Machine learning-based analysis reveals potential diagnostic gene biomarkers associated with immune infiltration in patients with osteoarthritis

doi:10.21203/rs.3.rs-7616797/v1

Download PDF

Research Article

Machine learning-based analysis reveals potential diagnostic gene biomarkers associated with immune infiltration in patients with osteoarthritis

https://doi.org/10.21203/rs.3.rs-7616797/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Objective

We aimed to find and validate reliable diagnostic markers in osteoarthritis (OA) and to explore the links with immunophenotypes.

Materials and Methods

For this study, we screened the Gene Expression Omnibus (GEO) dataset of three OA synovial samples for genes that differed between the disease and normal groups. Subsequently, multiple machine learning algorithms as well as screening core genes as diagnostic markers were employed. The diagnostic efficiency of the training and test sets was verified by ROC curves, and the correlation of core diagnostic genes with the immune microenvironment was compared by the CIBERSORT algorithm. Furthermore, in vitro and in vivo experiments were performed to validate the regulatory roles of selected genes in OA progression. Quantitative PCR and ELISA were used to detect gene expression and inflammatory cytokines in OA-simulated synoviocytes, while animal models were employed to assess joint pathology, synovitis, and cartilage degradation.

Results

We identified 81 DEGs. the results of LASSO and SVM-RFE identified 13 key diagnostic genes. the ROC analysis confirmed the diagnostic value of these 13 genes for OA. Immunocell infiltration analysis revealed activation of CD4 + T cells as well as T helper cells in synovial tissue, as well as downregulation of MHC class I and enhanced T cell co-repressor signaling, which may correlate with SCN4B expression. DLX2 knockdown and SCN4B overexpression significantly altered inflammatory mediator levels and cartilage catabolic markers in vitro, and ameliorated OA-related histopathological changes in vivo, confirming their functional involvement in disease progression.

Conclusion

Overall, this study identified 13 diagnostic biomarkers for OA, providing potential molecular targets for optimal treatment of OA. In particular, DLX2 and SCN4B may represent novel regulatory hubs involved in inflammation and cartilage degradation, offering new insights into OA pathogenesis and therapeutic strategies.

Osteoarthritis

Machine learning

Diagnostic biomarker

Immunocell infiltration

Osteoarthritis (OA) is a common degenerative joint disease. It is a complex disease characterized by cartilage destruction, joint stiffness and pain(1). Over 30 million people are affected in the United States alone, and the incidence is increasing(2, 3). Presently, the primary method of diagnosing OA is clinical evaluation by a physician, followed by radiography, magnetic resonance imaging (MRI), and in some cases, arthroscopy(4). Hence, there is an urgent need to break through the molecular pathways of OA development and identify biomarkers for diagnosis and prognosis.

Over recent years, high-throughput sequencing combined with bioinformatics analysis techniques with machine learning have been used to identify novel genes associated with a variety of diseases that may be used as diagnostic and prognostic markers(5, 6). With the rapid development and widespread application of microarray sequencing, bioinformatics has been increasingly applied to identify new biomarkers of various diseases, explore the epigenetic changes involved, and reveal the underlying molecular mechanisms. Many researchers have explored. Gong N et al. used bioinformatics analysis to identify and verify ferroptosis-related genes in osteoarthritis synovial tissue, and the AUCs of the identified ferroptosis-related central genes were all greater than 0.75 (7). Peng J et al. identified the effectiveness of 5 Hub genes in OA diagnosis, and evaluated the AUC values of VEGFA, FYN, GREM1, NRP1, and IL6R by ROC curves as 0.947, 0.795, 0.768, 0.775, and 0.776, respectively (8). Despite its excellent diagnostic performance, better methods are needed to identify optimal diagnostic markers. Recently, high-throughput sequencing combined with multiple machine learning modalities has been used to identify novel genes associated with diseases that may be used as diagnostic and prognostic markers for diseases(6, 9).

OA is not an autoimmune disease but has a plethora of inflammatory etiologies(10, 11). For example, in osteoarthritis, proinflammatory cytokines promote chondrocyte apoptosis(12) and suppression of inflammation reduces cartilage damage in osteoarthritis(13). The synovium in osteoarthritis contains a variety of immune cells, such as macrophage-like synoviocytes, fibroblast-like synoviocytes, plasma cells, and lymphocytes,(14) which produce inflammatory cytokines and lead to synovial matrix metalloproteinases (MMPs) that promote cartilage destruction in the development of meningitis(15). Exploring the immune basis of OA is, therefore, one of the valuable therapeutic strategies.

We acquired the osteoarthritis microarray dataset from the GEO database. Differentially expressed gene (DEG) analysis was used to compare osteoarthritic synovial tissue with normal tissue, followed by filtering and identification of osteoarthritis diagnostic biomarkers by multiple machine learning techniques, and quantification of immune cells and immune pathways in osteoarthritic synovial tissue and normal tissue using the ssGSEA algorithm. We identified and validated 13 diagnostic genes and provided an opportunity to investigate the association between these biomarkers and immune cells and immune pathways.

Data Acquisition

Publicly available gene-expression datasets related to osteoarthritis (OA) were systematically screened in the Gene Expression Omnibus (GEO) database. To ensure analytical consistency and biological relevance, datasets were included according to the following criteria: (1) samples must contain human osteoarthritic synovial tissue or cartilage, enabling direct comparison between OA and non-OA conditions; (2) datasets must include a sufficient sample size (≥ 20 total samples) to ensure statistical power; (3) transcriptomic profiling must be generated on commonly used microarray platforms (GPL570 or GPL6244), allowing dataset integration and cross-validation; and (4) datasets must contain both OA patients and healthy controls, ensuring appropriate group comparisons. OA patients' mRNA expression profiles were gathered from three GEO databases: GSE12021, GSE143514, and GSE206848. All three datasets were gathered from OA patients' synovial membranes. A dataset was integrated using GSE143514 and GSE206848 via the combat algorithm in the sva package(16) was used as a training cohort for variable screening and model training. GSE12021 was used as an external validation dataset for the model. All datasets were log2 standardized.

DEG identification and functional enrichment analysis

The differential expression analysis in this work was carried out using the R package "limma"(17). DEGs were screened at a P < 0.05 threshold to avoid omission, and DEG efficiency was measured using PCA. Subsequently, functional enrichment analysis was performed for DEGs including GO, KEGG and DO using the R package "clusterProfiler"(18), and pathways with FDR < 0.05 were considered significant. In addition, we used a technique called gene set enrichment analysis (GSEA)(19). Gene sets were considered highly enriched if P < 0.05 and FDR < 0.05 were satisfied.

Robust prediction models built by machine learning methods

The R package glmnet and e1071 was used to build the machine learning model. First, to filter the key DEGs, LASSO regression (nfold = 5, type.measure = "class") and SVM (number = 20) were used to the entire dataset. As a result, the crossover genes identified by the research were regarded as the important DEGs linked with OA and were utilized to build and train prediction models. We calculated the classifier's effectiveness using ROC curves and confusion matrices. Ultimately, we considered the 13 crossover genes as the best classifiers for constructing OA prediction models and applied them to an external validation cohort to evaluate the capability..

Determination of immune cell patterns in the microenvironment

GEO samples were analyzed for immune cell components and immune pathways using ssGSEA(20), and the list of these gene sets was obtained from a previous literature(21). In turn, we evaluated the association between these immune cells and pathways and the expression of key diagnostic genes.

Cell culture and OA model induction in vitro

Primary human synoviocytes were obtained from ScienCell (Carlsbad, CA, USA) and cultured in DMEM/F12 supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin at 37°C in a 5% CO₂ incubator. For experiments, cells were seeded at 5 × 10⁵ cells per well in 6-well plates for protein and RNA extraction and at 2 × 10⁵ cells per well in 12-well plates for transfection assays. To establish an OA-like inflammatory environment, cells were treated with 10 ng/mL recombinant human IL-1β (PeproTech, Cat. No. 200-01B) for 24 hours prior to downstream assays.

Gene knockdown and overexpression

Three small interfering RNAs (siRNAs) targeting human DLX2 and a scrambled negative control siRNA were synthesized by RiboBio (Guangzhou, China). The sequences are listed as follow.

Name	Sequence
siDLX2-1	5′-GCCTGAAATTCGGATAGTGAA-3′
siDLX2-2	5′-CGCACCATCTACTCCAGTTTC-3′
siDLX2-3	5′-TGATATGCACTCGACCCAGAT-3′

Transfection was performed using Lipofectamine 3000 (Thermo Fisher Scientific, Cat. No. L3000001) according to the manufacturer’s protocol. For siRNA knockdown, cells were transfected with 50 nM siRNA in 12-well plates (2 × 10⁵ cells/well). For plasmid overexpression, the full-length SCN4B coding sequence used for cloning is provided in Supplementary Information SI1, and the complete plasmid map of the OE-SCN4B construct is shown in Supplementary Figure S1 (SI2).

The SCN4B CDS was inserted into the pcDNA3.1(+) vector (Invitrogen, Cat. No. V79020), and 1 µg plasmid DNA per well (6-well plate) was transfected. Knockdown/overexpression efficiency was validated by RT-qPCR 48 h post-transfection. All transfection experiments were performed in biological triplicate (n = 3).

Quantitative real-time PCR (RT-qPCR)

Total RNA was extracted from cells or tissues using TRIzol Reagent (Invitrogen, Cat. No. 15596026), and reverse-transcribed using the PrimeScript RT reagent kit (Takara, Cat. No. RR037A). RT-qPCR was performed using SYBR Green Master Mix (Yeasen, Hieff™ qPCR SYBR Green Master Mix, Cat. No. 11184ES) on a QuantStudio 5 Real-Time PCR System (Applied Biosystems; Cat. No. A28574/A28568). Each reaction contained 10 ng cDNA in a 20 µL system, and all samples were run in technical triplicate.

The detailed PCR cycling conditions, reaction composition, and melting curve analysis are provided in Supplementary Information SI3. All qPCR primer sets generated a single, sharp melting-curve peak, confirming amplification specificity (Supplementary Fig. S2). Amplification efficiencies for all genes, calculated using a 5-point 10-fold serial dilution, ranged from 97.8% to 102.0%, with R² values ≥ 0.996, as summarized in Supplementary Table S1.

Relative mRNA expression levels were calculated using the 2^−ΔΔCt method, with GAPDH serving as the internal reference gene. Primer sequences used for qPCR are listed below.

Name	Primer sequence: forward primers (F), reverse primers (R)
CES3	Forward: 5′-CTGAGATGGTGCAGTGCCTTCA-3′ Reverse: 5′-GGAAGACAGTGCCATCAACGGT-3′
SCN4B	Forward: 5′-GACCAGGAGGAGGAGGACGA-3′ Reverse: 5′-CTGCTGCTGCTGCTGCTGC-3′
GNC8	Forward: 5′-CCAGCAGCTGACCAGCAGGAG-3′ Reverse: 5′-GCTGCTGCTGCTGCTGCTGC-3′
DLX2	Forward: 5′-TACTCCGCCAAGAGCAGCTATG-3′ Reverse: 5′-CGAATTTCAGGCTCAAGGTCCTC-3′
SIRPB2	Forward: 5′-GGATGAAGGCACCTCAGTGCTT-3′ Reverse: 5′-GTCTCCAAGCACTGTGCAGTTC-3′
EPHX1	Forward: 5′-GTTTTCCACCTGGACCAATACGG-3′ Reverse: 5′-TGGTGCCTGTTGTCCAGTAGAG-3′
DPEP3	Forward: 5′-AGGAGCAGGAGGAGCAGGAG-3′ Reverse: 5′-CTGCTGCTGCTGCTGCTGC-3′
ABCA3	Forward: 5′-ACGAGGAGGAGGAGGAGGAG-3′ Reverse: 5′-GTCTCCAAGCACTGTGCAGTTC-3′
VPREB3	Forward: 5′-ACCATCAGGGACTACGGTGTGT-3′ Reverse: 5′-CTCATCCTTGGCTGCCGAGAAT-3′
MMP13	Forward: 5′-CCTTGATGCCATTACCAGTCTCC-3′ Reverse: 5′-AAACAGCTCCGCATCAACCTGC-3′
ADAMTS5	Forward: 5′-CCTGGTCCAAATGCACTTCAGC-3′ Reverse: 5′-TCGTAGGTCTGTCCTGGGAGTT-3′
COL2A1	Forward: 5′-CCTGGCAAAGATGGTGAGACAG-3′ Reverse: 5′-CCTGGTTTTCCACCTTCACCTG-3′
SOX9	Forward: 5′-AGGAAGCTCGCGGACCAGTAC-3′ Reverse: 5′-GGTGGTCCTTCTTGTGCTGCAC-3′
GAPDH	Forward: 5’-AAGGTCGGAGTCAACGGATTTG-3 ’ Reverse: 5’-TGACAAAGTGGTCGTTGAGTCA-3 ’

ELISA assay for cytokines

Cell culture supernatants and serum samples were collected and analyzed for IL-6, TNF-α, and IL-10 levels using commercial ELISA kits (MultiSciences, China; IL-6: Cat. No. EK106; TNF-α: Cat. No. EK182; IL-10: Cat. No. EK110) according to the manufacturer’s instructions. Absorbance was measured at 450 nm using a microplate reader (BioTek, ELx800).

Ethical Considerations

All animal experiments were reviewed and approved by the Animal Ethics Committee of Nantong University (Approval No. P20230224-025) and were conducted in accordance with institutional guidelines and the ARRIVE guidelines.

Animal model of OA and treatment

The study was conducted in accordance with the ARRIVE guidelines. All animal experiments were approved by the Experimental Animal Center of Nantong University and the Animal Ethics Committee of Nantong University (Approval No. P20230224-025). Male C57BL/6 mice (8 weeks old) were obtained from the Experimental Animal Center of Nantong University. Mice were randomly divided into four groups (n = 6 per group): sham, OA, OA + siDLX2-2, and OA + OE-SCN4B. OA was induced via anterior cruciate ligament transection (ACLT) surgery on the right knee. Anesthesia was induced by intraperitoneal injection of ketamine (100 mg/kg) and xylazine (10 mg/kg) prior to surgery. siRNA (5 nmol) or plasmid DNA (20 µg) was delivered intra-articularly once a week for four weeks, starting one week after surgery. At 6 weeks post-surgery, all mice were euthanized by carbon dioxide inhalation under deep anesthesia. All procedures involving animals were performed in strict accordance with institutional guidelines and ethical standards.

Histological evaluation

Knee joints were fixed in 4% paraformaldehyde, decalcified, and embedded in paraffin. Sections (5 µm) were stained with Safranin O/Fast Green to assess cartilage integrity. The severity of cartilage damage was graded using the Osteoarthritis Research Society International (OARSI) scoring system. Synovitis was scored based on synovial lining hyperplasia, inflammatory cell infiltration, and subintimal fibrosis, as previously described.

Patient and Public Involvement

It was not appropriate or possible to involve patients or the public in the design, conduct, reporting, or dissemination of this laboratory-based study.

Data Availability Statement

The datasets analyzed during the current study are publicly available in the Gene Expression Omnibus (GEO) database under the accession numbers GSE12021, GSE143514, and GSE206848. All data are freely accessible at https://www.ncbi.nlm.nih.gov/geo/.

Statistical Analysis

R software (version 4.2.2) served for all statistical studies. The association between gene expression and immune cells and pathways was established with the help of Spearman correlation. The Wilcox test was applied to determine if any differences existed between the two groups. Statistically significant difference was defined as a p-value less than 0.05.

Identification of DEGs and aberrant pathways between OA samples and healthy samples

Our investigation used data from synovial samples from two GEOs (GSE143514 and GSE206848), which were performed retrospectively. After batch processing as described in the methods we performed limma package analysis and found a total of 81 DEGs. 50 genes were significantly increased and 31 genes were significantly decreased (Fig. 1A-B). To identify the major pathways of OA disease alterations, we performed GSEA analysis. GSEA showed that the enrichment pathway mainly upregulates the metabolism of arginine and proline, the calcium signaling pathway (Fig. 1C), while inhibiting olfactory transmission and steroid hormone biosynthesis (Fig. 1D).

Machine learning-based identification of diagnostic markers

We employed two different methods to uncover potential biomarkers. The DEGs were reduced to nine variables using the LASSO regression technique, which were identified as promising diagnostic biomarkers for OA (Fig. 2A-B). Similarly, the SVM-RFE technique reduced the DEGs to a selection of 81 genes (Fig. 2C). Thirteen overlapping features between these two algorithms (GNG8, CES3, DLX2, SIRPB2, VPREB3, EPHX1, ABCA3, DPEP3, SSTR5, LOC100506498, SCN4B, GIMAP1-GIMAP5, LOC102725072) were finally selected as markers of OA (Fig. 2D).

Functional characterization of 13 core diagnostic genes

To further explore the functional specificity of the 13 core diagnostic genes, we first mapped the correlation network describing the correlation between the 13 genes (Fig. 3A). Their chromatin locations are shown in Fig. 3B. The functional protein interaction network of the 13 genes was constructed by “Genemania”, and the functional pathways in which they are located are the GTPase complex, the extrinsic component of the cytoplasmic side of the plasma membrane (Fig. 3C). To determine the functions of these genes, we performed GO, KEGG enrichment analysis on the above functionally similar genes. GO assays showed that genes were mainly enriched in positive regulation of myeloid leukocyte mediated immunity, cargo receptor activity, superoxide-generating NAD(P)H oxidase activity, molecular sequestering activity (Fig. 3D). KEGG results showed that are mainly involved in PI3K-Akt signaling pathway, Cell adhesion molecules, Hematopoietic cell lineage, cAMP signaling pathway (Fig. 3E).

Validation of expression levels and diagnostic performance of 13 core diagnostic genes

We first checked the expression of 13 core diagnostic genes in the training set, where GNG8, DLX2, LOC100506498, GIMAP1-GIMAP5, LOC102725072 were upregulated in OA, while CES3, SIRPB2, VPREB3, EPHX1, ABCA3, DPEP3, SSTR5. SCN4B were down-regulated in OA (Fig. 4A). Immediately after, we performed ROC assays to explore the diagnostic value of the above 13 genes for OA. As shown in Fig. 4B-N, we observed that all 13 genes had a strong discriminatory power for OA (all AUC > 0.9). To further validate their expression changes and diagnostic performance. We validated these results in an independent test set. We determined that the expression differences of 13 genes could be reproduced in other data sets (Fig. 5A). The ROC curves confirmed that 13 genes could robustly diagnose the occurrence of OA (Fig. 5B-N). Further, to validate our diagnostic role of the hub gene in OA. We collected human serum samples from healthy (n = 20) and OA (n = 20) patients. The expression of ABCA3, EPHX and DXL2 was detected by ELISA kits. As envisioned, ABCA3, EPHX was significantly downregulated in the mid-term of OA patients (P < 0.05; Fig. 6A-B), and just the opposite, DXL2 was significantly upregulated in OA (P < 0.05; Fig. 6C).

Correlation of 13 diagnostic genes with immune cells and pathways

Infiltration of associated immune cells in TME is an independent predictor of OA and prognosis. To determine whether these diagnostic genes are associated with immune response. Therefore, our team first calculated the immune cell and pathway status in OA and normal samples by the ssGSEA algorithm. CD4⁺ T cells, T helper cells, and TIL were upregulated in OA samples, while the MHC class I pathway was downregulated (Fig. 7A-B). Further, we compared the correlation between these 13 genes and immune cells or pathways in OA. DPEP3, SSTR5, SCN4B appeared to be positively associated with immune active status, while LOC102725072 was associated with immunosuppressive status (Fig. 7C).

Construction and verification of Nomogram

Due to the excellent performance of these 13 core genes in diagnostics. To increase their clinical usability, we plotted the Nomogram (Fig. 8A). The calibration curve shows that the theoretical values overlap exactly with the predicted values confirming the accuracy and robustness of the Nomogram (Fig. 8B).

Experimental validation of core diagnostic genes in synoviocytes and OA mouse models

To experimentally validate the predicted biomarkers and their relevance to OA pathology, we conducted a series of in vitro and in vivo assays (Fig. 9). Firstly, the baseline expression of nine selected genes was measured in human synoviocytes and inflammatory-stimulated OA-like synoviocytes. Results showed consistent upregulation of DLX2 and GNG8, and downregulation of CES3, SCN4B, and ABCA3 in OA-like conditions (Fig. 9A). Among three siRNAs targeting DLX2, siDLX2-2 showed the highest knockdown efficiency (Fig. 9B), while SCN4B overexpression significantly increased its transcript levels (Fig. 9C). Subsequent analysis revealed that DLX2 knockdown or SCN4B overexpression significantly reduced the expression of cartilage catabolic markers including MMP13, ADAMTS5, and COL10A1, while upregulating SOX9 (Fig. 9D). ELISA assays of cell supernatants indicated reduced levels of pro-inflammatory cytokines (IL-6, TNF-α, and IL-10) upon DLX2 knockdown or SCN4B overexpression (Fig. 9E). In an OA mouse model, both DLX2 knockdown and SCN4B overexpression mitigated cartilage damage (OARSI scores, Fig. 9F) and synovitis (synovial inflammation scores, Fig. 9G). Furthermore, MMP13 and MHC-I mRNA levels in joint tissues were suppressed by SCN4B upregulation and DLX2 inhibition (Fig. 9H). Serum ELISA assays also showed systemic decreases in IL-6, TNF-α and IL-10 levels in the treatment groups (Fig. 9I). These findings strongly support the involvement of the DLX2-SCN4B axis in OA progression and highlight their potential as therapeutic targets.

OA is the most common type of arthritis that affects the health management of many people worldwide(22). timely diagnosis of OA is crucial for the management of such diseases, and such early diagnostic tools are still lacking in clinical practice(23, 24). Here, we provide a diagnostic gene screening strategy based on a machine learning approach. geo dataset and identified and validated 13 diagnostic genes. go,kegg was strongly enriched to PI3K-Akt signaling pathway, Cell adhesion molecules, Hematopoietic cell lineage These genes and pathways may be critical for the development of OA.

To screen for potential diagnostic biomarkers of OA, we performed two machine learning algorithms by using 81 DEGs between OA and normal tissue samples and identified only 13 core diagnostic genes. Distal-less homeobox 2 (Dlx2) is a transcription factor that belongs to the Dlx family and is critical for craniofacial tissue development. Previous research has linked DlX2 overexpression to postnatal condylar deformity, subchondral bone degradation, and abnormalities in condylar cartilage histology(25), which may be potentially associated with temporomandibular joint osteoarthritis. In a temporomandibular joint (TMJ) osteoarthritis (OA) marker study, the osteochondral surface was positive for EPHX1 and EPHX1 was also significantly upregulated in peripheral blood leukocytes (26), which is consistent with the results of our work. Differential expression of ABCA3 was found after post-traumatic osteoarthritis (PTOA) development via ligament reconstruction or ligament repair, suggesting that ABCA3 may be one of the potential therapeutic targets for the treatment of patients with osteoarthritis (27). The other diagnostic markers we identified have not been reported in osteoarthritis, which provides a new direction for basic scientific research.

Recently, an increasing number of studies have shown that immune cell infiltration is essential for the development and progression of osteoarthritis(28). OA joints have been found to display a marked infiltration of CD4⁺ T cells, which promote the polarization of stimulated Th1 cells and increase the excretion of immunomodulatory cytokines(29). This local inflammatory event exacerbates the OA process. Therefore, from an immune system perspective, assessing immune cell infiltration and identifying the diversity of infiltrating immune cell components is essential to unravel the molecular level causality of OA and to design new immunotherapeutic targets. In this study, we likewise found an upregulation of associated CD4⁺ T cells in OA, and LOC102725072 showed a significant negative correlation with CD4⁺ T. Therefore, further understanding of the relationship between the two would be of great benefit for OA treatment. In addition, we found that SCN4B, SSTR5 was associated with many immune activation signals, including NK cells, neutrophils, and cytokine activation. Therefore, this could suggest that SCN4B, SSTR5 is associated with the development and progression of OA by regulating multiple immune cells. These hypotheses require more studies to reveal the intricate interactions between genes and immune cells.

To validate the functional roles of our key diagnostic genes, we performed in vitro and in vivo experiments. Knockdown of DLX2 in OA-like synoviocytes significantly suppressed inflammatory cytokine secretion and downregulated cartilage degradation markers such as MMP13 and ADAMTS5. Conversely, overexpression of SCN4B attenuated inflammation and preserved cartilage-related gene expression (COL2A1, SOX9). ELISA analysis and OARSI scoring in animal models further confirmed the protective effects of SCN4B and the pro-degenerative role of DLX2 in joint tissue. These findings support their potential as therapeutic targets and validate our bioinformatics-based predictions in biological systems.

The present work still has some limitations that need to be acknowledged. First, the difficulty in obtaining samples leads to small sample sizes and requires larger clinical trials and cohort validation. Secondly, the data in this paper can only support the correlation analysis between OA and immune cells, and further biochemical experiments are necessary to reveal the causal relationship between the two.

Acknowledgments

Not applicable.

Author Contribution

Hang Zhang and ZiLiang Yu contributed equally to this work. Hang Zhang: Conceptualization, Methodology, Formal analysis, Writing – original draft, Visualization. ZiLiang Yu: Methodology, Validation, Writing – original draft. SuTing Zeng: Validation, Data curation. Wei Shi: Formal analysis, Investigation. Yuxin Zhang: Supervision, Writing – review & editing, Funding acquisition. HongJian Lu: Supervision, Project administration, Writing – review & editing.

Competing interests

The authors declare that they have no competing interests.

Funding

This study was supported by Funded Project of Jiangsu Association for Science and Technology Research (JSKXKT2022048) and University Special Clinical Research Fund Project (2024JY003)

Ethics approval and consent to participate

Consent for publication

Not applicable.

Clinical trial number

not applicable.

Data Availability Statement

Aho OM, Finnila M, Thevenot J, Saarakkala S, Lehenkari P (2017) Subchondral bone histology and grading in osteoarthritis. PLoS ONE 12(3):e0173726
Cisternas MG, Murphy L, Sacks JJ, Solomon DH, Pasta DJ, Helmick CG (2016) Alternative Methods for Defining Osteoarthritis and the Impact on Estimating Prevalence in a US Population-Based Survey. Arthritis Care Res (Hoboken) 68(5):574–580
Moskowitz RW (2009) The burden of osteoarthritis: clinical and quality-of-life issues. Am J Manag Care 15(8 Suppl):S223–S229
Wang SZ, Huang YP, Wang Q, Zheng YP, He YH (2010) Assessment of depth and degeneration dependences of articular cartilage refractive index using optical coherence tomography in vitro. Connect Tissue Res 51(1):36–47
Lai D, Ma W, Wang J, Zhang L, Shi J, Lu C et al (2023) Immune infiltration and diagnostic value of immune-related genes in periodontitis using bioinformatics analysis. J Periodontal Res 58(2):369–380
Huang D, Zheng S, Liu Z, Zhu K, Zhi H, Ma G (2022) Machine Learning Revealed Ferroptosis Features and a Novel Ferroptosis-Based Classification for Diagnosis in Acute Myocardial Infarction. Front Genet 13:813438
Xia L, Gong N (2022) Identification and verification of ferroptosis-related genes in the synovial tissue of osteoarthritis using bioinformatics analysis. Front Mol Biosci 9:992044
Cheng P, Gong S, Guo C, Kong P, Li C, Yang C et al (2023) Exploration of effective biomarkers and infiltrating Immune cells in Osteoarthritis based on bioinformatics analysis. Artif Cells Nanomed Biotechnol 51(1):242–254
Sun YC, Qiu ZZ, Wen FL, Yin JQ, Zhou H (2022) Revealing Potential Diagnostic Gene Biomarkers Associated with Immune Infiltration in Patients with Renal Fibrosis Based on Machine Learning Analysis. J Immunol Res 2022:3027200
Aspden RM, Scheven BA, Hutchison JD (2001) Osteoarthritis as a systemic disorder including stromal cell differentiation and lipid metabolism. Lancet 357(9262):1118–1120
Mohamed-Ali H (1995) Influence of interleukin-1 beta, tumour necrosis factor alpha and prostaglandin E2 on chondrogenesis and cartilage matrix breakdown in vitro. Rheumatol Int 14(5):191–199
Utomo L, Bastiaansen-Jenniskens YM, Verhaar JA, van Osch GJ (2016) Cartilage inflammation and degeneration is enhanced by pro-inflammatory (M1) macrophages in vitro, but not inhibited directly by anti-inflammatory (M2) macrophages. Osteoarthritis Cartilage 24(12):2162–2170
Kapoor M, Martel-Pelletier J, Lajeunesse D, Pelletier JP, Fahmi H (2011) Role of proinflammatory cytokines in the pathophysiology of osteoarthritis. Nat Rev Rheumatol 7(1):33–42
Smith MD (2011) The normal synovium. Open Rheumatol J 5:100–106
Siebuhr AS, Bay-Jensen AC, Jordan JM, Kjelgaard-Petersen CF, Christiansen C, Abramson SB et al (2016) Inflammation (or synovitis)-driven osteoarthritis: an opportunity for personalizing prognosis and treatment? Scand J Rheumatol 45(2):87–98
Leek JT (2014) svaseq: removing batch effects and other unwanted noise from sequencing data. Nucleic Acids Res 42(21):e161
Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W et al (2015) limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43(7):e47
Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z et al (2021) clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innov (Camb) 2(3):100141
Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA et al (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102(43):15545–15550
Hanzelmann S, Castelo R, Guinney J (2013) GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14:7
Xiang J, Huang W, He Y, Li Y, Wang Y, Chen R (2022) Construction of artificial neural network diagnostic model and analysis of immune infiltration for periodontitis. Front Genet 13:1041524
Rabago D, Nourani B (2017) Prolotherapy for Osteoarthritis and Tendinopathy: a Descriptive Review. Curr Rheumatol Rep 19(6):34
Braun HJ, Gold GE (2012) Diagnosis of osteoarthritis: imaging. Bone 51(2):278–288
Ebell MH, Osteoarthritis (2018) Rapid Evid Rev Am Fam Physician 97(8):523–526
Dai J, Si J, Zhu X, Zhang L, Wu D, Lu J et al (2016) Overexpression of Dlx2 leads to postnatal condyle degradation. Mol Med Rep 14(2):1624–1630
Zhang HY, Liu Q, Liu JQ, Wang J, Yang HX, Xu XJ et al (2019) Molecular changes in peripheral blood involving osteoarthritic joint remodelling. J Oral Rehabil 46(9):820–827
Donnenfield JI, Karamchedu NP, Proffen BL, Molino J, Fleming BC, Murray MM (2023) Transcriptomic changes in porcine articular cartilage one year following disruption of the anterior cruciate ligament. PLoS ONE 18(5):e0284777
Woodell-May JE, Sommerfeld SD (2020) Role of Inflammation and the Immune System in the Progression of Osteoarthritis. J Orthop Res 38(2):253–257
Kalaitzoglou E, Griffin TM, Humphrey MB (2017) Innate Immune Responses and Osteoarthritis. Curr Rheumatol Rep 19(8):45

No competing interests reported.

Download PDF

Editorial decision: Revision requested
09 Nov, 2025
Editor assigned by journal
16 Sep, 2025
Submission checks completed at journal
16 Sep, 2025
First submitted to journal
15 Sep, 2025

You are reading this latest preprint version

Machine learning-based analysis reveals potential diagnostic gene biomarkers associated with immune infiltration in patients with osteoarthritis

Status:

Version 1

Abstract

Objective

Materials and Methods

Results

Conclusion

Figures

Introduction

Materials and Methods

Data Acquisition

DEG identification and functional enrichment analysis

Robust prediction models built by machine learning methods

Determination of immune cell patterns in the microenvironment

Cell culture and OA model induction in vitro

Gene knockdown and overexpression

Quantitative real-time PCR (RT-qPCR)

ELISA assay for cytokines

Ethical Considerations

Animal model of OA and treatment

Histological evaluation

Patient and Public Involvement

Data Availability Statement

Statistical Analysis

Results

Identification of DEGs and aberrant pathways between OA samples and healthy samples

Machine learning-based identification of diagnostic markers

Functional characterization of 13 core diagnostic genes

Validation of expression levels and diagnostic performance of 13 core diagnostic genes

Correlation of 13 diagnostic genes with immune cells and pathways

Construction and verification of Nomogram

Experimental validation of core diagnostic genes in synoviocytes and OA mouse models

Discussion

Declarations

References

Additional Declarations

Status:

Version 1