Wnt-directed CXCL12-expressing apical papilla progenitor cells drive tooth root formation

Abstract

The tooth root is a critical component of the tooth anchored to surrounding alveolar bones. Tooth root formation is driven by cells in the apical papilla (AP) that generate new dentin-forming odontoblasts at the root-forming front. Mesenchymal stem cells have been isolated from AP for regenerative use; however, how AP cells physiologically coordinate tooth root formation remains undefined. We find that CXCL12⁺ cells emerge in AP under hypoxic environments at the onset of tooth root formation. Using Cxcl12-creER-based cell-lineage analysis, we further find that CXCL12⁺ AP cells contribute not only to odontoblasts but also to cementum-forming cementoblasts of the elongating root, while showing plasticity to alveolar bone osteoblasts under regenerative conditions. Canonical Wnt inactivation inhibits odontoblast fates of CXCL12⁺ AP cells and induces substantial root truncation, with their aberrant fibroblast fates suppressed by TGF-β receptor inhibitor galunisertib. Therefore, CXCL12⁺ AP cells maintain odonto-cementogenic fates in a Wnt-dependent manner, identifying these cells as pivotal dental mesenchymal progenitor cells driving tooth root formation with substantial plasticity.

The tooth root is a critical component of the tooth. Here they identify root-forming CXCL12+ apical papilla progenitor cells that provide odontoblasts and cementoblasts in a Wnt-dependent manner, with plasticity to form alveolar bone osteoblasts during regeneration.

Introduction

The tooth root is a critical component of the tooth, embedded in the socket and connected to the surrounding alveolar bone through the periodontal ligament¹. The tooth root transmits mechanical forces to the surrounding bone and maintains the function of the periodontium, which is composed of the cementum on the root surface, the periodontal ligament, and the alveolar bone, constituting the socket wall^2,3. The tooth root associated with the periodontal attachment apparatus is a unique feature of higher vertebrates, including mammals, in which a stable form of a tooth was adopted⁴. The periodontal attachment is at least in part enabled by the cementum, which provides the insertion site of the periodontal ligament fiber connecting the tooth root to the bone⁵. A structurally sound tooth root and its supporting structure are essential for the tooth to perform its fundamental function in mastication for efficient energy intake, speech, and esthetics for efficient communication and growth of the lower face.

The formation of the tooth root is induced by the down-growing dental epithelium termed Hertwig’s epithelial root sheath (HERS), which functions as a signal center and orchestrates the differentiation of dental mesenchymal cells in the adjacent dental papilla (DP) and the dental follicle (DF). Generation of the root dentin is a leading step of tooth root formation, in which DP cells differentiate into odontoblasts that form the dentin on the inner side of HERS⁴. The formation of the cementum occurs secondary to the formation of the root dentin. Therefore, the apical region of the dental papilla – apical papilla (AP) – in proximity to the root-forming front, possibly contains a critical cell population that drives the elongation of the tooth root^6,7.

However, the identities of AP progenitor cells in vivo and the molecular mechanisms regulating their cell fates toward odontoblasts remain undefined. Previous studies have isolated populations of “mesenchymal stem cells” from the apical region of the human tooth root, termed stem cells from apical papilla (SCAP), which have been studied primarily for regenerative purposes^8–10; these cells are heterogeneous and have yet to be demonstrated to fulfill stemness properties. In addition, several studies have utilized in vivo lineage-tracing approaches based on tamoxifen-inducible creER lines such as Gli1-creER¹¹, Osx-creER¹², and Axin2-creER¹³ to unveil the function of root-forming progenitor cells. However, a significant limitation of the prior research is that these lines simultaneously mark a broad spectrum of cells, including DF cells, dental epithelial cells, odontoblasts, and other mineralizing cell types^14,15. Our previous study unraveled an essential function of DF cells based on a DF-specific Pthrp-creER line¹⁶, which does not mark DP cells and odontoblasts. An AP-specific creER line would facilitate understanding the function of root-forming progenitors residing in DP.

From the signaling perspective, tooth root formation requires active canonical Wnt signaling, as inactivation of Wnt signaling using mineralizing cell type-specific Oc-cre or Col1a1-cre leads to complete failure of root formation^17–19. In addition, DP cells abundantly express chemokines and cytokines, including CXCL9, CXCL12 and CCL2^20,21. SCAPs abundantly express Chemokine C-X-C motif ligand 12 (CXCL12)^22,23. In the bone marrow, CXCL12 is expressed by reticular stromal cells termed CXCL12-abundant reticular (CAR) cells, which include precursors of osteoblasts and marrow adipocytes^24,25. A subset of CAR cells can contribute to the regeneration of the cortical bone in a manner controlled by canonical Wnt signaling²⁶. It remains unknown if CXCL12 expression can also define AP progenitor cells and how canonical Wnt signaling regulates their cell fates toward root dentin odontoblasts.

In this study, we sought to define the fundamental characteristics of tooth root-forming progenitor cells in the AP and how these cells coordinate tooth root formation. Our findings reveal that CXCL12⁺ cells in the AP continue to differentiate into root-dentin odontoblasts and cementoblasts in a canonical Wnt signaling-mediated manner during root elongation, identifying these cells as functionally critical dental mesenchymal progenitor cells that drive tooth root formation.

Results

CXCL12⁺ cells emerge in the hypoxic apical papilla at the onset of tooth root formation

We first sought to define markers for tooth root-forming progenitor cells of the apical papilla (AP) that emerge at the onset of tooth root formation. Chief among characteristics of the dental pulp is that it is relatively more hypoxic than other tooth-related tissues²⁷. First, we examined the expression of hypoxia-inducible factor 1α (HIF1α) in developing molars using quantitative immunohistochemical assays. In mice, tooth root formation of molars starts early postnatally, around postnatal day (P) 3²⁸. Immediately after birth at P0, the HIF1α signal was relatively low in the dental papilla (DP) within the tooth bud (Fig. 1a, a’). However, the HIF1α signal intensified between P3 and P6, then thereafter plateaued toward P15 (Fig. 1b, b’ and Supplementary Fig. S1a, b, quantification in Fig. 1c). Second, we injected pimonidazole (Pimo), a hypoxia-specific marker, to evaluate hypoxia. The Pimo signal also intensified between P0 and P6, particularly among odontoblasts at the root-forming front (Supplementary Fig. S1c). These data indicate that DP becomes hypoxic specifically at the onset of tooth root formation.

Fig. 1. CXCL12⁺ cells emerge in the hypoxic apical papilla at the onset of tooth root formation.

a–c HIF1α expression. Mandibular first molar (M1) sections at P0 (a) and P6 (b). a’, b’ High magnification. Yellow: HIF1α, gray: DIC/DAPI. c Quantification of HIF1α expression (mean intensity) from P0 to P15, n = 3 mice each time point. d–h Cxcl12-GFP expression. M1 sections of Cxcl12^GFP/+ mandibles at P0 (d), P3 (e), P6 (f, g). Green: Cxcl12-GFP, cyan: E11, gray: DIC/DAPI. (h): Quantification of Cxcl12-GFP⁺ DP cells from P0 to P25, n = 3 mice each time point. i–n Short chase analysis of Cxcl12-creER⁺ cells. i M1 sections of Cxcl12^GFP/+; Cxcl12-creER; R26R^tdTomato mandibles at P5 (pulsed at P3). i’ High magnification of distal M1. j Flow cytometry analysis of P5 CD45^neg tooth bud cells isolated from Cxcl12^GFP/+; Cxcl12-creER; R26R^tdTomato mice (pulsed at P3). k M1 sections of Gli1^GFP/+; Cxcl12-creER; R26R^tdTomato mandibles at P5 (pulsed at P3). l–n Immunostaining for CK5 (l), PDGFRα (m), SOX9 (n) in distal M1 of Cxcl12-creER; R26R^tdTomato at P5 (pulsed at P3). Green: Cxcl12-GFP (i), Gli1-GFP (k), red: Cxcl12^CE-tdT, cyan: CK5/PDGFRα/SOX9, gray: DIC/DAPI. o Cell proliferation. EdU was administered twice (6 and 3 h) to Cxcl12-creER; R26R^tdTomato mice (pulsed at P3) before analysis at P8. Arrowheads: EdU⁺Cxcl12^CE-tdT⁺ AP cells. Scale bar: 500 μm (a, b, d, e, f, i, k, o), 50 μm (d’, e’, f’, i’, k’, l, m, n, o’), 20 μm (a’, b’, g). M1: mandibular first molar, DP: dental papilla, DF: dental follicle, HERS: Hertwig’s epithelial root sheath. All data are mean ± SD. Representative images of at least three independent biological samples are shown in the figures. Source data are provided as a Source Data file.

CXCL12 is a classical target gene of HIF1α in bone marrow stromal and vascular cells^29–31. The CXCL12 and its receptor C-X-C chemokine receptor type 4 (CXCR4) axis play a pivotal role in stem cell homing and tumor development stimulated by hypoxia^32,33. Therefore, we next examined CXCL12 expression in developing molars. Analysis of the mandible of Cxcl12^GFP/+ knock-in mice³⁴ immediately after birth at P0 revealed that Cxcl12-GFP⁺ cells were absent in DP, while present in the dental follicle (DF) surrounding developing molars (Fig. 1d). However, at P3, a new group of Cxcl12-GFP⁺ cells emerged in the apical region of DP, which corresponded to AP (Fig. 1e). At P6, a broad type of DP cells started to express Cxcl12-GFP, not only in the apical region but also in the coronal region (Fig. 1f). These Cxcl12-GFP⁺ cells were distinct from odontoblasts or dental epithelial cells of the HERS, as they lacked E11 expression³⁵ (Fig. 1g) and exuberantly expressed HIF1α (Supplementary Fig. S1f). Notably, cells in many hypoxic regions did not express Cxcl12-GFP, indicating that hypoxia is not the sole driver of Cxcl12 expression. When tooth root formation progressed at P15 and P25, Cxcl12-GFP was broadly expressed in the dental pulp, particularly in the coronal portion (Supplementary Fig. S1d, e). Quantification revealed that the number of Cxcl12-GFP⁺ cells progressively increased in the DP during tooth root formation (Fig. 1h). Therefore, Cxcl12-GFP can mark a group of AP cells specifically at the onset of tooth root formation.

We subsequently sought to define the cell fates of CXCL12⁺ AP cells in tooth root formation using in vivo lineage-tracing approaches. For this purpose, we utilized a Cxcl12-creER transgenic line that we described previously²⁶. First, we characterized cells marked by Cxcl12-creER using a short-chase protocol. Interestingly, Cxcl12-creER specifically marked AP cells upon tamoxifen injection at the onset of tooth root formation. Analysis of Cxcl12^GFP/+; Cxcl12-creER; R26R^tdTomato mice at P5 after a single dose of tamoxifen at P3 revealed that Cxcl12-creER-marked tdTomato⁺ (thereafter Cxcl12^CE-tdT⁺) cells were exclusively located in the AP near the root formation front, immediately medial to HERS (Fig. 1i, i’). Cxcl12^CE-tdT⁺ cells did not localize to DF (Fig. 1i’). While a majority of Cxcl12^CE-tdT⁺ cells were also Cxcl12-GFP⁺, a small group of Cxcl12-GFP-negative Cxcl12^CE-tdT⁺ cells were observed toward the inner side of HERS, suggesting that these cells started to differentiate into odontoblasts at the root-forming front within the 48-hour timeframe.

Flow cytometry analysis of cells isolated from micro-dissected molars revealed that 15.3 ± 9.8% of Cxcl12-GFP⁺ cells were tdTomato⁺, whereas 67.7 ± 7.4% of tdTomato⁺ cells were GFP⁺ (Fig. 1j), indicating that Cxcl12-creER marks a subset of Cxcl12-GFP⁺ cells as well as other Cxcl12-GFP-negative cells. Histological examination of Cxcl12-creER; R26R^tdTomato mice at P5 without tamoxifen injection revealed some tdTomato⁺ ameloblasts were marked independently of tamoxifen injection (Supplementary Fig. S2a). In contrast, many fewer tdTomato⁺ cells were present in DP in these tamoxifen-negative mice. Flow cytometry analysis of the cells isolated from these tamoxifen-negative mice revealed a small population of tdTomato⁺ cells in a Cxcl12-GFP-negative fraction (Supplementary Fig. S2b), indicating a low level of leakiness of this Cxcl12-creER line in developing molars.

Glioma-associated oncogene homolog 1 (Gli1) is used as a marker of dental mesenchymal progenitor cells^11,36. At this stage, Gli1-creER marked dental mesenchymal both in the DP and DF and dental epithelial cells (Supplementary Fig. S2c, c’)(11). Histological analysis of Gli1^GFP/+; Cxcl12-creER; R26R^tdTomato/+ mice at P5 (pulsed at P3) revealed that Cxcl12-creER marked a subset of Gli1-GFP⁺ cells in the AP, but not in the DF (Fig. 1k, k’), indicating that CXCL12⁺ AP cells represent a subset of Gli1⁺ root progenitor cells. Further histological analysis of Cxcl12-creER; R26R^tdTomato mice at P5 (pulsed at P3) revealed that Cxcl12^CE-tdT⁺ cells did not express epithelial cell marker cytokeratin 5 (CK5), but a subset of these cells expressed PDGFRα (CD140a) and SOX9 (Fig. 1l, m, n). Cxcl12^CE-tdT⁺ cells proliferated actively near the HERS after 5 days of chase at P8, as they incorporated 5-ethynyl-2-deoxyuridine (EdU) administered shortly before analysis (arrowheads; 17.2 ± 5.3 EdU⁺Cxcl12^CE-tdT⁺ cells among 169.7 ± 10.4 Cxcl12^CE-tdT⁺ cells per 15 μm thickness; EdU⁺: 10.3 ± 3.8% of Cxcl12^CE-tdT⁺ cells per 15 μm thickness; n = 3) (Fig. 1o, o’).

We also pulsed Cxcl12-creER; R26R^tdTomato mice at P0 before the emergence of CXCL12⁺ AP cells. After 48 h, we unexpectedly found that cells of both DP and DF were marked by tdTomato, which contributed to relevant cell types in a manner similar to Cxcl12^CE-P3 cells, although to a lesser extent (Supplementary Fig. S2d). This could be due to either the extended bioavailability of tamoxifen marking emerging CXCL12⁺ AP cells at P2 or the specificity of our 207 kb Cxcl12-creER BAC transgene to AP cells, regardless of the induction time point, due to unknown reasons.

Therefore, CXCL12⁺ cells emerge in the hypoxic AP at the onset of tooth root formation, which Cxcl12-creER can effectively mark upon tamoxifen injection.

Cxcl12-creER⁺ AP cells function as odonto-cemento-progenitor cells

Subsequently, we traced the fate of CXCL12⁺ AP cells during tooth root formation using Cxcl12-creER. Interestingly, Cxcl12-creER⁺ AP cells contributed to both DP and DF in the elongating root. Analysis of Col1a1(2.3 kb)-GFP; Cxcl12-creER; R26R^tdTomato (pulsed at P3) molars revealed that, after 7 days of chase at P10, Cxcl12^CE-tdT⁺ cells robustly populated the dental mesenchyme surrounding the down-growing HERS, contributing not only to the newly formed portion of DP but also to the apical region of DF (Fig. 2a, a’). At this stage, Cxcl12^CE-tdT⁺ cells started to differentiate into Col1a1(2.3 kb)-GFP⁺ and E11⁺ odontoblasts of the newly formed root dentin (Fig. 2a’, b, arrowheads). Cxcl12^CE-tdT⁺ cells did not contribute to dental epithelial cells in HERS.

Fig. 2. Cxcl12-creER⁺ AP cells function as odonto-cemento-progenitor cells.

Cell fate analysis of Cxcl12-creER⁺ AP cells during tooth root formation. Mandibular first molar (M1) sections. a, b During active tooth root formation, after 7 days of chase at P10. Col1a1(2.3 kb)-GFP; Cxcl12-creER; R26R^tdTomato (a) and Cxcl12-creER; R26R^tdTomato (b) molars (pulsed at P3). (a’, b’): High magnification of distal M1. Arrowheads in (a): Col1a1-GFP⁺Cxcl12^CE-tdT⁺ odontoblasts. Green: Col1a1-GFP, red: Cxcl12^CE-tdT, cyan: E11, gray: DIC/DAPI. c–f At the completion of tooth root formation, after 22 days of chase at P25. c, d Col1a1(2.3 kb)-GFP; Cxcl12-creER; R26R^tdTomato (c) and Osteocalcin (3.8 kb) (Oc)-GFP, Cxcl12-creER; R26R^tdTomato (d) molars (pulsed at P3). Arrowheads in (c’): Col1a1-GFP⁺Cxcl12^CE-tdT⁺ odontoblasts, arrows in (d): Oc-GFP⁺Cxcl12^CE-tdT⁺ cementoblast. Green: Col1a1-GFP (c) or Oc-GFP (d), red: Cxcl12^CE-tdT, gray: DIC/DAPI. e Cxcl12-creER; R26R^RGBow/RGBow molars (pulsed at P3). Arrowheads: same-color clones of odontoblasts. Green: GFP, red: mOrange2, blue: mKate2. (f): EdU label-retention assay of Cxcl12-creER⁺ cells. Cxcl12-creER; R26R^tdTomato mice (pulsed at P3) were serially pulsed with EdU every 8 h for 3 days from P3 to P6. Arrows: EdU-retaining Cxcl12^CE-tdT⁺ cells at the neurovascular bundle. Green: S100, red: Cxcl12^CE-tdT, cyan: EdU, gray: DIC/DAPI. g Long-term cell fate after 6 months of chase. Oc-GFP; Cxcl12-creER; R26R^tdTomato molars. Asterisk: absence of tdT⁺ alveolar bone osteocytes. Arrows: Oc-GFP⁺Cxcl12^CE-tdT⁺ cementoblasts. Green: Oc-GFP, red: Cxcl12^CE-tdT, gray: DIC/DAPI. Scale bar: 500 μm (a, c, f, g), 200 μm (g’, g”), 50 μm (a’, b, c’, d, e, f’). DP: dental papilla/pulp, DF: dental follicle, HERS: Hertwig’s epithelial root sheath, Od: odontoblasts, D: dentin, C: cementum, PDL: periodontal ligament, AB: alveolar bone. Representative images of at least three independent biological samples are shown in the figures.

One of the most intriguing findings is that CXCL12⁺ AP cells robustly contributed to both odontoblasts and cementoblasts at the completion of tooth root formation. After 22 days of chase at P25, Cxcl12^CE-tdT⁺ cells populated the dental pulp and contributed to a majority of Col1a1(2.3 kb)-GFP⁺ odontoblasts of the root dentin (Col1a1-GFP⁺Cxcl12^CE-tdT⁺:72.4 ± 9.9 % of Col1a1(2.3 kb)-GFP⁺ odontoblasts, n = 3) (Fig. 2c, c’, arrowheads). In addition, analysis of Oc-GFP; Cxcl12-creER; R26R^tdTomato (pulsed at P3) molars revealed that Cxcl12^CE-tdT⁺ cells contributed to osteocalcin (Oc)-GFP⁺ cementoblasts on the root surface (Fig. 2d). In fact, a large proportion of Oc-GFP⁺ cementoblasts were Cxcl12^CE-tdT⁺ (Oc-GFP⁺Cxcl12^CE-tdT⁺: 47.7 ± 4.7% of Oc-GFP⁺ cementoblasts, n = 3), indicating that CXCL12⁺ AP cells robustly contribute to cementoblasts during tooth root formation. CXCR4 was expressed in the apical region of the tooth root (Supplementary Fig. S3a), indicating that CXCL12-CXCR4 signaling may play a role in cementoblast differentiation.

We further performed in vivo clonal analysis using biallelic multicolor RGBow reporter alleles³⁷, which can give rise to 6 distinct color combinations upon cre-mediated recombination. Analysis of Cxcl12-creER; R26R^RGBow/RGBow (pulsed at P3) molars revealed independently colored groups of odontoblasts on the dentin surface³⁸ (Fig. 2e, arrowheads), indicating that CXCL12⁺ AP cells might clonally contribute to root dentin odontoblasts. We further performed EdU label-retaining experiments to identify slow-cycling cells that may serve as the cell origin for root dentin odontoblasts. Cxcl12-creER; R26R^tdTomato mice (pulsed at P3) were serially pulsed with EdU from P3 to P6 and chased until P25. Within the apical portion of the tooth root, Cxcl12^CE-tdT⁺ EdU label-retaining cells were predominantly localized within or near the S100⁺ neurovascular bundle (Fig. 2f and Supplementary Fig. S3b, c, arrows). The same EdU label-retaining cells were observed after a more extended chase at 64 days after the serial pulse (Supplementary Fig. S3d). These findings are consistent with the concept that the neurovascular bundle provides a stem cell niche of an ever-growing incisor in mice³⁹.

We further chased Oc-GFP; Cxcl12-creER; R26R^tdTomato (pulsed at P3) molars further into the adult stage at 6 months. Cxcl12^CE-tdT⁺ cells continued contributing to dental pulp cells, odontoblasts of the root dentin, and cementoblasts on the root surface (Fig. 2g, g” arrows). However, these cells did not become osteoblasts and osteocytes of the alveolar cryptal bone in the interradicular and interseptal area (Fig. 2g, g’ asterisk), indicating that CXCL12⁺ AP cells are exclusively fated to odontoblasts and cementoblasts in the long term.

Therefore, these findings reveal that Cxcl12-creER⁺ AP cells function as root-forming odonto-cemento-progenitor cells that contribute to dental pulp cells, root dentin-forming odontoblasts, and cementoblasts in the long term, but not to alveolar bone osteoblasts.

Injury/insult-induced plasticity of CXCL12⁺ AP derivatives

Further, we asked whether CXCL12⁺ AP derivatives show plasticity under regenerative and pathological conditions, guided by our previous finding that CXCL12⁺ stromal cells convert their identity to a stem cell-like state in bone regeneration²⁶. We first investigated the periodontal regenerative capabilities of CXCL12⁺ AP derivatives. To this end, we utilized a surgical periodontal fenestration defect model previously described in rats^40,41. A 1 mm diameter drill-hole defect was created on the buccal side of the right mandible of Oc-GFP; Cxcl12-creER; R26R^tdTomato (pulsed at P3) mice at 8 − 10 weeks of age, while the left mandible was used as a contralateral control (Fig. 3a). The periosteum adjacent to the defect was entirely scraped off before the procedure, resulting in no pre-existing tdTomato⁺ cells outside the alveolar bone surface (Supplementary Fig. S4b, b” arrow). This drill-hole procedure completely removed tdTomato⁺ cells within the defect, while tdTomato⁺ cells in the adjacent PDL and cementum remained intact (Supplementary Fig. S4a, b). After 14 days of injury, Cxcl12^CE-tdT⁺ cells robustly participated in the defect healing and differentiated into Oc-GFP⁺ cells including cementoblasts on the repaired root surface (Fig. 3c, c’, arrowheads, Fig. 3d) and osteoblasts in the regenerated area of the alveolar bone (Fig. 3c, c’, arrows, Fig. 3e) (Oc-GFP⁺Cxcl12^CE-tdT⁺ cementoblasts: 4.4 ± 1.8 cells, osteoblasts: 9.1 ± 3.8, n = 4). This was in sharp contrast to the contralateral control side, wherein Cxcl12^CE-tdT⁺ cells were found among Oc-GFP⁺ cementoblasts and periodontal ligament cells but not among Oc-GFP⁺ osteoblasts and osteocytes in the alveolar bone (Fig. 3b, b’ asterisk) (Oc-GFP⁺Cxcl12^CE-tdT⁺ cementoblasts: 6.8 ± 3.0 cells, osteoblasts: 0.1 ± 0.2, n = 4). Therefore, CXCL12⁺ AP derivatives can convert their cell fates to alveolar bone osteoblasts in response to injury, demonstrating their plasticity.

Fig. 3. Injury/insult-induced plasticity of CXCL12⁺ AP derivatives cell fates.

a–e Impact of periodontal drill-hole injury on Cxcl12-creER⁺ AP derivatives. Surgery at 8–10 weeks-old Oc-GFP; Cxcl12-creER; R26R^tdTomato mice (pulsed at P3), analysis at 14 days after the procedure. a Experimental scheme of mandibular periodontal drill-hole surgery model. A 1.0 mm diameter drill hole is created on the buccal surface of the mandible to remove the alveolar bone and the cementum. b, c Mandibular first molar (M1) frontal section, left mandible (Control) (b), right mandible (Surgery) (c). Asterisk in (b’): absence of tdTomato⁺ alveolar bone osteocytes. Arrowheads in (c’): Oc-GFP⁺Cxcl12^CE-tdT⁺ cementoblasts newly formed within the defect. Arrows in (c’): Oc-GFP⁺Cxcl12^CE-tdT⁺ osteoblasts newly formed within the defect. d, e Quantification of Oc-GFP⁺Cxcl12^CE-tdT⁺ cementoblasts (d) and alveolar bone osteoblasts (e). n = 4 mice per each group. f–j Impact of ligature-induced periodontitis on Cxcl12-creER⁺ AP derivatives. Intervention at 12 weeks old Oc-GFP; Cxcl12-creER; R26R^tdTomato mice (pulsed at P3), analysis at 14 days after installment. f Experimental scheme of maxillary ligature-induced periodontitis model. A 6-0 silk suture was inserted between the first and second molar on the left side. Both ends of the suture string were knotted to prevent ligature loss. g–i Maxillary first and second molar (M1&M2) sagittal section, right maxilla (Control) (g, h), left maxilla (Ligature) (i, j). Asterisk in (b’): absence of tdTomato⁺ alveolar bone osteocytes. Arrows in (i"): tdTomato⁺ alveolar bone osteoblasts in fibrotic lesion. Green: Oc-GFP, red: Cxcl12^CE-tdT, gray: DIC/DAPI. Scale bars: 500 μm (g, h, i, j), 200 μm (b, c), 50 μm (b’, c’, g’, g”, i’, i”). B: buccal side, L: lingual side, D: dentin, C: cementum, PDL: periodontal ligament, AB: alveolar bone. *P < 0.05, Two-tailed, one-way ANOVA followed by the Mann-Whitney U test. All data are mean ± SD. Exact P-value is indicated in the figures. Representative images of at least three independent biological samples are shown in the figures. Source data are provided as a Source Data file.

Second, we investigated whether CXCL12⁺ AP derivatives respond to periodontal bone destruction. For this purpose, we utilized a simplified ligature-induced periodontitis model⁴² that acutely induces a vertical alveolar bone defect. A bacterially retentive silk suture was installed between the first molar (M1) and the second molar (M2) of the left maxilla of Oc-GFP; Cxcl12-creER; R26R^tdTomato (pulsed at P3) mice at 12 weeks of age to rapidly induce periodontitis, while the right maxilla was used as a contralateral control (Fig. 3f and Supplementary Fig. S4c). This procedure induced rapid alveolar bone loss as early as 7 days after ligature placement (Supplementary Fig. S4d–g). After 14 days of ligature placement, the interseptal bone was extensively destructed below the ligature, wherein Cxcl12^CE-tdT⁺ cells appeared to be accordingly lost compared to the contralateral control side (Fig. 3g–j). However, a striking difference was observed in the interradicular bone that was not immediately adjacent to the ligature. The distal side of the interradicular bone of M1 underwent fibrotic responses in which Cxcl12^CE-tdT⁺ cells exuberantly participated (Fig. 3i, i’, i”, arrows). Such a fibrotic response was not observed in the contralateral control side (Fig. 3g”, asterisk). Therefore, CXCL12 AP derivatives on the root surface can respond to inflammatory cues and convert their identities to fibrotic cell types under pathological conditions.

Single-cell RNA-seq characterization of CXCL12⁺ AP cells and their differentiation

We further aimed to define the molecular identity of CXCL12⁺ AP cells. For this purpose, we performed single-cell RNA-sequencing (scRNA-seq) analyses of Cxcl12-GFP⁺ cells at P6 when Cxcl12-GFP is ubiquitously expressed within the DP. We isolated Cxcl12-GFP⁺ cells from micro-dissected molars of Cxcl12^GFP/+ mice using fluorescence-activated cell sorting (FACS) and loaded these cells onto the 10X Chromium Single-Cell Gene Expression Solution platform (Fig. 4a and Supplementary Fig. S5a). UMAP-based visualization of 7,528 Cxcl12-GFP⁺ cells discovered 8 major clusters, including DF cells (Cluster 0: Runx2⁺, Acta2⁺, Hhip⁺), three clusters of dental papilla cells (Cluster 1,2,4: Sox9⁺, Sp7⁺, Msx2⁺), apical papilla/follicle cells (Cluster 3: Wif1⁺, Smoc2⁺, Pthlh⁺) and Schwann cells (Cluster 7: Plp1⁺), and endothelial cells (Cluster 6: Pecam1⁺) (Fig. 4b, c and Supplementary Fig. S5b). Importantly, Cxcl12-GFP⁺ DP, apical papilla/follicle, and DF cell clusters were closely related, indicating the contiguous nature of the DP and DF cell lineages. In addition, Cxcl12-GFP⁺ cells uniformly expressed Ctnnb1 encoding β-catenin, supporting a vital role of canonical Wnt signaling in Cxcl12-GFP⁺ cell differentiation (Fig. 4c).

Fig. 4. Single-cell RNA-seq characterization of CXCL12⁺ AP cells and their differentiation potential.

a–c Heterogeneity of Cxcl12-GFP⁺ cells. Single-cell RNA-seq analysis of Cxcl12-GFP⁺ cells isolated from Cxcl12^GFP/+ molars at P6. a Workflow of cell dissociation, FACS isolation of Cxcl12-GFP⁺ cells, and scRNA-seq. b UMAP plot of major classes of Cxcl12-GFP⁺ cells (Cluster 0–7, 7,528 cells). Pooled from n = 3 mice. c Feature plots of representative DF (Runx2, Acta2), DP (Sox9, Sp7), apical papilla/follicle (Wif1, Smoc2) and Schwann cell (Plp1) markers, and canonical Wnt (Ctnnb1) marker. High expression: violet, Low expression: yellow. d–j Putative cell origin and cell-cell interaction of Cxcl12-creER⁺ AP cell derivatives. d Workflow of integrative scRNA-seq analysis, merging datasets of Cxcl12^CE-tdT⁺ cells at P6 and P25 (pulsed at P3). DP: dental pulp, PDL: periodontal ligament. e, f UMAP plot of the merged dataset (Cluster 0–14, 5869 cells). e Colored by each dataset (Cxcl12^CE-tdT⁺ at P6: 358 cells, Col1a1GFP⁺ and Cxcl12^CE-tdT⁺ cells at P25: 5,511 cells), (f): major classes of Cxcl12^CE-tdT⁺ cells, osteoblast (C1), cementoblast (C6), PDL (C0), apical papilla/pulp (C4), dental pulp (C3, C7), stromal (C2), hematopoietic (C5, C10, C12), epithelial (C9) cells. g Feature plots of the representative pulp (Sox9), apical papilla/pulp (Wif1), PDL (Scx), cementoblast (Pthlh), osteoblast (Bglap) markers and tdTomato. h CellChat intercellular communication analysis. A chord diagram demonstrates the intercellular interaction network via the Wnt signaling pathways. The lines denote intercellular interaction strengths, linking sender cells (blunt end) with receiver cells (arrowhead). The color bars in the outer circle indicate cell states sending outgoing signals, while those in the inner circles represent cell types receiving incoming signals. i VeloVAE analysis after removing tdTomato^- cell populations (C1,5,9-14). Black arrows: dynamic velocity vectors. The orientation of black arrows denotes the developmental direction from progenitors to their derivatives. j scVelo-computed initial state and end point probability. Initial cell state probability (upper panel) and terminal cell state probability (lower panel). Arrowhead: putative points of root cells in AP cells.

We subsequently sought to define how CXCL12⁺ AP cells molecularly develop into their derivatives in the periodontium and dental pulp at the single-cell level. For this purpose, we performed an integrative scRNA-seq analysis by combining two datasets; first, short-chased Cxcl12^CE-tdT⁺ cells isolated from micro-dissected molars of Cxcl12-creER; R26R^tdTomato mice at P6 (pulsed at P3), and second, 3 weeks-chased Col1a1-GFP⁺ and Cxcl12^CE-tdT⁺ cells isolated from the periodontium of the mandibular molar of Col1a1(2.3 kb)-GFP; Cxcl12-creER; R26R^tdTomato mice (pulsed at P3) at P25 using LIGER⁴³ (Fig. 4d and Supplementary Fig. S5c). Leiden community detection of all 5,869 cells (Cxcl12^CE-tdT⁺ at P6: 358 cells, Col1a1GFP⁺ and Cxcl12^CE-tdT⁺ cells at P25: 5,511 cells) discovered 15 major clusters, PDL cells (Cluster 0), alveolar bone osteoblasts (Cluster 1), cementoblasts (Cluster 6), two clusters of DP cells (Cluster 3,7), and AP cells (Cluster 4) (Fig. 4e–g and Supplementary Fig. S5d). Cxcl12-tdT⁺ at P6 were mainly observed in AP (Cluster 4), the dental pulp (Cluster 7), and epithelial cells (Cluster 9), and robustly expressed Cxcl12 (Supplementary Fig. S5e–g). Importantly, Cxcl12^CE-tdT⁺ cells at P25 were found in cementoblasts (Cluster 6), apical papilla/pulp cells (Cluster 4), and dental pulp cells (Cluster 3,7) but not in alveolar bone osteoblasts (Cluster 1), consistent with our histological findings presented in Fig. 2. The epithelial-mesenchymal interaction is a primary mechanism that dictates tooth root formation⁴⁴. CellChat⁴⁵ intercellular communication analysis revealed that epithelial cells (Cluster 9) were the main cell population that communicated with other cells through Wnt signaling, including cells in the AP (Cluster 4) (Fig. 4h), indicating that Wnt ligands from dental epithelial cells in the HERS provide cues for their surrounding mesenchymal cells during tooth root formation. In fact, dental epithelial cells in Cluster 9 expressed major Wnt ligands, including Wnt4, Wnt6, Wnt7b, and Wnt10a, whereas AP cells expressed major Wnt receptors, including Fzd1-3 and Lrp5-6 (Supplementary Fig. S5h). In contrast, all cell groups except cementoblasts (Cluster 6) interacted with each other through CXCL12-CXCR4 signaling (Supplementary Fig. S5i).

To identify the developmental relationships among CXCL12⁺ AP cells and their descendants, we further refined the integrated data by applying the following two computational measures: 1) removing cells with no plausible lineage relationship to Cxcl12⁺ AP cells, including hematopoietic, endothelial, Schwann cells, alveolar bone osteoblasts (Supplementary Fig. S5j), and 2) extracting only the tdTomato-expressing cells (Supplementary Fig. S5k). Of note, we found alveolar bone osteoblasts not to be derived from Cxcl12⁺ AP cells (Fig. 2). We further analyzed the refined dataset with VeloVAE^46,47. VeloVAE analysis predicted a putative point of cell origins in the AP (Fig. 4i). Interestingly, AP cells were predicted to develop into cementoblasts through PDL cells and dental pulp cells. Tracing these velocity vectors by scVelo inferred putative points of root cells in AP cells with endpoints in dental pulp cells (Fig. 4j, red arrowhead), consistent with our VeloVAE observation.

To define the relationship between Cxcl12⁺ AP cells and Cxcl12-abundant reticular (CAR) cells in the bone marrow, we integrated the above-described single-cell RNA-seq dataset of Cxcl12-GFP⁺ AP cells with our previously published dataset of Cxcl12-GFP⁺ bone marrow stromal cells (GSE136970) using LIGER⁴³. As reported previously, CAR cells were heterogeneous, including subpopulations of Lepr^hiAdipoq^hiLpl^hi Adipo-CAR cells and cells with transcriptomic similarity to Cxcl12-GFP⁺ AP cells, which may correspond to Osteo-CAR cells⁴⁸ (Supplementary Fig. S6a, b). Importantly, Cxcl12-GFP⁺ AP cells were distinct from the Lepr^hiAdipoq^hiLpl^hi Adipo-CAR cell subset, supporting the notion that Cxcl12⁺ AP cells represent a cell type distinct from well-characterized Adipo-CAR cells in the bone marrow space.

Therefore, these scRNA-seq analyses corroborate our histological findings that CXCL12⁺ AP cells are fated to become dental pulp cells, odontoblasts, and cementoblasts, but not alveolar bone osteoblasts, under the influence of Wnt ligands secreted from the down-growing dental epithelial cells in the HERS.

Canonical Wnt signaling maintains odontogenic fates of CXCL12⁺ AP cells

Prompted by the identification of Wnt-mediated epithelial-mesenchymal interactions described above, we further performed functional analyses of canonical Wnt signaling in CXCL12⁺ AP cell proliferation and differentiation. Immunostaining analysis revealed that β-catenin was broadly expressed in the HERS, DP, and DF, as well as Cxcl12^CE-tdT⁺ cells (Supplementary Fig. S7a). We pulsed littermates of Cxcl12-creER; Ctnnb1^fl/+; R26R^tdTomato (AP-Ctnnb1 cHet/Cxcl12^CE-Ctnnb1^Het-P3 cells) and Cxcl12-creER; Ctnnb1^fl/fl; R26R^tdTomato (AP-Ctnnb1 cKO/Cxcl12^CE-∆Ctnnb1-P3 cells) mice at P3 and analyzed the mandibular molars after chase (Fig. 5a). RNAscope analyses confirmed that, after 5 days of chase at P8, Ctnnb1 (exon 2-6) mRNA was almost entirely abrogated in Cxcl12^CE-∆Ctnnb1-P3 cells (Supplementary Fig. S7b). Furthermore, Axin2, a canonical Wnt-responsive gene and a negative regulator of canonical Wnt signaling, was abundantly expressed by a peri-epithelial subset of Cxcl12^CE-P3 cells at the root-forming front of Control molars but was significantly downregulated in Cxcl12^CE-∆Ctnnb1-P3 cells (Supplementary Fig. S7c). Of note, Axin2 mRNA was broadly expressed across multiple cell types; therefore, Cxcl12^CE-P3 cells represent a subpopulation of Axin2⁺ cells. In contrast, Nfic, an important transcription factor in tooth root formation⁴⁹, was unchanged in Cxcl12^CE-∆Ctnnb1-P3 cells (Supplementary Fig. S7d). EdU assays revealed that, after 7 days of chase at P10, Cxcl12^CE-∆Ctnnb1-P3 cells were significantly less proliferative than Cxcl12^CE-Ctnnb1^Het-P3 cells, indicating that β-catenin is essential for maintaining CXCL12⁺ AP cell proliferation (Fig. 5b, c, f).

Fig. 5. Canonical Wnt signaling maintains odontogenic fates of CXCL12⁺ AP cells.

Impact of canonical Wnt inactivation in Cxcl12-creER⁺ AP cells on cell fates and tooth root formation. Cxcl12-creER; Ctnnb1^fl/+; R26R^tdTomato (AP-Ctnnb1 cHet). Cxcl12-creER; Ctnnb1^fl/fl; R26R^tdTomato (AP-Ctnnb1 cKO) mice were pulsed at P3. a Diagram of conditional Ctnnb1 deletion in Cxcl12-creER⁺ AP cells. b–h Cell proliferation and periodontal ligament cell differentiation. Mandibular first molars (M1) sections. b, c, f EdU assays at 7 days of chase at P10. EdU was administered twice (6 and 3 h) before analysis. Quantification of EdU⁺Cxcl12^CE-tdT⁺ cells is shown in (f). d, e, g, h POSTN staining at 22 days of chase at P25. Quantification of Cxcl12^CE-tdT⁺ odontoblasts (g) and POSTN⁺Cxcl12^CE-tdT⁺ periodontal fibroblasts (h). n = 4 mice per group (f, g), n = 4 mice (AP-Ctnnb cHet), n = 4 mice (AP-Ctnnb cKO). Green: EdU (a–c), POSTN (d, e), red: Cxcl12^CE-tdT, gray: DIC/DAPI. Scale Bars: 500 μm (b–e), 50 μm (b’–e’). i–p 3D-µCT analysis at 6 months. i Composite 3D surface model overlay, superimposition registered on mandibles (upper) and mandibular molars (lower). j–p Quantitative 3D-µCT analysis of M1&M2 root length (j, k), M1 root thickness (l), M1 dentin thickness (m), M1 crown length (n), M1 crown width (o), and M1 eruption height (p). M1: mandibular first molar, M2: mandibular second molar. n = 5 mice per each group. *P < 0.05, **P < 0.01, Two-tailed, one-way ANOVA followed by the Mann-Whitney U test. All data are mean ± SD. Exact P-value is indicated in the figures. Source data are provided as a Source Data file.

After 22 days of chase at P25, a large majority of Cxcl12^CE-∆Ctnnb1-P3 cells aberrantly converted to periostin (POSTN)⁺ periodontal ligament-like cells, particularly on the external tooth root surface on the side of the interseptal bone, associated with apparent lack of root dentin and cementum formation in the AP-Ctnnb1 cKO molar (Fig. 5e and Supplementary Fig. S7e, f). This was in stark contrast with Control Cxcl12^CE-Ctnnb1^Het-P3 cells, which showed orderly differentiation into Oc-GFP⁺ cementoblasts and POSTN⁺ PDL cells in proximity to the root surface (Fig. 5d, g, h).

We further asked whether aberrant cell fates of CXCL12⁺ AP cells in the absence of β-catenin are translated into tooth root morphology. For this purpose, we further chased AP-Ctnnb1 cHet and AP-Ctnnb1 cKO mice (pulsed at P3) into the adult stage of 6 months of age. We performed composite three-dimensional microCT analyses to define the morphological change. Notably, the AP-Ctnnb1 cKO mandible exhibited a defect specifically in the tooth root without involving the alveolar bone or mandibular body (Fig. 5i). The tooth roots of the AP-Ctnnb1 cKO mandibular molars (M1 and M2) were significantly truncated compared to those of the AP-Ctnnb1 cHet molars, associated with a significant reduction in the tooth root thickness and the dentin thickness (Fig. 5j–m). In contrast, no significant difference was noted in the tooth crown width/length or tooth eruption height between the two groups (Fig. 5n–p). The AP-Ctnnb1 cKO femur did not show any alteration in the femur length or other parameters at P25 (Supplementary Fig. S8a), indicating that the absence of β-catenin in CXCL12⁺ bone marrow stromal cells does not change long bone morphology or bone mass. Interestingly, histological examination of the AP-Ctnnb1 cKO mutant molars at 6 months revealed ectopic clusters of Cxcl12^CE-tdT⁺ cells with chondrocyte-like morphology in the apical region (SOX9⁺ chondrocyte-like cells: 3/3 AP-Ctnnb1 cKO mice examined) (Supplementary Fig. S7g, h), indicating cell fate shifts of CXCL12⁺ AP derivatives to a chondrocyte-like state.

To ensure that the AP-Ctnnb1 cHet molars present no phenotype, we next pulsed littermates of Cxcl12-creER; Ctnnb1^+/+; R26R^tdTomato (AP-Ctnnb1 WT); Cxcl12-creER; Ctnnb1^fl/+; R26R^tdTomato (AP-Ctnnb1 cHet) and Cxcl12-creER; Ctnnb1^fl/fl; R26R^tdTomato (AP-Ctnnb1 cKO) mice at P3 and analyzed the mandibular molars at P25. Importantly, no difference in the tooth root morphology or β-catenin immunoreactivity was observed between the AP-Ctnnb1 WT and AP-Ctnnb1 cHet molars (Supplementary Fig. S8b, c), supporting the validity of the use of the AP-Ctnnb1 cHet molars as a control. In AP-Ctnnb1 cKO molars, β-catenin immunoreactivity was specifically abrogated in tdTomato⁺ cells, as expected (Supplementary Fig. S8c).

In addition, we sought to define whether other signaling pathways also affect the tooth root-forming capability of CXCL12⁺ AP cells. We first examined the CXCL12-CXCR4 pathway. Immunostaining revealed that CXCR4, a cognate receptor of CXCL12, was broadly expressed at the presumptive root formation front in HERS, DP, DF, and Cxcl12^CE-tdT⁺ cells of P5 Cxcl12-creER; R26R^tdTomato (pulsed at P3) molars (Supplementary Fig. S9a). Flow cytometry analysis revealed that 2.5 ± 2.2% of CXCR4⁺ cells were Cxcl12^CE-tdT⁺, whereas 95.0 ± 4.0% of Cxcl12^CE-tdT⁺ cells were CXCR4⁺ (Supplementary Fig. S9b), indicating that CXCL12 secreted by Cxcl12^CE-tdT⁺ cells may interact with CXCR4 expressed by Cxcl12^CE-tdT⁺ and their surrounding cells such as dental epithelial cells in the HERS and dental mesenchymal cells in the DP and DF, both in autocrine and paracrine manners. We further investigated whether CXCL12 has a functional role in the differentiation of CXCL12⁺ AP cells. For this purpose, we analyzed littermates of Cxcl12^GFP/+; Cxcl12-creER; R26R^tdTomato (Control) and Cxcl12^GFP/fl; Cxcl12-creER; R26R^tdTomato (CXCL12-cKO) mice at P25 after a tamoxifen pulse at P3 (Supplementary Fig. S9c). Of note, the Cxcl12-GFP allele represents a loss-of-function allele²⁴. CXCL12-cKO mice did not show any phenotype in the tooth root or any alteration in CXCL12⁺ AP cell fates (Supplementary Fig. S9c), indicating that CXCL12 does not have a functional role in these processes. SOX9 is also expressed across many cell types at the presumptive root formation front at P5 (Fig. 1n). However, analysis of Cxcl12-creER; Sox9^fl/fl; R26R^tdTomato mice (pulsed at P3) revealed no phenotype in the tooth root at P25 (Supplementary Fig. S9d), demonstrating that SOX9 may also have no major functional role in the differentiation of the CXCL12⁺ AP cell subset ( ~ 15%) targeted by Cxcl12-creER.

Therefore, canonical Wnt signaling mediated by β-catenin has a vital role in maintaining physiological odontogenic fates of CXCL12⁺ AP cells and tooth root formation without alveolar bone formation or tooth eruption, supporting the notion that AP cells specifically regulate tooth root formation via the canonical Wnt pathway.

Canonical Wnt signaling prevents premature differentiation of CXCL12⁺ AP cells

We sought to define mechanisms underlying a cell fate shift of CXC12⁺ AP cells caused by defective canonical Wnt signaling. To this end, we performed RNA-seq analyses of β-catenin-deficient CXCL12⁺ AP cells. Cxcl12^CE-Ctnnb1^Het-P3 (Ctnnb1^Het) and Cxcl12^CE-∆Ctnnb1-P3 (∆Ctnnb1) cells were isolated by FACS at P10 from P3-pulsed AP-Ctnnb1 cHet and AP-Ctnnb1 cKO molars, respectively (Fig. 6a, b). Principal component analysis (PCA) revealed that biological quadruplicates of Ctnnb1^Het and ∆Ctnnb1 samples grouped by the genotype on PC1 (Supplementary Fig. S10a, Ctnnb1^Het: red dots, ∆Ctnnb1: blue dots), demonstrating that canonical Wnt signaling inactivation caused a consistent change in the transcriptomes of CXCL12⁺ AP cells. An examination of the Ctnnb1 locus revealed that the reads were significantly reduced between exons 2 and 6 in Cxcl12^CE-∆Ctnnb1-P3 samples, which are flanked by loxP sites in the Ctnnb1 floxed allele (Supplementary Fig. S10b), confirming the loss of β-catenin in these samples. Analyses of differentially expressed genes (DEGs) by DESeq2 revealed that there are 987 upregulated and 277 downregulated in Cxcl12^CE-∆Ctnnb1-P3 as compared to Cxcl12^CE- Ctnnb1^Het-P3 cells (FDR < 0.05 & Fold Change > 2) (Supplementary Fig. S10c). Importantly, the following genes were upregulated in Cxcl12^CE-∆Ctnnb1-P3 cells; non-canonical Wnt genes such as Wnt5a, Wnt antagonists such as Sostdc1 and Dkk1, chondrocyte marker genes such as Sox9 and Col9a1, PDL marker genes such as Postn and Mkx, osteoblast marker genes such as Sp7 and dental pulp/odontoblast marker genes such as Cxcl12 (Fig. 6c). The finding that POSTN was significantly upregulated in Cxcl12^CE-∆Ctnnb1-P3 cells in vivo (Fig. 5e) supports the validity of our RNA-seq findings. In contrast, chemokine genes such as Ccl3 and Ccl4 were downregulated in Cxcl12^CE-∆Ctnnb1-P3 cells (Fig. 6c). Gene Ontology (GO) enrichment analysis of DEGs revealed significant enrichment of GO terms, including extracellular matrix organization (GO:0030198; adjusted p < 1.4 × 10^-19), negative regulation of canonical Wnt signaling (GO:0090090; adjusted p < 7.8 × 10⁻¹³), cell differentiation (GO:0030154; adjusted p < 1.4 × 10⁻¹²) and cartilage development (GO:0051216; adjusted p < 1.7 × 10⁻¹²) (Fig. 6d). In addition, gene set enrichment analysis (GSEA) revealed that the relevant terms were enriched in Cxcl12^CE-∆Ctnnb1-P3 cells, including chondrocyte differentiation (NES 2.37), regulation of cartilage development (NES 2.27), collagen formation (2.44), and chemokine signaling pathway (NES − 1.60) (Fig. 6e and Supplementary Fig. S10d), supporting the status of Wnt-inactivated CXCL12⁺ AP cells with altered cell fates toward chondrocytes or fibroblasts. Therefore, these RNA-seq analyses revealed that defective canonical Wnt signaling induced premature differentiation of CXCL12⁺ AP cells associated with precocious matrix production.

Fig. 6. Canonical Wnt signaling prevents premature differentiation of CXCL12⁺ AP cells.

Identification of mechanisms underlying canonical Wnt inactivation-induced CXCL12⁺ AP cell fate shifts by RNA-req. a, b Workflow of cell dissociation, FACS isolation of Ctnnb1^Het (Control) and ∆Ctnnb1 Cxcl12-tdT⁺ cells and bulk RNA-seq. b FACS sorting of Cxcl12-creER⁺ AP cells at P10 (pulsed at P3). Blue box: Cxcl12^CE-Ctnnb1^Het-P3 (Ctnnb1^Het) cells isolated from AP-Ctnnb1 cHet molars, red box: Cxcl12^CE-∆Ctnnb1-P3 (∆Ctnnb1) cells isolated from AP-Ctnnb1 cKO molars. c–f RNA-seq analysis results. c Heatmaps of representative 20 differentially expressed genes (DEGs) associated with non-canonical Wnt signaling, Wnt antagonist, chondrocyte, periodontal ligament (PDL), osteoblast, and dental pulp/odontoblast (DP/Od) markers, as well as chemokine. Left 4 lanes: Ctnnb1^Het, right 4 lanes: ∆Ctnnb1. PDL: periodontal ligament, DP: dental pulp, Od: odontoblast. Yellow: higher expression, purple: lower expression, n = 4 mice per each group. d Top enriched GO terms – biological processes (GO-BP) overrepresented in DEGs (Adjusted p < 0.01, Fisher’s exact test). e, f GSEA analysis. Representative 4 pathways (e) and 2 up-regulated signaling pathways (f) in Cxcl12^CE-∆Ctnnb1-P3 cells. NES: normalized enrichment score (Adjusted p < 0.001, weighted Kolmogorov-Smirnov test). g, h Administration of TGF-β receptor inhibitor galunisertib (Gal) to AP-Ctnnb1 cHet and cKO mice. Gal or PBS was administrated into AP-Ctnnb1 cHet or cKO from P8 to P20. g Mandibular 1^st molar of AP-Ctnnb1 cHet and cKO at P25 (pulsed at P3). Scale bars: 500 μm (upper panels), 100 μm (lower panels). h Quantification of POSTN⁺tdTomato⁺ cells/tdTomato⁺ dental pulp cells. n = 3 mice (AP-Ctnnb cHet + PBS), n = 4 mice (AP-Ctnnb cHet + Gal), n = 6 mice (AP-Ctnnb cKO + PBS), n = 4 mice (AP-Ctnnb cKO + Gal), *P < 0.05, **P < 0.01, ***P < 0.001, Two-tailed, one-way ANOVA followed by the Mann-Whitney U test. All data are mean ± SD. Exact P-value is indicated in the figures. Source data are provided as a Source Data file.

Lastly, we set out to identify potential downstream targets of canonical Wnt signaling that govern the CXCL12⁺ AP cell fate. The KEGG pathway analysis by GSEA identified significant enrichment of several major signaling pathways other than the Wnt signaling pathway (NES: 1.69), including the TGF-β signaling pathway (NES: 2.22) (Fig. 6f). We further reanalyzed the single-cell RNA-seq dataset shown in Fig. 4b to establish the Wnt-TGF-β connection. Cells in the AP cluster (Cluster 3) expressed TGF-β-related genes, including TGF-β ligands (Tgfb1,2,3), receptor (Tgfbr2), and inducible genes (Tgfbi, Tgfb1i1), as well as its downstream effectors (Smad 3,6,7) and target transcription factor (Myc) (Supplementary Fig. S11a), indicating that TGF-β signaling is operative in AP cells under physiological conditions. We further performed SCENIC (single-cell regulatory network inference and clustering) analysis. Of the 248 transcription factors (TFs) identified by SCENIC, we particularly focused on the five well-described Wnt-responsive transcription factors, including Lef1, Tcf3, Tcf12, Tcf7l1 and Tcf7l2. These Wnt-responsive TFs regulated many TGF-β-related genes expressed in the AP, including Klf10, Sfrp4, Tbx3 (by Lef1), Bmpr1a, Gli3 (by Tcf3), and Yap1 (by Tcf7l2) (Supplementary Fig. S11b, c), indicating that Wnt signaling may regulate the TGF-β signaling pathway through transcriptional regulation in Cxcl12⁺ AP cells.

Subsequently, we experimentally test the functional requirement of the TGF-β signaling pathway in regulating CXCL12⁺ AP cell fate determination. To this end, we administered small-molecule TGF-β type I receptor inhibitor galunisertib (Gal) to littermates of Cxcl12-creER; Ctnnb1^fl/+; R26R^tdTomato (AP-Ctnnb1 cHet) and Cxcl12-creER; Ctnnb1^fl/fl; R26R^tdTomato (AP-Ctnnb1 cKO) mice (pulsed at P3) from P8 to P20 and analyzed these mice at P25. Importantly, TGF-β signaling inhibition by Gal did not affect AP-Ctnnb1 cHet molars without altering POSTN expression in Cxcl12^CE-P3 tdTomato⁺ cells or the tooth root length. In contrast, in AP-Ctnnb1 cKO molars, Gal significantly reduced POSTN expression in Cxcl12^CE-∆Ctnnb1-P3 cells (Fig. 6g, h). Therefore, the aberrant fibroblast fates of Wnt-inactivated CXCL12⁺ AP cells can be suppressed by TGF-β signaling inhibition.

These findings support that canonical Wnt signaling prevents premature differentiation of CXCL12⁺ AP cells and maintains their proper differentiation potential to root-forming odontoblasts and cementoblasts essential for establishing the tooth root and its periodontal attachment apparatus at least in part through suppressing TGF-β signaling (Fig. 7).

Fig. 7. Wnt-directed CXCL12-expressing AP progenitor cells drive tooth root formation.

Cxcl12⁺ AP mesenchymal progenitor cells differentiate into odontoblasts, dental pulp cells, PDL cells, and cementoblasts via canonical Wnt signaling during tooth root development. The TGF-β signaling pathway is a potential downstream target of canonical Wnt signaling in CXCL12⁺ AP cells. During periodontal injury response, descendants of Cxcl12⁺ AP cells were recruited to the bone defect and differentiated into osteoblast-lineage cells.

Discussion

Here, we identify that CXCL12⁺ AP progenitor cells provide a potent source of dentin-forming odontoblasts and cementum-forming cementoblasts and drive tooth root formation in a manner controlled by Wnt/β-catenin signaling. These CXCL12⁺ AP progenitor cells are not fated to alveolar bone-forming osteoblasts under normal conditions. However, these cells can convert their fates to osteoblasts under regenerative and pathological conditions, demonstrating salient features of cellular plasticity reminiscent of those occurring in bone marrow stromal cells, as reported previously²⁶. Our findings based on in vivo lineage-tracing approaches challenge the classical concept that cells in the dental papilla are exclusively fated to odontoblasts and dental pulp cells¹. The new concept we propose here is that cells in the apical papilla can generate both odontoblasts inside and cementoblasts outside during tooth root formation. It remains to be determined whether cells in the dental papilla can also give rise to those in the dental follicle in the earlier stages of tooth development.

We found that the dental papilla becomes increasingly hypoxic at the onset of tooth root formation, as demonstrated by HIF1α immunoreactivity (Fig. 1a–c). Coincident with the onset of hypoxia, CXCL12 becomes exclusively expressed in the AP, which then becomes widely expressed in the dental pulp, particularly in the coronal region. Increasingly hypoxic condition likely triggers cellular metabolic changes in dental papilla cells, which may lead to a switch in cell differentiation. CXCL12 has been described as a direct transcriptional target of HIF1α in other closely related cell types, such as bone marrow stromal cells³¹. Further investigation is required to determine if Cxcl12 also represents a direct transcriptional target of HIF1α in AP cells.

CXCL12 is a secreted chemokine that is crucial in stem cell regulation^25,50. CXCL12–CXCR4 signaling is essential for stem cell homing and tumor development stimulated by hypoxia^29,32,33. Furthermore, dental pulp cells express diverse chemokines and cytokines, including CXCL12, to defend against dentin-invading bacteria by inducing an immune response^20,21. Interestingly, our RNA-seq analysis revealed that several chemokines, such as Ccl3, Ccl4, Ccl6, and Ccl9, were downregulated in Wnt-inactivated CXCL12⁺ AP cells. CCL4 is one of the Wnt signaling target molecules in immunity escape in tumorigenesis^51,52. Moreover, CCL3 is involved in osteoclastogenesis during orthodontic tooth movement⁵³. These lines of evidence suggest that Wnt signaling may also regulate tooth root formation via chemotaxis. Further studies are needed to define how chemokines regulate tooth root formation.

Through single-cell RNA-seq analysis, we identified that Wnt ligands emanating from the down-growing dental epithelial cells in Hertwig’s epithelial root sheath act on their surrounding dental mesenchymal cells in the dental papilla and dental follicle (Fig. 4). Strategically located immediately adjacent to the dental epithelium, CXCL12⁺ AP cells likely represent the primary receiver of epithelium-derived Wnt ligands. Particularly among CXCL12⁺ AP cells, the Axin2⁺ subset likely represents the most Wnt-responsive cells due to its peri-epithelial localization. Wnt signaling is vital in regulating stem cells during development and homeostasis^54–57. The dental papilla originates from cranial neural crest cells (NCCs)⁵⁸, whose fates are controlled by canonical Wnt signaling^59,60. In tooth root formation, Wnt signaling is essential for regulating root progenitor cell proliferation, differentiation, and subsequent dentinogenesis and cementogenesis^61–63. Wnt inactivation in mature odontoblasts, cementoblasts, and osteoblasts disrupts tooth root formation^17–19. Our critical finding is that Wnt inactivation specifically in CXCL12⁺ AP cells is sufficient to halt tooth root formation by altering cell fates from odontoblasts to periodontal ligament fibroblast-like cells. This is consistent with the alteration of skeletal stem cell fate into chondrocytes due to Wnt inactivation⁶⁴. Interestingly, our findings demonstrate that Wnt inactivation in CXCL12⁺ AP cells causes failure of tooth root formation without affecting surrounding alveolar bones, successfully uncoupling the effect of Wnt signaling in two distinct biological processes. Alveolar bone formation likely enlists a different CXCL12-negative progenitor cell population. Our companion paper (Nagata et al.) shows that PTHrP⁺ DF cells coordinate alveolar bone formation via the Hedgehog-Foxf axis without affecting tooth root formation (Supplementary Fig. S12).

Our data show that CXCL12⁺ AP cells represent a small subset of Gli1⁺ cells (Fig. 1k), which have been reported to encompass a variety of root progenitor cells^11,28. Canonical Wnt signaling plays vital roles in Gli1⁺ progenitor cells, coordinating epithelial-mesenchymal interactions via SOX9¹³ and activating PDLSCs⁶⁵. Our study identifies a functionally essential subset of Gli1⁺ progenitor cells in the apical papilla, which plays pivotal roles in tooth root formation in a Wnt-dependent manner. Our scRNA-seq analysis suggests that Wnt signaling may regulate the TGF-β signaling pathway through transcriptional regulation in Cxcl12⁺ AP cells. However, the direct functional relationship between Wnt and TGF-β signaling pathways in these cells remains to be studied.

It is important to note that Cxcl12⁺ AP cells described in our study should be termed a cell type composed of heterogeneous cell populations as indicated by our scRNA-seq analysis. Further fractionation with cell surface markers or combinatorial genetic reporters will ultimately be needed to sift these cells into discrete cell populations with uniform functionality.

Translationally, the AP has been known to contain a stem cell population termed stem cells from the apical papilla (SCAP), which can be isolated from the dental pulp and applied for MSC-based regenerative therapy^8–10. Further investigation is warranted to define the relationship between SCAP and CXCL12⁺ AP cells that we identified in this study. Of note, our in vivo lineage-tracing analysis identifies that CXCL12⁺ AP cells can differentiate into odontoblasts inside the root and cementoblasts outside the root, highlighting their potential utility for tooth root regenerative therapies.

In conclusion, our results demonstrate that Wnt-directed CXCL12-expressing apical papilla progenitor cells are vital to tooth root formation by providing a robust source of dentin-forming odontoblasts and cementum-forming cementoblasts. These findings provide a mechanistic framework for tooth root formation and pave the way for developing innovative stem-cell-based regenerative therapies.

Methods

Ethical approval

All procedures were conducted in compliance with the Guidelines for the Care and Use of Laboratory Animals approved by the University of Texas Health Science Center at Houston’s Animal Welfare Committee (AWC), protocol AWC-21-0070 and AWC-24-0075, and the University of Michigan’s Institutional Animal Care and Use Committee (IACUC), protocol 8944 and 9496.

Mouse strains

Cxcl12-creER bacterial artificial chromosome (BAC)²⁶ and Cxcl12^GFP/+34 have been described previously. Rosa26-CAG-loxP-stop-loxP-tdTomato (Ai14: R26R^tdTomato, JAX007914), Osteocalcin (3.8 kb)-GFP (JAX017469), Col1a1-GFP (JAX013134), Ctnnb1-floxed (JAX004152), Sox9-floxed (JAX013106), Cxcl12-floxed (JAX021773), Rosa26-lsl-RGBow (JAX028583), FVB/NJ (JAX001800), and C57BL/6 J (JAX000664) mice were acquired from the Jackson laboratory. Gli1^GFP/+ mice were kindly provided by Dr. B. Allen (University of Michigan, MI, USA). All mice were housed in a specific pathogen-free condition and analyzed in a C57BL/6 background. Mice were housed in individually ventilated cages (Tecniplast, Buguggiate, Italy). Access to water and food (irradiated LabDiet 5053, St. Louis, Missouri) was ad libitum. Animal rooms were climate controlled to provide temperatures of 21 + /− 1 °C, 30–65% of humidity on a 12 h. light/dark cycle (lights on at 0600 Central Standard Time). Both male and female mice were used for the study, as we did not observe any sex bias. For all breeding experiments, creER transgenes were maintained in male breeders to avoid spontaneous germline recombination. Mice were identified by micro-tattooing or ear tags. Tail biopsies of mice were lysed by a HotShot protocol (incubating the tail sample at 95 ^oC for 30 min in an alkaline lysis reagent, followed by neutralization) and used for PCR-based genotyping (GoTaq Green Master Mix, Promega, and Nexus X2, Eppendorf). Perinatal mice were also genotyped fluorescently (BLS miner’s lamp) whenever possible. We occasionally observed spontaneously germline recombination when Cxcl12-creER and Ai14 were maintained in the same male breeder, resulting in ubiquitously tdTomato-fluorescing pups despite the absence of Cxcl12-creER. These “leaky” pups were identified early postnatally (~ P7) by a fluorescent scope and removed from the colony. To evaluate cell proliferation, two doses of 5-ethynyl-2’-deoxyuridine (EdU) (Invitrogen A10044) were injected shortly before analysis, at 6 and 3 h prior to sacrifice, or serially every 8 h for 3 days from P3 to P6 (dose per injection: 0.2 mg). To evaluate hypoxia, pimonidazole (Hypoxyprobe Inc, HP3-100) were administrated into Cxcl12-GFP mice shortly (1 h) before sacrifice. Mice were euthanized by over-dosage of carbon dioxide or decapitation under inhalation anesthesia in a drop jar (Fluriso, Isoflurane USP, VetOne).

Tamoxifen

Tamoxifen (T5648; Sigma-Aldrich) was mixed with 100% ethanol until dissolved completely. Subsequently, a proper volume of sunflower seed oil (Sigma S5007) was added to the tamoxifen-ethanol mixture and rigorously mixed. The tamoxifen-ethanol-oil mixture was incubated at 60 ^oC in a chemical hood until the ethanol evaporated completely. The tamoxifen-oil mixture was stored at room temperature until use. Tamoxifen (0.25 mg) was injected intraperitoneally into P3 mice or P0 mice.

Histology and Immunohistochemistry, and in situ hybridization

Samples were dissected under a stereomicroscope (Nikon SMZ-800) to remove soft tissues and fixed in 4% paraformaldehyde overnight at 4 ^oC, then decalcified in 15% EDTA for 1–14 days. Decalcified samples were cryoprotected in 30% sucrose/PBS solutions and then in 30% sucrose/PBS:OCT (1:1) solutions, each at least overnight at 4^oC. Samples were embedded in an OCT compound (Tissue-Tek, Sakura) under a stereomicroscope and transferred on a sheet of dry ice to solidify the compound. Embedded samples were cryosectioned at 16 μm using a cryostat (Leica CM1860) and adhered to positively charged glass slides (Fisherbrand ColorFrost Plus). Cryosections were stored at − 20 ^oC in freezers until use. Sections were postfixed in 4% paraformaldehyde for 20 min at room temperature.

For immunohistochemistry, sections were permeabilized with 0.25% TritonX/TBS for 30 min, blocked with 3% BSA/TBST for 30 min, and incubated with rabbit anti-POSTN polyclonal antibody (1:2000, EMD-Millipore, ABT280), rabbit anti-HIF1α polyclonal antibody (1:700, GeneTex, GTX127309), rabbit anti-cytokeratin 5 (CK5) polyclonal antibody (1:200, Abcam, ab24647), goat anti-PDGFRα polyclonal antibody (1:100, R&D, AF1602), rabbit anti-SOX9 polyclonal antibody (1:1000, EMD-Millipore, AB5535), rabbit anti-beta catenin polyclonal antibody (1:500, Abcam, ab16051), rabbit anti-S100 polyclonal antibody (1:500, Abcam, ab868), goat anti-podoplanin (E11) polyclonal antibodyd (1:100, R&D, AF3244) and goat anti-CXCR4 polyclonal antibody (1:400, GeneTex, GTX21670) overnight at 4^oC, and subsequently with Alexa Fluor 647-conjugated donkey anti-rabbit IgG (A31573) or Alexa Fluor 647-conjugated donkey anti-goat IgG (A21447) for 3 h at room temperature. For the EdU assay, sections were incubated with Alexa Fluor 647-azide (Invitrogen, A10277) for 30 min at 43 ^oC using Click-iT Imaging Kit (Invitrogen, C10337). Sections were further incubated with DAPI (4’,6-diamidino-2-phenylindole, 5 μg/ml, Invitrogen D1306) to stain nuclei.

RNAscope in situ hybridization was performed with RNAscope Multiplex Fluorescent kit V2 (323110) using probes of Ctnnb1 (Ex2-6) (428171), Nfic (448451-C2), Axin2 (400331), according to the manufacturer’s protocol. Briefly, sections were dehydrated with 100% ethanol and treated with hydrogen peroxide (H₂O₂) for 10 min at RT. Samples were subsequently treated with RNAscope Protease Plus or Protease III for 30 min. at 40 °C and washed with distilled water. Sections were treated with each target probe and hybridized for 2 hr. at 40 °C, followed by signal hybridization by AMP and amplification by HRP-C1 or HRP-C2. The Opal 690 fluorophore (AKOYA Biosciences, FP1497001KT) or the TSA Vivid fluorophore 650 (Tocris Bioscience, 7527) was added to label the HRP probe and sections were treated with HRP blocker. Stained samples were mounted in TBS with No.1.5 coverslips (Fisher).

For EdU assays, the Click-iT Imaging Kit 712 with Alexa Fluor 488-azide (Invitrogen, C10337) was used to detect EdU in cryosections. For pimonidazole assay, sections were incubated with rabbit anti-pimonidazole antibody (1:200, Hypoxyprobe, Inc., PAb2627AP) overnight at 4 ^oC, and subsequently with Alexa Fluor 647-conjugated donkey anti-rabbit IgG (A31573) for 3 h at room temperature.

Imaging

Images for fixed sections were captured by an automated inverted fluorescence microscope with a structured illumination system (Zeiss Axio Observer 7 with ApoTome.3 system) and Zen 3.4 (blue edition) software. The Filter Set 112 SBP was used with excitation (Ex), beam-splitter (Bs), and emission (Em) filter wheel consisting of Ex. 385/30, 469/38, 555/30, 631/33, 735/40, Bs. 405 + 493 + 575 + 654 + 761, Em. 425/30, 514/30, 592/25, 681/45, 788/38 nm. The objectives used were: Fluar 2.5x/0.12, EC Plan-Neofluar 5x/0.16, Plan-Apochromat 10x/0.45, Plan-Apochromat 20x/0.80, EC Plan-Neofluar 40x/0.75, and Plan-Apochromat 63x/1.40. Images were typically tile-scanned with a motorized stage, Z-stacked, and reconstructed by a maximum intensity projection (MIP) function. Differential interference contrast (DIC) was used for objectives higher than or equal to 10x. Representative images of at least three independent biological samples are shown in the figures.

Three-dimensional micro-computed tomography analysis of mouse samples

Mandibles, including molars and incisors, were placed in a 19 mm diameter specimen holder and scanned using a microCT system (µCT100 Scanco Medical, Bassersdorf, Switzerland). Scan settings were voxel size 12 µm, 70kVp, 114 µA, 0.5 mm AL filter, and integration time 500 ms. The original DICOM files were converted to de-identified files in gipl.gz format using the ITK-SNAP open-source software^66,67. Segmentations of each sample created by ITK-SNAP were then converted to 3D surface models in Slicer open-source software. Subsequently, each parameter was calculated using the Q3DC tool in Slicer. The details have been described previously¹⁶. Femur analysis was performed using Dragonfly software (Comet Technologies Canada Inc). Trabecular analysis was performed in a region beginning 120 μm below the growth plate and ending after 360 μm (30 slices). Cortical analysis was performed in the region beginning 600 μm below the growth plate and ending after 360 μm (30 slices), and a secondary ROI was used to exclude any trabecular bone located in the region of cortical analysis.

Periodontal drill-hole surgery

The periodontal drill-hole defects were created using the method modified from those reported by Padial-Molina et al. and Nagata et al.^40,41. Mice were anesthetized with 1.5–2.0% isoflurane (Fluriso, Isoflurane USP, VetOne) through a nose cone. A vertical extraoral incision was made at the right mandible under a stereoscopic microscope. The masseter muscle was excised to expose the buccal plate. A 0.9 mm diameter stainless steel bur was used with a rotary instrument (Fine Science Tools) to carefully drill through the buccal bone, periodontal ligament, cementum, dentin of the furcation, and the mesial root of the mandibular first molar. The dimension of the periodontal defect in the buccal area was 1.0 mm diameter in width 1.0 mm at maximum depth. The masseters and the skin were sutured with 7-0 resorbable sutures (Vicryl, Ethicon Inc., Somerville, NJ) and 6-0 silk (Surgical Specialties Corporation, Wyomissing, PA).

Ligature-induced periodontitis model

Localized periodontitis was induced by placing bacteria-retentive silk sutures between the maxillary molars, as previously described⁴². Mice were anesthetized with 1.5–2.0% isoflurane (Fluriso, Isoflurane USP, VetOne) through a nose cone. 6-0 Silk sutures were inserted between the first and second maxillary molars on the left side in 12-week-old mice. The contralateral molars were kept unligated and treated as an internal control. Both ends of the suture string were knotted to prevent ligature dislodgement. Mice were euthanized at the end points of each experiment. The mice were sacrificed by isoflurane overdose at the endpoints of each experiment, and both left and right maxilla, including molars and periodontal tissue samples, were collected.

Pharmacological inhibition of TGF-β signaling pathway

TGF-β receptor type I (TGF-bRI) kinase inhibitor galunisertib (Eli Lilly and Company, LY2157299) was reconstituted in DMSO as a stock solution (20 mg/ml) and kept at − 20 ^oC until use. LDE225 was further diluted in PBS and intraperitoneally administered at 2 μg/g b.w., once a day for 13 days (P8-P20).

Cell preparation

Gingival tissues of detached mandibles were removed entirely using sharp forceps, and dentoalveolar components, including molars, dental pulps, dental sacs, or periodontal tissues, were carefully resected using a disposable scalpel (No.15, Graham-Field). Molars (M1 and M2) were carefully extracted from sockets in a 35 mm dish containing 3 ml Ca²⁺, Mg²⁺-free Hank’s Balanced Salt Solution (HBSS, Sigma H6648) containing 2 Wunsch units of Liberase TM (Roche) and incubated at 37 ^oC for 15 min. on a shaking incubator (ThermomixerR, Eppendorf). Cells were obtained by rigorous pipetting and filtration through a 70 μm cell strainer (BD) into a 50 ml tube on ice to make a single cell suspension. Cells were pelleted and resuspended in an appropriate medium for subsequent purposes.

Flow cytometry

Dissociated dental pulp cells were stained by standard protocols with the following antibodies (1:500, Invitrogen/eBioscience). eFluor450-conjugated CD45 (30F-11, hematopoietic, 48-0451-82) and allophycocyanin (APC)-conjugated CD184 (CXCR4) (2B11, 17-9991-82). Flow cytometry analysis was performed using a four-laser BD LSR Fortessa (Ex. 405/488/561/640 nm) and FACSDiva software. Acquired raw data were further analyzed on FlowJo software (TreeStar). Representative plots of at least three independent biological samples are shown in the figures.

Single-cell RNA sequencing (scRNA-seq) analysis of FACS-sorted cells

Mice from the same litter were used to pool FACS-sorted GFP⁺ or tdTomato⁺ cells. Cell sorting was performed using a four-laser BD FACS Aria III (Ex.407/488/561/633 nm) high-speed cell sorter with a 100 μm nozzle. GFP⁺ or tdT⁺ cells were directly sorted into ice-cold Dulbecco’s phosphate-buffered saline (DPBS) /1% FBS. Cell numbers were quantified by Countless II automated Cell Counter (ThermoFisher) before loading onto the Chromium Single Cell 3’ microfluidics chip (10X Genomics Inc., Pleasanton, CA). cDNA libraries were sequenced by Illumina Hi-Seq 4000 or NovaSeq 6000. The sequencing data was first preprocessed using the 10X Genomics Cell Ranger software. For alignment purposes, we generated and used a custom genome fasta and index file by including the sequences of eGFP and tdTomato-WPRE in the mouse genome (mm10). Further downstream analysis steps were performed using the Seurat R package⁶⁸. We filtered out cells with less than 500 genes per cell and more than 20% mitochondrial read content. The downstream analysis steps include normalization, identification of highly variable genes across the single cells, scaling based on the number of UMI, dimensionality reduction (PCA and UMAP), unsupervised clustering, and the discovery of differentially expressed cell-type-specific markers. Differential gene expression was performed to identify cell-type-specific genes using the non-parametric Wilcoxon rank sum test in the rliger R package⁶⁹. Differentially expressed genes were defined as those passing thresholds with a p-adjusted value of 0.05 and logFC of 1. The scRNA-seq dataset presented herein has been deposited in the National Center for Biotechnology Information (NCBI)‘s Gene Expression Omnibus (GEO) and is accessible through GEO Series accession numbers GSM7911151, GSM7911152, GSM7911153.

We utilized VeloVAE^46,47 to infer the RNA velocity of the root-forming cells. We removed the tdTomato^- populations from our scRNAseq dataset (Clusters 1,5,9,10,11,12,13,14) before employing the velocity inference model to our data. Briefly, preprocessing steps include filtering cells and genes based on mRNA counts, computing a KNN graph based on PCA, and computing moments based on connectivities. By employing velovae.VAE() method, initializing the rate_prior parameter with priors for transcription, splicing and degradation rates, and setting full_vb = True, we defined a FullVB veloVAE model. The model was trained using the train() method with default parameters. The only hyperparameter used by the FullVB model is the coefficient of KL divergence between the rate parameter distributions, and is set to 1.0 by default. The RNA velocity embedding computed by the model was then projected onto the UMAP coordinates to plot the velocity vectors that define the direction of cell differentiation in the final velocity stream plot.

To compute a global map of cellular fate potential, we performed cell fate mapping inference by carrying out the protocol from scVelo⁷⁰. We used scv.tl.velocity_graph() to perform basic cell state categorization followed by scv.tl.terminal_states() to compute the root cells and end points, both using default settings. These terminal cell state probabilities were plotted using the scv.pl.scatter() function on top of the UMAP coordinates. We applied CellChat to identify and study the intercellular signaling networks between clusters in our scRNAseq data. We started with the creating a CellChat object using the processed Seurat object with createCellChat(). We then applied subsetData() to filter the expression data of signaling genes as per the protocol. After adding the CellChatDB.mouse database to the CellChat object, we applied identifyOverExpressedGenes() and identifyOverExpressedInteractions() to discover over-expressed genes as well as the overexpressed ligand-receptor pairs (statistical significance level defined by the Bonferroni corrected p-value < 0.05). The communication probability of the signaling pathways was inferred using computeCommunProb(), followed by computeCommunProbPathway and aggregateNet() for computing and aggregating networks of communication between the cells. Lastly, we used netVisual_chord_gene() to plot the Ligand-Receptor signaling network between dental Epithelial Cells and AP cells for the “WNT” signaling pathway.

We used the SCENIC workflow for identifying the enriched TFs and their target genes using python implementation of SCENIC (pySCENIC)⁷¹. Cytoscape 3.10.3⁷² were used to build a regulatory network.

RNA isolation of FACS-sorted apical papilla cells

Dissociated dental papilla cells harvested at P10 littermate mice were pooled based on the genotype [Cxcl12-creER; Ctnnb1^fl/+; R26R^tdTomato (Ctnnb1-cHet) or Cxcl12-creER; Ctnnb1^fl/fl; R26R^tdTomato (Ctnnb1-cKO)]. Cell sorting was performed using a four-laser BD FACS Aria II (Ex.407/488/561/633 nm). tdTomato⁺ cells were directly sorted into ice-cold DPBS/10% FBS and pelleted by centrifugation. Total RNA was isolated using the PicoPure RNA Isolation Kit (KIT0204, ThermoFisher), followed by a DNA-free DNA removal kit (AM1906, ThermoFisher) to remove contaminating genomic DNA.

Bulk RNA-seq analysis

RNA samples were submitted to the Cancer Genomics Center at The University of Texas Health Science Center at Houston, supported by CPRIT RP180734. Total RNA was quality-checked using the Agilent RNA 6000 Pico kit (#5067-1513) by Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, USA). RNA with an integrity number greater than 7 was used for library preparation. Libraries were prepared with SMART-Seq V4 PLUS Kit (R400752, Takara Bio, Japan) following the manufacturer’s instructions. The quality of the final libraries was examined using Agilent High-Sensitive DNA Kit (#5067-4626) by Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, USA), and the library concentrations were determined by qPCR using Collibri Library Quantification kit (#A38524500, Thermo Fisher Scientific). The libraries were pooled evenly and went for the paired-end 75-cycle sequencing on an Illumina NextSeq 550 System (Illumina, Inc., USA) using High Output Kit v2.5 (#20024907, Illumina, Inc., USA). The raw RNA sequencing data were processed starting with the concatenation of read files from each end into a single FASTQ file. Adapter trimming and quality control steps were performed using Trim Galore⁷³. The sequencing reads were then mapped to the GRcm39 reference genome from GENCODE⁷⁴ using STAR⁷⁵. Differential gene expression analysis was conducted using DESeq2⁷⁶, selecting genes with an FDR less than 0.05 and a fold change greater than 2 as differentially expressed genes. The Database for Annotation, Visualization, and Integrated Discovery (DAVID) (https://david.ncifcrf.gov/home.jsp) was used for Gene Ontology (GO) enrichment analysis. For the Gene Set Enrichment Analysis (GSEA)⁷⁷, the KEGG, GO, and REACTOME gene sets were utilized, taking test statistics from the DESeq2 as input in the pre-ranked test. Differential gene sets were identified with an FDR less than 0.25. The bulk RNA-seq data can be accessed under the accession number GSE248398 in the Gene Expression Omnibus (GEO) database.

Statistical analysis

Results are presented as mean values ± S.D. Statistical evaluation was conducted based on the One-way ANOVA, followed by Mann-Whitney’s U-test or paired t test. A p-value of < 0.05 was considered significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.N., N.O., and W.O. conceived the project. M.N., G.T.G., Y.Y., N.O. and W.O. wrote the manuscript. M.N., G.T.G., T.K., Y.A., H.M., A.K.Y.C., R.K., M.A., Y.Y., C.T.A., Y.N., Y.M., N.T., J.D.W., N.O. and W.O. performed the experiment. W.J.Z. Critiqued and critically revised the manuscript.

Peer review

Peer review information

Nature Communications thanks Matthew Greenblatt and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The raw imaging and scanning data generated during and/or analyzed during the current study are available from the corresponding author on reasonable request due to the large number and size of the files. The single-cell RNA-seq, and bulk RNA-seq data presented herein have been deposited in the National Center for Biotechnology Information (NCBI)’s Gene Expression Omnibus (GEO), and are accessible through the GEO Series accession number, SuperSeries GSE248300 including GSM7911151-GSM7911153, and SuperSeries GSE248398 including GSM7912965- GSM7912972. Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-61048-x.