Exercise-primed extracellular vesicles improve cell-matrix adhesion and chondrocyte health

Extracellular vesicles (EVs) have been suggested to transmit the health-promoting effects of exercise throughout the body. Yet, the mechanisms by which beneficial information is transmitted from extracellular vesicles to recipient cells are poorly understood, precluding a holistic understanding of how exercise promotes cellular and tissue health. In this study, using articular cartilage as a model, we introduced a network medicine paradigm to simulate how exercise facilitates communication between circulating EVs and chondrocytes, the cells resident in articular cartilage. Using the archived small RNA-seq data of EV before and after aerobic exercise, microRNA regulatory network analysis based on network propagation inferred that circulating EVs activated by aerobic exercise perturb chondrocyte-matrix interactions and downstream cellular aging processes. Building on the mechanistic framework identified through computational analyses, follow up experimental studies interrogated the direct influence of exercise on EV-mediated chondrocyte-matrix interactions. We found that pathogenic matrix signaling in chondrocytes was abrogated in the presence of exercise-primed EVs, restoring a more youthful phenotype, as determined by chondrocyte morphological profiling and evaluation of chondrogenicity. Epigenetic reprograming of the gene encoding the longevity protein, α-Klotho, mediated these effects. These studies provide mechanistic evidence that exercise transduces rejuvenation signals to circulating EVs, endowing EVs with the capacity to ameliorate cellular health even in the presence of an unfavorable microenvironmental signals.


INTRODUCTION
Extracellular vesicles (EVs) have emerged as a potent mechanism through which the bene cial effects of physical exercise are transmitted throughout the body 1 . EVs are nanovesicles that contain functional biomolecular cargoes, such as lipids, proteins, and nucleic acids 2 , which are highly responsive to physiological stressors and stimuli 3,4 . Several studies to date have shown that aerobic exercise modulates the bioactivity of circulating EVs, rendering EVs therapeutically bene cial in the context of disease such as cardiac ischemia-reperfusion injury, obesity, and type 2 diabetes mellitus 5,6 . However, the mechanism by which exercise alters the biological activity of circulating EVs and affects homeostasis of recipient cells remain unclear.
In silico computational approaches are useful for inferring the biological impact of EV biomolecules on recipient cells. This is particularly true for EV microRNAs (miRNAs), given that miRNAs can regulate posttranscriptional gene expression through complementary base pairing with mRNAs 7 . As such, the miRNA regulatory network has been used as a tool for dissecting the complex regulation by EVs on target biological processes 8, 9 . Previous miRNA regulatory network analyses, however, have lacked consideration of the downstream secondary response of miRNA target genes on the global gene regulatory network. To simulate these downstream effects of EV miRNAs, network propagation is a powerful tool for elucidating the global transcriptional response induced by miRNAs. Network propagation on miRNA regulatory networks is further strengthened by consideration of tissue speci city given that the precise actions of genes are frequently dependent on the tissue in which the cells reside 10 . Indeed, human diseases result from the disordered interplay of tissue-speci c processes 11 . The tissue-speci c functional gene network clari es biological functions of given genes to help identify disease driver genes that may exert important functional roles according to the tissue of interest With this in mind, this study introduced network medicine paradigms to simulate EV-mediated miRNA regulatory network perturbation in recipient cells and uncovered cellular rejuvenation effects induced by exercise-primed EVs in vitro. This study simulated the effects of EVs on chondrocyte aging, a precursor for knee osteoarthritis (KOA). KOA represents a common age-related pathology that typically drives considerable functional declines. The well-established use of exercise for the prevention and treatment of KOA, as recommended by clinical practice guidelines 12 , further supports the use of this disease model to examine the downstream role of aerobic exercise-primed EVs.

Aerobic exercise upregulates miRNAs that modify chondrocyte-matrix adhesion processes
Most studies assessing the effects of exercise on EV biomolecules to date have focused on isolated miRNAs of interest, thereby precluding a holistic understanding of exercise-sensitive EV miRNAs and their downstream effects on recipient cells. To address this issue, we rst performed a systematic literature search (Fig. 1A) and summarized EV miRNAs that are differentially regulated by exercise in older adults, as quanti ed by small RNA-seq. We identi ed only one study that performed small RNA-seq on circulating EVs in our target population, older adults, before and after exercise 3 . In that study, male participants (n = 10, mean age: 69.3 years old) completed an acute bout of aerobic exercise on a cycle ergometer for 40 minutes at under 70% of their maximum heart rate ( Fig. 1A) 3 . Circulating EVs were then isolated from the serum collected before exercise, immediately after exercise, and after 3 hours of recovery. Participants were separated into two groups based on baseline physical activity level (sedentary vs trained; n = 5 per group). The authors demonstrated that aerobic exercise primarily changed miRNAs targeting the IGF-1 signaling pathway in an exercise-dependent manner 3 . To assess the global EV miRNA response to acute aerobic exercise, we accessed and evaluated the raw small RNA-seq data (GSE144627) from all 10 participants in the study. We then applied a lter as proposed by Chen et al 13 in order to account for the bias caused by low expression count data. This ltering resulted in 442 miRNAs for analysis. Using this re ned data set from 10 participants, we identi ed 16 differentially (log2 fold > 1.0, p-value < 0.05) expressed miRNAs (Fig. 1B). Of the identi ed miRNAs, 15 were signi cantly upregulated after aerobic exercise. Among these 15 miRNAs, two out of three miRNA29 family members, miR29a-3p and miR29b-3p, were signi cantly upregulated (Fig. 1B). The miRNA29 family is the "master bromiRNA regulator" which targets at least 16 genes related to the extracellular matrix, including a large number of collagen isoforms and Itgb1 14 . This function is unique to miRNA29 given that no other miRNA targets more than actions in varied animal models of tissue brosis 16-18 , including chemically induced synovial brosis in murine knees 19 . The upregulated miR29a-3p and miR29b-3p after aerobic exercise is in line with the general understanding that exercise elicits anti-brotic effects 20 .
Next, we sought to determine the primary target genes of the 16 miRNAs. For this purpose, we used two different databases, miEAA.2.0 21 and miRTarBase 22 , for target gene prediction (Fig. 1C). The integration of these two databases allows for determination of robust target genes that have been experimentally validated. The analyses identi ed 27 genes implicated in the process of cell-matrix adhesion, including Itgb1 and Col4a1 23 , both of which are targets of the miR29 family 14 . Of note, when miR-4433b-3p, the only the miRNA downregulated by exercise, was used as an input, no target gene was consistently predicted. As such, all of the identi ed 27 target genes were the products of the remaining 15 miRNAs that were upregulated after aerobic exercise.
We then evaluated the biological function of these primary target genes. Given our focus on KOA, we generated a cartilage-speci c functional gene network from the 27 target genes using HumanBases software 24 (Fig. 1D). A functional relationship implies that two given genes participate in the same biological process in a speci c tissue or functional context 10 . The tissue-speci c functional gene network clari es the biological function of genes in a tissue speci c-manner 10 , thus helping identify biological processes that may exert a functional role in only the selected tissue(s) 11,25 . The generated cartilagespeci c network revealed gene ontology (GO) terms related to integrin-mediated cell-matrix adhesion processes, including "positive regulation of cell-substrate adhesion" and "cell-substrate adhesion" (Fig. 1E). In articular cartilage, altered cell-matrix interactions initiate mechanotransduction processes, activating downstream signals that direct the transcription of genes involved in the regulation of cellular and tissue homeostasis 26 . Together, these in silico analyses inferred that miRNAs signi cantly upregulated after aerobic exercise primarily direct modi cation of chondrocyte-matrix adhesion and initiate downstream secondary effects.
Perturbation of the cartilage-speci c miRNA regulatory network implicates exercise-primed EVs in the regulation of cartilage aging To infer the downstream effects of the upregulated EV miRNAs on signaling cascade in articular cartilage (i.e., a downstream effect of chondrocyte-matrix adhesion), we implemented a network propagation approach to the cartilage-speci c global functional gene network ( Fig. 2A). Network propagation explores the network vicinity of genes of interest to study their functions based on the premise that nodes with similar functions tend to lie close to each other in the networks 27 . The network paradigm introduced here assumes an interaction between articular cartilage and circulating EVs that have penetrated into the joint cavity through the synovium, where intercellular gaps of the blood-joint-barrier (0.1 ~ 5.5 µm) are greater than EVs (~ 0.15 µm) 28 . To experimentally con rm the ability of circulating EVs to penetrate the joint cavity, labeled EVs were systemically administered to aged (21 months old) mice via intravenous injection. A localized uorescence signal in the knee was observed 48 after systemic injection (see inset in Fig. 2A).
To simulate the EV-cartilage communication, we constructed a global cartilage-speci c network using HumanBases software 24 . With this global network in hand, we then performed a random-walk-restart (RWR) algorithm 29 to verify which gene nodes are most frequently visited on a random path in the cartilage-speci c network. The aforementioned 27 target genes were used as seeded (starting) genes.
Through this network medicine approach, we sought to determine the functional changes (i.e., KEGG pathways 30 ) after miRNA perturbation (Fig. 2B). Figure 2C shows an example of in silico activation of Itgb1, a primary target of miR29a-3p and miR29b-3p 14 , on the network. RWR applied to a cartilagespeci c network revealed that genes pseudo-activated (i.e., an a nity score > 0) by the 27 EV miRNAs target genes were highly associated with "PI3K/Akt signaling", "TGF-beta signaling", "Focal adhesion", and "ECM-receptor interaction" (Fig. 2D), pathways that have been implicated in the pathogenesis of KOA 31 . To show the relationship between the functional targets and cartilage aging, we de ned eight signaling pathways associated with cartilage aging according to mass-spectrometry proteomics data across the lifespan of male murine knee joints 32 . Logistic regression analysis revealed that functional targets with higher enrichment by network propagation displayed a signi cantly higher probability of an aged phenotype (Fig. 2E). These ndings indicate that exercise-primed EV miRNAs secondarily regulate the cartilage aging process.
Studies have shown that aging changes biophysical properties of the ECM, a quintessential feature of aged tissue and a recently proposed hallmark of aging 33,34 . This is particularly true in articular cartilage, as evidenced by the ndings that aged cartilage displays stiffness values that are more than twice the value of their young counterparts (E = ~ 50-100kPa in aged cartilage versus E = ~ 1-30kPa in young) 35 . A stiff ECM initiates, ampli es, and/or perpetuates aberrant ECM remodeling during aging 36 . This, in turn, disrupts chondrocyte function over time via mechanotransductive pathways 32,37−39 , eventually leading to KOA. Recent ndings revealed that the age-related increased ECM stiffness drives healthy young murine chondrocytes to an aged phenotype via adhesion-mediated mechanotransduction and epigenetic regulation of the gene encoding the longevity protein, Klotho 32 . Chondrocyte phenotypic changes were accompanied by morphological changes, which was attributed, at least partly, to alterations in focal adhesion kinase levels 32,40 . Building from this previous work, our current ndings from in silico analyses led us to ask two questions ( Fig. 2F): (1) do aerobic exercise-primed EVs counteract chondrocyte morphological changes as a surrogate measure of cell-matrix adhesion, and (2) can aerobic exerciseprimed EVs counteract the compromised chondrocyte health induced by an aged-like (stiff) ECM? The experiments that follow were designed to address these questions.
Exercise-primed EVs overrode chondrocyte morphological alterations induced by a stiff ECM As a rst step to address these questions, we conducted an exercise study in older individuals. Circulating EVs were isolated from older adults ages 65-85 years old (n = 4, 50% women) who completed 3-months of an aerobic exercise protocol that aimed to achieve a moderate-intensity based on perceived exertion (Fig. 3A). The exercise protocol was prescribed for 5 days per week and consisted of a combination of 1 supervised and 4 home-based, non-supervised sessions per week, which progressed from 70 minutes per week to 150 minutes per week across the intervention period. To facilitate engagement in the aerobic sessions, participants were provided with activity videos on a tablet that were developed by the study team. Exercises were taught in the supervised sessions and then the videos were used to facilitate the home-based, non-supervised sessions.
EVs were isolated using size-exclusion chromatography before and after 12 weeks of exercise exposure (EV pre and EV post , respectively), using a previously described protocol 41 . EV purity was con rmed by western blotting for the presence of EV-speci c markers, CD81, and the absence of non-EV marker, GM130 ( Figure S1). As determined by NanoSight nanoparticle tracking analysis, average EV size and number were similar across the two groups ( Fig. 3B-D), which is consistent with ndings from previous exercise cohorts 42 .
EV pre or EV post were then co-cultured with aged human chondrocytes seeded onto a substrate that mimics the matrix stiffness of aged human cartilage (100kPa) (Fig. 3E) 35 . We used this experimental design to test the ability of the exercise-primed EVs to counteract age-related pathogenic mechanotransduction induced by aged-like stiff substrate. Prior to analysis, we con rmed the rapid uptake of uorescentlylabeled EVs by aged human chondrocytes (Fig. 3F). This is consistent with a previous study demonstrating that the uptake of circulating EVs by recipient C2C12 occurs within 60 minutes 43 .
To interrogate the ability of exercise-primed EVs to modulate chondrocyte-matirx adhesion processes, we assessed morphological alterations of chondrocytes cultured on a stiff substrate that were exposed to either EV pre or EV post (Fig. 3G). Speci cally, we quanti ed chondrocyte morphology across 53 morphological variables using Cell Pro ler software 44 . Cell Pro ler simultaneously measures cell size, shape, intensity, and texture in a high-throughput manner. Figure 3G provides representative images of chondrocyte morphology across the different groups. We posited that exercise-primed EVs would interfere with the aged phenotype induced by a stiff matrix and drive a more youthful chondrocyte morpheme that resembles chondrocytes seeded on a soft substrate 32 . As such, the features of chondrocytes cultured on soft substrate were used as a reference value of a youthful pro le.
Principal component analysis (PCA) analysis con rmed signi cant differences between the morphology of chondrocytes treated with EV pre compared to those treated with EV post (Fig. 3H). Notably, chondrocytes seeded on a stiff matrix treated with EV post approximated cells cultured on a soft substrate (Fig. 3H), suggesting that the in uence of EV post on aged chondrocytes mimicks that of a soft (young-like) substrate. Figure 3I summarizes the changes in individual morphological features. Of the 19 morphometric variables signi cantly altered by EV post , the majority (12 [63.2%]) overlapped with the effect of a soft substrate (Fig. 3I). Among the individual morphogical variables of interest, we found that EV post increased chondrocyte "formfactor" (Fig. 3J). "Formfactor" is a metric of cellular sphericity or roundness and was the feature previously identi ed in chondrocytes as highly sensitive to mechanical input 32 . On the other hand, aged chondrocytes treated with EV pre did not display signi cant morphometric changes and more closely resembled untreated cells cultured on stiff substrate. Together, these results suggest that exercise-primed EVs can override the effects of a stiff matrix on chondrocyte morphology and support the network-based inference for the primary target of EV miRNAs (i.e., chondrocyte-matrix adhesion) (Fig. 1C-E).
Exercise-primed EVs counteract the compromised chondrocyte integrity and Klotho promoter hypermethylation induced by a stiff ECM To next evaluate the ability of exercise-primed EVs to promote chondrocyte health, we repeated the above in vitro experiment, this time testing whether exercise-primed EVs can restore a more youthful chondrogenicity to aged cells. We found that aged chondrocytes treated with EV post displayed increased type II collagen and Sox9 expression, two well-established markers of chondrocyte health (Fig. 4A). On the other hand, consistent with our morphometric analysis ( Fig. 3G-J), aged chondrocytes treated with EV pre displayed type II collagen and Sox9 levels comparable to cells without treatment (Fig. 4A). These ndings further verify the network-based inference for the downstream target of EV miRNAs (i.e., cartilage aging process) ( Fig. 2D-F).
Finally, we explored candidates that may drive the rejuvenating effect of EV post on chondrocyte integrity. Our recent study demonstrated that a stiff substrate increases methylation of the Klotho promoter in chondrocytes 32 . Klotho is an upstream regulator of PI3K/Akt signaling 30 and a so-called longevity protein that has been shown to attenuate the effects of aging on articular chondrocytes 32 . In response to a stiff substrate, transcriptional machinery is recruited to the Klotho promoter in murine chondrocytes, inhibiting the downstream anti-aging cascade 32 . One such transcriptional machinery is Dnmt1, an enzyme that catalyzes the transfer of methyl groups to CpG dinucleotides, usually repressing gene transcription by altering chromatin structure and blocking the access of transcription factors at the gene regulatory region 45 . Intriguingly, we found that elderly chondrocytes treated with EV post , but not those treated with EV pre , displayed signi cantly decreased Dnmt1 levels (Fig. 4B). While EV post -treated chondrocytes also displayed homogenous global DNA demethylation (Fig. 4C), the decreased Dnmt1 level was accompanied by decreased Klotho promoter methylation (Fig. 4D) and upregulated Klotho mRNA (Fig. 4E). These cumulative ndings suggest that aerobic exercise-primed EVs can override the compromised chondrocyte integrity and Klotho promoter hypermethylation induced by a stiff ECM.

DISCUSSION
While studies have evaluated EVs as a potent means to transpose the systemic bene ts of exercise onto target cells, mechanistic evidence that exercise enhances the accumulation of health-promoting biomolecules in EVs in the context of aging has been lacking. This knowledge gap is particularly critical in the development of effective management for people with KOA given that currently available treatments focus exclusively on symptom management. As a rst step to address this knowledge gap, this study introduced a network paradigm approach to infer functional targets of exercise-primed EVs in recipient aged chondrocytes. This approach revealed novel rejuvenating effects of exercise-primed EVs on aged chondrocytes (see graphical abstract; Fig. 5). Using the archived small RNA-seq data from EVs, the network medicine approaches inferred cell-matrix adhesion as a primary target of exercise-primed EVs, initiating downstream signals that direct cartilage aging. In support of this analytical prediction, we demonstrated that EVs isolated after aerobic exercise in older individuals overrode alterations in chondrocyte morphology and chondrogenicity induced by stiff ECM, a typical feature of aged articular cartilage 35 . The youthful phenotype induced following treatment with EVs that were isolated postexercise was due, at least in part, to epigenetic reprogramming of Klotho. Collectively, the ndings of this study suggest a mechanism by which circulating EVs released during/following chronic aerobic exercise may inhibit the aging process of distal joints.
The signi cance of the network paradigm introduced used in this study is the inference of downstream targets of EV miRNAs in a tissue-speci c manner. This novel approach is based on the premise that the biological impact of EV miRNAs on target genes should propagate across the global network, not just across miRNA target genes, and that this impact is most effectively interpreted when considering a tissue-speci c network. For example, miR29a-transfected chondrocytes displayed an upregulation of Col2a1 46 , even though Col2a1 is not a direct target of miR29a 22 . This discordance highlights the limitation of miRNA target gene prediction to anticipate overall cellular responses. By implementing the network propagation approach on a tissue-speci c network, we found that EV miRNAs that are modulated by exercise direct signaling pathways relating chondrocyte aging. PI3K/Akt signaling, which we found to be signi cantly inferred by network propagation as downstream of EV effects, has been implicated in the pathogenesis of human KOA 47  Building on the in silico network-based inference, we demonstrated that exercise-primed EVs abrogated the deleterious effects of a stiff matrix on chondrocyte aging. It is well established that an age-related increase in matrix stiffness disrupts chondrocyte functionality via mechanotransductive pathways [37][38][39] .
A recent study showed that a stiff matrix epigenetically represses the gene encoding the longevity factor, α-Klotho, resulting in chondrocyte dysfunction, a leading cause of OA 32 . This study provides evidence that exercise-primed EVs promote a youthful phenotype in aged chondrocytes, as evidenced by improved chondrogenicity and epigenetic reprograming of Klotho. This nding suggest that EV-based approaches may serve as a viable treatment for age-related KOA. Currently, no FDA-approved drug is available to target ECM mechanics by preventing or reversing tissue stiffening or interrupting the cellular response, although the eld of mechanomedicine has been rapidly expanding 50 . In keeping with the spirit of this eld, the identi cation of biomolecules that drive the exercise-induced epigenetic reprograming of Klotho is an interesting future work.
Although this study provides novel mechanistic insight into anti-aging effects of exercise, it has limitations. First, this study did not address how exercise alters EV bioactivity. It is reasonable to hypothesize that mechanical stress imposed by exercise alters the biogenesis and bioactivity of EVs, a hypothesis that is supported by studies demonstrating that mechanical stimulation of mesenchymal stem cells and lung epithelial cells promoted EV release and the expression of miRNAs in EVs 51,52 .
Further research is needed to establish the mechanism by which exercise promotes the bioactivity of circulating EVs. Similarly, the tissue sources of circulating EVs released in response to exercise remain unclear and a worthwhile area of investigation. Second, the experimental study that compared EV pre versus EV post were based on a small number of participants in older adults, which may contribute to bias depending on participant characteristics. As such, the ndings may not be generalizable to other age groups. Third, the exercise modality used in the current study was limited to aerobic exercise. As such, it is unclear whether the ndings of this study are also applicable to other type of exercise (e.g., resistance exercise). Finally, this study focused exclusively on analyzing miRNAs to infer functional targets of exercise-primed EVs. We cannot rule out the possibility that other EVs cargos, such as proteins and/or lipids, may also play a role in counteracting aging phenotypes in chondrocytes. Such studies would be an interesting extension to our current nding that EVs have the potential to counteract the compromised chondrocyte integrity and Klotho promoter hypermethylation that results from an aged ECM.

MATERIALS and METHODS
Steps to ensure methodological rigor While no established guideline is available for in vitro studies, we adapted Animal Research Reporting In Vivo Experiments (ARRIVE) essential 10 53 and excluded randomization (item4) and experimental animals (item8). This study also followed reporting guideline proposed by Emmerich et al 54 . Where possible, power analysis from pilot data was done to select the number of animals needed for the study using Power and Sample Size Program (version 3.1.2; Vanderbilt University Medical Center, Nashville, TN) 55 . For example, sample size calculation estimated that four EVs before and after aerobic exercise in older adults were required to achieve statistical power of 0.8 based on the type II collagen expression. The treatments and image analyses of EVs isolated before and after exercise were conducted under the same conditions.

Systematic literature search
We performed a systematic search to identify original articles investigating pro les of biomolecules in EVs (i.e., proteomics, RNA-sequencing, small RNA-sequencing, microarray) in older adults after aerobic exercise. Two independent reviewers (HI and ED) conducted an electronic search from the time of database inception to December 2021 using PubMed, Scopus, and Web of Science. A manual search was also performed using Google Scholar. The reference lists of relevant systematic reviews 1,6,42,56,57 were manually searched. The same two reviewers then assessed the titles and abstracts of records identi ed using the prespeci ed eligibility criteria. Articles that passed this initial screening were further reviewed using their full manuscripts. Exclusion criteria included non-peer review journal, non-original article, language other than English, animal study, adults with young or middle-aged (i.e., < 65 years old), no aerobic exercise intervention, no circulating EVs, no omics data, full text not available, and raw data not available. Disagreements regarding manuscript inclusion between the two reviewers was discussed until a consensus was achieved. Finally, a citation search of the identi ed articles was performed using Web of Science to screen for potential additional articles. Identi ed records were then screened and reviewed using the same criteria. Customized Google spreadsheets were used by reviewers to assess and record eligibility.

Differential expression analysis for archived EV small RNAseq data
We accessed the archived small RNA-seq data (GSE144627) collected from the serum in older adults before exercise, immediately after exercise, and after 3 hours of recovery 3 . Raw count data was normalized by count per million (CPM), lterByExpr function, and Trimmed Mean Mvalue (TMM) using R/Bioconductor package edgeR with default parameters 58 . Differential expression analysis was performed for miRNAs with a normalized CPM value using R/Bioconductor package limma 59 . The Benjamini-Hochberg FDR control for multiple hypothesis testing was used to produce q-values.

Target gene prediction of miRNAs
Primary target genes of differentially expressed miRNAs were determined using two different databases, miEAA.2.0 21 and miRTarBase 22 . miEAA.2.0 is a updated web-based application that offers a variety of commonly applied statistical tests such as over-representation analysis and miRNA set enrichment analysis 21 . The target genes signi cantly predicted by mmiRAA.2.0 were then cross-checked by miRTarBase 22 which is the experimentally validated miRNA-target interactions database 22 . Among the experimentally validated miRNAs, we have included genes with "strong" experimental evidence (i.e., genes supported by reporter assay, western blot, or qRT-PCR).

Functional characterization of target genes using GO enrichment analysis on the cartilage-speci c functional network
To determine the biological function of genes of interests, GO enrichment analaysis was performed on the cartilage-speci c network constructed by HumanBase web tool 24 with the target genes of interest used an input. Since the cartilage-speci c network was established based on human tissue, all the gene symbols were translated into human gene symbols prior to the analysis.
Cartilage-speci c networks were constructed using HumanBase software 24 . On cartilage-speci c global network, RWR was performed by R/Bioconductor package RandomWalkRestartMH 29 with the each target gene used as a seeded node. RWR simulated a walker starting from one node or a set of nodes (seed nodes) in one network, and such walker randomly moved in the network to deliver probabilities on the seed nodes to other nodes. After iteratively reaching stability, the a nity score of all nodes in the given network to each target gene node were obtained. Cartilage-speci c global network was constructed using a gold standard data set (i.e., already known gene interactions) downloaded from HumanBase software (https://hb. atironinstitute.org/download) 24 .
Functional characterization of in silico activated genes using pathway enrichment analysis To determine the functional targets of gene perturbation on miRNA regulatory network, KEGG enrichment analysis was performed 30 . In this analysis, pseudo-activated (i.e., an a nity score > 0) genes after each RWR were used as an input.

Determination of relationship between functional targets of gene perturbation and hallmarks of cartilage aging
Hallmarks of cartilage aging was de ned as the eight KEGG pathways signi cantly changed over time according to mass-spectrometry proteomics data of articular cartilage across the lifespan of male murine knee joints 32 . Logistic regression analysis was performed to assess the relationship between functional targets of gene perturbation on miRNA regulatory network (i.e., number of perturbation by RWR; continuous) and hallmarks of cartilage aging (i.e., the presence of hallmarks of cartilage aging; 0: absence, 1: presence). The results were provided as a probability of being aged phenotype. Involving Decedents (CORID). Isolated chondrocytes were seeded in tissue-culture asks at a density of 0.5 × 10 4 cells/cm 2 and maintained in low-glucose DMEM supplemented with 15% FBS, 1% Glutamax, and 1% Pen/Strep at 37℃ at 5% CO 2 . After cells were fully adhered to the culture substrate, the medium was changed every three days until cells reached 70-80% con uency. The cells were then detached with Trypsin/EDTA (Gibco/Thermo Fisher Scienti c) and passaged. For all experiment, second passage chondrocytes were used.

Aerobic exercise cohort
The aerobic exercise cohort was recruited from community-wide efforts, with the primary recruitment method being a clinical registry of individuals who expressed interest in engaging in this study.
Individuals who expressed interest completed a telephone screening to determine initial eligibility. Individuals who appeared to be eligible attended a study orientation session where additional details of the study were provided, and written informed consent was obtained prior to proceeding with additional study procedures. Upon obtainment of informed consent, participants completed additional eligibility screening procedures and other baseline assessment measures. Participants who remained eligible were assigned to participate in the intervention, with one of the intervention conditions being aerobic exercise.
The aerobic exercise intervention paradigm consisted of a combination of 1 supervised session and 4 unsupervised (home-based) sessions per week. The duration of each of the supervised sessions was 30 minutes and the unsupervised sessions progressed from 10 minutes per session during weeks 1-4 to 20 minutes per session during weeks 5-8 to 30 minutes per session during weeks 9-12. Thus, the weekly duration of aerobic exercise progressed from 70 to 150 minutes per week across this period. The aerobic exercise was designed to elicit a moderate-intensity. To facilitate this, aerobic exercise videos were developed by the study staff that we used to guide the supervised exercise sessions and were also placed on tablets provided to the participants to facilitate the unsupervised, home-based exercise sessions.

Serum collection
Blood was collected via venapuncture by a trained phlebotomist. Participants were instructed to fast for a period of 10-12 hours, except for water, and to also avoid strenuous physical activity for 10-12 hours prior to blood collection. Blood underwent initial processing, which included appropriate centrifuging and pipetting into 1 ml cryotubes prior to storage at -80 o C until used for analysis. Samples displaying hemolysis (as evidenced by pink/red coloration) were not included in the analysis.

EV isolation and characterization
EVs were isolated from serum using size-exclusion chromatography (cat no, SP6, qEVsingle 35-nm iZON columns) according to the manufacturer's protocol and as we described 41 . Brie y, after single wash of column by 1mL PBS, 100 µL of serum was added onto the top lter of the column. The eluted volume was collected in small Eppendorf centrifuge tubes. The rst ve fractions (in total 1,000 µL) were collected in a 1.5 mL tube. These fractions contain minimal amount of EVs and are considered to be void fractions according to manufacturer's guidelines. The majority of the EVs were eluted in fractions 6-11 (200 µL each). These fractions, a total of 1.2 mL, were collected together in a 2 mL tube. After isolation, EVs were stored at -20˚C.

Nanoparticle tracking analysis of EVs
The collected EVs were characterized for size and concentration by Nanoparticle tracking analysis (NTA) on a NanoSight NS300 (Malvern Panalytical). Ten µL from each EV sample was diluted 1:100 in EV-free water and infused through the ow-cell using a syringe pump (Harvard Apparatus 98-4730). Videos were recorded for three times in each sample, with the camera level set to 14. These videos were batch analyzed by the software (NTA 3.3) with the detection threshold set to 3. The ow-cell was washed with 1 mL of EV-free water between each sample.
In-well Western for EV purity A in-well western was performed to con rm EV purity. Brie y, 1×10 6 EVs were xed in 2% PFA for 10 minutes, followed by washing with PBS once. EVs were then blocked in PBS solution containing 3% BSA for 1 h at room temperature. After washing twice with PBS solution, EVs were incubated with Alex Fluor 647-labeled CD81(1:200, ThermoFisher) and GM130 (1: 200, Santa Cruz) overnight at 4°C, followed by secondary antibody incubation for 1 h at room temperature. EVs were washed twice with PBS, resuspended in 150 µL of PBS, and loaded to a 96-well plate. The uorescent imaging was performed using LI-COR ODYSSEY CLx and LI-COR Image Studio Acquisition Software (LI-COR Biosciences, NE). Mouse primary muscle broblasts were used as the positive control for GM130.

Preparation of bronectin-coated pAAm substrates
We prepared pAAm gels with different stiffness (5 kPa and 100 kPa) in accordance with a previous study 61 . The pAAm gels were made on glass coverslips pre-treated with 0.1 N sodium hydroxide (cat. SS255-1, Fisher Scienti c, IL), 0.5% 3-aminopropyltrimethoxysilane (cat no. AC313251000, Acros Organics, Belgium) and 0.5% glutaraldehyde (cat no. BP25481, Fisher Scienti c, IL) to improve gel adhesion. To facilitate cell adhesion, the surfaces of prepared hydrogels were further conjugated with bronectin (100 µg/mL, from bovine plasma, Sigma) using sulfo-SANPAH (cat no. NC1314883, Proteochem Inc., UT) as a crosslinker. Prior to seeding cells, gels were UV-sterilized in a cell culture hood for 30 minutes. Gels were kept hydrated in HEPES or PBS during all preparation steps.
Evaluation of the impact of matrix stiffness on cell fate Isolated primary chondrocytes from eldery were plated (5,000 cells per cm 2 ) on 5kPa and 100kPa pAAm gels and cultured in low-glucose DMEM supplemented with 15% FBS, 1% Glutamax, and 1% Pen/Strep at 37℃ at 5% CO 2 . On day 5, the culture medium was removed and seeded pAAm gels were xed in 2% PFA for 10 minutes. After a triple wash by PBS, cells were kept in PBS at 4˚C until used.

Assessment of therapeutic potential of EVs
To test the ability of circulating EVs isolated before and after exercise to modulate chondrocytes health, the isolated EVs were administrated (0.5×10 9 particles) on elderly human chondrocytes cultured on aged-like stiff pAAm substrate on day 3 after chondrocyte seeding. After additional 2-day culture with EVs (i.e., on day 5 after chondrocyte seeding), culture medium was removed and seeded pAAm gels were xed in 2% PFA for 10 minutes. After a triple wash by PBS, cells were kept in PBS at 4˚C until used.
EV uptake by chondrocytes Fluorescence intensity was quanti ed using Image J, in which integrated density was divided by number of cells in each image. We took 5-10 random images in each sample and averaged them for statistical analysis.
Quanti cation of cellular morphology F-actin images were obtained at 20x magni cation using a Zeiss Observer Z1 semi-confocal microscope with ZEN 2.3 software (Zeiss, Jena, Germany). Image processing and morphological feature extraction were performed using CellPro ler software (v4.0, The Broad Institute) 44 . Fifty-three shape features of chondrocytes were determined using the "identify primary objects" followed by the "measure object size shape" and "export to spreadsheet" modules.
Unsupervised machine learning PCA was performed for data reduction to identify the principal components that represent differences in the cellular morphology. To determine variables of cellular shape contributing to principal components, the loading matrix, a correlation between the original variables and PCs, was extracted. In addition, PCA was used to visualize the separation of (1) chondrocytes cultured on soft and stiff pAAm substrates, and (2) chondrocytes cultured on stiff pAAm supplemented with EV pre and EV post .
Global DNA methylation assay     EVs signi cantly decreased Klotho promoter methylation in elderly chondrocytes (n = 4/group). E, Exercise-primed EVs signi cantly increased Klotho gene expression in elderly chondrocytes (n = 4/group). Statistical analyses were performed using a two-tailed paired t-test (A-E). Data are presented as means ± 95% con dence intervals. Portions of the gures were created with biorender.com. Abbreviation: EVs, extracellular vesicles; EV pre , EV isolated at baseline; EV post , EV isolated after 3-month aerobic exercise.

Figure 5
Graphical abstract This study introduced network paradigms to simulate EV-mediated miRNA regulatory network perturbation in recipient articular chondrocyte and uncovered cellular rejuvenation effects induced by exercise-primed EVs in vitro. The observed rejuvenation effects was attributed, at least partly, to epigenetic reprogramming of the gene encoding the longevity factor, α-Klotho.