Glycoprotein non-metastatic melanoma protein B is a potential biomarker for arthroplasty aseptic looseningDownload PDF Download PDF ArticleOpen accessPublished: 12 September 2025Patrik Schadzek1,7 na1,Alexander Derksen2 na1,Wiebke Behrens3,7 na1,Maike Kosanke4,Oliver Dittrich-Breiholz4,Anika Hamm1,7,Kirsten Elger1,7,Yvonne Roger1,7,Ines Yang3,7,Meike Stiesch3,7,Yvonne Noll2,Marco Haertlé2,Lars-René Tuecking2,Christina Stukenborg-Colsman2,Henning Windhagen2,7,Doan Duy Hai Tran5,8,Anette Melk6 &…Andrea Hoffmann1,7 Scientific Reports volume 15, Article number: 32419 (2025) Cite this articleSubjectsBiomarkersDiseasesHealth careMedical researchMolecular biologyAbstractEndoprosthesis loosening is the major cause of arthroplasty failure. Currently, radiography, histo-pathological classification of the periprosthetic membrane, and microbiological examination are used retrospectively for diagnosis. Prospective options for early diagnosis are not available. The study presented here aimed to identify (prospective) molecular biomarkers of implant loosening in tissue samples from patients. Four patient cohorts (primary and revision arthroplasty due to aseptic loosening of hip or knee) were defined. Aseptic loosening of implants was assessed by the standardised approach of the Knee Society Total Knee Arthroplasty Roentgenographic Evaluation and Scoring System. Synovial fluid, bone marrow, and blood were collected from 96 patients. Mesenchymal stromal cells (BM-MSCs) were isolated from the bone marrow. Bulk RNA-sequencing of 28 samples (including a fifth patient cohort with knee arthrofibrosis as aseptic cause for revision surgery) showed that Glycoprotein Non-Metastatic Melanoma Protein B mRNA (GPNMB) was significantly upregulated in BM-MSCs derived from revision patients compared to patients with primary implantations. Elevated GPNMB mRNA levels were confirmed with qRT-PCR. Synovial fluid plasma study by ELISA revealed increased levels of GPNMB protein in patients undergoing revision surgery compared to patients undergoing primary arthroplasty. GPNMB concentration in synovial fluid plasma, which can be obtained non-invasively, could be a potential biomarker for early detection of implant loosening, ahead of current diagnostic procedures.IntroductionHip and knee endoprostheses (THA, TKA) are major implant types in orthopaedic and trauma care. In 2022, 177,826 hip and 137,030 knee primary surgeries were performed in Germany, and additionally 18,145 hip/14,379 knee revision surgeries1.Loosening and the resulting implant failures of endoprostheses pose a significant challenge2. For the diagnosis of looseness, either a skeletal scintigraphy or radiological images in at least two planes are performed3,4. In this context, radiolucent lines in the latter can indicate loosening2. If these progress over time, this indicates a nearly certain sign of implant failure2. With the help of radiographic evaluation systems, particularly the Knee Society Radiographic Evaluation System (KSRESS), signs of loosening in knee prostheses are assessed uniformly4,5. Aseptic loosening is responsible for 55% of cases, septic loosening due to infections accounts for about 15%, and other reasons like periprosthetic fractures or implant dislocations make up the remaining 30% 1,6. Due to the high proportion of aseptic loosening, the current study focuses on this pathology.Aseptic loosening is characterised by osteolysis due to increased osteoclast activity, resulting in insufficient osseointegration of the endoprostheses. Moreover, the process includes a reparative and inflammatory response of the immune system7finally resulting in fibrosis. This is histo-pathologically classified as type IV implant failure according to Krenn and Morawietz8. A major trigger seems to come from wear particles released during motion of the joint, histo-pathologically classified as type I implant failure. Type I and type IV failure occur in combination in most cases of joint arthroplasty aseptic loosening.Early detection of implant loosening is currently not possible. Thus, there is an urgent need for novel diagnostic parameters including individual biomarkers or complex signatures. Following this rationale, the study presented here focused on the unbiased transcriptome analysis of mesenchymal stromal cells from bone marrow (BM-MSCs) as one prominent type of regenerative cells for implant-tissue interaction. BM-MSCs of patients undergoing primary or revision surgery for hip or knee endoprostheses were analysed (in total 28 samples). Subsequent confirmation of potential biomarker candidates was performed in body fluid plasma on protein level to pinpoint protein biomarker candidates for revision arthroplasties, based on material from 96 patients.Materials and methodsPatient cohortsClinical informationThis study included 96 patients who received primary hip or knee arthroplasty due to osteoarthritis, or revision hip or knee surgeries due to aseptic implant loosening as diagnosed by roentgenographic assessment (cf. below) between 2021 and 2024 in the study hospital. Exclusion criteria were infections with hepatitis viruses or MRSA, depressions, strong smoking, and tumour diagnoses. Intraoperatively, synovial fluid, bone marrow, and blood were targeted in all patients but did not result in successful procuration in all cases. Condensed patient information is displayed in Table 1 (for more details, please refer to Suppl. Table 1). Five cohorts were defined from a subset of the study group for RNA-sequencing of BM-MSCs: primary hip (PH, 4 cases for the pilot run, 12 cases in total, colour code: light grey), revision hip (RH, 2 cases for the pilot run, 2 cases in total, colour code: dark grey), primary knee (PK, 2 cases for the second run, colour code: blue), revision knee (RK, 2 cases for the pilot run, 7 cases in total, colour code: violet), and revision knee due to arthrofibrosis (RK AF, 5 cases for the second run, colour code: pink). The four major cohorts (arthrofibrosis exempted) with inclusion of more patients were used for biochemical assays (ELISA, total protein).The BMI of the cohort with primary hip implantations was 28.0 ± 3.9 (range 22.1–34.9) for 23 patients (7 male, 16 female). The BMI of the cohort with primary knee implantations was 31.0 ± 5.2 (range 23.4–39.3) for 20 patients (10 male, 10 female). The BMI of the cohort with revision hip implantations was 27.2 ± 6.8 (20.0-51.9) for 25 patients (13 male, 12 female). The BMI of the two cohorts with revision knee implantations (arthrofibrosis included) was 31.8 ± 6.6 (21.3–54.9) for 28 patients (12 male, 16 female). In summary, the body mass index of patients with knee surgeries was higher than with hip surgeries.Table 1 Patient information.Full size tableInformed consent was received prior to participation. For the photo in Fig. 1B, written informed consent was received for the use of the photographs and publication. All research was performed in accordance with relevant guidelines/regulations.Sex as a biological variableOur study examined male and female patients. No significant differences in overall expression patterns were observed from RNA sequencing data in comparisons of all 28 female and male patients (PERMANOVA, p > 0.05), both with and without control for other technical and biological variables. Nevertheless, tested groups were either balanced for sexes, or sex was included as potential co-variate in RNA-sequencing analyses when testing primary and revision arthroplasty samples of hip or knee for overall differences in expression patterns (PERMANOVA) and differentially expressed genes (DESeq2).Age as a biological variableAll 28 patients whose BM-MSCs were used for RNA-sequencing were aged 60 years or above, with the exception of one younger person. The 87 patients whose body fluids were assessed by ELISA were aged 54 years or above, with the exception of 6 younger donors. No significant differences in overall expression patterns of RNA-sequencing data were observed in relation to age (PERMANOVA, p > 0.05). Tested groups were either balanced for age, or age was included as potential co-variate in analyses of overall differences in expression patterns (PERMANOVA) and of differentially expressed genes (DESeq2) when testing for differences in primary and revision arthroplasty samples of hip or knee.Roentgenographic assessment of implant looseningCharacteristic structural changes in osteoarthritis, their severity and their accompanying phenomena can be easily detected by X-rays, as radiolucent lines at the boundary between the prosthesis and the neighbouring bone or bone cement interface, as well as between the bone cement and the bone9. The radiolucent lines were assessed using the standardised approach of the KSRESS. The assessment was conducted with standardised X-ray pictures in the anteroposterior and axial positions of the knee (cf. Suppl. Fig. S1). All spots that were transparent to X-rays and had a diameter larger than 1 mm were noted and documented based on their corresponding zones9. Consequently, the overall defect length was determined by adding the measurements of each individual zone. In the anteroposterior pictures, three additional zones were included for cases with existing stems. This seems rational as evaluating tibial stems necessitates an assessment in imaging planes that are oriented differently10. In order to mitigate distortions caused by magnification effects, the OrthoView program (Jacksonville, FL, USA) was employed to examine and standardise the true magnification of the X-ray pictures.The radiological assessments were carried out separately by three skilled orthopaedic surgeons. A study on intra-observer reliability was conducted, yielding an intraclass correlation coefficient of 0.83, indicating a significant degree of measurement reliability.Biomaterial processingAn anatomical sketch of a hypothetical synovial joint with an implant is shown in Fig. 1A. Bone marrow, synovial fluid and blood were collected into heparin-containing tubes in order to prepare plasma and mononuclear cells (MNCs) from identical material as described in detail below.Fig. 1Tissue sampling. (A) Sagittal section through hypothetical synovial joint with hypothetical implant. (B) Blood sampling was optimised using an adapter (left) combined with a CPT™ tube (right) for density gradient preparation of plasma and MNCs.Full size imageBone marrowBone marrow was extruded from the femur and collected into syringes with heparin (2,500 IU/mL heparin per 10 mL of bone marrow). The bone marrow preparations were further diluted fourfold with phosphate-buffered saline without Ca2+, Mg2+ (PBS) and processed as described11. Bone marrow plasma was collected, thoroughly mixed, and then stored at − 80 °C in three aliquots until analysis. All plasma samples were slightly haemolytic.The MNCs floating on the density gradient were processed into BM-MSCs as described11. For RNA-Seq, BM-MSCs were used in passage 3.Synovial fluid and bloodSynovial fluid and blood (Fig. 1B) were collected in BD Vacutainer® CPT™ mononuclear Cell Preparation Tubes with an adapter. These tubes contain heparin and a density gradient material, hereafter called CPT™ tubes. The use of an adapter helped to minimise haemolysis and thereby notably improved the sample quality. After centrifugation at 1700 x g for 25 min the plasma was recovered and stored at − 80 °C until analysis. The overall degree of haemolysis of synovial fluid from TKA patients was lower than from THA patients. In case of low volumes of synovial fluid, PBS was added into the CPT™ tubes before further processing in the laboratory since the tubes are designed for use with volumes of 8 ml.The MNCs obtained during density gradient centrifugation were resuspended in residual plasma followed by centrifugation at 300 × g for 5 min at room temperature with brakes. The cell pellets were resuspended in freezing medium (95% FCS, 5% DMSO) and stored at − 140 °C in aliquots of 1 mL. For RNA isolation, aliquots were thawed in 9 mL ice-cold macrophage growth medium (M0-GM: RPMI-1640 (Sigma #R0883) with 10% FCS, 2 mM L-glutamine (Sigma G7513), 17.1 mM HEPES, 1% MEM non-essential amino acids (Sigma #M7145), 85.65 µM 2-mercaptoethanol, 100,000 U/L penicillin, 100 mg/L streptomycin, 25 ng/mL recombinant human M-CSF (Miltenyi Biotech #130-096-492, from E. coli)). Cells were incubated on ice for 10 min, then centrifuged at 300 x g for 5 min. The pellets were resuspended in 3 ml M0-GM, transferred into one well each of a 6-well plate and grown to about 90% of cell density until lysis for RNA isolation.RNA isolation, quantification and quality controlRNA was extracted from cell cultures via Trizol reagent (Invitrogen: 1 ml per 25 cm2 culture surface) according to the instructions by the manufacturer. Concentration was measured with a NanoDrop ND-1000 instrument. Samples for RNA-sequencing were subjected to a quality control with Agilent 2100 Bionanalyzer, using the Agilent RNA 6000 Pico Assay. Only samples with RNA integrity number > 7 were processed for RNA-sequencing.RNA-sequencing: library generation, quality control, and quantification300 ng of total RNA per sample were utilized as input for mRNA enrichment procedure with ‘NEBNext® Poly(A) mRNA Magnetic Isolation Module’ (E7490L; New England Biolabs) followed by stranded cDNA library generation using ‘NEBNext® Ultra II Directional RNA Library Prep Kit for Illumina’ (E7760L; New England Biolabs). All steps were performed as recommended in user manualE7760 (Version 1.0_02-2017; NEB) except that all reactions were downscaled to 2/3 of initial volumes.cDNA libraries were barcoded by dual indexing approach, using ‘NEBNext Multiplex Oligos for Illumina – 96 Unique Dual Index Primer Pairs’ (6440S; New England Biolabs). All generated cDNA libraries were amplified with 8 of final pcr.One additional purification step was introduced at the end of the standard procedure, using 1.2x ‘Agencourt® AMPure® XP Beads’ (#A63881; Beckman Coulter, Inc.). Fragment length distribution of individual libraries was monitored using ‘Bioanalyzer High Sensitivity DNA Assay’ (5067 − 4626; Agilent Technologies). The quantification of libraries was performed by use of the ‘Qubit® dsDNA HS Assay Kit’ (Q32854; ThermoFisher Scientific).Library denaturation and sequencing runEqual molar amounts of individually barcoded libraries were pooled for, in total, 2 sequencing runs, in which each analyzed library constituted around 7.8% of overall flowcell / run capacity. The library pool was denatured with NaOH and was finally diluted to 1.8 pM according to the Denature and Dilute Libraries Guide (Document # 15048776 v02; Illumina). 1.3 mL of the denatured pool was loaded on an Illumina NextSeq 550 sequencer using a High Output Flowcell (400 M cluster) for single reads (20024906; Illumina). Sequencing was performed with the following settings: Sequence reads 1 and 2 with 38 bases each; Index reads 1 and 2 with 8 bases each.BCL to FASTQ conversionBCL files were converted to FASTQ files using bcl2fastq Conversion Software version v2.20.0.422 (Illumina).Raw data processing and quality controlRaw data processing was conducted by use of nfcore/rnaseq (version 1.4.2), which is a bioinformatics best-practice analysis pipeline used for RNA sequencing data at the National Genomics Infrastructure at SciLifeLab Stockholm, Sweden. The pipeline uses Nextflow, a bioinformatics workflow tool. It pre-processes raw data from FastQ inputs, aligns the reads and performs extensive quality-control on the results. The genome reference and annotation data were taken from GENCODE.org (Homo sapiens: GRCh38.p13; release 34).Analyses of gene expression patterns based on RNA-sequencing dataIndividual genes were tested for differential abundances in different study groups with DESeq2 using the Wald test (version 1.32.0)12. The DESeq2 design formula contained the sequencing run as possible technical co-variate (set of 28 samples was distributed over two independent sequencing runs), and sex and age in years as non-technical co-variates to control for unwanted variation for the large cohort of 28 samples while this adjustment was not necessary for the balanced pilot discovery cohort of 8 samples. Analysing the large cohort of 28 samples, all genes with at least 10 reads in two samples were included in the DESeq2 analyses.For heatmap visualisation the software Qlucore Omics Explorer version 3.9 (Qlucore, Lund, Sweden) was used. The genes in the heatmap were sorted by their statistics (Welch´s t-test comparison between primary and revision cohorts).Gene set enrichment analysis was performed with the ShinyGO 0.80 graphical gene-set enrichment tool13with differentially expressed genes from DESeq2 analysis as input, including all genes with significant different counts indicated by adjusted p values 1000) were KLF2, BHLHE40, NR4A1, PPP1R15A, IER2, CTSK, SOCS3, ZFP36, JUNB, DUSP1, GPNMB, and JUN. Secondly, the occurrence of the encoded proteins in soluble forms (either secreted or by shedding of an extracellular domain) that can be quantitated by commercially available ELISA kits would be more advantageous than the choice of, e.g., intracellular transcription factors. This additional criterium was fulfilled by GPNMB within the 12 genes.GPNMB, also called osteoactivin, is an ideal target matching premises for biomarkers. It is a type I transmembrane glycoprotein whose extracellular domain can be shedded from the cell surface after proteolytic cleavage and is secreted into body fluids18. Hence, we examined the soluble form of GPNMB in bone marrow and blood plasma of 26 or 79 patients, respectively. The concentration of soluble GPNMB determined by ELISA and total protein concentration allowed to calculate relative GPNMB concentrations. Figures 3 and 4 display the results for the four patient cohorts for bone marrow plasma and blood plasma, respectively. Overall, the relative concentrations are about tenfold higher in bone marrow plasma than in blood plasma. However, in both body fluids no statistically significant differences were detected between the four patient cohorts which is substantiated by the Receiver Operating Characteristic (ROC) curve analyses (panels B) in Figs. 3 and 4 which indicate that the GPNMB concentrations do not perform much better than a random selection. The red dashed line in the ROC curves (AUC of 0.5) indicates a random process.Fig. 3Relative levels of GPNMB in bone marrow plasma (BM). The absolute GPNMB levels detected in ELISA were normalised by the total protein concentration of the individual samples. (A) Relative GPNMB levels displayed via four patient cohorts. The black bar represents the median value. ns: not significant.(B) ROC curves (Receiver Operating Characteristics) provide an overview of the diagnostic quality based on bone marrow plasma levels of GPNMB measured by enzyme-linked immunosorbent assay for prediction of a revision sample. AUC: Area under the curve. (C) Relative GPNMB levels for each individual patient.Full size imageFig. 4Relative levels of GPNMB in blood plasma (BP). The absolute GPNMB levels detected in ELISA were normalised by the total protein concentration of the individual samples. (A) Relative GPNMB levels displayed via four patient cohorts. The black bar represents the median value. ns: not significant. (B) Receiver operating characteristic curve (ROC) for prediction of a revision sample based on blood plasma levels of GPNMB measured by enzyme-linked immunosorbent assay. AUC: Area under the curve. (C) Relative GPNMB levels for each individual patient.Full size imageRelative GPNMB levels are significantly elevated in synovial fluid plasma at the time of knee revision arthroplastyThe missing correlation of soluble GPNMB in bone marrow plasma with the RNA-seq data from BM-MSCs prompted additional considerations. Figure 1A visualises that both THA and TKA not only directly contact the bone marrow within the bony tissue but also the synovial fluid within the joint cavity, both resulting in extensive tissue-implant interfaces. We therefore decided to measure GPNMB levels in the synovial fluid samples as well. Figure 5A demonstrates statistically significant differences for revision knee vs. primary knee arthroplasty (p = 0.0002) while no significant differences were documented for revision hip vs. primary hip arthroplasty. The Receiver Operating Characteristic curve analysis in Fig. 5B indicates that the GPNMB concentration in synovial fluid of knee samples is able to predict the cohort membership with a high sensitivity and specificity. This shows the diagnostic quality of GPNMB as biomarker for knee arthroplasty (AUC = 0.995), performing better than a random selection. Contamination with blood, observed in a number of samples, is unlikely to interfere with the measured values due to the notably lower relative GPNMB concentrations in blood plasma than in synovial fluid plasma. This interpretation was confirmed by artificial admixture of blood into blood-free synovial fluid samples which resulted in comparable results, confirming absence of interference with ELISA performance by blood or haemolysis (data not shown).The Spearman correlation was r = 0.4649 for the comparison between the relative GPNMB concentration in the synovial fluid and the summed roentgenographic defect length, indicating a moderate positive monotonic relationship between these variables, with statistical significance (p = 0.0449). No correlation was revealed with other roentgenographic parameters. For the knee revision surgeries, a moderate but significant correlation with the body mass index of the patients was found (r = 0.5607, p = 0.0125).Due to the comparably low number of biosamples and the fact that most aseptic revision surgeries are classified histo-pathologically as mixtures of type I + type IV according to Krenn and Morawietz, not as distinct types, a potential correlation analysis between the relative GPNMB levels in synovial fluid and histo-pathological grading was not attempted.Since GPNMB and SDC4 interact with each other as ligand and receptor19 the levels of SDC4 were determined in synovial fluid and bone marrow. The data is shown in Suppl. Fig. S2 and Suppl. Fig. S3. No statistically significant difference could be revealed between the four patient cohorts. Moreover, no correlation between relative GPNMB and relative SDC4 levels in both body fluids could be revealed.Fig. 5Relative levels of GPNMB in synovial fluid plasma (SF). The absolute GPNMB levels detected in ELISA were normalised by the total protein concentration of the individual samples. (A) Relative GPNMB levels displayed via four patient cohorts. The black bar represents the median value. ***p