Abstract
Background:
Diagnosing hip and knee periprosthetic joint infections (PJIs) is challenging, necessitating sensitive and specific biomarkers for accurate diagnosis. Cell-free microRNAs (miRNAs) are emerging as noninvasive biomarkers. We hypothesized that hip and knee PJIs are associated with unique cell-free miRNA profiles in synovial fluid, which can be used for the diagnosis of infection.
Methods:
Synovial fluid samples from 173 Caucasian patients undergoing septic or aseptic revision total joint replacement (TJR) of the hip or knee, as well as samples from 6 osteoarthritic knees, were analyzed. The samples were divided into a discovery group (40 samples; 50% septic) and a validation cohort (133 samples; 35% septic). Small RNA next-generation sequencing (NGS) was used to screen miRNAs in the discovery samples, with reverse transcription-quantitative polymerase chain reaction (RT-qPCR) used to confirm the NGS findings and to validate results in the independent, larger cohort. Logistic regression and cross-validation were applied to assess the diagnostic power of individual and combined miRNAs.
Results:
NGS identified 132 miRNAs with significant differences (false discovery rate < 0.05) between the septic and aseptic synovial fluid samples. Of these, 18 miRNAs were further analyzed with use of RT-qPCR in the independent cohort, with miR-223-3p and miR-338-5p showing the highest increases in septic synovial fluid (log2 fold change >4) and miR-151a-3p and miR-214-3p showing the most substantial reductions. To investigate the performance of the multivariable models, logistic regression was performed by dividing the cohort into a training set (60%) and a test set (40%), which showed improved performance relative to that of the univariate models (median area under the curve [AUC] for the multivariable models, 0.96). A subgroup analysis by joint type, gender, and synovial fluid sample preparation confirmed robust miRNA biomarker performance for PJI.
Conclusions:
Cell-free miRNA levels in the synovial fluid of patients undergoing septic hip or knee TJR were altered in response to infection, indicating immune cell activity in the joint. These miRNAs offer sensitive and specific pathogen-independent biomarkers with potential clinical applications in the diagnosis of hip and knee PJI.
Level of Evidence:
Diagnostic Level III. See Instructions for Authors for a complete description of levels of evidence.
The diagnosis of periprosthetic joint infection (PJI) of the hip and knee after total joint replacement (TJR) remains challenging1,2. Currently, the diagnosis of PJI is based on clinical and laboratory parameters in serum and synovial fluid, as well as based on microbiological results3,4. In recent years, synovial fluid markers have been described in the literature, with heterogenous results regarding their sensitivity and specificity5,6.
MicroRNAs (miRNAs) are small, endogenous, noncoding RNAs that regulate gene expression. They have gained considerable attention as potential diagnostic and prognostic biomarkers in various medical fields7-10. It has been demonstrated that miRNAs can be reliably detected in synovial fluid using next-generation sequencing (NGS) and reverse transcription-quantitative polymerase chain reaction (RT-qPCR)11. Importantly, the expression levels of miRNAs are known to change in response to physiological stimuli and pathological processes12, suggesting that miRNA profiles in synovial fluid may differ between patients undergoing septic versus aseptic revision TJR of the knee or hip.
A recent study investigated miRNA expression in the plasma and tissue of patients with hip PJI as compared with a control group13, but, to date, no study has specifically investigated miRNA profiles in the synovial fluid of patients with PJI. We hypothesized that synovial fluid may be a promising source of novel miRNA biomarkers of PJI that may aid in the diagnosis of joint infection.
This study provides strong preliminary evidence supporting the clinical utility of synovial fluid miRNAs as sensitive and specific biomarkers for diagnosing PJI.
Therefore, the purpose of this study was to discover and independently validate synovial fluid miRNAs that are potential biomarkers of infection in knee and hip revision TJRs. The 4 specific aims of the study were to (1) perform an NGS-based analysis of miRNAs in septic and aseptic synovial fluid, (2) replicate NGS data for select miRNAs with use of RT-qPCR, (3) perform an independent validation in new samples, and (4) apply machine learning to assess whether combinations of miRNAs in multivariable models can improve diagnostic performance compared with individual miRNAs.
Materials and Methods
Study Population and Experimental Design
After institutional review board approval, we retrospectively analyzed data from the prospectively maintained revision arthroplasty registry and corresponding biobank samples of our institution. Revision surgeries were classified as septic or aseptic per the 2018 International Consensus Meeting infection criteria for PJI4. Table 1 summarizes the cohort of 173 synovial fluid samples (151 knee and 22 hip revision TJRs) and the 6 osteoarthritic controls that were selected for this study. The cohort included 106 samples from female patients (61.3%) and 67 samples from male patients (38.7%), with a mean patient age of 71 years (range, 29 to 92 years) and a mean body mass index (BMI) of 29 kg/m2 (range, 18 to 54 kg/m2). Based on self-reported race and ethnicity, all patients were Caucasian. The cohort was divided into a discovery group (20 septic and 20 aseptic) and a validation group (46 septic and 87 aseptic). Table 1 provides detailed demographics for each group, including the number of hip and knee TJRs, the number of primary and revision TJRs, the number and types of surgical procedures and infections, and the laboratory parameters. PJIs were categorized as early acute, late acute, or chronic, as previously described14. Of the 66 patients in the septic PJI group who were analyzed, 56 (84.8%) had positive microbiological results, comprising 49 patients (87.5%) with monomicrobial infections and 7 patients (12.5%) with polymicrobial infections. The most common pathogens were Staphylococcus epidermidis (25.4%), other coagulase-negative staphylococci (17.5%), and Staphylococcus aureus (11.1%) (see Appendix Table 1). All samples were analyzed in a blinded manner.
Table 1.
Cohort Characteristics*
| Total Cohort | Discovery Group | Validation Group | Osteoarthritis Group | |
|---|---|---|---|---|
| Sample size | 173 | 40 | 133 | 6 |
| Demographics | ||||
| Age (yr) | 71 (29-92) | 73 (51-89) | 70 (29-92) | 72 (53-81) |
| BMI (kg/m2) | 29 (18-54) | 28.6 (19.1-50.2) | 29.02 (17.69-54.03) | 27.62 (24.01-34.93) |
| Female | 106 (61.3%) | 24 (60.0%) | 80 (60.2%) | 2 (33.3%) |
| Joint | ||||
| Hip | 22 (12.7%) | 0 (0%) | 22 (16.5%) | NA |
| Knee | 151 (87.3%) | 40 (100%) | 111 (83.5%) | NA |
| Surgery | ||||
| After primary TJR | 72 (41.6%) | 21 (52.5%) | 51 (38.3%) | NA |
| After revision TJR | 101 (58.4%) | 19 (47.5%) | 82 (61.6%) | NA |
| Surgical procedures | ||||
| Aseptic revision TJR | 107 (61.8%) | 20 (50.0%) | 87 65.4%) | NA |
| Total component replacement | 68 (63.6%) | 16 (80.0%) | 52 (59.8%) | NA |
| Partial component replacement | 39 (36.4%) | 4 (20.0%) | 35 (40.2%) | NA |
| Septic revision TJR | 66 (38.2%) | 20 (50.0%) | 46 (34.6%) | NA |
| Two-stage exchange | 30 (45.5%) | 7 (35.0%) | 23 (50.0%) | NA |
| DAIR procedure | 25 (37.9%) | 9 (45.0%) | 16 (34.8%) | NA |
| Single-stage exchange | 8 (12.1%) | 3 (15.0%) | 5 (10.8%) | NA |
| Amputation | 2 (3.0%) | 1 (5.0%) | 1 (2.2%) | NA |
| Girdlestone procedure | 1 (1.5%) | 0 (0%) | 1 (2.2%) | NA |
| Infection status | ||||
| Aseptic | 107 (61.8%) | 20 (50.0%) | 87 (65.4%) | NA |
| Septic | 66 (38.2%) | 20 (50.0%) | 46 (34.6%) | NA |
| Early acute | 16 (24.2%) | 5 (25.0%) | 11 (23.9%) | NA |
| Late acute | 11 (16.7%) | 4 (20.0%) | 7 (15.2%) | NA |
| Chronic | 39 (59.1%) | 11 (55.0%) | 28 (60.9%) | NA |
| Laboratory parameters | ||||
| Aseptic revision TJR | ||||
| Serum CRP (mg/L) | 3.0 (0.10-35.20) | 4.85 (0.23-32.60) | 2.90 (0.10-35.20) | NA |
| Serum leukocytes (109/L) | 7.03 (3.47-20.65) | 6.97 (3.79-12.04) | 7.03 (3.47-20.65) | NA |
| Synovial CRP (mg/L) | 0.50 (0.20-13.50) | 0.50 (0.20-12.20) | 0.50 (0.20-13.50) | NA |
| Synovial leukocytes (109/L) | 0.34 (0.02-4.16) | 0.38 (0.06-3.73) | 0.29 (0.02-4.16) | NA |
| Synovial PMN (%) | 17.50 (5.60-64.40) | 36.0 (7.2-64.4) | 16.1 (5.6-55.2) | NA |
| Septic revision TJR | ||||
| Serum CRP (mg/L) | 60.75 (0.40-410.60) | 12.70 (0.40-410.60) | 69.80 (1.20-351.90) | NA |
| Serum leukocytes (109/L) | 8.0 (3.95-161.72) | 7.96 (4.78-15.66) | 8.01 (3.95-161.72) | NA |
| Synovial CRP (mg/L) | 11.60 (0.20-217.40) | 3.70 (0.20-142.30) | 12.65 (0.20-217.40) | NA |
| Synovial leukocytes (109/L) | 26.13 (0.17-474.45) | 8.61 (0.17-357.38) | 26.90 (0.40-474.45) | NA |
| Synovial PMN (%) | 89.80 (8.80-97.80) | 75.1 (8.8-97.6) | 91.1 (23.2-97.8) | NA |
Values are given as the number of patients, with the percentage in parentheses, or as the mean, with the range in parentheses. BMI = body mass index; NA = not applicable; TJR = total joint replacement; DAIR = debridement, antibiotics, and implant retention; CRP = C-reactive protein; PMN = polymorphonuclear leukocytes.
Synovial Fluid Sample Collection
Synovial fluid samples were collected either preoperatively or intraoperatively, transported on ice, and stored at −80°C. Samples were divided into 500-µL aliquots within 30 minutes. We analyzed 46 native synovial fluid samples and 127 centrifuged (3,000 ×g at 4°C for 10 minutes) supernatant samples.
Total RNA Extraction
Synovial fluid samples were thawed and centrifuged at 12,000 ×g for 5 minutes at 4°C. Total RNA was extracted from 100 μL using an miRNeasy Mini Kit (QIAGEN) after dilution to 200 μL with nuclease-free water (NFW), as previously described11. Total RNA was eluted in 30 μL NFW and stored at −80°C. The RNA concentration was assessed using microcapillary electrophoresis (Agilent Technologies).
Small RNA Sequencing
Small RNA libraries were prepared using the RealSeq-Biofluids library kit (RealSeq Biosciences) according to the validated workflow previously developed in our laboratory11. Total RNA (1.5 ng) was mixed with 1 µL of miND Spike-Ins (TAmiRNA). Adaptor-ligated libraries were circularized, reverse-transcribed, and PCR-amplified using 21 cycles. DNA library size and yield were determined with the use of the DNA1000 Kit (Agilent Technologies), and the libraries were subsequently diluted and pooled into equimolar concentrations. Size selection was performed using 3% agarose cassettes (catalog number, BDQ3010; Sage Science). Pools were quantified by microcapillary electrophoresis prior to sequencing in 50-bp single-end mode on an Illumina NextSeq 2000, targeting 5 to 10 million total reads per sample (see Appendix Fig. 1).
RT-qPCR Analysis
Reverse transcription was performed using the miRCURY LNA RT Kit (QIAGEN) and 2 μL of total RNA input. qPCR was conducted using the miRCURY SYBR Green Master Mix and miRCURY LNA miRNA PCR assays (QIAGEN). cDNA was diluted 1:100 before qPCR. Synthetic spike-ins (QIAGEN) were added at specific steps to evaluate RNA isolation (UniSp4), reverse transcription (cel-miR-39-3p), and PCR efficiency (UniSp3). qPCRs were run on a LightCycler 480 Instrument II (Roche) using the following protocol: 95°C for 2 minutes, 45 cycles of 95°C for 10 seconds, and 56°C for 60 seconds. Melting curves were generated from 55° to 99°C, and quantification cycle (Cq) values were calculated using the second derivative maximum method, normalized to hsa-miR-22-3p. Quality metrics are shown in Appendix Figure 2.
Small RNA-Sequencing Data Analysis
Raw sequencing reads were processed using the miRNA NGS Discovery (miND) pipeline15. Read counts were normalized to total miRNA reads per sample in order to calculate reads per million (RPM). Unsupervised clustering and differential expression analysis were performed in R/Bioconductor using edgeR16. Independent filtering with DESeq2 was utilized to optimize cutoffs for low-abundance miRNAs. Significant differences between septic and aseptic samples were identified using a false discovery rate (FDR) threshold of 0.05.
Statistical Analysis
RT-qPCR data (delta Cq values) were analyzed for differences between the septic and aseptic groups with use of the Wilcoxon rank-sum test, with p values adjusted for multiple testing via the Benjamini-Hochberg method.
Statistical and Machine Learning Workflow for Multivariable Biomarker Model Development
A machine learning approach leveraging logistic regression models to assess the diagnostic performance of biomarker combinations was applied. All analyses were performed using the R package “Classification And REgression Training (caret).” For each set of biomarkers (i.e., combination of miRNAs), input data were preprocessed by selecting complete sets with no missing measurements. In the next step, the data set was partitioned into a training set (60%) and a test set (40%) using the createDataPartition() function and a fixed seed set by set.seed(). For each set of biomarkers, a model was trained by evaluating the effect of model tuning parameters on performance using resampling and tenfold cross-validation, choosing the optimal model settings across these parameters (i.e., the selection of an equation), and estimating model performance from the training set. The resulting model parameters were applied to the independent validation set to determine the area under the curve (AUC) from the receiver operating characteristic (ROC) analysis.
Results
NGS-Based Discovery of Synovial Fluid-Derived MiRNA Biomarkers of PJI
The discovery of miRNA biomarkers of PJI in synovial fluid was performed using small RNA sequencing to achieve an untargeted and genome-wide analysis of miRNAs. The discovery cohort included samples from patients with aseptic (n = 20) and septic (n = 20) knees, with 40% of samples from male patients (Table 1), and was complemented with samples from patients with osteoarthritic (n = 6) knees. No significant differences in age or BMI were observed between septic and aseptic groups.
The quality of small RNA-sequencing data was assessed on the basis of the total read count, the miRNA read count, the number of detected miRNAs, the recovery of RNA spike-ins, and the recovery of NGS spike-ins (see Appendix Fig. 1). Up to 800 miRNAs (median, 540) were detected per sample. No significant difference in the number of detectable miRNAs was observed between aseptic, septic, and osteoarthritic synovial fluids. The recovery of RNA and NGS spike-ins was high for all synovial fluids.
All 40 replicates (20 septic, 20 aseptic) were included in the statistical analysis. After adjusting for multiple testing (FDR < 0.05), 132 miRNAs were differentially expressed (Fig. 1-A); of these, 62 (47%) were upregulated and 70 (53%) were downregulated in septic joints. Some upregulated miRNAs showed effect sizes of more than twentyfold (e.g., miR-223-3p: log2 of the fold change [log2FC] = 4.54), whereas downregulations were more moderate, with a maximum sixfold reduction (e.g., miR-214-3p: log2FC = −2.09) compared with aseptic synovial fluid.
Fig. 1.
Figs. 1-A, 1-B, and 1-C: NGS-based analysis of miRNA biomarkers in the synovial fluid of aseptic and septic knees undergoing revision arthroplasty. Fig. 1-A: Volcano plot depicting the log2 fold change (FC) (x axis) and –log10 of the false discovery rate (FDR) (y axis) for the comparison of miRNA levels in synovial fluid between septic (n = 20) and aseptic (n = 20) knee arthroplasties. MiRNAs with an FDR of <0.05 are depicted in red. A subset of 18 miRNAs selected for further validation is depicted in green. Fig. 1-B: MA plot depicting average (A) synovial fluid abundance (log10CPM; x axis) versus log2 fold change (M; y axis). MiRNAs with an FDR of <0.05 are depicted in red. A subset of 18 miRNAs selected for further validation is depicted in green. Fig. 1-C: Heatmap based on NGS data (RPM values) for 18 miRNAs that were selected for further validation. Septic (n = 20), aseptic (n = 20), and osteoarthritic (n = 6) synovial fluid samples are shown. RPM values were ln(x + 1)-transformed. The rows are centered, and unit variance scaling was applied to the rows. Imputation was used for missing-value estimation. Rows and columns were clustered using the Pearson correlation distance and complete linkage.
A subset of 18 miRNAs with an FDR of <0.05, a log2FC of >|1|, and an average abundance of >4 log10-transformed counts per million (CPM) was selected (Figs. 1-A and 1-B) and was used to cluster septic, aseptic, and osteoarthritic synovial fluid samples. The resulting heatmap revealed a distinct grouping of most of the septic samples, with the osteoarthritic and aseptic synovial fluids forming a second homogeneous cluster (Fig. 1-C).
RT-qPCR Enabled Successful Technical Replication of NGS Results
To replicate the NGS results, the 18 miRNA candidates were analyzed in the same 40 samples with use of RT-qPCR. NGS and RT-qPCR data were compared by estimating the classification performance (the AUC from the ROC analysis) of each method for each of the 18 miRNAs. We observed a strong correlation and agreement in AUC values between the RT-qPCR and NGS results (Fig. 2), confirming the suitability of RT-qPCR for further validation studies.
Fig. 2.
RT-qPCR replication of the NGS results for 18 candidate miRNA biomarkers. AUC values obtained by RT-qPCR (y axis) are plotted against those obtained by NGS data (x axis) for the differentiation of septic from aseptic synovial fluid samples.
PJI MiRNA Biomarker Candidates May Have Been Released from Infiltrating Immune Cells or Joint Tissue
To explore the cellular origin of miRNAs that were differentially regulated between septic and aseptic joints, we investigated their cell type-dependent expression patterns using the publicly available Functional ANnoTation Of the Mammalian genome (FANTOM5) data repository (RIKEN)17. NGS data for the 18 miRNAs of interest and 2 potential reference miRNAs (miR-92a-3p and miR-15a-5p) were obtained for 21 different cell types, including those involved in innate immunity (monocytes, dendritic cells, macrophages, neutrophils, and mast cells), those involved in adaptive immunity (B cells, natural killer [NK] cells, T cells), and those present in joint tissue (endothelial cells, synoviocytes, chondrocytes, adipocytes, osteoblasts, myoblasts, mesenchymal stem cells, and muscle cells). The clustering of miRNAs by cell-type expression identified 3 distinct clusters (Fig. 3). The first cluster, associated with cell types that are present in joint tissue, was characterized by miR-151a-3p (endothelial cells) and miR-214-3p and miR-424-3p (mesenchymal cells). The second cluster, linked to adaptive immunity (B cells, T cells, and NK cells), showed high expression of miR-142-5p. The third cluster, linked to innate immunity (neutrophils and monocytes), was marked by miR-338-5p and miR-223-3p. These findings suggest that changes in miRNA levels in synovial fluid primarily reflect immune cell infiltration and host cell responses during joint infection, which alter the profile of cell-free miRNAs.
Fig. 3.
Heatmap showing the expression (log10CPM) of the selected miRNA panel in cell types present in joint tissue (bone, cartilage, tendon), as well as in adaptive and innate immune cell types. PJI miRNA biomarker candidates showed enriched expression in either hematopoietic or mesenchymal-derived cell types. PBMC = peripheral blood mononuclear cells.
Independent Validation of Biomarker Candidates by RT-qPCR
To validate the miRNA candidates that were identified by small RNA sequencing, the 18 miRNA candidates were analyzed using RT-qPCR in 133 independent synovial fluid samples, of which 46 (35%) were classified as septic. No significant differences in age, joint type, gender, or BMI were found between the septic and aseptic groups in the validation cohort (Table 1).
The effect size (log2FC) between the septic and aseptic groups and the adjusted p value (calculated using the Wilcoxon rank-sum test with the Benjamini-Hochberg correction) were obtained and compared side by side to the results from the discovery cohort (Table 2). We found that all miRNAs could be successfully validated, with highly similar effect sizes and significance levels.
Table 2.
Independent Validation of Candidate MiRNA Biomarkers in 133 Samples and Comparison to Discovery Study Results
| MiRNA | log2FC | Adjusted P Value* | ||
|---|---|---|---|---|
| Discovery (Septic/Aseptic, N = 20/20) | Validation (Septic/Aseptic, N = 46/87) | Discovery | Validation | |
| hsa-miR-142-3p | 2.39 | 2.23 | 0.0001 | <0.0001 |
| hsa-miR-142-5p | 2.68 | 2.99 | <0.0001 | <0.0001 |
| hsa-miR-151a-3p | −1.38 | −1.27 | 0.0006 | <0.0001 |
| hsa-miR-191-5p | 1.49 | 1.45 | 0.0012 | <0.0001 |
| hsa-miR-19a-3p | 1.64 | 0.97 | 0.0001 | 0.0006 |
| hsa-miR-214-3p | −2.00 | −2.15 | 0.0002 | <0.0001 |
| hsa-miR-223-3p | 4.41 | 4.93 | 0.0001 | <0.0001 |
| hsa-miR-30d-5p | 0.68 | 0.47 | 0.007 | 0.0106 |
| hsa-miR-338-5p | 3.7 | 4.66 | 0.0002 | <0.0001 |
| hsa-miR-345-5p | 1.57 | 1.72 | <0.0001 | <0.0001 |
| hsa-miR-3615 | 2.48 | 2.29 | 0.0001 | <0.0001 |
| hsa-miR-424-3p | 2.11 | 2.06 | <0.0001 | <0.0001 |
| hsa-miR-425-3p | 1.99 | 2.19 | 0.0002 | <0.0001 |
| hsa-miR-425-5p | 1.79 | 1.42 | <0.0001 | <0.0001 |
| hsa-miR-505-3p | 2.21 | 2.54 | <0.0001 | <0.0001 |
| hsa-miR-548ay-5p | 2.41 | 1.68 | <0.0001 | <0.0001 |
| hsa-miR-548d-5p | 2.53 | 1.98 | 0.0001 | <0.0001 |
| hsa-miR-629-5p | 2.16 | 1.94 | <0.0001 | <0.0001 |
Wilcoxon rank-sum test adjusted for multiple testing according to Benjamini-Hochberg.
Multivariable Analysis to Improve the Diagnostic Performance of MiRNA Biomarkers for PJI
To optimize diagnostic performance, we evaluated multivariable diagnostic models that combined miRNAs from the 18 biomarker candidates. Therefore, the discovery and validation cohorts were merged into 1 complete cohort (n = 173; Table 1), as independent validation had confirmed the robustness of the individual miRNA biomarkers. Low to moderate correlations among the 18 miRNAs (data not shown) supported a multivariable approach. The analysis workflow, visualized in Figure 4-A, involved randomly splitting the cohort into training (∼60%) and test (∼40%) data sets, using the training set to build a multiple logistic regression model and to define model parameters, and applying the model to the test set to evaluate performance via ROC analysis (i.e., the AUC values).
Fig. 4.
Figs. 4-A and 4-B: Development of multivariable miRNA models for the diagnosis of PJIs in the knee and hip. Fig. 4-A: A flowchart illustrating the process for the development of the classification models. For each univariate or multivariable model, the data were partitioned into a training set (∼60%, or 105 samples) and a test set (∼40%, or 68 samples). The training set was used to develop a logistic regression model using cross-validation to increase the robustness for parameter selection. The model parameters were then fixed and applied to the independent test set to assess model performance based on ROC analysis. Fig. 4-B: Box-and-whisker plot depicting the performance (AUC value) obtained from univariate (single miRNA) and multivariable (2, 3, or 4 miRNAs) models in the test data set. A nonparametric Kruskal-Wallis test was performed to compare the observed performance (AUC values) between the univariate and multivariable models. The whiskers represent the range, the bounds of the boxes represent the 1st and 3rd quartiles, and the horizontal lines represent the median.
Using this approach, the median AUC increased significantly from 0.83 in the univariate models to 0.96 in the multivariable models with 4 miRNAs (Fig. 4-B). To mitigate overfitting, we limited the model size and only analyzed pairwise miRNA models in detail. Figure 5-A shows the classification performance (AUC) of all pairwise combinations (bivariable models), consisting of 1 fixed miRNA and 1 variable miRNA from the set of 18. AUCs are presented for all 17 combinations for each fixed miRNA, with the single miRNA AUC superimposed for comparison. This analysis identified several miRNAs that were required to achieve high classification performance (AUC > 0.95), including miR-338-5p, miR-223-3p, miR-505-3p, and miR-142-5p.
Fig. 5.
Figs. 5-A through 5-E: Diagnostic performance of bivariable miRNA models for PJIs. Fig. 5-A: Model performance in the form of AUC values, which were determined from the test set of bivariable miRNA classification models, is shown. For each miRNA, its individual AUC value (blue dot, single miRNA) as well as the AUC values obtained from pairing the miRNA with each of the 17 remaining miRNAs (red box, paired model) are shown. Figs. 5-B through 5-E: The diagnostic performance of the logistic regression model, expressed as the probability of being septic (P | septic), derived from miR-338-5p and miR-214-3p in the validation cohort (n = 133) (Fig. 5-B) and total cohort (n = 173) (Fig. 5-C), as well as in the subgroups of total knee arthroplasties (n = 151) (Fig. 5-D) and total hip arthroplasties (n = 22) (Fig. 5-E).
A bivariable model combining miR-338-5p and miR-214-3p was selected to further analyze its performance. This selection was based on the association of miR-338-5p with neutrophils and the association of miR-214-3p with joint tissue (Fig. 3). In the validation cohort, the model achieved an AUC of 0.969 (Fig. 5-B). We further analyzed its sensitivity by selecting a cutoff that would achieve >95% sensitivity. This cutoff was identified as a probability (P) of 0.34, at which 95.65% sensitivity was achieved while retaining 89.33% specificity. Next, the model performance was investigated in the total cohort as well as in the subgroups of hip and knee PJI. The model showed no decline in performance, as measured with the AUC (Figs. 5-C, 5-D, and 5-E).
The model performance was equal in both genders and was independent of the collection method (centrifuged versus native) of the synovial fluid samples, with AUCs consistently exceeding 0.94 (Figs. 6-A through 6-D). Finally, we observed consistent performance of the model in culture-negative versus culture-positive PJIs (Fig. 6-E) and across early acute, late acute, and chronic infections, with the highest AUC of 0.987 observed in early acute PJIs (Fig. 6-F).
Fig. 6.
Diagnostic performance of miR-338-5p and miR-214-3p by subgroup. The diagnostic performance of the logistic regression model, expressed as the probability of being septic (P | septic), derived from miR-338 and miR-214 in female patients (n = 106) (Fig. 6-A), male patients (n = 67) (Fig. 6-B), synovial fluid (SF) samples collected by centrifugation (n = 127) (Fig. 6-C), and native SF samples (n = 46) (Fig. 6-D), as well as in culture-negative versus culture-positive PJIs (Fig. 6-E) and across early acute, late acute, and chronic infections (Fig. 6-F).
Discussion
In this study, we found that hip and knee PJIs were associated with unique miRNA signatures in synovial fluid, which were distinct not only from those in aseptic joints but also from those in osteoarthritic joints. Even some of the individual miRNAs achieved nearly 100% diagnostic accuracy for PJI, with several other miRNAs demonstrating AUC values of >0.90 in the ROC analysis, indicating a very promising novel approach in PJI diagnostics.
A cell-of-origin analysis suggested that the regulation of miRNAs in septic synovial fluid reflects 2 distinct biological mechanisms (Fig. 7). The upregulation of miRNAs that are enriched in innate immune cells, such as neutrophils (e.g., miR-338-5p, miR-223-3p), and adaptive immune cells (e.g., miR-142-3p, miR-142-5p) is likely due to the infiltration of immune cells into the joint space during infection and the subsequent release of cell-type-specific miRNAs that are absent from aseptic synovial fluid. Conversely, miRNAs that were transcribed in mesenchymal cells were repressed; for example, miR-214-3p was reduced fourfold in septic synovial fluid. This miRNA is known to be downregulated in chondrocytes under chronic inflammation in order to support nuclear factor kappa-B (NF-κB) signaling activation and to amplify inflammatory responses18. In the present study, a logistic regression model combining neutrophil-enriched miR-338-5p with mesenchymal-specific miR-214-3p achieved robust diagnostic accuracy for PJI (AUC > 0.95) in the test data set. This performance was consistent across joint types, genders, sample processing methods, and various types of PJI, independent of the microbiological results.
Fig. 7.
Proposed mechanism of action for miRNA biomarkers of PJIs in synovial fluid. During PJI, 2 main mechanisms may contribute to the observed increase or decrease in miRNA levels in synovial fluid: (1) the infiltration and activation of specific immune cell types, reflecting innate (granulocyte) and adaptive (T, B, or NK cell) immunity, results in the release of cell type-enriched miRNAs into synovial fluid, and (2) the exposure of joint-tissue cells to cytokines or pathogen-associated molecular patterns activates inflammatory response pathways and results in altered transcription and, consequently, the release of miRNAs into synovial fluid. EV = extracellular vesicle, RISC = RNA-induced silencing complex. Created in BioRender. Hackl, M. (2025) https://BioRender.com/a60j396.
A previous pilot study showed that the expression of miRNA in plasma and tissue changes in hip PJI13. However, the identified plasma miRNAs in that study showed lower effect sizes and higher variability than the synovial fluid miRNAs in the present study, indicating that the response to infection is not sufficient to robustly alter systemic miRNA levels in blood. While the observed number of miRNA responses in periarticular tissue in that study (23 miRNAs) was higher than that in the plasma of the same patients, it was lower than that identified in synovial fluid in the present study (132 miRNAs). This suggests that synovial fluid is the matrix of choice for PJI miRNA biomarker analysis.
These findings may have important clinical implications. First, miRNAs are emerging as promising biomarkers, providing a novel approach for the diagnosis of PJI. They can be quantified in synovial fluid using qPCR, a cost-effective and scalable method that enables near-patient testing and real-time clinical decision-making. Second, the identified miRNA biomarkers reflect the activation of both the innate and adaptive immune systems, making them independent of the infecting pathogen. Third, the combination of multiple miRNAs that represent distinct aspects of immune system activation enhances test specificity in distinguishing PJI from noninfectious chronic inflammation. Thus, synovial fluid miRNAs, when paired with sensitive microbial DNA detection methods such as NGS or PCR, may improve diagnostic specificity by indicating innate immune system activation in response to infection. This is particularly valuable for diagnosing culture-negative infections, which account for up to 45% of PJIs and remain a considerable diagnostic challenge19-21.
Early and accurate preoperative diagnosis of PJI, particularly in chronic cases, remains challenging. Commonly available parameters, including clinical features, serum biomarkers, and radiographic imaging, often fail to provide definitive results22,23. Joint aspiration, a highly recommended step in suspected PJI cases, can provide additional diagnostic information3,4. However, synovial tests such as the leukocyte count and neutrophil percentage often yield inconclusive results. Some synovial biomarkers offer additional diagnostic value, but their sensitivity and specificity remain insufficient for a reliable diagnosis5,6. Alpha-defensin (Synovasure; Zimmer Biomet) remains the only new diagnostic test listed in the current guidelines, both in the 2018 International Consensus Meeting infection criteria4 and in the European Bone and Joint Infection Society (EBJIS) definition of PJI3. Although this test is a rapid and highly specific synovial fluid test for PJI, there are limitations regarding its sensitivity, and it is mainly used as a confirmatory test rather than as a screening test24.
The generalizability of the present study may be limited, as the study was conducted at a single center. Additionally, the analysis focused on clearly septic and aseptic cases. Further validation in patients with rheumatological diseases or unclear preoperative diagnostic results leading to unexpected positive intraoperative cultures (UPIC) or negative intraoperative cultures (UNIC) is needed. Moreover, investigation is needed regarding the specificity of the model in cohorts in which other postoperative TJR complications, such as metallosis, are also present.
Conclusions
Our results from an untargeted discovery and blinded validation study involving a large cohort of patients with septic and aseptic knees and hips led to the identification of novel PJI biomarkers. These biomarkers achieved excellent diagnostic performances, with an AUC of >0.98 for individual miRNAs and an AUC of up to 1.0 for combinations of 3 to 4 miRNAs. This study provides strong preliminary evidence supporting the clinical utility of synovial fluid miRNAs as sensitive and specific biomarkers for diagnosing PJI. Future studies will be conducted to confirm the reproducibility of these findings across different centers and patient populations.
Appendix
Supporting material provided by the authors is posted with the online version of this article as a data supplement at jbjs.org (http://links.lww.com/JBJS/I837).
Footnotes
Investigation performed at the Michael Ogon Laboratory for Orthopaedic Research, Orthopaedic Hospital Speising, Vienna, Austria
A commentary by Mansour Abolghasemian, MD, FAAOS, and Elissa Rennert May, MD, MSc, FRCPC, is linked to the online version of this article.
Disclosure: No external funding was received for this work. T.L.K., A.B.D., and M.H. are employees of TAmiRNA, a microRNA biomarker company. The Disclosure of Potential Conflicts of Interest forms are provided with the online version of the article (http://links.lww.com/JBJS/I836).
Contributor Information
Bernhard J.H. Frank, Email: bernhard@franknet.at.
Teresa L. Krammer, Email: teresa.krammer@tamirna.com.
Jakob Pfeiffer-Vogl, Email: jakobp-v@hotmail.com.
Thi Thu Huong Nguyen, Email: nguyent@gmx.at.
Andreas B. Diendorfer, Email: andreas.diendorfer@tamirna.com.
Matthias Hackl, Email: matthias.hackl@tamirna.com.
References
- 1.Hipfl C, Mooij W, Perka C, Hardt S, Wassilew GI. Unexpected low-grade infections in revision hip arthroplasty for aseptic loosening: a single-institution experience of 274 hips. Bone Joint J. 2021. Jun;103-B(6):1070-7. [DOI] [PubMed] [Google Scholar]
- 2.Henderson RA, Austin MS. Management of Periprosthetic Joint Infection: The More We Learn, the Less We Know. J Arthroplasty. 2017. Jul;32(7):2056-9. [DOI] [PubMed] [Google Scholar]
- 3.McNally M, Sousa R, Wouthuyzen-Bakker M, Chen AF, Soriano A, Vogely HC, Clauss M, Higuera CA, Trebše R. The EBJIS definition of periprosthetic joint infection. Bone Joint J. 2021. Jan;103-B(1):18-25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Parvizi J, Tan TL, Goswami K, Higuera C, Della Valle C, Chen AF, Shohat N. The 2018 Definition of Periprosthetic Hip and Knee Infection: An Evidence-Based and Validated Criteria. J Arthroplasty. 2018. May;33(5):1309-1314.e2. [DOI] [PubMed] [Google Scholar]
- 5.Yilmaz MK, Abbaszadeh A, Tarabichi S, Azboy I, Parvizi J. Diagnosis of Periprosthetic Joint Infection: The Utility of Biomarkers in 2023. Antibiotics (Basel). 2023. Jun 15;12(6):1054. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Sharma K, Ivy M, Block DR, Abdel MP, Hanssen AD, Beauchamp C, Perry KI, Rosemark CL, Greenwood-Quaintance KE, Mandrekar J, Patel R. Comparative analysis of 23 synovial fluid biomarkers for hip and knee periprosthetic joint infection detection. J Orthop Res. 2020. Dec;38(12):2664-74. [DOI] [PubMed] [Google Scholar]
- 7.Nahand JS, Taghizadeh-Boroujeni S, Karimzadeh M, Borran S, Pourhanifeh MH, Moghoofei M, Bokharaei-Salim F, Karampoor S, Jafari A, Asemi Z, Tbibzadeh A, Namdar A, Mirzaei H. microRNAs: New prognostic, diagnostic, and therapeutic biomarkers in cervical cancer. J Cell Physiol. 2019. Aug;234(10):17064-99. [DOI] [PubMed] [Google Scholar]
- 8.Wang J, Chen J, Sen S. MicroRNA as Biomarkers and Diagnostics. J Cell Physiol. 2016. Jan;231(1):25-30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Larrea E, Sole C, Manterola L, Goicoechea I, Armesto M, Arestin M, Caffarel MM, Araujo AM, Araiz M, Fernandez-Mercado M, Lawrie CH. New Concepts in Cancer Biomarkers: Circulating miRNAs in Liquid Biopsies. Int J Mol Sci. 2016. Apr 27;17(5):627. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Starlinger P, Hackl H, Pereyra D, Skalicky S, Geiger E, Finsterbusch M, Tamandl D, Brostjan C, Grünberger T, Hackl M, Assinger A. Predicting Postoperative Liver Dysfunction Based on Blood-Derived MicroRNA Signatures. Hepatology. 2019. Jun;69(6):2636-51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Khamina K, Diendorfer AB, Skalicky S, Weigl M, Pultar M, Krammer TL, Fournier CA, Schofield AL, Otto C, Smith AT, Buchtele N, Schoergenhofer C, Jilma B, Frank BJH, Hofstaetter JG, Grillari R, Grillari J, Ruprecht K, Goldring CE, Rehrauer H, Glaab WE, Hackl M. A MicroRNA Next-Generation-Sequencing Discovery Assay (miND) for Genome-Scale Analysis and Absolute Quantitation of Circulating MicroRNA Biomarkers. Int J Mol Sci. 2022. Jan 22;23(3):1226. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Gutmann C, Khamina K, Theofilatos K, Diendorfer AB, Burnap SA, Nabeebaccus A, Fish M, McPhail MJW, O’Gallagher K, Schmidt LE, Cassel C, Auzinger G, Napoli S, Mujib SF, Trovato F, Sanderson B, Merrick B, Roy R, Edgeworth JD, Shah AM, Hayday AC, Traby L, Hackl M, Eichinger S, Shankar-Hari M, Mayr M. Association of cardiometabolic microRNAs with COVID-19 severity and mortality. Cardiovasc Res. 2022. Jan 29;118(2):461-74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Paksoy A, Meller S, Schwotzer F, Moroder P, Trampuz A, Imiolczyk JP, Perka C, Hackl M, Plachel F, Akgün D. MicroRNA expression analysis in peripheral blood and soft-tissue of patients with periprosthetic hip infection. Bone Jt Open. 2024. Jun 6;5(6):479-88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Frank BJH, Simon S, Aichmair A, Dominkus M, Hofstaetter JG. Clinical impact of microbiological results in two-stage revision arthroplasty with spacer exchange. Arch Orthop Trauma Surg. 2023. Aug;143(8):4741-54. [DOI] [PubMed] [Google Scholar]
- 15.Diendorfer A, Khamina K, Pultar M, Hackl M. miND (miRNA NGS Discovery pipeline): a small RNA-seq analysis pipeline and report generator for microRNA biomarker discovery studies. 2022. Accessed 2025 May 16. https://f1000research.com/articles/11-233
- 16.Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010. Jan 1;26(1):139-40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lizio M, Harshbarger J, Shimoji H, Severin J, Kasukawa T, Sahin S, Abugessaisa I, Fukuda S, Hori F, Ishikawa-Kato S, Mungall CJ, Arner E, Baillie JK, Bertin N, Bono H, de Hoon M, Diehl AD, Dimont E, Freeman TC, Fujieda K, Hide W, Kaliyaperumal R, Katayama T, Lassmann T, Meehan TF, Nishikata K, Ono H, Rehli M, Sandelin A, Schultes EA, ’t Hoen PA, Tatum Z, Thompson M, Toyoda T, Wright DW, Daub CO, Itoh M, Carninci P, Hayashizaki Y, Forrest AR, Kawaji H; FANTOM consortium. Gateways to the FANTOM5 promoter level mammalian expression atlas. Genome Biol. 2015. Jan 5;16(1):22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Cao Y, Tang S, Nie X, Zhou Z, Ruan G, Han W, Zhu Z, Ding C. Decreased miR-214-3p activates NF-κB pathway and aggravates osteoarthritis progression. EBioMedicine. 2021. Mar;65:103283. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Patel R. Periprosthetic Joint Infection. N Engl J Med. 2023. Jan 19;388(3):251-62. [DOI] [PubMed] [Google Scholar]
- 20.Goh GS, Parvizi J. Diagnosis and Treatment of Culture-Negative Periprosthetic Joint Infection. J Arthroplasty. 2022. Aug;37(8):1488-93. [DOI] [PubMed] [Google Scholar]
- 21.Kalbian I, Park JW, Goswami K, Lee YK, Parvizi J, Koo KH. Culture-negative periprosthetic joint infection: prevalence, aetiology, evaluation, recommendations, and treatment. Int Orthop. 2020. Jul;44(7):1255-61. [DOI] [PubMed] [Google Scholar]
- 22.Sigmund IK, Puchner SE, Windhager R. Serum Inflammatory Biomarkers in the Diagnosis of Periprosthetic Joint Infections. Biomedicines. 2021. Sep 1;9(9):1128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Abdel Karim M, Andrawis J, Bengoa F, Bracho C, Compagnoni R, Cross M, Danoff J, Della Valle CJ, Foguet P, Fraguas T, Gehrke T, Goswami K, Guerra E, Ha YC, Klaber I, Komnos G, Lachiewicz P, Lausmann C, Levine B, Leyton-Mange A, McArthur BA, Mihalič R, Neyt J, Nuñez J, Nunziato C, Parvizi J, Perka C, Reisener MJ, Rocha CH, Schweitzer D, Shivji F, Shohat N, Sierra RJ, Suleiman L, Tan TL, Vasquez J, Ward D, Wolf M, Zahar A. Hip and Knee Section, Diagnosis, Algorithm: Proceedings of International Consensus on Orthopedic Infections. J Arthroplasty. 2019. Feb;34(2S):S339-50. [DOI] [PubMed] [Google Scholar]
- 24.Renz N, Yermak K, Perka C, Trampuz A. Alpha Defensin Lateral Flow Test for Diagnosis of Periprosthetic Joint Infection: Not a Screening but a Confirmatory Test. J Bone Joint Surg Am. 2018. May 2;100(9):742-50. [DOI] [PubMed] [Google Scholar]