Immune Inhibitory Molecules LAG-3 and PD-1 Synergistically Regulate T-cell Function to Promote Tumoral Immune Escape

T-cell–mediated antitumor immune responses are essential for effective deletion of primary tumor lesions and for protection against metastases (1). Although the immune system can detect and destroy malignant cells, tumors escape surveillance by a variety of cell intrinsic and extrinsic mechanisms (1–3). As with chronic viral infection (4), tumor antigen-specific CD4⁺ and CD8⁺ T cells display impaired effector function and an exhausted phenotype characterized by decreased production of proinflammatory cytokines and hyporesponsiveness to antigenic restimulation (5). This is mediated by cell extrinsic mechanisms, such as regulatory T cells (T_reg), and cell intrinsic mechanisms, such as inhibitory molecules that are upregulated on exhausted, tumor-infiltrating lymphocytes (TIL). In combination, these inhibitory mechanisms represent a formidable barrier to effective antitumor immunity (6–10).

Inhibitory receptors such as cytotoxic T-lymphocyte–associated protein 4 (CTLA-4, CD152), lymphocyte-activation gene 3 (LAG-3, CD223), and programmed cell death 1 (PD-1, CD279) function at multiple levels to ensure appropriate T-cell homeostasis, activation, and differentiation (7, 11–17). Furthermore, all 3 inhibitory molecules also contribute to cell extrinsic regulation by controlling T_reg homeostasis and function, mediating induced T_reg development, and mitigating dendritic cell differentiation and function (13–16, 18, 19). Data from genetically manipulated mice indicate that CTLA-4 represents a basic and indispensible “off switch,” whereas PD-1 and LAG-3 play more subtle roles in immune regulation. Whereas Ctla4^−/− mice develop a severe lymphoproliferative disease and are usually moribund by 3 to 4 weeks of age (20), Pdcd1^−/− (which encodes PD-1) mice live beyond 1 year while developing subtle and variable immune-based disease manifestations depending on genetic background; Pdcd1^−/− BALB/c mice develop dilated cardiomyopathy 5 to 30 weeks of age, whereas Pdcd1^−/− C57BL/6 mice develop a protracted lupus-like condition that takes over 6 months to develop (21, 22). Unmanipulated Lag3^−/− C57BL/6 mice do not develop any disease manifestations within the first year of life (23).

Recent studies have revealed that LAG-3 and PD-1 are coexpressed on tolerized TILs suggesting that they may contribute to tumor-mediated immune suppression (5, 24). Preclinical models using antibody treatment to block LAG-3 for cancer treatment show enhanced activation of antigen-specific T cells at the tumor site and disruption of tumor growth (25). Abrogation of PD1 signaling in mice leads to enhanced CTL killing, cytokine production, and tumor-bearing animal survival over several different tumor models (26). On the basis of their roles in T-cell inhibition and antitumor immune regulation, individual antibody blockade of both CTLA-4 and PD-1 have been reported to show clinical utility (27, 28). Given this information, LAG-3 and PD-1 represent a potentially beneficial pairing for dual pathway blockade in cancer therapy. However, little is known about the extent of cooperative interaction between these regulatory pathways, information critical to the development of combinatorial immunotherapy based on simultaneous blockade of multiple receptors or ligands. In this study, we investigate whether there is synergy between LAG-3 and PD-1 by analysis of tumor growth and clearance in blocking antibody-treated mice and Lag3^−/−Pdcd1^−/− mice.

C57BL/6 mice were purchased from The Jackson Laboratory. Lag3^−/− mice were provided by Y.H. Chien (Stanford University) with permission from C. Benoist and D. Mathis (Joslin Diabetes Center; refs. 23, 29). Pdcd1^−/− mice were provided by Lieping Chen (Johns Hopkins University) with permission from T. Honjo (Kyoto University; ref. 30). At St. Jude Children's Research Hospital, the Lag3^−/−, Pdcd1^−/− and Lag3^−/−Pdcd1^−/− mice were backcrossed onto a C57BL/6 background an additional 5, 9, and 5 generations respectively, and a genome-wide single-nucleotide polymorphism analysis indicated that 100% of the markers were C57BL/6 for Lag3^−/− and Pdcd1^−/− mice and 90% for the Lag3^−/−Pdcd1^−/− mice. At Johns Hopkins, the Lag3^−/−Pdcd1^−/− were backcrossed 5 generations onto a B10.D2 background and crossed with Clone 4 (CL4) TCR transgenic mice. Animal experiments were carried out in specific pathogen-free facilities accredited by the American Association for the Accreditation of Laboratory Animal Care (AAALAC) at St. Jude Children's Research Hospital and Johns Hopkins Kimmel Cancer Center and approved by the respective Animal Care and Use Committees. The mice at St. Jude Children's Research Hospital are also Helicobacter- and MNV free. B16 melanoma cells were obtained from MJ Turk (Dartmouth College, Hanover, NH). This line has been authenticated by the RADIL at the University of Missouri (September 18, 2008) and maintained in continuous culture for no more than 6 months posttesting. It was also tested by IMPACT I PCR Profile at the RADIL at the University of Missouri (October 10, 2008). MC38 cells were obtained from J.P. Allison (Memorial Sloan-Kettering Cancer Center, NY), authenticated by the RADIL (March 10, 2003), and tested by IMPACT I at the RADIL at the University of Missouri (March 18, 2010). Sa1N cells were originally obtained from J.P. Allison (Memorial Sloan-Kettering Cancer Center, NY). Although these cells have not been authenticated, they perform as described in the literature in syngeneic A/J mice (14) and tested by IMPACT I at the RADIL at the University of Missouri (January 29, 2011).

Single-cell suspensions were prepared from spleens, inguinal, brachial, and axillary lymph nodes, and tumors. Cells were stained with fluorescent-labeled antibodies (BioLegend, BD-Bioscience Pharmingen, or eBiosciences) and analyzed by either FACSCalibur or LSR II flow cytometer (BD). The following clones were used: CD4 (GK1.5), CD8α (53–6.7), CD11c (N418), CD25 (PC61), CD44 (IM7), CD45R/B220 (RA3–6B2), CD69 (H1.2F3), PD-1 (RMP1–30), TCR-β (H57–597), Thy1.1 (HIS51), IFN-γ (XMG1.2), TNF-α (MP6-XT22), IL-17 (TC11–18H10), Foxp3 (150D), and LAG-3 (4–10-C9; ref. 31). For intracellular cytokine staining, cells were activated with phorbol 12-myristate 13-acetate (PMA; 100 ng/mL) plus ionomycin (500 ng/mL) for 4 hours in the presence of GolgiPlug (32), processed with a Cytofix/Cytoperm kit (32), and stained as indicated. Measurement of IFN-γ, TNF-α, and MCP-1 in serum was determined by IFN-γ- or TNF-α–specific ELISA kits (eBioscience) and a MCP-1–specific bead based kit (Millipore).

B16 melanoma and MC38 colon adenocarcinoma models were carried out as previously described with some modifications (33, 34). Briefly, on day 0 mice were injected with 1.25 × 10⁵ to 5.0 × 10⁵ B16 cells i.d. in the back or 2.0 × 10⁶ to 5.0 × 10⁶ MC38 cells subcutaneously (s.c) in the right flank. Lag3^−/−Pdcd1^−/− (and appropriate controls) were used at approximately 5 weeks of age. Tumor diameter was measured every 2 to 3 days with an electronic caliper and reported as volume using the formula m₁² × m₂ × π/6 (35). To isolate TILs, solid tumors were excised after 12 to 14 days, single-cell suspensions prepared by mechanical dissociation, followed by density gradient centrifugation on an 80%/40% Percoll (GE Healthcare) gradient. For CD4⁺ and CD8⁺ T-cell depletion experiments, anti-mouse CD4 (GK1.5) and anti-mouse CD8 (2.43) ascites were administered i.p. on days 1, 2, 5, 8, and 11 (pretitered for maximal deletion).

Sa1N fibrosarcoma cells or MC38 cells (2 × 10⁶) were implanted s.c. into A/J mice (Harlan) or C57Bl/6 mice (Charles River), respectively. Tumor volumes were measured with an electronic caliper (l × w × h/2) and randomized by size (10 mice per group). Mice with palpable tumors (Sa1N ∼60 mm³/2; MC38 ∼40 mm³/2) were injected i.p. at a dosage of 10 mg/kg for chimeric mouse anti-PD-1 (4H2, IgG1; ref. 36) and/or rat anti-mouse LAG-3 (C9B7W, IgG1; ref. 37). Control murine IgG1 (MOPC 21; BioXCell) was dosed at 20 mg/kg or added to individual anti–PD-1 or anti–LAG-3 antibody treatments at 10 mg/kg. Tumor growth inhibition (TGI) was calculated when all mice within a group were available for tumor measurement.

Splenocytes and lymph node cells from female mice (5–7 weeks old) were pooled, and 10⁷ cells injected i.v. into age-matched female Rag-1^−/− (5–6 weeks old) mice. CD4⁺ or CD8⁺ cells were depleted from splenocytes and lymph node cells with biotinylated anti-CD4 or anti-CD8 by macrophage separation using streptavidin-coupled beads (Milteny Biotec) to achieve purity above 95%.

Full necropsies were completed independently at St Jude Children's Research Hospital and Johns Hopkins. Tissues from knockout and control mice were fixed in 10% neutral buffered formalin, except eyes, which were fixed for 24 hours in Davidson's fixative, embedded in paraffin, sectioned at 4 μm and stained with hematoxylin and eosin for histopathologic examination. Collagen deposition was detected by a Masson's Trichrome stain. Macrophages and T cells were detected using rat anti-mouse MAC2 (CL8942AP; Accurate Chemical) and goat anti-human CD3 (sc-1127; Santa Cruz Biotechnology) antibodies, respectively. Heat-induced epitope retrieval was carried out by heating slides in ER2 (AR9640; Leica) for 30 minutes and the Refine system (DS9800; Leica Microsystems) used for detection. T_reg cells were detected with a rat anti-mouse FoxP3 antibody (14–5773-82; eBioscience) with heat-induced retrieval, pH 6.0, target retrieval buffer (S699; Dako) followed by a horseradish peroxidase–labeled streptavidin detection system (TS-125-HR; Thermo Shandon). In all immunohistochemical assays, 3,3′-diaminobenzidine was used as the chromogenic substrate with a light hematoxylin counterstain.

Mouse sera were analyzed by indirect ELISA, alongside a positive control (serum from a MRL/lpr mouse, a SLE disease model), using 96-well Nunc MaxiSorp plates (Nalgene Nunc). Multiple antigen blot assay (MABA) was conducted as described (38).

CL4 adoptive transfer and in vivo CTL studies were carried out as previously described (25, 39).

Summary statistics are presented as mean ± SEM. Group means were compared with 2-sample t tests. Event-free survival (moribund) estimates were calculated with the Kaplan–Meier method; mouse groups were compared by log-rank test. The proportions of tumor-free mice were evaluated with the binomial distribution; synergy hypotheses were tested based on the maximum likelihood method. Trends in weight over time and tumor growth over time among different mice groups were analyzed using mixed models. All P values are 2-sided, and statistical significance was assessed at the 0.05 level. Analysis was conducted by SAS (version 9.2).

PD-1 monoclonal antibody treatment has shown clinical efficacy against multiple malignancies including melanoma, prostate, renal cell, and lung cancer (27). LAG-3 has been suggested to directly modulate the activity of PD-1⁺ cells (5); furthermore, coexpression of LAG-3 and PD-1 has been shown in malignant mouse and human tissue (5, 24). Given these data, we hypothesized that LAG-3 and PD-1 act synergistically to control immune homeostasis and mediate tumor-induced tolerance. Consistent with previous reports, a significant percentage of CD4⁺ and CD8⁺ TILs from transplanted B16 melanoma, MC38 colorectal adenocarcinoma, and Sa1N fibrosarcoma expressed high levels of LAG-3 and PD-1 (32, 34), whereas similar upregulation was not observed on peripheral T-cell populations (Fig. 1). Next, we asked if antibody-mediated dual blockade of these pathways would reduce tumor growth by assessing the potential efficacy of combined anti–LAG-3 and anti–PD-1 blockade in mice with established tumors. Reduced growth of Sa1N fibrosarcoma and MC38 colorectal adenocarcinoma (32, 40–42) was observed in some but not all mice treated with the anti–LAG-3 or anti–PD-1 monotherapy (Fig. 2); only a few mice were tumor free after 50 days (0%–40%). For anti–LAG-3, this is the first demonstration of TGI with anti–LAG-3 as a monotherapy. In striking contrast, 70% and 80% of the Sa1N- and MC38-inoculated mice, respectively, were tumor free after 50 days following combinatorial anti–LAG-3/anti–PD-1 immunotherapy (Fig. 2). However, this regimen had no effect against established B16 tumors. Using the maximum likelihood method, there seemed to be a synergistic benefit of anti–LAG-3/anti–PD-1 combinatorial immunotherapy that is superior to either the additive effect of anti–LAG-3 and anti–PD-1 or monotherapy. Dual treatment with anti–LAG-3/anti–PD-1 did not result in immunopathologic manifestations such as lymphocytic infiltration in the Sa1N fibrosarcoma model as determined by detailed histologic analysis of multiple tissues. Despite efficient tumor clearance, no evidence of systemic or organ-specific autoimmunity was observed.

To investigate the mechanism underlying decreased tumor growth in antibody-treated mice, MC38 tumor-bearing mice were treated with the antibody combinations used above and draining lymph node (DLN) T cells, non-DLN (NDLN) T cells, and TILs analyzed by flow cytometry for phenotype and effector function. As expected, average tumor size of anti-PD1–treated or dual antibody–treated mice was significantly smaller than isotype control or anti-LAG3–treated mice (Supplementary Fig. S1). A significantly higher percentage of IFN-γ⁺CD8⁺ T cells were found in the tumor-associated DLNs of dual antibody–treated mice compared with the monotherapy groups, or cells analyzed from NDLNs (Fig. 3A). Likewise, a higher percentage of IFN-γ⁺CD4⁺ and IFN-γ⁺CD8⁺ TILs, and to a lesser extent TNF-α⁺ CD4⁺ and CD8⁺ TILs, were observed in anti–LAG-3/anti–PD-1–treated mice than in control groups (Fig. 3B). Taken together, these data suggest that anti–LAG-3/anti–PD-1 combinatorial immunotherapy may act synergistically to reduce tumor growth by increasing the proportion of effector T cells in the tumor and DLNs.

To further investigate the synergy between these 2 inhibitory molecules, we next assessed whether LAG-3 and PD-1 cooperate to control immune homeostasis and mediate tumor-induced tolerance with a genetic approach. Lag3^−/−Pdcd1^−/− C57BL/6 or B10.D2 mice were generated at 2 independent locations (see Materials and Methods), and disease manifestation and immune pathology analyzed over time. Lag3^−/−Pdcd1^−/− mice developed an early onset (∼4 weeks of age), lethal autoimmune condition that resulted in approximately 80% of the mice moribund by approximately 10 weeks (Fig. 4A and Supplementary Fig. S2). The major histopathologic manifestations included diffuse fibrosing lymphohistiocytic endocarditis, myocarditis, and pancreatitis (Fig. 4B, Supplementary Table S1 and Fig. S3). Extensive infiltration by CD3⁺ T cells, Foxp3⁺ T_reg cells, and Mac2⁺-F4/80⁺ macrophages was observed, in conjunction with substantial collagen deposition but limited B-cell and neutrophil infiltration; however, negligible autoantibody reactivity was seen in serum from Lag3^−/−Pdcd1^−/− mice but not single knockout or wild-type mice (Supplementary Fig. S4). Lag3^−/− and Pdcd1^−/− single KO mice lacked any disease manifestations or histopathology over this period of observation. These results show that the PD-1 and LAG-3 pathways synergistically regulate self-reactivity.

Consistent with the histopathology observed, substantially increased numbers of CD4⁺ and CD8⁺ T cells were observed in the regional LNs, but not the spleens, of Lag3^−/−Pdcd1^−/− mice (Fig. 4C, Supplementary Fig. S5). These cells possessed a predominantly activated/memory phenotype as indicated by CD69/CD44 staining. Nevertheless, there seemed to be minimal difference in the extent of division in vivo based on ex vivo Ki67 staining, even though Lag3^−/−Pdcd1^−/− T cells proliferate more in vitro following anti-CD3 stimulation (data not shown). The number of CD4⁺Foxp3⁺ T_reg cells, B cells, and CD11c⁺ dendritic cells were also increased in Lag3^−/−Pdcd1^−/− mice (Fig. 4C, Supplementary Fig. S5). Given that Lag3^−/− T_reg cells exhibit reduced suppressive activity (13, 14) and PD-L1 (PD-1 ligand) contributes to iT_reg development (15), it is possible that the combined loss of LAG-3 and PD-1 alters T_reg cell homeostasis.

To further probe the cellular defects in Lag3^−/−Pdcd1^−/− mice, we adoptively transferred splenocytes into lymphopenic Rag-1^−/− mice. In contrast to healthy wild-type and single knockout controls, Lag3^−/−Pdcd1^−/− splenocyte recipients started to lose body weight approximately 6 days posttransfer with 100% morbidity by day 20 (Supplementary Fig. S6A and S6B). Adoptive transfer experiments T-cell–depleted Lag3^−/−Pdcd1^−/− splenocytes clearly showed that both CD4⁺ or CD8⁺ T-cell populations contributed to the disease observed, with a dominant role for the former (Supplementary Fig. S6C and S6D). Consistent with these survival and weight loss data, histologic analysis of CD4⁺ T-cell–depleted Lag3^−/−Pdcd1^−/− splenocyte recipients revealed relatively normal bone marrow cellularity and density, whereas Lag3^−/−Pdcd1^−/− splenocyte recipients exhibited a near total absence of hematopoietic cell precursors in bone marrow and severe lymphoid depletion in the spleen, LNs, and Peyer's patches (Supplementary Table S2, Fig. S6E and S7). These data indicate that CD4⁺ T cells are primarily responsible for the pathology observed. Cytokine analysis revealed high levels of IFN-γ, TNF-α, and MCP-1 in the serum of Lag3^−/−Pdcd1^−/− recipients but not single knockout or wild-type control recipients (Supplementary Fig. S6F–S6H). Taken together, these data suggest that Lag3^−/−Pdcd1^−/− splenocyte recipients, in contrast with their single knockout and wild-type controls, develop an autoimmune GvHD-like syndrome with evidence of aplastic anemia and bone marrow failure as a cause of death.

The data thus far suggested that while a reasonable level of tolerance is maintained in single knockout Lag3^−/− or Pdcd1^−/− mice, dual loss of LAG-3 and PD-1 expression results in a loss of peripheral self-tolerance of CD4⁺ and CD8⁺ T cells. To test this in an antigen-specific system, we asked whether hemagglutinin-specific tolerance induced in transgenic mice expressing hemagglutinin as a self-antigen in multiple epithelial tissues (C3-HA^lo mice; refs. 13, 43), could also be broken if adoptively transferred hemagglutinin-specific T cells [from Clone 4 (CL4) TCR transgenics] lacked both inhibitory molecules. Compared with wild-type CL4 T cells, significant expansion of Lag3^−/−Pdcd1^−/− clonotypic CD8⁺ T cells was observed 5 days posttransfer (Supplementary Fig. S6I). Although this was not substantially greater than that seen with Pdcd1^−/− T cells, the Lag3^−/−Pdcd1^−/− T cells exhibited a significantly enhanced effector phenotype, as determined by intracellular expression of IFN-γ and IL-17, compared with their single knockout and wild-type controls (Supplementary Fig. S6J and S6K). Similarly, when tolerance was broken in C3-HA^lo transgenic mice (39), adoptive transfer of antigen-specific Lag3^−/−Pdcd1^−/− CD8⁺ T cells expanded significantly more than their single knockout and wild-type controls, although enhanced in vivo CTL activity was comparable in the Pdcd1^−/− and Lag3^−/−Pdcd1^−/− CD8⁺ T cell recipients (Supplementary Fig. S8). Collectively, these data suggest that the loss of LAG-3 and PD-1 also results in loss of tolerance to a model self-antigen.

To continue our analysis of Lag-3/PD-1 synergy in the regulation of antitumor immunity, we assessed tumor growth in Lag3^−/−Pdcd1^−/− mice and controls over time. Of the 3 transplantable tumor models examined in this study, B16 is regarded as the least immunogenic and thus the hardest to eliminate by immunologic intervention (32, 34). A low dose of B16 cells (1.25 × 10⁵) progressively grew in wild-type and Lag3^−/− mice inoculated intradermally at day 0, whereas limited growth was observed in Pdcd1^−/− and Lag3^−/−Pdcd1^−/− mice (Supplementary Fig. S9A). Although previous studies suggested that PD-1 deletion did not affect subcutaneously injected tumor growth (44), our experiments revealed reduced tumor growth in Pdcd1^−/− mice compared with wild-type mice. Paradoxically, Lag3^−/− mice developed slightly larger tumors. Whether this is due to reported defects in natural killer cell cytolysis (23), or an unexpected role of pDCs, which highly express LAG-3 (45), remains to be determined. Statistical analysis with the maximum likelihood method for synergy found that the lack of tumor growth in the Lag3^−/−Pdcd1^−/− mice was greater than the additive effects of tumor growth in Lag3^−/− mice and Pdcd1^−/− mice at day 11 (P < 0.05) and day 13 (P < 0.0005) suggesting that LAG-3 and PD-1 synergize to mediate tumor-induced tolerance. Depletion of CD4⁺ and CD8⁺ T cells restored normal B16 tumor growth in compound-deficient mice, indicating the necessity of adaptive immunity to the antitumor response (Supplementary Fig. S9B).

As the difference in resistance to B16 growth between Pdcd1^−/− and Lag3^−/−Pdcd1^−/− mice seemed small, we evaluated tumor B16 and MC38 growth at different doses. At the higher B16 dose (5 × 10⁵ cells per mouse), wild-type and Lag3^−/− mice show uncontrolled tumor growth and lethality with an average survival time of less than 20 days (Fig. 5A and Supplementary Fig. S10). Lag3^−/−Pdcd1^−/− mice (80%) eliminated tumors compared with only 40% of Pdcd1^−/− mice; however, B16 survivors did not display autoimmune vitiligo as is often seen with this model (46). We also investigated growth of subcutaneously implanted MC38 adenocarcinoma cells at 2 different doses. Whereas MC38 growth and survival were comparable in Lag3^−/−Pdcd1^−/− and Pdcd1^−/− mice when a low dose (2 × 10⁵ cells per mouse) was used (70% vs. 71%, respectively; Fig. 5B and Supplementary Fig. S10), Lag3^−/−Pdcd1^−/− mice were clearly more effective at preventing high-dose (5 × 10⁵ cells per mouse) MC38 tumor growth and ensuring survival (83% vs. 38%, respectively; Fig. 5C and Supplementary Fig. S10). Phenotypic analysis revealed enhanced IFN-γ expression by Lag3^−/−Pdcd1^−/− CD4⁺ and CD8⁺ T cells in tumor DLNs compared with wild-type and Lag3^−/− mice, and NDLNs from all groups (Fig. 6). These data suggest that combined loss of LAG-3 and PD-1 limits tumor-mediated tolerization and enhances tumor-specific immunity and resistance to tumor growth.

To further investigate the killing efficacy of Lag3^−/−Pdcd1^−/− T cells in vivo in the presence of an established tumor, clonotypic CL4 CD8⁺ T cells were transferred into ProTRAMP male mice, which develop prostate cancer driven by the probasin promoter and express hemagglutinin. After vaccination with hemagglutinin-expressing Vaccinia virus, Lag3^−/−Pdcd1^−/− recipients showed significantly increased killing ability in comparison with the wild type, and slightly increased killing efficiency in comparison with Pdcd1^−/− single knockouts (Supplementary Fig. S11). These data support the conclusion that Lag3^−/−Pdcd1^−/− CD8⁺ T cells are less susceptible to tumor-induced tolerance than wild-type cells. Taken together, these data clearly show that Lag3^−/−Pdcd1^−/− mice are more capable of resisting high-dose tumor growth than Pdcd1^−/− and wild-type mice.

The data presented here illustrate clear synergy between the inhibitory receptors LAG-3 and PD-1 in controlling immune homeostasis, preventing autoimmunity, and enforcing tumor-induced tolerance. First, we show coexpression of LAG-3 and PD-1 on tumor-infiltrating lymphocytes. Second, we show that dual blockade of these receptors leads to decreased tumor growth and enhanced antitumor immunity. Importantly, dual antibody–treated mice show more robust immune responses than either single-treated group. Third, analysis of mutant mice revealed a cooperative requirement for LAG-3 and PD-1 in maintaining immune homeostasis. Consistent with our observations following antibody-mediated blockade of LAG-3 and PD-1, Lag3^−/−Pdcd1^−/− mice prevented growth of high-dose B16 and MC38 tumors and ensured survival while single knockout controls and wild-type mice succumbed to disease. Taken together, these data reveal an unappreciated synergistic cooperation between LAG-3 and PD-1 in limiting tumor growth.

Although anti–LAG-3/anti–PD-1 combinatorial immunotherapy effectively cleared established Sa1N and MC38 tumors, this therapy was not effective against established B16 tumors. In contrast, B16 tumors were more difficult to establish in Lag3^−/−Pdcd1^−/− mice. B16 is a more difficult tumor to eradicate than MC38 and Sa1N and thus there could be several possible explanations for this apparent discrepancy. First, expression of LAG-3/PD-1 on TILs from B16 is lower than for MC38 and Sa1N, and this may be below a required threshold for efficacy upon antibody blockade in tumor-bearing mice. Second, as the knockout mice would lack LAG-3/PD-1 at the initiation of tumor inoculation, the immune system would not have this impediment at the commencement of an antitumor response. In contrast, the initial stages of LAG-3/PD-1 upregulation may already have occurred in tumor-bearing mice at the time of antibody treatment thereby establishing sufficient tolerance to prevent effective antitumor immunity. Third, there may be additional regulatory mechanisms that contribute a greater role with B16 that may be sufficient to prevent antitumor immunity when LAG-3/PD-1 blockade is initiated in mice that already have tumors. Lastly, we cannot rule out the possibility that the genetic deletion of LAG-3/PD-1 results in T_regs that have a cell intrinsic defect in their regulatory capacity which impacts early tumor establishment. Of course, a combination of these issues could also contribute to the difference in B16 clearance seen in LAG-3/PD-1 deficiency versus late stage antibody-mediated blockade. It is possible that alternate dosing regimens and therapeutic combinations could result in effective clearance of established B16 with anti–LAG-3/anti–PD-1 combinatorial immunotherapy.

It is noteworthy that the percentage of IFN-γ⁺ TILs is higher in Lag3^−/−Pdcd1^−/− mice versus anti–LAG-3/anti–PD-1–treated mice (Fig. 6 vs. Fig. 3). These differences could be reflective of the differences in B16 elimination in LAG-3/PD-1–deficient versus mAb-treated mice. Alternatively, as discussed above, these observations could be due to the different tumors analyzed or temporal differences between these experiments, as IFN-γ expression was determined 1 week after mAb treatment compared with 2 weeks after tumor inoculation into Lag3^−/−Pdcd1^−/− mice. Finally, we cannot rule out the possibility that there is a phenotypic difference in the immune cells in the Lag3^−/−Pdcd1^−/− mice. However, there does not seem to be an active systemic defect at the time of the experiment as high IFN-γ expression is not observed in the NDLN.

Although Lag3^−/−Pdcd1^−/− mice develop a lethal autoimmune condition, the disease is slower (∼10 weeks vs. 3–4 weeks) and less penetrant (80% vs. 100%) than the phenotype observed in Ctla4^−/− mice (20). Recently, analogous observations were reported by Honjo and colleagues in which BALB/c mice harboring a loss-of-function mutation in Lag3 combined with genetic deletion of Pdcd1 develop lethal myocarditis (47). Heart-infiltrating T cells from these compound-deficient mice were shown to produce high amounts of IFNγ compared with distal lymphoid organs such as the spleen. Our results are consistent with their data, as we also observed enhanced production of proinflammatory cytokines by T cells infiltrating sites of inflammation, such as the heart and pancreas in Lag3^−/−Pdcd1^−/− mice and in tumors and DLNs in knockout and antibody-treated mice. Furthermore, the Honjo group also observed accelerated autoimmune diabetes in NOD mice expressing a loss-of-function LAG-3 mutant, consistent with our recent observations in Lag3^−/− NOD mice (48). However, the mice used by Honjo and colleagues that lacked functional LAG-3/PD-1 expression were on a BALB/c background, whereas our data were derived from mice on a C57BL/6 or B10.D2 background. Strain-specific differences between these mice have been well documented (49, 50) and may have contributed to subtle differences in phenotypic and mechanistic observations reported. For instance, loss of LAG-3/PD-1 on a B10.D2 background can lead to an increase in IL-17⁺ cells which was not seen in mice on a Balb/c background.

Although CTLA-4, PD-1, and LAG-3 are all negative regulators expressed during T-cell activation, high level, dual LAG-3/PD-1 expression is largely restricted to infiltrating TILs. Thus, LAG-3/PD-1 combinatorial immunotherapy may promote tumor-specific responses relative to nonspecific or self-antigen–specific immune responses and thus may be less toxic than CTLA-4 blockade. Given the recent phase 3 results with anti–CTLA-4 treatment of patients with metastatic melanoma, showing a clear survival benefit (albeit with notable immune toxicity; ref. 28), our results suggest that combined blockade of PD-1 and LAG-3 is a highly promising combinatorial strategy for the immune-based therapy of cancer.

The authors declare competing financial interests. D.M. Pardoll, D.A.A.Vignali, C.G. Drake, and C.J. Workman have submitted patents that are pending and are entitled to a share in net income generated from licensing of these patent rights for commercial development. A.J. Korman, M. Selby, and J.F. Grosso are employees of Bristol-Myers Squibb. C.G. Drake has an ownership interest in Amplimmune and has served as a consultant to Dendreon, Bristol-Myers Squibb, and Pfizer.

The authors thank Many Jo Turk and Jim Allison for cell lines. At St. Jude, the authors thank Karen Forbes, Amy Krause, and Ashley Castellaw for maintenance and breeding of mouse colonies, Cliff Guy for help with cytokine analysis, Paul Thomas for anti-CD4 and anti-CD8 depleting Abs, Richard Cross, Greig Lennon and Stephanie Morgan for FACS, Song Wu and Hui Zhang for help with statistical analysis, the staff of the Shared Animal Resource Center at St Jude for the animal husbandry, the Veterinary Pathology Core Laboratory at St. Jude for histology and immunohistochemistry support, and the Hartwell Center for Biotechnology and Bioinformatics at St Jude for real-time PCR primer/probe synthesis and MOG synthesis and purification. At Johns Hopkins, the authors thank Dih-Dih Huang for the maintenance and breeding of the mouse colonies, and technical support. At Bristol-Myers-Squibb, the authors thank David Klitzing and the staff of the Milpitas animal facility for carrying out the tumor experiments and Rangan Vanganipuram, Brian Lee, and Shilpa Mankikar for provision of antibodies.

This work was supported by the NIH (R01 AI39480 to D.A.A. Vignali and R01 CA127153 and P50 CA58236–15 to C.G. Drake), a Hartwell Postdoctoral Fellowship (to M.L. Bettini), NCI Comprehensive Cancer Center Support CORE grant (CA21765, to D.A.A. Vignali), the American Lebanese Syrian Associated Charities (ALSAC, to D.A.A. Vignali), the Patrick C. Walsh Fund (to C.G. Drake), the Koch Fund (to C.G. Drake), NHLBI contract - HHSN-268201999934C and CIHR - 20R92141 to P.J. Utz, and NIAID F32 AI080086 to C.L. Liu. C.G. Drake is a Damon Runyon-Lilly Clinical Investigator.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.