Ferroptosis-related genes LPCAT3 and PGD are potential diagnostic biomarkers for osteoarthritis
Journal of Orthopaedic Surgery and Research volume 18, Article number: 699 (2023)
Osteoarthritis (OA) is the most common chronic joint disease and how ferroptosis contributes to OA has garnered much attention recently. Bioinformatics promoted the discovery of ferroptosis-related biomarkers for OA. But since OA is a whole-joint disease, sensitive biomarkers for OA are still limited. We herein focused on subchondral bone, a joint component often-ignored by existing bioinformatic reports, to identify ferroptosis-related diagnostic biomarkers for OA.
Microarray datasets GSE51588 and GSE55457 were downloaded from Gene Expression Omnibus database. Ferroptosis-related differential expression genes (Ferr-DEGs) between OA and normal samples were identified and their functional enrichment was analyzed. Common genes for OA diagnosis were selected from Ferr-DEGs using the combination of SVM-RFE, LASSO regression, and RandomForest machine learning algorithms. Common genes' diagnostic value was verified by receiver operating characteristic (ROC) curve and their association with immune infiltration was analyzed by CIBERSORT. Finally, candidate gene’s expression was verified in chondrocytes from OA patients and in an in vitro IL-1β-induced OA model, by RT-PCR.
Two ferroptosis-related genes, LPCAT3 and PGD, were identified as OA diagnostic biomarkers and confirmed by ROC diagnostic test. The association of LPCAT3 and PGD with the infiltration of several types of immune cells was identified. The decreased expression of LPCAT3 and PGD was both confirmed in OA chondrocytes and IL-1β-induced OA condition.
We identified ferroptosis-related genes LPCAT3 and PGD as potential diagnostic biomarkers for OA, which may offer insight into the role of ferroptosis in OA and provides useful information for the diagnosis and treatment of OA.
Osteoarthritis (OA) is the most common chronic joint disease characterized by cartilage damage, subchondral bone destruction, and synovium inflammation . It is estimated that more than 240 million people worldwide suffer from OA, and OA prevalence is even rising due to population aging, which causes a financial burden for patients and society . Several OA-related proteins, such as type 2 collagen (COL2A1), aggrecan (ACAN), and matrix metalloproteinase (MMP)-13, have been commonly acknowledged as OA biomarkers . However, the pathogenesis of OA is very complex in which mechanical loading, inflammation, and metabolic factors play a critical role corporately. Thus, the discovery of sensitive biomarkers, especially through high-throughput analysis, not only benefits the diagnosis and therapy but also helps understand the pathogenic mechanism of OA .
Ferroptosis, first discovered in 2012, is a new type of programmed cell death that is distinct from apoptosis, pyroptosis, and necroptosis in terms of morphological and biochemical features . Ferroptosis involves metabolisms of lipid, iron, and thiol, and is caused by the iron-dependent accumulation of lipid peroxidation that reaches the cellular lethal level. The number of studies about the role played by ferroptosis in OA started growing rapidly from two years ago, making it a research hotspot nowadays [6, 7]. Emerging evidence has suggested that iron dyshomeostasis and lipid peroxidation contribute to OA pathogenesis and ferroptosis could be a promising target in OA therapy.
In modern medicine, high-throughput sequencing combined with bioinformatic analysis becomes a popular research tool due to its strong power in uncovering valuable clues for the diagnosis, prognosis, and treatment of diseases. Increasing interest has been aroused to reveal diagnostic markers or therapeutic targets for OA based on bioinformatic data mining. For example, in the synovium or cartilage of the osteoarthritic joint, several ferroptosis-related genes such as SLC3A2, ATF3, IL6, etc. have been suggested as OA biomarkers in this way [8, 9].
However, sensitive biomarkers for OA are still limited and OA is a whole-joint disease involving cartilage, synovium, subchondral bone, and infrapatellar fat pad. Additionally, integrative analysis using multi-algorithms plus post hoc experimental validation is often necessary for obtaining more reliable predictions. Thus, focusing on subchondral bone, an easily-ignored joint component, we combined three machine learning algorithms to identify ferroptosis-related diagnostic biomarkers for OA and validated them in vitro, which may offer insight into the function of ferroptosis in OA together with the diagnosis and treatment of OA.
Data acquisition and processing
GSE51588 and GSE55457 microarray datasets were downloaded from Gene Expression Omnibus (GEO) data repository (https://www.ncbi.nlm.nih.gov/geo/). GSE51588 dataset contains 40 OA samples and 10 normal samples of human tibial plateaus (subchondral bone); GSE55457 dataset contains 10 OA samples and 10 normal samples of human synovium. The data was processed by mapping the probes to the gene names and log2-transforming quantile-normalized signal strength. Null probes were removed; replicate probes were mapped to the same gene.
Identification of ferroptosis-related differential expression genes (Ferr-DEGs)
Gene expression differences between OA and normal samples in GSE51588 dataset were analyzed by using R package “limma” (adjusted P value < 0.05). The entire set of differential expression genes (DEGs) was then intersected with the ferroptosis-related gene list retrieved from FerrDb database  (http://www.zhounan.org/ferrdb/) that contains 621 defined ferroptosis gene regulators, to obtain the Ferr-DEGs.
Functional enrichment analysis of the Ferr-DEGs
Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed to explore the major biological attributes of Ferr-DEGs. Enrichment maps were visualized by R packages “ggplot2” and “GOplot”. Enrichment analysis for Disease Ontology (DO) terms was run using R packages “clusterProfiler” and “DOSE”.
Selection of common genes for OA diagnosis
Potential OA diagnostic biomarkers were selected from the above Ferr-DEGs, via using the combination of three machine learning algorithms: support vector machine recursive feature elimination (SVM-RFE), the least absolute shrinkage and selection operator (LASSO) regression, and RandomForest. SVM-RFE is a machine learning technique that trains a subset of features from different categories to shrink the feature set and then backward select the most predictive features . LASSO regression was performed using R package “glmnet” to select the valuable variables in the linear model. The response type was set to binomial and “one standard error” rule was applied. RandomForest algorithm was applied to rank the candidate genes : relative importance value > 0.25 was classified as a typical chance cause. Finally, the most predictive genes were determined as the intersection of the results from SVM-RFE, LASSO regression, and RandomForest.
Diagnostic value of the candidate genes in OA
The comparison of candidate gene’s expression between OA and control group was determined by Student’s t test. Diagnostic value of the candidate genes was analyzed by receiver operating characteristic (ROC) curve in GSE51588 dataset.
For external validation, we screened all of the existing GEO datasets that contains osteoarthritic subchondral bone samples but did not find a suitable dataset for our study. Thus, we selected GSE55457 dataset, which contains 10 OA samples and 10 normal samples of human synovium, for external validation.
Immune cell infiltration analysis
To evaluate the role of immune microenvironment in OA development, CIBERSORT algorithm  (http://cibersortx.stanford.edu), which can perform linear support vector regression to deconvolute gene expression profiles, was employed to estimate the abundance of 22 infiltrated immune cell types in each sample. Student’s t test was employed to compare the difference in immune cell infiltration between the OA and the normal groups. Then, the correlation of the candidate genes to the abundance of infiltrated immune cells was determined by Spearman’s rank correlation test and visualized by R package “ggplot2”.
Experimental validation of the candidate genes
Chondrocytes from healthy people (CP-H107, Procell Life Science & Technology, Wuhan, China) and OA patients (402OA-05A, Cell Applications) were used for experimental validation of the candidate genes. Cells were maintained in DMEM/F12 medium (HyClone) supplemented with 10% fetal bovine serum at 37 °C, 5% CO2. Passage 2–3 was used for experiments and then cells were harvested for RT-PCR.
Additionally, we also performed an inflammatory-cytokine-induced OA model. Interleukin-1β (IL-1β) is one of the most popular cytokines for modeling OA in vitro because it can induce cartilage destruction through promoting the release of matrix metalloproteinases from chondrocytes . We used various concentrations of IL-1β (1, 5, 20 ng/ml where indicated) to stimulate CP-H107 chondrocytes and harvested the cells for RT-PCR after 48 h.
The expression of candidate genes and OA marker genes (COL2A1, ACAN, MMP13) was detected using RT-PCR. Cells were lysed by TRIzol (Invitrogen) for total RNA extraction. cDNA reverse-transcription was performed by using iScript™ cDNA Synthesis Kit (Bio-Rad). RT-PCR was carried out on StepOnePlus RT-PCR system (Applied Biosystems) using the iTaq™ Universal SYBR Green Supermix reagent (Bio-Rad). The relative expression of target genes was normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) based on the ΔΔCT method. Primers are listed in Table 1.
R program 4.3.1 was used for bioinformatic analysis and statistical computing. The comparisons between two groups were determined by Student’s t test while multiple-group comparisons were determined by one-way ANOVA plus Bonferroni post hoc test. Correlations were determined by Spearman’s rank correlation test. P < 0.05 was considered significant. For experimental validation assays, experiments were repeated three times and data were presented as mean ± SEM.
Identification of Ferr-DEGs and their functional enrichment
We used GSE51588 microarray dataset, which contains the transcriptome profile of human knee tibial plateaus (40 OA and 10 normal samples), to identify Ferr-DEGs. Ferr-DEGs were defined as the intersection of entire DEGs (adjusted P value < 0.05) and FerrDb-defined ferroptosis-related gene list . In total, there were 29 Ferr-DEGs in the OA group compared to the normal control group, including 3 upregulated and 26 downregulated genes (Fig. 1A, 1B).
In an attempt to investigate the biological activities and pathways associated with Ferr-DEGs among the OA cases, DO, GO, and KEGG enrichment analyses were performed. The top 10 enriched pathways for DO, GO, and KEGG terms, respectively, can be seen in Fig. 1C-1E. Taken DO, GO, and KEGG enrichment results together, metabolism-related pathways presented most frequently. For example, the “inherited metabolic disorder”, “metal metabolism disorder” and “carbohydrate metabolic disorder” in DO enrichment result (Fig. 1C); the “sulfur compound metabolic process”, “cellular modified amino acid metabolic process”, etc. in GO enrichment result (Fig. 1D); the “glutathione metabolism”, “carbon metabolism” etc. in KEGG enrichment result (Fig. 1E).
Selection of ferroptosis-related diagnostic biomarkers for OA
Aiming to obtain more reliable candidates, we used three machine learning algorithms to determine OA diagnostic biomarkers from Ferr-DEGs. SVM-RFE selected the most 2 predictive features (Fig. 2A) while LASSO regression selected 7 predictive genes from statistically significant univariate variables (Fig. 2B). RandomForest algorithm calculated the relationship between the error rate and the number of classification trees and ranked 10 genes with relative importance > 0.25 (Fig. 2C). We used a Venn diagram to display the intersection of three methods and eventually found 2 common genes, lysophosphatidylcholine acyltransferase 3 (LPCAT3) and phosphogluconate dehydrogenase (PGD), as potential OA diagnostic biomarkers (Fig. 2D).
Common gene’s diagnostic value for OA
We next verified the diagnostic value of these two common genes LPCAT3 and PGD for OA. In training set GSE51588 (Fig. 3), decreased expression of LPCAT3 and PGD was observed in OA condition and both of them showed a very good diagnostic accuracy for OA.
We also verified the diagnostic value of LPCAT3 and PGD in an external validation set GSE55457 that contains human synovium samples because no suitable subchondral bone GEO dataset was found. 14 Ferr-DEGs were found: 6 upregulated and 8 downregulated (Fig. 4A). In validation set GSE55457, LPCAT3 showed a decrease trend in OA, though not significant, and PGD expression was significantly decreased in OA (Fig. 4B), which is generally consistent with that of GSE51588. While LPCAT3 showed a poor diagnostic accuracy, PGD still showed a good diagnostic accuracy (Fig. 4C).
Immune cell infiltration pattern in OA and its association with the candidate genes
Inflammation and immune microenvironment play a critical role in OA pathogenesis. CIBERSORT-based immune cell infiltration analysis is a powerful tool to provide references for the prediction of disease course . We used CIBERSORT to generate the landscape of immune cell infiltration in the OA and the normal control samples in GSE51588 dataset, which displays the relative proportion of 22 types of immune cells (Fig. 5A). Correlations between every two cell types can be seen in Fig. 5B correlation matrix. As Fig. 5C shows, compared to the normal control, the OA group had higher infiltration of B cells naïve, T cells CD4 naïve, T cells regulatory, macrophages M1, macrophages M2, and dendritic cells resting. By contrast, the OA group had lower infiltration of T cells CD4 memory resting, mast cells resting, and neutrophils.
We next analyzed the association of LPCAT3 or PGD with immune cell infiltration in OA (Fig. 5D). The infiltration of several types of immune cells was found to have a significant correlation to LPCAT3 or PGD expression. Among them, both LPCAT3 and PGD had a positive correlation to neutrophils and mast cells resting (P < 0.05), while having a negative correlation to B cells naïve and dendritic cells resting (P < 0.05). This implies that LPCAT3 and PGD might have an important function on these immune cells and thus contribute to the pathogenesis of OA, which can be a good source for future studies.
In vitro validation of LPCAT3 and PGD as OA diagnostic biomarkers
We next validated LPCAT3 and PGD as OA diagnostic biomarkers in in vitro OA models. First, we detected LPCAT3 and PGD expression in chondrocytes from healthy people and OA patients (Fig. 6A). Chondrocytes-OA (402OA-05A chondrocytes from OA patients) are a useful OA model and showed a robust OA-like phenotype, demonstrated by the decreased expression of COL2A1 and ACAN and the increased expression of MMP13. Both LPCAT3 and PGD expression were reduced in chondrocytes-OA, which confirmed our bioinformatic findings.
We further confirmed LPCAT3 and PGD expression in an IL-1β-induced OA model. IL-1β stimulation on chondrocytes is one of the most popular models for studying OA in vitro . The OA-like response in CP-H107 chondrocytes upon IL-1β treatment (decreased expression of COL2A1 and ACAN and increased expression of MMP13) demonstrated the reliability of our experimental system (Fig. 6B). In IL-1β-induced OA condition, decreased LPCAT3 expression was confirmed. PGD expression also showed a decrease trend, though not statistically significant. Notably, the expression of LPCAT3 and PGD decreased as IL-1β treatment concentration increased (Fig. 6C), suggesting a negative association between LPCAT3 and PGD expression and the severity of OA.
Collectively, although LPCAT3 did not show good diagnostic accuracy in external validation set (Fig. 4), these in vitro results strongly supported the diagnostic value of LPCAT3 and PGD so we propose LPCAT3 and PGD as potential biomarkers for OA.
Over the past two years, studies focused on the role of ferroptosis in OA were increasing rapidly and thus made this a research hotspot. Clinical observations found that iron accumulation was commonly detected in the synovial fluid, synovia, and cartilage of OA patients and was positively correlated with OA severity [15, 16]. Ferroptotic alternations, such as iron accumulation, lipid peroxidation, and mitochondrial dysfunction are closely associated with OA progression where GPX4-mediated oxidative stress and extracellular matrix degradation serve as a core mechanism [16,17,18]. Nevertheless, how ferroptosis contributes to OA pathogenesis remains to be fully elucidated. In recent years, bioinformatics data mining greatly accelerated the discovery of new molecular targets for OA therapy. Xia et al. applied protein–protein interaction (PPI) network algorithm on a synovial tissue gene array dataset and screened 7 ferroptosis-related genes (ATF3, IL6, CDKN1A, IL1B, EGR1, JUN, CD44) to be possible biomarkers and therapeutic targets for OA . Liu et al. identified ferroptosis-related gene SLC3A2 as a therapeutic target for OA, via performing weighted correlation network analysis (WGCNA) and PPI analysis on a cartilage tissue RNA-seq dataset . Considering OA is a whole-joint disease involving cartilage, synovium, subchondral bone, and infrapatellar fat pad, our findings based on a subchondral bone tissue gene array is a good supplement to the previous synovium- and cartilage-derived discoveries [8, 9]. Additionally, in order to obtain more reliable outcomes, we combined the results from 3 machine learning algorithms (SVM-RFE, LASSO, RandomForest) and validated the common genes in vitro.
LPCAT3 is an enzyme that involves in the synthesis and decomposition of polyunsaturated fatty acids and can mediate the sensitivity of cells to ferroptosis . To date, we did not find any reports about its association with arthritis. By contrast, PGD is a vital oxidative carboxylase in the pentose phosphate pathway and its role in arthritis only has very limited reports. For example, salivary PGD level was altered in patients with rheumatoid arthritis compared to healthy subjects . PGD activity in synovial lining cells was suppressed by menadione epoxide administration in an experimental immune arthritis model . Based on the above, the role of LPCAT3 and PGD remains largely unknown in OA so it is worthy of further investigations.
Immune microenvironment is closely associated with OA development and immune cell infiltration in joint tissues can be significantly altered under OA condition [8, 22]. We found that the immune cell infiltration pattern in osteoarthritic subchondral bone was partly consistent with the results in osteoarthritic synovium or cartilage reported by other studies [8, 22]. We also revealed that ferroptosis-related genes LPCAT3 and PGD were correlated to the infiltration of several types of immune cells, especially neutrophils, mast cells resting, B cells naïve, and dendritic cells resting. This implies that LPCAT3 and PGD may contribute to OA pathogenesis via regulating the activities of these cells, which is a good point to test in the future.
Of course, this study also has some limitations. First, only one GEO dataset was included for bioinformatic analysis and the sample size is not quite large, which sometimes might cause bias. Future studies with a much larger sample size or if can combine multiple datasets are helpful to improve the validity. Second, comprehensive validation experiments, especially in vivo and clinical investigations, can greatly elevate the diagnostic and therapeutic values of the OA biomarkers identified herein and thus are worth exploring in the future.
In conclusion, based on integrative bioinformatic analysis and in vitro experimental validation, we identified ferroptosis-related genes LPCAT3 and PGD as potential diagnostic biomarkers for OA and revealed their association with immune cell infiltration in OA. These findings may offer insight into the role of ferroptosis in OA pathogenesis and also provide useful information for the diagnosis and treatment of OA.
Availability of data and materials
The datasets presented in this study can be found in online repositories (GEO data repository, https://www.ncbi.nlm.nih.gov/geo/, accession number GSE51588 and GSE55457). Further inquiries can be directed to the corresponding author.
Type 2 collagen
Gene Expression Omnibus
Ferroptosis-related differential expression genes
Differential expression genes
Kyoto Encyclopedia of Genes and Genomes
Support vector machine recursive feature elimination
The least absolute shrinkage and selection operator
Receiver operating characteristic
Lysophosphatidylcholine acyltransferase 3
Yue L, Berman J. What is osteoarthritis? JAMA. 2022;327:1300.
Chen N, Fong DYT, Wong JYH. Health and economic outcomes associated with musculoskeletal disorders attributable to high body mass index in 192 countries and territories in 2019. JAMA Netw Open. 2023;6: e2250674.
Hunter DJ, Collins JE, Deveza L, Hoffmann SC, Kraus VB. Biomarkers in osteoarthritis: current status and outlook—the FNIH Biomarkers Consortium PROGRESS OA study. Skeletal Radiol. 2023. https://doi.org/10.1007/s00256-023-04284-w.
Tachmazidou I, Hatzikotoulas K, Southam L, Esparza-Gordillo J, Haberland V, Zheng J, et al. Identification of new therapeutic targets for osteoarthritis through genome-wide analyses of UK Biobank data. Nat Genet. 2019;51:230–6.
Dixon SJ, Lemberg KM, Lamprecht MR, Skouta R, Zaitsev EM, Gleason CE, et al. Ferroptosis: an iron-dependent form of nonapoptotic cell death. Cell. 2012;149:1060–72.
Zhang Y, Huang X, Qi B, Sun C, Sun K, Liu N, et al. Ferroptosis and musculoskeletal diseases: “Iron Maiden” cell death may be a promising therapeutic target. Front Immunol. 2022;13: 972753.
Chang S, Tang M, Zhang B, Xiang D, Li F. Ferroptosis in inflammatory arthritis: a promising future. Front Immunol. 2022;13: 955069.
Xia L, Gong N. Identification and verification of ferroptosis-related genes in the synovial tissue of osteoarthritis using bioinformatics analysis. Front Mol Biosci. 2022;9: 992044.
Liu H, Deng Z, Yu B, Liu H, Yang Z, Zeng A, et al. Identification of SLC3A2 as a potential therapeutic target of osteoarthritis involved in ferroptosis by integrating bioinformatics, clinical factors and experiments. Cells. 2022;11:3430.
Zhou N, Yuan X, Du Q, Zhang Z, Shi X, Bao J, et al. FerrDb V2: update of the manually curated database of ferroptosis regulators and ferroptosis-disease associations. Nucleic Acids Res. 2023;51:D571–82.
Huang ML, Hung YH, Lee WM, Li RK, Jiang BR. SVM-RFE based feature selection and Taguchi parameters optimization for multiclass SVM classifier. Sci World J. 2014;2014: 795624.
Ishwaran H, Kogalur UB. Consistency of random survival forests. Stat Probab Lett. 2010;80:1056–64.
Newman AM, Steen CB, Liu CL, Gentles AJ, Chaudhuri AA, Scherer F, et al. Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol. 2019;37:773–82.
Johnson CI, Argyle DJ, Clements DN. In vitro models for the study of osteoarthritis. Vet J. 2016;209:40–9.
Ogilvie-Harris DJ, Fornaiser VL. Synovial iron deposition in osteoarthritis and rheumatoid arthritis. J Rheumatol. 1980;7:30–6.
Miao Y, Chen Y, Xue F, Liu K, Zhu B, Gao J, et al. Contribution of ferroptosis and GPX4’s dual functions to osteoarthritis progression. EBioMedicine. 2022;76: 103847.
Xu W, Zhang B, Xi C, Qin Y, Lin X, Wang B, et al. Ferroptosis plays a role in human chondrocyte of osteoarthritis induced by IL-1β in vitro. Cartilage. 2023. https://doi.org/10.1177/19476035221142011.
Al-Hetty H, Abdulameer SJ, Alghazali MW, Sheri FS, Saleh MM, Jalil AT. The role of ferroptosis in the pathogenesis of osteoarthritis. J Membr Biol. 2023;256:223–8.
Kagan VE, Mao G, Qu F, Angeli JP, Doll S, Croix CS, et al. Oxidized arachidonic and adrenic PEs navigate cells to ferroptosis. Nat Chem Biol. 2017;13:81–90.
Giusti L, Baldini C, Ciregia F, Giannaccini G, Giacomelli C, De Feo F, et al. Is GRP78/BiP a potential salivary biomarker in patients with rheumatoid arthritis? Proteom Clin Appl. 2010;4:315–24.
Pitsillides AA, Blake SM, Glynn LE, Frost GT, Bitensky L, Chayen J. The effect of menadione epoxide on the experimental immune arthritis in the rabbit. Int J Exp Pathol. 1991;72:301–9.
Li S, Wang H, Zhang Y, Qiao R, Xia P, Kong Z, et al. COL3A1 and MMP9 serve as potential diagnostic biomarkers of osteoarthritis and are associated with immune cell infiltration. Front Genet. 2021;12: 721258.
This study is supported by grants from National Natural Science Foundation of China (Grant No. 32000928), Sichuan Science and Technology Program (Grant No. 2023YFS0150), Guangxi Natural Science Foundation (Grant No. 2023GXNSFBA026125), and Basic Ability Improvement Project for Young and Middle-aged Teachers in Colleges and Universities of Guangxi (Grant No. 2023KY0122 and 2022KY0532).
The authors declare no potential conflicts of interest.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Wang, L., Ye, S., Qin, J. et al. Ferroptosis-related genes LPCAT3 and PGD are potential diagnostic biomarkers for osteoarthritis. J Orthop Surg Res 18, 699 (2023). https://doi.org/10.1186/s13018-023-04128-2
- Bioinformatic analysis
- Diagnostic biomarker
- Immune infiltration