Novel non-canonical role of STAT1 in Natural Killer cell cytotoxicity

ABSTRACT STAT1 is an important regulator of NK cell maturation and cytotoxicity. Although the consequences of Stat1-deficiency have been described in detail the underlying molecular functions of STAT1 in NK cells are only partially understood. Here, we describe a novel non-canonical role of STAT1 that was unmasked in NK cells expressing a Stat1-Y701F mutant. This mutation prevents JAK-dependent phosphorylation, subsequent nuclear translocation and cytokine-induced transcriptional activity as verified by RNA-seq analysis. As expected Stat1-Y701F mice displayed impaired NK cell maturation comparable to Stat1−/− animals. In contrast Stat1-Y701F NK cells exerted a significantly enhanced cytotoxicity in vitro and in vivo compared to Stat1−/− NK cells in the absence of detectable transcriptional activity. We thus investigated the STAT1 interactome using primary NK cells derived from Stat1ind mice that inducibly express a FLAG-tagged STAT1. Mass spectrometry revealed that STAT1 directly binds proteins involved in cell junction formation and proteins associated to membrane or membrane-bound vesicles. In line, immunofluorescence studies uncovered the recruitment of STAT1 to the target-cell interphase during NK cell killing. This led us to propose a novel function for STAT1 at the immunological synapse in NK cells regulating tumor surveillance and cytotoxicity.


Introduction
The signal transducer and activator of transcription 1 (STAT1) protein drives transcriptional programs induced by various cytokines such as interferons (IFNs) and IL-12. Accordingly, STAT1 is important for signal transduction in natural killer (NK) cells. NK cells are granular lymphocytes and build the front line against virally infected and malignant cells. Loss of STAT1 is associated with a pronounced impairment of NK cell maturation, cytotoxicity and tumor surveillance. 1,2 In the canonical JAK/STAT signaling cascade binding of cytokines to the respective receptor triggers Janus kinases (JAKs) to phosphorylate STAT1 on tyrosine (Y)701 that allows the formation of active STAT1 dimers. These translocate to the nucleus and bind to stimulus-specific DNA response elements to initiate or repress target gene transcription. Mutation of STAT1-Y701 to phenylalanine (F) precludes phosphorylation and the formation of nuclear translocation-competent STAT1 dimers, thereby abolishing the transcription of typical STAT1 target genes. 3 Cytokine stimulation induces the successive phosphorylation of STAT1 on Y701 and serine (S)727. 4 Interestingly, primary NK cells display a pronounced constitutive phosphorylation on S727 that restrains the cytotoxic capacity. 1 Compared to wild-type, NK cells with mutated STAT1-S727A protein show enhanced cytotoxicity and tumor surveillance. As the canonical function of STAT1-S727 phosphorylation is to increase transcriptional activity, negative regulation by S727-phosphorylated STAT1 (STAT1-pS727) presents a novel aspect of the protein's biology that is mechanistically not understood. As the negative regulation by STAT1-pS727 was observed in absence of discernible pY701, the possibility of STAT1 activity not requiring tyrosine phosphorylation was raised.
To unravel the non-canonical function of STAT1 in NK cells, we utilized Stat1-Y701F knock-in mice 5 lacking the ability to generate STAT1-pY701 protein. We report that the severely impaired NK cell cytotoxicity of Stat1-deficient animals is partially rescued in Stat1-Y701F mice. Consistent with its ability to confer NK cytotoxicity STAT1-Y701F partially restored NK cell-mediated tumor surveillance. Mass spectrometry analysis of NK cells expressing a doxycycline-regulated, FLAG-tagged STAT1 (Stat1 ind ) 6 revealed that incubation with target cells causes STAT1 to interact with a plethora of proteins that are important for cell junction formation and found in membrane-bound vesicles. Immunofluorescent staining of primary NK cells uncovered that a large proportion of STAT1 is present at the target cell interface.

STAT1-Y701F partially restores NK cell cytotoxicity
To investigate non-canonical functions of STAT1 in NK cells we generated Stat1-Y701F knock-in mice. 5 Ex vivo analysis of primary NK cells confirmed the lack of STAT1-Y701 phosphorylation (Fig. 1A) and of transcriptional activation of typical target genes, i.e. Mx1, Mx2, Irf7 and Gbp2 (Fig. 1B) upon type I IFN stimulation. Expression of the Stat1 gene is strongly reduced in cells expressing STAT1-Y701F, owing to the lack of a phosphotyrosine-dependent tonic signal. Despite the drastically reduced STAT1 protein levels in Stat1-Y701F NK cells ex vivo (Fig. 1A), constitutive phosphorylation on STAT1-S727 was clearly detectable (Fig. 1A), in line with previous observations. 1 Assessment by flow cytometry demonstrated that the number and maturation of splenic NK cells was impaired in Stat1-Y701F mice, comparably to Stat1 ¡/¡ NK cells (Fig. 1C). In contrast, we discovered a substantial difference between Stat1 ¡/¡ and Stat1-Y701F NK cells in their ability to kill tumor target cells. NK cell cytotoxicity was partially restored in Stat1-Y701F NK cells in in vitro assays upon IL-2 expansion ( Fig. 2A and  S2A). Noteworthy, we found that in vitro cultivation in IL-2 for 5 d enhanced STAT1-Y701F expression levels (Fig. S1). Most importantly the differences in cytotoxicity were not restricted to the in vitro situation but also extended to NK cell-dependent tumor surveillance in vivo. Upon intravenous injection of B16F10 melanoma cells, Stat1-Y701F mice developed only few pulmonary tumor nodules by day 14, whereas Stat1 ¡/¡ mice already showed pronounced signs of tumor burden. Tumor development was significantly delayed in Stat1-Y701F mice and only at day 19 post injection tumor nodules were clearly visible (Fig. 2B). A similar picture was observed in the liver; whereas Stat1 ¡/¡ mice showed clear signs of liver metastasis at day 14 and day 19, this was observed to a lesser degree in Stat1-Y701F mice indicating that the effects are not specific for the lung (Fig. S2). This led us to conclude that NK cell-mediated cytotoxicity and tumor surveillance is partially rescued in Stat1-Y701F mice.
Rescue of NK cell cytotoxicity in Stat1-Y701F mice in spite of mostly unaltered transcriptome We next wondered whether a distinct so far unrecognized transcriptional response may be induced in NK cells in the presence of Stat1-Y701F that may explain the rescue of NK cell-dependent cytotoxicity and tumor surveillance. To obtain a complete picture of transcriptional changes occurring in a STAT1-dependent manner we performed RNAseq analysis in Stat1 ¡/¡ , Stat1-Y701F and wild-type NK cells upon stimulation with IL-2 and IL-12. Our efforts are summarized in Fig. 3. In line with the established role of STAT1-pY701 as prerequisite for transcriptional activity, we failed to see any hint for substantial target gene transcription in Stat1 ¡/¡ or Stat1-Y701F NK cells. When comparing alterations in Stat1 ¡/¡ to Stat1-Y701F NK cells we obtained a list of seven genes that were significantly altered (either >2-fold upregulation or < 0.5 downregulation; p value < 0.01), among which Stat1 itself served as a positive control (Table S1). Current knowledge and  published literature could not provide any obvious link between the transcriptomic alterations in Stat1-Y701F NK cells (including changes in CamK2b, FAM20c or CD59a expression) and the rescue of NK cell cytotoxicity.

STAT1 interacts with proteins involved in cell junction formation and is associated with membrane or membrane-bound vesicles in NK cells
We thus wondered whether STAT1 exerts non-canonical functions in NK cells by forming protein signaling complexes with other proteins independent of its tyrosine phosphorylation to promote cytotoxicity. To study protein complexes of STAT1 in primary NK cells we made use of a knock-in mouse harboring a FLAG-tagged allele of STAT1 (Stat1 ind ). The use of Stat1 ind mice enables the dose-dependent and time-restricted expression of a FLAG-tagged STAT1 protein (STAT1a FLAG ) on a Stat1-deficient background. 6 Treatment of Stat1 ind mice for 10 d with 0.2 mg/mL doxycycline via drinking water induced the expression of STAT1 in NK cells comparable to wild-type levels (Fig. 4A). The expression of STAT1a largely rescued the impaired maturation of Stat1 ¡/¡ NK cells as assessed by flow cytometric analysis of CD27/CD11b on splenic NK cells ( Fig. 4B and S3). The partial normalization of CD27/CD11b expression was paralleled by an increase of KLRG1 C NK cells comparable to levels observed in wild-type mice (Fig. 4C). The phenotypic rescue induced by STAT1a expression was accompanied by a gain of function: The in vitro cytotoxicity of STAT1 re-expressing NK cells against YAC-1 target cells was comparable to wild-type (Fig. 4D). The fact that the maturation and cytotoxic function of Stat1-deficient NK cells are rescued by doxycycline-induced STAT1a FLAG expression to wild-type levels allowed us to proceed with proteomic studies.
To look for binding partners of STAT1 in primary NK cells we performed co-immunoprecipitation (Co-IP) studies of STAT1a FLAG in primary murine Stat1 ind NK cells either unstimulated or co-incubated with human Jurkat cells. The use of human target cells and murine NK cells allows minimizing the rate of false-positive hits as any human protein can clearly be assigned to the tumor cell. Jurkat cells are efficiently killed by mouse NK cells (Fig. S5). An optimized IP protocol was established ( Fig. S4), which allowed the precipitation of STAT1a-FLAG complexes followed by mass spectrometry analysis (n D 2 biological replicates). This approach detected a total number of 3,581 proteins in murine NK cells, approximately 2% of which were excluded from further validation as they belonged to the families of keratins, immunoglobulins, ribosomal and heat shock proteins and are frequently found as contaminants. Notably, 4.3% of the proteins detected in murine NK cells were found in the precipitated fractions indicating binding to STAT1a ( Fig. 5A and S5). Out of the 153 STAT1-binding proteins 18% were found to be constitutively associated to STAT1, 63% were increasingly bound upon target cell co-incubation, 11% were partially dissociated from STAT1 upon target cell coincubation and 8% were not assignable to any of these groups (Fig. 5B). Fig. 5A shows a short list of STAT1-binding proteins obtained in the mass spectrometry analysis (see Fig. S5 for the entire data set or visit ProteomeXchange: identifier PXD002206). Proteins highlighted in red have been described as STAT1 interaction partners and verified the successful pulldown and specificity of the experiment. Fig. 5C summarizes these well-known STAT1 interaction partners according to the String database (http://string-db.org/). Gene ontology enrichment analysis and visualization tool (GORILLA, http://cblgorilla.cs.technion.ac.il/) was used to search for enriched GO terms in our STAT1-interaction partner list compared to the background list of the overall proteome in the input sample. As expected the analysis revealed that STAT1a interacts with DNA and other macromolecular complexes in the nucleus. Consistent with our assumption of an extranuclear function however, several previously STAT1-unrelated GO terms were enriched including cell junction, membrane and membranebound vesicles (summarized in Fig. 5D).
To validate the mass spectrometry analysis we performed Co-IP of STAT1a FLAG -expressing NK cells either unstimulated or co-incubated with target cells followed by western blotting for distinct proteins. Fig. S6 summarizes these efforts and shows the constitutive and direct interaction of STAT1a FLAG with annexin A2 (Fig. S6A) and the short-term interaction with protein phosphatase 1B (PPM1B) and perforin, which is mainly induced upon target cell killing (Fig. S6B).

In NK cells STAT1 is recruited to site of target cell contact
The proteomic data indicate that STAT1 may contribute to the cytotoxic process by interacting with cell junction and membrane-associated proteins. We thus wanted to investigate whether the proposed non-transcriptional activity of STAT1 in the course of NK cell killing is reflected by changes in the subcellular localization of STAT1. We scrutinized the spatial distribution of STAT1 proteins in wild-type NK cells upon contact with target cells. Primary murine wild-type NK cells were coincubated for 30 min with leukemic cells followed by immunofluorescent staining. We chose Stat1 ¡/¡ leukemic cells as they represent prime NK cell targets due to their low expression levels of MHC class I. 7 In addition they served as optimal experimental control for STAT1 antibody specificity. The outcome of the experiment was clear-cut: as shown in Fig. 6A the majority of STAT1 proteins was polarized toward the tumor target cell GORILLA analysis (http://cbl-gorilla.cs.technion.ac.il/) of "cellular components" corroborated that STAT1 interacts with DNA and other macromolecular complexes in the nucleus and additionally revealed that STAT1 binds to proteins, which are associated to cell junction, membrane and membrane-bound vesicles with high significance. and assembled at the cell-cell interface. There STAT1 co-localized with F-actin (Fig. 6B), which is a well-established marker for the immunological synapse. 8 During NK cell killing the majority of the STAT1 protein pool remained cytoplasmic with a significant fraction being recruited to the area of the IS; only a minor fraction of STAT1 protein was present in the nucleus. Contrasting STAT1, STAT5 was not polarized toward the target cell and remained evenly distributed all-over the NK cell (Fig. 6C). This led us to conclude that STAT1 exerts a noncanonical transcription-unrelated function at the NK-target cell interface that is not shared by other members of the STAT transcription factor family.

Discussion
STAT1 is an essential mediator of immunity against microorganisms and tumors. Its vital importance derives from the employment by a number of different cytokine receptors that include IFN receptors and the IL-12 receptor. As both type I IFN and IL-12 are critical regulators of NK cell activity the impairment of cytotoxicity in Stat1 ¡/¡ mice meets the expectations. Surprisingly, however, STAT1 affects NK cell biology beyond its canonical role as an activator of immediate transcriptional responses. This notion emerged from our recent observation that phosphorylation at S727 bestows upon STAT1 the ability to restrict NK activity in the absence of detectable phosphorylation at Y701. 1 Following up on this finding, we now show that STAT1-Y701F partially rescues the defect of Stat1 ¡/¡ NK cells both in vitro and in vivo. The partial restoration of NK cell cytotoxicity occurred despite strongly reduced protein amounts underscoring its importance. In line with a non-nuclear and non-canonical activity of STAT1 in NK cells, the protein localized to the NK cell/target cell interface and copurified with membrane-associated and vesicular proteins.
Unphosphorylated STAT1 (U-STAT1) proteins predominantly localize to the cytoplasm as inactive homodimers. [9][10][11] In spite of this, nuclear and transcriptional activity has been assigned to U-STAT1. [12][13][14][15] This pY701-independent transcriptional function occurs upon upregulation of STAT1 through canonical signaling. As this feed-forward loop is impaired or even absent in Stat1-Y701F cells, insufficient STAT1 amounts are established for significant transcriptional U-STAT responses. Thus, the U-STAT1 pathway as originally described by Stark and colleagues is unlikely to rescue NK activity. Strong support for this argument stems from our RNA-seq data that showed only minor differences in transcriptional responses in Stat1-Y701F NK cells compared to Stat1 ¡/¡ controls. Seven genes were found significantly altered including Stat1, which served as control. Neither the altered expression of CD59a, nor of Fam20c or CamK2b provided a satisfactory explanation for the rescue of cytotoxicity. Fam20C is a kinase phosphorylating secreted proteins in a rather non-specific manner accounting for the majority of the secreted phospho-proteome. 16 CD59a ¡/¡ mice have been generated and show a mild hemolytic phenotype and an increased sensitivity to complementdependent lysis. 17 CaMKIIb is a serine-threonine kinase associated predominantly with neuronal functions. 18 We have currently no insights in the function of the so far solely RIKEN-annotated genes.
Despite the fact that we cannot formally rule out a transcriptional contribution to the observed phenotype we propose the existence of an unusual non-canonical function of STAT1 at the NK cell interface with its target cell. Our data provide strong evidence that the presence of STAT1 is required at the immunological interface to enable NK cell cytotoxicity- Figure 6. Polarization of STAT1 at the NK-target cell interface. (A) Murine wild-type LAK cells were co-incubated with CFSE-labeled Stat1 ¡/¡ v-abl C leukemic cells (green) in a ratio of 1:1 for 30 min prior to the immunofluorescent staining of STAT1 (red). Nuclear staining with DAPI (blue) was included as control. Scale bars: 10 mm. (B) Murine wildtype LAK cells were co-incubated with CFSE-labeled human Jurkat cells (green) for 30 min prior to the immunofluorescent staining of STAT1 (red) and F-actin (yellow). Upon tumor cell challenge STAT1 remained cytoplasmic and was partially found in the NK-immunological synapse, as it co-localized with F-actin. Nuclear staining with DAPI (blue) was included as control. Scale bars: 5 mm. (C) Murine wild-type LAK cells were co-incubated with CFSE-labeled human Jurkat cells (green) for 30 min prior to the immunofluorescent staining of STAT5 (red) and F-actin (yellow). Upon tumor cell challenge STAT5 remained evenly distributed over the NK cell and was not recruited to the NK immunological synpase as it did not co-localize with F-actin. Nuclear staining with DAPI (blue) was included as control. Scale bars: 5 mm.
thereby explaining the rescue that we observed upon expression of Stat1-Y701 in a Stat1-deficient background.
Recent years have seen the emergence of several phosphotyrosine-independent STAT functions in the cytoplasm, mitochondria or the cell nucleus. [19][20][21][22][23] Important in the context of our results, inactive Drosophila STAT was found associated with the apical membrane of epithelial cells in proximity to protein complexes regulating cell polarity. 24 In mammalian cells a fraction of U-STAT3 is in contact with the plasma membrane and with signaling endosomes. 25 These studies emphasize a role of STATs in membrane trafficking and organelleassociated signaling. At present, we cannot mechanistically explain the contribution of the U-STAT1 fraction at the target cell interface to NK cytotoxicity.
However, the presence of the protein in an IS is not unprecedented. The differentiation of Th precursors (Thp) to the Th1 lineage requires signaling via the T cell receptor (TCR) and the IFNg receptor (IFNGR). Activation of Thp cells induces the co-recruitment of TCR and IFNGR to the synapse and drives their differentiation into Th1 effector cells. 26 STAT1 is corecruited to the Thp-IS, binds to the IFNg receptor at the cell membrane and translocates to the nucleus. 27 Similar to NK cells, Thp STAT1 is constitutively phosphorylated on S727 in this situation, but remains Y701 unphosphorylated. Our purification of STAT1-associated NK cell proteins has not yielded any evidence for an association with the IFNg or other cytokine receptors, possibly due to the transient nature of the interaction, or due to limitations of the methodology. Our protocol was not optimized for the purification of membrane-bound proteins. In this light the high significance of STAT1 interaction with membrane proteins as revealed by GORILLA analysis is striking and similarities between STAT1 function in the IS of NK and Thp cells require further exploration.
In conclusion we provide the first evidence of U-STAT function in animals carrying a genomic mutation of the critical tyrosine residue. By demonstrating a membraneassociated function of U-STAT1 in NK cells we add an important new aspect to the complex biology and diverse employment of STATs in cell signaling and transcriptional regulation.

Co-immunoprecipitation (Co-IP) and Western Blotting
For Co-IP experiments 3 £ 10 6 NK cells were harvested and stimulated for 10 min with 100 U/mL rmIFN-b (Merck Millipore) or co-incubated for 30 min with 3 £ 10 6 Jurkat cells before the lysis in 200 mL IP lysis buffer. The composition of the optimal buffer to precipitate STAT1 complexes was determined experimentally (Fig. S4): 50 mM HEPES pH 7.5, 0.1% Tween-20, 150 mM NaCl, 1 mM EDTA, 10 mM b-glycerophosphate, 1 mM PMSF, 1 mM NaF, 500 nM Na 3 O 4 V and complete Protease Inhibitor Cocktail Tablets (Roche). STAT1a FLAG was precipitated with the use of the ANTI-FLAG M2 Affinity Gel (Sigma-Aldrich). As controls served 5% of the input (whole cell lysate) and 5% of the bead-supernatant (not bound to the beads). The proteins bound to the aFLAGbeads were eluted by cooking for 10 min in 4x Laemmli-buffer freshly complemented with b-mercaptoethanol.

Mass spectrometry
In-solution digestion: If not stated otherwise, all reagents were obtained from Sigma. The total lot of beads obtained per IP experiment was used for proteolytic digestion. Beads were washed with 50 mM ammonium bicarbonate (ABC buffer) on top of conditioned 3 kD MWCO filters (Pall Austria Filter GmbH) by centrifugation at 14,000 g for 15 min. After reduction with 200 mL of dithiothreitol solution (5 mg/mL dissolved in 8 M guanidinium hydrochloride in ABC buffer at pH 8) and alkylation with 200 mL of iodacetamide solution (10 mg/mL in 8 M guanidinium hydrochloride in ABC buffer), proteins were digested 18 h at 37 C using 10 mL trypsin solution (0.1 mg/mL). Clean up of peptide samples were performed using C-18 spin columns (Pierce, Thermo Scientific). Finally, the peptide samples were dried and stored at ¡20 C until MS analyses. For shotgun-analyses, dried samples were reconstituted in 5 mL 30% formic acid (FA) containing 10 fmol each of four synthetic standard peptides and diluted with 40 mL mobile phase A (98% H 2 O, 2% ACN, 0.1% FA). Synthetic peptides [Glu1-Fribrinopeptide B -EGVNDNEEGFFSAR; M28 -TTPAVLDSDGSYFLYSK; HK0 -VLETKSLYVR; HK1 -VLETK(e-AC)SLYVR] were obtained from Sigma and Peptide Specialty Laboratories GmbH and spiked in each sample as internal quality control for monitoring LC-MS-system stability.
Shotgun LC-MS analysis: 10 mL of the peptide samples were injected into a Dionex Ultimate 3000 nano LC-system coupled to a QExactive orbitrap mass spectrometer equipped with a nanospray ion source (Thermo Fisher Scientific). All samples were analyzed in technical duplicates. For pre-concentration, peptides were loaded on a 2 cm £ 75 mm C18 Pepmap100 precolumn (Thermo Fisher Scientific) at a flow rate of 10 mL/min using mobile phase A. Elution from the pre-column to a 50 cm £ 75 mm Pepmap100 analytical column (Thermo Fisher Scientific) and separation was achieved at a flow rate of 300 nL/min using a gradient of 8% to 40% mobile phase B (80% ACN, 20% Shotgun LC-MS data analysis: ProteomeDiscoverer 1.4 (Thermo Fisher Scientific, Austria) running Mascot 2.4 (Matrix Science, UK) was used for protein identification and qualitative data analysis. Protein identification was achieved searching against the SwissProt/UniprotKB Database (version 052014 including only reviewed proteins with 20,263 entries) allowing a mass tolerance of 5 ppm for MS spectra and 20 ppm for MS/ MS spectra as well as a maximum of two missed cleavages. Furthermore, search criteria included carbamidomethylation on cysteins as fixed modification and methionine oxidation as well as N-terminal protein acetylation as variable modifications. A list of lab-characteristic contaminants including various keratins was excluded manually. The FDR for peptide spectrum matches was set to < 0.01, in addition the Mascot significance threshold was set to 0.01.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium 30 via the PRIDE partner repository with the dataset identifier PXD002206 and 10.6019/ PXD002206. 31 Mass spectrometry data representation: The raw data was combined in a single Excel file and filtered for proteins found in the immunoprecipitate, i.e. the coverage in the IP > 0 in at least one of the four samples. Those hits (n D 153) were assigned to one of the following four groups: (i) proteins that are constitutively associated to STAT1a and found at comparable levels (coverage in the IP is constant) in unstimulated and target-stimulated NK cells (n D 40), (ii) proteins that are increasingly bound to STAT1a upon target cell co-incubation (coverage in the IP of target-stimulated > coverage in the IP of unstimulated; n D 80), (iii) proteins that are lost or less bound to STAT1a upon target cell co-incubation (coverage in the IP of target-stimulated < coverage in the IP of unstimulated; n D 23), and (iv) not assignable to either of these groups (n D 10). Within the four groups the hits were sorted according to the coverage in the IP in decreasing order. A colorimetric heatmap (three-color scale) was generated using the conditional formatting in Excel: green (low values), yellow (set at 75% percentile), red (high values). Zero values were displayed in black. The score of the corresponding input samples was displayed sideby-side for each protein respectively.

RNA-seq
Splenic NK cells were isolated using the MACS positive selection kit (DX5, Miltenyi) and cultured with 5000 U/mL rhIL-2 (Proleukin Ò , Roche) for 5 d. NK cells were sorted for CD3 ¡ NK1.1 C on a FACS AriaIII (BD) and samples were stimulated for 3 h with 5000 U/mL rhIL-2 § 5 ng/mL rmIL-12 (R&D). RNA was isolated by RNeasy Micro Kit (Qiagen) and RNA-seq has been analyzed based on the GRCm38 V17 mouse genome and gene annotation.
Sequencing and read processing: Paired-end 100 bp reads have been generated with the Illumina TruSeq protocol. On average 34 million read pairs have been generated for each sample. Reads were then aligned to the mouse reference genome (mm10) with the GSNAP aligner version 2012-12-20, which is an accurate splice-junction mapper for RNA-Seq data.
Differential gene expression analysis: To determine gene expression levels, we counted the number of reads for 38,293 mouse genes using HTSeq version 0.5.3 based on the Ensembl gene annotation version 71. We then used EDASeq (Version 1.4.0) to correct for GC-content bias and tested for differential gene expression using DESeq (Version 1.10.1).
Heatmaps and PCA clustering: Plots have been generated based on the 200 and 500 genes, which showed the strongest expression variance between the samples for heat map plot and PCA clustering respectively. Expression variances were calculated for samples which are shown in the plots only (all samples for PCA clustering and eight samples for heat map plots).

NK cytotoxicity assay
Flow cytometry-based in vitro cytotoxicity assays were performed as described previously. 1

B16F10 tumor model
Mice were injected via tail vein with 5 £ 10 4 B16F10 melanoma cells and sacrificed after 14 or 19 days, respectively. Lung and liver were harvested and captured on camera. Tumor nodules visible on their surface were counted under a microscope.

Statistical analysis
Statistical analysis was performed using GraphPad Prism Ò version 5.00 for Windows (GraphPad Software, San Diego, CA, www.graphpad.com). Where ANOVA showed a statistical difference, Tukey's multiple comparison testing was applied. The a level for all tests was set to 0.05 and p values were 2-tailed.

Disclosure of potential conflicts of interest
No potential conflicts of interest were disclosed.