DNAM-1 expression correlates with educating impact of MHC-I

Previously, we demonstrated that each individual MHC-I allele has a distinct educating impact on NK cells, with the D^d allele having the strongest influence, followed by K^b and a remarkably weak effect of D^b (ref. ³¹). Here, DNAM-1 expression was assessed in mice with no MHC-I (K^b−/−D^b−/−) gene, a single-MHC-I allele (K^b−/− or D^b−/−) and B6 wild-type (K^bD^b) mice. We found that about 45% of NK cells from K^b−/− D^{b−/−−/−} mice were DNAM-1⁺. Mice expressing only the weak educating MHC-I molecule D^b did not have an increased DNAM-1⁺ population. However, if one of the strong educating MHC-I molecules K^b or D^d was expressed, NK cells had a higher frequency of DNAM-1⁺ cells (Fig. 1d). The increase in DNAM-1⁺ cells thus followed the education hierarchy of the MHC-I alleles. In assessing DNAM-1 mean fluorescence intensity (MFI) on NK cells, expression of K^b or D^d alone, increased average DNAM-1 levels on the NK-cell pool (Fig. 1e). In conclusion, we found that the levels of DNAM-1 expression and the frequency of DNAM-1⁺ NK cells correlated with the educating impact of the MHC-I allele.

DNAM-1 expression on educated and uneducated NK cells

We next hypothesized that MHC-I-educated and non-MHC-I-educated NK-cell subsets within the same mouse strain would also have differential expression of DNAM-1. We examined the DNAM-1 expression on NK-cell subsets as defined by their pattern of self-MHC-I-specific inhibitory receptor expression: Ly49A, Ly49C, Ly49G2, Ly49I and NKG2A (Supplementary Fig. 2a). This allowed us to concentrate on single-positive (SP) cells, and to assess the impact of one specific Ly49r on DNAM-1 expression in the absence versus the presence of NKG2A (Fig. 2a). We found that among NK cells expressing only one inhibitory Ly49r, the Ly49C-SP and Ly49I-SP had a higher proportion of DNAM-1⁺ cells than the Ly49A-SP or Ly49G2-SP. The observed effect of Ly49r expression on DNAM-1 seemed to be additive, as the Ly49CI double-positive subset had a higher frequency of DNAM-1⁺ cells than either of the Ly49C-SP or Ly49I-SP. This could also be seen among the corresponding subsets co-expressing NKG2A. Interestingly, all NKG2A⁺ subsets had a higher frequency of DNAM-1⁺ cells than NKG2A⁻ subsets, with NKG2A-SP cells being 90% DNAM-1⁺. However, with the addition of Ly49r, there was a small but consistent reduction in the expression of DNAM-1 on the NK-cell subsets. Thus, the more inhibitory Ly49r expressed by an NK cell in addition to NKG2A, the proportion of DNAM-1⁺ cell was lower when compared to NKG2A-SP NK cells. This was more prominent if inhibitory Ly49r that do not bind strongly to self-MHC-I were co-expressed with NKG2A (Ly49A and/or Ly49G2) and less pronounced when self-MHC-I-specific Ly49r (Ly49C and/or Ly49I) were co-expressed (Fig. 2a).

Open in a separate window

Figure 2

DNAM-1 expression correlates with inhibitory receptors for self-MHC-I.

(a,b) Frequency of DNAM-1⁺ NK cells in B6 mice, (a) on NK-cell subsets that express a certain combination of inhibitory Ly49r in the presence (white dots) or absence (black dots) of NKG2A. Results are pooled from 26 mice taken from seven experiments. (b) DNAM-1 MFI on single-Ly49r⁺ NK cells in D^b−/−, D^d-transgenic, K^b−/−, B6 (K^bD^b), D^d-transgenic (K^bD^bD^d) and B2m^−/− mice. Results are shown as per cent of expression compared to the same NK-cells subset in K^b−/−D^b−/− mice, which is represented by the line at 100% relative DNAM-1 expression. Data are from three experiments with at least six mice in total. Significant differences are calculated against the corresponding NK-cell subset in K^b−/−D^b−/− mice by Mann–Whitney U-test and are depicted as *P<0.05, **P<0.01, ***P<0.001. Error bars denote s.d.

Interestingly, when we examined TIGIT, an inhibitory receptor for the DNAM-1-ligand CD155, we found that it is preferentially expressed on NKG2A⁻ NK cells (Supplementary Fig. 3a), K^b−/−D^b−/− NK cells had higher levels of TIGIT (Supplementary Fig. 3b) and the expression pattern on NK-cell subsets was opposite to that of DNAM-1 (Supplementary Fig. 3c), indicating that DNAM-1 and TIGIT may be regulated in a balanced fashion.

The higher expression of DNAM-1 on NK cells expressing self-MHC-I-binding inhibitory receptors could also be seen among other Ly49r-MHC-I allele combinations (Fig. 2b). Frequency and MFI of DNAM-1⁺ cells among Ly49C-SP or Ly49I-SP was higher compared to Ly49A-SP or Ly49G2-SP NK-cell subsets in D^b−/− mice (Fig. 2, Supplementary Fig. 2b). In D^d single mice, there was also a significant increase of DNAM-1 expression, but in a different pattern based on self-educating NK-cell subsets in this host (Fig. 2b). In addition, the NK-cell subsets expressing Ly49A-SP or Ly49G-SP had the highest frequency of DNAM-1⁺ NK cells among the NKG2A-negative subsets (Supplementary Fig. 2b). In K^b−/− mice, none of the SP subsets exhibited a significant increase of DNAM-1 (Fig. 2b), although there was a tendency in the Ly49A-SP NK cells, which could indicate a weak education impact of the D^b molecule on this subset due to weak binding to MHC-I (ref. ³²). In D^d-transgenic mice (K^bD^bD^d), as expected, the Ly49A, Ly49C and Ly49G2-SP subsets had a higher expression of DNAM-1.

Additive effect of multiple self-Ly49r on DNAM-1 expression

Next, we asked whether NK cells with more inhibitory receptors express more DNAM-1, and whether any pattern emerging would depend on MHC-I expression. We divided NK-cell subsets of B6 and B2m^−/− mice based on the number of inhibitory Ly49r and expression of NKG2A, which allowed us to identify and dissect two distinct patterns. DNAM-1 was associated with the number of Ly49r (regardless of specificity), both in the NKG2A⁺ and NKG2A⁻ subsets, albeit with a different impact. One component of this association is education-dependent (seen in NKG2A⁻ subsets in B6), which can be described as ‘the more Ly49r, the more DNAM-1' (Fig. 3a), irrespective of which Ly49r are expressed, as long as there is MHC-I expression. The more Ly49r that are expressed, the more likely it is, that this particular cell belongs to the educated pool of NK cells, thus corroborating our observation that DNAM-1 expression is associated with education.

Open in a separate window

Figure 3

Correlation of DNAM-1 expression to the number of inhibitory Ly49r.

The number of inhibitory Ly49r on NK-cell subsets has an impact on the proportion of DNAM-1⁺ cells within that subset. B6 or B2m^−/− NK cells, stained for inhibitory receptors as for Fig. 2a, have been grouped based on the number of inhibitory Ly49r they express (regardless of which combination of inhibitory receptors). The frequency of DNAM-1⁺ NK cells within each group is shown for NKG2A⁻ cells in a and for NKG2A⁺ cells in b. Significant differences are calculated comparing the corresponding NK-cell subset of B6 and B2m^−/− mice by unpaired t-test and are depicted as **P<0.01, ***P<0.001. Error bars denote s.d.

A second component of the association between the number of Ly49r and DNAM-1 expression was education-independent (seen in B2m^−/−), which can be described as ‘the more Ly49r, the less DNAM-1'(Fig. 3b). This was most pronounced in the NKG2A⁺ subset. We propose that this education-independent downregulation of DNAM-1 is associated with a higher probability of the cells being more mature, because NK cells acquire more Ly49r successively and cumulatively as they mature^33,34,35. This is only seen in the presence of a strong educating MHC-I allele (Supplementary Fig. 4a–c) and is in line with the recent findings that DNAM-1 is downregulated upon maturation²⁹.

Recent models for NK-cell education are based on shifts in activation threshold determined by the combined input of inhibitory and activating receptors^4,36,37, where inhibitory receptors decrease the activation threshold thereby increasing reactivity while activating receptors influence education in the opposite direction, that is, they tune down the responsiveness of an individual NK cell in continuous presence of the ligand for the activating receptor^37,38,39. We therefore assessed the influence of an activating receptor (Ly49D) in the presence or absence of the cognate MHC-I allele (D^d) on DNAM-1 expression. We analysed NK cells that either express the inhibitory Ly49A or the activating Ly49D or both and found that the proportion of DNAM-1⁺ cells was higher in the Ly49A-SP subset in D^d single mice when compared to K^b−/−D^b−/− or B6 mice. When Ly49D was expressed, the frequency of DNAM-1 was lower in the presence of the ligand D^d (Supplementary Fig. 4d). As previously published, both Ly49D and Ly49H were less frequent on DNAM-1⁺ cells²⁸, while other activating receptors such as NK1.1, NKG2D, NKp46 and 2B4 had a higher expression of DNAM-1⁺ NK cells compared to DNAM-1⁻ NK cells within B6 mice (Supplementary Fig. 4e), or on total NK cells from B6 mice compared to CD226^−/− mice (Supplementary Fig. 4f).

Educating molecules correlate with DNAM-1 expression

To dissect our data set in an unbiased way and to reduce the dimensionality, we performed a principal component analysis (PCA, Fig. 4) and an orthogonal projection and latent structures discriminant analysis (Supplementary Fig. 5). The PCA was calculated on the proportion of DNAM-1⁺ cells within a given NK-cell subset based on the expression of inhibitory Ly49r and NKG2A in mouse strains which express one defined MHC-I allele. The PCA revealed that the frequency of DNAM-1⁺ cells within each NK-cell subset separates these strains into three clusters (Fig. 4a) and identified three informative principal components (PC, two first shown) that together account for 87% of the variance. PC 1 (x axis) separates D^d single (red) from D^b−/− (blue) and K^b−/−D^{b−/−−/−} (yellow) and K^b−/− mice (green). The difference between the three clusters in the PC 1-direction seems to be the presence of an MHC-I allele with a strong educating impact. K^b−/−D^{b−/−−/−} (yellow) and K^b−/− mice (green) do not have an MHC-I allele that can measurably educate NK cells³¹. NK cells of D^b−/− mice (blue) are educated by Ly49C/I, while NK cells in D^d single mice (red) are educated by the presence of D^d via Ly49A/G. Interactions between Ly49A and D^d are of higher affinity, and confer a superior responsiveness as any other Ly49r-MHC-I pair^14,40,41. For separation into PC 1 (x axis), all subsets contribute positively (Supplementary Fig. 5a), while PC 2 (y axis) separates D^d single and K^b−/− mice (lower quadrants) from K^b−/−D^{b−/−−/−} mice (cluster at 0) and D^b−/− mice (upper quadrants). The main separator for PC 2 seems to be the possibility to receive education, either through Ly49C/I, or via Ly49A/G. This can be further supported by plotting the variables that contribute the separation in PC 2 (Fig. 4b), where NK-cell subsets expressing Ly49C/I exhibit positive values, while those with Ly49A/G have negative values. We conclude that the frequency of DNAM-1⁺ cells among NK-cell subsets based on Ly49r expression can cluster the mouse strains according to known interactions of certain MHC-I alleles with specific Ly49r.

Open in a separate window

Figure 4

PCA reveals that the mice spontaneously cluster into three main groups.

(a) PCA on K^b−/−D^b−/−, D^b−/−, K^b−/− or Kb^−/−Db^−/−Dd^+/+ based on the proportion of DNAM-1⁺ cells (DNAM-1 frequency) on the 32 NK-cell subsets defined by inhibitory Ly49r and NKG2A expression. NK cells are gated on live singlets NK1.1⁺CD3⁻ cells. Data are pooled from eight independent experiments with at least eight mice per group. The model has three significant principal components (two first shown). x axis corresponds to PC 1, y axis to PC 2. (b) Variables contributing to the separation in predictive PC 2 for the PCA. Error bars represent the 95% confidence interval.

DNAM-1⁺ NK cells kill MHC-I-deficient tumour cells

To test the hypothesis that DNAM-1-expressing NK cells represent the educated subsets endowed with capacity for missing self-recognition, we examined the ability of sorted DNAM-1⁺NKG2A⁺, DNAM-1⁻NKG2A⁺ and DNAM-1⁻NKG2A⁺ subsets of⁻ IL-2-stimulated NK cells to kill MHC-I-expressing (RMA) and MHC-I-deficient (RMA-S) cells. DNAM-1⁺NKG2A⁺ and DNAM-1⁺NKG2A⁻ NK cells could differentiate between RMA and RMA-S cells (Fig. 5a–c), while DNAM-1⁻NKG2A⁻ NK cells could not discriminate between the two cell lines. RMA and RMA-S cells have similar but low levels of CD155 and CD112 (Supplementary Fig. 6), indicating that the difference in detection of missing self is due to a distinct difference in education and not a result of dissimilar stimulation via DNAM-1. All three populations could kill YAC-1 equally well (Fig. 5d), and are therefore functional in recognizing and killing tumour cells in a system that is primarily dependent on recognition by NKG2D. IL-2-stimulated NK cells from CD226^−/− mice could discriminate between MHC-I⁺ and MHC-I⁻ target cells (Fig. 5e,f), which shows that the observed difference between DNAM-1⁺ and DNAM-1⁻ NK cells is not a result of direct interaction of DNAM-1 with the ligands on the target cells. This suggested that the presence of DNAM-1 on NK cells correlates with their ability for missing self-recognition but is not necessary to maintain that function. Along the same line, we observed that both DNAM-1⁺ and DNAM-1⁻ cells within educated subsets (Ly49A-SP in D^d single mice, Ly49C-SP in D^b−/− mice) performed equally well in the production of IFN-γ and CD107 release in response to crosslinking of NK1.1 (Fig. 6). Thus, we conclude that DNAM-1 is not required to induce education.

Open in a separate window

Figure 5

DNAM-1 is not required for missing self recognition.

The ability of NK cells to kill MHC-I-deficient tumour targets correlates with DNAM-1 expression. (a–d) In vitro cytotoxicity of sorted DNAM-1⁺NKG2A⁺ (squares), DNAM-1⁺NKG2A⁻ (circles) and DNAM-1⁻NKG2A⁻ (triangles) IL-2-stimulated NK cells. (a) Killing of RMA, one representative of three experiments shown. (b) Killing of RMA-S, one representative of three experiments shown. (c) Calculation of the differential killing (±s.d.) between RMA-S and RMA killing, n=3 experiments. (d) Killing of YAC-1 n=3 experiments. (e,f) In vitro cytotoxicity towards RMA (circles) and RMA-S (squares) by IL-2-stimulated CD226^−/− NK cells. (e) Shows one representative experiment of three and (f) shows differential killing between RMA-S and RMA, n=3 experiments. Error bars denote s.d.

Open in a separate window

Figure 6

DNAM-1⁺ and DNAM-1⁻ NK cells respond to receptor crosslinking.

(a) Expression of IFN-γ and CD107a on DNAM-1⁺ and DNAM-1⁻ Ly49A SP NK cells from a D^d-transgenic mouse on K^b−/−D^b−/− background (K^b−/−D^b−/−D^d+/+), after crosslinking with anti-NK1.1, one representative plot shown. (b) Frequency of IFN-γ⁺ NK cells on Ly49r SP subsets (n=6, 3 experiments). (c) Expression of IFN-γ and CD107a on DNAM-1⁺ and DNAM-1⁻ Ly49C-SP NK cells from a D^b−/− mouse (K^b+/+D^b−/−) after crosslinking with anti-NK1.1, one representative plot shown. (d) Frequency of IFN-γ⁺ NK cells on Ly49r SP subsets (n=6, 3 experiments).

Lack of DNAM-1 does not impair missing self recognition

On the basis of the rheostat model of education, NK cells are tuned by continuous interactions via their receptors with MHC-I and possibly other ligands. If DNAM-1 is crucial for reaching or maintaining the educated state, this could be tested by examining missing self-responses in mice lacking DNAM-1 or by blocking the interaction between DNAM-1 and it's ligands in vivo. CD226^−/− mice rejected MHC-I-deficient target spleen cells, although slightly less efficiently compared to B6 wild-type mice (Fig. 7a). This could indicate an education defect in the absence of DNAM-1, or that DNAM-1 is required for target cell killing. CD226^−/− mice had a normal NK-cell compartment both in terms of frequency and absolute number²⁰. Nevertheless, a reduction of NKG2A-SP cells could be observed in CD226^−/− mice when compared to B6 wild-type mice (Fig. 7b). To examine whether blockade of DNAM-1 can change the education status of NK cells, B6 mice were treated with anti-DNAM-1 monoclonal antibody for 2 or 14 days⁴². Treatment completely abrogated binding of several DNAM-1 antibodies (clones 10E5, 480.1 and TX-42.1; Supplementary Fig. 7) thus verifying that we could maintain complete blockade of DNAM-1 in vivo. Treated mice had no change in frequency and number of total NK cells and in vivo blocked NK cells could still efficiently kill MHC-I-deficient spleen target cells (Fig. 7c). We observed only small changes in the Ly49r repertoire, however, similar to CD226^−/− mice, the frequency of NKG2A-SP NK cells was markedly reduced at both time points tested (Fig. 7d). This demonstrates that DNAM-1 is neither necessary to achieve education nor to maintain the educated state of mature NK cells, and that DNAM-1 is not the major determinant for the rejection of MHC-I⁻ spleen cells.

Open in a separate window

Figure 7

Absence of DNAM-1 does not abrogate missing self killing.

In vivo killing of MHC-I-deficient spleen cells were assessed in CD226^−/− mice (a) or B6 wild-type mice after treatment with DNAM-1 mAb (c). CFSE-labelled B6 or MHC-I^−/− spleen cell suspensions were inoculated i.v. and rejection was assessed 44h later in the spleen. (b) Frequency of NKG2A-SP NK cells gated on live, singlet CD3⁻NK1.1⁺Ly49r⁻ cells in CD226^−/− mice. Bar graphs show data from six (B6, B2m^−/−), seven (2 days) or four (2 weeks) mice per group of two independent experiments with at least two mice per group. (c,d) DNAM-1 was blocked by injection of 200μg of anti-DNAM-1 (clone 3B3) for 2 days or 2 weeks every 5 days and rejection of CFSE-labelled B6 or MHC-I^−/− spleen cells was measured 44h after inoculation of target cells (d) Frequency of NKG2A-SP NK cells gated on live, singlet CD3⁻NK1.1⁺Ly49r⁻ cells in B6 mice treated with anti-DNAM mAb for 2 days or 2 weeks. Bar graphs show data from 6 (B6), 4 (B2m^−/−) or 14 (CD226^−/−) mice per group of two independent experiments with at least two mice per group. P values are calculated with t-test and are depicted **P<0.01, ***P<0.001. Error bars denote s.d.

Expression of DNAM-1 during NK-cell development

The notion that DNAM-1 expression may be related to education raises the question as to which stage during development of NK cells DNAM-1 expression first occurs. During differentiation, common lymphocyte progenitors (CLP, lin⁻c-kit⁻flk2⁺CD27⁺CD224⁺IL-7Rα⁺CD122⁻, Supplementary Fig. 8) develop into pre-NK precursors (pre-NKP, lin⁻c-kit⁻flk2⁻CD27⁺CD224⁺IL-7Rα⁺CD122⁻), progress into refined NKP (rNKP, lin⁻c-kit⁻flk2⁻CD27⁺CD224⁺IL-7Rα⁺CD122⁺) and ultimately give rise to NK1.1⁺NKp46⁺ NK cells, which egress from the bone marrow (BM)⁴³. Both DNAM-1 and NKG2A were absent on CLP, a few pre-NKP started to express DNAM-1 while rNKP were mostly positive for DNAM-1 (Fig. 8a,b). Expression of NKG2A, the earliest MHC-I-specific inhibitory receptor⁴⁴, was detected first on NK1.1⁺ immature BM NK cells.

Open in a separate window

Figure 8

DNAM-1 is expressed in early NK-cell progenitors before inhibitory receptors.

(a) DNAM-1 expression, (b) NKG2A expression on CLPs, pre-NKPs, R-NKPs and mature NK cells. Gating strategy for CLP pre-NKP, and rNKP is shown in detail in Supplementary Fig. 8. Mature BM or splenic NK cells were defined by gating on NK1.1⁺CD3⁻ cells. NKG2A or DNAM-1 expression was analysed in these subsets. Pooled data from at least three different experiments from 14 mice are shown, with s.d. (c) DNAM-1 frequency and (d) DNAM-1 MFI on NK cells of different maturation stages in the spleen. Pooled data from at least three different experiments from 12 mice are shown with s.d. Error bars denote s.d.

Therefore it appears that DNAM-1 expression is switched on before and independently of MHC-I-specific inhibitory receptors, opening the possibility that it may be involved in educating NK cells already in the BM. DNAM-1 expression was similar on NK cells from spleen, BM and liver from 10-day-old B6 and K^b−/−D^{b−/−−/−} mice, before the onset of Ly49r expression (Supplementary Fig. 9), suggesting that the reduced level seen in adult K^b−/−D^b−/− mice is imposed later during life. CD27 and CD11b are markers that dissect NK cells into functionally distinct maturation subsets⁴⁵. DNAM-1 expression increases when NK cells upregulate CD27, and DNAM-1 expression is highest on CD27⁺CD11b⁻ NK cells and then goes down again upon further maturation to CD27⁻CD11b⁺ cells (Fig. 8c,d).

Antibodies and flow cytometric analysis

For flow cytometry, single-cell suspensions were labelled with fluorescently conjugated anti-mouse antibodies. All antibodies were titrated and used at optimal dilutions. CD3 (145.2C11), CD27 (29A1.4), CD11b (M1/70.15), CD69 (H1.2F3), IFN-γ (XMG1.2), CD244 (m2B4(B6)458.1), cKit (2B8), CD127 (A7R34), CD122 (TM-β1), CD135 (A2F10), NK1.1 (PK136), KLRG1 (2F1), CD107a (1D4B), NKp46 (29A1.4), DNAM-1 (10E5, 480.1, TX-42.1 and 3B3 (ref. ²⁸)) and Ly49A (YEI/48) (Biolegend), Ly49G2 (4D11), and NKG2D (CX5) (BD Biosciences), and Ly49I (YLI-90) and NKG2A (20d5) (eBioscience). The 4LO3311 (Ly49C) hybridoma was a kind gift from Suzanne Lemieux. All surface staining was performed in phosphate-buffered saline (PBS) after blocking unspecific staining via FcγRII/III with purified anti-CD16/32 (2.4G2) (MabTech). Dead cells were excluded from the gating with the Aqua dead cell dye (Life Technologies). Flow cytometric analyses were performed on an LSRII or FACSCalibur (Becton Dickinson) and data analysis was performed using the FlowJo software (Tree Star). For comparison of relative expression levels of specific NK-cell subsets, we have normalized the expression data (MFI) of all NK-cell subsets in every individual mouse to the mean of the MFI of the same NK subset in all K^b−/−D^b−/− mice in the same experiment.

Tumours

RMA (H-2^b) is a subline derived in our laboratory from the B6 mouse-derived EL-4 T-cell leukemia⁶², a kind gift from Prof G. Klein and RMA-S is a TAP2-deficient variant of RMA⁶³, also derived in our laboratory. Leukaemia cells were propagated as ascites lines in irradiated B6 mice. YAC-1 is a Moloney murine leukaemia virus induced T-cell lymphoma of the A/Sn strain which was maintained in vitro. YAC-1 and EL-4 cells were obtained from G Klein, Karolinska Institutet.

Assessment of NK-cell activity in vivo

In vivo cytotoxicity assays were performed as previously described³⁶. Briefly, single-cell suspensions of spleen cells (B2m^−/−, B6) were labelled with a high (target cells: B2m^−/−) or low (control cells: B6) dose of CFSE (Invitrogen Molecular Probes). Target and control cells were mixed at a 1:1 ratio and co-injected intravenously (i.v.) 20 × 10⁶ (spleen cells) cells. Spleens were harvested after 2 days and relative percentages of CFSE-positive cells in each population were measured by flow cytometry. Rejection is given as the relative survival of target cells=(% remaining control cells/% control cells in inoculate)/(% remaining targets/% targets in inoculate).

In vivo blockade

In vivo blockade of DNAM-1 was performed as previously described²⁸. Briefly, mice were initially injected i.v. with 400μg anti-DNAM-1 (mAb 3B3). After this time point mice were repeatedly injected every 5 days with 200μg of mAb. After 48h, or 14 days, the in vivo capacity of NK cells to kill i.v. injected spleen cells and the maturation pattern of NK cells were assessed.

In vitro cytotoxicity assay

IL-2-stimulated NK cells were generated by culturing sorted (MACS Miltenyi) splenic NK cells in complete αMEM medium (αMEM, 10mM Hepes, 20μM 2-mercaptoethanol, 10% FBS, 100Uml⁻¹ penicillin, 100Uml⁻¹ streptomycin) for 5 days with 1,000U rIL-2 (Immunotools), previously described²⁸. Target cells were incubated for 1h in the presence of Na₂⁵¹CrO₄ (Cr; Amersham) and then washed 3 × in PBS and incubated with effector cells at indicated effector:target (E:T) ratios. After 4h, cell culture supernatants were collected and analysed by a radiation counter (Wallac, PerkinElmer). Specific lysis was calculated as follows: %specific lysis=((experimental release−spontaneous release)/(maximum release−spontaneous release)) × 100.

In vitro crosslinking assay

To measure degranulation and IFN-γ production upon NK1.1 or NKp46 crosslinking, U-bottom plates were coated with 20μg of anti-NK1.1 or anti-NKp46 for 1h at 37°C before being left overnight at 4°C. The wells were then washed and 5 × 10⁴ sorted NK cells (NK-cell isolation kit II, Miltenyi Biotec, cat.no: 130-096-892) were added to the wells. CD107a was added to the cultures for the duration of the assay (1:200). After 1h, brefeldin A and monensin (Biolegend) were added to culture and incubated for another 3h. Cells were collected and stained for surface markers and then fixed and permeabilized (BDCytofix/Cytoperm, BD Bioscience, cat. no: 554714). Intracellular IFN-γ was detected by staining with anti-IFN-γ antibody (Biolegend).

BM preparation and surface staining

BM was harvested from donor mice by crushing bones and removing debris using pre-separation filters (Milteny Biotech). Unfractionated BM cells (5 × 10⁶ per 100μl) were surface stained as indicated in Supplementary Fig. 8. The cells positive for the following markers: CD11b, Gr-1, Ter119, CD19, NK1.1, CD11c and CD3 were considered as lineage positive and excluded from further analysis. The source of all antibodies used is described above.

NK-cell preparation and adoptive transfer

Splenic 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. In total, 1–3 × 10⁶ NK cells were injected i.v. to irradiated (8Gy) mice.

Statistics and multivariate analyses

Statistical analyses (except for Fig. 4) were conducted using GraphPad Prism 5. Either non-paired or paired two-tailed Students t-test and one-way analysis of variance (ANOVA; for frequencies) or Mann–Whitney U-test and repeated measures ANOVA (for MFI) were performed for data from at least three independent experiments, unless indicated otherwise. Error bars denote s.d. P values are depicted as: *P<0.05, **P<0.01, ***P<0.001.

For multivariate analyses (Fig. 4 and Supplementary Fig. 5; see also Supplementary Note 1) SIMCA software version 14 (Umetrics) was used. PCA was used to investigate underlying trends in the data, while orthogonal projections to latent structures with discriminant analysis was used to investigate the difference between defined groups (mouse strains).

We thank Kenth Andersson and Helén Braxenholm for expert assistance with in vivo experiments, Birgitta Wester for flow cytometry help, Elina Staaf for introduction to and teaching of the multivariate analyses software and Maria H. Johansson for critical reading and comments on the manuscript. Members of Petter Höglund's, Klas Kärre's and Benedict Chamber's group are acknowledged for stimulating discussions. This work was supported by Swedish Cancer Society (B.J.C., K.K.), the Swedish Foundation for Strategic Research (B.J.C., K.K.), Stockholm County Council Theme Center Grant (B.J.C.), the Swedish Childhood Cancer Foundation (K.K.) and the Swedish Research Council (K.K.).