exactly from descriptions
Analysis details	Pipeline + statistical models described in Methods text	Moderate/High: allows method-by-method replication

Paper-level evaluation metrics (BGPT skeptical scoring)

Novelty: 8/10 — high-value because it quantifies pks+ prevalence longitudinally across full-term vs NICU infant cohorts with a clear gene-based operational definition.
Scientific quality: 7.5/10 — solid metagenomic pipeline description and reasonably large sample sizes, but key mechanistic steps remain indirect (gene presence not activity), and the excerpt does not show sensitivity analyses for mapping/thresholds.
Generality: 6/10 — strong for early-life infancy ecology in these cohorts, but generalizing to global/non-industrialized settings requires direct evidence and the paper explicitly flags this need.
Usefulness: 7/10 — provides actionable prevalence baselines for designing future mechanistic and longitudinal “gene-to-activity-to-host-damage” studies.
Reproducibility: 8/10 — raw read access via SRA is provided; methods are described; but “no original code” reduces ease of exact replication.
Explanatory depth: 7/10 — explains observed prevalence ecology and correlates with antibiotic exposure; however, mechanistic “imprinting/mutagenesis” remains a hypothesis pending direct functional measures.

9) What would most disprove the paper’s core narrative?

Show that gene prevalence does not translate to colibactin activity under infant gut conditions—e.g., demonstrate no colibactin metabolites / adducts even when gene sets are present (requires functional assays, not provided here). The paper itself calls for studies addressing when/conditions lead to mutagenesis.
Demonstrate that antibiotic exposure association collapses after adjusting for NICU severity, gestational age/maturation proxies, and other correlated NICU practices—i.e., gene prevalence is not specifically linked to antibiotics per se but to correlated variables. (The excerpt shows associations but causal identification is not established.)
Reproduce low/non-existent pks+ prevalence in diverse cohorts using comparable mapping thresholds and sequencing depth—if prevalence is not consistently high across settings, the “normal carriage” framing becomes weaker.

Author reviews (bespoke BGPT links)

Feedback:

Updated: April 15, 2026