Review Papers Against Their Own Results

Data source: counts reported for three 100,000-family sampling strategies (All Families, Non-singletons, Large Families) in the paper's Figure 5 and Supplementary Table; methodology uses MMseqs2 family clustering then prefix–suffix k=27 scan on RefSeq genomes and AllTheBacteria (2.4M) Kaul et al. sampling depth data

Interpretation: detection power (number of candidates) increases non-linearly with sampling depth; authors emphasize this as a major sampling bias and state their results are a conservative lower bound Kaul et al. sampling depth score

Paper summary: the 57.5 kb deletion producing yjcO-lysU is first visible around generation ~19,500 and fixes within ~500 generations; authors estimate selection coefficient s ≈ 0.065 from LTEE census data and classical sweep approximations (Kimura-based) — they use N_e ≈ 10^8 from Good et al. LTEE estimates and the fixation interval to compute s (Supplementary Note) Kaul et al. LTEE fusion fixation

Critical synthesis — strengths, limits, and high-confidence claims

• Strength — multi-scale evidence: the paper combines (i) explicit population simulations showing initial frequency (p_init) strongly predicts chance of functionalization, (ii) a concrete LTEE case with sequencing + RNA/Ribo evidence of expression, (iii) MTBC long-read analyses identifying two speciation-associated fusions, and (iv) a scalable, alignment-free survey across 2.4M genomes that identifies candidates — an unusually broad methodological sweep that supports the core claim that deletion-born fusions occur and can persist Kaul et al. methods summary
• Strength — conservative detection criteria: requiring exact 27-nt k-mer matches and DBSCAN1D multimodality makes false positives unlikely but increases false negatives; authors deliberately present a lower-bound catalog (explicitly acknowledged) Kaul et al. k-mer conservatism
• Limitation — functional evidence is weak for many candidates: most candidates lack biochemical/phenotypic validation; yjcO-lysU is expressed but shows no sequence evolution post-fixation and convergent deletions suggest benefit was from deletion rather than fusion product; MTBC fusions lack catalytic domains and need targeted functional work to show adaptive protein-level roles Kaul et al. functional caveat
• Limitation — reliance on sampling depth and assembly quality: detection requires both pre- and post-deletion states in dense within-lineage sampling and is hindered by contig breaks at repeats (common deletion junctions). The authors quantify this and show most bacterial proteins are singletons in RefSeq, so the reported set is conservative and skewed toward deeply sampled pathogens/model taxa Kaul et al. sampling bias
• Methodological caveat — ancestral state inference: Fitch parsimony on cluster labels can misclassify events when phylogenies are uncertain or outgroup choice is imperfect; the authors attempt outgroup rooting but parsimony remains sensitive to horizontal transfer and convergent insertions. This is acknowledged and mitigated by conservative filters but remains a potential blindspot Kaul et al. ancestral inference methods

Where I agree with the authors (high-confidence)

1. Deletions can create novel ORFs by fusing distal gene fragments — mechanistically obvious in gene-dense bacteria and supported by LTEE + MTBC case studies reported Kaul et al. case studies
2. Higher starting frequency (via hitchhiking on adaptive deletions) increases the probability that a nascent ORF persists long enough to accumulate beneficial mutations — simulation results are consistent with basic population genetics intuition Kaul et al. simulations
3. Their alignment-free prefix–suffix k-mer screen is a computationally efficient, practical tool to find recurrent structural variation at scale and can be generalized beyond fusion detection (MGEs, prophage, integrons) — the provided code and FM-index approach justify this claim Kaul et al. prefix-suffix method

Where the paper could be strengthened (concrete suggestions)

• Functional validation: prioritize a small set of candidates (e.g., yjcO-lysU-like events in other lineages or MTBC fusions) for biochemical assays (purified proteins, domain folding assays, localization, targeted knock-ins/knock-outs) to demonstrate protein-level function vs neutral persistence. Without this, claims about movement toward functionality remain suggestive.
• Relax k-mer exactness in a controlled sensitivity analysis: implement fuzzy-seed indices (e.g., spaced k-mers, minimizers with Hamming tolerance, or short gapped seeds) to quantify how many candidates are missed due to terminal point mutations; present ROC-like trade-offs between sensitivity and specificity.
• Phylogenetic uncertainty quantification: replace single-parsimony MRCA calls with probabilistic ancestral-state inference (stochastic mapping / Bayesian ancestral reconstruction) and bootstrapped trees to estimate robustness of 'split MRCA' assignments.
• Assess alternative explanations: perform targeted long-read re-assembly for key loci to rule out assembly artifacts or undetected insertions; more deeply quantify how many positive-distance peaks co-map to prophage/MGE signals versus true deletion-fusion signatures (paper partially does this; expand with false-discovery-rate estimates).

Actionable next-step experiments (concise)

1. CRISPR-assisted reconstruction: re-create the yjcO-lysU deletion and its parallel deletion in ancestral REL606 in the lab (CRISPR/Cas + recombineering) and test fitness across LTEE-like media to decouple deletion vs fusion benefits (compare deletion-only, fusion-only, and both) — falsifies whether fusion product contributes to fitness (see paper’s LTEE inference) Kaul et al. LTEE inference
2. Directed evolution on candidate fusion-carrying strains in the lab to observe whether fusions accumulate adaptive coding substitutions (sequence exploration) and whether dN/dS signatures converge to purifying selection — i.e., put the model to an experimental test.

Conclusions and confidence

This is a strong, well-documented, conservative paper that convincingly demonstrates deletion-born fusion events occur across bacteria, quantifies them conservatively at scale, and provides a robust computational tool to detect them. The main open question is functionalization: many candidates are at early stages (diversifying dN/dS signals but no conserved catalytic motifs), so claims about deletions as a generative source of adaptive protein-level novelty remain plausible but incompletely established without direct functional assay follow-up Kaul et al. conclusion