STAT5 Is a Key Regulator in NK Cells and Acts as a Molecular Switch from Tumor Surveillance to Tumor Promotion

Natural killer (NK) cells are innate lymphocytes that develop from a common lymphoid progenitor in the bone marrow (1). They represent the first line of defense against infected, stressed, and malignant cells. Recent evidence has assigned distinct features and functions to tissue-specific NK cells (2). NK cells have organ-specific properties, such as distinct profiles of receptor expression or cytokine production (3). Uterine NK cells secrete high levels of VEGFA and are involved in placental vascularization. The physiologic functions of other organ-specific NK-cell subsets are less well understood (4).

All aspects of NK-cell development are regulated by cytokines, their downstream signaling pathways, and transcriptional regulators. These include key cytokines such as IL2, IL12, IL15, IL18, and IL21 (5), most of which signal via the common γ chain (5) and activate the JAK–STAT pathway (6). JAK kinases (JAK1–3 and TYK2) bind to cytokine receptors and are activated by ligand/receptor binding. The activated kinase phosphorylates STAT transcription factors (STAT1–STAT6; refs. 6, 7).

Consistent with its function as the major STAT protein downstream of IL7, IL2, and IL15, STAT5 is absolutely essential for conventional NK-cell development and survival; Stat5^Δ/ΔNcr1-iCre^Tg mice lack NK cells (8). It is also important for lymphoid cell development (9): STAT5 is constitutively active in a plethora of lymphoid malignancies (10). Recent studies have described somatic Stat5b mutations as active drivers of lymphoid malignancies (11, 12), and considerable efforts are under way to develop STAT5 inhibitors. Such inhibitors will not only target malignant hematopoietic cells but will also have severe consequences for immune functions that depend on the activation of STAT5 by cytokines. A precise understanding of the detailed function of STAT5 in immune cells is required to predict the potential side effects of a STAT5-directed therapy.

We now provide the first evidence that STAT5 is more than a master regulator of NK-cell survival, proliferation, and cytotoxicity: It also inhibits NK-cell–mediated tumor angiogenesis by directly suppressing transcription of the angiogenic factor VEGFA. This finding forces us to reconsider the potential consequences of STAT5-directed therapies for NK-cell–surveilled tumors.

STAT5 regulates the antiapoptotic genes Bcl-xl, Bcl2, and Mcl1 in various cell types (13–17). The lack of expression of these genes would explain the almost complete absence of mature NK cells in Stat5^Δ/ΔNcr1-iCre^Tg mice. We investigated the level of the transcripts of these genes in NK cells derived from heterozygotic Stat5^Δ/+Ncr1-iCre^Tg mice. We found about 50% of the wild-type (WT) number of mature NK cells in the periphery of Stat5^Δ/+Ncr1-iCre^Tg mice (Fig. 1A), indicative of a gene dosage effect. The reduced levels of Stat5a and Stat5b are paralleled by decreased levels of Bcl-xl, Bcl2, and Mcl1 (Supplementary Fig. S1A S1B). To investigate the association between STAT5 levels, the expression of antiapoptotic genes and NK-cell survival, we crossed Stat5^Δ/ΔNcr1-iCre^Tg mice—which display a drastic reduction in NKp46⁺ NK-cell numbers (8)—to Vav-Bcl2 mice. Expression of the Bcl2 transgene rescues STAT5-deficient NK cells; the numbers of CD3⁻NKp46⁺ cells in the spleen, lymph nodes, liver, and lung of Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice are comparable with those in Stat5^fl/fl animals, although they did not reach the levels of Vav-Bcl2 transgenic animals. The overexpression of Bcl2 also leads to an increased number of NK cells in the bone marrow in both Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 and Vav-Bcl2 mice (Fig. 1B; Supplementary Fig. S1C).

QPCR analysis showed drastically reduced levels of both Stat5 isoforms in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells. The residual STAT5 most likely reflects nondeleting NK cells and “recent deleters,” which retain some STAT5 due to the long half-life of the protein (ref.8; Fig. 1C). PCR confirmed that the Stat5 gene had been deleted in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells (Supplementary Fig. S1D). We also found a consistent reduction of Stat5a and Stat5b levels in Vav-Bcl2 NK cells (Fig. 1C), which most likely results from a decreased expression of cytokine receptors in these mice (data not shown). To confirm the rescue of STAT5-deficient NK cells by Bcl2, we generated Stat5^fl/flMx1-Cre-Vav-Bcl2 mice, which allow Stat5 to be deleted conditionally by treatment with Poly (I:C). In contrast with Stat5^Δ/ΔNcr1-iCre^Tg mice, where deletion only affects late NK-cell stages, Stat5^fl/flMx1-Cre animals delete Stat5 in all NK cells irrespective of their developmental stage. In line with our previous observations, deletion of STAT5 in Stat5^Δ/ΔMx1-Cre mice resulted in the lack of splenic NK cells, whereas Stat5^Δ/ΔMx1-Cre-Vav-Bcl2 animals had comparable numbers of NK cells to Stat5^fl/fl mice (Supplementary Fig. S1E). Deletion of STAT5 and the levels of Bcl2 were verified by qPCR and Western blotting (Supplementary Fig. S1F–S1H). The results confirmed that Bcl2 expression enables cells to survive the deletion of STAT5.

NK-cell maturation takes place in distinct stages, which can be distinguished by the expression of CD27 and CD11b. NK cells acquire CD27 before they express CD11b. CD27⁺CD11b⁺ NK cells represent cytotoxic NK cells, which cease to express CD27 to become CD27⁻CD11b⁺ (18, 19). Loss of one allele of Stat5 results in reduced numbers of CD27⁺CD11b⁺ NK cells, paralleled by an increase of immature CD27⁻CD11b⁻ and CD27⁺CD11b⁻ NK cells (Fig. 1D and E). The absence of both STAT5 alleles in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice is associated with the appearance of highly immature NK cells (CD27⁻CD11b⁻; Fig. 1D). A similar pattern was observed in Vav-Bcl2 mice, as Bcl2 expression stabilizes the immature NK-cell stage (20). NK-cell maturation can be induced by type 1 interferon in Vav-Bcl2 mice, which is blocked by the absence of Stat5 in Stat5^Δ/ΔMx1-Cre-Vav-Bcl2 mice (Fig. 1E). Thus, STAT5 is involved in NK-cell maturation; NK cells overexpressing Bcl2 do not mature but can be stimulated to do so. The impaired NK-cell maturation in Stat5^Δ/+Ncr1-iCre^Tg, Vav-Bcl2, and Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice is paralleled by significantly decreased expression of the transcription factor Id2 (Fig. 2A; Supplementary Fig. S2A). It is also characterized by reduced mRNA and protein levels of the T-box transcription factor family members Tbet and Eomes (Fig. 2A; Supplementary Fig. S2B–S2E), although the results with Tbet did not consistently reach statistical significance (Supplementary Fig. S2B–S2E).

STAT5 regulates genes involved in cell-cycle control, including c-MYC and D-type cyclins (21, 22). To investigate the requirement for STAT5 in NK-cell proliferation, we analyzed growth curves of purified NK cells maintained in IL2. Although Stat5^fl/fl and Vav-Bcl2 NK cells proliferate with comparable kinetics, Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells have a pronounced growth disadvantage (Fig. 2B). The reduced proliferative ability of STAT5-deficient NK cells was confirmed in vivo using BrdUrd incorporation (Fig. 2C). Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells displayed reduced expression of cyclin D2 (Ccnd2) and c-MYC (Myc), prominent regulators of growth and cell-cycle control (Fig. 2D). In contrast with proliferation, apoptosis did not appear to be affected by the absence of Stat5 in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells, as revealed by Annexin V/7-AAD staining (Supplementary Fig. S2F).

NK cells produce high levels of perforin, granzymes, and IFNγ (23). To test whether STAT5 is involved in the regulation of NK-cell cytotoxicity, we analyzed mRNA and protein levels of Perforin (Prf1), Granzyme A/B (Grzma/b), and IFNγ (Ifng). Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells consistently showed a significant reduction in the levels of all effector molecules (Fig. 2E; Supplementary Fig. S3A–S3C). In line with the reduced STAT5 activation found in nonactivated ex vivo–derived Vav-Bcl2 NK cells, we found a decrease in levels of the effector molecules (Fig. 2E; Supplementary Fig. S3A). In contrast, no differences were observed upon activation in interferon-treated cells (Supplementary Fig. S3B and S3C). In vitro cytotoxicity assays confirmed the physiologic significance of these observations: NK cells with diminished levels of STAT5 cannot efficiently lyse target cells (Fig. 2F). Supporting evidence was provided by adoptive transfer of IL2-expanded NK cells into Rag2^−/−gc^−/− mice: in the absence of STAT5, the NK cells were unable to suppress the growth of RMA-S cells (Supplementary Fig. S3D).

To study NK-cell–mediated tumor surveillance in vivo, we injected 5 × 10⁴ B16F10 melanoma cells into the tail vein of Stat5^fl/fl, Stat5^Δ/ΔNcr1-iCre^Tg, Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2, and Vav-Bcl2 mice. Survival (Fig. 3A) and tumor nodules in the lung (Fig. 3B) were analyzed in two independent experiments. Despite the presence of normal numbers of NK cells in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice (Fig. 1B), no efficient rejection of the melanoma cells was observed: The onset of disease in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice was as accelerated as in Stat5^Δ/ΔNcr1-iCre^Tg mice (Fig. 3A). The majority of Vav-Bcl2 mice were able to reject the B16F10 cells efficiently, possibly related to the higher numbers of NK cells (Fig. 1B). To analyze tumor burden in the lung, mice were again injected with 5 × 10⁴ B16F10 tumor cells and sacrificed after 23 days. Numbers of tumor nodules in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice were similar to those in Stat5^Δ/ΔNcr1-iCre^Tg mice (Fig. 3B). The presence of NK cells expressing high levels of Bcl2 but lacking STAT5 was not sufficient to control tumor growth (Fig. 3A and B).

NK cells also play a prominent role in the surveillance of hematopoietic tumors (24–26). We s.c. injected the hematopoietic cell line RMA-S and v-abl–transformed cells into Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice. Both experiments had to be terminated after about a week (day 6 for v-abl and day 9 for RMA-S cells), as tumors developed extremely rapidly in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice. The accelerated tumor growth in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice gave rise to a high tumor mass, whereas Stat5^Δ/ΔNcr1-iCre^Tg mice had small tumors at the same time point (Fig. 3C–E for v-abl, Supplementary Fig. S4A and S4B for RMA-S). We reasoned that at this early tumor stage, the residual “nondeleting” NK cells in Stat5^Δ/ΔNcr1-iCre^Tg mice were able to control tumor growth. Importantly, we did see significant differences between Stat5^fl/fl and Stat5^Δ/ΔNcr1-iCre^Tg mice under challenge with a higher number of cells when the tumor burden was examined at later stages, confirming the critical role of NK cells (Supplementary Fig. S4C). Despite the presence of equal numbers of NK cells in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 and Stat5^fl/fl mice, a tumor-promoting effect was only observed in the presence of STAT5-deficient NK cells; the effect could be reversed by depletion of NK cells (Fig. 3F; Supplementary Fig. S4D).

This result suggests that NK cells that survive the loss of STAT5 are able to accelerate tumor formation and that STAT5 acts as a molecular switch in NK cells, regulating tumor suppression and tumor promotion. Irrespective of the ability of NK cells to recognize and eradicate tumor cells, lowering STAT5 levels in NK cells will exacerbate tumor growth. We tested this interpretation in a tumor model that is not subject to immune surveillance by NK cells (25, 27). Upon s.c. injection, RMA cells induced tumor development to a comparable extent and with similar kinetics in Stat5^Δ/ΔNcr1-iCre^Tg, Stat5^fl/fl, and Vav-Bcl2 animals. Tumor growth was significantly accelerated in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice (Fig. 4A and B). To verify that the tumor-promoting potential resides within the Stat5-deficient NK cells, we depleted NK cells using an anti-NK1.1 antibody in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice and controls. NK-cell depletion abrogated the tumor-promoting potential, and Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 animals had a tumor burden comparable to that of Stat5^fl/fl mice (Fig. 4C). This clearly demonstrates that the NK cells themselves are responsible for the enhanced tumor progression.

The development of subcutaneous tumors relies on the formation of new blood vessels (28). CD31 is a marker for angiogenesis, and we found significantly more CD31⁺ cells in tumors derived from Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice (Supplementary Fig. S4E). VEGFA plays a key part in tumor angiogenesis. We detected VEGFA expression in NK cells under basal conditions. Experiments with IL2-expanded NK cells showed that Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells have significantly enhanced levels of Vegfa mRNA compared with Stat5^fl/fl and Vav-Bcl2 cells (Fig. 4D). Comparable results were obtained in Stat5^Δ/ΔMx1-Cre-Vav-Bcl2 NK cells: levels of Vegfa mRNA were significantly increased upon Stat5 deletion (Supplementary Fig. S5A). To test the physiologic significance of VEGFA produced by NK cells, we embedded endothelial cells in a 3-D collagen matrix and enriched it with the supernatant of in vitro–expanded NK cells. After 24 hours, we observed basal sprouting with the supernatant of Stat5^fl/fl and Vav-Bcl2 NK cells in parallel with basal VEGFA production, whereas the supernatant of Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 NK cells drastically increased the average sprout number as well as the cumulative sprouting length (Fig. 4E and F). A neutralizing VEGFA antibody prevented the increased sprouting (Supplementary Fig. S5B). The proangiogenic effect was confirmed in assays with aortic rings embedded in 3-D collagen matrices. Supernatants of Stat5^fl/fl and Vav-Bcl2 NK cells induced a certain level of sprouting, but addition of supernatant from Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2–derived NK cells caused significantly enhanced sprouting (Supplementary Fig. S5C). The supernatants of NK-cell cultures thus contain proangiogenic factors, which are produced in higher concentrations in the absence of STAT5.

We have shown that NK cells produce angiogenic factors that trigger endothelial cell sprouting in vitro and that the tumor-promoting effect observed in Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice correlates with increased tumor angiogenesis. To test the in vivo significance of our findings, we crossed Vegfa^fl/fl mice to Ncr1-iCre^Tg mice to produce Vegfa^Δ/ΔNcr1-iCre^Tg mice, in which the deletion of Vegfa is restricted to NKp46⁺ NK cells. The numbers of NK cells in the periphery (spleen, blood, or lymph nodes) were unaffected despite the deletion of Vegfa in splenic ex vivo–derived NKp46⁺ NK cells (Supplementary Fig. S6A and S6B). We found no consistent or significant changes in NK-cell proliferation or maturation (Supplementary Fig. S6C and S6D). To investigate the role of NK-cell–derived VEGFA in tumor development, we injected v-abl⁺ tumor cells s.c. into the flanks of the mice. Vegfa^Δ/ΔNcr1-iCre^Tg mice showed a significantly reduced tumor burden (Fig. 5A–C). The reduction in tumor angiogenesis was confirmed immunohistochemically: tumors of Vegfa^Δ/ΔNcr1-iCre^Tg mice had significantly fewer CD31⁺ cells (Fig. 5D). VEGFA-deficient NK cells were not able to efficiently induce sprouting of endothelial spheroids (Supplementary Fig. S6E). The reduced tumor burden was abolished by depletion of NK cells (Supplementary Fig. S6F and S6G).

To confirm the protumorigenic potential of NK-cell–derived VEGFA, we injected RMA-S cells s.c. into the flanks of mice (Supplementary Fig. S6H). In addition, we used a retrovirally induced leukemia model that triggers a slowly and oligoclonally evolving disease that closely reflects the situation in human patients. We injected Vegfa^fl/fl and Vegfa^Δ/ΔNcr1-iCre^Tg newborn mice with a replication-incompetent ecotropic retrovirus encoding v-abl. Vegfa^Δ/ΔNcr1-iCre^Tg mice displayed a significantly increased disease latency and survived longer than Vegfa^fl/fl controls (Fig. 5E). Consistent with the results in Fig. 3A and B, Vegfa^fl/fl and Vegfa^Δ/ΔNcr1-iCre^Tg mice showed no differences in tumor burden in the B16F10 melanoma model (Supplementary Fig. S6I). This probably stems from the fact that B16F10 melanoma forms multiple small nodules in the lung that require less angiogenesis. In addition, the lung is well oxygenized, reducing the need for angiogenesis in evolving small tumors.

Myeloid cells are a source of VEGFA in the tumor microenvironment (29, 30). We found that both myeloid cells and NK cells infiltrate v-abl⁺ tumors in comparable numbers between all genotypes (Fig. 5F; Supplementary Fig. S7A–S7C). Although both NK cells and myeloid cells are present around blood vessels, the results of experiments with VEGFA-deficient NK cells show the in vivo significance of NK-cell–derived VEGFA for hematopoietic tumors.

We have shown that loss or absence of STAT5 is associated with increased levels of VEGFA. To investigate how STAT5 interferes with the production of VEGFA, we stimulated ex vivo–derived NK cells with IL2 or IL2+IL15. Both cytokines strongly activate STAT5, accompanied by a decrease in Vegfa mRNA (Fig. 6A). STAT5 chromatin immunoprecipitation (ChIP) analysis of primary NK cells after stimulation showed an increase in STAT5 binding to the Vegfa promoter (Fig. 6B). To investigate which cytokines affect STAT5 activation in NK cells, we stimulated IL2-expanded WT NK cells with IL6, IL12, IL15, IL18, IL21, IL23, or IFNβ. Although IL6 and IL23 do not alter STAT5 activation, IL10, IL12, IL18, IL21, and IFNβ lead to a clear downregulation of STAT5^Y694 phosphorylation (Fig. 6C). In line with the decreased STAT5 activation, there is a clear increase in Vegfa mRNA levels in NK cells after stimulation with IL10, IL12, IL18, IL21, or IFNβ (Supplementary Fig. S8A). STAT5 occurs in two isoforms, STAT5A and STAT5B, which are encoded by adjacent genes and share at least 90% homology. They are generally considered to have highly redundant functions (31), but using NK cells purified from Stat5a and Stat5b knockout mice, we found that the amount of Vegfa mRNA was increased only in the absence of STAT5B (Fig. 6D).

Human peripheral blood NK cells are also able to produce VEGFA, with immature CD16⁻CD56^bright NK cells secreting higher levels than mature CD16⁺CD56⁺ NK cells (4). We FACS-sorted both NK-cell populations from healthy human donors (n= 8) and analyzed the levels of STAT5 and VEGFA. We found that CD16⁻CD56^bright NK cells contain high levels of VEGFA, accompanied by low levels of STAT5A and B and the downstream target BCL2 when compared with CD16⁺CD56⁺ NK cells (Fig. 6E; Supplementary Fig. S8B). Within the CD16⁻CD56^bright NK-cell population, we detected an inverse correlation between VEGFA and STAT5B, the relevant STAT5 isoform regulating VEGFA (Pearson correlation: R= 0.689, P= 0.003 comparing STAT5A and VEGFA and R = −0.547, P= 0.028 comparing STAT5B and VEGFA; Supplementary Fig. S8C). The data were confirmed by a scatter plot and a microarray expression analysis comparing resting and IL2-activated human NK cells (Fig. 6F; Supplementary Fig. S8D). Upon stimulation with IL2, human NK cells upregulate STAT5 target genes, such as EOMES, CCND2, and CISH, whereas the expression of VEGFA is reduced (Fig. 6F; Supplementary Fig. S8D).

A number of inhibitors of STAT5 and its upstream kinases are currently the subject of preclinical investigations and clinical trials. Treatment of murine IL2-expanded NK cells with a preclinical STAT5 inhibitor caused a significant upregulation of Vegfa expression and a clear downregulation of the STAT5 target genes Bcl2, Mcl1, and Cish (Supplementary Fig. S8E). Treatment of mouse NK cells with ruxolitinib, an inhibitor of the upstream kinases JAK1/2/3, significantly increased the levels of Vegfa mRNA (Fig. 7A). Treatment of FACS-sorted CD3⁻CD56⁺ peripheral blood mononuclear cells with 0.5 μmol/L ruxolitinib was sufficient to prevent activation of STAT5, as indicated by STAT5^Y694 phosphorylation (Fig. 7B). After a 3-hour treatment with the inhibitor, human peripheral blood NK cells expressed significantly higher levels of VEGFA (Fig. 7C), suggesting that ruxolitinib not only inhibits STAT5 signaling but also stimulates the secretion of proangiogenic factors. This finding is of direct relevance to patients. The effect is observed in vivo: Ruxolitinib treatment of Vegfa^fl/fl mice s.c. injected with v-abl⁺ tumor cells gave rise to an increased tumor burden that was not observed in Vegfa^Δ/ΔNcr1-iCre^Tg mice (Fig. 7D and E). This experiment confirms that ruxolitinib accelerates tumor progression by enhancing NK-cell–derived VEGFA expression.

We report that STAT5-deficient NK cells are rescued by overexpression of the antiapoptotic gene Bcl2. This finding enabled us to investigate the functions of STAT5 in NK cells. STAT5 is a key regulator of NK-cell maturation, proliferation, and cytotoxicity. Not only does STAT5 regulate the transcription of genes that have been identified as targets in other cell types (Myc, Ccnd2, or Prf1; ref. 32), but also it is involved in the regulation of the transcription factors Eomes and Id2. Vav-Bcl2 mice show reduced basal levels of STAT5 in nonactivated NK cells, together with an impairment of NK-cell maturation and reduced levels of effector molecules, whereas NK-cell proliferation and cytotoxicity are largely unimpaired. Stimulation with cytokines causes maturation, consistent with the idea that the delay in maturation is related to a decreased level of cytokine receptors in nonactivated Bcl2–transgenic NK cells (data not shown). Vav-Bcl2 mice are comparable with wild-type mice in many key functions, such as NK-cell tumor surveillance. In contrast, Stat5^Δ/ΔNcr1-iCre^Tg-Vav-Bcl2 mice that lack STAT5 in NK cells show severely hampered NK-cell function.

The most striking finding is that NK cells promote tumor formation in the absence of STAT5. We show that wild-type cytotoxic NK cells secrete significant levels of Vegfa and that transcription of Vegfa is suppressed by STAT5, especially by STAT5B. Loss of STAT5 causes increased production of Vegfa in two mouse models. The increased levels of Vegfa promote tumor formation. Experiments with conditional knockout mice have confirmed the significance of the protumorigenic role of NK-cell–derived Vegfa: Loss of NK-cell–derived Vegfa decreases tumor burden while increasing disease latency.

Immune cells such as tumor-associated macrophages promote tumor angiogenesis and lymphangiogenesis (29, 30). However, tumor-infiltrating NKp46⁺ NK cells are also found in the surroundings of blood vessels, and VEGFA produced by NK cells enhances the formation of lymphoid tumors. Infiltrating immature NK cells that secrete VEGFA have been detected in patients with non–small cell lung cancer, breast tumors, and colon tumors (33, 34). The presence of VEGFA-producing NK cells is associated with a poor prognosis, showing that NK-cell–derived VEGFA plays a part in human disease. Which cells are primarily responsible for producing VEGFA in a particular context presumably depends on the type and stage of the tumor. It is currently unclear whether tumor-associated NK cells and macrophages interact to promote tumor growth and angiogenesis. It will be of great interest to test whether NK cells contribute to the polarization of macrophages.

Depending on the context, NK cells may have either cytotoxic or angiogenic functions. We propose that STAT5 regulates the switch between the two effects. There is ample evidence that tumor-infiltrating NK cells have low cytotoxic potential. It has been speculated that treatment with IL2 and IL15 reactivates noncytotoxic NK cells (35). Our data support the idea: These cytokines signal via STAT5 and may help to revert tumor-infiltrating but exhausted NK cells into killers that produce perforin and IFNγ. Decidual NK cells have previously been shown to possess two types of activity: They normally produce significant amounts of VEGFA, but stimulation with IL15 switches them to cytotoxic cells (36). We find that cytokines present in the tumor microenvironment, such as IL10, IL12, IL18, IL21, and IFNβ, decrease the activity of STAT5 (reflected by reduced phosphorylation of STAT5^Y694) while enhancing expression of VEGFA. These results are consistent with the notion that cytokine-induced STAT5 regulation acts to convert NK cells from cytotoxic killers to tumor promoters.

The development of STAT5 inhibitors is currently at the preclinical stage. To the best of our knowledge, no specific STAT5 inhibitor has entered clinical trials. Nevertheless, several inhibitors of the upstream JAK kinases have been approved by the FDA, e.g., for the treatment of rheumatoid arthritis, and are currently being tested in a wide variety of clinical studies. In patients with advanced myelofibrosis, the JAK1/2/3 inhibitor ruxolitinib is of clinical benefit. It reduces the tumor burden associated with a reduction in the levels of cytokines, including VEGFA, which is most likely produced by the tumor cells themselves (37). Previous work has shown that long-term ruxolitinib treatment impairs the maturation and cytolytic functions of NK cells. Patients with myeloproliferative neoplasm treated with the inhibitor suffer from recurrent infections (38). We now extend these findings by showing that ruxolitinib significantly increases the amount of VEGFA in NK cells. Ruxolitinib treatment of mice injected with v-abl⁺ lymphoma cells caused a significantly enhanced tumor burden that depends on NK-cell–derived VEGFA. This NK-cell–mediated effect is presumably important in cases of minimal residual disease controlled by NK cells or in other situations where limited tumor burden is controlled by the immune system, as described in the Elimination, Equilibrium, and Escape model (39). Under these circumstances, NK cells keep the tumor in check, and there is a delicate balance between tumor and immune cells. The balance can be disrupted by the presence of factors that suppress STAT5, such as cytokines or JAK–STAT inhibitors.

We conclude that treatment with inhibitors of STAT5 or its upstream kinase not only hampers the cytotoxic function of NK cells but might also induce Vegfa expression, thereby worsening the prognosis. NK cells have also been shown to inhibit angiogenesis by secreting IFNγ (40). The reduced release of IFNγ in STAT5-deficient NK cells may contribute to angiogenesis, although the molecular details remain to be elucidated. It is becoming clear that NK cells are not mere killing machines but complex players with tightly controlled effector responses. These need to be thoroughly investigated before we can predict with confidence the outcome of STAT5-directed therapy. In the meantime, it is important to monitor very closely the tumor development in patients undergoing clinical trials of JAK–STAT inhibitors, with complementary anti-VEGF therapy indicated if tumor growth is accelerated.

Mice were bred on C57BL/6 background and maintained at the University of Veterinary Medicine Vienna under pathogen-free conditions. The following mice were studied: Vegfa^fl/fl, Vegfa^Δ/ΔNcr1-iCre^Tg, Stat5^fl/flMx1-Cre [inducible Cre expression after Type 1 interferon response, e.g., after Poly(I:C) injection], Stat5^fl/flMx1-Cre-Vav-Bcl2, Stat5^Δ/+Ncr1-iCre^Tg, Stat5^Δ/ΔNcr1-iCre^Tg, Stat5^Δ/ΔNcr1-iCre^Tg−Vav-Bcl2, Stat5^fl/+, Stat5^fl/fl, and Vav-Bcl2 littermates (8, 41–43). B10;B6-Rag2^tm1Fwa II2rg^tm1Wjl mice (Taconic) were used for adoptive transfer experiments.

All experiments were carried out with age-matched 6-to-12-week-old mice and were approved by the institutional animal care committee and review board conforming to Austrian law (license 66.009/0019-II/10b/2010, 14.1.10,68.205/0218-II/3b/2012, and BMWFW-68.205/0103-WF/V/3b/2015). Stat5a^−/− and Stat5b^−/− (44, 45) were backcrossed with C57BL/6 once, and then littermates were crossed for this study.

Splenic NK cells were isolated using the MACS-positive selection kit (DX5; Miltenyi) and cultured with 5,000 U/mL rhIL2 (Proleukin; Roche). NK-cell stimulations were performed in the presence of 5,000 U/mL rhIL2 ± 5 ng/mL rmIL12 (R&D), 50 ng/mL rmIL15 (PeproTech), 100 U/mL rmIFNβ (Sigma), 5 ng/mL rmIL10 (R&D), 100 ng/mL rmIL21 (Immunotools), 50 ng/mL rmIL23 (R&D), 100 ng/mL rmIL18 (R&D), 50 ng/mL rmIL6 (R&D), 50 ng/mL rmIL6Rα (R&D), or 10 ng/mL Phorbol-12-myristat-13-acetat (PMA; Sigma)/250 ng/mL ionomycin (Sigma). For in vitro activation of Cre recombinase in NK cells expressing Mx1Cre, IL2-expanded NK cells were treated with 1,000 U/mL rmIFNβ (pbl Assay Science).

For inhibitor studies, splenic mouse or human peripheral blood NK cells were treated for 3 hours with 0.5/1 μmol/L ruxolitinib (Chemietek) or 4 hours with a small-molecule STAT5 inhibitor (46), and DMSO was used as vehicle control.

RMA and RMA-S cell lines were kindly provided by A. Cerewenka and authenticated by flow cytometry. V-abl⁺ cell lines were generated in the laboratory of Professor V. Sexl. All lines were tested for the absence of Mycoplasma by PCR and surface marker expression by flow cytometry every 6 months (last authentication: November 2015).

Murine endothelial cells were suspended in 80% endothelial cell growth medium (RPMI containing 10% FCS), mixed 1:1 with Endothelial Cell Growth Medium MV (PromoCell) and 20% methylcellulose (Sigma), and seeded as drops (800 cells/100 μL) in nonadherent dishes. The dishes were incubated upside down as hanging drops for 24 hours. Under these conditions, all suspended cells contribute to the formation of a single spheroid per drop of defined size and cell number. Spheroids were harvested and seeded in a collagen matrix, and supernatant of IL2-expanded NK cells or 30 ng/mL mVEGFA (PeproTech) was added. Two μg/mL LEAF purified anti-mouse VEGFA Antibody (2G11-2A05; BioLegend) was added for the neutralization experiments. After 24-hour incubation, spheroids were scanned and photographed with an Olympus IX71 microscope using cellSens Dimension Software (Olympus). The sprout number and the cumulative length of the sprouts from each spheroid were calculated with ImageJ software.

Mice were treated three times by i.p. injection of 200 μg Poly(I:C) (InvivoGen) over 14 days. Seven days after the last injection, mice were analyzed. Activation of the Cre recombinase and deletion of Stat5 resulted in a global Stat5 deletion in Stat5^fl/flMx1-Cre and Stat5^fl/flMx1-Cre-Vav-Bcl2, including the entire hematopoietic compartment, and are denoted as Stat5^Δ/ΔMx1-Cre and Stat5^Δ/ΔMx1-Cre-Vav-Bcl2.

In vitro cytotoxicity assays were performed as previously published (25, 47).

Mice were treated two times by i.p. injection of 100 μg PK136 (anti-NK1.1.) antibody before s.c. injection of tumor cells and three times during the tumor progression. Successful deletion of CD3⁻NK1.1⁺NKp46⁺ NK cells was checked in the blood of treated animals by flow cytometry.

In the B16F10 melanoma model, mice were injected i.v. with 5 × 10⁴ B16F10 cells. After 23 days, lungs were analyzed. For survival assays, mice were sacrificed at first signs of dyspnea and health detractions.

In the v-abl, RMA and RMA-S tumor model, 5 × 10⁵ to 5 × 10⁶ tumor cells were injected s.c. into the flanks of the mice, and the health status was controlled daily. After 6 to 12 days, mice were sacrificed and tumor weight was determined. Tumor diameters were measured with a caliper, and the tumor volume in mm³ was calculated by the formula: volume = (width)²× length/2.

In the A-MuLV model, newborn mice were injected with 100 μL of replication-incompetent ecotropic retrovirus encoding for v-abl by s.c. injection as described previously (48). Mice were checked daily for disease onset.

Ruxolitinib (Chemietek) was dissolved to make a stock solution in DMSO and further diluted for oral gavage in PBS containing 0.5% methylcellulose (w/v) and 0.1% Tween 80. Mice were treated continuously once a day after tumor cell injection by oral gavage with 95 mg/kg ruxolitinib.

The following antibodies (clones) were purchased from BD Biosciences: CD3ϵ(145-2C11), CD3(UCHT1), CD56(B159), CD16(3G8), CD11b(M1/70), CD27(LG.3A10), IFNγ(XMG1.2) or purchased from eBioscience/Affymetrix: CD3ϵ(17A2, 145-2C11), CD49b(DX5), NK1.1(PK136), NKp46(29A1.4), Granzyme B(NGZB), T-bet(4B10), Eomes(Dan11mag), and Perforin (eBioOMAK-D). Granzyme A(3G8.5) was purchased from Santa Cruz Biotechnology. Liver and lung were perfused, and liver lymphocytes were obtained by a Percoll (GE Healthcare) gradient. Lung tissue was Collagenase/DNaseI (Sigma) digested prior to FACS staining. Bones were smashed with a mortar and filtered through a nylon mesh to obtain single-cell suspension. Whole blood and splenocytes were depleted of erythrocytes, and anti-CD16/CD32 (eBioscience) was added to all samples prior to staining. Intracellular stainings were performed with the Foxp3/Transcription factor staining buffer set (eBioscience) after application of a fixable viability dye (eBioscience). BrdUrd (Sigma) was injected i.p. 16 hours prior to analysis, and incorporation was checked using an anti-BrdUrd antibody (BU20A; eBioscience) and DNase1 (Sigma) digestion according to eBioscience's protocol. Apoptosis stainings with IL2-expanded NK cells were performed with the Annexin V Apoptosis Detection Kit (eBioscience).

All samples were recorded on a FACS Canto II flow cytometer or sorted on a FACS Aria III (BD Biosciences) and analyzed with BD FACS Diva software version 6.1.2 and 7.0.

Blood of healthy donors was diluted 1:2 with 1× PBS, and lymphocytes were isolated using LymphoPrep (Axis Shield). CD3⁻CD56⁺, CD3⁻CD16⁻CD56^bright, and CD3⁻CD16⁺CD56⁺ cells were sorted on a FACS AriaIII (BD), and RNA was isolated by the RNeasy Micro Kit (Qiagen).

MACS-purified and IL2 cultured NK cells (10⁷) were either left untreated or stimulated with IL15 for 30 minutes followed by cross-linking using 1% formaldehyde for 10 minutes at 37°C. The reaction was stopped by the addition of 0.5 mol/L glycine for 5 minutes. Cell nuclei were prepared and lysed in 1 mL of lysis buffer on 4°C overnight (o/n). Chromatin was sheared by sonication yielding chromatin fragments between 200 and 500 bp and diluted 2.5-fold in ChIP dilution buffer. Immunoprecipitations were performed at 4°C o/n with an anti-STAT5 antibody (C-17, sc-835; Santa Cruz Biotechnology). Chromatin was precleared using 25 μL salmon sperm DNA/protein A-agarose beads and incubated with the antibody o/n. Immune complexes were collected with 25 μL beads for 3 hours and washed with RIPA buffer, high salt buffer, LiCl buffer, and TE buffer. Samples were eluted twice in elution buffer (2% SDS, 10 mmol/L DTT, and 100 mmol/L NaHCO₃). DNA cross-linking was reversed by heating at 65°C o/n followed by proteinase K digestion. DNA was extracted with phenol-chloroform, precipitated in isopropanol, and resuspended in TE Buffer. The predicted binding site was obtained using Ensemble for the sequence of the Vegfa promoter and blasting for the potential STAT binding motif TTC(N)_2-4GAA. Obtained DNA fragments were analyzed by qPCR using the following primer pairs: Vegfa: Fw: 5′-GCATGCATGTGTGTGTGTGT-3′ and Rev: 5′-GGCAGGGACGTATGAGGATA-3′, Cish: Fw: 5′-CGCGCTGCTATTGGCCCTCCC-3′ and Rev: 5′-GTCTGGGGCCCTGAGCAGTG-3′, CD19^down: Fw: 5′-CCCTCTTCTCATTCGTTTTCCA-3′ and Rev: 5′-CCAGGAAAGAATTTGAGAAAAATCA-3′. N-fold enrichment was calculated relative to a negative region downstream of the CD19 gene (CD19^down).

Tumors were either paraformaldehyde-fixed and paraffin-embedded or frozen in optimal cutting temperature compound (OCT). Sections (3 μm) were stained with hematoxylin/eosin according to standard histologic procedures. CD31 (Cell Signaling; #9654S) stainings were performed as previously reported (49). The stainings were scanned and photographed with an Olympus IX71 or a Nikon Eclipse E1000 microscope using cellSens Dimension (Olympus) or NIS-elements (Nikon) software. Stainings of CD31 and NKp46 (29A1.4; BioLegend) and CD31 and F4/80 (CI:A3-1; ABD Serotec) were performed with frozen tumor sections as previously reported (50) and scanned with a LSM 5 Exciter (Zeiss) or a Nikon Eclipse E1000 microscope. Tumors were photographed and counted at random areas by three individual researches in a blinded manner.

Data were extracted from ArrayExpress dataset E-GEOD-8059 and E-GEOD-50838 (Affymetrix Human Gene 1.1 ST Arrays). The heatmap was generated using ClustVis software (51).

Student t test, one-way ANOVA, Tukey post-hoc test, and Mantel–Cox log-rank tests (survival curves) were performed using GraphPad Prism Software version 5.04 and 6.02. For multiple comparison of survival curves, Bonferroni correction was applied. Statistical analysis is indicated for each experiment specifically (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P ≤ 0.0001).

No potential conflicts of interest were disclosed.

Conception and design: D. Gotthardt, Veronika Sexl

Development of methodology: D. Gotthardt, E.M. Putz, R. Moriggl

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Gotthardt, E. Grundschober, E. Straka, P. Kudweis, Z. Bago-Horvath, A. Witalisz-Siepracka, A.A. Cumaraswamy, P.T. Gunning, C. Stockmann

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Gotthardt, E.M. Putz, M. Prchal-Murphy, E. Straka, G. Heller, Z. Bago-Horvath, C. Stockmann, V. Sexl

Writing, review, and/or revision of the manuscript: D. Gotthardt, E.M. Putz, E. Grundschober, Z. Bago-Horvath, A. Witalisz-Siepracka, A.A. Cumaraswamy, P.T. Gunning, B. Strobl, M. Müller, C. Stockmann, V. Sexl

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Kudweis, B. Strobl, C. Stockmann, V. Sexl

Study supervision: M. Müller, V. Sexl

The authors thank L. Edlinger, S. Fajmann, P. Jodl, H. Schachner, and G. Asfour for all their help, G. Tebb for careful reading and revision of the manuscript, and C. Moschner for helping with graphical illustrations of the sprouting assay. They are also grateful to the mouse facility.

The authors were supported by the Austrian Science Fund FWF (grants SFB F28 and SFB F47) and the PhD program “Inflammation and Immunity” FWF W1212.

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.