Research ArticleTransplantation

Apoptosis in mesenchymal stromal cells induces in vivo recipient-mediated immunomodulation

See allHide authors and affiliations

Science Translational Medicine  15 Nov 2017:
Vol. 9, Issue 416, eaam7828
DOI: 10.1126/scitranslmed.aam7828

MSC sacrifice for immunosuppression

Transfer of mesenchymal stromal cells (MSCs) induces immunosuppression, although the cells are undetectable shortly after transfer. The immunosuppressive mechanism of MSCs has been somewhat of a mystery, but Galleu and colleagues now suggest that MSC apoptosis is crucial. They observed in a mouse model of graft-versus-host disease that cytotoxic cells rendered the MSCs apoptotic shortly after transfer. Moreover, cells from patients that responded to MSC therapy had more cytotoxic activity against MSCs. These findings not only provide important mechanistic insight but also suggest that patients could be screened for responsiveness to MSC therapy before transfer.


The immunosuppressive activity of mesenchymal stromal cells (MSCs) is well documented. However, the therapeutic benefit is completely unpredictable, thus raising concerns about MSC efficacy. One of the affecting factors is the unresolved conundrum that, despite being immunosuppressive, MSCs are undetectable after administration. Therefore, understanding the fate of infused MSCs could help predict clinical responses. Using a murine model of graft-versus-host disease (GvHD), we demonstrate that MSCs are actively induced to undergo perforin-dependent apoptosis by recipient cytotoxic cells and that this process is essential to initiate MSC-induced immunosuppression. When examining patients with GvHD who received MSCs, we found a striking parallel, whereby only those with high cytotoxic activity against MSCs responded to MSC infusion, whereas those with low activity did not. The need for recipient cytotoxic cell activity could be replaced by the infusion of apoptotic MSCs generated ex vivo. After infusion, recipient phagocytes engulf apoptotic MSCs and produce indoleamine 2,3-dioxygenase, which is ultimately necessary for effecting immunosuppression. Therefore, we propose the innovative concept that patients should be stratified for MSC treatment according to their ability to kill MSCs or that all patients could be treated with ex vivo apoptotic MSCs.


Mesenchymal stromal cells (MSCs) have received center-stage attention because they exhibit potent immunosuppressive and anti-inflammatory activities (1) that have been extensively tested in several medical conditions, ranging from autoimmune diseases to the immunological complications of clinical transplantation (25). The extensive clinical use has been undeterred by the fact that the mechanisms underlying MSC therapeutic activity remain largely unresolved. MSC-mediated immunosuppression is major histocompatibility complex (MHC)–independent, is non–antigen-specific (1), and targets virtually all immune cells by arresting cell cycle progression (68). Interference with amino acid metabolism in the inflammatory microenvironment has been suggested as a crucial mechanism, with indoleamine 2,3-dioxygenase (IDO) being one of the major candidates (9, 10), along with transforming growth factor–β1, hepatocyte growth factor (11), prostaglandin E2 (12), and soluble human leukocyte antigen G (13, 14). Tumor necrosis factor–α (TNF-α)–stimulated gene 6 protein (TSG-6) has been identified as a recent addition to the anti-inflammatory armamentarium of human MSCs (15). However, there is consensus that the immunosuppressive activity does not solely rely on MSCs but may involve the active engagement of other immunomodulatory cells, such as regulatory T cells (16, 17) and immunosuppressive macrophages (18, 19).

The imperfect knowledge of MSC immunobiology may well explain why the results of the clinical trials have often been controversial and why conclusive proof of efficacy has not yet been provided. Two major unresolved challenges undermine progress in the field. The first is that only a proportion of patients, although affected by the same disease, responds to MSC infusions, and this response cannot be predicted. The second is that, to be efficacious, MSCs are not required to engraft. The vast majority of infused MSCs resides transiently in the lungs before becoming undetectable within a few hours (20). Because our current knowledge cannot provide an explanation to this paradox (6, 9, 10, 15), a better understanding of the mechanisms underlying MSC therapeutic activity would be highly desirable to improve clinical efficacy and make therapeutic achievements more reproducible. We selected to address these challenges in graft-versus-host disease (GvHD) because there is proof of principle that MSCs are clinically efficacious (4, 21). By using a mouse model and clinical samples, we have tested the hypothesis that MSCs undergo in vivo apoptosis after exposure to the GvHD environment.


MSCs undergo apoptosis in recipient GvHD animals

We used a mouse model of GvHD in which lethally irradiated C57BL/6 male mice were transplanted on day 0 with bone marrow (BM), polyclonal purified CD4+ T cells from female syngeneic donors, and purified CD8+ T cells transgenic for a T cell receptor specific for the male mouse HY antigen Uty [Matahari (Mh)] as GvHD effectors (fig. S1A) (22). The addition of CD4+ T cells is necessary to facilitate the expansion of the CD8+ T cells. In this model, the expansion of the T cells effecting GvHD (CD8+Vβ8.3+) can be precisely enumerated and correlates with the clinical severity of the disease. Human MSCs were injected 3 days after the transplant. To explain the mechanism by which MSCs are rapidly cleared after injection (20, 23), we tested the hypothesis that MSCs undergo apoptosis.

In vivo MSC caspase activation was evaluated as a readout of apoptosis. MSCs were transfected with the pGL3-control vector for the expression of firefly luciferase (Luc+; luc-MSCs). Caspase activation was measured as luciferase activity using injection of Z-DEVD-aminoluciferin. In this system, caspase 3 activation could be quantified on the basis of emitted light because Z-DEVD is cleaved upon activation of caspase 3, leading to release of aminoluciferin, which, in turn, can be metabolized by the firefly luciferase expressed in MSCs. Luc-MSCs were injected into recipients of BM transplant with CD8+ Mh T cells (GvHD group), and 1 hour later, caspase activity was measured in vivo as total luminescence signal (TLS). Two groups of control mice received MSCs. One consisted of untreated males (naïve group), and the second was a group of mice that were irradiated and also received CD4+ T cells and BM cells (BM group) without the transgenic CD8+ T cells to reproduce the condition of MSC infusion in the absence of activated cytotoxic T cells (fig. S1A). We observed high caspase activity only in MSCs injected into GvHD mice (Fig. 1, A and B). High signal could be detected from the lungs of all animals when the control d-luciferin (firefly luciferase substrate) was used (fig. S1, B and C), thus confirming that luc-MSCs can also be tracked in the lungs when caspase activity could not be detected.

Fig. 1. MSCs undergo in vivo apoptosis after infusion without affecting immunosuppression.

(A) Luc-MSCs were injected intravenously into naïve, BM, and GvHD mice 3 days after transplantation. All animals were then injected intraperitoneally with Z-DEVD-aminoluciferin and imaged 1 hour later. n = 6 (one to three mice per group), grouped from three independent experiments. In each experiment, a different MSC expansion was used. White lines separate multiple photographs assembled in the final image. (B) TLS was measured from the images of mice in (A) and shown as mean ± SD. (C and D) Percentage of GvHD effector cells (CD8+Vβ8.3+) calculated in the lymphocyte gate (defined by the physical characteristics of the cells) in the spleen (C) and lungs (D) of GvHD mice (black circles) and GvHD mice treated with MSCs (black squares) 4 days after MSC injection. n = 15 (GvHD) and 13 (GvHD + MSCs) mice, grouped from four independent experiments. Means ± SD are shown. Statistics in (B): one-way analysis of variance (ANOVA) with Tukey’s multiple comparison test (**P < 0.01 and ***P < 0.001). Statistics in (C) and (D): unpaired t test (**P < 0.01). ns, not significant.

The evidence that MSCs undergo apoptosis after infusion prompted the question of whether they are still capable of suppressing antigen-driven T cell expansion. Therefore, we analyzed their immunosuppressive effect by enumerating CD8+Vβ8.3+ Mh T cells (GvHD effector cells) in MSC-treated or untreated GvHD mice. MSCs produced a substantial reduction in GvHD effector cell infiltration in both the spleen and lungs (Fig. 1, C and D). These results indicate that, despite the presence of MSC apoptosis after infusion (Fig. 1, A and B), MSC immunosuppression still occurs.

We can likely exclude the possibility that the observed immunosuppressive activity could be the consequence of the recipient inflammatory cytokines because, in our xenogeneic combination, murine inflammatory cytokines will not cross-react with the corresponding human receptors and will not activate immunosuppressive molecules in human MSCs (2426) while retaining the ability to expand murine effector cells mediating GvHD (27). Accordingly, human MSCs were not able to inhibit concanavalin-A–induced proliferation of murine splenocytes unless preactivated by human cytokines (fig. S2, A to C). Furthermore, exposure of human MSCs to murine inflammatory cytokines did not up-regulate IDO, TSG-6, or PTSG2 (prostaglandin-endoperoxide synthase 2) and considered major effectors of human MSC-mediated in vitro immunosuppression (fig. S2D).

In vivo MSC apoptosis depends on activated recipient GvHD effector cells

Our results show that MSCs rapidly undergo apoptosis after infusion, providing an explanation for the rapid clearance of transplanted MSCs in the recipient. The absence of in vivo MSC apoptosis in the control groups (naïve and BM mice) demonstrates that MSC apoptosis is not the result of xenogeneic recognition of human MSCs because it is detected only in GvHD mice. When we enumerated GvHD effector cell infiltrate (CD8+Vβ8.3+) in the lungs of mice, where MSC apoptosis occurs, we found that only the lungs of GvHD, but not those of naïve and BM mice, contained a large proportion of CD8+Vβ8.3+ cells (Fig. 2A), thus confirming the correlation between caspase activation in MSCs and the presence of GvHD effector cells.

Fig. 2. MSC apoptosis is important for immunosuppression and requires functionally activated cytotoxic cells in the recipient.

(A) The percentage of CD8+Vβ8.3+ cells in lung cell suspensions from naïve C57BL/6 male, BM, or GvHD mice was analyzed in the lymphocyte population. Means ± SD are shown. n = 12 (GvHD), 3 (BM), and 3 (naïve) mice, grouped from three independent experiments. (B) CD8+ cells were sorted from the lungs and spleens of naïve female Mh (gray bars) or GvHD mice (not treated with MSCs; white bars) 7 days after the transplant and tested for their ability to induce MSC apoptosis in vitro. The results show annexin-V+/7-AAD MSCs (mean ± SD) in three independent experiments (n = 10 per group; the black bar represents the value of apoptosis in MSCs cultured alone used as control (n = 3). (C) Luc-MSCs were infused in three independent experiments in GvHD (n = 7) and GvHDPerf−/− (n = 7) mice 3 days after transplantation. One hour later, mice were injected with Z-DEVD-aminoluciferin and imaged. White lines separate multiple photographs assembled in the final image. (D) TLS was obtained from (C) and expressed as mean ± SD. (E and F) The percentage of effector GvHD cells (CD8+Vβ8.3+) in the lymphocyte population was measured in the spleen (E) and lungs (F) of untreated GvHDPerf−/− (n = 16) and GvHDPerf−/− (n = 17) mice treated with MSCs (mean ± SD of four independent experiments). Statistics in (A) and (B): one-way ANOVA with Tukey’s multiple comparison test (*P < 0.05 and ***P < 0.001). Statistics in (D) to (F): unpaired t test (***P < 0.001).

To test the hypothesis that GvHD effector cells were responsible for MSC apoptosis, we cultivated MSCs with CD8+ T cells purified from the lungs or spleens of GvHD (in vivo–activated) or naïve Mh (in vivo–resting) mice. Activated, but not resting, Mh CD8+ T cells induced MSC apoptosis (Fig. 2B). Similar proportion of cytotoxicity against MSCs could be elicited by naïve Mh CD8+ T cells stimulated in vitro by CD3/CD28 beads (fig. S3A).

The requirement of cytotoxic cells in the induction of MSC apoptosis and the consequent immunosuppression was evaluated using Mh/perforin knockout (Mh/Perf−/−) mice as donors of defective cytotoxic GvHD effector cells (GvHDPerf−/− group). Luc-MSCs were infused into GvHDPerf−/− or control GvHD mice that had received Mh CD8+ T cells. Mice were imaged 1 hour later, and caspase activation was measured as described above. We observed a much lower caspase activity in GvHDPerf−/− mice compared to GvHD controls (Fig. 2, C and D). High signal was detected in the lungs of all animals when the control d-luciferin was used (fig. S3, B and C), thus confirming that luc-MSCs were also in the lungs when caspase activity could not be detected. Infiltration of GvHD effector cells in the spleen and lungs was not reduced in the GvHDPerf−/− group receiving MSCs (Fig. 2, E and F). We conclude that MSC apoptosis is indispensable for immunosuppression and requires functionally activated cytotoxic cells in the recipient.

Cytotoxic activity against MSCs is associated with clinical response to MSCs in GvHD patients

On the basis of these findings, we inferred that the presence of cytotoxic cells in the recipient could be predictive of MSC therapeutic activity. Sixteen patients (mean age, 40.5 years; range, 10 to 69 years), with severe steroid-resistant grade 3 to 4 GvHD received a total of 17 doses of MSCs. Patient characteristics are summarized in table S1.

Clinical responses to MSCs were defined by an improvement of at least 50% in at least one organ affected by GvHD, as previously described (4, 21, 28). Five patients obtained a clinical response. Peripheral blood mononuclear cells (PBMCs) were freshly collected within the 24 hours preceding the MSC infusion and tested directly for their ability to induce MSC apoptosis ex vivo in a 4-hour cytotoxic assay. One patient received two doses of MSCs, and the cytotoxic assay was performed before each dose independently. MSCs were sourced from the same donor used for the infusion (n = 8) or from a different donor (n = 9). At the time of performing the assay and cytofluorimetric analysis, the operator was blind to patients’ clinical details. PBMCs from healthy donors (n = 5) were used as controls.

Overall, PBMCs from GvHD patients and control PBMCs exhibited similar cytotoxic activity against MSCs (mean ± SD, 10.63 ± 8.76% and 3.82 ± 2.50%, respectively; P = 0.10). However, cytotoxicity between clinical responders and nonresponders to MSCs was markedly different, with the proportion of apoptotic MSCs (annexin-V+/7AAD) exhibiting a fourfold difference (Fig. 3, A and B). The discrimination threshold of apoptotic MSCs between responders and nonresponders calculated using the receiver-operating characteristic curve revealed that a 14.85% cutoff was predictive of clinical response with the highest sensitivity and specificity. Cytotoxicity did not vary among MSC preparations because when we tested patients’ PBMCs against the MSCs used for the infusion as compared to another preparation obtained from an unrelated donor, no difference in apoptosis induction could be detected (fig. S4A). To further confirm the irrelevance of the specific MSC preparation, we evaluated the susceptibility of MSCs sourced from different unrelated donors to undergo apoptosis after exposure to four different mixed lymphocyte reaction (MLR) combinations. The proportion of apoptotic MSCs was similar among the different MSC preparations when the same MLR was tested. Conversely, the cytotoxic activity against the same MSCs varied among different MLR (fig. S4B).

Fig. 3. Cytotoxic activity against MSCs predicts clinical responses to MSCs in GvHD patients.

(A and B) PBMCs obtained from healthy controls (HC) or patients with GvHD receiving MSCs in the following 24 hours were incubated in 24-well plates with MSCs at a PBMC/MSC ratio of 20:1 for 4 hours. Apoptosis was measured in MSCs assessing the percentage of annexin-V+/7-AAD cells by flow cytometry. (A) Representative plots for HC, clinical responders (R), and nonresponders (NR). Left panels show the background apoptosis of MSCs alone used in the corresponding cytotoxic assay. (B) Apoptosis was compared among HC (circles; n = 5), R (triangles; n = 5), and NR (squares; n = 12). Statistics: one-way ANOVA and Tukey’s multiple comparison test (***P < 0.0001).

Finally, we ruled out the possibility that different proportions of CD8+ and CD56+ cells could account for the differing cytotoxic activity because the average frequency in the PBMCs of responders and nonresponders was similar (fig. S4, C and D). Therefore, we conclude that the presence of activated cytotoxic cells in the recipient is predictive of MSC therapeutic activity.

MSC apoptosis induced by cytotoxic cells is the result of a bystander effect

To define the mechanisms that drive apoptosis in MSCs, we used in vitro–activated PBMCs from healthy donors as effector cells. We found that activated, but not resting, PBMCs induced extensive early apoptosis (annexin-V+/7AAD) in MSCs (Fig. 4A), which peaked at 4 hours and shifted toward late apoptosis (annexin-V+/7AAD+) by 24 hours (fig. S5A). In accordance with our in vivo observations (Fig. 1, A and B), only activated PBMCs induced caspase activation in MSCs, with a peak at 90 min (videos S1 to S3), and this was completely abrogated by the pan-caspase inhibitor Z-VAD-FMK (Fig. 4B, fig. S5B, and video S4).

Fig. 4. MSC apoptosis is mediated by activated CD8+ and CD56+ cytotoxic cells and is the result of a bystander effect.

(A) PBMCs from healthy donors (each independent experiment used a different PBMC donor) were activated using phytohemagglutinin (PHA; PHA-aPBMCs) or MLR (MLR-aPBMCs). Resting PBMCs (gray bars), PHA-aPBMCs (black bars), or MLR-aPBMCs (dashed bars) were incubated with MSCs at the indicated ratios and MSC apoptosis (annexin-V+/7-AAD) calculated after 4 hours. ND, not done. (B) Apoptosis in MSCs cultivated with MLR-aPBMCs in the presence or absence of the pan-caspase inhibitor Z-VAD-FMK (10 μM) or the corresponding concentration of its vehicle (dimethyl sulfoxide). (C and D) Apoptosis in MSCs cultivated with MLR-aPBMCs used as unfractionated or positively selected for CD11b+, CD4+, CD8+, or CD56+ cells (C) or depleted of CD56+, CD8+, or both (D). (E) Apoptosis in MSCs cultivated with MLR-aPBMCs in the presence or absence of the GrB inhibitor Z-AAD-CMK (300 μM) or the perforin inhibitor EGTA (4 mM). (F) Apoptosis in MSCs cultivated with PHA-aPBMCs in the presence or absence of neutralizing concentrations (10 and 100 μg/ml) of FAS-L monoclonal antibody anti-CD178. (G) Apoptosis in MSCs after culture with autologous (black bars) or allogeneic (gray bars) PHA-aPBMCs in the presence or absence of neutralizing doses of anti–HLA-A, anti–HLA-B, anti–HLA-C, or anti-HLA-DR antibodies. The white bar shows spontaneous apoptosis in MSCs plated alone. (H) Apoptosis in MSCs cultivated with MLR-aPBMCs in direct contact or in a Transwell. (I) Apoptosis in MSCs cultivated with PHA-aPBMCs in the presence or absence of escalating doses (10 to 75 μM) of PKCζ-PS (protein kinase Cζ–pseudosubstrate). In (B) to (I), the PBMC/MSC ratio was 20:1. Results represent the mean ± SD of three or six (H) independent experiments. Statistics: one-way ANOVA with Tukey’s multiple comparison test (*P < 0.5, **P < 0.01, and ***P < 0.001).

To identify the cells inducing apoptosis in MSCs, we performed selective enrichment and depletion experiments among activated PBMCs. We found that CD56+ natural killer (NK) and CD8+ T cell populations were the only cells responsible for initiating MSC apoptosis (Fig. 4, C and D). To characterize the mechanisms mediating MSC apoptosis induced by activated cytotoxic cells, we studied potential factors involved in caspase 3 activation. Inhibition of either granzyme B (GrB) or perforin completely abolished the ability of activated PBMCs to kill MSCs (Fig. 4E) and activate caspase 3 (fig. S5C and videos S5 and S6). We also observed reduced PBMC-mediated cytotoxicity when the CD95 ligand [CD95L; also known as the Fas ligand (Fas-L) or apoptosis antigen 1 ligand] was neutralized (Fig. 4F), but not when TNF-α or TNF-related apoptosis-inducing ligand was inhibited, even in the presence of very high concentrations of their respective inhibitors (fig. S5D).

We then interrogated the nature of the MSC-cytotoxic cell interaction. We observed that apoptosis was not affected by the presence of anti–human leukocyte antigen (HLA) class I or anti-HLA class II neutralizing antibodies. Consistently, the cytotoxic activity of activated PBMCs against autologous or allogeneic MSCs did not differ (Fig. 4G). However, although PBMCs required physical contact with MSCs to induce apoptosis (Fig. 4H), blocking immunological synapse formation by inhibiting the polarization of microtubule-organizing center (29) had no effect (Fig. 4I). These results demonstrate that MSC killing by activated cytotoxic cells is a bystander effect that does not involve the immunological synapse.

MSC apoptosis does not interfere with the recognition of the specific target of cytotoxic cells

Having determined that the MSC apoptosis induced by cytotoxic cells is MHC-independent and non–antigen-specific, we asked whether MSCs could exert their immunosuppressive effects by competing with and antagonizing antigen-specific recognition. New York esophageal squamous cell carcinoma 1 (NY-ESO1)–specific human CD8+ T cell clone (4D8) or interleukin-2 (IL-2)–activated polyclonal human CD56+ purified NK cells were used as effector cells against NY-ESO-1 peptide–pulsed T2 or K562 target cells, respectively. T2 cells are deficient in a peptide transporter for antigen processing and therefore can only present exogenously pulsed peptides (30). K562 is a leukemia cell line selectively sensitive to NK cytotoxicity (31). Two different sets of experiments were performed. In the first set, 4D8 or NK cells were tested against fixed numbers of putative (susceptible) target cells in the presence of escalating numbers of MSCs used as a cold target. The alternative condition consisted of escalating the numbers of the putative specific target cells—now used as cold targets—in the presence of a fixed number of MSCs then considered as the susceptible target. MSCs did not compete with antigen-specific T cell cytotoxicity because the killing of peptide-pulsed T2 cells was not affected by the presence of MSCs (Fig. 5A). The same results were obtained using NK cells (Fig. 5B). In contrast, the presence of the putative target cells markedly reduced MSC killing in a dose-dependent manner in both systems (Fig. 5, C and D). Our data show that MSC killing does not interfere with the primary recognition of the cognate antigen.

Fig. 5. MSCs do not compete with cytotoxic cell recognition of the cognate target.

(A) Apoptosis in T2 cells after culture with 4D8 cells at a 4D8/T2 ratio of 20:1. Where indicated, increasing concentrations of MSCs (used as cold target) were added. Apoptotic T2 cells were identified as annexin-V+/7-AAD+ cells. (B) Apoptosis in K562 cultured with NK cells (NK/K562 ratio of 20:1). Where indicated, increasing concentrations of MSCs (used a cold target) were added. (C) Apoptosis in MSCs cultured with 4D8 cells (4D8/MSC ratio of 20:1). Where indicated, increasing concentrations of T2 cells (used as cold target) were added. (D) Apoptosis in MSCs cultured with NK cells at an NK/MSC ratio of 20:1. Where indicated, increasing concentrations of K562 (used as cold target) were added. In all experiments, apoptosis of MSCs, T2 cells, or K562 cells was assessed after 4 hours of coculture by flow cytometry. Results represent the mean ± SD of three independent experiments. Statistics: one-way ANOVA and Tukey’s multiple comparison test (*P < 0.05).

Apoptotic MSCs are immunosuppressive in a TH2-type inflammation model

Our data imply that, because MSC killing does not interfere with the primary recognition of the cognate antigen, induction of apoptosis must be prominently involved in the immunosuppressive activity. Accordingly, in the GvHD model described above, MSC apoptosis produced by recipient cytotoxic cells is required for immunosuppression. Therefore, we asked whether this causative relationship remains valid in a different disease model associated with noncytotoxic type 2 T helper cell (TH2)–mediated inflammation. We selected the model of ovalbumin (OVA)–induced allergic airway inflammation (32) summarized in fig. S6A. Although cytotoxic immune cells have been implicated as contributing to the induction of this condition (33, 34), CD8+ and NK1.1+ cells infiltrating the bronchoalveolar lavage (BAL) and lung tissues were less than 2% 1 hour after the last OVA challenge, when MSCs were infused (fig. S6, B to E). To confirm the absence of MSC killing, mice received luc-MSCs to assess caspase activation after infusion and imaged 1 hour later. No caspase activation was detected in any of the mice (Fig. 6, A and B). High signal could be detected in all animals receiving control d-luciferin (fig. S6, F and G).

Fig. 6. Apoptotic MSCs exert in vivo immunosuppression in a TH2-type inflammation model in the absence of cytotoxic cells.

(A) Luc-MSCs were injected into naïve (n = 3) and OVA + MSCs (n = 6) mice 1 hour after the last challenge. One hour later, mice received Z-DEVD-aminoluciferin and were imaged in three independent experiments. White lines separate multiple photographs assembled in the final image. (B) TLS was measured from (A) (mean ± SD). (C) Eighteen hours after MSC infusion, eosinophil infiltration was assessed in the BAL of naïve (n = 3) and those infused with MSCs (n = 3), OVA (n = 6), and OVA + MSCs (n = 6) in two independent experiments. Means ± SD are shown. (D) Eosinophil infiltration (mean ± SD) evaluated in BAL cytospin preparations from OVA-sensitized mice treated with apoMSCs. Groups were as follows: OVA without apoMSCs (n = 6), OVA treated with apoMSCs (1 × 106; n = 7), and naïve mice receiving apoMSCs (1 × 106; n = 2). Results represent the mean ± SD of three independent experiments. Statistics in (B): unpaired t test. Statistics in (C) and (D): one-way ANOVA and Tukey’s multiple comparison test (*P < 0.05).

The therapeutic activity, assessed by quantitating the eosinophil infiltration in the BAL, showed no difference between MSC-treated and untreated mice (Fig. 6C). Together, these results indicate that, also in this model, MSC immunosuppression relies on the presence of recipient cytotoxic cells that mediate MSC apoptosis. Therefore, we decided to test whether MSCs, made apoptotic in vitro by exposure to anti-FAS and GrB (apoMSCs), could bypass the need of cytotoxic cells in the recipient and ameliorate eosinophil infiltration. When apoMSCs were administered to recipient mice, we observed that the eosinophil infiltration in BAL was much reduced (Fig. 6D).

Apoptotic MSCs infused in GvHD are immunosuppressive and induce IDO production in recipient phagocytes

We subsequently investigated whether apoMSCs could be immunosuppressive also in the GvHD model. ApoMSCs were administered either intravenously or intraperitoneally, and the infiltration of CD8+Vβ8.3+ Mh T cells was assessed and compared to untreated GvHD mice. ApoMSCs produced a substantial reduction in GvHD effector cell infiltration in both the spleen and lungs, but this could only be observed in those mice treated with apoMSCs infused intraperitoneally (Fig. 7, A to D).

Fig. 7. ApoMSCs exert immunosuppressive activity in GvHD and elicit IDO in engulfing recipient phagocytes.

(A to D) The percentage of GvHD effector cells was assessed in the lymphocyte gate in the spleen (A and C) and lungs (B and D) of GvHD mice (black circles) and those treated with apoMSCs (black squares). ApoMSCs were infused intraperitoneally (ip; GvHD mice, n = 10; GvHD + apoMSCs mice, n = 8) (A and B), or intravenously (iv; GvHD mice, n = 9; GvHD + apoMSCs mice n = 7) (C and D). Results represent the mean ± SD of three independent experiments. Statistics: unpaired t test (*P < 0.05 and **P < 0.01). (E to K) MSCs were labeled using CellTrace Violet and subjected to apoptosis induction using GrB/anti-FAS (5 and 10 μg/ml, respectively). ApoMSCs were injected intraperitoneally (E, F, and J) or intravenously (G to I and K) into GvHD mice 3 days after the transplant. After 2 hours, animals were sacrificed, and mesenteric lymph nodes (LN) (E, F, and J) or lungs (G to I and K) were harvested. Cells engulfing apoMSCs were identified as Violet+ cells within the CD11b+ (E), CD11c+ (F), CD11bhighCD11cint (G), CD11c+CD11b (H), and CD11bhighCD11c (I) subpopulations. The corresponding subpopulations were gated in GvHD mice, which had not received Violet-labeled apoMSCs and used as controls. IDO expression was assessed in CD11c+ and CD11b+ (J) or CD11bhighCD11cint, CD11c+CD11b, and CD11bhighCD11c (K) cells positive for CellTrace Violet (engulfing apoMSCs) and compared with the corresponding populations in GvHD mice that had not received apoMSCs. Data are representative of similar results obtained from three mice in two independent experiments.

It has been reported that the injection of irradiated thymocytes into animals results in their phagocytosis by recipient macrophages and induction of IDO (35). We therefore tested whether apoMSCs followed the same destiny by eliciting in vivo efferocytosis by recipient phagocytes and by inducing IDO production. For this purpose, labeled apoMSCs were traced in recipient phagocytes after injection. After intraperitoneal administration, apoMSCs were largely identified inside CD11b+ (Fig. 7E) and CD11c+ (Fig. 7F) phagocytes in the peritoneal draining lymph nodes (36) but absent when searched for in the lungs and spleen. When the intravenous route was used, among the several phagocytic populations investigated (37), CD11bhighCD11cint, CD11bhighCD11c, and CD11bCD11c+ were detected as engulfing apoMSCs in the lungs (Fig. 7, G to I, respectively). The analysis of IDO expression in the phagocytes engulfing apoMSCs in both the intravenous and intraperitoneal groups revealed that only the phagocytes in the intraperitoneal group were able to increase IDO expression in comparison with their counterparts in untreated GvHD mice (Fig. 7, J and K). These findings strongly suggest that the immunosuppressive effect of apoMSCs involves recipient phagocytes and IDO as crucial effector mechanisms.

Recipient-derived IDO-producing phagocytes are indispensable for MSC immunosuppression in GvHD

To directly test the importance of recipient-derived phagocytes or recipient-produced IDO in MSC immunosuppressive activity, we depleted phagocytes, inhibited IDO activity in GvHD mice before MSC treatment, and evaluated the effect of live MSCs on the expansion of GvHD effectors. To deplete phagocytes, liposome clodronate was given to mice 72 hours before MSC injection. The treatment substantially impaired the ability of MSCs to suppress Mh T cell infiltration (Fig. 8, A and B). Finally, animals were given the IDO inhibitor 1-methyl-d-tryptophan (1-DMT) (38) before MSC injection. In addition, in this case, the beneficial effect of MSCs on Mh T cell infiltration was much reduced in mice receiving 1-DMT compared to controls (Fig. 8, C and D). We therefore conclude that the immunosuppressive effect of MSCs requires the presence of recipient phagocytic cells or IDO production.

Fig. 8. Recipient phagocytes and IDO production are required for MSC immunosuppressive activity in GvHD.

(A and B) GvHD mice were treated with liposomal clodronate 10 min after the transplant. Where indicated, MSCs were infused 3 days later. The percentage of GvHD effector cells (CD8+Vβ8.3+) was calculated in the lymphocyte gate in the spleen (A) or lungs (B) after 4 additional days. Mean ± SD was obtained by grouping three independent experiments with n = 12 (GvHD) and 10 (GvHD + MSCs) mice per group. (C and D) GvHD effector cell infiltration was studied in the spleen (C) and lungs (D) of GvHD mice treated with the IDO inhibitor 1-DMT. In the treated mice, MSCs were infused 3 days after the transplant (n = 11). Controls consisted of GvHD mice that did not receive MSCs (n = 9). Percentage of CD8+Vβ8.3+ cells refers to the lymphocyte population. Results refer to the mean ± SD of three independent experiments. Statistics: unpaired t test (*P < 0.05 and **P < 0.01).


This study sheds light on the controversial topic of MSC therapeutics by identifying a crucial mechanism that potentially explains several unresolved issues in the field. The first striking piece of information provides the resolution to the paradox that MSCs are therapeutically efficacious despite the lack of engraftment (3941). We demonstrate that MSCs undergo extensive caspase activation and apoptosis after infusion in the presence of cytotoxic cells and that this is a requirement for their immunosuppressive function. Although other recipient-dependent reactions have been described as mediating MSC lysis in vitro (42) and MSC clearance in vivo (40, 41, 43, 44), our study has shown the instrumental role of in vivo MSC apoptosis in delivering immunosuppression after infusion. Furthermore, although several studies (38, 4547) have reported the ability of apoptotic cells to modulate immune responses, here, we provide evidence that in vivo naturally occurring cell death drives immunosuppression.

MSC apoptosis requires and is effected by cytotoxic granules contained in recipient cytotoxic cells that also mediate GvHD in recipient mice. The cytotoxic activity against MSCs can also be detected in the PBMCs of GvHD patients, and it is predictive of clinical responses. Patients displaying high cytotoxicity respond to MSCs, whereas those with low or absent cytotoxic activity do not improve after MSC infusion. Therefore, the ability of the recipient to generate apoptotic MSCs appears to be a requirement for the therapeutic efficacy and could be used as a potential biomarker to stratify patients for MSC infusions. However, the limited number of patients analyzed prompts further validation in a clinical study. Moreover, characterizing the phenotype of the cytotoxic cells mediating MSC apoptosis in patients can provide important information for the development of a routine biomarker assay.

MSC recognition by cytotoxic cells is not antigen-specific because it neither requires HLA engagement nor results from an alloreactive rejection, thus supporting the current practice of using third-party MSCs. MSCs must be in physical contact with the activated cytotoxic cells to undergo apoptosis, although immunological synapse is not required. This supports a bystander role for the cytotoxic granules released by the activated cytotoxic cell. Accordingly, it has been described that lytic granule secretion precedes the formation of cytotoxic T lymphocyte/target cell synapse (29). Furthermore, such a nonspecific mechanism can mediate tissue damage in the context of HIV replication (48) or atherosclerosis (49), whereby activated CD4+ or NK T cells have been implicated in the progression of HIV infection or the atherosclerotic disease, respectively. In these studies, bystander cells are not of mesenchymal origin, thus raising the question (that we have not addressed here) of whether nonspecific induction of apoptosis and subsequent immunosuppression is selective for MSCs.

Our data suggest an approach to MSC therapeutics that highlights the key role of MSC recipient to orchestrate and determine MSC effector functions. Not only are cytotoxic cells in the recipient required to initiate apoptosis in infused MSCs but also phagocytes that, by engulfing apoptotic MSCs and producing IDO, ultimately deliver MSC immunosuppressive activity. Similar mechanisms have been described to explain how apoptotic cells of different lineages, generated in vitro, induce immune modulation in GvHD (4547) and macrophage IDO production in other systemic autoimmune diseases (38). This is also consistent with the described ability of MSCs to stimulate recipient immune tolerance networks, such as regulatory T cells and macrophages (18, 19).

The depletion of recipient macrophages or the inhibition of IDO activity also impairs the therapeutic activity of live MSCs, thereby linking in vivo MSC apoptosis with immunosuppression. It is unlikely that any particular phagocyte population (macrophages or dendritic cells) is selectively involved in engulfing apoMSCs because they similarly display such an activity in vivo. However, because clodronate exhibits a preferential depleting activity on macrophages compared to dendritic cells, our data suggest that macrophages play a more important role.

One of the impacts of our study is that, although MSCs remain as the necessary starting point for therapeutic immunosuppression, patient-derived cells play a crucial role in delivering such an immunosuppression. Therefore, the efforts aimed at identifying the most clinically effective MSC subpopulation and the potency assays to validate such a selection may prove futile. A further proof supporting this concept is that the administration of ex vivo–generated apoMSCs can circumvent the requirement for cytotoxic cells in a TH2 inflammatory model and that apoMSCs can be effective at suppressing the expansion/infiltration of the GvHD effector cells. ApoMSCs were mostly effective in the GvHD model only when administered intraperitoneally. Despite being phagocytosed, apoMSCs injected intravenously did not induce IDO production, thus suggesting that the site at which MSC apoptosis occurs may influence the immunosuppressive function, perhaps by engaging with a subpopulation of phagocytes. Therefore, a more thorough characterization of the administration modality is required before testing apoMSCs in the clinical setting.

A final question is whether a cytokine-dependent “licensing” (25, 50) coexists with the generation of apoptotic MSCs. Although our data indicate that cytokine licensing is not required for the therapeutic activity, we cannot exclude that, before undergoing apoptosis, MSCs directly inhibit inflammatory reactions through the conventional pathways. Furthermore, caspase activation in MSCs may trigger cell death–independent pathways that stimulate the synthesis of immunomodulatory molecules independently of the generation of signals for phagocytosis (51). Consistent with this, it has been shown that MSCs activate caspase-dependent IL-1 signaling that enhances the secretion of immunomodulatory molecules (52).

Our study constitutes a paradigm shift in MSC therapeutics, whereby their apoptotic demise is a key step in the effector mechanism of immunosuppression exerted by MSCs. A further impact of our discovery is that the principle underpinning this mechanism could be used as a biomarker to predict clinical responses to MSCs and therefore stratify GvHD patients for MSC treatment. We therefore believe that the next generation of clinical trials should shift the focus from choosing the best MSC population to choosing the patients most likely to respond. Furthermore, the intriguing possibility that apoMSCs may be effective in patients refractory to MSCs paves the way to new avenues in the clinical manufacturing of MSCs.


Study design

This study aimed to verify whether MSCs undergo apoptosis after infusion and to test the role played by MSC apoptosis in the initiation of recipient-derived tolerogenic immune response. A mouse model of GvHD, in which the disease is mediated by the expansion and activation of Mh CD8+ T cells in the recipient, was chosen for three important reasons: It recapitulates a minor mismatch between donor and recipient, T cells effecting GvHD can be precisely enumerated, and there is proof of principle that MSCs are effective in treating GvHD. Furthermore, human MSCs were used to avoid the confounding effects of recipient cytokines on MSC immunomodulating function (10, 50). In this system, murine inflammatory cytokines will not cross-react with the corresponding human receptors and will not activate immunosuppressive molecules in human MSCs (2426) while retaining the ability to expand murine effector cells mediating GvHD. Depletion of phagocytes and inhibition of IDO production were conceived as loss-of-function experiments to assess the requirement of these factors in the delivery of MSC apoptosis–dependent immunosuppression. A mouse model of OVA-induced allergic airway inflammation was used to assess whether the causative relationship between cytotoxic cells and MSC apoptosis in the delivery of MSC immunosuppression is valid in a disease associated with TH2-type inflammation. All animal procedures were carried out in compliance with the UK Home Office Animals (Scientific Procedures) Act of 1986.

In all experiments, animals were randomly allocated to control or experimental groups. No blinding approach was adopted. No statistical method was used to predetermine sample size, which was estimated only on previous experience with assay sensitivity and the different animal models. Unless otherwise specified, three independent experimental replicates were performed.

To demonstrate that the presence of cytotoxic cells against MSCs in GvHD patients could be predictive of MSC therapeutic activity, samples from GvHD patients were collected and tested for their ability to induce MSC apoptosis in a cytotoxic assay within 24 hours before MSC infusion. At the time of performing the assay and cytofluorimetric analysis, the operator was blind to patients’ clinical details. All patients were affected by steroid-resistant GvHD and received MSCs for compassionate use. PBMCs from healthy donors were used as controls. All samples were collected after informed consent was obtained in accordance with the local ethics committee requirements. Primary data are located in table S2.

Mice and disease models

Acute GvHD was induced as previously described (22). Briefly, after lethal irradiation (11 gray), recipient C57BL/6 male mice were transplanted intravenously with 1 × 106 purified CD8+ T cells from female Mh mice, together with 5 × 106 unfractionated BM and 2 × 106 purified CD4+ T cells from female C57BL/6 wild-type donors. The control group received BM and purified CD4+ T cells only. C57BL/6-Prf1tm1Sdz/J (Perforin–/–) mice were purchased from the Jackson Laboratory and bred with Mh Rag2–/– mice, and the resulting offspring was intercrossed for two generations to obtain Mh Rag2–/– Perf knockout F3 mice. For the depletion of all phagocytes, mice received 1 mg of liposome clodronate ( intravenously 72 hours before MSC infusion (35). Recipient IDO activity was inhibited by using 1-DMT treatment (2 mg/ml; Sigma-Aldrich) in the drinking water starting from 6 days before MSC injection until animals were sacrificed (38).

OVA-induced airway inflammation was induced as previously described (32). Briefly, female Balb/C mice (Harlan Laboratories) were injected intraperitoneally with 30 μg of chicken egg albumin (OVA type V; Sigma-Aldrich) on days 0 and 7. Controls received vehicle (aluminum hydroxide) only. On days 14, 15, and 16, animals were challenged with an aerosolized solution of OVA (3%) administered with a DeVilbiss UltraNeb 90 for 25 min. MSCs or apoMSCs were injected 1 hour after the last challenge. After additional 18 hours, mice were terminally anesthetized, a cannula was inserted into the exposed trachea, and three aliquots of sterile saline were injected into the lungs to obtain BAL fluid. The total number of cells in the lavage fluid was counted.

MSC preparations

Clinical-grade BM-derived human MSCs were generated from BM aspirates collected from the iliac crest of healthy donors. The cells were plated at a density of 10 to 25 million/636 cm2. After 3 days at 37°C and 5% CO2, nonadherent cells were discarded. When cell confluence of 90 to 100% was achieved, cells were detached with trypsin-EDTA and reseeded at a density of 5000 cells/cm2. MSCs from different donors were used at passage 2 for all in vivo experiments, whereas they were used at passage 8 for the in vitro experiments. In the latter case, we did not observe any difference in terms of apoptosis susceptibility between different passages. In each experiment, MSCs were derived from a single expansion and not pooled.


Between November 2012 and July 2016, 16 patients affected by steroid-resistant GvHD were treated with MSCs according to Regulation (European Commission) no. 1394/2007. All patients received GvHD prophylaxis. Of the 16 patients included in the study, 13 developed GvHD after hematopoietic stem cell transplantation, and the remaining 3 developed GvHD after donor lymphocyte infusion. Twelve patients were affected by acute GvHD, 3 were affected by late onset acute GvHD, and 1 was affected by chronic GvHD. The diagnosis of GvHD was made on histological criteria and GvHD staged according to standard criteria (53, 54). Patient characteristics are summarized in table S1. Samples were collected within 24 hours before MSC injection.


Results were expressed as mean ± SD. The unpaired Student’s t test was performed to compare two mean values. One-way ANOVA and Tukey’s multiple comparison tests were used to compare three or more mean values. A probability of null hypothesis less than 5% (P < 0.05; two-sided) was considered statistically significant. See the Supplementary Materials for the full experimental procedures.


Materials and Methods

Fig. S1. MSCs can be traced in the lungs of mice after infusion.

Fig. S2. Human MSC immunosuppression is not licensed by murine cytokines.

Fig. S3. MSC apoptosis is induced by cytotoxic cells.

Fig. S4. Cytotoxicity against MSCs varies among PBMC donors but is independent on the percentage of CD8+ or CD56+ in GvHD patients.

Fig. S5. MSC killing is mediated by caspase 3 and effected by GrB and perforin.

Fig. S6. Infused MSCs can be imaged in the lungs of mice with TH2-type lung inflammation.

Table S1. Clinical features of GvHD patients.

Table S2. Primary data.

Video S1. Living-cell imaging of fluorescence resonance energy transfer (FRET)–MSCs plated alone.

Video S2. Living-cell imaging of FRET-MSCs plated with PHA-aPBMCs.

Video S3. Living-cell imaging of FRET-MSCs plated with resting PBMCs.

Video S4. Living-cell imaging of FRET-MSCs plated with PHA-aPBMCs in the presence of the pan-caspase inhibitor Z-VAD-FMK.

Video S5. Living-cell imaging of FRET-MSCs plated with PHA-aPBMCs in the presence of the GrB inhibitor Z-AAD-CMK.

Video S6. Living-cell imaging of FRET-MSCs plated with PHA-aPBMCs in the presence of the perforin inhibitor EGTA.

References (5557)


  1. Acknowledgments: We would like to thank D. Spina (Institute of Pharmaceutical Science, King’s College London) for the help and support for the statistical analysis. Funding: This work was supported by the Bloodwise Specialist Programme (14019 and 12006). A.G. is a recipient of a Bloodwise Clinical Research Training Fellowship (15029). Author contributions: A.G, H.W., and F.D. conceptualized the study. A.G., F.D, C.B., Y.R.-V., and R.C. developed the methodology. A.G. and L.D. carried out the formal analysis. A.G., Y.R.-V., C.T., C.L., L.B., S.W., K.H., T.S.C., and L.W. conducted the investigation. F.D., K.O., D.I.M., J.A., M.v.B., and M.B. provided the resources. A.G. and L.D. carried out the visualization. A.G. and F.D. wrote the original draft of the manuscript. A.G., C.B., G.L., F.M.W., H.W., and F.D. reviewed and edited the manuscript. F.D. acquired the funding and supervised the study. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Reasonable requests for additional data or materials will be fulfilled under appropriate agreements.

Stay Connected to Science Translational Medicine

Navigate This Article