Natural killer (NK) cells and some T cells use activating receptors such as NKG2D to recognize and eliminate infected and transformed cells that up-regulate ligands for these receptors (
1). There are six to eight different NKG2D ligands, which are poorly expressed by normal cells but up-regulated in cancer cells (
2). Many tumor cells release soluble NKG2D ligands through proteolytic shedding, alternative splicing, or exosome secretion (
2,
3). Numerous reports conclude that excreted NKG2D ligands modulate NKG2D from the cell surface and desensitize antitumor effector cells (
4,
5), although a functional impact of soluble NKG2D ligands is not always observed (
6–
9). To study shed NKG2D ligands in a controlled setting, we focused on the mouse ligand MULT1, which is commonly up-regulated in primary tumors (
10) and is a transmembrane protein, as are the human ligands major histocompatibility complex (MHC) class I chain–related protein A (MICA), MICB, ULBP4, and ULBP5 (
11). Analysis of fibroblasts transduced with either N- or C-terminally tagged MULT1 revealed an N-terminal species (23 kD after deglycosylation) shed into the culture supernatant (fig. S1A) and a 24 kD membrane “stub” in the cell lysates, in addition to full-length (around 42 kD) MULT1 (fig. S1B). Inhibiting matrix metalloproteinases blocked MULT1 shedding (fig. S1C).
Hemagglutinin (HA)–MULT1–transduced fibroblasts produced nearly eightfold more shed MULT1 than did untransduced fibroblasts (fig. S2). WEHI-7.1 and C1498 but not human 293T cell lines excreted MULT1 produced endogenously. We detected serum MULT1 (mean concentration ~250 ng/ml) in most tumor-bearing
Eμ-myc transgenic mice, which frequently develop MULT1
+ tumors (
10), but not in most nontransgenic littermates (
Fig. 1A). Very high concentrations of soluble MULT1 were also detected in sera of
Apoe−/− mice fed a high-fat diet (
Fig. 1A). Given that atherosclerosis and liver inflammation in such mice are largely dependent on NKG2D function (
12), it seemed unlikely that soluble MULT1 inhibits NKG2D function. Thus, MULT1 is released from cell lines that naturally or ectopically express MULT1 and accumulates in sera of animals with spontaneous tumors and NKG2D-dependent inflammatory disease.
Purified shed HA-MULT1 bound to NKG2D with high affinity [average dissociation constant (
Kd) of 13 ± 3.8 nM] (fig. S3), similar to the affinity reported for recombinant MULT1 (
13). In parallel, we engineered fibroblasts to secrete an ectodomain fragment of HA-MULT1 (which we call “secMULT”). SecMULT1 also bound to NKG2D with high affinity (19 ± 4.3 nM) (fig. S3).
To test the function of soluble MULT1, we engineered two NKG2D ligand-negative B6 strain tumor cell lines (the B16-BL6 melanoma and RMA lymphoma) to secrete secMULT1. Surprisingly, both cell lines were rejected by syngeneic B6 mice compared with cells transduced with empty vector (
Fig. 1B and fig. S4A), despite the absence of cell surface MULT1 (fig. S4B). Tumor cells transduced with full-length MULT1 [mutated in the cytoplasmic tail to optimize cell surface expression (fig. S4B) (
14)] were also rejected (
Fig. 1B). B16-secMULT1 cells were still rejected in B6 hosts that had been depleted of CD8
+ cells but grew progressively in B6 and
Rag1−/− hosts that had been depleted of NK1.1
+ cells (
Fig. 1C and fig. S5). Hence, NK cells but not CD8
+ cells participate in the rejection of B16-secMULT1. B16 cells with inducible secMULT1 (fig. S6) were also partially rejected (
Fig. 1D). In this case, the secMULT1 lacked an epitope tag, showing that rejection occurs without a tag. A mixture of B16 (90%) and B16-secMULT1 (10%) cells was also rejected, demonstrating that sec-MULT1 acts extrinsically (
Fig. 1E). These data ruled out the possibility that rejection was solely due to intrinsic stress responses in sec-MULT1–expressing tumor cells. Instead, the data suggested that secMULT1 mobilizes or activates antitumor effector cells.
To address whether tumor cells secreting secMULT1 activate NK cells, we adapted a short-term in vivo NK induction protocol (
15,
16) by injecting irradiated tumor cells intraperitoneally in normal mice. Injection of B16-secMULT1 or B16 cells induced similar modest increases in the percentages of NK cells in the peritoneal washes 3 days later (fig. S7), but B16-secMULT1 cells induced more potent ex vivo killing activity against NK-sensitive YAC-1 tumor cells (
Fig. 2A). Similar results were obtained with RMA-secMULT1 cells (fig. S8A). Furthermore, higher percentages of peritoneal NK cells from mice injected with B16-secMULT1 cells produced interferon-γ (IFN-γ) after stimulation ex vivo with YAC-1 tumor cells (
Fig. 2B) or immobilized antibodies against NK-activating receptors (
Fig. 2C and fig. S8B). To allow the recovery of intratumoral NK cells at early times after subcutaneous transfer, we implanted 3 × 10
5 to 5 × 10
5 tumor cells mixed with matrigel. Seven days later, NK cells extracted from B16-secMULT1 tumors exhibited stronger IFN-γ responses after stimulation ex vivo (
Fig. 2D). Therefore, soluble MULT1 stimulated NK cell functional capacities in both subcutaneous and peritoneal tumors.
Recombinant MULT1 (rMULT1) is similar in size to shed MULT1. When injected with B16 tumor cells (fig. S9A), rMULT1 resulted in partial tumor rejection (
Fig. 2E and fig. S9B), and NK cells extracted from the tumors exhibited increased functional activity after stimulation ex vivo (
Fig. 2F). The rMULT1 sample was devoid of endotoxin or other PAMPs that activate macrophages (fig. S9C). These data established that soluble MULT1 causes tumor rejection, likely by activating NK cells.
secMULT1 and shed MULT1 are monomeric (fig. S10, A to C) and should not cross-link NKG2D, which is typically necessary for immune receptor activation. Indeed, monomeric rMULT1 failed to stimulate IFN-γ production when incubated with peritoneal NK cells for 4 hours (fig. S11A). Soluble MULT1 may form a multivalent array in vivo, but preliminary staining analyses failed to detect such arrays. These data argue that soluble MULT1 stimulates NK cells by other mechanisms.
Target cells bearing membrane NKG2D ligands, including MULT1, caused down-regulation of cell surface NKG2D, presumably by aggregating the receptor and triggering receptor endocytosis (
Fig. 3A) (
8,
17). Tumor cells secreting secMULT1, in contrast, caused NKG2D up-regulation on NK cells in vivo (
Fig. 3, A and B), without affecting an irrelevant receptor, DNAM-1 (
Fig. 3B). NKG2D up-regulation occurred without increases in NKG2D mRNA (fig. S12) or intracellular protein. These findings suggested the following hypothesis (
Fig. 3C): that untransformed host cells express membrane NKG2D ligands that persistently engage NK cells, cause NKG2D down-regulation, and globally desensitize the NK cells as the tumors progress; and that soluble MULT1 enhances responsiveness and cell-surface NKG2D expression by blocking these interactions.
Consistent with the hypothesis, host CD11b
+F4/80
+ myeloid cells associated with either peritoneal or subcutaneous B16 or B16-secMULT1 tumors displayed the NKG2D ligand RAE-1 (but not MULT1) on the cell surface (
Fig. 3, D and E). Monocytes in patients with several types of cancer also expressed NKG2D ligands (
18). Tumors may cause an increase in RAE-1 expression by myeloid cells in the peritoneum (
Fig. 3D) and possibly other sites (
19). Hence, myeloid cells, and possibly other nontumor cells, express NKG2D ligands in vivo. We further demonstrated that rMULT1 competitively blocks binding of RAE-1ε-Fc fusion protein to NKG2D on NK cells (
Fig. 3F), confirming a distinct prediction of the hypothesis.
To test whether RAE-1 expressed on endogenous cells caused NK cell inactivation, we used the clustered regularly interspaced short palindromic repeat (CRISPR)/Cas9 method to disrupt both B6 strain RAE-1 genes,
Raet1d and
Raet1e. Peritoneal NK cells from
Raet1d−/−Raet1e−/− mice exhibited significant increases in membrane NKG2D expression as well as functional responses (
Fig. 4, A to C). Greater differences occurred when the mice were injected with irradiated B16 tumor cells, suggesting that greater NK desensitization and NKG2D down-regulation occurs in the presence of tumors. The increase did not fully account for the effects of secMULT1 (
Fig. 4, A and B), suggesting that secMULT1 may also operate by other mechanisms. NK cells in
Raet1-deficient mice exhibited greater responses to both NKG2D-dependent stimuli (for example, to YAC-1 cells) (
Fig. 4B) and NKG2D-independent stimuli (for example, to antibody to NKp46) (
Fig. 4C), which is consistent with published reports that persistent stimulation by cells expressing NKG2D ligands results in a global desensitization of NK cells (
17,
20). In mice injected with B16-secMULT1 tumor cells, smaller differences were observed, as predicted if secMULT1 blocks NKG2D interactions with RAE-1-expressing cells. Indeed, addition of rMULT1 to cultures of peritoneal wash cells, which contain NK cells and RAE-1
+ myeloid cells, resulted in increased responses of NK cells to stimulation, when tested 8 to 20 hours later (fig. S11, B and C). These data indicated that interactions of NK cells with RAE-1 molecules on nontumor cells cause NKG2D down-regulation and functional desensitization, and that this may be accentuated in tumor-bearing mice.
Our model further predicts that NKG2D receptor deficiency, or blocking with antibody, should have a similar effect as soluble MULT1. NK cells from NKG2D-deficient (
Klrk1−/−) mice exhibited a modest increase in functional activity (fig. S13A), confirming recent findings (
21,
22). A larger effect was evident in NKG2D-deficient NK cells on a
Rag2−/− background (
Fig. 4D). Similarly, intraperitoneal injections of B6 mice with F(ab’)
2 fragments of blocking NKG2D antibody resulted in enhanced functional responses ex vivo (
Fig. 4E). Most remarkably, NKG2D-deficient
Rag2−/− mice exhibited a strongly enhanced rejection response against B16 and B16-secMULT1 tumors, compared with the responses of
Rag2−/− mice (
Fig. 4, F and G). Furthermore, incorporation of F(ab’)
2 fragments of blocking NKG2D antibody into B16 tumors that were established in subcutaneous matrigel plugs resulted in partial tumor rejection and augmented responsiveness of NK cells within the residual tumors (fig. S13, B and C). Thus, NKG2D deficiency, or blockade, results in enhanced NK cell responsiveness and tumor rejection. These data strongly support the proposed model (
Fig. 3C).
The finding that NK cells persistently stimulated through NKG2D or other receptors are broadly desensitized is consistent with published data (
20,
23) and may reflect a defect in mitogen-activated protein kinase (MAPK)/extracellular signal–regulated kinase (ERK) signaling (
23). Blocking or disabling NKG2D restores killing of B16 cells because NK cells use receptors distinct from NKG2D to target B16 cells. A more complex outcome should pertain with tumor cells that express membrane NKG2D ligands because the NK cells, although more active, will be partly blocked in tumor cell recognition. Tumor cells often express multiple NKG2D ligands, suggesting that therapeutic efficacy may be maximized by blocking only the specific NKG2D ligands expressed by host cells, rather than by blocking NKG2D altogether.
Our results are surprising because they show that soluble NKG2D ligands in vivo stimulate tumor rejection and increase membrane NKG2D, whereas the literature suggests that they should suppress tumor rejection and decrease membrane NKG2D. However, NKG2D down-regulation is frequently not observed in patients with soluble MICA/MICB (
6–
9). Moreover, MULT1 and MICA/MICB ligands differ in a key respect: affinity. Soluble MULT1 is a high-affinity (
Kd ~ 10 nM) monomeric ligand. Soluble MICA and MICB are low-affinity ligands (
Kd ~ 1 μM), which are present in patient sera at concentrations below 1 nM (
4,
24), meaning that NKG2D occupancy is predicted to be extremely low. This consideration suggests that systemic effects of soluble MICA and MICB may be indirect, or that the active form of soluble MICA or MICB is actually a multimeric exosome form (
25,
26), which can bind and cross-link the receptors despite a low affinity and low concentration. Binding and cross-linking are conditions known to cause modulation of other immune receptors from the cell surface (
27). B16 cells secreting the low-affinity MICA ligand, when injected intraperitoneally, failed to induce significant NKG2D up-regulation or increased NK functional activity (fig. S14), which is in line with our expectations. The results identify an unexpected mechanism of immune activation and support efforts to evaluate the potential of soluble NKG2D ligands or antibodies that block NKG2D or its ligands for immunotherapy of cancer. Studies suggest that engagement of other NK-activating receptors, such as NKp46, may also lead to NK cell desensitization, suggesting that multiple targets exist for amplifying NK function (
28).
Acknowledgments
We thank H. Nolla and A. Valeros for help with cell sorting; C. Kang for targeting eggs for creating the Raet1 knockout mice; Q. Yan for assistance with Biacore assays; and T. Trevino, L. Bai, S. Li for technical assistance. We thank Raulet Lab members and R. Vance for useful discussion and comments on the manuscript. The data presented in this manuscript are tabulated in the main paper and in the supplementary materials. D.H.R. is an inventor on a patent (U.S. 6821522) filed by University of California, Berkeley describing the use of soluble ligands for immunotherapy of cancer. W.D. was supported by a Cancer Research Institute postdoctoral fellowship. B.G.G. was supported by a National Science Foundation Graduate Research Fellowship and the Hirth Chair Graduate Fellowship of University of California, Berkeley. L.W. and A.I. were supported by Leukemia and Lymphoma Society Fellowships, and S.L. was supported by a NIH–Initiative for Maximizing Student Development grant. This work was supported by NIH grant R01 CA093678 to D.H.R.