Femoral bones were explanted, muscles largely removed in a way that osteotomized bone parts maintained one entity. Tissue was fixed using 4% electron microscopy-grade PFA in PBS for 4 h at 4°C, washed in PBS, and ran through a sucrose gradient (10%, 20%, 30%; á 12–24 h). The fixators were removed from the fixed samples, underwent μCT measurement, bones were frozen in SCEM medium (Sectionlab, Japan), cut into slices of 7 μm using Kawamoto‘s film method (35), and stored at −80°C.

Movat's Pentachrome staining was conducted as descried previously (21, 24). TRAP staining for quantification was performed using a kit following the manufacturer's instructions (Thermo Scientific, 386A-1KT, MA, US). Individual slides were stained using small volumes of staining solutions on a heating plate at 37°C. For immunofluorescence, individual sections were thawed, rehydrated in PBS, blocked with 10% donkey serum, and stained with antibodies in PBS/0.1% Tween 20/5% donkey serum containing DAPI for 1–2 h. Target proteins were identified using antibodies against CD31/PECAM-1 (goat polyclonal unconjugated, AF2628, R&D Systems, 1:100), CD206/MMR (C068C2 conjugated to AF594, BLD-141726, 1:100), CD80 (goat polyclonal unconjugated, AF740-SP, 1:100), Endomucin (Emcn) (V.7C7 unconjugated, sc-65495, 1:100), F4/80 (Cl:A3-1 unconjugated, MCA497G, 1:400), GFP (goat polyclonal conjugated to AF488, 600-101-215, 1:100), Ly-6C (ER-MP20 biotinylated, MA5-16666, 1:20), Ly-6G (1A8 biotinylated, BLD-127603, 1:200), Mac-2/Galectin-3 (M3/38 unconjugated, BLD-125401, 1:100), Osx (rabbit polyclonal, sc-22536-R, 1:200), Runx2 (EPR14334 conjugated AF647, ab215955, 1:100), Sox9 (EPR14335 unconjugated, ab185230, 1:200), VEGFA (rabbit polyclonal unconjugated, ab46154, 1:100). Primary antibodies were stained with secondary antibodies when unconjugated (1:500, Thermo Fisher, anti-rat conjugated AF488, A21208; anti-rat conjugated AF546, A11081; anti-rat conjugated AF594, A21209; anti-rabbit conjugated AF488, A21206; anti-rabbit conjugated AF546, A10040; anti-rabbit conjugated AF647, A31573; anti-goat conjugated AF647, A21447; or streptavidin conjugated AF546, S11225). Samples were washed between steps and after staining with PBS/0.1 % Tween 20 for 3 × 5 min. Stained samples were kept in PBS for 5 min and embedded using aqueous mounting medium (Fluoromount, Thermo Fisher, MA, US) and analyzed microscopically within 6 days. Simultaneous detection of Ly6C and Ly6G was considered to indicate presence of the Gr-1 protein.

Movat's Pentachrome images were taken with a light microscope in a 2.5 × magnification and the program AxioVision (both Carl Zeiss Microscopy GmbH, Germany). For Figure 2, images (CD31 & Emcn) were taken with a Keyence microscope (BZ 9000) using a 10x or 4x objective. All other images were acquired at a Zeiss LSM880 in tile scan mode at a resolution of 2048 × 2048 using a 20x objective, unless specified otherwise. For display, pictures were background subtracted and contrast adjusted using ImageJ 1.52i.

Image analysis of cell and tissue distribution (Figure 2) was performed with ImageJ and an own developed pipeline which has been published and described in detail previously (21). Mean intensity was determined with ImageJ within the marked ROIs distinguishing between the gap, the adjacent to the gap and the bone marrow area. Mean intensities were normalized to the maximum intensity of the image.

Quantification was performed in CellProfiler 3.1.8. (36) (Supplementary Figure 1). Macrophage subsets were described via the co-localization of identified CX3CR1⁺, Gr-1⁺, and F4/80⁺ objects. CX3CR1⁺F4/80⁺Gr1⁺ objects, representing cells were divided by roundness based on object shape features (FormFactor, Perimeter, Min Feret Diameter, Supplementary Figure 5). Objects were only considered cells when they overlapped with a nucleus (DAPI) signal. In a second pipeline localization of F4/80⁺ macrophages towards the Emcn⁺ endothelium was analyzed by examination of the direct (≤3.5 μm) and distant (>7 μm) neighborhood of the subset. Localization of the identified subsets was analyzed by counting the identified objects in regions of interests (ROIs). ROIs for the osteotomy gap were determined as shown in Figure 5E. For neighborhood analysis and frequency determination (Figures 3E,F) either Emcn^hi or Emcn^lo areas within the osteotomy gap or distant thereof were encircled freehand using ImageJ 1.52i, respectively (Figure 3D). Areas close to Sox9⁺ chondrocytes were excluded as they did contain few, if any macrophages. Emcn^hi areas largely overlapped with F4/80^hi areas.

As a GRIN lens system, we used custom singlet GRIN needle microendoscope (length ca. 5.07 mm, diameter = 0.60 mm; NEM-060-10-10-850-S-1.0p, GRINTECH Jena, Germany). The GRIN lens was glued into the lens tubing to a final penetration depth of 650 μm when screwed into the fixator plate. CX3CR1:GFP mice were anesthetized and mounted to the microscope as previously described (10). Qtracker 655 Vascular Label (Thermo Fisher, MA, US) were injected and images were acquired (505 × 505 px, 500 × 500 μm, unidirectional, line average 4, step size 4.5–6.5 μm, ca. 18 steps, stack time 60 sec). Image stacks were loaded into Imaris 9.3.0 software (Bitplane Zürich, Switzerland), median filter (3 × 3 × 3) was applied. Videos were exported (1024 × 1024, 4 fps) and images were angled maximum intensity projections. Individual mice were measured at 940–950 nm on day 2, 3, 4, 5 after osteotomy.

Statistical analysis was carried out with GraphPad Prism V.5 or V.8 software. All values are expressed as the mean ± SD if not stated otherwise. Mann-Whitney U-test, Wilcoxon-signed rank test and Friedman test with Dunn's post-hoc test were mainly used, since Gaussian distribution was not expected due to inter-individual variations. A p < 0.05 was considered statistically significant. Image analysis was blinded for time points.

Fracture healing consists of consecutive phases. Progression into each phase depends on the undisturbed and error-free course of the respective previous phase (38). In the mouse-osteotomy model used in this study, residuals of cells in the fracture hematoma are visible in Movat's Pentachrome staining at day 3, while the osteotomy gap is filled with a mixture of bone marrow cells including hematopoietic cells at day 7, and endochondral bone formation occurs between day 14 and day 21 (Figure 1A). Using immunofluorescence histology, a positive signal for CD31 (Platelet endothelial cell adhesion molecule 1) is seen in elongated cells inside the fracture hematoma, which include no detectable nuclei (Supplementary Figures 2A–C), probably indicating that these residuals represent endothelial cells that may have lost their integrity. In order to further characterize the cellular composition within the fracture area, immunofluorescence staining for key transcription factors of mesenchymal differentiation was performed on serial sections of the same bones (Figure 1B). Already at day 3, osteoblast progenitors characterized by nuclear expression of Runt-related transcription factor 2 (Runx2) are dispersed in the fracture gap. By day 7, their number has increased, leading to a dense population of the gap, whilst sparing a region in the center. The peripheral borders of the gap and adjacent periosteal regions are populated by cells co-expressing Runx2 and SRY-box transcription factor 9 (Sox9) in the nucleus, indicating their potential for either osteogenic or chondrogenic differentiation (Figure 1B). This is in accordance with the blue-greenish color in corresponding regions of Movat's Pentachrome staining, indicative of cartilaginous tissue (Alcian blue positive). At day 14, Runx2⁺ cells are found to localize on both sides of the gap. Single Runx2⁺ cells are surrounded by areas characterized by the presence of only a few nuclei, in line with the presence of mineralized bone in those areas, as visualized by Movat's Pentachrome staining (compare: Figures 1A,B). Sox9⁺ cells are exclusively present in the center of the gap at this time point. Notably, the majority of these cells do not co-express Runx2. In contrast, at day 21, a time point when mineralization of the gap is complete, the progenitors identified in this area are almost exclusively Runx2⁺. In a next step, Osterix (Osx) and CD31 stainings were performed on serial slides from the same individuals. Direct comparison reveals that at day 3, only some Runx2⁺ cells are also Osx⁺, indicating various degrees of osteogenic differentiation. From day 7 on, Runx2⁺ cells are also positive for Osx⁺, consistent with ongoing maturation of osteoblasts. Osx⁺ osteoblasts localize in the fracture gap alongside existing and newly formed bone, as indicated by differential interference contrast (DIC) signal. Sox9⁺ cells can be found at day 14 and 21 in areas with cartilage and close to Osx⁺ cells, indicating ongoing further differentiation and mineralization, which is accompanied by vascularization. For further quantitative analysis of the vasculature sprouting into the gap, we defined the gap as a rectangular region between the cortical ends and analyzed CD31⁺ vessels and Osx⁺ bone cells in this region (Figure 1C). At day 3, CD31⁺ matrix (residuals of the fracture hematoma) and some CD31⁺ cells are present (compare to Supplementary Figures 2A–C), while Osx⁺ cells enter the osteotomy gap at later time points, between day 7–21 (Figures 1C,E). At later phases of fracture healing, endothelial cells displaying the phenotype of type H vessels (Emcn^hiCD31^hi) closely associate with Osx⁺ osteoprogenitors/pre-osteoblasts (Figure 1C). Investigating the co-localization of Runx2⁺ and Sox9⁺ cells with respect to CD31⁺ endothelium reveals a close localization of Runx2⁺ cells around vascularized areas, while Sox9⁺ cells are only found in less vascularized areas (Figure 1D). Our results confirm a close relationship between vessel formation and osteoprogenitors/osteoblasts and extend previous reports of this phenomenon, which focused on bone development (7), to a regenerative scenario.

While Emcn^hiCD31^hi endothelium can be found directly in the osteotomy gap and in the adjacent tissue, Emcn^loCD31^lo endothelium is prominent in bone marrow regions not affected by the injury. Pixel intensity analysis confirms the differences between the Emcn⁺ and/or CD31⁺ cells in and adjacent to the osteotomy gap, compared to the bone marrow (Figures 2A,B).

We further characterized the type H vessels in the fracture gap (Figures 2C,D). During vascularization of cartilaginous tissue, invading vessel buds (distal loops), previously described as a morphological criterion for type H endothelium (7), can be identified (Figure 2E, first row marked by arrows). Furthermore, CD31⁺ arterioles, which are negative for Emcn, are found within the osteotomy gap. However, a second morphological criterion described to be typical for type H vessels in the growth plate areas, namely their columnar structure, does not appear during regeneration, suggesting that this ordered arrangement is a specific feature of bones undergoing longitudinal growth. Taken together, our data reveal the appearance of Emcn^hiCD31^hi in the fracture gap, which share additional distinct morphological features with the previously described type H vessels, although they are not organized in columnar structures. This finding suggests an important role of this osteogenic vessel type not only during bone development, but also during regenerative processes of the bone.

In order to determine the role of macrophages in the context of vascularization during bone regeneration, we analyzed macrophage subsets and their spatial localization relative to type H vessels. Based on the pan-macrophage marker F4/80, which includes also osteomacs, we defined M1- and M2-like macrophages labeled by CD80 and CD206 (Mannose receptor), respectively, as well as Mac-2 (Galectin-3), a marker associated with pro-inflammatory macrophages (22). Over the course of endochondral bone regeneration, F4/80⁺ cells localize throughout the bone marrow, as well as on the border to bone surfaces, accumulating in various areas (Figure 3A). At day 3 post-osteotomy, high numbers of F4/80^hi cells localize in periosteal regions adjacent to the osteotomy gap. They can be found periosteal in varying numbers throughout the regeneration process, until the bone remodeling phase at day 21. At day 14, F4/80^hi cells are found exclusively in periosteal regions, whereas they are found in both the medulla and the periost at day 21. Extra-medullar F4/80⁺ macrophages are positive for the anti-inflammatory M2(a) macrophage marker CD206 (Figure 3A, blue framed inset). Those M2-like cells are not abundant in the marrow or at sites of vascularization, where expression of CD206 is generally lower and macrophage morphology rather ramified (Figure 3A, yellow and orange framed insets). CD206 expression is also found in cells resembling the morphology of endothelia (Emcn⁺), which are F4/80⁻ throughout the regeneration process as well as under homeostatic conditions (Figure 3A, yellow framed inset; Supplementary Figures 3C,D). M1-like macrophages, defined by expression of CD80, are extremely rare and only few are found at day 3, with their abundance comparable to control tissue (Supplementary Figures 3B,D). Using the marker Mac-2 in addition, almost all CD206⁺ macrophages are found to be positive for Mac-2 at day 3, however, not all Mac-2⁺ cells are CD206⁺ (Figure 3B). Mac-2 positive cells localize in the fracture gap and in proximity to the bone surfaces. Over the course of regeneration, double-positive Mac-2⁺ and CD206⁺ cells vanish and cells which are single positive for each of the two markers are detected at day 21, a time point when remodeling is ongoing (Figure 3B). Quantitative pixel-based area analysis of F4/80 macrophages, CD31 and TRAP shows that CD31⁺ and F4/80⁺ cells are reduced over time, while the number of TRAP⁺ cells (osteoclasts or activated macrophages) increases (Figure 3C; Supplementary Figure 3A).

CD206⁺ macrophages are present throughout the healing process with no apparent preferential localization towards the vasculature, and they are mainly Mac-2⁺ in the early phase of regeneration. Over the course of regeneration, F4/80-expression decreases in regions adjacent to the callus, and at later time points only few areas, which contain cells expressing F4/80 at intermediate levels, are observed.

Since we could not find substantial amounts of CD80⁺ cells at day 3 post-osteotomy, we analyzed the regeneration at day 1 and day 2. Some CD80⁺ cells were found at day 1, and they were abundant in higher numbers at day 2 post-osteotomy (Figure 4A). Some of those are F4/80⁺CD80⁺ macrophages, which localize in areas adjacent to the osteotomy gap. CD80⁺ cells are located directly at the injury site (Figure 4A, day 2). Macrophages are known to support vascularization via VEGFA during tissue regeneration, which is why next, we analyzed VEGFA using immunofluorescence histology over the time course of bone regeneration (Figures 4B–D). VEGFA is not expressed by F4/80⁺CD80⁻ macrophages throughout the entire process, but CD80⁺F4/80⁻ and a fraction of CD80⁺F4/80⁺ cells express VEGFA. This expression was restricted to very early time points, namely day 1 and day 2 (Figures 4B,C). From day 3 on, VEGFA is found to be expressed in the fracture gap and along bone surfaces (endosteal, periosteal) and in the bone forming areas (Figure 4D). The signal for VEGFA at day 14 is pronounced in areas, which also contain Sox9⁺ and Runx2⁺ cells (Figures 1B, 4D). Taken together, VEGFA expression is observed in CD80⁺ cells, including a fraction of F4/80⁺ cells, in the early phase until day 3, after that, VEGFA is expressed predominantly by osteoblasts and chondrocytes in bone forming areas.

High abundance and expression of the vascular markers CD31 and Emcn is observed in both, the damaged tissue in the early regeneration, and in bone forming tissue over the subsequent course of regeneration in the osteotomy gap (Figure 2C). These areas rich in type H vessels, simultaneously contain F4/80^hi cells (Figures 3A,D). In order to evaluate the proximity in localization between macrophages and type H endothelium, we defined two regions for each sample, the Emcn^hi region and the Emcn^lo region (Figure 3E). Objects in these regions were segmented based on marker expression of Emcn, F4/80 and DAPI. Over the course of regeneration, similar frequencies of 25.0 ± 11.6 % F4/80⁺ cells are present in both, the Emcn^hi and Emcn^lo regions. The frequencies of F4/80⁺ cells decrease significantly from day 3 until day 21 post-injury (Figure 3E), confirming the qualitative results from Figure 3C. Next, we analyzed the position of macrophages relative to the endothelium, and found that, a two- to three-fold higher number of objects (F4/80⁺ objects or identified nuclei, which serve as a control for all cells) localize in proximity to the endothelium in the Emcn^hi region as compared to the Emcn^lo region (Figure 3F). Cells are defined as “in proximity” when they are located less than half a nuclear diameter in distance (<3.5 um) away from the endothelium.

During analysis of immunofluorescence staining in samples of osteotomized, LIMB-implanted bones, we noticed a transient, strongly polarized distribution of newly formed vessels to the distal area of the osteotomy gap (Figure 5A; Supplementary Figure 4A). A similar polarization occurs in samples of the osteotomy-model (Supplementary Figure 4B). In some cases, this phenomenon is accompanied by a variation in cell density in the respective area, as shown by differences in the abundance of the DAPI signal. Scoring according to the criteria described in the methods section for the abundance and intensity of the markers in the gap-adjacent regions reveals that CD31⁺ vessels are significantly brighter and more abundant adjacent distally to the gap, as compared to the proximal site (Figure 5B). This spatial polarization with respect to bright CD31⁺ vessels is pronounced at day 4 and decreases quickly afterwards (Figure 5B). A similar trend is observed for F4/80⁺ cells in the same regions (Figure 5B). Taken together, these analyses reveal a transient, directional spatial polarization (regarding the localization from the fracture gap) of CD31^hi vessels, adjacent to the fracture gap.

In order to understand the dynamics of macrophages during bone regeneration, we took advantage of the fractalkine receptor (CX3CR1) reporter mouse strain CX3CR1:GFP. These mice were implanted with a Gradient Refractive INdex (GRIN) lens, enabling longitudinal imaging of the bone marrow during regeneration (10). At the same time, these mice underwent osteotomy surgery. Since the first wave of vascularization occurs during resolution of the fracture hematoma, bones that had been treated using the procedure described were fixed and analyzed at day 4, 6, and 8 using immunofluorescence histology. In order to be able to exclude monocytes and granulocytes from the macrophage analysis, we stained sections using antibodies against Ly-6C/Ly-6G (Gr-1). We find that almost all F4/80⁺ macrophages in the osteotomy gap are also CX3CR1⁺ (Figures 5C,D, yellow), confirming that the reporter mice are suitable for tracking macrophage dynamics during bone regeneration. GFP⁺ cells are present within the osteotomy gap at early time points and become less abundant over time. They locate in close proximity to CD31^hi endothelium, do not express Gr-1 (Figure 5D), excluding the possibility that some of them are granulocytes, which are also described to be CX3CR1⁺. We analyzed the osteotomy gap between the cortices and the bone marrow as described in Figure 5E. At day 6, of all identified cells in the osteotomy gap, the CX3CR1⁺ cells display the highest increase in cell number, as compared to homeostasis. A minor increase is detected in F4/80⁺ cells and a strong reduction of Gr-1⁺ cells is observed compared to homeostasis (Figure 5F). The space is largely occupied by F4/80⁺ myeloid cells, which are not monocytes (Gr-1⁻). The strongest increase in cell number is observed in the CX3CR1⁺Gr-1-F4/80⁺ subset, which make up to 12% in ratio to all nuclei at day 6 (Figure 5G). Almost 6% of CX3CR1⁺ objects are F4/80⁻ and only few are Gr-1⁺. Morphologically, the cell phenotype is of non-round shape (Figure 5E, Supplementary Figure 5). We then longitudinally sampled time-lapse videos with two-photon microscopy from individual mice over the course of the early regeneration process. On the first and second day, almost no signal is detected inside the osteotomy gap (data not shown). On day 3, individual CX3CR1⁺ cells enter the hematoma (Figure 5Hd3). Those cells display a non-round, but non-ramified shape and move through the tissue (Supplementary Movie 1). A front of CX3CR1⁺ cells forms at the edge of the field of view at day 4, containing both round and non-round cellular phenotypes, forming a dense, partially resident population which expands into the entire field of view (Figure 5H, Supplementary Movies 2, 3). This invasion is accompanied by the occurrence of perfused vessels that expand, following the CX3CR1⁺ cell front until the field of view appears vascularized (Supplementary Movie 4). Under full vascularization, sessile CX3CR1⁺ cells localize towards the vasculature and motile cells can be observed to move along the endothelium (Supplementary Movie 3). Generally, the abundance of CX3CR1⁺ in the gap increases with progression of the regeneration process (Figure 5F, Supplementary Movies 3, 4).

These results indicate that most cells in the fracture gap are myeloid, non-monocytic, non-granulocytic, non-round CX3CR1⁺ which invade the osteotomy hematoma starting day 2–3, become gradually sessile in the fracture gap at day 3–4, where they precede the vasculature until it becomes fully perfused. After vascularization, CX3CR1⁺F4/80⁺ cells persist until the onset of the remodeling phase.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.