Meta-analysis Integrated With Multi-omics Data Analysis to Elucidate Pathogenic Mechanisms of Age-Related Knee Osteoarthritis in Mice

Abstract

Increased mechanistic insight into the pathogenesis of knee osteoarthritis (KOA) is needed to develop efficacious disease-modifying treatments. Though age-related pathogenic mechanisms are most relevant to the majority of clinically presenting KOA, the bulk of our mechanistic understanding of KOA has been derived using surgically induced posttraumatic OA (PTOA) models. Here, we took an integrated approach of meta-analysis and multi-omics data analysis to elucidate pathogenic mechanisms of age-related KOA in mice. Protein-level data were integrated with transcriptomic profiling to reveal inflammation, autophagy, and cellular senescence as primary hallmarks of age-related KOA. Importantly, the molecular profiles of cartilage aging were unique from those observed following PTOA, with less than 3% overlap between the 2 models. At the nexus of the 3 aging hallmarks, advanced glycation end product (AGE)/receptor for AGE (RAGE) emerged as the most statistically robust pathway associated with age-related KOA. This pathway was further supported by analysis of mass spectrometry data. Notably, the change in AGE–RAGE signaling over time was exclusively observed in male mice, suggesting sexual dimorphism in the pathogenesis of age-induced KOA in murine models. Collectively, these findings implicate dysregulation of AGE–RAGE signaling as a sex-dependent driver of age-related KOA.

“What doesn’t kill me makes me stronger.” In 1888, Friedrich Nietzsche made this proclamation; a proclamation that has echoed long beyond his years, perhaps because it resonates with our inherent desire to not only survive but to thrive. Over the last century, medical research has championed increasing the life span of our population, with encouraging success. However, a new set of epidemic diseases has made clear the all-too-common disconnect between surviving and thriving. One of the most debilitating, yet nonlethal, diseases is knee osteoarthritis (KOA), which represented the 15th leading cause of disability worldwide in 2019 (1). The number of years lived with KOA-induced disability has increased by 39% from 1990 to 2019 (2), suggesting that the impact of KOA on society is rapidly expanding. Despite this, progress in developing effective interventions for KOA has been slow, with current therapeutic strategies primarily focusing on symptom management, such as pain reduction and compensatory approaches to improve physical function (3). The development of effective and sustainable disease-modifying treatment is, therefore, a critical challenge to reduce the economic burden of KOA and to extend the health span of our aging population.

Several clinical trials targeting molecules implicated in the pathogenesis of KOA have been tested, though none have proven effective. For example, inducible nitric oxide synthase (iNOS) has been shown in a dog model of posttraumatic OA (PTOA) to be a key driver of cartilage damage (4). Unfortunately, clinical translation of cindunistat, a drug targeting iNOS, did not slow the rate of joint space narrowing in older subjects with KOA when compared to placebo (5). Likewise, inhibitors of a-disintegrin and metalloproteinase with thrombospondin motifs (ADAMTS) have also been investigated owing to their reported role in the degradation of proteoglycans in a murine model of PTOA (6). However, most phase I and II clinical trials involving ADAMTS have not reported findings and/or have ceased further development (7). The lack of success in these trials is attributed, at least in part, to our inadequate understanding of the cellular and molecular mechanisms driving KOA.

What has impeded progress toward mechanistic understanding of KOA pathogenesis? Due to challenges obtaining longitudinal human OA cartilage samples and difficulties securing valid and reliable control samples, animal models are a staple of KOA mechanistic studies. In these efforts, a variety of animal models with over 20 different OA induction methods and across numerous animal species have been used. Clearly, disease pathology and progression are dependent on the model used (8). Models in which OA is induced surgically or mechanically (ie, PTOA) are most commonly investigated owing to their convenience (9). However, the translatability of findings from PTOA animal models to age-related OA in humans is unclear. For instance, unlike in PTOA models, most individuals with OA do not present with a clear inciting event. Instead, aging is the single greatest predictive factor for KOA (10). PTOA, on the other hand, accounts for only 12% of the total KOA burden (11). The study of natural-aging KOA models is therefore imperative for enhanced mechanistic insight into disease processes affecting the majority of people with OA.

One valuable method for elucidating pathogenic mechanisms is meta-analysis of the existing literature. Meta-analysis allows for comparison of contrasting results across different studies (12). Another advantage of meta-analysis is that it allows for relative weighting of the robustness of candidate pathways implicated in disease pathogenesis (12), findings that may be further confirmed using systems biology approach. In this study, we integrated meta-analysis together with analyses of publicly available transcriptomic and proteomic data sets with the goal of identifying common molecular denominators of age-related KOA. Specifically, we performed a systematic literature search to identify all relevant age-related KOA studies in murine models and compiled findings via meta-analysis. For this, we limited studies to a mouse model, as this is the most common model given the relatively fast disease progression, cost-efficiency, and ease of handling (8). Furthermore, several assessment tools, including microcomputed tomography, gait analysis, and pain assessments, are available for mouse models, thereby allowing for a more comprehensive assessment of pathology (13). From the included studies, we then integrated histology for cartilage degeneration, RNA (ie, RNA-seq), and protein (eg, Western blots and immunohistochemistry) data to characterize aged cartilage and determine the most robust pathways associated with age-related KOA. The identified pathways were then cross-checked by analyzing archived mass spectrometry proteomics data from young and aged male and female mice. The culmination of these analyses revealed that advanced glycation end product (AGE)–receptor for AGE (RAGE) signaling was the most statistically robust pathway associated with age-related KOA in male, but not female, mice.

Materials and Methods

The full methodology for this manuscript is provided in the Supplementary Material and includes information on: eligibility criteria, literature search, study selection, data collection, meta-analysis, functional characteristics of transcriptome, protein–protein interaction (PPI) network, risk of bias SYstematic Review Centre for Laboratory animal Experimentation (SYRCLE), publication bias, and quality of evidence (Grades of Recommendation, Assessment, Development, and Evaluation [GRADE]). This study was conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statement (14) (Supplementary Appendix S1), PRISMA protocols (PRISMA-P) (15), Meta-analysis of Observational Studies in Epidemiology checklist (16) (Supplementary Appendix S1), Cochrane Handbook for Systematic Reviews of Interventions (17), and the practical guide for meta-analysis from animal studies (18).

Literature Search, Study Selection, and Data Collection

We included articles characterizing aged articular cartilage in the knee joints of mice (ie, each study had to include both aged and young mice or aged and middle-aged mice). Young, middle-aged, and aged mice are commonly defined as 3–6, 9–15, and 18–24 months old, respectively, and correspond to approximately ages 20–30, 38–47, and 56–69 years in humans, respectively (19). However, to increase the number of included articles, we expanded the range of young mice to 2–7 months old, which allowed us to include an additional 4 studies (20–23). Given that 2-month-old mice are juvenile and not skeletally mature, we performed a sensitivity analysis and confirmed that the overall conclusion of the meta-analysis was unchanged when studies with 2-month-old mice were omitted. This review focused on articles written in English. The predefined outcome measurements used in our meta-analysis consisted of (i) morphology, (ii) morphometry, (iii) composition, (iv) biomechanical characterization, (v) biomarker, and (vi) molecular biology.

An electronic search was conducted on March 10, 2020 using PubMed, Physiotherapy Evidence Database, Cumulative Index to Nursing and Allied Health Literature, and Cochrane Central Register of Controlled Trials electronic databases. An additional search was conducted in June 2021 to include all the relevant articles prior to publication. A manual search of the reference lists within previously published narrative reviews was also performed. Finally, a citation search was performed on the original records with Web of Science. Two independent reviewers (H.I. and G.G.) assessed eligibility and screened titles and abstracts yielded by the search. Full manuscripts of the articles that met the eligibility criteria were then reviewed. Disagreements regarding manuscript inclusion between the 2 reviewers were discussed until consensus was achieved.

A single reviewer (H.I.) extracted data regarding basic study information (authors, publication year, and country of the corresponding author), experimental condition (ie, mice strain, age, sample size, and sex), target joint (tibiofemoral or patellofemoral joints), outcome measures (cartilage morphology, morphometry, composition, biomechanical characterization, biomarker, and molecular biology), funding, and presence of conflict of interests. These data were extracted because of their potential to influence key outcome measures (with the exception of basic study information) (24,25). If the outcome measures from multiple time points were reported within same age category (eg, 3 and 6 months old from the young group), we averaged the effect size as a means to address multiplicity (27). Averaging multiple effect sizes within studies is a straightforward way to handle multiplicity if the effect sizes within studies are equivalent in the sense that each would answer the research question with similar relevance and they are not expected to differ in relation to any moderators of interest (27). We further confirmed that overall results of meta-analysis and trajectory of cartilage degeneration were similar when using the effect size only from the older aged mice (eg, if effect size was provided for 3 and 6 months old mice, the effect size of 6 months old mice was used). If outcome measures from multiple compartments (eg, medial and lateral compartments in tibiofemoral joint and patellofemoral joint) were presented, data from the most severe region were extracted.

Meta-analysis

To characterize aged articular cartilage and the underlying mechanism of age-related cartilage degeneration, pooled estimates and 95% confidence intervals (CIs) for standardized mean differences (SMDs) of outcome measures were calculated using a random-effect model. SMDs were calculated using the mean between-group difference (aged versus middle-aged or young) divided by the pooled standard deviation. Study heterogeneity, the inter-trial variation in study outcomes, was assessed using I², which is the proportion of total variance explained by inter-trial heterogeneity.

To address the trajectory of age-related changes in articular cartilage, a mixed linear regression analysis with random slopes and random intercepts was performed for the outcome of cartilage morphology. In this analysis, age category (1: young, 2: middle-aged, 3: aged) and standardized semiquantitative scores of cartilage degeneration were included as independent and dependent variables, respectively. To standardize semiquantitative scores of cartilage degeneration, all histological scores provided in each included study were converted to 0–100 and recalculated as in a previous meta-analysis (28), with higher scores indicating more severe cartilage degeneration. For example, if the Osteoarthritis Research Society International (OARSI) score (ranging from 0 to 24 points) was 10 points, these data were converted to 41.7 points. This approach gave us the advantage of an intuitive and easily interpretable histological assessment across different studies. Of the 11 studies included in the meta-analysis for cartilage morphology, 4 studies used OARSI scoring. To account for possible biases associated with the histological scoring method, we performed a sensitivity analysis, in which we included only OARSI scores as an outcome. We found that the overall effect size was similar to original analysis, suggesting that the scoring method did not influence the overall effect size.

Multi-omics Computational Analysis

To evaluate the function of transcripts that were significantly altered across cartilage aging, we performed a single-sample gene set enrichment analysis. R/Bioconductor package fgsea was used with included gene scores defined by log2 fold change of gene expression profile available from 1 study (23). This analysis can determine whether a given gene set is significantly enriched in a list of gene markers ranked by their correlation with a phenotype of interest. We used gene ontology (GO) terms (GO_Biological_Process_2018) as a gene set downloaded from Enrichr (https://amp.pharm.mssm.edu/Enrichr/). Subsequently, REVIGO software (29) was applied to summarize redundant GO terms and visualize the summarized results. RNA-seq data (23) with a false discovery rate (FDR) adjusted p value less than .05 was used.

Mass spectrometry data were generated by Iijima et al. (30). Briefly, knee cartilage samples from the medial and lateral tibia and femur were microdissected under a microscope from male and female young (4–6 months), middle-aged (10–14 months), and aged (21–24 months) C57/BL6 mice using a protocol similar to that described by Gardiner et al. (31). For microdissection, the femur and tibia were separated, and all fat, muscle, and ligament tissues were removed. The femoral condyles were visualized under the microscope, and small scissors and tweezers were used to scrape cartilage off the condyles. Collected samples were placed in phosphate buffered saline (PBS) on ice. The tibial plateau was similarly placed under the microscope and visualized. Small scissors and tweezers were used to scrape cartilage off the plateau. Samples were washed separately in PBS after completion of the dissection. Dissectors carefully examined cartilage samples to ensure minimal fat, muscle, ligament, and tendons were included in the tissue harvest. We previously confirmed that over 95% of cells isolated by the same method expressed type II collagen (ie, chondrogenic marker) (30), suggesting minimal contamination by other tissues. Cartilage samples were then lyophilized and processed for liquid chromatography-mass spectrometry, such that peptides were isolated in 6 fractions. Protein searches were performed using Proteome Discoverer 2.5 (Thermo Scientific, Waltham, MA) utilizing the Sequest HT search algorithm with the mouse proteome. Peptide quantification is based on TMT MS2 reporter ions generated from fragmentation. Normalization was performed at the peptide level, and protein ratios were calculated from grouped ion abundances, with protein FDR set to a maximum of 0.05. The mass spectrometry proteomics data are located on ProteomeXchange Consortium via the PRIDE partner repository62 with the data set identifier PXD024062 and 10.6019/PXD024062. Full methodology of the mass spectrometry sample preparation and analyses are available in the Supplementary Methods.

We used the Search Tool for the Retrieval of Interacting Genes database (STRINGdb) (32) as a means to evaluate the PPI across proteins that were significantly associated with cartilage aging. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses using the STRINGdb (32) were then performed to determine whether proteins associated with cartilage aging were significantly enriched in specific signaling pathways. In these analyses, we included significantly expressed (p < .05) proteins identified by the meta-analysis. To support the KEGG pathways driven by meta-analysis, the same analyses were performed using mass spectrometry-based proteomics data (young, 4–6 months vs aged, 21–24 months; male and female; n = 5/age/sex) available from 1 study (30). To confirm that expected age-related changes in cartilage integrity were detectable from the mass spectrometry data, we performed principal component analysis in which 155 extracellular matrix (ECM) proteins were included.

Overall Quality of Evidence

Two independent reviewers (H.I. and K.W.) assessed risk of bias, publication bias, and quality of evidence (GRADE) (26). Disagreements between the 2 reviewers were discussed until consensus was achieved. For GRADE assessment, the evidence quality was downgraded if (i) outcomes have a high risk of bias; we defined this as a lack of blinding outcome assessment in more than 50% of the included studies (risk of bias domain); (ii) heterogeneity between trials was more than substantial (I² ≥ 50%) with nonoverlapping 95% CI (inconsistency domain); (iii) sample size was inadequate; we defined this as optimal information size (33) and wide 95% CI that included 0 (precision domain); and (iv) publication bias existed, as identified by the Egger’s regression test (publication bias domain). The publication bias domain was applied only for outcomes with ≥10 studies. The indirectness domain was not considered, as all studies used mouse models and outcomes were not directly related to human clinical trials and clinical decisions. The quality of evidence was judged as “high,” “moderate,” “low,” or “very low.”

Results

A systematic search identified a total of 1 222 articles from electronic databases relating to KOA and aging, of which 41 met inclusion criteria for this systematic search (20–23,30,34–69) (Figure 1; Supplementary Table S1). The rationale for excluding studies during the full-text screening process is provided in Supplementary Table S2. The most common reason for exclusion was the lack of an “aged” comparison group over 18 months old. Supplementary Table S3 summarizes the characteristics (eg, outcome variables) of the included studies. The most commonly used murine strains were from a C57BL/6 background. The median ages of young, middle-aged, and aged mice were 5 months (interquartile [IQ] range: 3–6), 12 months (IQ range: 12–12), and 22 months (IQ range: 18–24), respectively. Of the 30 studies that reported the sex of the mice, only 7 studies included both male and female mice. The remaining 23 studies used either only male (22 studies) or female (1 study) mice. Outcome measurements in the included studies primarily examined the tibiofemoral joint. Of these, 1 study reported transcriptomic (23) and 1 study reported proteomic data over time (30). Aside from this 1 study, all protein data presented in the included studies were immunohistochemical in nature. Of note, throughout this manuscript, we use official gene symbols to refer to proteins (all caps) and genes (italicized), for example, for type X collagen: COL10A1 (protein) and Col10A1 (gene).

Figure 1.

Flow diagram of literature search results. Electronic database and Google Scholar searches yielded a total of 161 studies. After duplicates were removed (n = 1), the titles and abstracts of 160 studies were screened, and the remaining 96 studies were assessed for eligibility by full-text screening. From full-text screening, 33 studies met eligibility criteria, and citation search from the 33 studies identified 8 additional studies. Ultimately, a total of 41 studies were included. Of 41 studies, 1 study included RNA-seq data (23) and 1 study included mass spectrometry-based proteomics data (30).

Age-Induced Cartilage Degeneration is Accompanied by Elevated Inflammation, Impaired Autophagy, and Cellular Senescence

A meta-analysis of 16 histological studies confirmed the expected age-related increase in cartilage degeneration, defined as a progressive loss of cartilage ECM over time (Figure 2A and B). These findings were further supported by mixed linear regression analysis (Figure 2C). Of note, age-related cartilage degeneration was generally less severe compared to the anterior cruciate ligament-rupture PTOA model (Figure 2C) (70). From 10 studies, gross morphologic alterations were supported by non-pooled histology data (i.e., data that were mot included for meta-analysis because of methodological heterogeneity in the studies) (Supplementary Table S4). Data were also associated with lower cellularity and a higher percentage of apoptotic chondrocytes as well as lower antiapoptotic activity (34,46,48). Further supporting evidence of age-related cartilage degeneration, 3 studies evaluated matrix metalloproteinase 13 (MMP13) and/or type X collagen (COL10A1) (43,46,64), 2 markers of hypertrophic chondrocyte activity (71). Aged cartilage showed higher MMP13 and COL10A1 protein levels compared to young cartilage (43,46,64). One other study demonstrated that aged cartilage showed higher levels of MMP3 (40), a well-known marker of synovial inflammation in rheumatoid arthritis that is involved in the breakdown of the ECM of cartilage (72).

Figure 2.

Progression of cartilage degeneration with aging. (A and B) Effect size for cartilage morphology and age-related cartilage degeneration compared to young (A) and middle-aged cartilage (B). The forest plot displays relative weight of the individual study, standardized mean differences (SMDs), and 95% confidence interval (CI). The red diamond indicates the global estimate and its 95% CI. The green bar indicates the prediction interval. (C) The trajectory of age-related alteration in cartilage morphology (standardized articular cartilage [AC] score) in individual studies. p value of linear mixed-effect model is provided. As reference values, we also provide the standardized AC score in anterior cruciate ligament-rupture posttraumatic osteoarthritis (OA) model at week 1 and week 8 (70).

Previous epidemiological and biological evidence suggests a link between age-related systemic inflammation and development of OA in humans (73). Given that human articular chondrocytes from older adults release higher levels of pro-inflammatory cytokines than young counterparts (74), we searched for evidence of local inflammatory factors in aged cartilage. Although protein-level evidence of local inflammation in aged murine samples is generally lacking, 1 study did demonstrate that interleukin 36 receptor antagonist, an activator of Nuclear factor kappa-light-chain-enhancer of activated B cells (NF-kB) signaling (75), expression was reduced with aging (52).

As a next step toward a more comprehensive view of KOA, we assessed global transcriptomic changes across the life span of mice. Only 1 study to date has performed RNA-seq on whole-joint tissue samples from different age groups (23). This study demonstrated that aging results in upregulation of inflammatory response-related genes and downregulation of genes associated with cartilage development and homeostasis (23). To further assess the functions of the transcript products at the systems level, we accessed the study with archived RNA-seq data (23) and performed gene set enrichment analysis (Figure 3A). The data revealed that aging upregulated inflammation-associated biological processes, such as “antigen receptor-mediated signaling pathway” and “regulation of interleukin-8 production,” while ECM-related biological processes including “ECM organization” were downregulated with aging (Figure 3B; see Supplementary Tables S5–S8 for details). Notably, these molecular changes were evident at middle-age (Figure 3C).

Figure 3.

Transcriptomic signature in aged knee joints shows upregulated inflammation and downregulated extracellular matrix (ECM) synthesis. (A) Analysis flow. Normalized RNA-seq data from 1 original study (23) were used for gene set enrichment analysis (GSEA). REVIGO was used to summarize redundant pathway. (B) Gene ontology (GO) enrichment tree-map of biological processes for transcripts upregulated or downregulated between young versus middle-aged and middle-aged versus aged. Each rectangle represents a supercluster GO, visualized with different colors. Size of rectangles was adjusted to reflect the p value of the GO term calculated by Top GO (ie, the larger the rectangle, the more significant the GO term). (C) Trajectory of inflammation and ECM synthesis pathways predominantly identified by the GO enrichment tree-map. (D) GO enrichment analysis from aging knee osteoarthritis (KOA)-specific genes.

To determine whether the upregulated inflammation and downregulated ECM-related pathways are unique to age-related KOA, we accessed archived RNA-seq data of age-related KOA and anterior cruciate ligament-rupture PTOA models (23) and compared differentially expressed (DE) transcripts across the 2 groups (Supplementary Figure S1). With the goal of identifying genes that contribute to the trajectory of change in age-related KOA, we focused on overlapping DE genes when comparing young versus middle-aged and middle-aged versus aged groups (yielding 1 173 genes; Supplementary Figure S1). Similarly, for PTOA data, we focused on overlapping DE genes across all available time points (uninjured vs PTOA day 1, week 1, week 2, and week 6; Supplementary Figure S1). Notably, only 35 (2.9%) of the 1 173 DE transcripts associated with age-related KOA overlapped with the PTOA model, suggesting that aging and PTOA have markedly distinct transcriptional profiles (Figure 3D). GO enrichment analysis revealed that “inflammatory response” was uniquely identified in the age-related KOA model (Figure 3D; Supplementary Figure S1). In contrast, “inflammatory response” was not significantly upregulated in the DE genes associated with the PTOA model (p = .246), which is inconsistent with the general understanding that PTOA is associated with inflammatory cascades (76).

To address the possibility that differences in the transcriptomic signatures of the 2 models reflect disease severity rather than distinct pathophysiological mechanisms, we performed a sensitivity analysis in which we limited DE genes to day 1 in the PTOA model (1 177 genes; Supplementary Figure S2). This is the earliest time point for transcriptomic data available. PTOA knees at day 1 displayed minimal structural changes in cartilage (77). Consistent with our primary analysis, “inflammatory response” was uniquely identified in the age-related KOA model but not in the PTOA model (inflammatory response in PTOA compared to control, p = .311). To rule out the possibility that highly restrictive gene selection for GO enrichment analysis may have missed inflammatory genes associated with PTOA, we performed another sensitivity analysis that considered all DE genes across all time points for the age-related KOA (3 876 DE transcripts) and PTOA (2 262 DE transcripts) models (Supplementary Figure S3). Again, “inflammatory response” was not significantly associated with genes uniquely expressed in the PTOA model (p = .859). It is worthwhile to note that, in these 2 sensitivity analyses, “inflammatory response” was, however, significantly increased for DE transcripts shared across aging and PTOA models (p = .002 in both sensitivity analyses). These data indicate that, while both aging and PTOA models display alterations in inflammatory responses, the aging KOA model displays distinct inflammatory responses which are driven by genes unique to the aging process. The distinct mechanistic trajectories when comparing PTOA versus age-related KOA are also consistent with the divergent clinical phenotypes between the two pathologies (78).

Considering the growing body of evidence related to the role of inflammation–autophagy–senescence in the pathogenesis of KOA (79), we next evaluated alterations in autophagy and senescence as a function of time. These data were generated from three individual studies investigating protein-level changes (20,60,80). Two studies showed that impaired autophagy (as determined by decreased autophagy related 5) and decreased resistance to oxidative stress (as determined by decreased heme oxygenase 1) were observed in aged cartilage (60,80). Although no study has quantified the trajectory of chondrocyte senescence over time, one study found synovial fluid in aged knee joints contained elevated senescence-related microRNAs, such as miR-34-5p (20). The increase in senescence-related microRNA generally supports a previous study, which demonstrated that p16, a marker of senescence, was increased in hip joint chondrocytes of aged mice (81). Increased production of pro-inflammatory mediators is a prominent feature of the senescence-associated secretory phenotype (73), which further highlights the need to evaluate the pro-inflammatory cytokines in the context of aging KOA.

Construction of PPI Networks from Data Across Individual Studies and Validation by Mass Spectrometry Data Analysis Reveals Enrichment of the AGE–RAGE Signaling Pathway

Complex signal transduction occurs through the dynamic modulation of PPIs, which play an important role in the development, onset, and progression of disease (82). Our meta-analysis identified 25 DE proteins (Supplementary Table S9) evaluated across 11 articles (35, 37,43,46,48,52,55,57,59,60,69). To determine whether these proteins converge to specific pathways, we used the KEGG enrichment analysis, available in the STRINGdb (32) (Figure 4A). Given the quality of evidence of each protein was “very low” (15 proteins) or “low” (10 proteins) according to the GRADE (26) (Supplementary Table S10), we also performed a sensitivity analysis focusing on 10 DE proteins with higher GRADE (i.e., higher quality of evidence (83)). These two different KEGG enrichment analyses consistently identified transforming growth factor beta (TGF-B) signaling and AGE–RAGE signaling in diabetic complications as falling within the top 5 pathways identified (Figure 4A). Supplementary Figures S4–S5 contain detailed information on the risk of bias and publication bias assessments used for the GRADE assessment. Briefly, rigor and reproducibility were rated as low mostly because of a high risk of bias and imprecision (i.e., inadequate sample size). Overall, all domains in the risk of bias were judged as “unclear of risk of bias” when sufficient details to assess the criterion were lacking.

Figure 4.

Advanced glycation end product–receptor for AGE (AGE–RAGE) signaling pathway involved in cartilage aging. (A) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis to identify specific pathways associated with cartilage aging based on 25 differentially expressed (DE) proteins (p < .05) between young versus aged identify by previous articles. As a sensitivity analysis, we excluded 15 proteins of lower Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) score. The interactome network was generated by Search Tool for the Retrieval of Interacting Genes database (STRINGdb) for DE proteins between young versus aged. Each node represents a protein identified from meta-analysis and each line represents an interaction. The color of nodes refers to the GRADE score; red represents low GRADE and blue represents very low GRADE. Of note, 1 protein (BAG1) is not visualized because of no interaction with other proteins. (B) Mass spectrometry of the DE proteins between young versus aged C57/BL6 mice to support the 2 pathways identified by meta-analysis (ie, TGF-beta and AGE–RAGE signaling pathways). We performed KEGG enrichment analysis and identified AGE–RAGE signaling pathway in male mice. (C) Visualization of AGE–RAGE signaling pathway. Each node represents a protein, and the red node represents the DE proteins (p < .05) confirmed with cartilage aging in male mice.

To further evaluate the involvement of the 2 identified pathways by meta-analysis, we accessed archived mass spectrometry proteomic data (30). These data were obtained from the cartilage tissue of young (4 months old) and aged (21 months old) C57/BL6 male and female mice (n = 5/age/sex; Figure 4B) and are available via ProteomeXchange under identifier PXD024062. Mass spectrometry analysis identified a total of 44,689 peptides associated with 6,694 unique proteins (30). We confirmed the expected age-related changes in cartilage ECM proteins including COL2A1 (Supplementary Figure S6), suggesting that the mass spec data can detect age-related cartilage integrity alterations. KEGG enrichment analysis using the STRINGdb (32) confirmed immunohistochemistry and transcriptomic findings that the AGE–RAGE signaling pathway in diabetic complications was the most robust pathway significantly changed over time (Figure 4B). These findings are further supported by the RNA-seq data showing that aged joints display an altered “inflammatory response” (Figure 3B–D), which is a known trigger for AGE accumulation and downstream RAGE signals (Figure 4C). Notably, changes were observed in male, but not female, mice (Figure 4B). Although AGE–RAGE signaling is a well-known signaling cascade involved in diabetes complications (84,85), these results indicate that the AGE–RAGE signaling pathway may also be involved in age- and sex-dependent cartilage degeneration.

Discussion

Using an integrated approach of meta-analysis together with multi-omics analyses of publicly archived data, this study synthesized age-related changes in murine knee joint articular cartilage integrity based on available evidence from 41 different murine studies (Graphical abstract). By excluding PTOA models, our findings provide mechanistic insights into the unique pathways associated with aging, the most common risk factor for KOA (10). Consistent with clinical reports of cartilage structural abnormalities in people aged 45–55 years old (86), meta-analysis revealed that age-related cartilage degeneration was apparent in middle-aged mice, both histologically and at the transcriptional level. Subsequent bioinformatic analysis of archived RNA-seq data (23) revealed that the onset of impaired cartilage integrity at middle-age was accompanied by upregulation of genes associated with inflammation, a prominent feature in age-related KOA. Importantly, the molecular profiles of the inflammatory response with aging were unique compared to those observed following PTOA, even after adjusting for disease severity. Early changes in inflammatory processes and ECM regulation at the transcript level were further supported by a previous microarray study that showed inflammation-related genes were upregulated while ECM-related genes were downregulated in middle-aged mice (87). Next, construction of PPI networks from data across 11 individual studies, which included 25 DE proteins, revealed enrichment of AGE–RAGE signaling pathway. Analysis of archived mass spectrometry data (30) further implicated AGE–RAGE as the most robust pathway associated with age-related KOA.

We appraised and collated data from 41 available murine studies to provide insight into unique pathways associated with age-related KOA using multi-scale (RNA, protein, and tissue scales) with multi-omics approach. Meta-analysis revealed that age-related progressive cartilage degeneration was apparent in middle-aged mice (tissue scale), which is accompanied by transcriptomic signature of aging-driven upregulated inflammation (RNA scale). Our approach of bioinformatics with metaanalysis of aging-associated proteins revealed altered AGE-RAGE signaling as a key driver of age-related KOA (protein scale). Most notably, mass spectrometry revealed the AGE-RAGE signaling pathway to be identified only in male mice, highlighting the importance of considering sex as a biological variable in clarifying the pathogenesis of age-related KOA. This evidence indicates that AGE-RAGE signaling pathway may be uniquely involved in age- and sex-dependent cartilage degeneration with the elevated inflammatory response.

Animal studies typically focus on individual proteins or pathways, thereby precluding a holistic understanding of disease pathogenesis. Here, through an integrated approach of meta-analysis of aging-associated proteins and RNA-seq analysis, we identified cellular senescence, inflammation, and autophagy as major hallmarks of KOA in aged animals. Identification of these hallmarks is consistent with previous studies demonstrating that expression of the p16 gene, a major regulator of the senescent program (88), was upregulated in aged murine cartilage (81). Moreover, it has been shown that removing senescent cells partially reversed the effect of time on cartilage degeneration (89). At the nexus of these three hallmarks, we found that AGE–RAGE signaling emerged as a major driver for age-related KOA. Evidence for the involvement of the AGE–RAGE pathway in murine knee cartilage supports the accumulation of AGE in aged human knee cartilage (90). Additionally, in vitro studies have shown that accumulated AGE suppresses proteoglycan synthesis in mouse chondrocytes (91), a hallmark of age-related cartilage degeneration. These previous findings suggest that AGE–RAGE signaling may be a mechanistic driver for age-related KOA. Given that few transcripts overlapped between the age-related KOA and PTOA models, we found AGE–RAGE signaling to be associated with age-related KOA but not PTOA. In support of these findings, proteomics data revealed that cartilage in PTOA porcine knees was not significantly associated with AGE–RAGE signaling (92). Still, it should be noted that knockout of RAGE did decrease cartilage degeneration after surgically induced PTOA in mice (93), suggesting that further studies are warranted.

Although the signaling cascade involved in the AGE–RAGE pathway is not fully understood in age-related KOA, RAGE signaling activates inflammatory pathways and increases MMP13 expression in human chondrocytes (94–96). These findings are consistent with the fact that AGE increases the stiffness of the collagen network in human knee cartilage (90), altering mechanotransductive-dependent chondrogenic pathways (97). In addition, the accumulated AGE may compromise maintenance of cartilage integrity in the context of injury. This possibility is supported by evidence that aged cartilage displays high susceptibility to mechanical overloading compared to young cartilage (87). This evidence from murine and human studies indicates that limiting AGE accumulation may be a promising therapeutic target to prevent or slow the progression of age-related KOA. Despite the demonstrated role of AGE accumulation in cartilage aging, the mechanisms leading to AGE synthesis and accumulation in aged cartilage remain poorly understood. The current study supports the previous in vitro and human studies and highlights age-related inflammation as a potential trigger for AGE accumulation in cartilage. It is well established that aging and age-associated diseases display chronic and low-grade inflammation, and that this low-grade inflammation leads to tissue damage and degeneration (98). This cascade is distinct from trauma-related, acute, and transient inflammation (98). It should be noted that the distinct transcriptomic signature across the aging and PTOA models is consistent with our sensitivity analysis adjusting for disease severity, suggesting that disease severity cannot entirely explain the different inflammatory responses across the two models. Investigation into the mechanistic link between age-related inflammatory response, AGE accumulation, and cartilage degeneration is an interesting area for future studies.

In addition to the mechanistic insights gained through integrating meta-analysis and a systems biology approach, our work also highlights the need for study designs that ensure methodological rigor and consider sex as a biological variable. Despite the fact that women have a greater risk of developing KOA than men (99), the overwhelming majority of studies investigating age-related KOA only included male mice. While we found no published studies directly evaluating the pathogenesis of age-related KOA according to sex, two studies considering sex in the pathogenesis of PTOA demonstrated that male mice displayed greater cartilage degeneration than female counterparts (100,101). Although the majority of female mice do not experience menopause (102) and therefore may not be a reliable model of human postmenopausal KOA, sex as a biological variable is clearly an essential consideration as the field strives to increase rigor, promote discovery, and expand the clinical relevance of age-related KOA murine models. Indeed, mass spectrometry-based proteomic analysis identified enrichment of the AGE–RAGE signaling pathway only in males. These findings are consistent with a clinical study demonstrating that AGE was associated with KOA-related joint space narrowing in men but not in women (103). We also acknowledge that male mice generally display more severe cartilage degeneration than female mice, which is also confirmed in the study we accessed for the mass spectrometry proteomics data (30). Therefore, we cannot rule out that higher disease severity in male mice may have contributed to higher enrichment on AGE–RAGE signaling compared to female mice.

Although this study provides a new perspective to the pathogenesis of KOA, it has limitations. Our meta-analysis and systems biology approach were based on a small number of studies, which may contribute to bias depending on the methodology used in the original studies. Also, RNA-seq data were generated from whole-joint tissue samples (e.g., cartilage, bone, meniscus, ligament, synovium, and fat), rather than isolated cartilage (23). Therefore, we cannot discount the possibility that elevated inflammation in aging KOA mice may be attributed to noncartilage joint tissue. The generalizability of our findings to cartilage biology is, thus, unclear. However, it is well established that KOA is a disease of the whole joint rather than solely a cartilage disease, and findings from the RNA-seq analysis suggest that elevated inflammation is a feature of aged joint tissue pathology. Additionally, given the lack of female mice in current studies, we were not able to clarify mechanisms underlying age-related KOA in female mice, and this remains an important area for future work. It should also be acknowledged that sex-dependent enrichment of AGE–RAGE signaling, which was validated via mass spectrometry analysis, may be specific to the C57BL6 strain given that males display a higher susceptibility to the development of diet-induced obesity and insulin resistance than females (104). Another limitation is that most protein data in our meta-analysis were converted from graphical data to numerical data. Although our approach displays excellent inter-rater reliability (agreement of 0.999), there is the potential for bias in these analyses. Finally, in the assessment of protein data, we did not account for the cartilage region analyzed (e.g., medial tibia vs lateral tibia) or the timing of sacrifice, since these details were not available from the included studies. The timing of sacrifice is of interest because it has been shown that circadian changes influence the cartilage proteome (105).

According to both histological observation and transcriptomic analysis, data reveal that the pathogenesis of KOA in natural-aging murine models follows a similar relative timeline to that observed in humans. Our findings also suggest that declines may be driven initially by increased inflammation, and the AGE–RAGE signaling pathway may be uniquely involved in age-related cartilage degeneration in male mice. The notable differences in transcriptional profiles when considering age-related KOA versus PTOA, even after adjustment for disease severity, suggest that, in preclinical models, increased attention should be paid to the differences between age-associated pathology and trauma-related pathology. Indeed, our findings are in line with subgroupings among clinical phenotypes (78). Finally, we identified a lack of consideration of sex as a biological variable as an important shortcoming of the current state of the science, which has likely contributed to delays in translational efforts. Taken together, this work highlights the need for scientists to adopt targeted steps to enhance scientific rigor in the field in order to accelerate the pace of discovery toward the development of safe and effective KOA interventions.

Funding

This study was supported in part by (i) a Grant-in-Aid from the Japan Society for the Promotion of Science for Overseas Research Fellowships for H.I. (grant no. N/A), (ii) National Institute of Aging of the National Institutes of Health R01AG052978, National Institute of Aging of the National Institutes of Health R01AG061005, and National Institute of Aging of the National Institutes of Health R01AG066198-01 to F.A., (iii) the National Institute of General Medical Sciences of the National Institutes of Health under Award Number T32GM008208 for G.G., and (iv) the John and Posy Krehbiel Professorship in Orthopedics for C.E. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conflict of Interest

None declared.

Author Contributions

H.I., G.G., Y.M., and F.A. provided the concept, idea, and experimental design for the studies. H.I., G.G., and F.A. wrote the manuscript. H.I., G.G., K.W., S.S., C.E., Y.M., and F.A. provided data collection, analyses, interpretation, and review of the manuscript. H.I., G.G., and F.A. obtained funding for the studies.