Secreted Factors and EV-miRNAs Orchestrate the Healing Capacity of Adipose Mesenchymal Stem Cells for the Treatment of Knee Osteoarthritis

Mesenchymal stem cells (MSCs) derived from adipose tissue and used either as expanded cells or minimally manipulated cell preparations showed positive clinical outcomes in regenerative medicine approaches based on tissue restoration and inflammation control, like in osteoarthritis (OA). Recently, MSCs’ healing capacity has been ascribed to the large array of soluble factors, including soluble cytokines/chemokines and miRNAs conveyed within extracellular vesicles (EVs). Therefore, in this study, 200 secreted cytokines, chemokines and growth factors via ELISA, together with EV-embedded miRNAs via high-throughput techniques, were scored in adipose-derived MSCs (ASCs) cultivated under inflammatory conditions, mimicking OA synovial fluid. Both factors (through most abundantly expressed TIMP1, TIMP2, PLG and CTSS) and miRNAs (miR-24-3p, miR-222-3p and miR-193b-3p) suggested a strong capacity for ASCs to reduce matrix degradation activities, as those activated in OA cartilage, and switch synovial macrophages, often characterized by an M1 inflammatory polarization, towards an M2 phenotype. Moreover, the crucial importance of selecting the target tissue is discussed, showing how a focused search may greatly improve potency prediction and explain clinical outcomes. In conclusion, herein presented data shed light about the way ASCs regulate cell homeostasis and regenerative pathways in an OA-resembling environment, therefore suggesting a rationale for the use of MSC-enriched clinical products, such as stromal vascular fraction and microfragmented adipose tissue, in joint pathologies.


Introduction
Osteoarthritis (OA) is a heterogeneous disease with a wide range of underlying pathways and involves structural alterations mainly in the articular cartilage, subchondral bone and synovial membrane [1]. In particular, OA is characterized by a complex biological imbalance between the repair and destruction of joint tissues, rather than a stand-alone mechanical degeneration [2]. The initial loss of cartilage integrity leads first to superficial erosions and later to fissures followed by the expansion of calcifications. Matrix degradation products and proinflammatory mediators deregulate chondrocyte function and stimulate synovium proliferative and inflammatory responses of both resident synoviocytes and macrophages. Eventually, bone turnover is increased with the development of subchondral bone marrow lesions [3].
Due to the strong biological fingerprint, the ideal OA management should rely on the restoration of tissue homeostasis. Nevertheless, to date, available treatments are not supposed to have a direct

Characterization of ASC-Derived Extracellular Vesicles
ASCs treated with OA cytokines for 48 h release around 22,500 extracellular vesicles (EVs) per cell. Isolated EVs exhibited the characteristic cup-shape morphology ( Figure 4A) and were within the literature-reported size range (50-400 nm in diameter), with enrichment in the small particles (mode 118 ± 19 nm) ( Figure 4B). To detect particles within the nanometer range, the flow cytometer has been calibrated with FITC-fluorescent beads of predetermined size (100 to 900 nm, Figure 4C). Purified EVs felt within the 100 nm size, confirming NTA data ( Figure 4D). EVs were strongly positive for extracellular vesicle CD63 and CD81 markers ( Figure 4E), 78.9% ± 1% and 81.6% ± 0.9%, respectively, consistent with previously reported characteristics of extracellular vesicles. On the contrary, CD9 staining was low (2.4% ± 0.2%).

EV-Associated miRNAs
Two-hundred and fifty-seven miRNAs were identified as EVs embedded at various levels of expression in all the four ASCs under analysis (Supplementary Table S1). EV-miRNA hierarchical clustering output was superimposable to the one obtained for factors with higher similarity for ASC 1 and 3 ( Figure 5). Nevertheless, a pattern of overall conservation was again present, with correlation analysis showing a mean R 2 of 0.97 ± 0.01. Therefore, an average C RT value was calculated in order to provide a guide to the level of each miRNA (Supplementary Table S1). Given the presence on average of one abundant miRNA molecule per EV in MSC, 100 MSC-EVs would be needed to shuttle its genetic message to target cells [42,43]. Therefore, only highly abundant miRNAs laying in the first quartile of expression were considered for further analysis (65 miRNAs, covering 96.7% of the genetic message) (Supplementary Table S1, yellow background). Ingenuity pathway analysis (IPA) of the 65 miRNAs used to score only experimentally observed targets identified 818 mRNAs (Supplementary Table S2). An unbiased GO enrichment analysis against the whole genome, with a stringent p-value threshold (< 10 −6 ) due to the high number of scored mRNAs, was able to identify BP related to cell cycle regulation ( Figure 6), with the most significantly (p-value < 10 −9 ) enriched terms being: regulation of the cell cycle process (GO:0010564, p-value 1.31 × 10 −11 , 45 associated genes) and regulation of the mitotic cell cycle (GO:0007346, 8.14 × 10 −10 , 39 genes). Notably, negative regulation of the cell cycle process term (GO:0010948, 6.12 × 10 −7 , 24 genes) suggested that shuttled miRNAs may reduce the activity of genes delaying the cell cycle events, thus promoting proliferation. To define better this influence, the enriched molecular function (MF) GO search was able to specify target terms as the cyclin-dependent protein serine/threonine kinase regulator activity term (GO:0016538, 2.02 × 10 −9 ), encompassing several cyclins (CCNA2, CCNE2, CCNF, CCNE1, CCND1 and CCND3); cyclin-dependent kinases (CDK4) and inhibitors (CDKN2A, CDKN1C, CDKN1B and CDKN1A) and an apoptosis-related cysteine peptidase (CASP3).

Target and Effect Prediction of EV-miRNAs on OA-Cartilage
To frame the power of EV-miRNAs in the OA setting, the 818 target mRNAs were filtered through 2368 abundantly expressed genes, identified as laying in the first quartile of expression (out of 9474 total genes), in cartilage biopsies from OA patients [44]. After filtering, 224 EV-miRNA targets were left and shared (Supplementary Table S3). First, identified genes were searched against those in the IPA database of the OA-related pathway. Nineteen genes resulted targets of EV-miRNAs (ATF4, BMPR2, FGF2,  FGFR1, FGFR3, FN1, HIF1A, ITGA5, JAG1, MEF2C, MMP3, SMAD3, SMAD4, SMAD5, SP1, TGFBR1,  TGFBR2, TIMP3 and VEGFA). Most represented protein classes were transcription factors (PC00218: HIF1A, MEF2C, SMAD3/4/5 and SP1); receptors (PC00197: BMPR2 and TGFBR1/2) and signaling molecules (PC00207: FGF2, FN1 and VEGFA). Notably, many of these molecules specified molecular cascades that are directly involved with OA insurgence and development, like TGF-beta (P00052: BMPR2, SMAD3/4/5 and TGFBR1/2); FGF (P00021: FGF2 and FGFR1/3) and Wnt (P00057: SMAD4/5 and TGFBR1) signaling pathways. Second, GO enrichment (p-value < 10 −6 for enriched GO term threshold) was conducted by scoring the 224 mRNAs against the entire first quartile of OA cartilage as background. Several BP terms were identified, suggesting that when a tissue and/or disease-focused search is performed, the specificity and power of the outcomes increase (Supplementary Figure S1). In particular, five terms were more significantly enriched (p-value < 10 Finally, the list of 65 abundant EV-miRNAs was compared with the database of miRNAs reported to be directly involved in OA pathogenesis, both protective and destructive [45]. Seventeen miRNAs resulted to be involved with protective mechanisms, seven regulate degenerative pathways and four are controlling both mechanisms (Table 3 and Figure 7). Interestingly, scoring the genetic weight of the EV-shuttle and miRNA-related message, protective molecules account for almost 40%, whereas degenerative ones less than 10%. Overall, this might indicate a preponderance of EV-miRNAs in cartilage and synovium protection.

Target and Effect Prediction of EV-miRNAs on Synovial Macrophages
In a recent publication, the transcriptional landscape of infiltrated synovial macrophages has been defined in OA patients [46]. Two types of macrophages were identified: inflammatory-like OA (iOA) macrophages, which are closely aligned to rheumatoid arthritis (RA) macrophages, and classical OA (cOA) ones displaying, together with lower inflammatory features, aberrant tissue repair mechanisms and secretion of factors that may directly contribute to cartilage degradation. Since, in a previous publication, the macrophage transcriptional fingerprint was able to clearly distinguish RA and OA cells [47], we opted to consider newly defined cOA macrophages' mRNA dataset for further analysis. Eight-hundred and eighteen EV-miRNA target mRNAs were filtered against the first quartile of expression (2498 transcripts) of synovial macrophages (total of 9991), and 214 shared genes were identified (Supplementary Table S4). Only 112 genes were shared between the OA cartilage and synovial macrophages, suggesting a divergent response.
First, identified genes were searched against those in the IPA database of the inhibition of metalloproteases pathway, a subclass of the previously scored OA pathway, given the role of synovial macrophages in cartilage ECM destruction. Interestingly, MMP1, 3 and 9 resulted experimentally verified targets, together with RECK and TIMP3 inhibitors, suggesting a profound role of EVs on OA macrophages remodeling activity. Further, searching the GO term macrophage activation (GO:0042116), out of 94 annotated genes, six emerged as possible targets (APP, IL10, JUN, TLR4/7 and TNFa). Intriguingly, abundant miRNAs did not target directly the expression of canonical M1 chemokines (CCL2/3/4/8/20 in the top expressed OA-synovial macrophages mRNAs) and cytokines (IL1A/B), with the exception of IL6 and TNFa. On the contrary, EVs might regulate macrophage polarization by targeting TLR4 that, through NF-kb, suppresses STAT3, whose predominance over STAT1 increases M2 polarization, associated with immune tolerance and tissue repairing. Then, to have a more comprehensive view, the genes were scored against the OA synovial-macrophages first quartile mRNAs (p-value < 10 −6 ) as background to identify enriched GO terms. Again, several BP emerged, with a completely different pattern with respect to cartilage. In particular, six terms were strongly enriched (p-value < 10 −9 ): the anatomical structure development (GO:0048856, 6.05 × 10 −12 , 84 genes); animal organ development (GO:0048513, 3.66 × 10 −11 , 46 genes); positive regulation of multicellular organismal process (GO:0051240, 1.51 × 10 −10 , 68 genes); negative regulation of cell death (GO:0060548, 4.49 × 10 −10 , 54 genes); regulation of smooth muscle cell proliferation (GO:0048660, 7.47 × 10 −10 , 19 genes) and negative regulation of programmed cell death (GO:0043069, 9.95 × 10 −10 , 50 genes). Interestingly, and differently from cartilage, in macrophages, EV-miRNAs may preferentially target genes regulating apoptosis and cell death-related pathways, as emphasized by other significantly enriched GO terms like the negative regulation of the apoptotic process (GO:0043066, 1.47 × 10 −9 , 1.81 × 10 −6 , 49 genes) and regulation of cell death (GO:0010941, 6.29 × 10 −9 , 71 genes). In fact, narrowing the GO enrichment analysis (p-value < 10 −9 as the threshold for GO enrichment), cell death and its regulation emerged as a main constituent of a GO term branch (Supplementary Figure S2). Further, EV-miRNAs are also able to target genes defining myeloid lineage, as shown by GO terms: the regulation of myeloid cell differentiation (GO:0045637, 1.87 × 10 −7 , 19 genes) and positive regulation of myeloid cell differentiation (GO:0045639, 2.04 × 10 −7 , 13 genes). As for cartilage, the different BPs rely on a high number of molecular and transcriptional cascades, as emphasized by the MF GO term DNA-binding transcription factor activity, RNA polymerase II-specific (GO:0000981, 8.45 × 10 −9 , 31 genes).
Eventually, a database of miRNAs involved in M1 vs. M2 macrophage polarization [48] was scored to weight the influence of embedded EV-miRNAs on macrophage phenotypes (Table 4). Although four miRNAs are involved in M2 with respect to seven controlling M1 phenotype, the presence of miR-24-3p, which is responsible for the M1 to M2 switch, and miR-222-3p, which controls M2 polarization, tip the balance towards a more pronounced influence on M2 phenotype ( Figure 8). Consistently, as a global EV fingerprint, 26% of first-quartile miRNA-dependent signals are encompassed by M2-related molecules vs. 7% represented by M1 players. Intriguingly, when IPA CCR5 signaling in the macrophages pathway was searched against the 214 target mRNAs, JUN, FOS and MAP2K4 kinases were identified, suggesting an EV-dependent downregulation of the CCR5 cascade, whose inactivation resulted in a switch towards an M2 phenotype [49]. Moreover, to support the influence of EV-miRNAs on monocyte/macrophage homeostasis, miR-16-5p (0.5%), regulating the monocyte to macrophage developmental transition, and miR-34a-5p (0.7%) together with miR-132-3p (0.5%), participating in macrophage maturation, were also found in the frame of the most expressed miRNAs.

Discussion
In this work, both EV-shuttled miRNAs and soluble factors have been characterized in the secretome of ASCs cultivated in vitro under OA-resembling inflammatory conditions. Several molecules were identified as both cartilage-protective and M1 towards M2 macrophage switchers. Herein presented results provide a strong molecular basis for the promising clinical results observed with ASC-containing products, as MFAT or SVF, in OA settings.
In a murine OA model, injection of MSC secretome, similarly to injection of MSCs, resulted in early pain reduction with a protective effect [15]. Further, both the whole conditioned medium and purified EVs resulted to be effective in cartilage surface restoration and ECM deposition in several in vivo OA defect models [25,27,[50][51][52], mainly thorough EV-associated matrix-remodeling enzymes and regulators as a novel means of matrix remodeling in physiological and pathological conditions [53]. In humans, phase I clinical trials demonstrated the safety and preliminary evidence of efficacy and pain reduction by using local injections of autologous ASCs in OA patients [54,55]. Similarly, autologous and minimally manipulated MSC-enriched tissues [56], such as BMAC or MFAT/SVF, showed the same outcomes. Of note, a partially overlapping array of pro-regenerative and anti-inflammatory molecules was recently shown to be secreted from both expanded ASCs and MFAT or lipoaspirates [57], suggesting that signals secreted by MSCs may explain, at least in part, the promising results of MSC-enriched products in OA management.
Although it is quite well-accepted that MSCs act through paracrine mechanisms on resident cells, nevertheless, a thorough characterization of soluble factors and EV content is lacking and would be essential to define MSCs and MSC-enriched products. In this perspective, in the last years, researchers have tried to couple the array of secreted molecules with the observed protective effects in clinical experiences [58,59]. Despite valuable hints, very often, the input datasets had two major flaws. First, secreted factors and EV-miRNAs data were not generated from the same experimental workflow; therefore, technical and/or biological differences were underestimated. Second, for both types of molecules, no inflammatory effects resembling those in the OA joint were considered, being cells cultured in a control medium or under a heavy priming (in the order of ng/mL), which is not close to the concentration of cytokines usually found in synovial fluids (pg/mL). Thus, the present study offers a more homogeneous and comprehensive dataset, being aware that the combination of the three inflammatory cytokines used is a simplified in vitro condition due to the greatly higher number of factors and the wide range of concentration in the joints of OA patients.
Cartilage thinning and degradation are one of the major fingerprints in OA, resulting from an imbalance of the cartilage homeostasis characterized by excessive catabolic activity that overcomes repair capacity. In particular, increased action of the matrix metalloproteinases (MMPs) in combination with reduction of their inhibitors (TIMPs) levels are largely responsible for the earliest osteoarthritic changes [60], suggesting that a restoration of the healthy balance would help in residual cartilage protection and regeneration. Interestingly, in pioneering studies on OA patients, SVF injection was able to stimulate partial regeneration of the articular cartilage, as described by positive changes in the imaging outcomes (MRI or CT) [61][62][63][64][65]. Consistently, in the last years, few clinical trials based on intra-articular injection of autologous ASCs in OA joints reported improvement in pain, function and cartilage volume [66,67]. These effects on cartilage may be, at least partially, ascribed to those protective molecules and miRNAs we found in the ASC secretome. In fact, three out of four of the most abundantly secreted factors (FST, TIMP2 and SERPINE1) are connected with extracellular matrix assembly and homeostasis and are further reported to have a direct influence on cartilage. TIMPs can regulate ECM remodeling and the activities of growth factors and their receptors by inhibiting MMPs [68]. In dogs, TIMP2 expression is decreased in OA cartilage [69] and synovial fluids [70], whereas, in mice, TIMP2 depletion leads to accelerated OA [71]. Similar protective effects on cartilage homeostasis were demonstrated in mice for the activin receptor follistatin, whose injection reduced both proteoglycan erosion and synovial macrophage infiltration [72]. Further, Serpin E1, also called PAI-1 or plasminogen activator inhibitor 1, positively correlates with the synthesis of cartilage during pathophysiologic processes [73] and counteracts the activity of urokinase/tissue-type plasminogen activators (uPA/tPA), both elevated in OA cartilage in contrast to reduced serpin E1 [74]. Moreover, the presence of the urokinase plasminogen activator surface receptor (PLAUR) in the secretome (2.4 ng per 10 6 ASCs) may also contribute to reduce plasminogen (either naturally occurring or secretome-shuttled) activation in plasmin, which can activate MMPs and plays an important role in modulating cartilage function [75]. Furthermore, elafin (PI3) that was found in the secretome at 12 ng per million ASCs may directly inhibit MMP9 [76].
Nevertheless, other abundant molecules, although at lower levels (between 10 and 100 ng with the exception of IGFBP4), have detrimental or ambiguous effects on cartilage homeostasis, making an overall potency prediction less straight-forward. IGFBPs (insuline growth factor-binding proteins) bind chondrocyte-secreted IGF-1, which induces the production of cartilage matrix components and has a protective role in OA [77]. Accordingly, IGF-1 deficiency, partly caused by IGFPBs seizure, caused an increased severity of OA articular cartilage lesions in both rats and guinea pigs [78,79]. Additionally, cathepsin S is a cysteine protease which has been implicated in the pathogenesis of OA and is involved in acute inflammatory conditions of cartilage [80]. Interestingly, Dkk1, a Wnt pathway suppressor, was reported to have dual activity: its supplementation has been shown to protect mice from experimental OA [81], whereas, in rats, the silencing alleviated chondrocyte apoptosis and cartilage destruction [82]. Thus, a stringent control over Wnt-signaling is required to maintain cartilage homeostasis [83]. Further, interleukin 6 is associated with increased levels of MMPs, as well as with the radiographic severity of OA [84]; although, in the secretome, the concomitant presence of its receptor IL6ST may mitigate the detrimental effects. Similarly, the detection of receptors such as IL2RB and TNFRSF1A, with TNFRSF1B/11B and IL1RL1 and CD40 between 1 and 10 ng, may reduce the amount of free OA-related cytokines and chemokines [85]. Thus, the global message of ASC-secreted factors supports the protective and postulated pro-regenerative effects observed in the clinical experience.
This indication is also endorsed by the EV-miRNAs. It clearly emerged that only a sharp definition of the disease context may allow a potency prediction. In fact, each miRNA may target multiple mRNAs which may be absent, weakly expressed or very abundant. In this work, we decided to study the effect of shuttled miRNAs on highly expressed transcripts, in either cartilage or macrophages, being aware that those with low levels might also have important direct regulatory or, through their protein products, active functions. Interestingly, EV-miRNAs appeared to regulate several processes connected with both OA-related transcriptional cascades and migration/cell motility. In particular, the TGFB pathway was demonstrated to play multiple roles in the development, growth and maintenance of articular cartilage [86]. EV-miRNAs appeared to potentially downregulate the TGFBR1/2 branch of the pathway, targeting in addition the downstream effectors SMAD3 and 4. This is of relevance, since TGFBR1 levels and activation correlated with aggrecan and collagen type II levels in human OA cartilage [87]. Nevertheless, a clear picture is again difficult to be drawn, since SMAD5, effector of the alternative ALK1-dependent branch of the TGFB pathway and highly correlating with MMP-13 mRNA levels [87], is also targeted by EV-miRNAs. Thus, the net effect of miRNAs on the protective TGFBR1/Smad2/3 vs. destructive ALK1/Smad1/5 signaling balance may not be evaluable in silico. Under a different point of view, a tight regulation of TGFB may be beneficial, since its supplementation as a therapeutic agent, although leading to prolonged elevation of proteoglycan synthesis and content in articular cartilage, may also induce inflammation, synovial hyperplasia and osteophyte formation [88]. Therefore, ASC-EVs might be an interesting adjuvant therapy to circumvent the undesirable effects of TGFB on cartilage repair. On the contrary, in silico-predicted effects on chondrocyte movements are much sharper, with a strong target on genes involved in cell motility and chemotactic response. Interestingly, in OA cartilage, it has been postulated that chondrocytes may move towards one another and cluster together [89] to secrete inflammatory factors like TNF-α or IL-6. Thus, ASC-EVs could suppress chondrocyte movement and cluster formation observed in pathologic tissues. Eventually, miRNAs effects get even clearer when only those directly involved in cartilage protective or destructive mechanisms are considered (Figure 7). The net weight of shuttled miRNAs is toward protection, especially counteracting inflammatory signals. In particular, miR-24-3p represents 19% of EV-miRNAs and was proposed as therapeutic, since its reduction led to p16INK4a increase and MMP1 secretion, indicating that miR-24-3p is protective against senescence and cartilage catabolism [90]. Additionally, miR-24-3p inhibits chondrocyte apoptosis and helps in reducing OA pathogenic processes in rats [91]. In addition, among the most abundant ones, miR-125b-5p (12.5%) prevents aggrecan loss and miR-222-3p (5.8%) cartilage degradation. Therefore, in agreement with secreted factors, miRNAs have a strong balance towards protective messages on OA cartilage, although few degenerative signals may be identified.
Recently, together with cartilage repair [57], MFAT was also demonstrated to mediate the activity of synovial macrophages [32]. In fact, MFAT released mediators with long-lasting anti-inflammatory properties, unraveling the deep interplay with the OA synovial membrane and/or inflammatory cells presence/homeostasis. Supporting this paradigm, herein reported data showed a clear effect of ASCs factors and miRNAs. In fact, 23 secreted molecules control, at different levels, leukocyte chemotaxis, and 16 are involved with positive regulation. Among the most expressed, the alpha chemokine CXCL10, naturally present in OA synovial fluid, has a pivotal role in the first wave of leukocyte recruitment to the synovium [92]. Of note, CXCL10 was also shown to effectively recruit human subchondral progenitors, suggesting that postulated purely inflammatory molecules may contribute to the migratory effect of synovial fluid, not only on inflammatory cells but also on human subchondral progenitors with healing properties [93]. Similar properties on mesenchymal progenitors were shown for the neutrophil chemoattractant CXCL5, also found at > 10 ng, and T cells-attractant CXCL12, able to enhance the migration of bone marrow MSCs [94,95]. In contrast, CCL2 (> 10 ng) does not recruit human MSCs [96], being a potent soluble mediator involved in macrophage chemotaxis and migration together with other beta chemokines found in ASC secretome as CCL5/8 (> 1 ng), CCL1/3/7/13/21 and MIF (< 1 ng) [96]. Intriguingly, and supporting the effectiveness of the herein proposed in vitro model, MSCs treated with OA synovial fluid showed high expression of CCL2 and IL6 [97], that, together with migration, can direct M2 polarization of monocyte/macrophages [98], suggesting not only an attractant but also a potential anti-inflammatory mechanism. In fact, once monocyte/macrophage accumulate within the inflamed synovial joints as the initial and critical step for OA pathogenesis, their survival, activation and, most importantly, polarization are crucial. In this path, TGFB, CSF1 and IL13 are potent stimulator of the M2 phenotype. They can explain, at least in part, how MSCs facilitate the monocyte to macrophage transition, attenuate already activated M1 macrophages and enhance M2 activation. In this frame, in previous reports, bone marrow MSCs promoted changes in the M1/M2 balance in vitro [99]. This paradigm may also explain the in vitro observation that both ASCs and SVF induced macrophage polarization toward a pro-wound healing M2 type [100]. Therefore, thanks to secreted factors, ASCs in enriched tissues as MFAT or SVF can attenuate M1 macrophages by shifting them to an M2 state, laying the foundations for injury resolution in inflammatory-stimulating environments, such as OA joints.
Intriguingly, in the last years, not only MSCs but also their purified EVs showed similar effects on macrophages with cartilage restoration, pain reduction and a M1 to M2 polarization [27][28][29]. Mining the transcriptome of OA synovia-isolated macrophages, although EV-miRNAs targeted the macrophage-secreted remodeling enzymes, such as MMP1/3/9, they did not directly target and shutdown M1 signals. Albeit, they appeared to induce M2 polarization by interfering with TLR4, which suppresses STAT3, whose predominance over STAT1 increases M2 phenotype [101]. This capacity was emphasized for miRNAs involved in M1 vs. M2 macrophage polarization [48]. Twenty-six percent of miRNA molecules embedded in ASC-EVs are directly involved in M2 regulation (Table 4). Although less in number with respect to M1 miRNAs, the presence of miR-24-3p and miR-222-3p (5.8%) tipped the balance towards a protective phenotype. In particular, miR-24-3p was reported to have a wide field of action in macrophages. First, it is able to mediate the transition between M1 and M2 states through p110δ [102]. Second, it enhanced the ability of IL-4 and IL-13, both found in ASC secretome, to generate a M2 phenotype eventually able to resist M1 re-conversion under inducing stimuli [102]. Third, miR-24-3p is a negative regulator of TLR-mediated pro-inflammatory cytokine production in M1 macrophages [103,104]. Furthermore, miR-222-3p was demonstrated to suppress the expression of SOCS3, an alternative negative regulator of the M2 promoter STAT3 [105]. Remarkably, ingenuity pathway analysis of the CCR5 signaling in macrophages identified JUN, FOS and MAP2K4 kinases as targets, suggesting an EV-dependent downregulation of the CCR5 cascade, whose inactivation resulted in a switch towards an M2 phenotype [49]. Eventually, EV-miRNAs also target genes of the negative regulation of programmed cell death (GO:0043069) and of the apoptotic process (GO:0043066). This means that cell death and turnover might be tightly regulated in macrophages, being thus a further beneficial effect since, in mice, unbiased macrophage depletion did not attenuate the severity of OA but induced systemic inflammation with a massive infiltration of inflammatory T cells and neutrophils into the injured joints [106]. Therefore, not only macrophage polarization but also the maintenance of specific macrophage subtypes, may be necessary to mitigate inflammation and OA.
Thus, being aware that other factors and EV-embedded miRNAs secreted by ASCs were not assayed in the present study and will need further characterization in future experiments, the global message of ASC secretome with a strong contribution of highly secreted TIMP1, TIMP2, PLG and CTSS or miR-24-3p, miR-222-3p and miR-193b-3p supports the protective and postulated pro-regenerative effects observed in the clinical experience and could be potential therapeutic candidates in organ/tissue healing.

Ethics Statement
The research was performed at IRCCS Istituto Ortopedico Galeazzi under Institutional Review Board approval (San Raffaele Hospital Ethics Committee approval in date 8 March 2018, registered under number 6/int/2018). Specimens were collected after obtainment of patient informed consent (CI_REGAIN_adulto_v2) and following the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

ELISA Assays
Concentrations of 200 soluble cytokines, chemokines, receptors and inflammatory and growth factors were determined by Quantibody ® Human Cytokine Array 4000 Kit (https://www.raybiotech. com/quantibody-human-cytokine-array-4000/) in the ASC secretomes after, according to the manufacturers' instructions (RayBiotech, Norcross, GA, USA). Dilutions (1:2) allowed to have absorbance readings within the standard curve values. The amount of each factor in pg/mL was converted into pg/million cells by multiplying the original value for the total volume in ml and, finally, diving by the total number of cells. Values are shown as million cells normalized pg and mean values ± SD calculated.

EVs Isolation and Characterization
Five milliliters of clarified secretomes were 1:2 diluted with PBS and centrifuged at 100,000× g for 9 h at 4 • C in a 70Ti rotor (Beckman Coulter, Fullerton, CA, USA), and EV pellets were processed as follows: (i) Flow cytometry: Before ultracentrifugation, secretomes were additioned with 10 µM CFSE (Sigma-Aldrich) and incubated for 1 h at 37 • C. After ultracentrifugation, as previously described, labelled EVs were suspended in 50 µl PBS per 5 mL of processed medium. Particles were 1:10,000 diluted in PBS and 100 µl stained with anti-5 µL CD9-APC clone HI9a, anti-CD63-APC clone H5C6 and CD81-APC clone 5A6 (Biolegend, San Diego, CA, USA) for 30 min at 4 • C in the dark. Antibodies were used individually. Events collection was performed with a CytoFLEX flow cytometer collecting a minimum of 30,000 events. A reference bead mix (Biocytex, Marseille, France) composed of a FITC fluorescent mixture of spheres (100 nm, 300 nm, 500 nm and 900 nm) was used to set up the flow cytometer. Gains were: FSC = 106, SSC = 61, FITC = 272, PE = 116 and PC7 = 371. FITC threshold was set at 500 to include 100 nm beads and some smaller debris in the FITC channel. (ii) Transmission electron microscopy: Five microliters of PBS-suspended EVs were absorbed on Formvar carbon-coated grids for 10 min. Filter paper was used to blot drops. Negative stain was performed with 2% uranyl acetate aqueous suspension for 10 min, and excess was removed by filter paper before drying the grid at RT. Samples were examined with a TALOS L120C transmission electron microscope (Thermo Fisher Scientific, Waltham, MA, USA) at 120 kV. (iii) Nanoparticle tracking analysis (NTA): Purified EVs in PBS (1:100 diluted) were visualized by Nanosight LM10-HS system (NanoSight Ltd., Amesbury, UK). Three recordings of 30 s were performed for each sample. NTA software was used to analyze the data and provided both the concentration measurements and the high-resolution particle size distribution profiles.

Screening of EV-Embedded miRNA Expression
After ultracentrifugation, Trizol was used to dissolve EV pellets, and miRNeasy Kit and RNeasy CleanUp Kit were used sequentially to obtain RNA enriched in small molecules < 200 nt (Qiagen, Hilden, Germany). During extraction, 6 pg of a nonhuman synthetic miRNA (arabidopsis thaliana ath-miR-159a) were added to each sample as a spike-in to monitor the technical variability during both the isolation and the following reactions. The spike-in was further used to equalize A and B panels of the OpenArray ® platform (Life Technologies). Briefly, cDNAs were prepared by standard reverse transcription (RT), and preamplification was performed with A and B independent kits, followed by real-time RT-PCR analysis with the QuantStudio™ 12 K Flex OpenArray ® Platform (QS12KFlex), as previously described [107]. To process miRNA expression data from the A and B miRNA panels, together covering 754 human miRNA sequences from the Sanger miRBase v21 Gene, the Expression Suite Software (Life Technologies) was used. C RT of 28 was considered as a burden for the absence or presence of amplification. The global mean was selected as a normalization method [108] and was calculated from miRNAs positively amplified in all samples. Normalized miRNA expression was determined using the relative quantification 2−∆ C RT . Values are shown as normalized Ct and mean values ± SD calculated.

Pathway Analysis
Identification of functional annotations: (i) Protein: Secretome-identified factors were subjected to functional enrichment analysis to provide insight into the functional associations of these protein subsets. The analysis was performed using Gorilla tool (http://cbl-gorilla.cs.technion.ac.il/) with two unranked lists of genes (target and background lists) running mode and 200 factors of the ELISA array as background [109,110]. The list of proteins was also submitted to the PANTHER web interface (http://www.pantherdb.org/) to identify proteins belonging to the same functional classifications, following default settings [111]. (ii) miRNA: The predicted miRNA targets were annotated into functional BP using the microRNA Target Filter tool in ingenuity pathway analysis (IPA; Ingenuity ® Systems, http://www.ingenuity. com). Filters were: confidence "experimentally observed", with all sources available in the IPA database (TarBase, miRecords, TargetScan Human and Ingenuity Expert Findings). To obtain the list of all experimentally validated mRNA targets, no disease filter was set.

Hierarchical Clustering
Heat map plots were generated, scoring factors or miRNAs with ClustVis package (https: //biit.cs.ut.ee/clustvis/) [112]. Clustering options were: distance for rows-correlation, method for rows-average and tree ordering for rows-tightest cluster first. Log 2 [(pg) factors per 10 6 cells] or miRNA C RT after normalization values were used.

Statistical Analyses
In each experiment, four independent ASCs were analyzed. Only proteins or miRNAs present and quantified in all conditions were considered as positively identified. GraphPad Prism software (GraphPad, San Diego, CA, USA) was used for statistical analysis. Kolmogorov-Smirnov normality test assayed normal data distribution. Grubbs' test identified outliers. The level of significance was set with a minimum p-value < 0.05. Data was presented as mean ± SD (Appendix A).

Conclusions
Secreted factors and EV-embedded miRNAs may explain the promising results observed for both expanded ASCs and ASC-enriched products as SVF or MFAT. Both molecules and miRNAs may profoundly alter cartilage homeostasis and macrophage polarization, supporting the protective and pro-regenerative effects observed in the clinical experience. We think that this report shed light on some mechanisms that mesenchymal stem cells, alone or in their niche, activate to heal a diseased joint. Future studies will be needed, especially under the light of new transcriptomes being daily released for the other cellular players, as the identification of the targets appeared as a mandatory sieve to deeply understand cellular and molecular interactions.  Acknowledgments: Authors wish to thank Salvatore Criniti and Gaia Lugano for data management and Giuseppe Banfi for useful discussions.

Conflicts of Interest:
The authors declare no conflicts of interest.