GARP is a key molecule for mesenchymal stromal cell responses to TGF‐β and fundamental to control mitochondrial ROS levels

Abstract Multipotent mesenchymal stromal cells (MSCs) have emerged as a promising cell therapy in regenerative medicine and for autoimmune/inflammatory diseases. However, a main hurdle for MSCs‐based therapies is the loss of their proliferative potential in vitro. Here we report that glycoprotein A repetitions predominant (GARP) is required for the proliferation and survival of adipose‐derived MSCs (ASCs) via its regulation of transforming growth factor‐β (TGF‐β) activation. Silencing of GARP in human ASCs increased their activation of TGF‐β which augmented the levels of mitochondrial reactive oxygen species (mtROS), resulting in DNA damage, a block in proliferation and apoptosis. Inhibition of TGF‐β signaling reduced the levels of mtROS and DNA damage and restored the ability of GARP−/lowASCs to proliferate. In contrast, overexpression of GARP in ASCs increased their proliferative capacity and rendered them more resistant to etoposide‐induced DNA damage and apoptosis, in a TGF‐β‐dependent manner. In summary, our data show that the presence or absence of GARP on ASCs gives rise to distinct TGF‐β responses with diametrically opposing effects on ASC proliferation and survival.


| INTRODUCTION
Multipotent mesenchymal stromal cells (MSCs) are non-hematopoietic, perivascular cells that can be found in virtually all organs and tissues in the body. Isolated MSCs have been defined as plastic adherent, CD73 + CD90 + CD105 + CD11b − CD31 − CD45 − cells that can differentiate into adipocytes, chondroblasts, and osteoblasts in vitro 1 . in vitro expanded MSCs have been shown to promote tissue repair and inhibit inflammatory/autoimmune responses in preclinical [2][3][4] and human clinical trials. 5,6 However, despite promising results and the observed safety, the beneficial effects of MSCs in many clinical trials have been disappointing. 7 A major hurdle for MSC-based therapies is the need to expand MSCs in vitro due to the large number of cells needed/patient (1-10 × 10 6 cells/kg) 8 and the low frequency of MSCs in tissues (from 0.3% to 4% in adipose tissue 9 to 0.001%-0.01% in the bone marrow 10 ). The expansion of MSCs is associated with several problems including the loss of homing capacity, 11 onset of cellular senescence, 12 decrease in differentiation capacity, 13 and susceptibility to genomic instability and malignant transformation, 14,15 limiting the utility of MSCs as a cell-based therapy. Another problem is the low survival and engraftment of injected cells in vivo, 16 and it has been estimated that up to 99% of the injected MSCs will die shortly after administration. 17,18 Several of the above obstacles can be circumvented using the MSC-derived secretome which has shown beneficial effects in inflammatory/autoimmune diseases and in regenerative medicine. However, the therapeutic efficacy of the secretome depends on the functional properties of the MSCs from which it was obtained. Thus, research on optimizing the culture conditions of MSCs is an important endeavor when developing therapies using either MSCs or the MSC-derived secretome. 19,20 Understanding the factors governing MSC expansion in vitro and the development of an MSC product with high-expansion capacity and increased genomic stability would most likely increase the success of MSCs in cell therapy. 21 Glycoprotein A repetitions predominant (GARP), also known as LRRC32, is a type I transmembrane protein which belongs to the leucine-rich repeat family of proteins which encompasses a large number of intracellular and extracellular proteins with distinct functions in neural development, innate immunity, and inflammation, including tolllike receptors and cell adhesion molecules. 22,23 GARP binds and activates latency-associated peptide (LAP)/transforming growth factor (TGF)-β1 complexes on the surface of regulatory T cells (Tregs) [24][25][26] and B-cells, 27,28 supporting their immunosuppressive capacity and isotype switching to IgA production, respectively. We have previously shown that GARP binds LAP/TGF-β1 to the surface of murine ASCs, regulating the activation of TGF-β and modulating their immunomodulatory capacity. Interestingly, silencing of GARP in ASCs significantly decreased their proliferative capacity, but the mechanisms remain unknown. 29 GARP has been shown to both inhibit 30 and promote 31,32 T-cell proliferation. Also, GARP expression in HeLa cells correlated positively with their proliferative capacity, whereas silencing of GARP in the NmuMG breast cancer cell line did not affect their proliferation, 32,33 suggesting that the impact of GARP on proliferation depends on cell type. GARP can also regulate cell proliferation/survival via its activation of TGF-β. TGF-β1 has been found to both promote 34,35 and inhibit 36,37 the proliferation of MSCs in vitro, and some studies have suggested that TGF-β1 can exert a biphasic effect on MSC proliferation, where low concentrations (≤0.25 ng/mL) increase proliferation while higher concentrations (≥1 ng/mL) inhibit proliferation. 38,39 The aim of the current study was to investigate the mechanism behind the block in proliferation of GARP-silenced human ASCs and analyze the effect of GARP-overexpression on ASC proliferation and survival. We found that silencing of GARP in ASCs (GARP −/low ASCs) increased the activation of TGF-β and induced an increase in basal mitochondrial reactive oxygen species (mtROS) and γ-H2AX levels, indicating higher levels of DNA damage. Blocking TGF-β signaling in GARP −/low ASCs reduced the mtROS levels and DNA damage and significantly reverted their block in proliferation, suggesting that in the absence of GARP, the activation of TGF-β by ASCs induces a deleterious response. In contrast, GARP-overexpressing ASCs exhibited an increased proliferative capacity and an increased resistance to DNA damage and apoptosis, through a TGF-β-dependent mechanism. Our data indicate that GARP is required to regulate TGF-β activation/signaling, allowing ASCs to grow in stress conditions, such as in vitro culture.
To overexpress GARP, referred to in the manuscript as LV-GARP, 0.7 × 10 6 ASCs (passage 3-6) were mixed with the concentrated viruses (LV-GARP) and subsequently seeded in 6-well plates and maintained at 37 C. After 5 hours, the media were replaced by fresh medium. The next day, cells were transduced again with the concentrated virus and after 5 hours were expanded in T75 flasks at 21% O 2 /5% CO 2 at 37 C.
To rescue GARP expression in GARP-silenced cells, referred to in the manuscript as LV#19 + LV-GARP, 0.7 × 10 6 ASCs (passage [3][4][5][6] were mixed with concentrated LV#19 and LV-GARP for 5 hours. The LV-containing media were then replaced by fresh medium. The next day, cells were transduced again with concentrated LV-GARP for 5 hours and were subsequently expanded in T75 flasks at 21% O 2 /5% CO 2 at 37 C. GARP expression was analyzed by flow cytometry 4 days after transduction.

| Detection of surface GARP and sorting of GARP ++ ASCs
ASCs were stained with an anti-human GARP antibody (GARP-eFluor660) from eBioscience (San Diego, California). A rat IgG2a kappa-eFluor660 Isotype control (eBioscience) was used in order to determine the background staining, and dead cells were excluded using 7AAD (eBioscience). Cells were acquired on a FACS Canto II flow cytometer and analyzed using the FACS Diva Software (BD Bioscience). After GARP staining of LV-GARP ASCs, the nontransduced (NT(S) and GARP-overexpressing (GARP ++ ) ASCs were separated using a FACS Aria flow cytometer (BD Bioscience).

| Analysis of ASC proliferation
To analyze the effect of GARP on ASC proliferation, the xCelligence real-time cell analyzer system from Roche (Roche Applied System, Penzberg, Germany) was used. In brief, 1000 cells/well of NT, LV-CTRL, LV#18, LV#19, or LV#19 + LV-GARP ASCs were added to 16-well E-plates (Roche Applied System) as previously described. 43 The E-plates were then placed on the device station in the incubator   Raw data were exported from Illumina GenomeStudio and processed in R using negative control probes for background correction and quantile normalization. 44 Probes with detection P-values <.05 in at least two replicates were discharged, and expression values of the remaining probes corresponding to the same gene were aggregated by the median value. Differential expression analysis was carried out using linear models implemented in the limma R package. 45 Genes with an FDR-adjusted P-value <.05 and absolute fold-change >0.5 were selected as significantly differentially expressed. Analysis of gene functions and canonical pathways was performed using the Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems Inc., Redwood City, California).

| Analysis of H2AX phosphorylation
In order to quantify the amount of double-strand DNA breaks ASCs were seeded in 12-well plates (50 000 cells/well), 5 hours after silencing of GARP. The following day, SB431542 (10 μM) or MitoTEMPO (25 μM) were added to the cells. Cells were harvested 3 days later, stained for γ-H2AX, and analyzed by flow cytometry. In order to analyze in more detail the number of γ-H2AX foci/cell, NT, LV-CTRL, LV#18, and LV#19 ASCs were stained for γ-H2AX as described above and acquired on an ImageStream X Mark II Imaging flow cytometer (Millipore) and analyzed using the IDEAS software (Spot Wizard).
In order to study the effect of TGF-β signaling on the induction of DSBs in GARP-overexpressing ASCs, 50 000 NT(S) and GARP ++ ASCs were plated in 12-well plates. After 5 hours, 10 μM of SB431542 was added to the cells. The following day, cells were treated with etoposide (25 μM) and after 6 or 24 hours, the cells were stained for γ-H2AX as explained above.

| Statistical analysis
The statistical analysis was performed using the GraphPad Prism software (GraphPad Software, Inc, La Jolla, California). All data are represented as mean (SD) of three independent experiments unless otherwise stated in the figure legend. Multiple comparisons of the data were performed using the one-way analysis of variance, followed by the Dunnett's or Bonferroni post-test. Correlation of data was performed using the two-tailed Pearson test. P values ≤.05 were considered statistically significant.
F I G U R E 1 Silencing of GARP inhibits the expansion of ASCs in vitro and induces apoptosis. Human ASCs were transduced with LVs expressing two GARP-specific shRNAs (LV#18 and LV#19) targeting distinct sequences of the coding region of the GARP mRNA. Non-transduced (NT) and LV-CTRL-transduced ASCs were used as controls. A, The proliferation of NT, LV-CTRL, LV#18, and LV#19 ASCs were analyzed using the xCelligence real-time cell analyzer system. Proliferation is represented by cell index, and the data show one representative experiment out of three. B, NT, LV-CTRL, LV#18, and LV#19 ASCs were pulsed with BrdU for 3 hours and subsequently stained for BrdU-incorporation and analyzed by flow cytometry. The data are shown as mean (SD) of three independent experiments. *P < .05 vs LV-CTRL. C, NT, LV-CTRL, LV#18, and LV#19 ASCs were stained with 7AAD and Annexin V, 5 days (black bars) and 11 days (white bars) after transduction and analyzed by flow cytometry. Data are shown as mean (SD) of four independent experiments. *P < .05 vs LV-CTRL day 11, **P < .01 vs LV-CTRL day 11. D, Representative dot plots of 7AAD/Annexin V-stained NT, LV-CTRL, LV#18, and LV#19 ASCs on day 5 after transduction. E, The proliferative capacity of NT, LV-CTRL, LV#19, and LV#19 + LV-GARP ASCs was analyzed using the xCelligence real-time cell analyzer system. F, The proliferation data are represented as the cell index at 160 hours. The results are shown as mean (SD) of four independent experiments. **P < .01 vs LV#19. G, The percentages of apoptotic (Annexin V + ) NT, LV-CTRL, LV#19, and LV#19 + LV-GARP ASCs were analyzed on day 5 (black bars) and day 11 (white bars) after transduction by flow cytometry. Results are shown as mean (SD) based on four independent experiments. *P < .05 vs LV#19 day 5, **P < .01 vs LV#19 day 11. ASCs, adipose-derived mesenchymal stromal cells; GARP, glycoprotein A repetitions predominant; LVs, lentiviral vectors 3 | RESULTS

| GARP is required for ASC proliferation and survival
We have previously shown that GARP is important for the expansion of murine and human ASCs in vitro, 29 and we wanted to understand the mechanisms behind this observation. In order to silence GARP, we transduced ASCs with LV vectors encoding for two distinct GARPtargeting shRNAs (LV#18 and LV#19) or a control shRNA (LV-CTRL).
Using the xCelligence real-time cell analyzer system ( Figure 1A)  LV-CTRL and NT), both 5 and 11 days after GARP silencing. Overexpression of GARP in GARP −/low ASCs rescued their block in proliferation ( Figure 1E and F) and prevented their death by apoptosis ( Figure 1G). This effect was seen either when simultaneously co-F I G U R E 2 Silencing of GARP affects the transcriptional program of ASCs, modulating genes involved in cell cycle regulation, apoptosis, and DNA repair. Total RNA was isolated from NT, LV-CTRL, LV#18, and LV#19 ASCs 6 days after transduction, and gene expression was analyzed using the HumanHT-12v4 Expression BeadChip. An ingenuity pathway analysis (IPA) of the modulated genes was performed and revealed several biological categories and biofunctions significantly affected in GARP −/low ASCs compared with NT and LV-CTRL ASCs. In-depth analysis of the biological categories "Cell Cycle" (A), "Cell Death and Survival" (B), and "DNA Replication, Recombination and Repair" (C), showing the most prominent biofunctions (black bars) and their predicted activation (positive z-score) or inhibition (negative z-score). Red circles show the statistical significance for each biofunction and the red line represents P = .01. D, Heatmap showing the top significantly changed genes (LV#18/LV#19 vs NT/LV-CTRL) in the biofunction "DNA Replication, Recombination and Repair." E, IPA prediction of activated/inhibited canonical pathways that were significantly overrepresented in GARP −/low ASCs compared with NT and LV-CTRL ASCs. Bar colors represent the predicted activation (red), inhibition (blue), z-score = 0 (no color), and no activity pattern available (grey) based on the z-score. The values next to the bars represent the zscores when available. The red line represents P = .01. F, IPA prediction of upstream regulators, activated (positive z-score) or inhibited (negative z-score), responsible for the obtained gene expression profile in GARP −/low ASCs. Red circles show the statistical significance for each biofunction and the red line represents P = .01. ASCs, adipose-derived mesenchymal stromal cells; GARP, glycoprotein A repetitions predominant; LVs, lentiviral vectors transducing ASCs with LV#19 and LV-GARP (expressing codonoptimized hGARP, resistant to the shRNAs) or when firstly silencing GARP using LV#19 and subsequently overexpressing GARP the following day (data not shown).
3.2 | Silencing of GARP affects the transcriptional program of ASCs, modulating genes involved in cellular fitness, cell cycle regulation, and DNA repair In order to analyze the impact of GARP silencing on the transcriptional program of ASCs and identify potential mechanisms involved in the observed phenotype, we performed a microarray analysis on NT, LV-CTRL, LV#18, and LV#19 ASCs. We used ASCs from three unrelated donors and silenced GARP using three different batches of LVs ( Figures S1A and B). The mRNA was isolated 6 days post-transduction, a time point at which the proliferative block of GARP −/low-ASCs was evident. Using linear models, we identified genes differentially expressed in both LV#18-and LV#19-transduced ASCs compared with NT and LV-CTRL ASCs and obtained a list of 378 genes that were upregulated (>0.5 logFC) and 556 genes that were downregulated (<−0.5 logFC) ( Table S1).
As expected, the IPA software identified several inhibited biofunctions in the "Cell Cycle" category required for cell cycle as the top activated upstream regulator ( Figure 2F). TP53 contributes to the maintenance of the G2/M checkpoint via the transcriptional repression of CDC25C, cyclin B, and CDK1. 46 In agreement, these genes were downregulated in GARP −/low ASCs compared with NT and LV-CTRL ASCs (Table S1). In addition, the expression of several TP53-inducible antioxidant genes were upregulated in the GARP −/low-ASCs, including SESN1, SPATA18, and GPX1 47,48 (Table S1). All together, these data suggested that GARP silencing could lead to DNA damage and proliferation block in the G2/M phase through activation of TP53.

| GARP −/low ASCs are blocked in the G2/M phase and exhibit increased DNA damage
In order to verify the predicted G2/M block, we analyzed the cell cycle kinetics of GARP −/low ASCs using BrdU and 7AAD. As predicted by the IPA, higher percentages of BrdU + cells were observed in the G2/M phase of GARP −/low ASCs, compared with LV-CTRL and NT ASCs ( Figure 3A and B). As the IPA also suggested a downregulation of the DNA repair machinery, we hypothesized that the observed G2/M block could be a transitory state of the GARP −/low ASCs before entering apoptosis due to excessive DNA damage. We therefore studied the degree of DSBs in GARP −/low ASCs. H2AX is a key factor in the repair process of damaged DNA and its phosphorylated form (γ-H2AX) serves as a biomarker for DSBs. 49 Using flow cytometry, we found that both basal ( Figure 3C) and etoposide-induced ( Figure 3D) γ-H2AX levels were significantly higher in GARP −/low ASCs compared with NT and LV-CTRL ASCs indicating that GARP −/low ASCs are more exposed or susceptible to DNA damage. In addition to the FACS analysis, we quantified the number of γ-H2AX foci in the nuclei in NT, LV-CTRL, and GARP −/low ASCs using an Imaging flow cytometer. These data confirmed our previous results, showing a fourfold increase in foci formation in GARP −/low ASCs compared with control cells ( Figure 3E). ASCs ( Figure 4B). In MSCs, ROS is mainly produced by the mitochondrial complexes I and III (mtROS) and, to a lesser extent, by the NADPH oxidase (NOX)-4. 52 We showed that GARP −/low ASCs produced higher levels of mtROS F I G U R E 5 Inhibition of TGF-β signaling in GARP −/low ASCs reduced mtROS, DNA damage, and partially reverted the proliferation block. NT, LV-CTRL, LV#18, and LV#19 ASCs were cultured without (blue bars) or with 10 μM SB431542 (red bars) and incubated with (A) MitoSOX to analyze the levels of mtROS or (B) stained for γ-H2AX. Data are shown as mean (SD) of three independent experiments. *P < .05 vs LV#18 or LV#19 control, **P < .01 vs LV#18 control, and ***P < .001 vs LV#19 control. We used the same controls as in Figure 5C (MitoSOX) and Figure 5D (γ-H2AX) as SB431542, and mitoTEMPO were added simultaneously in the same set of experiments. C, Graph showing the proliferation of NT, LV-CTRL, LV#18, and LV#19 ASCs cultured without (blue bars) or with SB431542 (red bars). Proliferation was calculated as the cell index at 160 hours/cell index at 10 hours. Data are shown as mean (SD) of three independent experiments. *P < .05 vs LV#18 control, *P < .05 vs LV#19 control. D, Graph showing the proliferation of NT, LV-CTRL, LV#18, and LV#19 ASCs cultured without (blue bars) or with an anti-TGF-β1/2/3 Ab (magenta bars). Proliferation was calculated as the cell index at 160 hours/cell index at 10 hours. Data are shown as mean (SD) of three independent experiments. *P < .05 vs LV#18 control, **P < .01 vs LV#19 control. ASCs, adipose-derived mesenchymal stromal cells; GARP, glycoprotein A repetitions predominant; LVs, lentiviral vectors; TGF-β, transforming growth factor-β F I G U R E 6 GARP-mediated activation of TGF-β protects ASCs against DNA damage and apoptosis and promotes their proliferation. A, ASCs were transduced with LV-GARP and LV-EGFP, and the size (%) of the GARP-and EGFP-overexpressing populations in the bulk culture was followed over time by flow cytometry. Data are shown as mean (SD) of three experimental replicates. *P < .05 LV-GARP vs LV-EGFP at week 5, **P < .01 LV-GARP vs LV-EGFP at weeks 6 and 7. B, NT (black bar) and LV-GARP-transduced ASCs (orange bar) were stained with Annexin V/7AAD on days 5 and 11 after transduction, and the amount of apoptotic cells (Annexin V + ) was analyzed by flow cytometry. Data are shown as mean (SD) of four independent experiments. *P < .05 LV-GARP vs NT. C, GARP-overexpressing cells (GARP ++ , upper right dot plot) were separated from NT ASCs (NT(S), lower left dot plot) using an FACSAria cell sorter, and their proliferative capacity was analyzed comparing the total cell number of GARP ++ ASC relative to NT(S) ASCs after 21 days of culturing. Results are shown as mean (SD) of three independent experiments. **P < .01. E, NT, NT(S), and GARP ++ ASCs were cultured with or without etoposide (25 μM) for 2 days and subsequently stained for Annexin V and analyzed by flow cytometry. Results are shown as mean (SD) of three independent experiments. *P < .05, ns = not significant. F, NT(S) and GARP ++ ASCs were stimulated with etoposide (25 μM) for 0, 2, and 6 hours and subsequently stained for γ-H2AX and analyzed by flow cytometry. Data are shown as mean (SD) of three independent experiments. *P > .05 vs LV-CTRL,^P < .05 LV-GARP vs NT at 24 hours. G, Recombinant TGF-β1 (1 ng/mL) and conditioned medium (CM) from NT(S) and GARP ++ ASCs were added to SBE-HEK293 cells for 18 hours, and luminescence were read on a Glomax Multi Detection System (Promega). Data are shown as mean (SD) of three independent experiments. *P < .05. H, NT(S) and GARP ++ ASCs were cultured with 25 μM etoposide for 48 hours in the absence (black bars; control) and presence of SB431542 (white bars), stained for AnnexinV/7AAD and analyzed by flow cytometry. ***P < .001. I, NT(S) and GARP ++ ASCs were cultured with 25 μM etoposide for 0, 6, and 24 hours in the absence (black bars; control) and presence of SB431542 (white bars), stained for γ-H2AX and analyzed by flow cytometry. *P < .05, **P < .01. ASCs, adipose-derived mesenchymal stromal cells; GARP, glycoprotein A repetitions predominant; LVs, lentiviral vectors; TGF-β, transforming growth factor-β compared with control cells (Figures 4C, plots) that could be reversed by the addition of mitoTEMPO (a mitochondrially targeted antioxidant) ( Figure 4C, graph, green bars). Importantly, inhibition of mtROS reduced the levels of DSBs in GARP −/low ASCs ( Figure 4D), pointing to mtROS as the main trigger of the DSB increment observed after downregulation of GARP. Finally, we found that mitoTEMPO, but not apocynin (a NOX inhibitor), rescued the proliferation of GARP −/low-ASCs to values near to control cells ( Figure 4E). In addition, there was an inverse correlation between the proliferative capacity of the ASCs and their mtROS levels ( Figure 4F). In summary, these data suggest that the loss of GARP results in an increase in mtROS which damage the DNA, resulting in an inhibition of cell proliferation.

| Inhibition of mtROS reduced the DSBs and reversed the proliferation block in GARP −/low ASCs
3.5 | Inhibition of TGF-β signaling in GARP −/ low ASCs reduced mtROS levels, DNA damage, and partially reversed the block in proliferation We have previously shown that silencing of GARP in murine ASCs increased their secretion and activation of TGF-β1 29 and similarly, silencing GARP in human ASCs increased their production of active TGF-β ( Figure S2). Recent reports have demonstrated that TGF-β1 can induce mtROS production in MSCs with effects on their proliferation and survival. 53,54 We found that blocking TGF-β signaling using SB431542 55 in GARP −/low ASCs, just 1 day after GARP-silencing, reduced their mtROS levels ( Figure 5A), H2AX-phosphorylation ( Figure 5B), and significantly increased their proliferative capacity ( Figure 5C). Furthermore, addition of an anti-TGF-β1/2/3 antibody (1D11), 1 day after GARP silencing, also reversed the block in proliferation of GARP −/low ASCs ( Figure 5D). These data suggest that, in the absence of GARP, TGF-β1 increases the levels of mtROS which induce DSBs and inhibit ASC proliferation.

| Overexpression of GARP in ASCs promotes their proliferation and enhances their resistance to apoptosis and DNA damage via TGF-β signaling
To further investigate the effect of GARP on cell proliferation and survival, we generated several ASCs lines overexpressing GARP in 30%-50% of the population using LV-GARP and analyzed changes in the percentage of the GARP-overexpressing cells along time in culture and the level of apoptosis. We found that the percentage of GARPoverexpressing ASCs increased in culture with time suggesting a proliferative advantage of these cells over NT and EGFP-overexpressing (LV-EGFP) ASCs ( Figure 6A). We also observed significantly less LV-GARP ASCs undergoing apoptosis compared with NT ASCs, both 5 and 11 days after transduction ( Figure 6B). We then separated the non-transduced (NT(S)) and GARP-overexpressing (GARP ++ ) ASCs ( Figure 6C) and analyzed their proliferation and susceptibility to apoptosis-inducing stimuli. As seen in the bulk cultures, GARP ++ ASCs proliferated more compared with NT(S) ASCs ( Figure 6D). Interestingly, GARP ++ ASCs were more resistant to etoposide-induced apoptosis ( Figure 6E) and DNA damage, as visualized by γ-H2AX ( Figure 6F), compared with NT(S) ASCs. In summary, these data show that, in contrary to the silencing of GARP, overexpression of GARP in ASCs promotes their proliferation and increases their resistance to apoptosis and DNA damage.
Finally, using a TGF-β-responsive reporter cell line, we found that CM of GARP ++ ASCs contained significantly higher levels of active TGF-β compared with NT(S) ASCs, showing that GARP also promote TGF-β activation by ASCs ( Figure 6G). Inhibition of TGF-β signaling using SB431542 significantly reversed the protective effects of GARP on etoposide-induced apoptosis ( Figure 6H) and DNA damage ( Figure 6I). Our data suggest that presence or absence of GARP gives rise to two distinct TGF-β responses with diametrically opposing effects on ASC proliferation and survival.

| DISCUSSION
MSCs represent a promising cell-based therapy in regenerative medicine, inflammatory/autoimmune diseases, and cancer. However, basic research on how to improve the proliferation, survival, and immunomodulation of MSCs is of fundamental importance for their successful translation into clinical applications. We have previously shown that murine ASCs express GARP/LRRC32 which is important for their immunosuppressive and proliferative capacities. 29 The aim of the current study was to investigate the mechanism behind the block in proliferation of GARP-silenced human ASCs and analyze the effect of GARP-overexpression on ASC proliferation and survival.
We performed a microarray analysis comparing gene expression in GARP-positive and GARP −/low ASCs. In accordance to the phenotype, several genes important for MSC proliferation were downregulated in GARP −/low ASCs, including six of seven recently identified predictive gene markers for MSC proliferation (PLK1, CDC20, BIR5C, SPC25, PBK, and CCNA2). 56 An IPA identified TP53 as a possible activated upstream regulator that could explain the obtained gene expression profile in GARP −/low ASCs. This is in agreement with Zhou and colleagues who showed that GARP low HeLa cells proliferated less and exhibited an increased expression of TP53, p21/Cip1, and p27/Kip1 compared with GARP intermediate HeLa and GARP high HeLa cells. 32 Our IPA analysis also revealed an inhibition of the DNA repair biofunction, suggesting an increase in DNA damage in the absence of GARP. In line with these predictions, we found that GARP −/low ASCs exhibited higher basal levels of DSBs and were also more susceptible to etoposide-induced DNA damage.
Considering the potential links between GARP and DNA damage, several pieces of evidence suggested the involvement of TGF-β and ROS. Firstly, we have previously shown that silencing of GARP in mASCs resulted in an increased activation of TGF-β1, an increase in SMAD2/3-phosphorylation, and the induction of TGF-β1-responsive genes in GARP −/low mASCs. 29 Secondly, it was recently reported that TGF-β1 can increase the ROS production in murine BM-MSCs, which resulted in senescence 53 and apoptosis. 54 We thus measured the ROS levels in GARP −/low ASCs and detected a significant increase in the levels of mtROS compared with control cells. Although physiological levels of ROS in MSCs are necessary for their proliferation 57 60 and a reduction in their immunomodulatory 61 and osteogenic capacities. 62 We found that reducing the levels of mtROS in GARP −/low ASCs using the mitochondria-targeted antioxidant mitoTEMPO, decreased the amount of DSBs, and partially reversed their block in proliferation. Importantly, inhibition of TGF-β signaling in GARP −/low ASCs reduced their levels of mtROS and DSBs and almost completely reversed the block in proliferation. These data suggest that, in the absence of GARP, TGF-β activation increases which induces a deleterious response in ASCs through its induction of mtROS, DNA damage, and apoptosis.
In stark contrast, overexpression of GARP in ASCs also increased their TGF-β activation, which however rendered the cells more resistant to etoposide-induced DNA damage and apoptosis.
Thus, the question arises as to how both the absence and overexpression of GARP can increase TGF-β activation which then induces diametrically opposing responses. Considering the two scenarios regarding the bioavailability and activation of TGF-β, we believe that, in the absence of GARP, ASC-produced LAP/TGF-β is secreted and targeted to the extracellular matrix where it has been shown to be activated by thrombospondin-1. 63,64 In contrast, GARP-mediated activation of TGF-β occurs on the cell surface in a cell-cell contact-dependent manner, 21,65 which results in a qualitatively distinct response due to a higher local concentration of the ligand in the synapse 66 and/or to the presence of additional cell-cell interactions. 67 Although we have not characterized the nature of the TGF-β responses in GARP −/low and GARP ++ ASCs, TGF-β has been shown to cause cellular redox imbalance by inducing mtROS via the mTOR pathways 68 and depleting glutathione via the JNK and induced activating transcription factor 3 pathway, 69 which can result in DNA damage. In contrast, TGF-β has also been shown to protect against genomic instability by enhancing nonhomologous endjoining repair, 70 ATM activity, 71 and the SMAD3/β2spectrin/Fanconi anemia DNA repair pathway. 72 Importantly, three TGF-β isoforms (TGF-β1, TGF-β2, and TGF-β3) exist which bind to the same receptors but exhibit distinct functions in vivo, in part due to their different tissue distributions and expression levels. 21 Most of the work on GARP-mediated activation of TGFβ has focused on the TGF-β1 isoform but GARP can also bind TGF-β2 with low affinity 25 and activate TGF-β3. 73 Future studies should investigate the role of GARP in the activation of TGF-β2, and TGF-β3, especially during MSC differentiation into chondrocytes where these isoforms play important roles. 21 Whether GARP, apart from its effects on TGF-β activation, is directly involved in the DNA damage or antioxidant responses is unknown. However, since GARP contains the highly frequent proteinprotein interacting domain, the leucine-rich repeat (LRR) motif, the possibility for GARP to interact with other proteins is broad. So far, GARP has been described to interact with LAP/TGFβ, 74 lysosomal-associated transmembrane protein 4B (LAPTMP4B) 75 and with HSP90A 75 and HSP90B1 (GP96), 76 members of the HSP90 family of chaperones, in the cytosol and endoplasmic reticulum, respectively. It would be interesting to investigate if GARP interacts with other HSP90 members, since the HSP90 chaperone machinery plays a key role in orchestrating stress responses and regulating the activity of proteins involved in cell cycle and the DNA damage response, such as BRCA1, BRCA2, RAD51, p53, and CHK1. 77 Regardless of the mechanisms, our data show that GARP converts a potentially deleterious TGF-β response into an increased proliferative/anti-apoptotic response in ASCs during their expansion in vitro.
Apart from inducing mtROS and DNA damage, uncontrolled TGF-β signaling in MSCs has been shown to induce the formation of stress fibers and aberrant differentiation in vitro and to promote osteoarthritis 39,78,79 and fibrosis-related diseases 80 in vivo, which also might be inhibited by GARP. The development of MSC products with high expansion capacity and resistance to in vivo infusion/transplantation would most likely increase the success of MSCs in cell therapies for regenerative medicine, autoimmune/inflammatory diseases, and cancer. Since overexpression of GARP in ASCs protected them from DNA damage and reduced their sensitivity to apoptosis, we propose that GARP overexpression in MSCs could be a potential alternative to generate improved MSCs for therapeutic applications. In this sense, GARP-overexpressing MSCs could be advantageous for the treatment of inflammatory/autoimmune diseases as the GARP/TGF-β axis promotes immune tolerance, in part, through the induction of Tregs. 81 However, GARP/TGF-β was recently shown to promote the osteogenic differentiation of MSCs in vitro. 82 Thus, GARP ++ MSCs could be ideal for bone regeneration applications while giving rise to unwanted differentiation when used in inflammatory/autoimmune diseases. In addition, GARP has also been found to promote the growth and spread of tumor cells, classifying it as a possible oncogene. 33 In the context of MSCs, this is probably not a problem as our data suggest that GARP protects against DNA damage and transformation of human MSCs has not yet been observed in vivo. Future studies should analyze the survival and therapeutic efficacy of GARP ++ MSCs in preclinical models of tissue regeneration and inflammation/autoimmunity, also focusing on possible adverse events associated with increased TGF-β activation, including unwanted differentiation, tumorigenesis and fibrosis. 83 It will also be interesting to address whether the observed role of GARP in protecting against TGFβ-induced DNA damage is specific to ASCs or if these results are reproducible in other GARP expressing cells, including GARPexpressing tumor cells. 33 In this sense, GARP could emerge as a therapeutic target for certain cancers known to exhibit an increased GARP expression, such as breast and colorectal cancer.

| CONCLUSIONS
In summary, our data suggest that GARP protects ASCs against deleterious TGF-β activation in vitro, preventing oxidative DNA damage and cell death. In contrast, GARP-overexpressing ASCs exhibit increased proliferation and are less susceptible to DNA damage. We propose that overexpression of GARP in MSCs could increase their therapeutic efficacy in clinical applications.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.