Prognostic and therapeutic potential of microRNAs for fracture healing processes and non‐union fractures: A systematic review

Abstract Background Approximately 10% of all bone fractures result in delayed fracture healing or non‐union; thus, the identification of biomarkers and prognostic factors is of great clinical interest. MicroRNAs (miRNAs) are known to be involved in the regulation of the bone healing process and may serve as functional markers for fracture healing. Aims and methods This systematic review aimed to identify common miRNAs involved in fracture healing or non‐union fractures using a qualitative approach. A systematic literature search was performed with the keywords ‘miRNA and fracture healing’ and ‘miRNA and non‐union fracture’. Any original article investigating miRNAs in fracture healing or non‐union fractures was screened. Eventually, 82 studies were included in the qualitative analysis for ‘miRNA and fracture healing’, while 19 were selected for the ‘miRNA and fracture non‐union’ category. Results and conclusions Out of 151 miRNAs, miR‐21, miR‐140 and miR‐214 were the most investigated miRNAs in fracture healing in general. miR‐31‐5p, miR‐221 and miR‐451‐5p were identified to be regulated specifically in non‐union fractures. Large heterogeneity was detected between studies investigating the role of miRNAs in fracture healing or non‐union in terms of patient population, sample types and models used. Nonetheless, our approach identified some miRNAs with the potential to serve as biomarkers for non‐union fractures, including miR‐31‐5p, miR‐221 and miR‐451‐5p. We provide a discussion of involved pathways and suggest on alignment of future research in the field.


BACKGROUND
Despite surgical and treatment improvements for traumarelated diseases, approximately 10% of fractures do not heal fully. 1 The main traditional risk factors related to nonunion or delayed fracture healing are older age, female sex, smoking, diabetes mellitus (DM) and obesity. 2 However, the accuracy of predictions based on these factors remains poor and may not be used to guide early interventions to prevent non-unions. MicroRNAs (miRNAs) are small, noncoding RNAs involved in the regulation of gene expression pathways 3 by driving messenger RNA (mRNA) degradation and translational repression, influencing pivotal cellular processes, such as cell proliferation, differentiation, apoptosis and cell migration. [4][5][6] Recently, miRNAs have been discussed as promising predictive markers since they are indicative of cellular processes and can be assessed non-invasively in blood and other body fluids in the form of a 'liquid biopsy'. 3,4,7 For example, miRNAs are already investigated to be used as biomarkers for cancer 8 and are involved in maintaining vasculo-protective functions. 9 As they are functional molecules, they also hold great promise in theranostic approaches. Until now, these miRNAs are still in the early stages of investigation and have to be validated before translation into clinical practice.

The process of fracture healing
For bone healing processes, two different ossification mechanisms may take place: intramembranous and endochondral ossification. 10 During intramembranous ossification, bone regenerates directly by differentiation of mesenchymal stromal cells (MSCs) into osteoblasts. Intramembranous ossification occurs within a few days at the periosteal sites characterised by low strain and hydrostatic pressure 11 at distal edges of the fracture site and leads to a hard callus formation. 12 Bridging across the central fracture gap provides initial stabilisation, leading to first biomechanical functions. 13 Subsequent differentiation of the MSCs into end-stage osteoblasts leads to new bone formation. 14 In contrast, endochondral ossification is a bone regeneration process in which bone heals indirectly through the formation of a cartilage intermediate. 10,15 Endochondral ossification occurs primarily in long bones such as femur, tibia or humerus, which are not rigidly fixed and therefore allow motion between the bony ends of the fracture. Cartilage formation, as the first step during endochondral ossification, occurs in less stable regions with higher strains, where no direct ossification can take place and thus occurs mostly in regions close to the fracture site. 12 During endochondral ossification, MSCs differentiate into chondrocytes and start building a cartilaginous extracellular matrix. This produces a callus that subsequently mineralises, and the mineralised callus is remodelled into bone. Bone repair in general is divided into three different phases: the inflammatory, the reparative or proliferative and the remodelling phases. 16 Initially, a haematoma is formed and inflammation in the fracture region occurs. The haematoma acts as a source of signalling molecules that initialise fracture healing, including interleukins (IL-1 and -6), tumour necrosis factor alpha 13,17 and growth factors, such as transforming growth factor-β1 (TGF-β1), fibroblast growth factors (FGFs), platelet-derived growth factor (PDGF) and bone morphogenic proteins (BMPs). 18,19 BMPs are part of the TGF-β superfamily, and key players in MSC proliferation and differentiation. 20 For example, BMP-2 directs the differentiation of cells from the periosteum or marrow cavity into a chondrogenic or osteogenic phenotype. 21 The following reparative phase is defined by vascular remodelling and recruitment of mesenchymal progenitor cells that will differentiate into chondrocytes or osteoblasts. 16 The differentiation of MSCs into bi-potential osteochondral progenitor cells is initially regulated by sex determining region Y-box 9 (SOX9) expression. 22 Remodelling is dynamically regulated by the activity of osteoblasts, osteocytes and osteoclasts. During remodelling, the degradation of callus tissue by osteoclasts is followed by replacement of woven bone with lamellar bone. 23 Including the remodelling phase, the whole fracture healing process can last up to several years. 24 During the formation of new bone tissue, the expression of genes encoding for collagen type I and II, as well as other extracellular matrix components, including osteocalcin, osteonectin and osteopontin, change over time and marker genes can be detected in in vitro experiments and indicate either differentiation towards chondrogenesis or osteogenesis. For osteogenic differentiation, common markers are osteocalcin (BGLAP), osteopontin (SPP1), runt-related transcription factor 2 (RUNX2), alkaline phosphatase (ALPL) and collagen type I (COL1A1 and COL1A2). 25 Well-established chondrogenic markers are collagen type II (COL2A1), aggrecan core protein (ACAN), cartilage oligomeric matrix protein (COMP) and SOX9. 26 In addition, a number of miRNAs have been identified to regulate central osteogenic differentiation markers. 27 For example, miR-9 inhibition increases mRNA levels of RUNX2 and BMP7 in bone tissue at the fracture site. 27

1.2
The role of vascularisation, innervation and mechanical load during fracture healing Fracture healing is supported by vascularisation, innervation and mechanical loading. The initial haematoma is a temporary matrix for the invasion of the vascular network, 28 which provides oxygen and nutrients and removes waste, including necrotic bone tissue resorbed by osteoclasts. 29 Here, the vascular endothelial growth factor (VEGF) signalling pathway is the principal mediator, stimulating angiogenesis, bone formation and callus mineralisation, 30 with BMP2 promoting angiogenesis by increasing VEGF production in osteoblasts. 31 Bone is a highly innervated tissue and the peripheral nervous system is directly involved in osteogenesis through secretion of neuropeptides, such as vasoactive intestinal peptide and calcitonin gene-related peptide, [32][33][34] which modulate osteogenic differentiation. 35 Mechanical loading, and particularly the strain across the fracture gap, is one major determinant of the fracture healing process and influences the time for the fracture to heal, the ossification route and the stability of the newly formed bone. 36 Mechanical forces influence the differentiation of MSCs by improving or preventing angiogenesis, 37 as well as activating the TGF-β/BMP pathway during the fracture healing process. [38][39][40] miRNAs are also involved in the control of bone remodelling, particularly by regulating osteoclast and osteoblast differentiation and function. Changes in miRNA expression levels influence the function, apoptosis and proliferation of bone cells, and can regulate differentiation processes. 41,42

miRNAs and bone diseases
Bone diseases, such as osteoporosis or osteoarthritis, are a common and increasing problem in the ageing population and miRNAs have already been investigated as predictive markers for individual outcomes of bone diseases. 43 For example, miR-146a/b has been shown to regulate the expression of FGF2, which is associated with bone mineral density (BMD), the main diagnostic variable for osteoporosis. 44 Higher levels of FGF stimulate osteoclastogenesis, which enhances bone resorption, leading to lower BMD. 44 Of note, miR-21, miR-23a and miR-24 have been found to be upregulated in the serum of patients who endured a bone fracture, and a similar miRNA profile was detected in osteoporotic bone tissue, 45 indicating that blood miRNA levels resemble tissue miRNA composition. Today, many studies have reported on changes in miRNA expression in osteoporotic fractures in animal models (e.g., induced by bilateral ovariectomy in rodents) and investigated miRNA expression during healing. Together, these studies indicate that miR-21 promotes early bone repair in rat models of osteoporosis and miR-21-3p improves the healing of osteoporotic fractures in mice. 46,47 The increasing understanding of the pivotal roles of miRNAs in time and special bone healing processes has set the stage for miRNAs as predictive markers for delayed fracture healing and non-unions. However, current knowledge on miRNAs in bone healing originates from diverse clinical populations, a wide range of different tissues and cell populations and a multitude of animal and cell models, with partly conflicting findings.

Objective
This review aimed to summarise and structure the findings from clinical populations, animals and cell models to identify miRNAs with the potential to be used as biomarkers to monitor the fracture healing process. Several studies have already investigated the role of miRNAs in fracture healing processes to find potential biomarkers for non-union fractures or fracture healing in general. This review aims to detect and discuss the unknown main regulators and highlight promising miRNAs that have the potential to be used for clinical diagnosis and treatment. 48,49 An advantage of miRNAs as biomarkers is that they can be detected in biofluids, and they can be analysed in blood samples by using a small amount of blood. They are very specific, as they can directly be connected to signalling pathways and their role in target gene regulation can be assessed.

METHODS
A systematic review was used to identify common miR-NAs involved in fracture healing or non-union followed by a qualitative analysis. We screened for all miRNAs that were validated as involved in the fracture healing process and possible biomarkers for non-union fractures. All included studies had to either (1) screen patient samples (which we defined as Type 1 Study) or (2) implement an animal model (Type 2 Study), as those studies have a high translational potential. Studies that only focused on the in vitro validation of miRNAs during chondrogenesis and osteogenesis were not selected for further qualitative analysis, as they are lack validation on a higher translational or clinical model.

Literature search and inclusion criteria
A systematic review (CRD42022344974) in accordance with the PRISMA guidelines 50 and following the suggestions for reporting on qualitative summaries was performed. 51,52 Literature search was conducted using PubMed, Web of Science, EBSCO and Scopus, including variations and combinations of the following keywords: 'microRNAs and fracture healing' and 'microRNAs and fracture non-union'. Any original article investigating miR-NAs in fracture healing or non-union fractures was eligible for inclusion. Specific inclusion criteria were as follows: (1) studies investigating miRNAs in patient samples, (2) studies investigating miRNAs in animal models of bone healing and (3) studies investigating miRNAs in in vitro models of bone healing. Studies that (1) reported only on in vitro analysis, (2) only focused on small interfering RNA/long noncoding RNA, (3) were not available in English (full text), (4) were not available as full-text or (5) retracted articles were excluded from further analysis. Conference abstracts and grey literature were not included. All records published until 28 February 2022 were eligible for inclusion.

Study selection, data extraction and aggregation
Data were extracted by two reviewers (Franziska Lioba Breulmann and Luan Phelipe Hatt), and tables were created including information on first author, year of publication, number of patients included/animals analysed, type of intervention, underlying diseases, differentially expressed miRNAs, miRNA analysis method (sequencing, microarray, quantitative polymerase chain reaction [qPCR]), clinical screening, type of in vitro experiment, animal fracture model and cell type. Strand information (-3p/-5p) was not included in the selection process since some studies indicated identical regulation independent of strands and strand information is not always provided by the authors. Direction of miRNA regulation was extracted as indicated by the authors (i.e., if statistical significance was reported). In the case of imprecise, uncommon, unclear/conflicting or missing descriptions of methods, or participants, studies were excluded.

Grouping of studies and synthesis
To provide a structured qualitative summary, studies were grouped into two main categories: (1) fracture healing and (2) non-union fractures. The validity of the reported findings was assessed using categories: clinical population, animal model and cell model. Studies that were found in both literature searches were included in only one of the two main categories according to the main subject of the study. For example, some studies were found in the general search for fracture healing but included nonunion fracture patients or non-union animal models. 53 The certainty of the evidence was addressed using an evaluation of how directly the included studies addressed the planned question/applied methodology (measurement validity), the number of studies and the consistency of effects across studies. The risk of bias of the studies was not assessed since only studies investigating patient populations followed by animal/cell model validation were included. We qualitatively analysed the included studies to evaluate which miRNAs have already been validated to be involved in fracture healing and non-union and which miRNAs are most promising as biomarkers for healing progress. Figure 1 summarises the literature search process. In brief, the search 'microRNAs and fracture healing' resulted in the following: n = 130 records on PubMed, n = 80 on Web of Science, n = 28 on EBSCO and n = 323 on Scopus. On PubMed, 74 full-text articles were screened, 45 on Web of Science, 25 on EBSCO and 65 on Scopus. From this search, 88 full-text articles were identified as fulfilling the selection criteria. However, six full texts found using the keywords 'microRNA and fracture healing' were categorised in the non-union group, as they focused on screening in non-union fracture patients or investigating a non-union model. In summary, a total of 82 fulltext articles were included in the fracture healing group (Table 1). For 'microRNA and fracture non-union', records identified were n = 17 on PubMed, n = 9 on Web of Science, n = 7 on EBSCO and n = 63 on Scopus, and 13 full texts fulfilled the inclusion criteria (Table 2). Additionally, six results from the keyword search 'microRNA and fracture healing' were included in the table of non-union fractures, as they focused on non-unions. In summary, as depicted in Table 2, only 19 full-text articles were found focusing specifically on miRNAs in fracture delay or non-union.

TA B L E 1 MicroRNAs (miRNAs) regulated in fracture healing
Author (

TA B L E 2 Differentially expressed microRNAs (miRNAs) in non-union fractures
Author (   F I G U R E 1 Literature search review process according to the PRISMA criteria. Reviews or record duplicates were excluded from analyses. Six records were excluded after screening of full-text articles according to following criteria: reported only on in vitro analysis, only focused on small interfering RNA (siRNA)/long noncoding RNA (lncRNA) and not available in English (full text) or retracted articles. The included studies are summarised in Table 1, 2 and 3. In summary, 82 studies were included in the 'microRNA and fracture healing' analysis, while 19 studies were selected for the 'microRNA and non-union' screening.

miRNAs in fracture healing: clinical screening
Twenty-four studies analysed miRNAs in blood or tissue samples (bone biopsies) of fracture patients. 45, Blood or tissue samples were collected during surgery, subsequently RNA was extracted, and validation of miRNAs was performed. In detail, n = 18 studies 45 Since the different studies provide different opportunities to find possible miRNAs involved in fracture healing, Figure 2 gives a complete overview about the investigations carried out and the different study types.

F I G U R E 2
Different study types to investigate microRNAs (miRNAs) during fracture healing process. Type 1 studies focused on patient screenings using bone biopsy or blood sample analysis. Type 2 studies used animal fracture models and analysed callus tissues. Analyses were either performed by miRNA sequencing or microarray, followed by in vitro validation of the miRNA and an animal study. Some studies performed an interventional animal study by injecting miRNAs into the animal fracture site for validation.
Thirty-nine of the studies that included an animal fracture model examined the effect of selected miRNAs directly on fracture healing by injection of miRNAs or their inhibitors into the fracture site. 27,[46][47][48]54,65,69,73,77,79,83,84,[88][89][90][92][93][94][95]98,99,101,102,105,[109][110][111][112][113][114][115][116][117][118][119][120][121][122][123] Seven studies used exosomes to deliver the identified miRNAs to the fracture site. 49,76,77,79,91,97,99 Osteoporosis patients with a lower BMD presented lowimpact fractures, such as spine fractures or femoral neck fractures. 124 Nine studies 46,47,65,81,102,104,106,111,125 used an osteoporosis fracture model to screen and validate differentially expressed miRNAs in vivo. This was done by ovariectomy in mice or rats, causing a loss in BMD to study miRNAs being differentially expressed in fractures under osteoporotic conditions. The osteoporosis model was either implemented to isolate bone marrow-derived MSCs (BMSCs) afterwards and perform transfection with miRNAs in vitro, 81,104 or the model was directly used to inject miRNAs or their inhibitors to study the fracture outcome in osteoporosis-induced animals. 65,111 To assess the fracture healing process in animal models, X-ray and micro-computed tomography (μCT) are commonly used, with the majority of studies combining both methods. X-ray was usually performed to control the model immediately after the first surgery where the fracture was implemented. Subsequently, X-ray was used during the follow-up to monitor the fracture healing, analysing callus formation, bridging of the fracture gap and callus volume. 64 There are only a few common timepoints during the experiment to perform an analysis based on μCT and usually the animals are euthanised for this purpose. Recently, in vivo μCT protocols have been developed that allow for longitudinal monitoring of healing progression. 126 This could be worth to be included in further studies investigating the fracture healing process. Common μCT analysis includes bone volume (BV), tissue volume (TV), BV/TV and BMD. 121 In terms of mouse fracture models, C57BL/6N was the most common mouse strain (n = 20). Twenty studies performed a fracture model in rats, without clearly specifying the animal strain, and four studies implemented an osteoporotic fracture model. 46,104,106,111 Two studies performed a fracture model in rabbits. 115,127

miRNAs in fracture healing: in vitro experiments
To validate miRNAs involved in differentiation processes, 65 studies included in vitro experiments. Fifty-seven studies performed transfection of BMSCs or osteoblasts with the miRNAs of interest, with lipofection as a common method to deliver miRNA agonists or antagonists. 54 A large heterogeneity in cell origins and cell types used was detected. Thirty-three studies used primary bone marrow mesenchymal stem cells isolated from human subjects, mice or rats. Only five studies worked with human primary cells to validate the effect of miRNAs on the differentiation process, which is probably the most clinically translatable in vitro experiment. 47,66,87,91,99 Most studies (n = 43) validated their findings using cell lines. Two studies performed experiments with human embryonic kidney cells HEK293, 84,121 one used the osteosarcoma cell line MG-63 90 and one study used human umbilical vein endothelial cells (HUVECs). 87 Studies on mouse cells used either BMSCs (n = 13) or the precursor osteoblast cell line MC3T3-E1 (n = 18). One study performed experiments on mouse embryonic C3H10T1/2 cells. 120

3.6
Summary of validated miRNAs in fracture healing Table 1 lists all the miRNAs described as differentially expressed during fracture healing processes. The miRNAs that are more often found to be regulated during bone remodelling in screenings and sequencing should be further validated and examined to find possible biomarkers predicting the outcome of the fracture healing process. Table 3 presents an overview of the most described miR-NAs that are differentially expressed in fracture healing (A) and non-union fractures (B). Overall, 121 different miRNAs were identified in 82 different studies as differentially expressed in fracture healing. Eighteen miR-NAs were identified in at least two independent studies, eight were described in three independent studies, four in four different studies and one in five independent studies (Table 3A). Three miRNAs were most frequently identified and analysed in the context of fracture healing ( Figure 3). miR-21 has been described as differentially expressed during fracture healing in nine different publications. 45,46,71,86,101,103,109,115,128 miR-140-3p/5p was investigated in seven different reports 48,78,85,103,104,120,129 and miR-214-3p/5p was described in eight different studies. 55,63,81,104,106,111,113,130
One report studied 727 miRNAs at different timepoints during fracture healing process in rat fracture model implementing closed femoral shaft fractures. 139 A total of 368 miRNAs were identified as upregulated early during fracture healing on day 5 (top four: miR-339-3p, miR-451-5p, miR-532-5p and miR-551b-3p), while 207 were increased on day 11 (top four: miR-221-3p, miR-376a-3p, miR-379-3p and miR-379-5p). The top four miRNAs for both timepoints were selected for further validation by reverse transcription-quantitative polymerase chain reaction (RT-qPCR). 139 Similarly, Waki et al. 140 performed microarray analysis on 680 miRNAs from tissue samples harvested on day 14 after surgery to implement a non-union fracture model or a healing fracture model in rats. They selected five miRNAs, miR-140-3p, miR-140-5p, miR-181a-5p, miR-181d-5p and miR-451a, according to the microarray analysis, for further validation in tissue samples. 140 Another study performed a microarray analysis on callus tissue samples from patients with identified non-union fractures 6-8 months after tibial fracture. 125 miRNAs identified as differentially expressed in nonunion fracture tissue samples were subsequently validated by RT-qPCR. A similar approach was used by Long et al., 138  found 557 miRNAs that were differentially expressed between the two groups and functionally validated the expression of the most differentially expressed miRNA target genes by transfecting BMSCs with miRNA-mimics or -inhibitors. 138

miRNAs in non-union fractures: fracture models
The fracture models used were comparable among studies since a femoral fracture model in mice or rats was commonly used. To study non-union fracture healing, the most common method was cauterisation or removal of the periosteum on the fracture site to avoid normal fracture healing. 129,140,142,143 Another approach was the creation of a critical size defect. 141 In this study, the authors used a fracture gap of 1.8 mm for the non-union fracture model in mice and only 0.25 mm for the normal healing fracture model. A metallic clip was implanted to keep the gap size during the experiment. 141 One study investigated the effect of DM on the fracture healing process and the miR-NAs involved in this disease. 139 DM was induced by an intraperitoneal injection of 40 mg/kg streptozotocin. Rats with blood glucose levels over 300 mg/dl at 1 week postinjection were included in the DM group. Non-DM mice served as a control group. All animals were subjected to femoral shaft fractures.
Out of 14 studies that implemented a fracture model, eight fracture models were performed in rats, 129,132,[138][139][140][142][143][144] and six were performed in mice 130,131,135,141,145,146 with C57BL/6J being the most common strain. 130,131,136,145 Two studies did not indicate the mouse strain used. 135,141

miRNAs in non-union fractures: in vitro experiments
Fourteen out of 19 studies performed an in vitro experiment to validate clinical or in vivo findings. Five of these studies 130,133,134,136,145 used osteoblasts, while six studies 53,131,132,135,138,143 used (B)MSCs with cells cultured under osteogenic conditions. Transfection of MSCs with miRNA-mimics or -inhibitors was also performed. 74,131 One study used adipose-derived stem cells (ASCs) to investigate the effect of different miRNAs on the differentiation process. 142 Even in the case of non-union-related studies, a large heterogeneity in cell origins and cell types used was detected. Six studies worked with human primary cells to validate the effect of miRNAs on the differentiation process, which is likely the most clinically relevant experiment. 53,131,132,135,138,142 Three studies used primary BMSCs/MSCs from rats. 79,143,144 In total, seven studies used cell lines for in vitro validations. Five studies used the mouse osteoblast cell line MC3T3-E1, 74,130,133,136,145 whereas two studies used the human osteosarcoma cell line MG-63. 134,137

Summary of validated miRNAs in non-union fractures
Sixty miRNAs were identified in 18 different studies. Three miRNAs were identified in two different reports: miR-31-5p, 140 (Table 3B). Three of the most common miRNAs involved in disturbed fracture healing including known targets are depicted in Figure 4.

DISCUSSION
This review aimed to systematically summarise findings from clinical populations, animals and cell models to identify miRNAs with the potential to be used as biomarkers to monitor the fracture healing process. Our approach identified miR-21, miR-140 and miR-214 as potential biomarkers for fracture healing in general (Figure 3), while miR-31-5p, miR-221 and miR-451-5p have a potential in the monitoring of fracture healing and non-union fractures (Figure 4). The discussion is focused on those six miRNAs, as they have been validated in several studies. In addition, their targets and the known regulated pathways linked to bone healing are presented, further supporting their promising role as biomarkers.

miRNAs in fracture healing: miR-21
miR-21 was identified in a microarray study as differentially expressed in osteoporotic compared to non-osteoporotic patient samples. 45 Nine independent studies 45,46,71,86,101,103,109,115,128 provided evidence for a pivotal role of miR-21 in rat and mouse fracture models. Treatment of the fracture site with miR-21 mimics increased callus formation and induced bridging of the fracture gap compared to an injection of scrambled miRNA. 46 In addition, overexpression of miR-21 in MSCs promoted osteogenesis and accelerated bone fracture healing, 101 consistent with the finding that miR-21 is involved in early stages of intramembranous ossification as an important step in fracture healing. 128 miR-21 promotes fracture healing by activating the phosphoinositid 3-kinase/serine-threonine-kinase/RAC-alpha serine/threonine-protein kinase (PI3K/AKT) signalling pathway, 109 which regulates a number of processes, such as cell survival, proliferation, growth, metabolism and angiogenesis. 147 Controversially, Sheng et al. 115 showed that miR-21 activates the extracellular signal-related kinase pathway concluding that a downregulation of miR-21 promotes fracture healing in rats. In addition, miR-21 regulates the TGF-β pathway by targeting the phosphatase and tensin homologue deleted on chromosome 10 (PTEN) 148 and by targeting the TGF-β-receptor 2 (TGFBR2). 149 Furthermore, miR-21 facilitates osteoclastogenesis by inhibiting programmed cell death 4 (PDCD4), 150 inhibits cell migration and promotes cytoskeletal organisation. The members of dedicator of cytokinesis (DOCK) 180-related protein superfamily have also been identified as target genes. 151 Downregulation of DOCK inhibits cell migration and promotes cytoskeletal organisation.

miRNAs in fracture healing: miR-140
miR-140 has been shown to promote osteogenic differentiation and calcium deposition in bone 120 and to play a role in skeletal development by regulating endochondral ossification. 152 Its role has been validated in a mouse F I G U R E 4 MicroRNAs (miRNAs) in fracture non-union. Mechanism of action of miR-31-5p, miR-221 and miR-451-5p involved in non-union fractures. '+' indicates promotion, while '-' indicates inhibition of the target gene or pathway. This regulation does not necessarily represent the direct target of the miRNA but rather the overall downstream pathway regulation. Main effect of miR-31-5p is the reduction of osteogenesis by inhibition of runt-related transcription factor 2 (RUNX2) and bone morphogenic protein 2 (BMP2) expression. miR-221 leads to increased fibroblast proliferation by upregulation of platelet-derived growth factor (PDGFA). miR-451-5p in general decreases cell growth, migration and angiogenesis, all critical factors for fracture healing. fracture model where it promoted fracture healing after injection in the fracture site. 120 Waki et al. 129 identified miR-140 as differentially expressed in a screening of fracture patients with healed fractures, compared to non-healed fractures. Orth et al. 141 confirmed that miR-140-3p and miR-140-5p were significantly downregulated in a non-union mouse fracture model. These observations may be explained by involvement of miR-140-3p in activation of the Wnt signalling pathway, which promotes osteogenesis 153 and the fracture healing process. 48 miR-140 may also control fracture healing by regulating the expression of Toll-like receptor 4 (TLR4) and BMP2. 103,142 Dnpep was identified as a target gene of miR-140-5p 141,152 and an upregulation of Dnpep caused skeletal defects by inhibiting BMP signalling. 152 However, Orth et al. 141 also predicted an inhibitory effect of miR-140 on BMP2 through the downregulation of stromal cell-derived factor 1α (SDF-1α) (CXCL12).
In contrast to the reported positive effects of miR-140 in fracture healing, it was also shown that the inhibition of miR-140-5p promoted osteogenesis in ASCs and enhanced fracture healing in an atrophic non-union rat model. 142 A transfection of chondrocytes with miR-140 mimics increased SMAD1 expression and suppressed the hypertrophy of chondrocytes by controlling the BMP pathway. 154 The role of miR-140 during fracture healing is still not completely understood. In particular, it is unclear which outcome is promoted by an increased miR-140 level at an early timepoint after a fracture. However, it is likely that the regulation of miR-140 during fracture healing process is time-dependent or the different strands are regulated differently.

miRNAs in fracture healing: miR-214
A review of the literature indicates contrasting findings regarding the role of miR-214 in fracture healing delay or non-union. Screening of fragility fracture patients showed decreased miR-214 levels in blood and tissue samples, and miR-214 was shown to regulate proliferation and apoptosis of osteoblasts and bone formation by inhibiting the expression of SRY-box transcription factor 4 (SOX4). 55 miR-214 was increased in patients directly following a fracture and inhibition of miR-214 promoted cell survival and extracellular matrix formation in the early phase of fracture healing by targeting type IV collagen (COL4A1), 63 a component of basement membranes. 155 Upregulated miR-214 led to improved fracture healing by regulating the BMP/Smad signalling pathway. 111 miR-214 is also involved in the modulation of Wnt/β-catenin pathway and inhibits endochondral ossification, which led to delayed fracture healing in fracture patients. 113 Functional studies suggested that suppressing miR-214 in MSCs enhances RUNX2 levels and promotes osteogenic differentiation. Li et al. 106 identified CTNNB1 as a target of miR-214, which is relevant since CTNNB1 encodes β-catenin, a mediator in the Wnt signalling pathway that activates the osteogenic transcription factor RUNX2. Thus, miR-214 may function as a possible therapeutic target to improve the fracture healing process and should be further investigated to decrease the risk for delayed fracture healing or non-union.

miRNAs in non-union fractures
Compared to investigations on miRNAs involved in physiological bone healing processes, data on miRNAs involved in non-union fractures are scarce. Nonetheless, some miR-NAs have been identified and validated for impaired bone healing. Table 2 includes miRNAs that have been described as differentially expressed during non-union fractures. Of note, miR-31-5p, miR-221 and miR-451-5p have not been reported in studies with undisturbed fracture healing and may thus be specifically regulated in non-union fractures and may represent promising miRNAs with biomarker potential. Those miRNAs should be further validated to examine their exact role during the fracture healing process.

miRNAs in non-union fractures: miR-31a-5p
Profiling miRNAs in rat non-union fractures showed an upregulation of miR-31a-3p/-5p. 140 miR-31a-3p/-5p were among the most upregulated miRNAs in the non-union fracture group compared to healing fractures on postfracture day 14. Orth et al. 141 reported a comparable observation from a mouse non-union model with a 1.8 mm femoral fracture gap, suggesting that miR-31-5p is a negative regulator of MSC osteogenic differentiation. 156 Accordingly, miR-31 was reported to target the expression of special AT-rich sequence-binding (SATB) homeobox 2 gene (SATB2), 156 which encodes a pro-osteoblastogenic transcription factor. 157 In addition, RUNX2 and BMP2 are known targets of miR-31a-5p and inhibition of miR-31a-5p promoted the osteogenic differentiation of MSCs in vitro. 158

miRNAs in non-union fractures: miR-221
Microarray analysis of callus tissue samples from atrophic non-union fracture patients revealed an increased expression of miR-221, 53 and functional studies in MSCs provided evidence that miR-221 overexpression inhibits the expression of PDGF subunit A (PDGFA). 53 This is of relevance since inhibition of PDGFA stimulates fibroblast proliferation in the early stages of fracture healing leading to improved bone formation. 159 Using a diabetic rat model of delayed fracture healing and non-union, Takahara et al. 139 also observed an altered expression level of miR-221-3p, which was highly upregulated on post-fracture days 5, 7 and 11 in the DM group.

miRNAs in non-union fractures: miR-451-5p
miR-451-5p was identified using microarray analysis in a rat DM model after analysing callus samples from the fracture site indicating elevated levels from days 5 to 14 after a surgery-implemented fracture. 139 Functional studies showed that miR-451-5p inhibits cell growth, migration and angiogenesis via downregulation of IL6R. 160 However, in a fracture model implemented in non-diabetic rats, a stronger upregulation of miR-451-5p in standard healing fractures was observed in comparison to nonunion fractures. 129 Furthermore, upregulation of miR-451-5p increased cyclooxygenase 2 (COX2) protein levels linking miR-451 to endochondral ossification by increased chondrocyte hypertrophy. 161,162

miRNAs in osteoporotic fractures
Even though this was not a category that was systematically analysed within this review, some miRNAs associated with osteoporotic fractures have also been identified. Patients suffering from osteoporosis have a higher incidence of non-traumatic fractures and have an elevated risk for non-union. The internal fixation to treat fractures in osteoporotic patients suffers from insufficient strength and stability because of low BMD. 163 Nine of the included studies 46,47,81,102,104,106,111,164 implemented an osteoporotic fracture model to focus on this group of patients. miR-140, miR-214, miR-21 and miR-26a were shown to improve osteogenic differentiation in osteoporotic bone defects. 46,102,130 miR-22 stimulated osteogenic differentiation by inhibition of MYC protooncogene (MYC)/PI3K/AKT pathway and was beneficial in an osteoporotic bone model to promote bone healing. 47 Possible evidence for a role of miR-214 in fracture healing in an ovariectomised mouse model was reported by Wang et al., 81 who transfected HUVECs with a miR-214 inhibitor, leading to improved tube formation and cell migration. Mechanical stimulation across the fracture gap in an osteoporotic mouse model decreased miR-214 levels improving fracture healing. In addition, miR-214 suppression enhanced RUNX2 expression and promoted osteogenic differentiation in an osteoporotic rat model, 106 while enhanced miR-214 expression delayed healing of osteoporotic model fractures likely by inhibiting the BMP/Smad signalling pathway. 111 In terms of improved fracture healing in osteoporosis, miR-187-induced osteogenic differentiation, bone reconstruction and healing in a mouse osteoporotic fracture model. 65 A subsequent analysis in a model of ovariectomy to induce osteoporosis confirmed the importance of miR-187. 65 It has also been suggested that knockdown of miR-100 can improve the osteogenic differentiation of BMSCs by promoting the AKT/mechanistic target of rapamycin kinase (mTOR) pathway. 164 The identified miRNAs may be used to predict the risk for a non-union in osteoporotic fracture patients and to improve the treatment of fragility fractures due to a low BMD.

Future perspectives
Our analyses suggest that specific miRNAs with biomarker potential exist, which may be used to predict disturbed fracture healing alone or in combination. However, translation into clinical practice requires a standardised approach for future studies, also with respect to minimally invasive sampling procedures such as liquid biopsies. Also, it needs additional evidence that data from the osteoporotic models are relevant in humans with osteoporosis.
Since there is a lack of knowledge concerning the comparability of local versus circulating miRNA levels, we suggest combined analysis of the herewith identified miRNAs in non-union fracture patients' callus tissue collected during revision and concurrent blood sampling. Correlation with clinical outcomes should be performed over at least 6 months after fracture. Additional individual factors with influence on miRNA levels should be thoroughly investigated and reported, including age, DM, menopausal state and T-score, in the case of osteoporotic patients. Further research including functional analysis of identified miR-NAs using established in vivo or in vitro models is needed to identify the involved target molecules and pathways and to understand miRNA expression profiles over time. The overall vision is to use these miRNAs as diagnostic markers, alone or in combination with other factors, to predict the risk of fracture healing disturbances and improve the decision making and the individual treatment options for the patients. Moreover, these miRNAs or the pathways they affect also represent promising therapeutic targets to improve fracture healing.

CONCLUSIONS
Despite large heterogeneity in the field of miRNAs and fracture healing, this systematic analysis of clinical screenings and functional validation studies revealed a set of miRNAs with biomarker potential in disturbed fracture healing. Based on our investigation, miR-31-5p, miR-221 and miR-451-5p appear to be involved in processes linked to fracture non-union and could be used to predict disturbed bone healing. For fracture healing in general, focusing on miR-21, miR-140 and miR-214 is promising for future investigations. Further studies will have to focus on those miRNAs and on their further validation during fracture healing before miRNA-based theranostic approaches become an option.

A C K N O W L E D G E M E N T S
This work was supported by AO Foundation, AO CMF and AO Trauma. Figures were created with BioRender.com.

C O N F L I C T O F I N T E R E S T
The authors declare they have no conflicts of interest.