MMP9 integrates multiple immunoregulatory pathways that discriminate high suppressive activity of human mesenchymal stem cells

The mechanisms underlying mesenchymal stem cells’ (MSC) suppressive potency are largely unknown. We here show that highly suppressive human adipose tissue-derived MSC (AdMSC) display and induce a differential immunologic profile, upon ongoing AdMSC suppressive activity, promoting: (i) early correlated inhibition of IFN-γ and TNF-α production, along IL-10 increase, (ii) CD73+Foxp3+Treg subset expansion, and (iii) specific correlations between gene expression increases, such as: MMP9 correlated with CCL22, TNF, FASL, RUNX3, and SEMAD4 in AdMSC and, in T cells, MMP9 upregulation correlated with CCR4, IL4 and TBX21, among others, whereas MMP2 correlated with BCL2 and LRRC31. MMP9 emerged as an integrating molecule for both AdMSC and T cells in molecular networks built with our gene expression data, and we confirmed upregulation of MMP9 and MMP2 at the protein level, in AdMSC and T cells, respectively. MMP2/9 inhibition significantly decreased AdMSC suppressive effect, confirming their important role in suppressive acitivity. We conclude that MMP9 and 2 are robust new players involved in human MSC immunoregulatory mechanisms, and the higher suppressive activity correlates to their capacity to trigger a coordinated action of multiple specific molecules, mobilizing various immunoregulatory mechanisms.

AdMSC with high suppressive capacity induce early and simultaneous decrease in TNF-α and IFN-γ in supernatants. AdMSC/PBMC interactions increased IL-10 in the supernatant at day one of cocultures, and decreased IFN-γ and TNF-α at all time points (days 1, 3, 5 of culture) ( Fig. 2A), but no significant changes for the other cytokines (data not shown). Cytokine kinetics allowed us to detect positive correlations among cytokine production changes during immunoregulation. Although IFN-γ and TNF-α decreased in assays with AdMSC displaying high or low suppressive capacity, only in the highly suppressive group we observed a positive correlation between the inhibition of these two cytokines at day one of coculture (Correlation coefficient, ρ = 1.000, p = 0.0000) (Fig. 2B). For the assays displaying low suppression, the positive correlation between IFN-γ and TNF-α decrease appeared later, at days 3 (ρ = 0.8627, p = 0.0270) and 5 (ρ = 0.9963, p = 0.0000) (data not shown).

AdMSC/PBMC interactions induce a dominant regulatory profile of gene expression modifications.
We evaluated gene expression changes of several immune-relevant molecules, during ongoing suppressive activity, in vitro, in AdMSC and T cells, and checked for differential profile regarding the magnitude of AdMSC suppressive activity. Both AdMSC and T cells displayed robust gene expression modifications during Production of cytokines TNF-α, IFN-γ and IL-10 (pg/ml) in the supernatant of cocultures of AdMSC + PBMC at AdMSC: PBMC ratio (1:10, 1:50 and 1:100) after 1, 3 and 5 days of culture, measured by kit "Human Th1/Th2 CBA" (Cytometric bead Array) and analyzed by flow cytometry (n = 9 experiments with AdMSC derived from different individuals). AdMSC/PBMC interactions decreased TNF-α at all time points; induced an increase of IL-10 at day 1, and a decreased of IFN γ at days 3 and 5 (A). Detection limits of 5 pg/ml 5000 pg/ml. Mann Whitney test *p < 0.05, **p < 0.001. Only in assays with highly suppressive AdMSC we observed a positive correlation between the inhibition of TNF-α, and IFN-γ at day one of coculture (Correlation coefficient, ρ = 1.000, p = 0.0000) (B). For the assays displaying low suppression, the positive correlation between IFN-γ and TNF-α decrease appeared later, at days 3 (ρ = 0.8627, p = 0.0270) and 5 (ρ = 0.9963, p = 0.0000) (data no shown).
The overall functional profile of gene expression modifications was predominantly regulatory for both T cells (Regulatory/Inflammatory gene expression modification ratio: 5.50) and AdMSC (Regulatory/Inflammatory gene expression modification ratio: 2.72). We detected both upregulation of predominantly REG genes and downregulation of predominantly INFLAMMA genes. MMP9 is a central node in specific networks integrating genes with upregulated expression in AdMSC and T cells during suppression. We used IPA to identify molecular networks among differentially expressed genes in AdMSC and T cells during suppression. The software built 4 networks for T cells and 7 for AdMSC. We summarize the networks for AdMSC (Fig. 4A) and T cells (Fig. 4B); with score ≥ 2 or p ≤ 0.01 they overlap and share various genes. The networks 2 and 5 displayed higher number of common molecules, 18, (Network 2: AdMSC data; Network 5: T cell data; Fig. 4). The metalloproteinase 9 (MMP9) appeared as a central node, for both AdMSC (Fig. 4A) and T cells (Fig. 4B), when the networks were represented in a radial layout, placing the most connected node(s) in the center. Protein Upregulation of MMP9 MMP2 upon ongoing AdMSC suppressive activity. We determined the expression of MMP2, MMP9, in AdMSC, and T cells, at the protein level, upon ongoing suppressive activity, for some assays, comparing the basal expression in AdMSC and T cells alone and following AdMSC -PBMC interactions, during suppressive activity.
In concordance with the gene expression data, we found significant protein expression upregulation of MMP9 ( Fig. 5A and B; p = 0.01, paired t-test for median fluoresecence intensity -MFI and p = 0.03, paired t-test for fold increase) for AdMSC, and of MMP2 ( Fig. 5C and D; p = 0.006, paired t-test for median fluoresecence intensity -MFI and p = 0.03, paired t-test for fold increase) for T cells. The increase of MMP9 did not reach statistical significance for T cells (p = 0.05, paired t-test for median fluoresecence intensity -MFI) nor did the increase of MMP2 for AdMSC (p = 0.07, paired t-test for the percentage of positive AdMSC).
In addition, also in concordance with our gene expression data, we found significant upregulation of PD1L (p = 0.0008, paired t-test for median fluoresecence intensity -MFI) on AdMSC (Suplementary Fig. 3J), and an increase in the percentage of CCR4 (p = 0.04, paired t-test) and CTLA4 (p = 0.007, paired t-test) (Suplementary Fig. 4) positive T cells, at the protein level.

MMP2/9 are important new players in human AdMSC immunoregulatory.
We selected four molecules whose expression dominantly increased during AdMSC suppressive activity (MMP2/MMP9, CD73, HLAG, IDO), to further evaluate their participation and potential synergy/interactions in immunosuppressive mechanisms over T cell proliferation, using their inhibitors.
All inhibitors partially reestablished T cell proliferation in a dose dependent manner, indicating a negative impact on AdMSC's immunoregulatory effect (Fig. 6). However, none of the inhibitors used alone abolished AdMSC suppressive activity. In line with the IPA analysis, MMP2/9 inhibitor significantly restored T cell proliferation ( Fig. 6), indicating that MMP9 and MMP2 have a central role in human AdMSC immunoregulatory mechanisms and their high suppression potency. We combined iMMP with iIDO, iCD73 and iHLAG to test synergic action further decreasing AdMSC suppressive activity, but found no additional impact (data not shown).
Highly suppressive AdMSC display/induce a differential immunomolecular profile indicating interplay of multiple immune molecules and mechanisms. We found various specific correlations of gene expression modifications, only in cocultures displaying high (>50% proliferation inhibition) immunosuppressive activity (Hi-Sup), for both AdMSC and T cells. We highlight some of these specific positive correlations in Hi-Sup assay, such as between the increase in gene expression of MMP9 and CCL22 (  8A) and BCL2 and LRRC32, and also between the increase of MMP9 (Fig. 8B), and IL-4 and CCR4 and TBX21. Only in Hi-Sup assays, we found any correlation between increased mRNA expression of important immunoregulatory molecules, namely, IL4, GATA3, HLAG, FOXP3, IDO, BCL2, CCR8 and IL10 in T cells. Likewise, in AdMSC, only in Hi-Sup assays, we detected correlations between increased mRNA expression of CCL22, IL-13 and TNF.
We used IPA to build interaction pathways of the genes whose upregulation occurred in a correlated manner, exclusively in Hi-Sup assays. Again, MMP9 appears as a nodal molecule in both networks, directly or indirectly connecting most molecules with high correlation to IL4 and TNF in the T cells and AdMSCs networks, respectively.
We show two networks for the genes with exclusive positive correlations with TNF in AdMSC (MMP9, HLAG and LIF) (Fig. 9A) and with IL4 in T cells (MMP9, LGALS1, LRRC31, SOCS3, TGFB1, BCL2, CCR8, FOXP3, GATA3, HLAG and IDO) (Fig. 9B), representing subcellular and spatial location of molecules. For T cells, the network (Fig. 9B) shows functions related to "Cellular Movement, Hematological System Development and Function, Immune cell trafficking and cell-mediated immune response", with a highly significant score of 12 (p = 10 −12 ) (Fig. 9). In the T cell network, IL4 appears directly connected with IL1-complex and indirectly to TGFB1 and with the cytoplasmic complex NFat (Nuclear factor of activated T cells). In the nucleus, the transcription factors GATA-3 and FOXP3 are also directly connected to IL-4.
For AdMSCs, the network (Fig. 9A) shows functions related to "Nervous System Development and function, tissue morphology and inflammatory response", with a highly significant score of 12 (p = 10 −12 ). In the AdMSC network, TNF is directly connected to IL1-complex and MMP9 in the extracellular space, and to the transcription regulators complexes: NFκB and AP1 in the nucleus. In the plasma membrane, TNF indirectly connects to the NADPH oxidase complex, which is a nodal molecule connected to a great number of molecules, with a central role in the network.

Discussion
Despite current discussion on MSC phenotypic/functional heterogeneity 14,37,38 , little is known about the functional relationship among the various molecules/mechanisms involved in MSC activity and the intensity of MSC suppressive capacity, especially in human MSC. We have started tackling these questions, by evaluating the effect of AdMSC activity on the expression of various immune-relevant molecules, during AdMSC/PBMC interactions and ongoing suppressive activity, in both AdMSC and T cells, and the impact on cytokine production and on various effector and regulatory immune cell subpopulations, comparing AdMSC's high or low suppressive capacity over T cell proliferation. We here show that AdMSC suppressive activity, in vitro, involves the simultaneous activation/mobilization of multiple molecules in both AdMSC and T cells, indicating that multiple molecular pathways act simultaneously. AdMSC with high suppressive activity is determined by the capacity to mobilize a differential set of multiple imune-related molecular pathways, in a correlated manner. Interestingly, various important molecules involved in immunoregulation or cell survival, such as IL-4, FOXP3, GATA-3, IL-13, IL-10 and BCL2, only presented upregulation in a correlated manner in assays with highly suppressive AdMSC. Within this complex network, we found that MMP9 is a robust new player involved in the molecular network of MSC's high suppressive capacity. In addition, we show that MSC with high suppressive capacity induce an increase of IL-10 and early inhibition of IFN-γ and TNF-α production, in a correlated manner, not found in assays with low suppressive activity.
Some gene expression changes were strikingly dominant -occurring in all experiments -such as, increased IDO in T cells and AdMSC, and IL1β and MMP2 in T cells. In fact, nineteen out of the 30 genes studied for AdMSC displayed a dominant upregulation, including MMP9, HLA-G, IL10, PDL1, SEMAD4, FASL, RUNX3, TNF, IFNg and several chemokines. Although IDO 20 , HLA-G 23 , PDL1 25, 39 and IL10 22 have been previously reported as relevant to AdMSC suppressive activity, it was not known that they are simultaneously and dominantly mobilized. In line with the highlighted importance of TFNα increase for MSC suppressive activity, inducing NFkB activation 40 , we found a dominant TNF mRNA increase in AdMSC, in all experiments. Although the upregulation of TNF in AdMSC, itself, was unrelated to AdMSC immunosuppressive intensity, correlated upregulation was only found in assays with highly suppressive AdMSC, namely with HLAG, LIF and MMP9.
The IPA analysis using all differentially expressed genes in AdMSC and T cells, showed MMP9 as a central node during AdMSC suppressive activity, in both networks, for T cells and AdMSC. Since central node molecules have been shown to play a crucial role in connecting the molecular network 35,36 , our IPA analysis points MMP9 as an important player in AdMSC suppressive activity. Moreover, MMP9 appeared as an integrating molecule in the networks built with the genes displaying positive correlation with the upregulation of IL4 in T cells and TNF in AdMSC, exclusively in assays with high suppressive activity. The significant upregulation of MMP9 expression during AdMSC suppressive acitivity, was also confirmed at the protein level.
We tested whether MMP9 and other dominantly upregulated molecules (IDO, HLAG), played a critical role in AdMSC suppressive activity. The inhibition of these molecules, in vitro, decreased but did not abolish AdMSC suppressive activity, indicating immunoregulatory activity occurring in concert with other players. The inhibition of MMP9, indeed, significantly decreased AdMSC high suppressive activity, supporting its relevant participation in immunoregulatory mechanisms and in suppressive potency. TNF, dominantly upregulated in AdMSC, has a direct relationship with MMP9, increasing MMP9 enzymatic activity 41 , and this MMP9-TNF correlated upregulation was only found in highly suppressive AdMSC, suggesting the need of an integrated action of these molecules for higher suppressive activity. MMP9 upregulation also correlated with the increase of CCL22, exclusively in highly suppressive AdMSC, suggesting favorable conditions for Treg recruitment. In the IL4 network build for T cells, we found a direct relationship of IL4 with TGF and TGF with MMP9. IL-4 increases TGF, which alone increases MMP9 activity 42 . Again, only in assays with highly suppressive AdMSC, we found correlated upregulation of IL-4, in T cells, with several molecules, including with MMP9, FOXP3 and GATA-3.
Noteworthy, MMP2 is also affected by the inhibitor used and also integrates our IPA networks. We confirmed upregulation of MMP2 at the protein level, in T cells, upon AdMSC/PBMC interactions. It is, therefore, likely that MMP2 also contributes to AdMSC high suppressive activity, through T cell activity. Only in assays with highly suppressive AdMSC, we found correlated upregulation of MMP2 and BCL2 and LRRC32 (GARP), in T cells, suggesting that mobilizing Tregs acting through TGF-beta and promoting cell survival are involved in AdMSC high  suppressive activity. Moreover, the expression of GARP on human Tregs has been reported to be related to their higher suppressive acitivy. Although, both proinflammatory 43 and immunoregulatory activities have been suggested for MMP2 44,45 , immunoregulation seems predominant in the context of AdMSC activity. MMP9 has been implicated in murine bone marrow MSC suppressive activity, in vitro and in vivo 46 , and reported in human bone marrow MSC as an important molecule for homing/migration into tissues, in response to inflammatory stimuli 47 . Moreover, bone marrow MSC prevented murine islet allograft rejection by MMP2/9 suppressive activity 46 . We now show the importance of MMP/9 as novel relevant and dominant participants in human AdMSC suppressive mechanisms and potency.
The increased numbers of Tregs expressing CD73, an ectoenzyme that induces adenosine -an immunosuppressive molecule 49 , indicates another MSC immunoregulatory mechanism. Accordingly, others have reported the induction of CD4 + CD25 + CD73 + T cells by human MSC 50 . We now show that the expansion of CD73 + CD4 + CD25 + Tregs during AdMSC immunoregulatory activity, actually occurs in cells with stable FOXP3 expression, characteristic of Tregs but not activated T cells. Using the CD73 inhibitor, we observed some decrease in AdMSC suppressive activity in individual experiments, but no statistical significance, either alone or in combination with IDO inhibitor. This suggests that, at least in vitro, the expression of CD73 by Tregs may not be a dominant feature in the AdMSC suppressive activity.
We also found decreased numbers of activated/effector CD4 + and CD8 + cells expressing ICOS, and CD8 + cells expressing OX40, molecules implicated in effector T cell costimulation and activation 51 , respectively. As ICOS requires de novo induction on T cell surface, being upregulated in a late phase of T cell activation 52 , our data suggest that AdMSC also act upon a late phase of T cell activation.
The increased TNFα, IFNγ and IL1B mRNA expression in AdMSC underscores the importance of proinflammatory cytokines, enhancing MSC immunosuppressive activity, as reported 26,53 . Indeed, many proinflammatory molecules trigger a network of immunoregulatory molecules, as reported for IFN-γ inducing IDO and PD-L1 20,39,44 in the context of MSC immunosuppression. In addition, IFN-γ along with TNF-α, IL-1α and IL-1β induces iNOS and CXCL10 and CXCL9, also important in cell trafficking in immunoregulation 30 . The marked mRNA upregulation of CCL22 and CCL17 and of CXCL10 and CCL5 at the protein level, in AdMSC during suppressive activity, suggest a favorable context for Treg migration and recruitment, as reported 30,45 , corroborating these interpretations.
The magnitude of MSC suppressive activity and implicated mechanisms have been largely overlooked in the literature particularly in clinical trials. Considering the potential impact on immunodulatory cell therapy outcome, we believe it is crucial to determine whether high or low magnitude of MSC suppressive capacity provides the best beneficial therapeutic effect, in specific pathological contexts, and underlying mechanisms. The recent findings that T cells from autoimmune disease patients have reduced sensitivity to immunoregulation by MSC 54 further highlights the importance of determining these issues for immunomodulation cell therapy in the clinic.
Besides adding novel players to the scenario of MSC immunobiology, our data bring a relevant contribution, providing a broader view of the interactive immunoregulatory molecular networks involved in MSC immunologic activities. The more potent MSC suppressive activity correlates to their capacity to trigger a coordinated action of multiple molecules, mobilizing various immunoregulatory mechanisms. We highlight MMP9 emerging as an integrating molecule in human AdMSC immunregulatory network and the importance of the early inhibition of proinflammatory cytokines, along with the induction/increase of IL-10, contributing to a more potent immunoregulatory capacity. Taken that MSC are present throughout the body, we may interpret that the multiple MSC interactions with a variety of immune cells -and other cell types -along evolution, have favored these stromal cells to develop an enormous variety of immunoregulatory mechanisms, likely to contribute, in vivo, to the maintenance of in situ and peripheral tolerance, and the control of homeostasis, supporting the idea that MSC may integrate the immune system.

Material and Methods
Experimental Design. To measure AdMSC suppressive capacity over T cell proliferation (anti-CD3 stimulated), we cocultured AdMSC (different individuals, n = 11) with PBMC from a single healthy individual, favoring the evaluation of AdMSC immunoregulatory variability individual-dependent. PBMC from an additional individual were used for some experiments. We classified AdMSC as displaying high (>50% inhibition) or low (<50%) suppressive activity. We evaluated several immunologic features following AdMSC/PBMC interactions and tested for correlations of gene/protein expression modifications considering high (>50% inhibition) or low (<50%) suppressive capacity. Features analysed: (i) T cell proliferation inhibition; and changes in (ii) gene expression of immune-relevant molecules in AdMSC and T cells; (iii) proinflammatory and immunoregulatory cytokine production; (iv) regulatory and effector T cell subpopulations. Using inhibition assays, we evaluated the contribution of selected molecules to AdMSC suppressive activity.   Tables S1 and S2). We used a Master Mix with SYBR Green (SuperArray Bioscience Corporation, USA) (final volume: 25 µl), the ABI Prism 7500 Real Time PCR System (Applied Biosystems), and GAPDH for normalization. Gene expression changes following PBMC/AdMSC interactions were calculated relative to PBMC or AdMSC alone. We considered increased or decreased mRNA expression when the relative expression (R.E.) was ≥2.0 and ≤0.5, respectively and statistically significant (P < 0.05, Wilcoxon). For gene expression quantification we used the SABiosciences software (http://www.sabiosciences.com/pcrarraydataanalysis.php). Undetectable: Ct >35. Gene expression changes were also characterized as exhibiting a dominant regulatory (REG: R.E ≥ 2 for regulatory genes or ≤0.5 for inflammatory genes) or an inflammatory (INFLAMMA: the opposite of REG) profile.

Study Subjects.
Network analysis using Ingenuity Pathway Analysis (IPA). IPA Network maintains a graphical database of molecular network interactions (Ingenuity Knowledge Base, IKB). We uploaded in the IPA all genes differentially expressed in AdMSC and T cells, following AdMSC/PBMC interactions, and genes with positive correlations in T cells and AdMSC, only in the assays with high suppressive activity. The genes were mapped to their corresponding gene objects in the IKB and overlaid onto a global network based on the IKB. Smaller networks (<35 nodes) containing a significant number of focus genes were algorithmically selected and scored; a score ≥ 2 indicated a probability p ≤ 0.01 that the focus genes in a network were found together by chance.
Evaluation of AdMSC and Regulatory or Effector T cell subpopulations. We evaluated the effect of AdMSC/PBMC interactions under anti-CD3 stimulation, on the expression of various immune-related Protein expression of MMP 2 and MMP9 upon AdMSC/PBMC interactions. We analysed the intracellular protein expression of MMP2 and MMP9 by FACS, in both AdMSC and T cells, alone, or following coculture, in the same conditions, using: anti-human MMP9-FITC antibody (Cat N. IC9111F, R&D Systems, Inc) and anti-human MMP2-PE antibody (Cat N. IC9023P, R&D Systems, Inc). We used FACS CantoII TM analyzer, DIVA software (BD biosciences, USA), and the FlowJo software 9.1 (Tree Star, OR) for analysis. At least 300,000 events were acquired within the lymphocyte gate or within the AdMSC gate. We compared the expression in co-cultures to AdMSC or T cells alone.
Effect of AdMSC/PBMC interactions on cytokine production. We evaluated IL-2, IL-4, IL-5, IL-10, IFN-γ and TNF-α production in coculture supernatants (days 1, 3 and 5), using Cytometric Bead Array (CBA Th1/Th2) (BD Company, USA), the FASCanto II (BD Company,USA), the BD FACS DIVA TM program and the BD CBA Software (BD Company,USA). Concentration was calculated using the standard curves (FCAP Array Software BD Company,USA). Detection limits: 5 to 5000 pg/ml. For samples showing concentrations below the lower detection limit, we considered the value just below this limit to quantify cytokine production change in cocultures.
Statistical analysis. We used the following Statistical tests: Wilcoxon test for: AdMSC suppressive capacity over T cell proliferation, expression changes in the percentage and MFI (FACS), gene expression changes in T cells and AdMSC during PBMC/AdMSC interactions; One-Way ANOVA: to compare the % of proliferation inhibition in the 3 conditions tested; Spearman correlation test: for correlations among gene expression changes; Fisher's exact test: for significance in the IPA network analyses; Mann-Whitney test: for differences in cytokine production during AdMSC/PBMC interactions, and AdMSC suppressive capacity with and without inhibitors of selected molecules. Significant differences: *(p < 0.05), **(p < 0.01) and ***(p < 0.001).