Circulating CD38⁺ regulatory T cells are decreased in bone fracture non-union with chronic subclinical infection

Abstract

Fracture non-union represents a complex clinical challenge resulting from an incompletely understood interplay between mechanical and biological factors. Infection frequently contributes to non-union but many cases are misdiagnosed due to a lack of classical clinical symptoms. This study characterized peripheral blood mononuclear cells (PBMCs) from aseptic non-union (NU-AS, n = 24) and fracture-related infected non-union (NU-FRI, n = 20), and compare them to healed controls (H, n = 18). High-dimensional mass cytometry (CyTOF) revealed significant elevations of regulatory T cells (Tregs; p = 0.0028) and T helper 1 (Th1) cells (p = 0.0073), and reduced expression of the activation marker CD38 in CD4⁺ T cells (p = 0.0016) and Tregs (p = 0.0017) in NU compared to H. In a subgroup analysis between NU-AS and NU-FRI, monocyte and CD38⁺ Treg cell counts provided excellent diagnostic potential, with the combination achieving a sensitivity of 100% and a specificity of 91.7%. These findings highlight an important role of the activation marker CD38 in diagnosing chronic subclinical infection, which promises earlier identification of appropriate management of these patients.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-50028-w.

Introduction

Every year, approximately 178 million bone fractures occur worldwide¹, primarily due to events such as motor vehicle accidents and falls. While most fractures heal successfully, up to 10% of long bone fractures may experience delayed healing or failure to heal, often referred to as non-union (NU)^2,3. Clinically, NU is defined as failure of a fracture to heal without additional surgical intervention, independent of the duration of previous treatment^1,4,5. Patient-dependent risk factors for NU include advanced age, smoking, use of non-steroidal anti-inflammatory drugs, and comorbidities such as various genetic disorders, diabetes, and nutritional deficiencies⁶. Patient-independent factors include the type, location and displacement of the fracture, the degree of soft tissue injury, extent of bone loss, quality of surgical treatment, and the presence of infection^6,7. Moreover, mechanical instability is a well-recognized factor contributing to NU, disrupting normal healing by impairing callus formation and remodelling^8,9.

The immune system plays an important role in fracture healing, as the initial injury triggers an early inflammatory response that recruits a range of immune cells to the injured tissue^10–12. The roles of innate immune cells, particularly neutrophils and macrophages, as early responders and key drivers of the initial inflammatory phase, are firmly established^13,14. Macrophages subsequently release growth factors that support tissue repair^15,16. In contrast, the role of adaptive immune cells is still emerging. T helper 1 (Th1) cells are critical in the early inflammatory phase, while T helper 2 (Th2) cells contribute to the later tissue repair phase^17,18. Studies have shown that elevated Th1 and T helper 17 (Th17) cells may contribute to bone healing complications, as these cells are major sources of receptor activator of nuclear factor kappa-B ligand (RANKL), promoting osteoclast activity and bone resorption, especially under inflammatory conditions^18–20. Regulatory T cells (Tregs), in contrast, control inflammation in the early phase, facilitating proliferation and differentiation of osteogenic precursor cells^20,21 and inhibiting osteoclast differentiation and bone resorption^21–23. These T cell subsets are driven by dendritic cells (DCs), which are critical for establishing immunological memory and tolerance²⁴, and therefore significantly contribute to bone healing²⁵. Early dysregulation between immune cells may negatively affect bone repair, with, for instance, patients with impaired mandibular fracture healing exhibiting a higher preoperative ratio of CD8⁺ T effector cells to CD4⁺ Tregs²⁶. During infection, activated T cells typically upregulate CD38 expression, which enhances proinflammatory cytokine production, supports cell survival, and sustains metabolic activity^27–29. Following the clearance of infection, CD38 expression typically returns back to baseline, however, in chronic infection CD38 remains elevated on PD-1⁺ exhausted CD8⁺ T cells^27,30. Dysregulated Bcell activity has also been implicated in healing complications, as patients with delayed bone healing displayed reduced regulatory B-cell activity and secretion of Interleukin 10 (IL-10)^31–33. Thus, the immune response is crucial for proper healing but is also implicated in healing complications.

Currently, a comprehensive understanding of how immune cells contribute to NU in human patients is lacking. The relative infrequency of NU, coupled with prolonged timeframes between initial fracture stabilization surgery and NU diagnosis, makes large-scale studies challenging. The goal of the present study was therefore to immunophenotype peripheral blood mononuclear cells (PBMCs) from patients with NU and compare the findings with control patients experiencing uneventful fracture healing. A primary objective was to identify differences in immune profiles of PBMCs between infected and aseptic NU, as this distinction substantially impacts treatment decisions but currently lacks adequate diagnostic modalities in patients without clinical symptoms of infection.

Results

In total, PBMCs from 18 healed (H) and 44 non-union (NU) patients underwent high-dimensional immune phenotyping by mass cytometry (CyTOF; 43-colour panel), examining innate and adaptive immune cell populations as well as activation and exhaustion markers. Thanks to group matching, no significant differences in age or gender distribution between groups were observed. However, NU patients had significantly higher BMI (p < 0.0001), and a greater prevalence of diabetes (p < 0.0001) compared to H patients, while H patients were more likely to be smokers (p = 0.018).

T cell and granulocyte clusters differ between healed and non-union patients in peripheral blood

Initial analyses considered all NU patients as a single cohort. Uniform manifold approximation and projection (UMAP) representations of all PBMCs from H (Fig. 1A) and NU patients (Fig. 1B) illustrate the most abundant cell types in each group. UMAP plots indicated greater abundance of Tregs, granulocytes, and monocytes in NU compared to H patients.

Fig. 1.

Uniform manifold approximation and projection (UMAP) of PBMCs from all Healed and Non-union patients. UMAP representation of all cells from (A) H and (B) NU patients coloured by lineage. (C) Volcano plot comparing proportions of 28 metaclusters (MC) between H and NU patients. The x-axis shows the difference in means of log2-transformed MC proportions; the y-axis shows the -log10-transformed Benjamini-Hochberg (BH)-adjusted p-value obtained using Welch’s t-test. Green points above the horizontal line indicate significantly different clusters. Downregulated clusters are marked within a grey box. (D) Significantly different clusters with corresponding cell types and markers. UMAP representation of significantly different clusters in (E) H and (F) NU patients. Corresponding heatmaps are provided in Figure S1-3.

Unsupervised clustering using PhenoGraph and edgeR identified 28 metaclusters, of which 7 showed significant differences in cell enrichment between H and NU patients (Fig. 1C). A detailed heatmap with individual marker expression is available in Figure S1-3. Clusters enriched in NU patients included granulocytic/neutrophilic populations (C04, C15, C17), CD4⁺ T cells (C18, C24), and CD8⁺ T cells (C08) (Fig. 1D). In contrast, γδ T cells (C13) were decreased in NU patients. UMAP visualizations of significantly altered clusters are shown for H (Fig. 1E) and NU patients (Fig. 1F).

Increased CD4⁺ T helper cell subsets in peripheral blood in non-union patients compared to healed patients

Because cluster analyses revealed significant differences in T cell populations, CD4⁺ T helper cell subsets were further investigated at the individual patient level (Fig. 2). NU patients had significantly higher proportions of Tregs (Fig. 2A) and Th1 cells (Fig. 2B) compared to H patients (p = 0.0028 and p = 0.0073, respectively). Conversely, T follicular helper (Tfh) cells (Fig. 2D) were reduced in NU (p = 0.0031). No significant differences were observed in Th2 cells (Fig. 2C), γδ T cells (Figure S5I) or neutrophils (Figure S7B). Regarding activation markers, CD38⁺ CD4⁺ T cells (Fig. 2E) and Tregs (Fig. 2F) were significantly decreased in NU patients (p = 0.0016 and p = 0.0017, respectively). Additional immune cell analysis can be found in the supplements (Figure S4- S8).

Fig. 2.

CD4⁺ T helper subsets and CD38⁺ cells in Healed versus Non-union patients. (A) Tregs, (B) Th1 cells, (C) Th2 cells, (D) Tfh cells, (E) CD38⁺ CD4⁺ T cells, and (F) CD38⁺ Tregs. Data shown are from individual patients (H: n = 18; NU: n = 44). Statistical analyses performed used Mann-Whitney or Welch’s tests. Gating strategy: CD45⁺CD66b^-, CD19^-CD20^-, CD3⁺, CD4⁺; Tregs: CCR4⁺, CD45RA^-CD45RO⁺, CD127^lowCD25^high; Th1 cells: CD25^-, CXCR3⁺CCR6^-; Th2 cells: CD25^-, CXCR3^-CCR6^-; Tfh cells: CD25^-, CXCR3⁺CCR6⁺.

Fracture-related infection is associated with increased neutrophils and double positive T cells in peripheral blood

The NU cohort was subsequently stratified into aseptic NU (NU-AS) and fracture-related infected NU (NU-FRI) subgroups to examine how infection affects the systemic immune responses in NU. UMAP representation of all PBMCs from NU-AS (Fig. 3A) and NU-FRI (Fig. 3B) patients showed no visually apparent differences between the groups. However, clustering analysis identified 2 metaclusters significantly elevated in NU-FRI compared to NU-AS (Fig. 3C): one cluster of neutrophils (C15) and a distinct double-positive (DP: CD4⁺CD8⁺) T cell cluster (C24; Fig. 3D). UMAP representations of these significantly changed clusters are shown for NU-AS (Fig. 3E) and NU-FRI patients (Fig. 3F).

Fig. 3.

UMAP of PBMCs from NU-AS and NU-FRI patients. All cells from (A) NU-AS and (B) NU-FRI coloured by lineage. (C) Volcano plot comparing proportions of 28 metaclusters (MC) between NU-AS and NU-FRI patients. The x-axis shows the difference in means of log2-transformed MC proportions; the y-axis shows the -log10-transformed Benjamini-Hochberg (BH)-adjusted p-value obtained using Welch’s t-test. Green points above the horizontal line indicate significantly different clusters. (D) Significantly different clusters with corresponding cell types and markers. UMAP representation of significantly different clusters in (E) NU-AS and (F) NU-FRI patients. Corresponding heatmaps are provided in Figure S1-3.

CD38⁺ Tregs of NU-FRI are decreased in peripheral blood compared to NU-AS and H patients

T helper subsets were also analysed among H, NU-AS, and NU-FRI patients (Fig. 4). Tregs (Fig. 4A) and Th2 cells (Fig. 4C) were increased in NU-AS compared to H (p = 0.0163 and p = 0.0421; respectively). Th1 cells were elevated in NU-FRI compared to H (p = 0.0022; Fig. 4B), and Th17 cells were increased in NU-AS compared to NU-FRI patients (p = 0.0015; Fig. 4D). Tfh cells were more abundant in H than NU-AS patients (p = 0.0188; Fig. 4E). CD38⁺ CD4⁺ T cells were significantly reduced in NU-AS compared to H patients (p = 0.0010; Fig. 4F), while CD38⁺ Tregs were significantly lower in NU-FRI compared to both H and NU-AS patients (p < 0.0001 for both; Fig. 4G).

Fig. 4.

CD4⁺ T helper subsets and CD38⁺ cells in H, NU-AS, and NU-FRI patients. (A) Tregs, (B) Th1 cells, (C) Th2 cells, (D) Th17 cells, and (E) Tfh cells, (F) CD38⁺ CD4⁺ T cells, and (G) CD38⁺ Tregs in Healed (H), NU-AS, and NU-FRI patients. Data shown are from individual patients (H: n = 18; NU-AS: n = 24; NU-FRI: n = 20). Statistical analyses used one-way ANOVA or Kruskal-Wallis tests. Gating strategy: CD45⁺CD66b^-, CD19^-CD20^-, CD3⁺, CD4⁺; Tregs: CCR4⁺, CD45RA^-CD45RO⁺, CD127^lowCD25^high; Th1: CD25^-, CXCR3⁺CCR6^-; Th2: CD25^-, CXCR3^-CCR6^-; Th17: CD25^-, CXCR3^-CCR6⁺; Tfh: CD25^-, CXCR3⁺CCR6⁺.

Monocytes in peripheral blood are elevated in NU-FRI compared to NU-AS and H patients

In addition to adaptive immune cells, our CyTOF panel enabled us to examine innate immune populations, given the observed differences between H and NU subgroups. Significant differences in monocyte abundance were also observed, with increased total monocytes levels in NU-FRI patients compared to H (p = 0.0232) and NU-AS (p = 0.0005) patients (Fig. 5A). No significant differences were seen in classical (Fig. 5B), non-classical (Fig. 5C), or transitional monocyte subsets (Fig. 5D).

Fig. 5.

Monocytes and subsets in H, NU-AS, and NU-FRI patients. (A) Monocytes, (B) classical monocytes, (C) non-classical, and (D) transitional monocytes in Healed (H), NU-AS, and NU-FRI patients. Data shown are from individual patients (H: n = 18; NU-AS: n = 24; NU-FRI: n = 20). Statistical analyses used one-way ANOVA or Kruskal-Wallis tests. Gating strategy: Monocytes: CD45⁺CD66b^-, CD19^-CD20^-, CD3^-CD56^-, CD14⁺; classical: CD38⁺ CD14⁺; non-classical: CD14^low, CD38^-; transitional: CD14⁺, CD38^low.

Diagnostic potential of CD38⁺ Treg and monocyte counts

Given the distinct infection-induced changes in Tregs and Monocytes, we next assessed their potential as biomarkers of immune dysfunction. Receiver Operating Characteristic (ROC) analysis assessed the diagnostic potential of the counts of CD38⁺ Treg and monocyte fraction to distinguish NU-FRI from NU-AS (Fig. 6A). The areas under the curve (AUC) were 0.965 for CD38⁺ Tregs and 0.808 for monocytes (Fig. 6B). Combined analysis showed an AUC of 0.9965 and a high Youden index of 0.92 with a sensitivity of 100%, and a specificity of 91.7%.

Fig. 6.

Receiver operating characteristic curve. (A) ROC curve for the discrimination of NU-AS and NU-FRI based on the number of monocytes and CD38⁺ Treg cells, as well as their combination; (B) overview of the area under the ROC curve, the optimal cut-off at the maximum Youden index with the corresponding sensitivity and specificity for each biomarker.

Discussion

From a clinical perspective, managing fracture NU is challenging due to its multifactorial nature and significantly different protocols dependent upon aetiology (infection or non-infectious reasons). Recent investigations have highlighted immune dysregulation as another potential factor. Impaired immune responses, possibly triggered or exacerbated by mechanical stress, may hinder bone regeneration and sustain chronic inflammation at fracture sites^8,9,34. Aside from providing insights into the biological and immunological processes underlying NU, there is also the potential for improved diagnostics by providing a more complete assessment of the systemic immune responses of patients with NU caused by infection or aseptic NU.

In this context, this is the first study to use samples from a large multicenter clinical study on NU to thoroughly investigate the underlying immune mechanisms associated with NU in humans. Initial comparisons comparing NU and H patients revealed substantial differences between these cohorts, with NU notably associated with increased Tregs. The elevated Treg levels in NU patients suggest an ongoing attempt to control inflammation and facilitate healing. In normal fracture healing, Tregs typically increase transiently, returning to baseline within one-month post-fracture³⁵. Previous studies have shown that patients with impaired mandibular fracture healing (as late as 6 months after surgery) exhibit a higher ratio of CD8⁺ T effector to CD4⁺ Treg preoperatively, indicating an early immune dysregulation negatively impacting bone repair²⁶. Our cohort lacked preoperative samples after initial trauma for such comparison; however, at NU diagnosis, we did not observe this altered ratio (Figure S4K). In NU-FRI patients, persistent Treg activation may be driven by chronic antigen exposure, whereas in NU-AS patients it likely represents a response to sustained low-grade inflammation^21,35,36. It may be worth noting that, despite chronic antigenic exposure, exhaustion markers did not increase, indicating that immune suppression is more likely driven by Tregs than by T cell exhaustion. Since the prolonged elevation of Tregs was especially pronounced in NU-AS patients, this suggests that chronic infection was not the primary factor driving Treg expansion. The source of inflammation driving T reg persistence was not linked with diabetic or other known comorbidities but may at least in part be linked with mechanical instability. However, this cannot be reliably accounted for in the analysis of our outcomes but may be more amenable to follow-up in a preclinical in vivo study where mechanics can be altered by implant fixation⁸.

Crucially, we found fewer CD38⁺ Tregs in NU patients compared to H, which sub-group analysis revealed to be driven by the NU-FRI cohort. Typically, activated T cells upregulate CD38 expression during infection, which enhances proinflammatory cytokine production and promotes cell survival and maintains metabolism^27–29. Reduced CD38 expression in T cells in chronic inflammation is associated with impaired immune function and a diminished inflammatory response^27,29. Our observation of more numerous reduced CD38⁺ and, therefore, inactive Tregs in NU-FRI patients suggests impaired functionality or failure to resolve inflammation. However, it has been shown that bone fracture leads to an increase in chemokines such as C-C motif chemokine ligand 22 (CCL22) at the fracture site^37,38, which could orchestrate the selective recruitment and spatial reorganization of CD38⁺ Tregs to the fracture zone, where they fine-tune inflammation and promote bone repair. This phenotype may hypothetically reflect excessive CCL22-driven recruitment to the fracture site, this may likely indicate that local accumulation leads to a corresponding decrease in the periphery. This hypothesis may be addressed in future by investigations into CD38 marker expression in T cells at healing fracture sites. Another contrast to typical acute immune responses was the reduced Th17 cell levels in NU-FRI compared to NU-AS patients. Th17 cells typically defend against bacterial infections by promoting inflammation and coordinating broader immune responses^39,40. Their reduction in NU-FRI might reflect again an increase at the site of infection or chronicity of infection. While Th17 cells drive pro-inflammatory responses, Tregs suppress them dependent on the cytokine environment. The relative balance between Tregs and Th17 cells is a critical determinant of the immune response. As Treg-associated cytokines (IL-10, TGF-β) inhibit IL-17, which is needed for Th17 differentiation^21,41, reduced Th17 cell levels in NU-FRI compared to NU-AS patients could also be influenced by a higher number of (activated) Tregs. Measuring local tissue cytokine levels may aid in deciphering the precise local balance between Th17 and Tregs in NU. The comparison between NU-FRI to NU-AS patients showed a cluster of CD4⁺CD8⁺ T cells significantly elevated in NU-FRI compared to NU-AS. Double positive CD4⁺CD8⁺ T cells are often pro-inflammatory and show a T helper like phenotype, although they can also acquire cytotoxic functions. In chronic viral infections, rheumatoid arthritis, and graft-versus-host disease, these cells are associated with increased pro-inflammatory responses and heightened tissue damage^42,43. However, to date, the specific contribution of double-positive (DP) CD4⁺CD8⁺ T cells to bone infection and repair remains unexplored, as no studies have directly examined their functional role in fracture healing.

The prolonged elevation of Th1 cells observed in our NU cohort, especially in NU-FRI patients, indicates an ongoing inflammatory response driven by Th1-type immunity. Th1 cells are known to secrete proinflammatory cytokines which promotes cellular immune responses and contributes to tissue inflammation. However, during fracture healing, this prolonged Th1 dominance may contribute to a dysregulated immune environment that impairs tissue regeneration. By maintaining an inflammatory milieu, elevated Th1 activity may inhibit the resolution of inflammation and interfere with bone repair^44,45. These findings support the notion that immune imbalance, rather than simple infection or mechanical factors alone, plays a pivotal role in the pathophysiology of non-union. A further finding from the sub-division of NU patients into NU-FRI and NU-AS was a significant increase in monocytes in NU-FRI patients compared to NU-AS and H. Similar to Tregs, monocytes are key players in normal bone regeneration. Monocyte levels typically peak within the first few days following fracture and return to baseline within 2–4 weeks, depending on the progression and resolution of the remodelling phase^46–48. Persistently elevated monocyte levels, therefore, reflect abnormal inflammatory persistence. Our observation suggests chronic infection-driven inflammation in NU-FRI patients, potentially impeding the transition to bone formation. Thus, immune responses underlying NU differ substantially between NU-FRI (monocyte-driven) and NU-AS (Treg-mediated).

Clinically, distinguishing NU-FRI from NU-AS is essential, as their management significantly differs. NU-FRI management requires a staged surgical procedure, including surgical removal of the implant and eradication of the infection through resection and debridement combined with a systemic antibiotic therapy^1,2. Only after the successful eradication of infection subsequent treatment of the non-union can be undertaken⁴⁹. NU-AS, on the other hand, is typically treated in a single-stage procedure using reamed intramedullary exchange nailing^50,51. ROC curve analysis of CD38⁺ Treg counts demonstrated strong potential for differentiating NU-FRI from NU-AS. High specificity and sensitivity, along with a high Youden index, suggest that CD38⁺ Treg may be a potential biomarker for distinguishing infection-driven NU from aseptic cases. The discriminatory ability was even higher when CD38⁺ Treg counts were combined with monocyte counts. Currently, FRI diagnosis relies on both suggestive and confirmatory criteria, including preoperative parameters such as the patient’s clinical presentation, serum inflammatory markers, and radiological imaging, as well as intraoperative microbiological and histopathological findings⁵². However, a reliable preoperative diagnosis of FRI is not possible using these criteria and even the intraoperative diagnostics do not provide satisfactory sensitivity⁵³. Thus, immune profiling of monocytes and CD38⁺ Tregs could significantly improve preoperative non-union diagnostics, enabling early targeted therapeutic interventions. Future, mechanistic studies would be of interest to determine of these cells suppress or activate inflammation, produce cytokines, or regulate repair. Functional assays would provide critical additional insights into T cell functionality, enabling additional understanding of the underlying biology as an adjunct to the diagnostic potential. Interestingly, CD38⁺ T cells are already used clinically as a biomarker in oncology, with their increased numbers being associated with disease progression in multiple myeloma patients⁵⁴, and similar testing facilities are presumably possible for routine diagnosis of NU-FRI. Overall, these results identify a promising preoperative biomarker for distinguishing infection-driven from aseptic NU, though validation in larger, independent cohorts is needed.

While our study provides a range of valuable insights into the immune landscape of NU patients that have not been previously described, several limitations should be acknowledged. The cohort size (62 patients) limits generalizability. Additionally, samples were collected at a single time point before surgical revision, and earlier or longitudinal sampling might offer greater insights. Given the logistical difficulties, acquiring larger cohorts or longitudinal samples remains challenging. It would also be advantageous to compare local immune profiles at the site of NU, which will likely contrast with the findings in the peripheral blood. PBMCs do offer a window into the involvement of immune cells in NU, however, local tissue samples would also be desirable. Of course, matched bone samples from healthy healed patients may not be easy to acquire, leading to lack of important control tissues in such circumstances. As open fractures are a recognized risk factor for the development of FRI⁵⁵, it is not surprising that this association is also observed in FRI occurring in late-stage non-union. Consequently, our study showed also that the healed patient group displayed a significant lower percentage of open fractures and a lower percentage of plate fixations. However, because of the small sample size, a subgroup analysis based on implant used or bone affected was not possible. Furthermore, our study primarily focused on immune phenotyping rather than functional assays such as cytokine production or proliferation tests, which might provide further mechanistic insights. We also observed discrepancies between UMAP and statistical analysis, this may mask differences within specific cell clusters, such as the γδ T cell cluster, due to aggregating multiple distinct subsets.

Conclusions

Despite limitations, this study significantly advances understanding of immune mechanisms underlying fracture NU and identified potentially clinically relevant diagnostic biomarkers. This represents the first application of high-dimensional mass cytometry (CyTOF) to NU patients, enabling comprehensive immune profiling, far surpassing conventional flow cytometry approaches in depth and detail. Future research should explore mechanistic roles of Th1, Th2, and Treg cells in fracture healing and NU, alongside mechanisms sustaining monocyte elevation and CD38 expression over time in NU-FRI.

NU patients exhibit distinct T helper subsets compared to healed patients. While differentiating FRI from aseptic NU clinically remains challenging due to subtle clinical signs, immune characterization provides new insights into the nuanced immune landscape of NU fractures. Importantly, our study identified distinct immune biomarkers with clinical potential to differentiate NU-FRI from NU-AS, potentially aiding future diagnostics and patient management.

Methods

Patient cohort

This prospective multicenter clinical study characterised immune cell phenotypes in the peripheral blood from patients with aseptic non-union (NU-AS) and fracture-related infected non-union (NU-FRI), comparing them to patients with complication-free fracture healing outcomes (H). More than one hundred patients were enrolled at eight level-I trauma centers in Germany between January 2019 and April 2022⁵⁶. After matching for gender and age between the groups, 62 patients from whom at least 5 × 10⁶ PBMCs could be isolated from preoperative peripheral blood samples were included in this analysis (Fig. 7). The study was approved by the Ethics Committee of the Institutional and National Medical Board (Bavarian State Chamber of Physicians, ID 2016–16041, initial vote) and registered in the German Clinical Trials Register and WHO International Clinical Trial Registry Platform (DRKS00014657, https://trialsearch.who.int/Trial2.aspx?TrialID=DRKS00014657, first date of registration 26/04/2018). Each of the 8 study centers obtained local ethical approval (secondary votes: BG Klinikum Duisburg: Ethik-Kommission der Ärztekammer Nordrhein (approval number: 2018285), BG Unfallklinik Frankfurt am Main: Ethik-Kommission bei der Landesärztekammer Hessen (approval number: MC 112/2018), BG Unfallklinik Ludwigshafen: Ethik-Kommission der Landesärztekammer Rheinland-Pfalz (approval number: 2018-13316-Klinische Forschung), BG Klinik Tübingen: Ethik-Kommission an der Medizinischen Fakultät der Eberhard-Karls-Universität und am Universitätsklinikum Tübingen (approval number: 722/2016BO2), BG Klinikum Bergmannstrost Halle: Ethikkommission der Ärztekammer Sachsen-Anhalt (approval number: 33/18) and BG Klinikum Hamburg Ethik-Kommission der Ärztekammer Hamburg (approval number: MC-184/18), BG Klinikum Unfallkrankenhaus Berlin did not require a secondary vote from its local ethics committee, the initial vote of Ethik-Kommission der Bayerischen Landesärztekammer (approval number: 16041) was sufficient) and all patients provided written informed consent before inclusion.

Fig. 7.

Group assignment, and clinical data. (A) NU patients sub-divided into NU-AS and NU-FRI. (B) Summary of clinical data of included patients.

Peripheral heparinized blood (Plus Blood Collection Tubes, BD Vacutainer, Plymouth, UK) was collected preoperatively from patients with femoral or tibial shaft non-union requiring revision surgery (index surgery), as well as from patients with healed femoral or tibial fractures undergoing implant removal, who served as controls. Non-union was diagnosed based on clinical symptoms, clinical evaluation, and radiographic assessment. Radiographic criteria for non-union included the absence of a bony bridge in at least three out of four cortices on anteroposterior and lateral views of conventional radiographs⁵⁷. Consequently, non-union was defined as a fracture failing to heal without additional surgical intervention⁵. In the healed group, implants were removed after complication-free fracture consolidation, without clinical or laboratory signs of infection or delayed healing. Exclusion criteria, previously described by Trenkwalder et al., 2025⁵⁶, included long-term antibiotic use before surgery, ongoing antimicrobial treatment, pregnancy, acute infection of other origins, simultaneous non-union at different sites, immunosuppression, autoimmune disease, or malignancy.

NU-FRI was diagnosed according to the diagnostic criteria for FRI^1,58, including the presence of a fistula, identification of the same pathogen in two separate tissue cultures, detection of the same pathogen in both tissue culture and sonication fluid, or identification of microorganisms in a single tissue culture confirmed by histopathology. Cases with only one positive tissue or sonication culture were classified as NU-AS⁵⁶. Histopathological evidence alone (Histopathological Osteomyelitis Evaluation Score; HOES) was insufficient to confirm infection, as previously described⁵⁹. NU diagnosis (AS or FRI) was based on microbiological and histopathological findings from the index surgery and any subsequent surgeries within 12 months of inclusion. On average, 4.5 (IQR 3–6) samples for microbiological analysis (including tissue cultures and sonication, if available) and 1 (IQR 1–2) sample for histopathology were collected during revision surgery for infection diagnostics.

In total, 18 healed (H) patients and 44 NU patients were included (Fig. 7A). NU patients were divided into NU-AS (n = 24) and NU-FRI (n = 20). Clinical data including sex, age, body mass index (BMI), diabetes, and smoking status as well as injury and fixation details (polytrauma, affected bone, implant used and injury type) are summarized in Fig. 7B. NU patients had more open fractures (52%) compared to the healed patients (16.6%), however, this was due to the high rate of open fractures in FRI patients (50%), while AS patients had an open fracture rate (17.2%) similar to the healed patients. Plating fixation was lowest in the healed cohort (5.6%), while it was elevated in both the NU-AS (33.3%) and NU-FRI (45%) cohort.

A chi-square test showed that the healed patient group displayed a significant lower percentage of open fractures and a lower percentage of plate fixations (p = 0.375, p = 0.0331, respectively). It is recognised that open fractures are a risk factor for developing FRI⁵⁵. And this is also observed in the FRI associated with later stage non-union. To add is that non-union is diagnosed after 6 to 9 months and the revision is considered on a case-by-case basis as needed. Not all patients are in the early post-fracture phase, as all were diagnosed at least 94 days after fracture stabilization. NU patients ranged from 94 days to 1334 days, NU-AS from 162 days to 1254 days and NU-FRI from 94 days to 1334 days. The non-union revision was, on average, 93 days earlier in NU-AS patients than in NU-FRI patients, but the difference was not significant.

Isolation of PBMCs

PBMCs were isolated from peripheral blood by density gradient centrifugation. Heparinized blood was diluted with Dulbecco’s phosphate-buffered saline (DPBS; Bio&SELL GmbH, Germany). The blood-DPBS mixture was layered over BioColl separation medium (Bio&SELL GmbH, Germany) in separation tubes (Kisker Biotech GmbH & Co. KG., Germany). After centrifugation (800 × g, 20 min, room temperature (RT)), the PBMC layer was washed first with DPBS (500 × g, 10 min, RT), then with complete Roswell Park Memorial Institute Medium (cRPMI: RPMI supplemented with 10% fetal calf serum (FCS); 300 × g, 10 min, RT). PBMC pellets were resuspended in cRPMI, and viable cell counts were determined using trypan blue (0.4%, Sigma-Aldrich, USA) and a Neubauer improved counting chamber (Karl Hecht GmbH & Co KG, Germany).

Cells were cryopreserved in cold freezing medium (20% DMSO/FCS) at 5–35 × 10⁶ cells/mL, frozen overnight at -80 °C in Mr. Frosty freezing containers (Thermo Fisher Scientific, USA) containing 2-propanol (AppliChem GmbH, Germany), and stored in liquid nitrogen until use.

Antibody labelling and titration

Antibodies were conjugated with lanthanide and cadmium metals using Maxpar Antibody Labeling Kits (Fluidigm, USA) according to the manufacturer’s instructions. Cadmium-labelled antibodies were eluted in W-buffer (Fluidigm) and HRP-Protector peroxidase stabilizer; lanthanide-labelled antibodies were eluted in W-buffer (Fluidigm) and antibody stabilizer buffer (Candor Bioscience, Germany) supplemented with 0.05% sodium azide. The final antibody volume was adjusted by adding an equal volume of antibody stabilizer buffer (Fluidigm; supplemented with 0.05% sodium azide or HRP-Protector). Antibodies were stored at 4 °C. All antibodies were titrated on healthy donor PBMCs to determine optimal staining concentrations (Table S1).

Antibody staining

Cryopreserved PBMCs were thawed and stained in batches using an immune profiling kit (Standard BioTools, USA) supplemented with additional conjugated antibodies (Table S1). PBMCs were incubated with Human TruStain FcX (BioLegend, USA) for 10 min at RT to block Fc receptors (FcR). After centrifugation (300 × g, 5 min), cells were stained with metal-conjugated antibodies for 30 min at RT. After staining, cells were washed twice with Maxpar^® Cell Staining Buffer (CSB; Fluidigm; 300 × g, 5 min), fixed with 1.6% formaldehyde solution (Thermo Fisher Scientific) for 10 min at RT, and centrifuged (800 × g, 5 min). Cells were then incubated in 250 µM Cell-ID™ Intercalator-Ir (Fluidigm) in Maxpar Fix and Perm Buffer (Fluidigm) overnight at 4 °C. The next day, cells were centrifuged (800 × g, 5 min), resuspended in 100 µL residual volume, and frozen at -80 °C until acquisition.

Data acquisition

Cells were washed twice with CSB and cell acquisition solution (CAS; Fluidigm), filtered through a 35 μm cell strainer, and mixed with EQ Four Element Calibration Beads (Fluidigm). Acquisition was performed on a Helios™ mass cytometer (Fluidigm) at 250–300 events/second. Data normalization was done using EQ beads. Samples were acquired across eight days, mixing patient groups to minimize batch effects.

Data processing and analysis

Data were analysed using FlowJo V10 (Treestar, USA) and GraphPad Prism 10 (GraphPad Software Inc., USA). Gating strategy can be found in the supplements (Figure S9-S13). Statistical analyses included Mann-Whitney U test or Welch’s tests (H vs. NU), and one-way ANOVA with Tukey’s multiple comparisons or Kruskal-Wallis tests (H vs. NU-AS vs. NU-FRI), depending on normality distribution testing. Significance was set at p < 0.05.

Normalized data was uploaded to OMIQ (OMIQ, USA) and an arcsinh transformation with a coefficient of 5 was applied. A standardized cleanup strategy, as described by Bagwell et al.., 2020⁶⁰, was applied. Further analysis was performed on the transformed data. Data consistency was evaluated over time; non-conforming sections were gated out using flowCut. Viable CD45⁺ single cells were selected by manual gating. For lineage population analysis, total live CD45⁺ single cells were down-sampled to 2,000 events per subject. Unsupervised analysis employed uniform manifold approximation and projection (UMAP; neighbours = 15, minimum distance = 0.4, components = 2; metric=Euclidean, embedding initialization=spectral), followed by PhenoGraph clustering (k = 100). Significant cluster differences were identified using EdgeR (OMIQ) or visualized as heatmaps with Euclidean clustering. Metacluster proportions were analysed by Welch’s t-test with Benjamini-Hochberg (BH)-adjusted p-values.

Immunological parameters distinguishing NU-FRI from NU-AS were evaluated using receiver operating characteristic (ROC) curve analysis in SPSS (ver. 26; IBM, USA). Optimal cut-off values, sensitivity, and specificity were determined using the maximum Youden index. A ROC curve combining two predictors was generated based on binary logistic regression predictive probabilities.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

P.F. : Data collection, data curation, data visualization, manuscript writing, manuscript revision; C.S., TF.M: supervision; K.T. and S.E.: PBMC isolation; P.F., C.S., K.T., S.E., S.H., F.W., G.M., CA.A., M.R., JJ.GV., G.M. , SAJ.Z., TF.M., EC.dJ: manuscript revision; K.T. and S.E; S.H.: Conceptualization and implementation of the multicenter study ; F.W. : Recruitment of patients, patient survey collection, G.M. , SA.JZ., EC.dJ: Data interpretation, CA.A., M.R.: Input on data Interpretation, J.J.GV.: Supervision for data analysis and data visualization; S.H., SAJ.Z., EC.dJ., TF.M.: funding.

Funding

The study was supported by the German Social Accident Insurance, DGUV grant number FF-FR 0276 and AO Trauma grant number AR_2021_08.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki. The study was approved by the Ethics Committee of the Institutional and National Medical Board (Bavarian State Chamber of Physicians, ID 2016–16041). Each study center received an ethics vote from its local responsible ethics committee. All patients provided their written consent before study inclusion.

Competing interests

The authors declare no competing interests.