Transcriptome analysis of early fracture healing. Bones heal well in healthy individuals, but aging and confounding diseases can negatively impact this process (24–26). Despite our understanding of the critical role of osteogenic inducers in healing, little is known about the factors that regulate early events within the injured bone. Therefore, we focused our attention on day 3 following femoral fracture, as the expansion and commitment of SSPCs form the basis for the formation of mature lineages. To determine the potential molecular mechanisms regulating fracture healing, we performed scRNA-seq on periosteal cells isolated from intact and fractured bones (3 dpf) (Figure 1A). We sorted live, nonhematopoietic (CD45–) and hematopoietic (CD45+) cells (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.181073DS1) and performed scRNA-seq (10× Genomics).

Transcriptional profiling of periosteal nonhematopoietic cells from intact and fractured bones. (A) Experimental design for mouse treatment aimed at collecting samples for scRNA-seq. Femur fractures were induced in SMACre–ER/NICD1 and SMACre+ER/NICD1 male mice. To induce overexpression of NICD1, animals were treated with tamoxifen (Tx) on the day of fracture and 2 dpf. Femur samples of intact or injured periosteum were collected, digested, and cells were sorted for CD45– and CD45+. Subsequently, scRNA-seq was performed. (B) Clusters of periosteal CD45– cell populations with (C) violin and feature plots presenting characteristic conserved gene expression for each cluster from integrated intact and fractured Cre– and Cre+ samples are shown. (D) Proportion of cells within the control intact and fractured sample of each cluster. (E) Periosteal cells from Cre– intact and fractured samples were analyzed for cell cycle phases and cell proportion in the G2M phase. (F) GSEA indicates that MSCs1 is a stem/progenitor cell population within the periosteum and (G) Monocle3 trajectory analysis (https://cole-trapnell-lab.github.io/monocle-release/) shows cell differentiation from the MSCs1 cluster to mature chondrocytes and osteoblasts from integrated intact and fractured Cre– and Cre+ samples and (H) trajectory of the clusters in pseudotime.

Unsupervised clustering defined cell populations based on conserved gene expression in the clusters. Within the CD45– periosteal populations, we identified 16 clusters, including ECs with characteristic expression of Cdh5, satellite cells expressing Pax7, and smooth muscle cells expressing Myh11 (Figure 1, B and C). Several populations expressed Prxx1 and were classified as MSCs, whereas more mature cells expressed typical genes; chondrocytes (Col2a1), reticular cells (Adipoq and Lepr), and osteoblasts (Ibsp and Dmp1) (Figure 1, B and C). We employed cell sorting to enrich MSC and EC populations within periosteal samples, although there was some contamination with hematopoietic cells (clusters 3 and 10). Clusters 13, 14, and 15 exhibited low numbers of cells (under 25 cells) and were therefore not analyzed (Figure 1D). Periosteal cells from cluster 1 (MSCs1), cluster 2 (MSCs2), and cluster 5 (satellite cells) were expanded by day 3 after fracture (Figure 1D). However, ECs, MSCs1, MSCs3, satellite cells, smooth muscle cells, and osteoblasts exhibited an increasing proportion of cells in the G2M phase (Figure 1E). The MSCs1 cluster showed characteristics of quiescent stem cells, with a gene expression profile showing inhibition of cell cycle genes and mitosis (Cenpm, Rpa2/2, Rfc3/4, Ccnb2, Cenpa, and Spc24) as well as DNA replication (Gins2/4, Pold3, and Anapc15). Based on the gene expression profile and proliferative capacity before and after the fracture and gene set enrichment analysis (GSEA), we concluded that the least differentiated cluster is MSCs1, which was set as a root population in trajectory analysis. Based on a trajectory analysis, the MSCs1 cell population rapidly proliferated after fracture and transitioned into a highly proliferating MSCs2 cluster (Figure 1, E–G). GSEA showed that this MSCs1 cluster was increased in gene sets involved in asymmetric localization of planar cell polarity proteins (Scrib, Psmf1, and E2f3), extracellular matrix, collagen formation (Col2a1 and Col11a1), and degradation of the extracellular matrix, indicating a potential stem cell population (Figure 1F). These cells differentiate through MSC intermediate stages into chondrocytes or osteoblasts, as indicated by a trajectory analysis (Figure 1, G and H).

Upon fracture, ECs showed higher expression of Col18a1 (endostatin), Plvap, and genes encoding extracellular proteins like Col3a1, Col1a2, Col1a1, and Col15a1 in comparison with the intact periosteum (Figure 2A). These genes are involved in the healing process (27). The MSCs1 cluster showed a significant increase in the Cxcl5 chemokine, which recruits and activates leukocytes upon fracture. Genes that modulate the inflammatory response, such as Prg4, Ptx3, Cxcl12, Cxcl1, and Il4ra, were increased at the fracture site (Figure 2, A and B). With early tissue reorganization following injury, the expression of Timp1 and Mmp3 was increased. Many genes (e.g., Acta2, Ptx3, Cxcl12, Il4ra, and Cxcl5) were differentially expressed in all mesenchymal clusters of fractured samples compared with the intact periosteum. Only a few genes were differentially expressed in mature populations of chondrocytes (Figure 2A), whereas osteoblasts and CXCL12-abundant reticular (CAR) cells did not show any statistically significant differences in gene expression following the fracture. These mature populations were likely present at the time of a fracture, might have not gone through differentiation, and the bone injury did not significantly affect their transcriptional profile.

Periosteal cells from intact and fractured bones have distinct transcriptional profiles. (A) Heatmaps of EC, MSCs1, MSCs2, satellite cell, and chondrocyte clusters showing differentially expressed genes with significantly increased or decreased expression of Cre– control fractured compared with intact samples. Color intensity represents the mean gene expression of all cells within the cluster. (B) Feature plots of the specific genes identified in A that were found to be significantly increased in periosteal cells upon fracture compared with intact samples.

We performed CellChat (v2.1.2; https://github.com/jinworks/CellChat) analysis on our samples. More than 80% of signaling pathways in our samples (Cre– and Cre+, both intact and fractured) had significant cell-cell interactions (activin, BMP, Notch, PDGF, IGF, MMP, and BSP) in the same signaling pathways (Supplemental Table 1). IL-6 signaling interactions were specific only to the control fractured Cre– group. NEGR, SLIT, and WNT signaling pathways were specific pathways in the Cre– intact sample compared with fractured Cre– (Supplemental Figure 2A). IGF, collagen, BSP, VCAM, CXCL, PDGF, NOTCH, and periostin are some of the signaling pathways specific to the fractured compared with intact samples. In Notch interactions from ECs to other mesenchymal clusters of Cre– intact and fractured samples, we found that EC communication probabilities were strongest among JAG2 and DLL4 ligands interacting with Notch2 and Notch3, while interactions with Notch 1 were more probable in fractured samples. JAG1 did not interact with Notch receptors in intact samples (Supplemental Figure 2B).

Effect of Notch1 overexpression on fracture healing. To address the role of Notch1 signaling in the early regulation of bone healing, we isolated intact and fractured periosteal cells (3 dpf) from αSMACre+/NICD1 and Cre–/NICD1 mice. We observed an increase in the proportion of ECs (cluster 0) and some MSCs in clusters 6 (MSCs4) and 7 (MSCs5) within fractured samples with Notch1 overexpression (Figure 3A). scRNA-seq data indicated that fracture (Cre– and Cre+ samples) increases the proportion of mesenchymal populations, with a decreased proportion of ECs in both intact Cre– and Cre+ samples (Supplemental Figure 3). We postulate that this is not due to reduction of EC number, but due to a greater increase in MSC proliferation. The MSCs2 cluster from the injured periosteum of NICD1-overexpressing animals (Cre+) exhibited significantly increased expression of Ibsp and Alpl compared with the injured Cre– callus (Figure 3, C and D). Overexpression of NICD1 in fractured samples led to increased expression of genes involved in IFN signaling (i.e., Ifit1, Isg15, Ifit3, Iigp1, Ifi203, and Ifi27l2a) in MSC and EC clusters (Figure 3, B–D).

Overexpression of NICD1, during fracture healing, induces osteogenic and IFN signaling gene expression. (A) Proportion of cells within each cluster in control Cre– and Cre+ αSMACreER/NICD1 fractured samples. (B) Volcano plots showing differentially expressed genes in NICD1 fractured periosteal samples compared to NICD1 intact samples of EC, MSCs1, and MSCs2 clusters. (C) Heatmaps with the complete list of differentially expressed genes. Color intensity represents the mean gene expression of all cells within the cluster. (D) Feature plots of αSMACreER/NICD1 fractured Cre– and Cre+ samples showing increased osteogenic genes (Ibsp, Alpl) and IFN signaling genes (Isg15, Ifit1).

Despite the high proliferative activity of mesenchymal populations at 3 dpf, CD45+ hematopoietic cells remained a dominant population (~88% CD45+ cells). RNA-seq analysis revealed the presence of primarily myeloid cells (neutrophils, macrophages, and dendritic cells), with a small proportion of lymphocytes (Supplemental Figure 4). Neutrophils and macrophages expressed significantly higher levels of Tnf and Il1b upon fracture (Supplemental Figure 4C). However, overexpression of NICD1 in the αSMA-expressing cells resulted in a decrease in proinflammatory signals in CD45+ cells (Il1b, Tnf, and Il6) after bone injury, confirming interactions between MSCs and hematopoietic cells (Supplemental Figure 4D).

EC-derived DLL4 as a regulator of bone healing. We confirmed the expression of Dll4 and Jag2 in ECs, while Jag1 was expressed in MSCs (Supplemental Figure 5). We did not detect expression of Dll1 or Dll3 in periosteal populations. Notch1 showed strong expression in ECs, committed mesenchymal progenitors, CAR cells, satellite cells, and smooth muscle cells, while a low level of Notch2 was present in all nonhematopoietic cells within the callus (Supplemental Figure 5). Hes1, a downstream signaling gene of Notch, exhibited the highest expression in periosteal cells.

The only population expressing Dll4 was endothelial (Figure 4A), prompting us to pursue experiments disrupting Dll4 (Dll4Δ) in ECs using Cdh5CreER. We performed a lineage tracing experiment of Cdh5 cells using Cdh5CreER/Ai9 (Supplemental Figure 7). We injected tamoxifen on the day of fracture. Cdh5/Ai9 cells are ECs positive for the CD31 endothelial marker (Supplemental Figure 7). Lineage tracing of fractured femurs revealed already-formed blood vessels with Cdh5Cre/Ai9-labeled ECs in the bone marrow and in the muscle surrounding the fracture (Supplemental Figure 6a′). Cdh5/Ai9 cells were present within the expanded periosteum at 3 dpf (Supplemental Figure 6a′′). We observed a lack of infiltrating blood vessels within the cartilaginous tissue on days 7 and 10 (Supplemental Figure 6, b′ and C), but abundant ECs were present within the mineralized callus (Supplemental Figure 6, D and E). At later time points of fracture healing, blood vessels surrounded the callus tissue and penetrated the remodeling callus (Supplemental Figure 6e′).

Notch signaling in fracture healing. (A) Feature plot of Dll4 expression within periosteal cells. (B) Experimental design of recombination efficiency evaluation by determining mRNA expression of Dll4 and Notch downstream signaling genes (Hes1 and Hey1) in male and female mice. To induce recombination, tamoxifen (Tx) was injected at 0, 2, and 4 dpf, and gene expression was evaluated at 7 dpf. Dll4 was successfully deleted in male mice, with decreased expression of Hes1 and Hey1 in the Cre+ callus (C) and lungs (D). Males: Cre–n = 6, Cre+n = 6; females: Cre–n = 5, Cre+n = 5. (E) Proportion of thymocyte subpopulations in the thymus, where deletion of Dll4 in ECs on day 5 after the first tamoxifen injection induced a small decrease in CD4+ cells and an increase in CD8+ cells, with representative dot plots. (F) Flow cytometry analysis showing dot plots of CD31+ cells within the callus at 5 dpf expressing DLL4. (G) Proportion of callus cells expressing Sca1, CD90, PDGFRα, and CD51 at 5 dpf, with representative overlaid histograms of Sca1 and CD90 cell expression (unstained control, Cre– and Cre+ sample). In E–G, Cre–n = 3, Cre+n = 4. Unpaired, 2-tailed Student’s t test. *P < 0.05.

We evaluated fracture healing in young adult (10- to 12-week-old) mice lacking Dll4 in ECs using Cdh5CreER. We induced Cre activity by intraperitoneal injection of tamoxifen (75 mg/kg) on 0, 2, and 4 dpf (Figure 4B). Real-time PCR confirmed efficient Dll4 deletion in callus samples collected 7 dpf from male Cre+ mice compared with Cre– mice (Figure 4C) and in lung tissue abundant in blood vessels (Figure 4D). This deletion of Dll4 within ECs led to a decrease in expression of Hes1 in the bulk calluses of Cre+ mice (Figure 4C). Notably, deletion of Dll4 was less efficient in female mice (not statistically significant), and the Hes1 and Hey1 genes were not downregulated in female animals (Figure 4C).

Although Notch signaling is crucial for T lymphocyte differentiation (28), our inducible model of Dll4 deletion with tamoxifen did not lead to differences in B220+ and CD3+, CD4+, and CD8+ proportions within the bone marrow of an intact bone at 5 dpf (Supplemental Figure 8, A and B). We observed differences in differentiating T cell populations (CD4+CD8+, CD4+, and CD8+ T cells) within the thymus, where deletion of Dll4 decreased the proportion of CD4+ cells and increased CD8+ cells (Figure 4E).

We further evaluated the phenotype of the MSCs within the callus tissue at 5 dpf. With Dll4 deletion within ECs, the proportion of nonhematopoietic (CD45–Ter119–) cells was significantly decreased (23.3% fewer CD45–Ter119– cells, with 40.6% in Cre+ vs. 31.2% Cre– animals; P < 0.01). The proportion of ECs (CD45–CD31+) within the groups was unchanged (4.30% ± 0.91% in Cre+ and 5.47% ± 1.11% in Cre–). The lack of DLL4 expression in CD45–CD31+ cells was confirmed by flow cytometry (Figure 4F). The MSC population (CD45–Ter119–CD31–) within the callus of Cre+ mice exhibited a significantly decreased proportion of Sca1+ cells (P < 0.05) and a lower proportion of CD90+ cells (P = 0.05) (Figure 4G). The Sca1+CD51+ population was also decreased in Cre+ animals compared with Cre– littermate controls in the periosteal callus at 5 dpf (Supplemental Figure 8C).

We evaluated how the deletion of Dll4 in ECs affects callus healing through histology, micro-computed tomography (microCT), and mechanical testing. Histological evaluation of femur fractures revealed a smaller callus area at 4, 7, and 14 dpf (Figure 5, A and B). The reduced callus area indicated lower proliferation under DLL4 deficiency, which was confirmed by a significant decrease in EdU+ cells within the callus area of Cre+ animals at 4 dpf (Figure 5C). However, at 7 dpf, Cre+ animals exhibited more EdU+ cells. Despite the increased proliferation later in the healing process, the callus area did not reach the size of their littermate control Cre– animals, indicating a phenotype of delayed healing. We observed less cartilage area in Cre+ animals at 4 and 7 dpf (Figure 5, D and F), and significantly less mineralized tissue at 14 dpf (Figure 5, E and G). MicroCT analysis confirmed decreased bone mass within the callus at 14 dpf (Figure 5H). By day 21, there was no difference in bone mass and callus volume (Figure 5H), nor any difference in the biomechanical properties of the fractured femurs (Figure 5I). Although we observed decreased expression of Bglap in Cre+ mice within the callus at 7 dpf (Figure 5J), histological analysis of osteocalcin on 14 dpf did not show any difference in osteocalcin+ cells within the callus (Figure 5K).

Deletion of Dll4 in ECs impairs fracture healing. (A) Experimental design. Deletion of Dll4 in ECs was induced by injecting tamoxifen (Tx) at 0, 2, and 4 dpf. Fractured bone samples were evaluated on day 4, 7, and 14 by histology and day 21 by microCT and torsion testing. (B) Dll4 deletion led to a decreased callus size at 4 dpf (Cre–n = 6, Cre+n = 9), 7 dpf (Cre–n = 9, Cre+n = 8), and 14 dpf (Cre–n = 8, Cre+n = 10). (C) Dll4 deletion also resulted in decreased proliferation at 4 dpf (Cre–n = 5, Cre+n = 9), which increased by 7 dpf (Cre–n = 9, Cre+n = 8) in Cre+ compared with Cre– animals. (D and F) Cre+ mice had significantly less cartilage. Sample numbers at 4 dpf (Cre–n = 5, Cre+n = 9), 7 dpf (Cre–n = 9, Cre+n = 8), and 14 dpf (Cre–n = 8, Cre+n = 10) were analyzed by evaluating Safranin O–stained sections, as observed on representative sections. (E and G) Mineralized area was analyzed by von Kossa staining (Cre–n = 8, Cre+n = 10), as shown on Cre- and Cre+ representative sections. (H) MicroCT analysis showed decreased callus bone mass on day 14 and no difference in callus volume, with representative 3D reconstructions of Cre– and Cre+ fractures on the right. At 7 dpf, Cre–n = 8, Cre+n = 6; and at 21 dpf Cre–n = 7, Cre+n = 9. (I) Biomechanical properties were evaluated by torsion testing and are presented as bone strength (maximum torque), stiffness as a measure of torsional rigidity, and toughness as a work to fracture measure, with no change between Cre– and Cre+ fractured bones at 21 dpf. Cre–n = 9, Cre+n = 6. (J) Bglap gene expression analysis at 7 dpf (Cre–n = 6, Cre+n = 6). (K) Histological analysis of osteocalcin-stained samples at 14 dpf (Cre–n = 8, Cre+n = 10). (L) Evaluation of CD31+ cells within the periosteal callus at 7 dpf (Cre–n = 9, Cre+n = 12) and 14 dpf (Cre–n = 7, Cre+n = 9). Deletion of Dll4 in ECs with (M) representative magnified images of CD31 staining within the mineralized callus at 14 dpf. Scale bar: 200 μm. (N) Proportion of osterix-stained cells within the periosteal callus (Cre–n = 5, Cre+n = 6). (O) Representative images of osterix-stained fractured femurs with magnified areas of cartilaginous callus (a and b), mineralized callus (a’ and b’), and cortical bone with the area next to the pin insertion (a” and b”). The analyzed periosteal callus is shown by the yellow line. Scale bars: 1 mm (left) and 100 μm (right). PT, pin trace; CB, cortical bone; P, periosteum. Unpaired, 2-tailed Student’s t test. *P < 0.05, **P < 0.01.

Given the important role of DLL4 in vasculogenesis, we evaluated the presence of autocrine effects of Dll4 deletion on EC number within the callus. We stained femur fracture sections at 7 and 14 dpf for the EC marker CD31 (Figure 5, L and M), which colabels Cdh5/Ai9 cells within the callus area (Supplemental Figure 7). We observed no difference in CD31+DAPI+ cells within the periosteal callus when comparing the Cre– and Cre+ groups (Figure 5, L and M). Next, we evaluated the effects on the population of osteoprogenitors and determined a 43% decrease in osterix+ (Osx+) cells in the periosteal callus upon deletion of Dll4 in ECs (33.4% in Cre– compared with 19.1% in Cre+) (Figure 5, N and O).

As an important control, we also crossed the Dll4-floxed mice with αSMACreER to verify the EC-specific requirements for DLL4 in fracture healing. We evaluated the expression of Dll4 in the fractured callus at 7 dpf upon tamoxifen treatment (at 0, 2, and 4 dpf) in both Cre– and Cre+ animals. We observed no change in Dll4 expression when Dll4 was deleted under the αSMA promotor using αSMACreER/Dll4Δ mice (Supplemental Figure 9), and no effect on Hes1 and Hey1 expression, which is consistent with our scRNA-seq data showing the lack of Dll4 expression in all mesenchymal clusters (Supplemental Figure 5).

Notch ligands as a therapeutic approach for bone healing. We used a clinically relevant critical size femoral defect model to examine the role of Notch signaling during bone healing. We utilized DLL4 and JAG1 alone or in combination with BMP2. Because Jag1 is more broadly expressed within periosteal tissue (MSCs, muscle cells, preosteoblasts, and ECs) than Dll4 (Supplemental Figure 5), we expected it to exhibit a stronger healing phenotype with JAG1 treatment. JAG1 (6 μg) alone was not sufficient to induce bone bridging of a 2 mm defect within 9 weeks of healing. However, when combined with 5 μg of BMP2, JAG1 significantly increased bone volume within the defect area compared with BMP2 treatment alone (Figure 6). Similarly to JAG1, DLL4 treatment (5 μg) alone did not result in bridging of a defect. However, when combined with BMP2, bridging of the defect was achieved. The cortical bone morphology within the defect resembled physiological appearance more than with BMP2 treatment alone.

Notch ligands improve critical size femoral defect healing. BMP2 (5 μg), JAG1 (6 μg), and DLL4 (5 μg) alone or in combination were applied to a collagen scaffold (Medtronic). Femurs were evaluated by microCT and bone volume within the defect area was determined. JAG1 in combination with BMP2 resulted in significantly more bone within the defect compared with BMP2 only, with cortical bone formation in the whole length of a defect. On the right, representative 3D reconstructions of healing femoral defects 9 weeks after defect surgery. Saline, n = 5; BMP2, n = 12; JAG1, n = 11; BMP2 + JAG1, n = 10; DLL4, n = 4; BMP2 + DLL4, n = 11. One-way ANOVA with Bonferroni’s post hoc test. ***P < 0.001, statistically different from saline; ##P < 0.01, ###P < 0.001, statistically different from BMP2; †††P < 0.001, statistically different from BMP2 + JAG1; ‡‡‡P < 0.001, statistically different from BMP2 + DLL4.