What happened
The Georgia Institute of Technology study of accelerated whole-tooth replacement in Lake Malawi cichlids has completed public review at eLife and was published as the version of record on 24 September 2026 (Mubeen et al., eLife 15:RP110584, doi:10.7554/eLife.110584, open access). We covered the bioRxiv preprint on 10 September; the paper was posted there on 7 January 2026, sent for review at eLife on 30 January, and went through two reviewed preprint versions (20 April and 7 September) before the version of record. What makes this publication unusual, and useful for the field’s record, is that the three public reviews, the authors’ response, and the editorial assessment are published alongside the paper itself, so anyone can read exactly what the referees pushed on and what changed in response.
The eLife assessment rates the study “valuable” with “solid” strength of evidence, and then qualifies it: the evidence is “only partially supported, as the conclusions are primarily supported by computational inference, without experimental validation of key findings.”
What the review changed
All three reviewers made the same central demand: validate the single-nucleus RNA-seq inferences, whether cell-type annotations, trajectory claims, differentially expressed genes, or ligand-receptor pairs, ideally with RNAscope, immunostaining, or serial histology of plucked versus control jaw halves. No new wet-lab data were added. Instead, the authors leaned on their published cichlid histology, in situ hybridization, and pathway-perturbation work (they tally 29 figures with 273 panels across the cited papers), adopted the suggested semantic qualifier “putative cell types,” and strengthened the limitations section.
The one substantive new analysis is in mesenchyme. The revised Figure 4 and a new figure supplement identify a putative dental ectomesenchyme (DEM) population within the dental follicle, marked by co-expression of twist1, dnmt1, and runx2, overlapping stromal markers Lum, Col6a3, Aspn, and Vegfc. A CellChat comparison found significantly increased signaling from this population to cycling ameloblasts on the plucked side at day 0, enriched for SEMA4, EPHB, SLIT, and SPP1 pathways; the authors focused on day 0 because it had enough replicates and cells for a robust comparison. This stromal microenvironment result was the direct answer to Reviewer 1’s major point on supporting-cell populations.
One detail in our preprint coverage needs flagging here. The preprint described two mesenchymal progenitor domains, one in the dental follicle marked by twist1 and one in the dental papilla marked by twist2. The version of record no longer contains a twist2 papilla domain: the word twist2 does not appear, and the progenitor story is consolidated into the single putative DEM domain, with pseudotime starting points in the DEM, dental follicle, and dental papilla (Figure 4C). The papilla remains a trajectory origin, but the specific twist2 marker claim did not survive review.
What the review did not change
The quantitative core of the study is identical to the preprint: tooth gain runs about 3 to 4 times higher on the plucked jaw half across all three species, with roughly 60 percent of plucked-side teeth new after 15 days; the single-nucleus dataset resolves 27,114 nuclei into 25 cell types across four time points (60 minutes, 1 day, 3 days, 7 days). The temporally sequenced readout, an early immune and inflammation phase giving way to epithelial construction and later vascular and nerve pathfinding signals, stands. So does the version-of-record abstract’s pathway list: Collagen, BMP, MMP, Semaphorin, and Slit-Robo. What peer review established is the status of that readout: it is computational inference, and the authors now state explicitly that the inferred ligand-receptor pairs “represent hypotheses to be tested in future studies.” Nobody has yet shown, in this system, that any of the inferred signals are necessary for the accelerated replacement.
The repair-versus-regeneration caveat is now on the record
Reviewer 3 raised the sharpest conceptual objection: plucking is severe manipulation. Tooth remnants and unintended damage to surrounding tissue may trigger repair mechanisms rather than pure replacement, and oral-cavity manipulation loads the site with inflammatory stimuli that physiological exfoliation does not. The authors kept the word “plucking” (cichlid teeth lack roots and are numerous, they argue) and kept the title, but they conceded the substance: the revised limitations state that they do not directly document the histological progression of tissue responses across the time course or the degree of damage plucking causes at each sampled point, and they note the control side likely sees some bone remodeling and immune response too, only less of it. For a field that wants to read this model as a clean window on natural tooth replacement, that caveat belongs in every summary.
Where this leaves the third dentition question
The plucking paradigm itself, a locally confined threefold acceleration of one-for-one replacement that is conserved across species with wildly divergent dentitions, remains a genuinely manipulable vertebrate regeneration model, and both the reviewers and the authors point to functional follow-up (zebrafish tests of specific pathways, spatial transcriptomics across the plucking time course) as the obvious next step. For tooth germ bioengineering, the study still offers the field’s most detailed time-resolved parts list of a regenerating tooth. What it does not offer, even after peer review, is any mammal data, any human-relevant validation, or any experimentally confirmed causal signal. The record should say “valuable, solid, inference,” and now, on the journal’s own page, it does.
Provenance: verified against the version of record, its published reviews, and the author response at eLife; sourcing and method at /method/.