S. aureus ATCC 6538 and P. aeruginosa ATCC 27853 biofilms were grown on 1.75 cm × 1.75 cm × 0.85 cm grade 5 titanium simulated fracture fixation plates using a modified CDC biofilm reactor as previously described (Kay et al., 2022; Williams et al., 2025).

A total of 24 male (wether) and female Rambouillet sheep aged 1–3 years and weighing ∼ 50–70 kg underwent surgery. Surgeries were performed with approval and oversight from the Institutional Animal Care and Use Committee (IACUC) at the University of Utah and from the United States Army Medical Research and Development Command's Animal Care and Use Review Office (ACURO). Each sheep was treated with their first dose of rifampin and cefazolin or cefepime, depending on the group (Table 1 “Animal treatment groups”), 30–120 min before surgery.

After the sheep was anesthetized, the surgical limb (right hind) was draped with a protective cloth, hearing protection was inserted into the animal's ear, and the limb was exposed to a pressurized air blast as described previously to simulate soft tissue trauma that is characteristic of fracture injuries. The blast was performed using a Martin Engineering Tornado Air Cannon (Williams et al., 2025, 2019). The sheep was transferred to the operating table, and the right hind limb underwent a presurgical scrub.

The surgical procedure was performed as described (Williams et al., 2025, 2012b). A representative image can be seen in Fig. 1. Briefly, an incision was made in the anterior region of the proximal medial aspect of the right hind tibia, the skin and associated subcutaneous tissue were separated from the bone surface, simulated fractures were created on the bone surface, and biofilm-ridden simulated fracture fixation plates were secured to the bone using cortical bone screws.

Animals in the study were randomly organized into one of four treatment groups (Table 1). All sheep received the same surgical procedure.

The veterinary and research team members remained with the animals after each procedure until they were standing, weight-bearing, and eating. Animals were monitored at least twice daily for signs of pain or distress throughout the monitoring period.

Fluorochrome labelling techniques have previously been used in ovine models to demonstrate bone formation dynamics (Williams et al., 2025). Animals received an IV injection of calcein green (9.9 mg kg⁻¹) via the jugular vein or a forelimb catheter, 5 and 16 d before sacrifice, per published protocols (Epperson et al., 2021; Williams et al., 2012a, b).

All animals were euthanized 21 d post-surgery with an overdose IV injection of a pentobarbital-based solution (1 mL/4.5 kg). A veterinary technician or trained lab member verified euthanasia with a vital sign check, i.e. heartbeat, eye reflex, and breathing. The animal's right hind (surgical) limb was disarticulated at the coxofemoral joint by excising soft tissue with a scalpel.

All elements of the microbiological sample collection were performed by a lab member donning sterile gloves using an aseptic technique as previously described (Williams et al., 2025). Samples collected included a tissue section directly above the distal fixation plate, the distal simulated fracture fixation plate and associated screws, a bone section in the centre of the area under the distal plate, the bone surrounding each screw track, and the bone marrow directly under the section of drilled bone. Photographs were taken before and throughout the necropsy (Fig. 2).

All samples were placed into pre-weighed conical tubes. Tissue samples were transferred to a small Ninja blender and homogenized without overheating. The conical tubes were vortexed for 1 min, sonicated at >40 kHz for 10 min, and vortexed again for 1 min. 100 µL of the sample mixture was pipetted from the conical tube and immediately plated on a TSA plate as a “0” dilution. Samples were quantified thereafter using a 10-fold dilution series. 50 µL of the final dilution tube was plated on TSA; the 50 µL was dispersed in 5 × 10 µL spots. Plates were incubated at 37 °C for 24–48 h, and colonies were counted to calculate CFU per gram or CFU per sample.

Soft tissues surrounding the proximal simulated fracture fixation plate were dissected away, leaving ∼ 1 cm of soft tissue around the plate. An oscillating bone saw was used to cut the bone down to a smaller section, and the sample was fixed in 10 % neutral buffered formalin with three changes of solution every 48 h. The sample was placed in 70 % ethanol until further processing. μ-CT images were captured and used to create three-dimensional (3D) reconstructions of the bone surface directly underneath the proximal simulated fracture fixation plate (90 kV, 88 µA, 36 mm field of view, high-resolution scan for 4 min with a Cu 0.1 mm X-ray filter). Before imaging, the two distal screws were removed from the proximal plate to reduce artefacts. The topographical analysis was completed by selecting a 10 × 10 × 1.5 mm region of interest (ROI) centred on the bone under the proximal fracture fixation plate. Additionally, two smaller ROIs (4 × 4 × 1.5 mm) adjacent to the osteotomy were analysed to quantify the percentage of bone post-treatment or post-infection.

After μ-CT analysis, the sample was dehydrated in increasing grades of ethanol and xylene using a Tissue-Tek VIP (Miles Scientific). The sample was then embedded in PMMA and cut down with a bandsaw, sectioned, and ground to an optical finish using protocols previously described (Williams et al., 2025, 2012a, b). The bone under the simulated fracture fixation plate was analysed using backscattered electron imaging to examine the bone response and calculate the percentage of woven bone. The sections were adhered to plastic slides, then they were further ground and polished to ∼ 75 µm for fluorescence microscopy to visualize the calcein green label and determine the percentage of reactive bone. Sections were lastly stained with Sanderson's rapid bone stain for qualitative light microscopy analysis. The number of osteoclasts in each section was counted, as osteoclast count correlates with infection severity (Croes et al., 2019).

Statistical analysis and processing were performed using the Stata Statistical Software: Release 18 (StataCorp LLC, College Station, TX, USA). Error bars represent ± standard error unless otherwise noted.

Microbiological analysis compared the mean log₁₀ reductions in CFUs (normalized to sample or by weight) between the treatment groups. The control plate CFUs were averaged for each animal, and a log₁₀ transformation was applied to provide a baseline inoculum. Recovered samples that did not grow any CFUs were reported at the limit of detection (LOD) by artificially adding one colony to the least diluted plate. The normalized CFUs for each sample were summed for each animal, and the log₁₀ transformation was applied to determine the total recovered bioburden. The mean log₁₀ reduction was calculated as the difference between the log₁₀-transformed control plate doubled CFUs (to account for two inocula plates) and the total sample CFUs. Positive log₁₀ reductions indicated bacterial bioburden was reduced from the initial inoculum. Group means were tabulated and compared using a two-sample t test. A p value < 0.05 was considered statistically significant. Comparisons between individual sample bioburden and inocula were similarly compared with two-sample t tests, where a p value < 0.05 was considered statistically significant.

Reactive (fluorescent microscopy) and woven bone (backscattered electron imaging) analyses were also compared using two-sample t tests. Group means were tabulated and compared, with a p value < 0.05 used to indicate statistical significance. Osteoclast and μ-CT analyses used multiple locations, which introduced data clustering. A mixed-effect linear regression (multilevel model) accounted for nested locations within each sheep. The experimental condition (group) was a fixed effect, and the sheep was a random effect. Analyses were based on the t statistic due to the experimental sample size.

Robust biofilms of 8.68 ± 0.88 log₁₀ CFU for S. aureus and 9.05 ± 0.31 log₁₀ CFU for P. aeruginosa reliably grew on inocula fixation plates. Representative scanning electron microscopy (SEM) images of biofilms on simulated fracture fixation plates were published previously (Kay et al., 2022). Outcomes suggested that all animals were inoculated with mature biofilms at the time of surgery.

Microbiological outcomes and subdermal tissue observations at necropsy suggested that, despite suture lines showing signs of healing, all positive control of infection animals, and all those treated with antibiotics, had infected sites. There was a statistically significant difference in initial S. aureus inocula values between positive control of infection animals and those treated with systemic antibiotics (p=0.0126; Table 2), yet animals were inoculated with >8log⁡10 CFU, which was well above an infectious dose (Fig. 3).

S. aureus-inoculated animals treated with cefazolin and rifampin (Group 3) did not have a significant difference in total bioburden compared to positive controls (p=0.9904; Table 2 and Fig. 4). The antibiotic therapy was less effective at reducing S. aureus bioburden in avascular or more inaccessible spaces, such as the screw threads and hard tissue (bone and screw tracks). In these particular areas, therapy failed to achieve statistical significance compared to the positive control (p=0.0650 and p=0.0502; Table 2 and Fig. 3). Conversely, antibiotic therapy significantly outperformed the positive controls in more superficial locations, including the fixation plate and soft tissue (p=0.0132 and p=0.0002, respectively; Table 2 and Fig. 3).

Initial inocula baselines in animals inoculated with P. aeruginosa that served as positive controls or that received cefepime and rifampin had similar bioburden (p=0.1588; Table 2 and Fig. 3). When pooled, P. aeruginosa bioburden was reduced significantly more in animals treated with antibiotics than in positive controls (p=0.0026; Table 2 and Fig. 4). However, similarly to the S. aureus outcomes, systemic therapy was limited in less accessible regions, such as hard tissue and screws; bioburden levels remained similar to the positive controls (p=0.2331 and p=0.8439; Table 2 and Fig. 3). Additionally, bioburden levels were nearly identical on the fixation plate between the positive control and antibiotic therapy groups (p=0.9773; Table 2 and Fig. 3). Onboarding antibiotics demonstrated efficacy against P. aeruginosa in the soft tissue, where bacterial counts were nearly 3log⁡10 lower than in positive controls (p=0.0022; Table 2 and Fig. 3). Despite this, cefepime and rifampin failed to reduce bacterial bioburden in hard tissue below the 10⁵ CFU g⁻¹ of tissue infection benchmark (6.20 log₁₀ units).

Two sheep groups that were worked on in a previously published study were included in the datasets of the sections below for comparison (Williams et al., 2025). The groups included animals sacrificed at time = 0 (immediately after surgery; Group 5) and a negative control group that received no bacteria (21 d endpoint; Group 6). These animals received the same surgery and histological processing described in the methods of this study but without the introduction of bacteria or antibiotic therapy. The histological data for each group are reproduced in Figs. 6–9 for both S. aureus and P. aeruginosa to allow visual and statistical comparison. A representative histological image can be seen in Fig. 5.

μ-CT was used to assess the percentage of remaining bone volume beneath the proximal fracture fixation plate (Fig. 6). Previously published data showed that animals displayed a high bone volume (96.1 % ± 2.6 %) at baseline (time = 0; Group 5), which was largely preserved in the negative control group after 21 d (90.7 % ± 4.0 %; Group 6) (Williams et al., 2025). The slightly lower bone volume in negative controls compared to baseline was attributable to the periosteum removal during surgery and natural bone porosity.

S. aureus infection significantly reduced bone volume. Specifically, positive control animals (Group 1) retained only 38.5 % ± 6.7 % of the bone beneath the proximal fixation plate, a statistically significant reduction compared to negative controls (p=0.0000; Table 3: 1 vs. 6). This confirmed that infection successfully developed in positive controls. However, onboarding cefazolin and rifampin mitigated bone loss; Group 3 animals retained 74.9 % ± 4.2 % of bone. This was statistically significant compared to positive controls (p=0.0020; Table 3: 1 vs. 3) and approached, but did not quite reach, statistical significance with negative controls (p=0.0570; Table 3: 3 vs. 6).

Similarly, P. aeruginosa infection led to significant bone loss. Positive control animals (Group 2) retained only 49.5 % ± 12.1 % of bone, a statistically significant reduction compared to the negative control (p=0.0010; Table 4: 2 vs. 6). However, unlike its efficacy against S. aureus, antibiotic therapy was far less effective against P. aeruginosa. Animals treated with cefepime and rifampin (Group 4) retained just 45.9 % ± 6.8 % of bone, which remained significantly lower than the negative control (p=0.0000; Table 4: 4 vs. 6) and was not significantly different from positive controls (p=0.7480; Table 4: 2 vs. 4).

SEM was used to assess the percentage of woven bone as an indicator of new, healthy bone remodelling (Fig. 7). Previously published data showed that at baseline, no woven, i.e. healing, bone was observed (Williams et al., 2025). This was anticipated, since no time had elapsed for a healing response to occur. In negative control animals, substantial new bone formation (38.1 % ± 6.2 %) was observed in surgically impacted areas, indicating a natural healing response to the surgical procedure, which included periosteal stripping, simulated fracture placement, and fixation plate attachment. This suggested that the surgical model did not impair the animal's natural bone healing capacity.

In contrast, healing was severely attenuated in positive control animals (groups 1 and 3). Positive control animals inoculated with S. aureus (Group 1) exhibited significantly less woven bone (1.9 % ± 1.1 %) and increased bone resorption, consistent with infection development. This decrease was statistically significant versus the negative control (p=0.0025; Table 3: 1 vs. 6). A similar response was observed in animals inoculated with P. aeruginosa, where new woven bone formation (2.1 % ± 1.2 %) was also significantly less compared to the negative control (p=0.0024; Table 4: 2 vs. 6).

Systemic therapy moderately improved woven bone formation. Animals treated with cefazolin and rifampin had significantly more new woven bone growth (16.7 % ± 3.1 %) compared to the positive control animals (p=0.0096; Table 3: 1 vs. 3), though significantly less than negative controls (p=0.0079; Table 3: 3 vs. 6). A similar trend was observed in animals with P. aeruginosa infection: more woven bone formation was observed when cefepime and rifampin (Group 4) were onboarded (25.8 % ± 7.6 %) than in positive controls. However, due to increased variability in the response, bone formation was not statistically different from the negative or positive controls (p=0.2321 and p=0.0570; Table 4: 4 vs. 6 and 2 vs. 4).

Fluorescent labelling was used to quantify reactive bone growth as an indicator of unhealthy bone formation (Fig. 8). Animals at baseline (time = 0) had minimal reactive bone under the proximal fracture fixation plate (0.85 %). The negative control animals that only underwent the surgical procedure showed a mild increase in reactive bone (7.5 % ± 2.3 %) (Williams et al., 2025).

Infection with S. aureus elevated reactive bone formation. Positive controls (Group 1) displayed a modest increase in bone formation (8.8 % ± 4.6 %) that was not statistically significant versus negative control animals (p=0.7867; Table 3: 1 vs. 6). Surprisingly, cefazolin and rifampin therapy (Group 3) exacerbated the response (17.3 % ± 6.1 %), though it remained statistically insignificant versus negative controls (p=0.1538; Table 3: 3 vs. 6).

A similar trend was seen in P. aeruginosa-infected animals. Positive control animals exhibited a slight increase in reactive bone (11.0 % ± 5.3 %), which was elevated by cefepime and rifampin (12.7 % ± 3.5 %). Like with S. aureus, neither displayed significance versus the negative control animals (p=0.4876 and p=0.2310; Table 4: 2 vs. 6 and 4 vs. 6).

Light microscopy was used to enumerate osteoclast count (osteoclasts mm⁻²), a known indicator of infection severity (Fig. 9) (Croes et al., 2019). Negligible osteoclast activity was observed at baseline (0.1 mm⁻²) and remained low in negative control animals (0.3 ± 0.1 mm⁻²).

Infection significantly increased osteoclast activity. In positive control animals, osteoclast counts rose sharply in both S. aureus (22.9 ± 3.8 mm⁻²) and P. aeruginosa (21.9 ± 5.0 mm⁻²) infections versus negative controls (p=0.0000 and p=0.0000; Tables 3 and 4: 1 vs. 6 and 2 vs. 6). Systemic therapy had varying effects. S. aureus animals treated with cefazolin and rifampin (9.9 ± 1.7 mm⁻²) demonstrated significantly reduced osteoclast activity compared to positive controls (p=0.0010; Table 3: 1 vs. 3) but remained elevated relative to negative controls (p=0.0030; Table 3: 3 vs. 6). However, in P. aeruginosa-infected animals (15.5 ± 2.9 mm⁻²), cefepime and rifampin treatment did not significantly reduce osteoclast counts versus positive controls (p=0.1150; Table 4: 2 vs. 4) and levels remained higher than in negative control animals (p=0.0000; Table 4: 4 vs. 6).