What was tested
A group led by Shubin Chen, with corresponding authors Yaofeng Wang, Duanqing Pei, and Jinglei Cai, reporting in Cell Regeneration (a Springer open access journal, Creative Commons BY, article 51, online December 2025), asked a practical laboratory question: can you keep tooth-forming mesenchyme tooth-forming once you take it out of the embryo and put it in a dish? The cells are mouse dental mesenchymal cells isolated from embryonic day 14.5 molar mesenchyme. Everything here is mouse, embryonic, and either in a dish or transplanted to an ectopic site. There is no human cell in this paper and no tooth in a jaw.
The result, in order
Fresh embryonic mouse dental mesenchyme can instruct an epithelial partner to build a tooth, but that ability collapses fast in ordinary serum-containing (FBS) medium: by day 4 the cultured cells form cysts, not teeth. The paper’s contribution is a serum-free medium, an N2B27 base, that holds the cells odontogenically competent for up to about 14 days, including one passage. That is the honest headline: a longer shelf life for competence, not a higher ceiling.
The efficiency numbers make the point and also guard against reading too much into them. Competence is scored by recombining the cultured mesenchyme with a separate epithelial layer and transplanting the pair under a mouse kidney (renal) capsule for three weeks, then counting how many grafts form a tooth-like structure. Fresh FBS cells at day 1 actually score highest, 71.79 percent (28 of 39), but FBS cells are useless by day 4. In the new medium, unpassaged cells hold up well across the two weeks: 83.33 percent (10 of 12) at day 1, 72.73 percent (16 of 22) at day 4, 79.37 percent (50 of 63) at day 7. The paper’s often-quoted 40.74 percent (11 of 27) is the harder case, cells that have been passaged once, and it is a measure of how much competence survives a passage rather than a success rate for making teeth. A second passage fails to proliferate at all. So the medium buys time and one passage, and no more.
The structures themselves need to be described precisely. They are ectopic tooth-like structures under a kidney capsule, carrying an ameloblast layer, an enamel space, dentin, and dental pulp, but sitting alongside bone and cysts. They are not erupted, not loaded, not in a mouth, and the mesenchyme cannot make them alone: an epithelial partner is required every time. In one striking control, a non-dental renal epithelium worked as that partner in 3 of 7 grafts (42.9 percent), which says the cultured mesenchyme carries the instructive signal, but N equals 7 and it does not remove the need for some epithelium.
The mechanism they pin down
The medium works, the authors argue, by keeping BMP4 low. Serum drives excess Bmp4, and that excess is what strips the cells of competence. Two clean tests anchor the claim. Adding the BMP receptor inhibitor dorsomorphin to FBS cultures partly rescues them (5 of 10 grafts form a tooth-like structure). Adding recombinant BMP4 back into the good N2B27 medium abolishes tooth formation entirely (0 of 10). That loss-of-function and gain-of-function pair is the strongest result in the paper. Single-cell and bulk RNA sequencing (about 70,000 cells) line up with it: the maintained cells turn up a mineralization program and an odontogenic program, while the serum cells turn up negative regulators tied to a runaway BMP state.
Two cautions belong on the mechanism. First, the paper’s title also credits retinoic acid and osteopontin, but those arms are weak: single-tooth rescues, 1 of 12 and 1 of 14 respectively, which the authors themselves call partial. The load-bearing lever is BMP4 dose, not those two. Second, the effect is a dose response, not an on/off switch. The authors are explicit that moderate BMP4 supports the maintenance state and only excess suppresses it.
Why the BMP4 result does not contradict the anti-USAG-1 route
This site’s flagship program, the anti-USAG-1 antibody TRG-035 (/programs/anti-usag-1-trg035/), points the other way on BMP: USAG-1 is a BMP antagonist, and neutralizing it raises BMP signaling to release a tooth in a living jaw. Here, raising BMP4 destroys tooth formation. It is worth naming that tension plainly rather than smoothing it over: in this assay more BMP means fewer teeth, and in the antibody work removing a BMP brake means a tooth appears.
What dissolves the apparent paradox is that the two act at different points of a non-monotonic dose curve, in different cells and stages. This paper is about keeping isolated embryonic mesenchyme paused and competent in a dish, a state that needs BMP kept low; too much BMP4 pushes the cells to differentiate or die and spends their potential. The antibody acts later, in an intact postnatal jaw, on an already-specified successional tooth held in check, where a local, timed rise in BMP (and Wnt) lets that tooth progress. The clean illustration is inside this very paper: USAG-1 (Sostdc1) is one of the genes that goes up in the competent, low-BMP condition. High USAG-1 here is a marker of the low-BMP state the cells need to stay poised, not a lever anyone is pulling. Same molecule, opposite sign, because the stage and the required BMP level differ. Neither result undercuts the other.
Where it sits, and what not to read into it
This is a bioengineered tooth germ result (/programs/bioengineered-tooth-germ/): it is classical epithelial and mesenchymal recombination with a subrenal readout, the whole-germ lineage, and its practical value is a reagent, a medium that widens the handling window for primary dental mesenchyme. It is not a third dentition result. The third dentition, a latent successional tooth released in a real jaw, is the USAG-1 story, and this paper does not touch that biology; conflating the two is the main way to misread it.
No general press has covered this paper, so there is no coverage to correct yet. When it is covered, the likely misread is “scientists grew new teeth”. Five facts bound that claim: the cells are mouse and embryonic, they needed an epithelial partner, the teeth grew ectopically under a kidney capsule, the structures are tooth-like and not erupted or functional, and there is no human data. The useful next steps are the ones the authors flag: a human cell source, a way past the one-passage proliferation wall, and eventually an orthotopic result. The state-of-the-field summary at /field/ does not change on this paper.
Provenance: every claim above traces to the article at the DOI cited, read from its open access full text, per our method at /method/.