The PTEN pathway in Tregs is a critical driver of the suppressive tumor microenvironment

See allHide authors and affiliations

Science Advances  06 Nov 2015:
Vol. 1, no. 10, e1500845
DOI: 10.1126/sciadv.1500845


The tumor microenvironment is profoundly immunosuppressive. We show that multiple tumor types create intratumoral immune suppression driven by a specialized form of regulatory T cell (Treg) activation dependent on the PTEN (phosphatase and tensin homolog) lipid phosphatase. PTEN acted to stabilize Tregs in tumors, preventing them from reprogramming into inflammatory effector cells. In mice with a Treg-specific deletion of PTEN, tumors grew slowly, were inflamed, and could not create an immunosuppressive tumor microenvironment. In normal mice, exposure to apoptotic tumor cells rapidly elicited PTEN-expressing Tregs, and PTEN-deficient mice were unable to maintain tolerance to apoptotic cells. In wild-type mice with large established tumors, pharmacologic inhibition of PTEN after chemotherapy or immunotherapy profoundly reconfigured the tumor microenvironment, changing it from a suppressive to an inflammatory milieu, and tumors underwent rapid regression. Thus, the immunosuppressive milieu in tumors must be actively maintained, and tumors become susceptible to immune attack if the PTEN pathway in Tregs is disrupted.

  • tumor immunology
  • tumor microenvironment
  • Regulatory T cells
  • Tregs
  • PTEN
  • indoleamine 2,3-dioxygenase


Tumors are not invisible to the immune system (14). Analysis of patients’ T cells suggests that many patients are chronically attempting to mount an immune response against their own tumors (58). However, in almost all cases, this attempted response is suppressed by potent inhibitory mechanisms within the tumor microenvironment. Multiple mechanisms contribute to this suppression (9), but one key component is the Foxp3+ regulatory T cell (Treg) system. Tumors actively recruit Tregs, and the Tregs in tumors become highly activated (10). If Tregs are transiently ablated in tumor-bearing mice, then the immune system rapidly attempts to attack the tumor (11, 12). However, in these models, complete depletion of Tregs leads to lethal autoimmunity, yet partial depletion of Tregs is not sufficient to reverse the suppressive milieu in the tumor (13). To date, it has not been possible to preferentially target the Tregs that are protecting the tumor.

To improve this selectivity, we asked whether tumors might preferentially depend on certain molecular pathways to activate Tregs. This activation step is a critical control point in Treg biology, because even fully mature Tregs require specific activation signals to become functionally suppressive (14). These signals include engagement of the T cell antigen receptor (TCR) (15, 16). However, antigen specificity alone does not confer selectivity for tumors, because many of the Tregs in tumors appear to recognize normal self-antigens (17). Besides antigen, Tregs also integrate a variety of modulating signals from the local microenvironment during activation (18). This can produce “tailored” forms of activation, with distinct patterns of transcription factors and functional activity (19). We hypothesized that the unique environment of tumors might preferentially elicit certain characteristic pathways of Treg activation.

The lipid phosphatase PTEN (phosphatase and tensin homolog) has recently been identified as an important signaling pathway in Tregs (20, 21). Targeted deletion of PTEN causes Treg instability (20), leading to progressive conversion into proinflammatory effector cells (so-called ex-Tregs), and eventual lupus-like autoimmunity. However, the mechanism by which PTEN stabilizes Tregs and the role that this might play in tumor biology remain unknown. Here, we show that many Tregs in the tumor microenvironment constitutively express PTEN and rely on this pathway to maintain their suppressive function. PTEN signaling in Tregs is required for the immune system to suppress responses to apoptotic cells, including apoptotic tumor cells. If this PTEN pathway is interrupted, then tumors lose the ability to create their usual highly suppressive milieu. The Tregs in tumors are destabilized and reprogram into inflammatory effector cells, the tumor milieu becomes proinflammatory and immunogenic, and even large established tumors undergo rapid regression.


Connection between IDO, PD-1, and PTEN in Tregs

Melanoma tumors in mice contained a large population of activated Tregs expressing the cell surface receptor programmed cell death-1 (PD-1) (Fig. 1A and fig. S1). PD-1 is a marker for activated Tregs in tumors (10), although its functional role remains unknown. The same PD-1+ Tregs also coexpressed high levels of the Foxp3 binding partner Eos (Ikzf4), which can be up-regulated in tumors by the immunoregulatory enzyme indoleamine 2,3-dioxygenase (IDO), as we have shown (22). IDO promotes tolerance and immunosuppression in the immune system (23), and it can directly activate Tregs in tumors (22, 24). These same PD-1+ EosHI Tregs also coexpressed PTEN phosphatase (Fig. 1A, right-hand panels).

Fig. 1 PTEN is required to maintain IDO-induced Treg activation.

(A) Analysis of peripheral Tregs from LNs of normal mice without tumor or from TDLNs and tumor of mice with B16F10 tumors, gated on CD4+Foxp3+ Tregs. (B) Proposed model for regulation of Akt by IDO and the PD-1→PTEN pathway during Treg activation, based on data from figs. S2 to S8. (C and D) Tregs from PTENTreg-KO mice or wild-type (WT) controls were activated in vitro for 2 days with IDO+ DCs from TDLNs, using the coculture system described in fig. S2A. (C) Tregs were analyzed for phosphorylation of Akt by FACS at the end of activation. (D) After activation, Tregs were re-sorted and tested in readout assays for their ability to maintain FoxO3a and PD-1. (E) PTENTreg-KO Tregs or WT (parental) control Tregs were activated either with IDO+ DCs from TDLNs or using conventional anti-CD3 mitogen with IDO blocked. After 2 days, Tregs were re-sorted and tested in readout assays for functional suppressor activity, as in fig. S2A. Each point is the mean of triplicate cocultures; error bars show SD (most are less than ±5%, smaller than the symbols). *P < 0.01 by analysis of variance (ANOVA) versus PD-1/L blockade (all other groups not significant versus PD-1/L blockade). (F and G) WT Tregs were activated with either IDO+ TDLN DCs or conventional αCD3 mitogen and then sorted and tested for functional suppressor activity (F) in the presence of PTEN inhibitor VO-OHpic. Mean of triplicate cocultures; SD bars are smaller than the symbols; *P < 0.01 by ANOVA versus no VO-OHpic. (G) In parallel experiments, the re-sorted Tregs were recovered at the end of the suppression assay and assessed for their level of Akt phosphorylation and detectable FoxO3a expression by FACS. Panels are representative of three to five independent experiments each.

To help elucidate the connection between IDO, PD-1, and PTEN in activated Tregs, we used the in vitro culture system shown in fig. S2. Resting Tregs were activated by coculture with IDO-expressing dendritic cells (DCs) and activated effector cells. After initial activation by IDO, the Tregs then became strictly dependent on PD-1 to maintain their functional activity (fig. S2A). We hypothesized that the IDO and PD-1 pathways might converge at the level of Akt kinase. In most T cells, high Akt signaling is required for normal activation; however, Tregs are unusual in that they must keep Akt signaling low during activation or else they will lose their suppressive phenotype (25). PD-1 is known to inhibit Akt via the activation of PTEN phosphatase (26, 27), but the effect of IDO on Akt was unknown. We found that IDO inhibited phosphorylation of Akt on Ser473 (fig. S2B). Ser473 is the target of the mTOR TORC2 complex; consistent with this, pharmacologic inhibition of mTOR could fully substitute for IDO during Treg activation (fig. S2C). In vivo, administration of an IDO inhibitor drug to tumor-bearing mice increased pAkt473 phosphorylation in Tregs (fig. S3). Thus, both IDO and PD-1 appeared to restrain excessive Akt signaling in activated Tregs.

The phosphorylation status of Akt is only a proxy marker and is very labile. We sought a more stable biomarker that might reflect the longer-term impact of dysregulated Akt signaling on the overall Treg phenotype. The transcription factor FoxO3a is important in Treg function (28), and it is sensitive to the activity of Akt (2931). In the short term, phosphorylation by Akt drives nuclear exclusion of FoxO3a and blocks its function (32), but in the long term, overactivity of Akt leads to degradation of FoxO3a (33). Fluorescence-activated cell sorting (FACS) analysis of Tregs showed that FoxO3a expression was highly bimodal, with some Tregs expressing none and some having strongly positive expression (fig. S4). In vitro, after activation cocultures, the fraction of Tregs that became FoxO3a-positive tracked with the suppressor activity controlled by IDO and mTOR (fig. S5A), whereas on a cell-by-cell basis, those Tregs that expressed PD-1 also coexpressed FoxO3a (Fig. 5B). Functionally, Tregs from FoxO3-deficient mice (28) were selectively unable to mediate the form of suppressor activity created by the IDO and PD-1, even though conventional CD3-induced Treg activity remained intact (fig. S5C). Thus, although FoxO3a itself was not unique or restricted only to IDO and PD-1 (29, 30), we hypothesized that loss (down-regulation) of FoxO3a could be an informative biomarker for loss of the suppressive Treg phenotype of interest.

IDO and PD-1 acted sequentially. IDO up-regulated expression of PD-1 on the Tregs via a process requiring tryptophan depletion and signaling via the amino acid–sensitive GCN2 kinase (34) (figs. S6 and S7, A to C). PD-1 was then required to maintain the IDO-induced activation state (fig. S8). If the PD-1 pathway was blocked, then Akt phosphorylation rapidly became high, FoxO3a was lost, and suppression activity was abrogated. On the basis of these findings, we hypothesized the model shown in Fig. 1B, in which IDO and PD-1 sequentially act to maintain the suppressive Treg phenotype. The key implication of this model was that maintenance of the suppressive phenotype required continuous, ongoing control of Akt: If this control was disrupted, then the suppressive phenotype was rapidly lost. Thus, the critical leverage point for continued suppression became PD-1 signaling via PTEN.

Maintenance of the Treg activation state by PTEN

PTEN is an important target for multiple upstream pathways, because it inhibits phosphatidylinositol 3-kinase (PI3K) and thus limits phosphorylation of Akt (26, 27). In addition to PD-1, PTEN is also downstream of neuropilin-1, a potent activator of Tregs (29). To assess the role of PTEN, we crossed mice bearing a floxed PTEN allele (35) with Bac-transgenic mice expressing a Cre-GFP (green fluorescent protein) fusion protein under the Foxp3 promoter (36). This Cre strain deletes in ~95% of Foxp3+ Tregs (36). Consistent with this, our PTENTreg-KO mice expressed Cre-GFP in ~95% of Foxp3+ cells and showed no detectable PTEN expression in Foxp3+ Tregs from tumors or tumor-draining lymph nodes (TDLNs) (fig. S9). The phenotype of this particular Cre/lox intercross was somewhat less penetrant than other PTEN-deficient strains that have been described (20), in that our mice were healthy when young and did not develop spontaneous autoimmunity until later in life (which is an important advantage for tumor studies). However, even in our young, healthy PTENTreg-KO mice, the Tregs were unable to control phosphorylation of Akt at Thr308 during Treg activation in vitro (Fig. 1C) and could not maintain the activated FoxO3a+ PD-1+ suppressive phenotype after re-sorting (Fig. 1, D and E). This was not due to a global defect in all Treg activity, because conventional CD3-induced Treg activity remained intact in these mice (Fig. 1E). Thus, the defect in PTENTreg-KO Tregs was a selective one, but—at least in the case of IDO—it profoundly compromised their ability to maintain the activated Treg phenotype.

We next asked whether PTEN could be pharmacologically targeted. PTEN inhibitor drugs are under active preclinical investigation for their neuroprotective and cardioprotective effects; we tested VO-OHpic, a high-affinity small-molecule inhibitor of PTEN (37), for its ability to block suppression by IDO-activated Tregs in vitro (Fig. 1F). In the absence of any Tregs, VO-OHpic had no effect on the readout T cells (neither toxic nor stimulatory). When control CD3-activated Tregs were added, VO-OHpic had no effect on their ability to suppress. However, when the same Tregs were activated by IDO, VO-OHpic fully blocked their suppressor activity, in a dose-dependent fashion. This was accompanied by progressive increase in phosphorylation of Akt in the Tregs and progressive loss of detectable FoxO3a (Fig. 1G). (In these studies, Thr308 was the direct target of PTEN→PI3K, and there was no IDO in the system to inhibit phosphorylation of Ser473; hence, both increased together as the Tregs became activated.)

Failure to create a suppressive tumor microenvironment in the absence of pten-Tregs

We next tested the effect of PTENTreg-KO hosts on tumor growth. Aggressive melanoma tumors implanted in PTENTreg-KO hosts grew much slower than the same tumors implanted in wild-type parental strains (Fig. 2A). Slower growth was also seen with E.G7 (EL4-OVA) and LLC tumors (fig. S10). Analysis of immune cells infiltrating the tumors (Fig. 2B) showed that wild-type hosts contained many PTEN+ Tregs that also coexpressed FoxO3a and PD-1, consistent with a suppressive phenotype. In contrast, the Tregs in PTENTreg-KO tumors did not express FoxO3a or PD-1; instead, they appeared unstable, with many expressing proinflammatory markers such as interleukin-2 (IL-2), CD40L, and IL-17 (Fig. 2B, lower panels, and fig. S11). All of these “reprogrammed” Tregs continued to express residual Foxp3 (Fig. 2B, bottom graph), thus showing that they were derived from former Tregs. The presence of these unstable, reprogrammed Tregs is consistent with our previous descriptions of Treg reprogramming in tumors when IDO is blocked (22, 38). Similar unstable ex-Tregs have been described during autoimmunity in mice with targeted deletion of PTEN in Tregs (20).

Fig. 2 Tumors in PTENTreg-KO hosts lose the ability to create a suppressive intratumoral milieu.

(A) Growth of B16F10 tumors in PTENTreg-KO hosts and WT B6 hosts. Pooled data from four experiments, n = 6 to 8 tumors per time point. *P < 0.05 versus WT, and all points thereafter. (B to D) Analysis of tumor-infiltrating immune cells in B16F10 tumors after 10 days of tumor growth in either PTENTreg-KO or parental Foxp3-GFP-Cre hosts: (B) Tregs, (C) CD8+ T cells, and (D) CD11c+ DCs. Representative of a total of nine experiments on days 10, 15, and 22. Intracellular cytokines were measured after 4 hours of activation with phorbol 12-myristate 13-acetate (PMA)/ionomycin. (E) Tumors were implanted on one side of PTENTreg-KO hosts; then, on day 14, the phenotype of T cells and DCs within the tumor was compared with pooled contralateral LNs, distant from the tumor. (F) Carboxyfluorescein diacetate succinimidyl ester (CFSE)–labeled pmel-1 cells, recognizing tumor-associated gp100, were transferred into WT or PTENTreg-KO hosts after a single dose of cyclophosphamide (CTX) to release tumor antigens. (G) PTENTreg-KO hosts or WT controls were implanted with B16F10 tumors or B16-OVA tumors bearing a nominal antigen. Mice then received a mixture of two CFSE-labeled responder cells: pmel-1 recognizing gp100 and OT-I recognizing OVA. Mice were then vaccinated against only the gp100 antigen, and tumors were analyzed 4 days later for evidence of epitope spreading (OT-I activation). (E) to (G) are representative of at least three experiments each.

The tumor-infiltrating CD8+ T cells in PTENTreg-KO hosts appeared more activated (Fig. 2C). Fewer CD8+ cells were PD-1+ (implying an “exhausted” phenotype) and more had an activated phenotype expressing CD103+, CD69+, and interferon-γ (IFN-γ). DCs in tumors from PTENTreg-KO mice also showed a more activated phenotype (Fig. 2D). More DCs expressed an activated myeloid DC phenotype of Ly6c+CD11b+CD103+, which has been associated with antitumor immune surveillance (39). Many of these CD103+ DCs produced IL-6 (bottom panels). This was significant because, as we will show, IL-6 proved to be a key driver of Treg reprogramming. Similar changes were seen when E.G7 lymphoma tumors were grown in PTENTreg-KO hosts (fig. S12).

These inflammatory changes were physically localized only to the tumor and TDLNs (Fig. 2E). Elsewhere in these (young, healthy) PTENTreg-KO hosts, the T cells and DCs appeared normal. This was informative, because it showed that the tumor became a potent local stimulus for inflammation when the host lacked the pten-Treg pathway.

In PTENTreg-KO hosts, antigen-specific CD8+ T cells became aware of tumor-associated antigens after chemotherapy. Mice with established B16F10 tumors were treated with a single dose of CTX to release tumor antigens and then received a CFSE-labeled cohort of resting pmel-1 T cells, reactive with a tumor-associated gp100 antigen. In PTENTreg-KO hosts, the resting pmel-1 cells became activated and proliferated in tumors, whereas proliferation was suppressed in wild-type hosts (Fig. 2F). Similarly, when tumor-bearing mice were treated with T cell adoptive transfer and antitumor vaccination, PTENTreg-KO hosts supported generalization of the immune response to new, additional T cells, recognizing antigen derived only from the tumor (Fig. 2G). No such epitope spreading occurred in wild-type hosts. (In all of these studies, the readout T cells were purified by negative selection, because this was important to prevent artifactual activation, as shown in fig. S13.) Thus, taken together, in PTENTreg-KO hosts, tumors were unable to create their usual suppressive tumor microenvironment and instead were regarded by the immune system as spontaneously inflammatory and immunogenic.

Reconfiguration of the suppressive microenvironment in established tumors

We next asked whether a similar antitumor inflammatory response could be induced in wild-type mice by pharmacologic inhibition of PTEN. This was a more demanding setting because the tumor had already established a suppressive milieu. Administering PTEN inhibitor by itself had no effect on growth of established tumors; however, if mice also received immunotherapy (vaccine plus pmel-1 T cells), then blocking PTEN had a potent effect, allowing rapid regression of tumors (Fig. 3A). Similar results were seen with ovalbumin (OVA)–expressing EL4 tumors and anti-OVA vaccine (Fig. 3B).

Fig. 3 Vaccination drives rapid reconfiguration of the suppressive tumor microenvironment when PTEN is blocked.

(A and B) WT C57Bl/6 mice with established B16F10 tumors (A) or E.G7 tumors (B) were treated with tumor-specific T cells (pmel-1 or OT-I) plus cognate vaccine, with or without VO-OHpic [10 mg kg−1 day−1 intraperitoneally (ip)] as indicated. Each curve represents pooled data from a total of 6 to 22 tumors in three to five independent experiments, measured serially. *P < 0.01 versus all other curves by ANOVA. (C) B16F10 tumors received vaccine and pmel-1 cells as in (A), with or without either VO-OHpic or CAL-101 (PI3K-δ inhibitor, 30 mg kg−1 day−1 ip). Tumors were harvested 4 days after vaccine and stained for Akt phosphorylation in gated GFP+ Tregs and for evidence of Treg destabilization and reprogramming. Representative of five experiments. (D) Effect of VO-OHpic on intratumoral OT-I responses and activation of tumor-associated DCs in mice with E.G7 tumors, treated with OT-I/vaccine as in (B), with or without VO-OHpic. Similar results were seen using B16F10 with pmel-1/hgp100 vaccine. (E) Requirement for both VO-OHpic and activated OT-I/vaccine to successfully drive Treg destabilization and DC activation in established E.G7 tumors. (D) and (E) are representative of at least three experiments each.

Analysis of treated tumors revealed that phosphorylation of Akt in Tregs remained low when tumors received only vaccine/T cells alone, but Akt phosphorylation became high when the PTEN inhibitor was added (Fig. 3C, red histograms). Phosphorylation was increased at both Thr308 and Ser473 sites. This is the expected pattern during activation of conventional (non-Treg) CD4+ cells (40), but it should not occur in Tregs, which need to keep Akt activity low. Thus, blocking PTEN caused dysregulated control of Akt in Tregs. (Notably, the Tregs were analyzed after 4 days of treatment; thus, Akt phosphorylation reflected the whole process of activation and destabilization, over and above the direct effects of the PTEN inhibitor on Thr308.)

Consistent with their inability to control Akt activity, intratumoral Tregs in PTENTreg-KO mice lost detectable FoxO3a and PD-1 (Fig. 3C, dot plots). They also became destabilized and up-regulated pro-inflammatory IL-2 and CD40L (that is, underwent reprogramming). As a biologic control for these effects, we included the PI3K-δ inhibitor CAL-101 (41), which inhibits Treg activity but does not affect PTEN. When tested in our in vitro model, CAL-101 inhibited Treg activation as expected, but it was not selective, inhibiting both IDO-induced and CD3-induced suppression (fig. S14). In vivo, CAL-101 did not increase Akt phosphorylation in Tregs, did not cause loss of FoxO3a and PD-1, and did not drive Treg destabilization and reprogramming (Fig. 3C). In our previous studies of Treg reprogramming, we had shown that TCR-mediated activation is strictly required for reprogramming to occur (22, 38, 42). Thus, CAL-101 appeared to inhibit all Treg activation (41), including the “destabilizing” activation that was required to drive reprogramming, whereas VO-OHpic allowed activation, but in a destabilizing fashion.

Further characterization of the tumor milieu after VO-OHpic treatment showed that vaccine-specific OT-I cells were suppressed when PTEN was active but were able to proliferate and up-regulate granzyme B when PTEN was blocked (Fig. 3D). Tumors also contained more activated, proinflammatory Ly6c+CD11b+ myeloid DCs when PTEN was blocked, and the DCs expressed more CD86 and less PD-L1 (Fig. 3D, lower panels). The contribution of the immunotherapy regimen (T cells and vaccine) was important, because successful transformation of the tumor microenvironment required both vaccine-activated T cells and inhibition of PTEN (Fig. 3E).

Finally, we noted that the PTEN inhibitor itself had no effect on a normal vaccine response in the absence of tumor (fig. S15A). The same was true for genetic ablation of PTEN (fig. S15B): In the absence of a suppressive tumor, ablating PTEN had no effect on the response to vaccine. Thus, the effect of PTEN inhibition was to reverse the specific tumor-induced suppression, rather than to nonspecifically augment all immune responses.

pten-Tregs and tolerance to apoptotic cells

PTEN+ Tregs were crucial for tumors, but it was not clear what role these cells normally played in the immune system. Other strains of mice with Treg-specific deletion of PTEN develop a spontaneous lupus-like autoimmunity as they age (20). Such “lupus-prone” phenotypes can reflect an inability to maintain tolerance to apoptotic cells (43). We have previously shown that apoptotic cells are potent inducers of IDO, and IDO-deficient mice fail to create tolerance to apoptotic cells and instead develop lupus autoimmunity (23). We triggered apoptosis in EL4 tumor cells by in vitro treatment with staurosporine for 4 hours, and then injected the apoptotic tumor cells into normal, tumor-free mice. Apoptotic tumor cells drove extensive up-regulation of IDO in draining LNs (DLNs) (Fig. 4A). This IDO expression was functionally important, because mice treated with an IDO inhibitor became responsive to antigen from the apoptotic cells, whereas responses were suppressed if IDO was active (Fig. 4B). Apoptotic tumor cells elicited a prominent population of PTEN+ Tregs expressing FoxO3a+ and PD-1+ in local DLNs, and induction of these Tregs was blocked by the IDO inhibitor (Fig. 4C). When Tregs were sorted from DLNs after challenge with apoptotic cells, these Tregs mediated potent, spontaneous suppression ex vivo, in the PD-1–dependent fashion characteristic of IDO-induced activation (Fig. 4D). In contrast, Tregs from resting LNs showed no spontaneous suppressor activity. [Readout assays in these experiments were driven by cognate antigen, with no CD3 mitogen, and under these conditions, resting Tregs do not show spontaneous suppression (14).] Thus, apoptotic tumor cells appeared to directly elicit the IDO-induced, PD-1–dependent PTEN+ FoxO3a+ PD-1+ form of Treg activation.

Fig. 4 The PTEN pathway in Tregs is required to suppress immune responses to apoptotic cells.

(A to D) Normal mice without tumors received footpad injection of dying EL4 tumor cells (treated for 3 hours in vitro with staurosporine to induce apoptosis). (A) After 48 hours, DLNs and unaffected contralateral LNs (CLNs) were stained for IDO (red chromogen). (B) OT-I response to cell-associated antigen from apoptotic EL4-OVA (E.G7) cells, or parental EL4 controls, with or without treatment in vivo with the IDO inhibitor drug in drinking water. (C) After apoptotic cell injection, Foxp3+ Tregs were compared in DLNs versus control distant CLNs for activation of pten-Tregs, with or without IDO inhibitor (indoximod). (D) Two days after challenge with apoptotic cells, Tregs were sorted from DLNs (or control LNs without injection) and directly tested ex vivo for constitutive suppressor activity, as described in fig. S2A, with or without PD-1/L blockade in the readout assay. Each point is the mean of triplicate cocultures; bars show SD (all are less than ±5%, smaller than the symbols). (E) Effect of PTEN inhibitor (VO-OHpic) on the immune response to apoptotic EL4-OVA cells (E.G7). Responses were compared in WT (parental) hosts and PTENTreg-KO hosts, with and without VO-OHpic. (F) Old mice (1- to 2-year-old retired breeders) and young mice (<6 months) from the PTENTreg-KO strain were tested for spontaneous antinuclear autoantibodies in serum (ANA, left) and deposition of immunoglobulin G (IgG) and complement C3 in the kidney (right). (G and H) Young, healthy PTENTreg-KO mice, or control WT B6 mice, were challenged with 2 × 107 apoptotic thymocytes weekly for four doses intravenously. Splenic B cells were analyzed for CD24 expression on day 28 (G). Serum titers of autoantibodies were compared at days 0, 14, and 28 (H). Pooled data from four to five mice; bars show SD.

To test whether these Tregs were mechanistically controlled by PTEN, we treated mice with the PTEN inhibitor during exposure to apoptotic cells, and we compared wild-type hosts to PTENTreg-KO hosts (Fig. 4E). When wild-type mice received VO-OHpic at the time of challenge, apoptotic cells elicited no FoxO3a+PD-1+ Tregs (Fig. 4E, top row). Instead, the apoptotic cells now induced inflammatory CD11b+CD103+ myeloid DCs, and the readout OT-I T cells were able to respond to antigens from apoptotic cells (lower rows). Similar results were seen when genetically defined PTENTreg-KO mice were challenged with apoptotic cells (Fig. 4E, right-hand plots). In these PTENTreg-KO mice, there was no additional effect of adding VO-OHpic, suggesting that the PTEN expressed in Tregs was an important target of VO-OHpic in this model.

On the basis of the response to apoptotic tumor cells, we asked whether the PTENTreg-KO mice had a generalized defect in tolerance to all apoptotic cells. Although the mice in our strain were healthy when young, some of the older mice (>1 year) spontaneously developed antinuclear antibody and immune complex deposition in the kidney, consistent with lupus-like autoimmunity (Fig. 4F). Therefore, to test the younger, asymptomatic PTENTreg-KO mice for an underlying defect in tolerance, we challenged them with intravenous injection of apoptotic thymocytes. These were syngeneic cells, and wild-type mice were fully tolerant to them; however, the PTENTreg-KO mice showed widespread B cell activation (assessed as loss of CD24, Fig. 4G) and rapidly developed multiple autoantibodies against classic apoptotic cell antigens (nucleosome, double-stranded DNA, and histone; Fig. 4H). Thus, when confronted with large numbers of apoptotic cells, the PTEN pathway in Tregs appeared critical for the host to create immune suppression and tolerance.

Synergy between chemotherapy and PTEN inhibition

Chemotherapy releases a wave of dying tumor cells. We hypothesized that the tumor might become critically dependent on the PTEN pathway after chemotherapy. Mice bearing established B16F10 tumors received a single injection of CTX, with or without VO-OHpic (Fig. 5A). The dose of CTX was modest and by itself had no effect on these resistant tumors. VO-OHpic alone also had no effect on tumor growth. However, together these agents displayed striking synergy, causing rapid shrinkage of the tumor within 4 to 5 days. Mice received no other immunotherapy or T cell transfer; hence, any immunologic effects were the spontaneous contribution of the host immune system. Chemotherapy plus VO-OHpic was well tolerated, without weight loss or other obvious toxicity. In validation studies, mice without tumors were treated with VO-OHpic for up to 21 days and did not develop detectable autoantibodies (fig. S16).

Fig. 5 Blocking PTEN allows rapid, spontaneous reconfiguration of the suppressive tumor milieu after chemotherapy.

(A) WT B6 mice with established B16F10 tumors were treated with a single dose of CTX (150 mg kg−1 ip) with or without VO-OHpic (10 mg kg−1 day−1 ip). Tumor volume is shown. Pooled data from five independent experiments; n = 10 to 16 tumors in each group; bars show SD. *P < 0.01 versus all other groups by ANOVA. (B to D) Mice with B16F10 tumors were treated as in (A) with CTX plus either VO-OHpic, CAL-101, or vehicle. (B) Phosphorylation of Akt in tumor-associated Tregs was assessed 4 days after CTX [numbers show the mean fluorescence intensity (MFI) of the positive population]. (C) Treg reprogramming in tumors after chemotherapy. Inset histograms show reduction in Foxp3 fluorescence with VO-OHpic. (D) DC activation in tumors after chemotherapy with VO-OHpic. (E) Rag1-KO hosts or WT B6 controls treated with CTX plus VO-OHpic. Mean of 16 tumors from two independent experiments; bars show SD (the bars in the WT group are smaller than the symbols). *P < 0.01 by ANOVA. (F) B6 mice received 2 × 106 Thy1.1-congenic resting pmel-1 cells and then were implanted with B16F10 tumors. After 9 days, mice received either CTX + VO-OHpic or no treatment, and the phenotype of pmel-1 cells in tumors was analyzed on day 14. Representative of three experiments. (G) The effect of CTX + VO-OHpic is abrogated by antibody-mediated depletion of CD8+ cells. FACS analysis of tumor-infiltrating cells from representative tumors is shown on the right. n = 6 tumors per group, pooled from three independent experiments, *P < 0.01 by ANOVA. (H) Aggregate phenotyping data for immune cells infiltrating tumors treated with CTX, with or without VO-OHpic. The number of independent experiments for each marker is given, with P value by two-tailed paired t test. Lines connect the two groups in each experiment.

In this chemotherapy model, inhibiting PTEN caused much more potent synergy than inhibiting IDO, which causes growth delay but not tumor regression as we have previously shown (44). However, combining an IDO inhibitor with VO-OHpic conferred additional benefit, delaying the regrowth of the tumor and prolonging the duration of response (fig. S17A). This was consistent with our in vitro model, in which PTEN maintained the function of Tregs that were already activated, but IDO contributed to activation of new Tregs. LLC lung carcinoma tumors also showed synergy between chemotherapy and inhibition of PTEN (fig. S17B).

Immune activation after chemotherapy when PTEN is inhibited

When PTEN was inhibited, the response of the host immune system to chemotherapy became fundamentally changed. Phosphorylation of Akt in Tregs went up (Fig. 5B), with associated loss of FoxO3a expression (Fig. 5C). Tregs were destabilized, as shown by reprogramming (expression of IL-2 and CD40L, Fig. 5C, lower panels). DCs acquired the activated myeloid DC phenotype (CD11b+CD103+) and many began to express IL-6, whereas PD-L1 was reduced and CD86 was markedly increased (Fig. 5D). In contrast, the Tregs and DCs from control mice receiving the PI3K-δ inhibitor CAL-101 showed none of these changes (Fig. 5, B to D). Functionally, both CAL-101 and VO-OHpic enhanced the antitumor effect of CTX, but VO-OHpic drove more sustained tumor regression (fig. S18). We hypothesized that the response to VO-OHpic might be related to the beneficial inflammatory effects of the destabilized (reprogrammed) Tregs. To test this, tumors were grown in mice with a transgenic diphtheria toxin receptor (DTR) under the control of the Foxp3 promoter, thus allowing global depletion of Tregs by administration of diphtheria toxin (DT, fig. S19). Ablation of Tregs with DT modestly enhanced the effect of CTX, but this effect was much less than the effect of VO-OHpic, and tumors showed no regression. When mice received both VO-OHpic and Treg ablation, the superior efficacy of VO-OHpic was lost, and the response became equivalent to DT ablation alone. Thus, the destabilized Tregs appeared to be an important driver of the antitumor response.

To directly test this, we bred mice that lacked the ability to undergo Treg reprogramming by ablating the IL-6 receptor on Tregs (IL6RTreg-KO mice). We have previously shown that IL-6 is required to drive Treg reprogramming (22, 38), which has also been seen by others (45, 46). Tregs from IL6RTreg-KO mice were unable to reprogram in vitro (fig. S20A). In vivo, treatment with CTX + VO-OHpic lost all antitumor effect (fig. S20, B and C). Thus, Treg reprogramming was functionally required for the synergy between VO-OHpic and chemotherapy, consistent with the role of PTEN in maintaining Treg stability (20, 21).

We next tested the contribution of effector CD8+ T cells. Studies in Rag-deficient mice showed that adaptive immunity was strictly required (Fig. 5E). To follow a known population of tumor-reactive CD8+ T cells during treatment, we pretransferred a cohort of Thy1.1 congenic pmel-1 T cells into normal wild-type host mice before implanting tumors (Fig. 5F). In untreated mice, pmel-1 cells in the tumor appeared unactivated (no expression of CD69, 1B11, or IFN-γ) and many expressed PD-1, suggestive of an exhausted phenotype. In contrast, when mice were treated with CTX + VO-OHpic, the pmel-1 cells in tumors down-regulated PD-1, and many now expressed CD69, 1B11, and IFN-γ. Finally, we used antibody-mediated depletion to determine whether endogenous CD8+ T cells were required. Depletion of CD8+ cells completely abrogated the antitumor effect of CTX + VO-OHpic (Fig. 5G), and the intratumoral Tregs did not lose their FoxO3a+ PD-1+ (suppressive) phenotype. This was consistent with our previous reports (22) showing that Treg destabilization is an active process, driven by activated effector T cells.

Together, across multiple experiments using B16F10, EL4, LLC, and autochthonous Tg(Grm1)Epv tumors, we observed consistent reconfiguration of the tumor milieu into a proinflammatory phenotype after CTX + VO-OHpic (Fig. 5H). These changes closely recapitulated the spontaneous changes seen when tumors were grown in genetically defined PTENTreg-KO mice (cf. Fig. 2). Similarly, fig. S21 shows that tumors in wild-type mice treated with CTX + VO-OHpic became indistinguishable from tumors grown in PTENTreg-KO hosts, whereas, conversely, tumors grown in PTENTreg-KO mice showed no further discernible effect of CTX + VO-OHpic. Thus, although PTEN can be expressed in multiple cell types, its expression in Tregs appeared to be an important target of VO-OHpic in tumor-bearing hosts.

Rapid regression of autochthonous tumors

Autochthonous tumors coevolve with the host immune system. This creates profound immune tolerance, and such tumors are much more resistant to the immunologic effects of chemotherapy than transplantable tumors (47). To test the role of PTEN in a demanding autochthonous setting, we used Tg(Grm1)Epv mice (48), which express the oncogenic Grm1 receptor under the melanocyte-specific Trp2 promoter. These mice progressively develop extensive multifocal melanomas on the ears and tail. When tumors became extensive (4 to 6 months of age), mice were treated with one dose of CTX plus VO-OHpic for 6 days. After treatment, even large, multifocal confluent tumors rapidly regressed (Fig. 6A). As a measure of regression, we used ear thickness as a proxy, because all ears were extensively involved with tumor (inset graph, Fig. 6A). Control groups receiving chemotherapy alone, VO-OHpic alone, or no treatment all showed no detectable effect (shown as the pooled controls on the graph). All the effects of CTX + VO-OHpic were lost when mice received depleting monoclonal antibody (mAb) against CD8 (upper inset graph), confirming that the effect was immune-dependent. Histologically, regression was accompanied by a selective elimination of the melanotic tumor cells, leaving underlying tissue intact (Fig. 6B). FACS analysis of regressing tumors showed emergence of the characteristic Ly6c+CD11b+-activated myeloid DCs (Fig. 6C), identical to those in the transplantable tumor models mentioned earlier. Tumor-infiltrating host CD8+ T cells lost their PD-1+ exhausted phenotype and up-regulated granzyme B (Fig. 6D). All Tregs in the tumor lost detectable FoxO3a and PD-1 expression after treatment (Fig. 6E), and the Tregs became unstable and began to express CD40L and IL-2 (Fig. 6F). Thus, even in mice with extensive, multifocal autochthonous tumors, a single dose of chemotherapy caused rapid and widespread tumor regression when PTEN was inhibited.

Fig. 6 Extensive autochthonous melanoma tumors regress after chemotherapy plus PTEN inhibitor.

Tg(Grm1)Epv mice with extensive, multifocal autochthonous melanomas were treated with a single dose of CTX plus VO-OHpic as shown. Some mice also received concurrent CD8-depleting antibody as described in Materials and Methods. (A) Regression of tumors on the ears and tail (ears are from representative mice; tails are before and after). Inset graph: quantitation of ear thickness (proxy for tumor involvement) after treatment. n = 10 to 16 ears per group serially measured from three independent experiments; bars show SD. *P < 0.01 by ANOVA versus all other groups. The “pooled controls” group contains CTX alone and VO-OHpic alone, neither of which showed any effect. To estimate tumor regression, we subtracted the normal thickness of healthy mouse ear (0.6 ± 0.03 mm, n = 12) and then we expressed the serial tumor measurements as percent change relative to the original pretreatment baseline for that ear. (B) Histology on day +4 of CTX/VO-OHpic [hematoxylin and eosin (H&E) stain; tumor cells are visible as black melanotic lesions]. Representative of four experiments. (C to F) Phenotype of cells from disaggregated tumor (several pooled tumors from the same mouse). Representative of five independent experiments.


Here, we identify PTEN signaling in Tregs as an important, centrally positioned driver of the immunosuppressive milieu in tumors. When this pathway was active, Tregs in tumors were highly suppressive; the antigen-presenting cell (APC) population was dominated by PD-L1–expressing DCs, with little evidence of inflammation or cross-presentation, and CD8+ T cells appeared unactivated and exhausted in the tumor. In contrast, if the PTEN pathway was ablated in Tregs, then the same tumors became spontaneously inflammatory and immunogenic: Tregs in the tumor (but not elsewhere) lost their suppressive phenotype and converted into proinflammatory helper cells (ex-Tregs); the major DC population was changed into a characteristic population of activated myeloid DCs expressing Ly6c, CD11b, and CD103 and producing IL-6; costimulation (CD86) went up and co-inhibition (PD-L1) went down; and CD8+ T cells became activated in the tumor. Thus, the PTEN pathway in Tregs dominantly regulated a suite of critical downstream pathways.

Tumors rely on multiple immunosuppressive mechanisms; thus, pten-Tregs are a part of a larger network, but it was striking that a single pathway in a single cell type was so crucial for the overall immunologic milieu of the tumor. pten-Tregs were found in three different transplantable tumors and in each of the spontaneous tumors tested in an autochthonous tumor model. It is not yet known how these different tumors all converged on the same suppressive pathway, but insight may come from our finding that PTENTreg-KO mice were unable to suppress immune responses to apoptotic cells. Tumors are aberrant and have a high rate of constitutive cell turnover—in many respects, they resemble chronic wounds (49). This chronic, ongoing cell death can become massively increased after chemotherapy or immunotherapy. Thus, the ability to enforce immune suppression in response to apoptotic cells may be an important immune-evasion pathway for tumors, particularly under the stress of chemotherapy or vaccination.

PTEN had connections to the IDO pathway. IDO-activated Tregs became dependent on PD-1→PTEN signaling to maintain their suppression, whereas apoptotic tumor cells actively induced pten-Tregs in an IDO-dependent fashion. However, the tumor-resistant phenotype of PTENTreg-KO mice was much more robust than that of IDO1-KO mice, and blocking PTEN had a more acute antitumor effect than blocking IDO alone. PTEN is downstream of other important Treg-activating signals as well, such as neuropilin-1 (29). Thus, PTEN appears to be an important nexus for multiple upstream signals in Treg biology. This makes it an attractive target for therapy.

Pharmacologic inhibition of PTEN was synergistic with chemotherapy. In the case of the melanoma models [B16F10 and Tg(Grm1)Epv], this synergy was dramatic. Different tumor types displayed different degrees of regression after CTX + VO-OHpic, and some tumors may recruit more myeloid-derived suppressor cells or other mechanisms that are not as dependent on PTEN. Part of the susceptibility of various tumors may be dictated by the strength of the spontaneous, preexisting CD8+ T cell response against the tumor. Using depletion studies, we found that endogenous host CD8+ T cells were strictly required for the antitumor effect of chemotherapy + PTEN inhibitor. In addition to being cytotoxic effector cells, these CD8+ cells appeared to play an important role as drivers of inflammation. Without the endogenous CD8+ cells, intratumoral Tregs never lost their suppressive phenotype, and there was no downstream activation of inflammatory myeloid DCs. Thus, although blocking PTEN rendered the Tregs unstable in the face of inflammation, the CD8+ T cells were required to actually trigger that inflammation itself and thus drive reprogramming.

This role for inflammation and T cells differs from the older concept that the immune system might contribute some of the intrinsic efficacy of standard chemotherapy (that is, chemotherapy used by itself) (50). Our model does not speak to this, because we deliberately chose a chemotherapy regimen that had no efficacy by itself against the tumor. It was only when the PTEN inhibitor was added, and new immunologic mechanisms were now unleashed, that the combined chemoimmunotherapy regimen showed efficacy. It was this new, synergistic effect of the combination that was strictly immune-dependent.

pten-Tregs were indispensable for regulating the immune response to apoptotic cells. In mice without tumors, simply injecting apoptotic tumor cells caused rapid local activation of pten-Tregs, and T cell responses to cell-associated antigens were suppressed. However, if PTEN was blocked or ablated, then the same apoptotic tumor cells were now treated as proinflammatory and caused activation of DCs and cross-presentation of antigen to T cells. Thus, whether the host treated apoptotic cells as immunogenic or tolerogenic was not intrinsic to the dying cells themselves, but rather was a choice enforced by the activated pten-Tregs. Consistent with this, when PTENTreg-KO mice were challenged with repetitive injections of apoptotic thymocytes, they rapidly developed multiple lupus-like autoantibodies. This phenotype is very similar to IDO1-deficient mice, challenged with the same regimen of apoptotic cells (23), which is consistent with our finding that apoptotic cells induced pten-Tregs via IDO. Thus, we hypothesize that one normal role of the pten-Treg pathway is to maintain self-tolerance to apoptotic cells, and this natural tolerogenic pathway is pathologically co-opted by tumors.

Other strains of mice lacking PTEN in Tregs spontaneously develop lupus autoantibodies as they age (20, 21). Our mice, using a different Cre system and different PTEN-floxed mice, had a somewhat milder phenotype and did not develop autoantibodies until late in life. However, both strains developed the same lupus-like autoimmunity, with its characteristic autoantibody response against antigens derived from apoptotic cells. The fact that our strain had late onset of disease was crucial, because it allowed us to demonstrate the markedly and fundamentally altered underlying response to tumors, even in healthy, asymptomatic young mice.

The role of PTEN in self-tolerance introduces the possibility of autoimmune side effects when PTEN inhibitors are used therapeutically. Autoimmunity is always a potential “on-target” toxicity for strategies that aim to break tolerance to tumor cells (51). However, the PTEN pathway has several attributes that make it attractive for clinical therapy. First, the required duration of inhibitor treatment is brief, being focused just on the specific period when tumor cells are undergoing cell death after chemotherapy or vaccination. Second, the tumor appears to be heavily dependent on the pten-Treg pathway for immunosuppression, much more so than normal tissues. This was shown by the spontaneous inflammation seen within the tumor in young, healthy PTENTreg-KO mice but not seen anywhere else outside the tumor. Third, tumors are mutated and hence potentially more immunogenic than normal tissues (52). This may contribute to the “driver” CD8 response needed to initiate inflammation within the tumor.

Together, these factors may allow brief, focused dosing with the PTEN inhibitor to achieve selective effects on the tumor while minimizing autoimmune toxicity. The use of intermittent dosing also mitigates the hypothetical concern that PTEN inhibitors might be oncogenic. PTEN is a tumor-suppressor gene, but it is not a transforming oncogene; thus, to have an oncogenic effect, it must be inhibited for a prolonged period and also combined with additional driver oncogenes. Hence, transient intermittent interruption and restoration of PTEN should not be an oncogenic regimen and may actually cause senescence in tumor cells (53). Together, our findings identify the PTEN pathway in Tregs as an important driver of the immunosuppressive tumor microenvironment and a potentially attractive target for therapy.



Indoximod (1-methyl-d-tryptophan) and NLG919 were supplied as clinical-grade material by NewLink Genetics Inc. For in vitro studies, indoximod was dissolved in alkaline pH as previously described (34) and diluted in phosphate-buffered saline (PBS) at a final pH of 7.4. Rapamycin (Sigma #R8781) was used at 85 nM. PP242 (Seleckchem #S2218) was used at 100 nM. Human gp10025–33 (KVPRNQDWL) and SIINFEKL peptides were synthesized by Southern Biotechnology from the published sequence (54, 55). VO-OHpic (BioVision #1801-5) was used at 1 μM in vitro unless otherwise specified (37) and at 10 mg kg−1 day−1 in vivo, administered in 10% dimethyl sulfoxide (DMSO). CAL-101 (Seleckchem #S2226) was used at 200 nM in vitro and at 30 mg kg−1 day−1 in vivo, administered in 10% DMSO.

Treg/DC cocultures

Initial activation cocultures. Our Treg coculture system has been previously described (22, 24). Cocultures were performed in V-bottom wells (Nunc 249952 V96) to ensure cell-cell contact. Rapid isolation of Tregs was important; hence, LNs and spleen were mechanically disaggregated by passing once through a 40-μm mesh and then briefly stained on ice and FACS-sorted into ice-cold medium using low-shear fluidics and a large nozzle. IDO+ DCs (5 × 103 per well) were enriched from TDLNs of B16F10 tumors (days 7 to 11) by sorting for CD11c+ cells, which typically contained 30 to 40% IDO+ DCs (24); because the effects of IDO were dominant, further fractionation was not necessary. Resting Tregs (2 × 104 per well) were sorted as CD4+GFP+ cells from spleens of Foxp3GFP, Foxp3DTR-GFP, or Foxp3GFP-Cre mice; all three strains gave equivalent results. CD8+ effector cells (5 × 104) were FACS-sorted from spleens of OT-I or pmel-1 mice and added to cocultures with 100 nM cognate peptide (either SIINFEKL or hgp100). All cultures received a feeder layer of 1 × 105 T cell–depleted B6 spleen cells (CD4NEGCD8NEG) to maintain Treg viability, as previously described (24). Cultures received either indoximod (200 μM) or NLG919 (1 μM) to inhibit IDO. For αCD3-induced activation, cocultures received αCD3 mAb (0.1 μg/ml; clone 145-2C11, azide-free); in addition, these cultures also always received an IDO inhibitor to ensure that IDO was not active or inducible during coculture.

Treg re-sorting and readout assays. After 2 to 3 days (the time was not critical), activation cocultures were harvested, cells were stained for CD4, and the activated Tregs were re-sorted as CD4+GFP+ cells. For readout assays, sorted Tregs were added to V-bottom wells containing 5 × 104 FACS-sorted CD8+ effector cells (OT-I or pmel-1, as used in the original activation cultures) plus 5 × 103 CD11c+ DCs sorted from normal LNs of mice without tumors. Some wells received blocking antibodies against the PD-1 pathway, as a cocktail (each 50 μg/ml) of anti–PD-L2, clone TY25 (56); anti–PD-1, clone J43 (57); and anti–PD-L1, clone MIH7 (58) (a gift of M. Azuma). After 3 days, cocultures were analyzed by flow cytometry or were pulsed with [3H]thymidine to measure proliferation.

Ex vivo Treg assays from vaccine-draining LNs. To measure spontaneous Treg activity in vaccine-draining LNs (VDLNs), we implanted B16F10 tumors in Foxp3-GFP reporter mice. On day 11, mice received hgp100 vaccine in CpG/IFA (incomplete Freund’s adjuvant) in the contralateral footpad, with intravenous pmel-1 adoptive transfer. Three days later, the VDLNs were harvested and CD4+GFP+ Tregs were sorted and added to readout assays as described in the preceding section.

Antibodies and FACS staining

LNs were prepared by rapidly passing through a 40-μm mesh and then stained using short incubation times (10 min on ice), as previously described (22). Tumors were disaggregated by treating for 1 hour with collagenase (1 mg/ml; C5138, Sigma), deoxyribonuclease (0.1 mg/ml; D5025, Sigma), and hyaluronidase (0.1 mg/ml; H3884, Sigma) in RPMI 1640 medium.

Conjugated mAbs against the following were from BD Biosciences: CD4 (clone RM4-5), CD8α (clone 53-6.7), CD80 (clone 16.10A1), CD86 (clone GL1), CD11c (clone HL3), Ly6c (clone AL-21), PTEN (A2B1), CD25 (clone PC61), IFN-γ (clone XMG1.2), CD24 (clone M1/69), and B220 (clone RA3-6B2). Conjugated antibodies against the following were obtained from eBioscience: Foxp3 (clone FJK-16s), granzyme B (clone NGZB), PD1 (clone J43), PD-L1 (clone MIH5), PD-L2 (clone 122), CD103 (Ber-ACT8) and Ly6c (clone HK1.4), CD278 (ICOS, clone 7E.17G9), CD69 (clone H1.2F3), IL-2 (clone JES6-5H4), IL-17A (clone 17B7), CD40L (clone MR1), IL-6 (clone MP5-20F3), CD11b (clone M1/70), and CD126 (clone D7715A7). Clone 1B11 (glycosylated CD43) was from BioLegend.

Intracellular antigens were detected using a fixation-permeabilization reagent and a matching perm-wash buffer from eBioscience (catalog #00-5521), with blocking using 5% normal donkey serum, and then acquired immediately after staining. IL-2, IFN-γ, and IL-6 were measured after 4 hours of activation with PMA/ionomycin in the presence of brefeldin A as previously described (22). Unconjugated anti-FoxO3a (4 μg/ml; rabbit mAb, clone 75D8, Cell Signaling Technology) was used in the perm-wash buffer and was detected with donkey anti-rabbit PE (Jackson ImmunoResearch #711-116-152) at 1:100 dilution. All washes were performed in the perm-wash buffer in the cold. Goat anti-Eos (sc-132308) (2 μg/ml) from Santa Cruz Biotechnology was used, followed by cross-adsorbed secondary donkey anti-goat APC (705-136-147) from Jackson ImmunoResearch as previously described (22).

For phospho-specific staining, antibodies were from Cell Signaling Technology, either Alexa Fluor 647–conjugated pAkt-Thr308 (catalog #3375), pAkt-Ser473 (catalog #4075), or anti-pS6 (catalog #4851). Cells were washed in PBS, fixed with 2% paraformaldehyde for 10 min at 37°C, prechilled for 1 min, and then permeabilized by slow addition of ice-cold methanol to a final concentration of 90%. Cells were then incubated on ice for 30 min, washed with 1% fetal calf serum/PBS, blocked with the same solution for 10 min at room temperature and then for 1 hour at room temperature, and washed. Cells were acquired immediately after staining.

Mouse strains

All animal studies were approved by the Institutional Animal Care and Use Committee of Georgia Health Sciences University. FoxO3-deficient mice (FoxO3Kca strain) have been previously described (28) and were the gift of S. Hedrick and the Ludwig Institute for Cancer Research. The following were obtained from the Jackson Laboratory and bred in our colony: OT-I mice [CD8+, recognizing the SIINFEKL peptide of OVA on H2Kb (55)] and pmel-1 mice [B6.Cg-Thy1a/CyTg(TcraTcrb)8Rest/J, recognizing a peptide from human gp100 (54)]. IDO-deficient mice (B6 background) have been previously described (59, 60). PD-1–deficient mice have been previously described (26).

Tregs were identified using GFP reporter mice. To ensure that results were not influenced by a particular reporter strain, we confirmed them using multiple reporter systems. Foxp3GFP mice (61, 62), bearing a Foxp3-GFP fusion protein in the coding region of the Foxp3 locus, were the gift of A. Rudensky. Foxp3DTR mice, with a DTR-GFP fusion construct knocked in to the 3′ untranslated region of the Foxp3 gene (but with a normal Foxp3 coding sequence), were the gift of A. Rudensky (63). Mice with a BAC-transgenic GFP-Cre fusion protein under the Foxp3 promoter (Foxp3GFP-Cre) (36, 64) were obtained from the Jackson Laboratory [NOD/ShiLt-Tg(Foxp3-EGFP/cre)1Jbs/J] and backcrossed in our colony onto the B6 background. These were used for intercrosses with floxed alleles to create Treg-specific deletions, as follows.

PTENTreg-KO mice: Foxp3GFP-Cre mice, as described above, were crossed with mice bearing loxP sites flanking exon 5 of the PTEN gene (B6.129S4-Ptentm1Hwu/J, Jackson Laboratory) (35). The resulting strain was maintained as hemizygous for GFP-Cre and homozygous for PtenloxP/loxP.

FoxO3Treg-KO mice: Foxp3GFP-Cre mice were crossed with mice bearing loxP sites flanking exon 2 of the Foxo3 gene (a gift of R. DePinho and the Dana-Farber Cancer Institute) (65). The resulting strain was maintained as hemizygous for GFP-Cre and homozygous for Foxo3loxp/loxp. These mice were used as controls to validate the specificity of FoxO3a intracellular staining.

IL6RaTreg-KO mice: Foxp3GFP-Cre mice were crossed with mice bearing loxP sites flanking exons 4 to 6 of the IL-6 receptor α-chain (Il6ra) gene (B6;SJL-Il6ratm1.1Drew/J, Jackson Laboratory) (66). The resulting strain was maintained as hemizygous for GFP-Cre and homozygous for IL6raloxp/loxp.

Tumor studies

The Tg(Grm1)Epv mouse strain (48) was the gift of S. Chen (Rutgers University). The B16F10, EL4, E.G7, and LLC cell lines were from the American Type Culture Collection. B16-OVA (B16F10 transfected with full-length chicken OVA) clone MO4 (67) was the gift of A. Houghton (Memorial Sloan Kettering). Tumor implantation was performed as previously described (24), using 1 × 105 cells for B16F10 and 1 × 106 cells for other cell lines (large inocula were used to ensure rapid tumor engraftment and immune suppression). Tumor volume was calculated from orthogonal diameters using the formula V = L × W2 × π/6. Indoximod (2 mg/ml) was administered in drinking water as previously described (24), and NLG919 (6 mg/ml) was similarly administered. VO-OHpic was given at 10 mg kg−1 day−1 ip in 10% DMSO. CAL-101 was given at 30 mg kg−1 day−1 ip in 10% DMSO. For in vivo CD8 depletion studies, mice received anti-CD8 antibody clone YTS 169.4 (BioXcell) given at 200 μg ip 1 day before CTX and then at 100 μg ip every other day throughout the experiment. Indoximod (2 mg/ml; D-1MT) was administered in drinking water as previously described (24). Mice received approved euthanasia when tumors reached a size of 300 mm2 (product of orthogonal diameters); death was not used as a planned end point in any study.

Vaccines and T cell adoptive transfers

CpG-1826 (phosphorothioate oligo 5′-TCCATGACGTTCCTGAGCTT-3′) was synthesized from the published sequence (68) by TriLink Biotechnologies. Whole OVA protein was from Sigma (#A-5503). Human gp10025–33 was synthesized from the published sequence (54). Vaccines were prepared with 100 μg of OVA protein or 25 μg of peptide, with 50 μg of CpG-1826 in IFA (Sigma F-5506) and administered in the hindlimb footpad. Popliteal LNs were harvested 4 days later. For all adoptive transfers, OT-I or pmel-1 spleen cells were enriched by negative selection using magnetic beads (mouse CD8 isolation kit II, #130-095-236, Miltenyi Biotec). Staining for bead isolation was performed on ice, with short incubation times. Mice received 2 × 106 enriched CD8+ cells via the tail vein.


Multiple treatment groups were compared by ANOVA with Tukey’s honestly significant difference correction. Independent replicates for each experiment in the manuscript are indicated in the figure legends. Error bars always show SD.


Supplementary material for this article is available at


Fig. S1. Extended phenotyping of Treg population in LNs of normal mice, in TDLNs of mice with B16F10 tumors, and in the disaggregated tumors themselves.

Fig. S2. Role of IDO, PD-1, and mTOR pathways during Treg activation in vitro.

Fig. S3. In vivo phosphorylation status of S6 and Akt473 in Tregs from TDLNs of tumor-bearing mice, after vaccination in the presence or absence of IDO inhibitor drug (2 mg/ml in drinking water, or vehicle control, as indicated).

Fig. S4. Confirmation of specificity of the FoxO3a staining antibody (rabbit monoclonal, clone 75D8, Cell Signaling Technology).

Fig. S5. Foxo3a in activated Tregs in vitro.

Fig. S6. Activation step cocultures using IDO+ TDLN DCs were supplemented with excess tryptophan at the concentrations shown.

Fig. S7. IDO was important in controlling Akt phosphorylation, but it was unclear how IDO, which was expressed in the DCs, influenced the mTOR→Akt473 pathway expressed in Tregs.

Fig. S8. Role of PD-1 in maintaining IDO-induced suppressor activity.

Fig. S9. Validation of PTENTreg-KO mice.

Fig. S10. Growth of EL4-OVA (E.G7) and LLC tumors in PTENTreg-KO hosts.

Fig. S11. Spontaneous reprogramming of Tregs in tumors of PTENTreg-KO mice.

Fig. S12. Inflammatory intratumoral milieu in E.G7 tumors grown in PTENTreg-KO hosts.

Fig. S13. Artifactual preactivation by CD8 bead isolation obscures immunosuppression by tumors.

Fig. S14. Tregs were activated using IDO+ TDLN DCs (left) or with αCD3 mitogen and IDO blocked (right).

Fig. S15. Effect of PTEN blockade or genetic deletion on response to vaccination.

Fig. S16. Absence of autoantibodies in mice receiving VO-OHpic.

Fig. S17. Combination of VO-OHpic with indoximod.

Fig. S18. Comparison of VO-OHpic versus CAL-101 when given in combination with CTX.

Fig. S19. B16F10 tumors were grown in mice bearing an insertion of a fusion protein of GFP with the human DTR into the Foxp3 locus (63).

Fig. S20. Requirement for IL-6 receptor expression on Tregs for Treg destabilization and anti-tumor response.

Fig. S21. PTENTreg-KO mice lose any further effect of the VO-OHpic PTEN inhibitor.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank J. G. Clohessy (Harvard Medical School) for advice on dosing of VO-OHpic, S. Chen (Rutgers University) for the gift of Tg(Grm1)Epv mice, and M. Azuma for the gift of anti–PD-L1 mAb. Funding: This study was supported by NIH R01 grants CA096651, CA103320, and CA112431; the Lovick P. and Elizabeth T. Corn Foundation (to D.H.M.); R01 HL11879 and CA72669 (to B.R.B); R01 AI077610 (to J.D.P.); R01 AI083005 and AI103347 (to A.L.M.); and grant 26290059 from the Ministry of Education, Culture, Sports, Science, and Technology, Japan (to H.Y.). Author contributions: D.H.M. supervised the study and wrote the manuscript. M.D.S. and D.H.M. designed, performed, and interpreted all studies with tumors and apoptotic tumor cells. T.L.M., in conjunction with R.S., designed and interpreted the lupus model studies. R.S. performed and analyzed the lupus model studies. R.B.H. and J.D.W. discussed the design and interpretation of the PD-1 blocking studies. E.C. assisted with the design and interpretation of the CD8 bead studies. A.L.M., with L.H., discussed the design and interpretation of the PD-1 and IDO studies. M.R.M. provided reagents and discussed the interpretation of the IDO studies. A.H.S. and L.M.F. provided PD-1 and PD-L1/2 knockout mice. H.Y. provided blocking antibodies. J.D.P. participated in the design and interpretation of the mTOR studies. B.R.B. participated in the design of the PD-1, IDO, and mTOR studies and discussed results. B.R.B., A.H.S., J.D.P., M.R.M., E.C., and A.L.M. provided critical comments on, discussed, and helped revise the manuscript. Competing interests: D.H.M., A.L.M., and B.R.B. have intellectual property interests in the therapeutic use of IDO inhibitors. D.H.M. and A.L.M. receive consulting income and research support from NewLink Genetics Inc. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article