## Introduction

The objective of this study was to assess the influence of individualized cyclic mechanical loading on the remodelling phase of fracture healing via longitudinal in vivo micro-CT imaging. For this purpose, we developed a femur defect loading model, that allows application of individualized cyclic mechanical loading based on computed strain distribution in the mineralized callus using animal-specific real-time micro-finite element analysis with 2D/3D visualizations and strain histograms. We hypothesized, that individualized cyclic mechanical loading improves structural and dynamic properties of the mineralized callus during the remodelling phase of fracture healing.

The study showed that the remodelling phase of fracture healing is highly responsive to cyclic mechanical loading leading to significantly accelerated and larger mineralized callus formation and a higher degree of mineralization. Loading-mediated maintenance of callus remodelling was associated with distinct effects on Wnt-signalling-associated molecular targets Sclerostin and RANKL in callus sub-regions and the adjacent cortex of one randomly selected animal per group. The tightly controlled femur defect loading model could in the future widen our knowledge on the local spatio-temporal mechano-molecular regulation of all fracture healing phases.

## Results

### General physical observation

All mice recovered rapidly from surgery and there were no dropouts in both groups (loading, control) during the complete experimental period. The body weight of the animals did not significantly change over time (see Supplementary Fig. S1). Similar to observations from a previous femur defect study25, nest building started on the day of surgery and all animals were found in groups in the nest in the morning of post-operative day 1. In both groups, social interaction between mice and nesting behaviour during the pre-loading healing period (week 0–3) and the loading period (week 3–7) did not differ from pre-surgical observations.

### Volumes of interest (VOI) for evaluation by time-lapsed in vivo micro-CT

In order to exclude bias in the further micro-CT analyses, we compared the size of the defects and different VOIs (depicted in Fig. 3; for details on VOI generation see Tourolle né Betts et al.27). The defect sizes were 0.63 ± 0.07 mm for the control and 0.60 ± 0.07 mm for the loading group (n = 10/group). The VOIs encompassed the following volume for the control (n = 10) and the loading group (n = 10): 1.14 ± 0.13 mm3 versus 1.04 ± 0.16 mm3 for the defect centre (DC), 6.68 ± 0.77 mm3 versus 6.35 ± 0.82 mm3 for the defect periphery (DP), 2.28 ± 0.17 mm3 versus 2.31 ± 0.16 mm3 for the cortical fragments (FC), 15.44 ± 0.70 mm3 versus 15.84 ± 0.99 mm3 for the fragment periphery (FP). No significant differences in volume were detected in any of the VOIs between groups.

From week 0 to week 3, in both groups, strong resorptive activities (Figs. 2, 3g) were seen in the adjacent cortical fragments (FC). BRR showed significant weekly increases whereas no significant weekly changes in bone formation were observed in this VOI. The strong resorptive activities let to decreased bone volume by week 3 reaching statistical significance in the loading group (p < 0.0001). Furthermore, the fraction of highly mineralized bone significantly decreased from week 1 to week 3 in both groups (p < 0.0001), indicating cortical reorganisation.

### Individualized cyclic mechanical loading improves properties of the mineralized callus during the remodelling phase of fracture healing

Loading also affected the bone turnover in the adjacent cortex (FC). Already after 1 week, loaded animals showed significantly lower cortical bone resorption compared to controls (Fig. 3j). Whereas cortical volume significantly declined (p < 0.0001) in control animals from week 3 to week 7 by 27%, it significantly increased (p < 0.0001) by 20% in the loaded animals during the same time period (Fig. 3k). From week 4 to week 7 loading was also associated with significantly higher cortical mineralization (+ 4% in week 4 to + 54% in week 7) compared to controls (week 4: p = 0.0459, week 5–week 7: p < 0.0001; Fig. 3l).

In week 7, the FC VOI comprised 46% and 34% of the osseous tissue in the total VOI (TOT) for the control and loading group, respectively. In the DC VOI, 20% (control group) and 17% (loading group) of the total bone volume were seen. In the two peripheral VOIs, 9% (control group) and 14% (loading group) of the total osseous tissue were seen in the DP and 26% (control group) and 35% (loading group) in the FP VOI. Loading was associated with a shift in the bone distribution from the cortical fragments to the peripheral VOIs.

### Histology

In order to also visualize the tissue composition of the callus, we performed end point histological stainings of serial sections from one randomly selected animal of each group. Histology supported our micro-CT findings with complete cortical bridging seen in the animal from the loading and from the control group (Fig. 4b-i). The loaded animal showed a larger callus with some cartilage fractions suggesting ongoing endochondral ossification processes, whereas in the control animal the callus was largely remodelled with restoration of the medullary cavity indicating proceeding towards the end of healing (Fig. 4b-e) further supporting the observations from micro-CT cross-sections (n = 10/group) shown in Supplementary Fig. S3.

Furthermore, differences in the shape of the cell nuclei were observed between the osteocytes in the fracture callus (round shape) of the loaded animal in comparison to the cell nuclei seen in the cortical bone of the same animal (ellipsoidal shape) and the cell morphology seen in the control animals (Fig. 5a + b), suggesting an association between the local mechanical environment and cellular morphology. In regions with lower strains visualized by micro-FE (endosteal fracture callus; Figs. 4a, 5b), rounder cell nuclei were observed, compared to ellipsoidal nuclei shape in higher strained regions (cortex; Figs. 4a, 5b). To capture potential underlying mechano-molecular targets, we performed immunohistochemistry of Sclerostin (inhibitor of the mechano-responsive and osteoanabolic Wnt signalling pathway) and RANKL (negatively-regulated target gene of Wnt signalling) which have both previously been associated with the mechanical regulation of bone adaptation and healing16,29,30,31,32,33. Less abundant Sclerostin staining was visible in the fracture callus compared to the cortical bone of the loaded animal (Fig. 4 f and 5b) with no region-specific differences in staining patterns being seen in the control animal (Figs. 4g, 5b), suggesting that Wnt-signalling contributes to the loading-mediated osteoanabolic effects on fracture healing seen in this study. At the same time RANKL, which is down-regulated by Wnt-signalling, was more abundantly and stronger expressed in the callus of the loaded animal compared to the control animal (Fig. 4h + i and 5b). In line with previous studies, strong RANKL expression was seen in lowly strained callus regions as visualized via micro-FE analysis (Figs. 4a, 5b), indicating ongoing bone remodelling.

## Methods

### Animals

All animal procedures were approved by the Commission on Animal Experimentation (license number: 181/2015; Kantonales Veterinäramt Zürich, Zurich, Switzerland). We confirm that all methods were carried out in accordance with relevant guidelines and regulations (Swiss Animal Welfare Act and Ordinance (TSchG, TSchV)) and reported considering ARRIVE guidelines. Animal experiments were performed using previously established protocols for osteotomy surgery25, in vivo micro-CT imaging26, analgesia/anaesthesia25,26 and post-operative monitoring25. To study adult fracture healing female 12 week-old C57BL/6J mice were purchased from Janvier (Saint Berthevin Cedex, France) and housed in the animal facility of the ETH Phenomics Center (EPIC; 12 h:12 h light–dark cycle, maintenance feed (3437, KLIBA NAFAG, Kaiseraugst, Switzerland), 5 animals/cage) for 8 weeks. At an age of 20 weeks, all animals received a femur defect by performing an osteotomy with a 0.66 mm Gigli wire saw as previously described26 (group 1: control group, n = 10; group 2: loading group, n = 10; housing after surgery: 2–3 animals/cage; for details on study design see Supplementary Table 1). All defect surgeries were performed by the same veterinarian. Perioperative analgesia (25 mg/L, Tramal®, Gruenenthal GmbH, Aachen, Germany) was provided via the drinking water two days before surgery until the third post-operative day. For surgery and micro-CT scans, animals were anaesthetized with isoflurane (induction/maintenance: 5%/1–2% isoflurane/oxygen). Perioperative handling, and monitoring, micro-CT imaging and loading application was performed by the surgeon.

The four parts of the loading fixators (n = 20) were assembled as depicted in Fig. 1. To allow optimal identification throughout the experiments, one side part was engraved with a fixator-specific number (Fig. 1a,b). The stiffness of each fixator was measured using a Zwick testing machine and the fixators were assigned to the two groups, to allow similar distributions of fixator stiffness in the loading and control group (Supplementary Fig. S4).

### Femur osteotomy

In all animals an external fixator (Mouse ExFix, RISystem, Davos, Switzerland; mean stiffness: 17 N/mm; Supplementary Fig. 4) was positioned at the craniolateral aspect of the right femur and attached using four mounting pins. First, the most proximal pin was inserted approximately 2 mm proximal to the trochanter, followed by placement of the most distal and the inner pins. Subsequently, a femur defect was created using a Gigli wire (diameter: 0.66 mm).

### Time-lapsed in vivo micro-CT

Immediate post-surgery correct positioning of the fixator and the defect was visualized using a vivaCT 40 (Scanco Medical AG, Brüttisellen, Switzerland) (isotropic nominal resolution: 10.5 µm; 2 stacks of 211 slices; 55 kVp, 145 µA, 350 ms integration time, 500 projections per 180°, 21 mm field of view (FOV), scan duration ca. 15 min). Subsequently, the fracture callus and the adjacent bone between the inner pins of the fixator were scanned weekly using the same settings. Scans were registered consecutively using a branching scheme (registration of whole scan for bridged defects; separate registration of the two fragments for unbridged defects). Subsequently, morphometric indices (bone volume—BV, bone volume/total volume—BV/TV, bone formation rate – BFR, bone resorption rate—BRR) were computed (threshold: 395 mg HA/cm3; for details on methods see Tourolle né Betts et al.27). To assess mineralization progression, a second threshold (645 mg HA/cm3) was applied and the ratio between highly and lowly mineralized tissue (BV645/BV395) was calculated. The two selected thresholds are included in our recently developed multidensity threshold approach27. According to the standard clinical evaluation of X-rays, the number of bridged cortices per callus was evaluated in two perpendicular planes (UCT Evaluation V6.5-1, Scanco Medical AG, Brüttisellen, Switzerland). A ‘‘healed fracture’’ was considered as having a minimum of at least three bridged cortices per callus.

For evaluation, four volumes of interest (VOIs) were defined, which were created automatically from the post-operative measurement as described in Tourolle né Betts et al.27 (Fig. 3): defect centre (DC) containing endosteal callus and newly formed cortices, defect periphery (DP) containing the central periosteal callus, cortical fragment centre (FC) containing the medullary cavity and old cortices on both sides of the defect, and fragment periphery (FP) containing the periosteal callus adjacent to the old cortices. Data were normalised to the central VOIs: DC/DC, DP/DC, FC/FC, FP/FC.

Defect sizes (h) for each animal were calculated in MATLAB (version R2018b) based on the DC volume and the cross-sectional area of the proximal (CSAP) and distal cortices (CSAD): $${\text{h}} = \frac{2DC }{{\left( {CSAP + CSAD} \right)}}.$$

From week 4 to week 7, individualized cyclic loading (8–16 N, 10 Hz, 3000 cycles; 3 × /week; controls—0 N) was applied via the external loading fixator (Fig. 1b) based on computed strain distribution in the callus using animal-specific RTFE analysis (for detailed description of methods see24). Briefly, after weekly micro-CT measurements of each animal, the images were pre-processed using threshold-binning to create a high-resolution multi-density FE mesh, which was then solved on a supercomputer within the same anaesthetic session for each mouse. 2D and 3D visualizations of the mineralized callus were generated and a strain histogram was plotted. This information allowed load scaling based on the strains induced in the mineralized callus in individualized animals. Specifically, the distribution was scaled to achieve a median effective microstrain of 700. To assess regions that posed a failure risk the 99th percentile strains of each simulation was visualised in both image based and histogram forms. If more than 50 voxels exceeded 1% then the optimised load was reduced by 1 N.

### Mechanical testing

3 point bending tests were performed on a Zwick compression tester (ZwickRoell GmbH & Co, Ulm, Germany). Tests were conducted on a setup of 8 mm support span and up to 3 N. Tests were performed quasi-statically with a cross-head speed of 0.4 mm/min. Stiffness was then calculated from the linear region of the resultant force–displacement curve. Each sample was tested three times and the mean stiffness was taken.

### Histology

Histological stainings were performed in one randomly selected animal per group. On day 49, femora were excised, the femoral head was removed and the samples were placed in 4% neutrally buffered formalin for 24 h and subsequently decalcified in 12.5% EDTA for 10–14 days. The samples were embedded in paraffin and the complete fracture callus was cut in 10 µm longitudinal serial sections. Every 10th section was stained with Safranin-O: Weigert’s iron haematoxylin solution (HT1079, Sigma-Aldrich, St. Louis, MO)—4 min, 1:10 HCl-acidified 70% ethanol—10 s, tap water—5 min, 0.02% Fast Green (F7258, Sigma-Aldrich, St. Louis, MO)—3 min, 1% acetic acid—10 s, 0.1% Safranin-O (84,120, Fluka, St. Louis, MO)—5 min. Images were taken with Slide Scanner Pannoramic 250 (3D Histech, Budapest, Hungary) at 20× magnification. Sections in between were stained for Sclerostin and RANKL.

Immunohistochemical staining of Sclerostin was performed according to Wehrle et al.18. Nonspecific sites were blocked (1% BSA/PBS + 1% rabbit serum) for 60 min at room temperature. Subsequently, the sections were incubated with the primary antibody against Sclerostin (AF1589, R&D Systems, Minneapolis, MN; 1:150 in 1%BSA/PBS + 0.2% rabbit serum) overnight at 4 °C. To detect the primary antibody, a secondary biotinylated rabbit anti-goat-IgG antibody (BAF017, R&D Systems, Minneapolis, MN) was added for 1 h at room temperature. For signal amplification, the slides were incubated with avidin–biotin complex (PK-6100 Vector Laboratories, Burlingame, CA) for 30 min. Diaminobenzidine (Metal Enhanced DAB Substrate Kit, 34,065 ThermoFisher Scientific, Waltham, MA) was used as detection substrate. Counterstaining was performed with FastGreen (F7258, Sigma-Aldrich, St. Louis, MO).

For immunohistochemical staining of RANKL, slides were placed into TBS-Tween buffer and inserted into the Dako-Autostainer (Dako, Ft. Collins, USA) using the following staining protocol: incubation with primary antibody against RANKL (1:100; ab 9957, Abcam, Cambridge, UK) at 4 °C overnight, rinsing with TBS-Tween, peroxidase blocking for 10 min at room temperature, rinsing with TBS-Tween, Envision + System HPR Rabbit (Dako K4003) for 60 min at room temperature, rinsing with TBS-Tween and incubation with DAB (Dako K3468) for 10 min at room temperature. The slides were rinsed in A. dest and counterstained with FastGreen for 2 s.

For both immunohistochemical stainings, species-specific IgG was used as isotype control (Supplemetary Fig. S5). Images were taken with Slide Scanner Pannoramic 250 (3D Histech, Budapest, Hungary) at 40 × magnification.

### Statistics

Data were tested for normal distribution (Shapiro–Wilk-Test) and homogeneity of variance (Levene-Test). Depending on the test outcome, group comparisons (loading vs. control group) of data derived at single time points (VOI size) were performed by two-tailed Student’s t test or Mann–Whitney U-test (IBM SPSS Statistics Version 23). For statistical evaluation of repeated measurements two-way ANOVA with Geisser-Greenhouse correction and Bonferroni correction (GraphPad Prism 8) were performed. The level of significance was set at p < 0.05.