Retuning of Mouse NK Cells after Interference with MHC Class I Sensing Adjusts Self-Tolerance but Preserves Anticancer Response

Natural killer (NK) cells belong to the innate immune system and are important in responses to infections, cancer, and transplants. They kill virus-infected or transformed cells and secrete cytokines that regulate other immune responses (1). NK cell activation is mediated by germline encoded receptors that bind to ligands expressed by normal and malignant cells (2, 3). Activation is balanced by inhibitory MHC class I–specific receptors: killer cell immunoglobulin-like receptors (KIR) in humans, Ly49 in mice, and NKG2A/CD94 receptors expressed in both species (2). The presence of inhibitory MHC class I–specific receptors allows NK cells to react to “missing self,” i.e., insufficient amounts of self MHC class I on target cells (4, 5).

Because cancer cells often overexpress ligands for activating receptors and concurrently downregulate MHC class I expression, missing self-recognition by NK cells has been proposed as an immunotherapeutic principle against cancer. There are two settings in which the balance between activation and inhibition has been explored in immunotherapy. First, in vivo blockade of NK cell inhibitory receptors has been attempted in mice using F(ab′)₂ fragments. This approach, analogous to checkpoint blockade in T cells (6), resulted in increased rejection of MHC class I–expressing malignant cells caused by a liberated NK cell response to leukemia cells (7–9). This strategy has now been tested as anti-KIR therapy in clinical trials: a phase I trial for acute myeloid leukemia (AML) and a phase II trial for myeloma (10–12). The second immunotherapy setting is adoptive transfer of mature NK cells across a “missing self-barrier,” i.e., to cancer patients lacking MHC class I molecules of the donor (13, 14). Importantly, in both approaches, very few incidents of autoimmunity or graft-versus-host reactivity were observed, arguing against missing self-attacks on normal host cells. Although some antitumor effects have been documented, the overall impression from these early trials is that there is much room for improvement (15, 16).

Optimal antitumor reactivity may be hampered by (re-)education of NK cells after interfering with MHC class I recognition in vivo. During NK cell maturation, self MHC class I molecules instruct NK cells to acquire appropriate missing self-recognition capacity while remaining self-tolerant, a balancing process termed NK cell “education” (17–21). Thus, only NK cells that receive the educating signal from MHC class I, via inhibitory receptors, become fully functional effector cells (17, 19, 20). Thus, MHC class I–deficient mice have normal numbers of NK cells, but they are all hyporesponsive (22, 23). Similarly, NK cells that do not express inhibitory receptors for a self-MHC class I molecule are hyporesponsive (17, 19, 20, 24). Initially, NK cell education was thought to take place in a defined period during maturation, but the notion of a static endpoint of education has been challenged. Using MHC class I mosaic mice, we found that mature NK cells changed their reactivity when the surrounding MHC class I setup was altered in vitro (25). More recently, two groups reported a rapid adaptation in self-reactivity of NK cells after transfer to hosts with a novel MHC class I environment (26, 27). These observations are in line with our proposal that NK cell education operates like a “rheostat” that continuously tunes the NK cell activation threshold (18). This model postulates that (i) NK cells are not just “on” or “off” but operate along a continuum of responsiveness levels; (ii) responsiveness of single cells depends qualitatively and quantitatively on the inhibitory MHC class I input; and (iii) responsiveness is reversible even in mature NK cells, by quenching the inhibitory input to the NK cell receptors (18).

Predictions from the rheostat model are highly relevant for NK cell therapy, both in settings of inhibitory receptor blockade and adoptive NK cell therapy over MHC class I–barriers. In both scenarios, NK cells may adapt to lower responsiveness when they are prevented from sensing the appropriate inhibitory input.

Antibody blockade of self-specific inhibitory Ly49 receptors in vivo leads to release from NK cell inhibition in the effector–target cell interaction and increased rejection of MHC class I–expressing leukemia cells. Rejection of normal syngeneic spleen cells was not increased in the same setting (9). This is surprising given the strong rejection of normal cells that occurs when Ly49/MHC class I interactions are perturbed by MHC class I removal on the target cell (22, 23). One explanation is that Ly49 receptor blockade affects the NK cell rheostat and thus retunes the activation threshold in NK cells to secure self-tolerance to normal cells (26, 27). Furthermore, when KIR ligand–mismatched NK cells are transferred between individuals in NK cell therapy, the rheostat model predicts rapid induction of tolerance due to loss of appropriate HLA input. It has not been investigated whether this leads to tolerance also toward leukemic cells.

In this study, we have explored the effects of a changed MHC class I perception on missing self-rejection and antitumor reactivity in parallel to determine the relationship between these two types of NK cell responses. We approached this question in two established experimental models in B6 mice: in vivo blockade of Ly49I/C receptors and adoptive transfer of mature NK cells to an environment with a different MHC class I phenotype.

Mice were maintained at Karolinska Institutet (Stockholm, Sweden). Experiments were performed according to governmental and institutional regulations and were approved by the local ethical committee. Animals were 6 to 12 weeks of age at the start of the experiments. A complete description of the mouse strains used can be found in the Supplementary Material. As MHC class I–deficient target cells in the in vivo cytotoxicity assay, spleens from K^b−/−D^b−/−, β₂m^−/− or TAP1^−/− mice were used and are referred to as MHC class I⁻ cells. For NK cell depletion, mice were administered 200 μg αNK1.1 (PK136, Mabtech). RMA (H-2^b) originates from the EL-4 T-cell leukemia (28) and RMA-S is a TAP2-deficient variant of RMA (5, 29). RMA and RMA-S cells were propagated as ascites lines in sublethally irradiated (4-Gy dose) B6 mice. MHC class I⁻ C1498⁻ variant cells were generated from C1498⁺ (30) and were a kind gift of Andrew Makrigiannis (University of Ottawa, Ontario, Canada).

In vivo cytotoxicity assays on i.v. injected target cells were performed as previously described (31). Briefly, spleen cells were isolated and labeled with different concentrations of CFSE (Invitrogen Molecular Probes). Cell populations (spleen or tumor) were mixed and coinjected i.v. For 5E6 treatment (F(ab′)₂ or Fab), mice were administered at 48 hours or 24 hours, respectively, pre-CFSE inoculation. Spleens were harvested, and total or relative percentages of CFSE-positive cells in each population were measured. Survival is given as a percentage of recovered CFSE-labeled cells of total spleen cells. Alternatively, in the case of transfer experiments, the relative survival of target cells is given as a ratio of MHC I⁻ versus MHC I⁺ inoculated cells. In vivo killing of C1498⁻ NKT cell lymphoma cells was performed as described by Belanger and colleagues (30), with small changes. Tumor cells were mixed with NK1.1-depleted, BMQC-labeled (10 μmol/L; Invitrogen Molecular Probes) spleen cells of impaired missing self-recognition (IMSR) mice (32) as a reference cell population, and coinjected i.p. Peritoneal lavage was performed 6 hours after injection and the survival of target cells was calculated. For calculations regarding data presented as ratios, see Supplementary Methods. For tumor growth experiments, ascites-grown RMA and RMA-S cells were inoculated s.c. into either the opposite flanks of the same mouse (dose 10²) or separately into one flank (dose RMA 3 × 10², RMA-S 10⁴). The growth of solid tumors was followed and measured by palpations three times weekly. When indicated, mice were treated with 600 μg of 5E6 F(ab′)₂ at days −2 and +1 (referred to tumor injection). In vitro degranulation assay was performed as previously described (32).

5E6 F(ab′)₂ fragments and 4D11 (Novo Nordisk) for Ly49C/I or Ly49G2 blockade respectively have previously been used and described (9). The monovalent 5E6 Fab fragments were custom-produced (Bio-Rad AbD Serotec). Mice were administered a dose of 5E6 F(ab′)₂ fragments equivalent to blocking saturation of 70% to 80% or 500 μg of Fab fragments i.p. The 5E6 hybridoma was a kind gift from Prof. Michael Bennett (deceased; previously at University of Texas Southwestern Medical Center, Dallas, TX) and the 4LO3311 hybridoma (Ly49C) from Suzanne Lemieux (retired; Institut Armand Frappier, Laval, Québec, Canada). For flow cytometry, cells were stained, acquired, and analyzed as previously described (32). A description of all antibodies used is provided in Supplementary Material.

Either whole-spleen solutions were used or NK cells were isolated by magnetic sorting with the NK cell isolation kit II (Miltenyi Biotec) according to the manufacturer's instructions. The purity of the isolate was assessed by Flow Cytometry. NK cells (1–3 × 10⁶) were injected i.v. to irradiated (8 Gy) mice.

Statistical analyses were conducted using GraphPad Prism 5. Either nonpaired or paired two-tailed Student t test (inhibitory receptor blockade), and one-way ANOVA (adoptive transfer) with Bonferroni multiple comparison test were performed for data from at least three independent experiments, unless indicated otherwise.

A more detailed description of the methods is provided in Supplementary Material.

The retuning hypothesis predicts that antibody blockade of self-specific inhibitory receptors (Ly49C/I in B6 mice) should lead to reduced NK cell responsiveness. Indeed, B6 mice treated with 5E6 F(ab′)₂ against Ly49C/I showed reduced NK cell–mediated rejection of MHC class I⁻ spleen cells. There was no effect on rejection of syngeneic spleen cells (Fig. 1A and B; ref. 9). There was also no effect when mice were treated with F(ab′)₂ against the non–self-specific inhibitory receptor Ly49G₂ (clone 4D11; Fig. 1C), supporting an argument that the interference between self MHC class I and Ly49 receptors was required for reduced rejection of MHC class I⁻ spleen cells. 5E6 F(ab′)₂ treatment did not change the size or cellularity of the spleen (Supplementary Fig. S1). NK cells thus appear to adapt to reduced inhibitory input, leading to partially impaired missing self-rejection of spleen cells. In contrast, strong rejection of the MHC class I⁻ leukemia cell RMA-S was maintained after 5E6 F(ab′)₂ treatment, while rejection of the MHC class I–expressing control cell RMA was increased (Fig. 1D and E; ref. 9). Because the two tumor cell lines show no differences with respect to expression of several NK cell–activating ligands, the tumor rejection is well explained by the missing self-hypothesis (Supplementary Fig. S2).

Next, we tested if the maintained missing-self killing of tumor cells after blockade of inhibitory receptors could be extended to other tumor models than RMA/RMA-S. For this purpose, we used the NKT lymphoma cell line C1498, from which an MHC I⁻ variant was recently generated (30). The parental C1498 cells have been shown to be more efficiently rejected upon 5E6 F(ab′)₂ treatment (7). We injected MHC I⁺, NK-depleted spleen cells as reference cells together with MHC I⁻ C1498⁻ cells in a 6-hour peritoneal assay. The tumor cells were rejected in B6 mice, while NK1.1-treated B6 or β₂m^−/− mice did not reject these cells. Importantly, rejection of these MHC I⁻ tumor cells was not affected by 5E6 F(ab′)₂ treatment of B6 mice (Fig. 1F).

To test if direct inhibition of NK cells by F(ab′)₂-induced cross-linking of inhibitory receptors could explain the reduced rejection of MHC class I⁻ spleen cells, we generated monovalent Fab fragments. These blocked the receptors in vivo with faster kinetic (highest saturation observed at 12 to 55 hours (Supplementary Fig. S3) compared with 48 to 96 hours for F(ab′)₂) (9), and induced a similarly reduced rejection of MHC I⁻ spleen cells (Fig. 1G and H). Thus, inhibitory receptor blockade resulted in reduced rejection of MHC class I⁻ spleen cells, consistent with retuning of NK cells.

We then tested whether the pattern of in vivo tumor cell killing seen after altered inhibitory input would also emerge in a long-term tumor outgrowth assay. Mice were inoculated with a threshold dose of RMA cells, titrated out to yield tumors in some but not all mice in order to detect possible changes after inhibitory receptor blockade. A reduced proportion of mice developed RMA tumors in the group treated with 5E6, while virtually all (treated and control) mice rejected the same dose of RMA-S cells. The stronger rejection capacity against RMA after inhibitory receptor blockade was also seen at the minimal dose required to yield tumors in all nontreated mice (3 × 10² cells; Table 1). In order to challenge the NK cell capacity to reject RMA-S cells maximally, we also inoculated a 100-fold higher dose (10⁴), known to lead to outgrowth of this tumor in about half of the animals. We now observed a reduced capacity to reject RMA-S cells in mice treated with 5E6 (Table 1). Overall, these data suggest that (i) the blockade of inhibitory receptors induces a missing self-response against the MHC I⁺ tumor cells that can also lead to complete rejection in a tumor outgrowth model; (ii) the discrimination between MHC I⁺ and MHC I⁻ tumor cells and the vigorous rejection of the latter are also maintained in this setting; and (iii) a weakened capacity to reject MHC I⁻ tumor cells can actually be revealed by calibrating the experimental setup to achieve maximal sensitivity.

We next asked if the reduced rejection of MHC class I⁻ spleen cells was associated with hyporesponsiveness in NK cell subsets expressing the targeted Ly49C and Ly49I receptors. We analyzed the response by Ly49C– and Ly491–single positive (sp) or Ly49C/I-double positive (dp) NK cells negative for all other inhibitory receptors by flow cytometry (Supplementary Fig. S4). The total response (IFNγ and CD107a) to NKp46 receptor stimulation by Ly49I-sp or Ly49C/I-dp NK cells from 5E6 F(ab′)₂-treated mice was significantly reduced (Fig. 2A). Interestingly, there was no significant reduction in the response performed by Ly49C-sp cells from F(ab′)₂-treated mice, indicating that 5E6 blocking was primarily affecting Ly49I-positive NK cells. The response to PMA/ionomycin stimulation was not affected, arguing against a generally lowered capacity to degranulate or produce cytokines (Fig. 2B), and Ly49A-sp NK cells, not targeted by the 5E6 Ab, were unaffected (Fig. 2C). In summary, the data indicate a selective adaptation to reduced inhibitory input in the targeted NK cell subsets, resulting in reduced responsiveness to stimulation. The reduced (but not abolished) responsiveness can be reconciled with the in vivo results, showing no missing self-response to normal cells but maintaining response to tumor cells.

There is a correlation between education and expression of killer cell lectin-like receptor G1 (KLRG1), with a higher frequency of KLRG1⁺ cells among NK cells with receptors for self MHC class I (33, 34). We therefore tested expression of KLRG1 on NK cells targeted by receptor blockade (Fig. 2D and E). Ly49I-sp and Ly49C/I-dp NK cells from F(ab′)₂-treated mice showed a significantly reduced proportion of KLRG1⁺ cells. As in the responsiveness assay, there was no difference in the Ly49C-sp and Ly49A-sp subsets. There was also no difference in the Ly49A/G₂-dp subset regarding KLRG1 expression.

Treatment did not change frequencies of NK cells expressing inhibitory Ly49 and NKG2A receptors, neither in total Ly49 receptor expression (each receptor analyzed independently of other receptors) nor in NK cells expressing one single inhibitory receptor. Furthermore, expression of 2B4, NKp46, NK1.1, and NKG2D was unaffected, as was NK cell maturation assessed by CD27 and Mac-1 (CD11b; data not shown).

We next investigated how the in vivo missing self-response toward cancer cells and normal cells is affected in an environment where the inhibitory input from self MHC class I is abolished (as opposed to targeting NK cells with particular receptors): transfer of mature NK cells from wild-type (wt) B6 mice to β₂m^−/− mice. The mice were challenged with a mixture of MHC class I⁻ and control B6 CFSE-labeled spleen cells at day 7 after transfer. B6 NK cells transferred to MHC class I–deficient recipients (but not to B6 recipients) lost their capacity to mediate selective rejection of MHC class I⁻ spleen cells, both when NK cells were transferred as total spleen cells and as enriched NK cells (Fig. 3A and C; ref. 27). The rejection was mediated by mature donor NK cells, because NK cell depletion of donors abolished the rejection (Supplementary Fig. S5A). There was no difference in activation status, as the levels of CD69 on transferred NK cells were comparable between the two groups (data not shown).

When irradiated B6 mice received syngeneic bone marrow rather than mature spleen cells, there was no missing self-rejection at day 7, indicating that this time point was insufficient for de novo development and education of NK cells (Supplementary Fig. S5B). The loss of responsiveness toward MHC class I⁻ spleen cells could also be seen when NK cells from D8 mice (K^bD^bD^d) were transferred to β₂m^−/− mice. Here, both the effect of a strong educating ligand (D^d; ref. 24) and the influence of remaining MHC class I molecules in the recipient (D8 to B6) could be tested. Transfer to β₂m^−/− mice resulted in a reduction of responsiveness, and the MHC class I⁻ target spleen cells were tolerated. Transfer to B6 mice, however, did not lead to a decrease in responsiveness toward MHC class I⁻ spleen cells (Fig. 3B).

It was of special interest to investigate how transfer of NK cells would affect responsiveness to cancer cells with the missing self-phenotype. Mice adoptively transferred with spleen cells were thus challenged with a mixture of RMA and RMA-S cells. B6 NK cells transferred to B6 mice rejected RMA-S cells to the same extent as naïve B6 mice (Fig. 3D). Importantly, B6 NK cells transferred to a β₂m^−/− environment, rejected MHC class I⁻ leukemia cells equally well, and much better than naïve β₂m^−/− mice (35). Similar to NK cells subjected to inhibitory receptor blockade, NK cells could thus still distinguish between MHC class I–sufficient and -deficient malignant cells, despite a reduced rejection of MHC class I⁻ healthy cells.

We next asked whether it was possible to increase the tumor cell rejection capacity of mature NK cells by transferring them to a host expressing MHC class I molecules absent from the original educating environment. It was previously reported that this procedure can increase their in vitro responsiveness (26, 27). Whether it also affects in vivo killing of tumor (or normal) cells is a critical question for immunotherapy that has not been investigated. We set up a model based on transfer of enriched β₂m^−/− NK cells into MHC class I–sufficient RAGcγ^−/− mice, which lack NK cells and therefore allow the survival of the transferred MHC class I–negative NK cells as well as a clear in vivo readout to measure their activity. Responsiveness of B6 NK cells was maintained after transfer to RAGcγ^−/− mice, whereas transferred β₂m^−/− NK cells did not reject MHC class I⁻ spleen cells substantially: there was only a slight, nonsignificant increase compared with rejection by the same NK cells transferred to β₂m^−/− mice (Fig. 4A). In contrast, β₂m^−/− NK cells transferred to MHC class I–sufficient hosts efficiently rejected cancer cells lacking MHC class I to a degree comparable with that of naïve B6, and significantly better than naïve β₂m^−/− mice (Fig. 4B).

We also studied a different donor–recipient combination: donor TAP1^−/− NK cells and recipient Prf^−/− mice (K^bD^b). These recipient mice lack perforin and thus accept grafts of adoptively transferred MHC class I–deficient cells (Fig. 4C). Seven days after transfer, the TAP1^−/− NK cells were still tolerant toward TAP1^−/− spleen cells, but rejected RMA-S cells equally well as naïve B6 mice (Fig. 4D). We conclude that transfer of mature MHC class I–deficient NK cells to an MHC class I–sufficient host leads to increased responsiveness toward leukemia cells lacking MHC class I. However, tolerance to MHC class I⁻ healthy cells remains, suggesting that retuning reverts tolerance to an extent allowing tumor cell recognition but not recognition on nonmalignant cells with a missing self-phenotype.

B6 and β₂m^−/− NK cells were phenotyped after transfer to MHC class I–sufficient or –deficient mice. The expression of NKG2D was reduced after transfer as such (B6 cells to B6 mice or β₂m^−/−cells to β₂m^−/− mice) but it was most affected after transfer of B6 NK cells to β₂m^−/− mice. Conversely, the expression of NKG2D was increased on β₂m^−/− NK cells after transfer to RAGcγ mice. Note that the expression of NKG2D differed between naïve B6 and naïve β₂m^−/− mice (Fig. 5A and B). The frequency of KLRG1⁺ NK cells remained stable when B6 NK cells were transferred to B6-irradiated mice or β₂m^−/− cells to β₂m^−/− mice, while both the frequency and the mean fluorescence intensity (MFI) decreased upon transfer of B6 NK cells to β₂m^−/− mice to a level similar to naïve β₂m^−/− mice and increased upon transfer of β₂m^−/− NK cells to MHC I⁺ hosts (Fig. 5C and D). There were no significant changes in expression of other activating receptors or in the frequency of NK cell subsets based on expression of inhibitory receptors (data not shown).

We have investigated how in vivo killing of malignant cells representing the missing self-phenotype is affected by previous interference with NK cell sensing of MHC class I molecules via antibody-mediated blockade of self-specific inhibitory receptors or adoptive transfer of mature NK cells to MHC class I–disparate hosts. A reduction in inhibitory input to NK cells resulted in hyporesponsiveness toward MHC class I–deficient spleen cells. In both experimental models, NK cells were, however, still capable of discriminating between MHC class I⁺ and MHC class 1⁻ malignant target cells.

The induction of hyporesponsiveness has not been studied before in the context of antibody-mediated inhibitory receptor blockade, a selective interference with NK cell/MHC class I interactions that does not involve in vitro handling of cells and irradiation. Our results thus strengthen the notion that NK cells adapt to altered MHC class I–dependent input under physiologic conditions. In vitro studies at the single-cell level indicated that the effect is due to reduced responsiveness of the targeted NK cells. However, this was observed only for Ly49I⁺ NK cells (with or without Ly49C), indicating a stronger binding of antibody fragments to Ly49I or that the functional consequences of blocking the two receptors differ. Our results suggest that the “retuning” effect should be taken into account in further development of immunotherapy based on NK checkpoint (inhibitory receptor) blockade.

Transfer of mature hyporesponsive NK cells that have not been exposed to MHC class I molecules to an MHC class I–sufficient host resulted in increased capacity to kill MHC class I–deficient leukemia cells in vivo, an essential observation. While previous studies reported an increased in vitro responsiveness of NK cells after transfer, such alterations may reflect an increased activation status without a distinct functional consequence (26, 27). In a clinical setting, this translates to transfer of hyporesponsive NK cells from a donor lacking a particular KIR ligand to a recipient where it is expressed, leading to rejection of recipient cancer cells with reduced MHC class I expression. Emergence of such MHC-loss variant tumor cells is not uncommon (36).

In both models, NK cells retained reactivity to tumor cells while becoming partially or completely tolerant to MHC class I⁻ spleen cells, even though both types of target cells represent the same “missing self” phenotype. How can these findings be reconciled with previous observations, explained in the context of a coherent model, and what are the implications for further development of NK cell–based immunotherapy strategies? We propose that retuning of NK cell responsiveness after reduction of the inhibitory input increases the overall activation threshold for the targeted NK cell, adapting it to the signals received from normal cells in the environment; this can explain the reduced rejection of normal MHC class I⁻ cells, as well as the robust tolerance observed against normal MHC class I–expressing cells after inhibitory receptor blockade (Fig. 6). Cells presenting more activation stimuli, such as cancer cells, still overcome the activation threshold. MHC class I–expressing RMA cells are killed more efficiently after inhibitory receptor blockade (9), despite a higher activation threshold, because the blockade also acts in the effector–target interaction phase. This suggests that therapy should aim for intermittent rather than continuous blockade to avoid adaptation of the NK cells and to achieve even more efficient killing of MHC class I⁺ cancer cells. Previous studies support that high expression of activating ligands, like Rae1γ, can override MHC class I–dependent inhibition (32, 37, 38). A key aspect in our study is the use of MHC class I⁻ spleen cells to monitor the activation threshold against normal cells, providing a more sensitive and quantitative assay for NK responsiveness in vivo and excluding effects of the blockade in the effector–target interaction phase.

In vivo rejection of MHC class I–deficient spleen cells was affected more strongly in the transfer model, as it was abolished completely in the MHC class I–deficient recipients. This may be due to a more profound change of the MHC input or to effects related to experimental transfer or irradiation of the recipients. Nevertheless, the mice still showed vigorous rejection of RMA-S cells. In the transfer model, a contribution to education by MHC class I on donor-derived hematopoietic cells cannot be excluded, even if the results indicate that ligand-deficient recipient cells dominate and dictate the tuning of NK cells.

In one study of human hematopoietic transplantation, NK cells developing in the recipient following transplantation were more responsive in vitro if they expressed inhibitory receptors for donor MHC class I (39). This may be interpreted in terms of education dictated by hematopoietic donor cells. However, all hematopoietic cells were presumably of donor origin in the human study, while some recipient hematopoietic cells were still present in the transfer experiments in the present study. Further, and in contrast to our study, education or tolerance was not assessed against normal cells in the human transplantation study, but with an in vitro assay based on triggering with tumor target cells.

Additional comparative studies are required to obtain a complete picture of the role of hematopoietic versus nonhematopoietic cells. Whether receptor blockade of inhibitory KIRs affects NK cell responsiveness has not been studied in humans, but in a mouse model transgenic for one KIR as well as its HLA ligand, in vivo KIR blockade did not significantly affect ex vivo IFNγ or CD107 responses to αNK1.1 or tumor cell stimulation after 7 to 21 days of continuous blockade (40). However, the possibility of retuning was not tested with the in vivo assay used in our study. Further studies are needed to elucidate whether KIR blockade leads to retuning in treated patients.

The results from studying the effects of introducing a “gain” in inhibitory input may be interpreted within the same explanatory model. In this case, it is postulated that transfer of hyporesponsive NK cells to a recipient expressing novel MHC class I⁻ molecules in the environment leads to decreased activation threshold, not enough to break tolerance and perform missing self-rejection of healthy cells, but sufficient to cause increased in vivo rejection of tumor cells.

If our observations can indeed be interpreted in the context of retuning as a part of a continuous education process, this should be testable by studying alterations in the mechanisms or molecules involved in education. However, these have not been elucidated. Education has been associated with the lectin-like receptor KLRG1 (33, 34), an adhesion molecule that acts as an inhibitory receptor on both NK and T cells and also as a marker for maturation or activation during a virus infection (41, 42). The KLRG1 expression pattern has usually been considered as a downstream consequence of education rather than a mechanistic determinator. Interestingly, in the transfer model, reduced inhibitory input correlated with decreased expression of KLRG1 and vice versa, similar to observations by Joncker and colleagues (27). In addition, we report the novel finding of KLRG1 downmodulation after inhibitory receptor blockade. Overall, our data suggest that KLRG1 expression is reversible and more tightly associated with education than previously considered.

A similar pattern was seen for NKG2D in the transfer model (Fig. 5), similar to the observations of Elliot and colleagues (26). As one possible explanation, we speculate that NKG2D modulation occurs continuously as a result of interactions between NK cells and surrounding cells (43). The latter are assumed to express low levels of NKG2D ligands, which would increase upon irradiation (38), hence the downmodulation of NKG2D upon simple transfer from B6- to B6-irradiated recipients. The lower NKG2D levels in β₂m-deficient mice (recipients as well as untreated animals) may be due to NK cells interacting differently (e.g., longer) with MHC-deficient cells because of the lack of inhibitory ligands, via NKG2D and/or other receptors.

We have assumed above that all education processes as well as the adaptation of the mature NK cells to altered inhibitory input are controlled by a common reversible mechanism, the rheostat. This is the simplest, but not the only model to explain our data. For example, adaptation of mature NK cells could be determined by a different mechanism than that responsible for the original education (44). This can be tested only when the molecular mechanisms involved in education are clarified. Regardless of the mechanisms involved, the conclusions and interpretations relating to immunotherapy still hold. Hence, one firm general conclusion is that NK cells exert antitumor effects even if they are adapted to become tolerant to normal cells lacking critical MHC class I molecules. In a clinical perspective, one may speculate whether the divergent results reported by different transplantation centers on the benefit of a KIR/KIR–ligand mismatch in bone marrow transplantation (45–47) or NK cell adoptive transfer (15) may be due to retuning effects of different strength, perhaps profound enough to also impair NK cell antitumor effects. Although the missing self–based NK cell discrimination between a small s.c. administration of MHC class I⁺ and MHC class I⁻ tumor cells was largely intact after altered inhibitory input, we did observe slightly reduced rejection capacity of MHC class I⁻ RMA-S cells when the mice were challenged with a 100-fold higher tumor cell dose. Retuning may thus also affect tumor cell killing negatively in some situations. This leaves room to improve the therapeutic effects in cell therapy as well as inhibitory receptor blockade settings. Especially during the early phase of stem cell or NK cell infusion therapy in which retuning and antileukemic effects occur in parallel, small differences in conditioning regimen, T-cell depletion, or immunosuppression may influence the activation threshold of educated donor NK cells. It would thus be important to prevent adaptation of NK cells to reduced inhibitory input, or to combine the therapy with another NK cell–activating regimen, such as cytokines or ADCC-inducing antibodies (8, 9, 48).

In conclusion, we have shown that educated NK cells adapt and reduce responsiveness when perceiving decreased Ly49/MHC class I interactions in both the inhibitory receptor blockade and the transfer model. Importantly, missing self-reactivity to tumor cells is not disturbed, while mature NK cells acquire tolerance toward normal host cells. NK cell–mediated killing of cancer (but not normal) cells of the missing self-phenotype can even be gained upon transfer of noneducated NK cells to an environment presenting novel MHC class I molecules. A possible explanation for these observations is provided by the rheostat model for NK cell education. Regardless of the mechanisms involved, the data provide evidence that missing self-recognition by NK cells can be harnessed for immunotherapy despite adaptation of NK cells to the environment, leading to tolerance to normal cells.

No potential conflicts of interest were disclosed.

Conception and design: A.K. Wagner, S.L. Wickström, P. Höglund, M.H. Johansson, K. Kärre

Development of methodology: A.K. Wagner, S.L. Wickström, S. Salam, T. Lakshmikanth, M.H. Johansson, K. Kärre

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.K. Wagner, S.L. Wickström, S. Salam, T. Lakshmikanth, E. Carbone

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.K. Wagner, S.L. Wickström, R. Tallerico, S. Salam, T. Lakshmikanth, H. Brauner, P. Höglund, E. Carbone, M.H. Johansson, K. Kärre

Writing, review, and/or revision of the manuscript: A.K. Wagner, S.L. Wickström, R. Tallerico, T. Lakshmikanth, H. Brauner, P. Höglund, E. Carbone, M.H. Johansson, K. Kärre

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.L. Wickström, S. Salam

Study supervision: P. Höglund, E. Carbone, M.H. Johansson, K. Kärre

Other (supervision of post doc experimental work): E. Carbone

The authors thank Margareta Hagelin, Maj-Britt Alter, Kenth Andersson, Helén Braxenholm, and Anna-Karin Persson for expert assistance with in vivo experiments, Birgitta Wester for flow cytometry help, Benedict Chambers and Ci Song for experimental assistance, and Katja Lindholm and Gustaf Vahlne for initial work in the inhibitory receptor blockade model. Members of Petter Höglund's, Björn Önfelt's, and Klas Kärre's group are acknowledged for stimulating discussions.

This work was supported by grants from Karolinska Institutet (M.H. Johansson, P. Höglund, and K. Kärre), the Swedish Foundation for Strategic Research (K. Kärre and P. Höglund), the Swedish Research Council (K. Kärre and P. Höglund), the Swedish Cancer Society (K. Kärre and P. Höglund), the Swedish Childhood Cancer Foundation (K. Kärre), the Åke Wiberg Foundation (M.H. Johansson), the Magnus Bergwall Foundation (M.H. Johansson), the Royal Swedish Academy of Sciences (M.H. Johansson), the Syskonen Svensson Foundation (M.H. Johansson), the Swedish National Board of Health and Welfare (M.H. Johansson), AIRC (grant 15521), and UICC International Cancer Technology Transfer Fellowship (E. Carbone). A.K. Wagner and S.L. Wickström are PhD candidates at Karolinska Intitutet and are supported by KID grants, S. Salam is supported by a grant from the Higher Education Commission Pakistan, R. Tallerico by a FIRC Post Doc Fellowship, and H. Brauner by a Karolinska Research Internship.

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.