Phosphodiesterase 4A confers resistance to PGE2-mediated suppression in CD25+ /CD54+ NK cells.

Inadequate persistence of tumor-infiltrating natural killer (NK) cells is associated with poor prognosis in cancer patients. The solid tumor microenvironment is characterized by the presence of immunosuppressive factors, including prostaglandin E2 (PGE2), that limit NK cell persistence. Here, we investigate if the modulation of the cytokine environment in lung cancer with IL-2 or IL-15 renders NK cells resistant to suppression by PGE2. Analyzing Cancer Genome Atlas (TCGA) data, we found that high NK cell gene signatures correlate with significantly improved overall survival in patients with high levels of the prostaglandin E synthase (PTGES). In vitro, IL-15, in contrast to IL-2, enriches for CD25+ /CD54+ NK cells with superior mTOR activity and increased expression of the cAMP hydrolyzing enzyme phosphodiesterase 4A (PDE4A). Consequently, this distinct population of NK cells maintains their function in the presence of PGE2 and shows an increased ability to infiltrate lung adenocarcinoma tumors in vitro and in vivo. Thus, strategies to enrich CD25+ /CD54+ NK cells for adoptive cell therapy should be considered.

Introduction

Due to their important role in immune surveillance and eradication of tumor cells, natural killer (NK) cells are increasingly used in cancer immunotherapy. Early studies demonstrate that allogeneic NK cells contribute to graft‐versus‐leukemia effects in the setting of hematopoietic stem cell transplantation (Passweg et al, 2004; Ruggeri et al, 2008). More recent studies provide evidence of therapeutic benefit with low toxicity profiles upon infusion of donor alloreactive haploidentical NK cells in patients with acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS) (Shaffer et al, 2016; Dolstra et al, 2017; Bjorklund et al, 2018). However, infusion of NK cells in patients with solid tumors has yet to result in beneficial clinical responses. Several studies have highlighted that inadequate persistence and dysfunctionality of the infused NK cells represents rate‐limiting factors for successful clinical outcome (Bachanova et al, 2014; Liu et al, 2018).

The microenvironment of solid tumors is a complex network of stromal cells consisting of several different cell types and immunosuppressive factors. In addition, specialized subsets of immunosuppressive myeloid and lymphoid cells including tumor‐associated macrophages, myeloid‐derived suppressor cells (MDSC), and regulatory T cells are often found in abundance within the stroma of solid tumors (Woo et al, 2001; Gabrilovich & Nagaraj, 2009; Solinas et al, 2010). Such cells and tumor cells themselves produce a variety of factors that inhibit NK cell activity (Barrow & Colonna, 2017). The arachidonic acid pathway is important in regulating inflammatory responses. Arachidonic acid is catalyzed by the cyclooxygenases (COX‐1/2) into the reactive intermediate prostaglandin (PG) H2 which is further processed into different prostaglandins, including PGE2, by several terminal synthases. Early studies show that melanoma‐derived fibroblasts produce PGE2 to downregulate NK cell activating receptors and thereby inhibit the function of NK cells (Balsamo et al, 2009). More recent studies show that also other cell populations including mesenchymal stromal cells are able to inhibit NK cell function through the secretion of PGE2 (Reinders & Hoogduijn, 2014; Hu et al, 2019). We and others have shown that tumor cells produce PGE2 to suppress NK cell activity via the Prostaglandin E2 receptor 2 (EP2) and EP4 receptors (Holt et al, 2011; Wennerberg et al, 2014; Li et al, 2016). Thus, strategies to render NK cells less susceptible to PGE2 to increase the persistence and activity of NK cells are needed to improve clinical responses of NK cell adoptive cell therapy in patients with solid tumors.

Interleukin‐2 (IL‐2) and IL‐15 are type I cytokines that are commonly used to activate and expand NK cells for cellular therapy. Both cytokines activate NK cells via shared beta and gamma receptors but use distinct alpha receptors to transmit signaling (Castro et al, 2011; Qiao & Fu, 2020). Although IL‐2/IL‐15 receptor complexes activate similar JAK/STAT signal transduction cascades, they display distinct activities in vivo where IL‐2 preferentially expands regulatory T cells and CD4+ helper T cells and IL‐15 supports the development of central memory T cells and NK cells (Zhang et al, 1998; Kennedy et al, 2000; Castro et al, 2011; Liao et al, 2011; Conlon et al, 2015). While activation with IL‐15 or IL‐2 results in similar steady‐state mRNA profiles in murine CD8+ T cells (Ring et al, 2012), we recently showed that IL‐15 activates mTOR and primes stress‐activated gene expression programs in human NK cells (Mao et al, 2016). Consequently, infusion of IL‐15 primed NK cells show a longer in vivo persistence compared with IL‐2 primed NK cells.

In this study, the ability of IL‐2 and IL‐15 to modulate the susceptibility of NK cell to PGE2‐mediated suppression was investigated. Collectively, we found that IL‐15 activates NK cells to become less susceptible to PGE2‐mediated suppression compared with IL‐2. Mechanistically, our data demonstrate that IL‐15 enriches for CD25+/CD54+ NK cells and this population show an mTOR‐dependent upregulation of the cAMP hydrolyzing enzyme phosphodiesterase 4A (PDE4A). This population of NK cells shows increased ability to form cell clusters and kill tumor targets in the presence of PGE2 and to infiltrate lung adenocarcinoma tumor spheroids and reduce tumor burden in zebrafish larva xenograft model. We furthermore show that high expression of PTGES, encoding for the enzyme responsible for PGE2 biosynthesis, affects the prognostic value of NK cells in patients with lung adenocarcinoma. Collectively, our data position CD25+/CD54+ NK cells as a key subpopulation involved in the persistence within an immunosuppressive tumor microenvironment.

Results

Expression of PDE4A renders human NK cells less susceptible to PGE2‐mediated suppression

Since PGE2 is abundant within the microenvironment of solid tumors and that IL‐15 enhances NK cell persistence in vivo, we sought to investigate if NK cells activated by IL‐15 or IL‐2 differ in their sensitivity to PGE2‐mediated suppression. Purified NK cells from healthy individuals were activated with either IL‐2 or IL‐15 in the presence of PGE2. While exposure to PGE2 inhibited the proliferation and cytotoxicity of IL‐2 activated NK cells, IL‐15 activated NK cells maintained their proliferation and cytotoxicity in the presence of PGE2. While both proliferation and cytotoxicity were significantly higher in IL‐15 stimulated NK cell compared with IL‐2 stimulated NK cells in the presence of PGE2, proliferation but not cytotoxicity was significantly higher in IL‐15 stimulated NK cells in the absence of PGE2. Yet, the difference between IL‐2 and IL‐15 stimulated NK cells in the presence of PGE2 was greater than in the absence of PGE2 (Fig 1A and B). PGE2 did not affect the viability of either IL‐2 or IL‐15 activated NK cells Fig 1C).

Figure 1

Expression of PDE4A renders human NK cells less susceptible to PGE2‐mediated suppression

Purified NK cells were activated with IL‐2 or IL‐15 in the presence or absence of PGE2 (1 µM) and tested for (A) proliferation by flow cytometry for Ki‐67 staining after 4 days of culture (n = 6, biological replicates), (B) ability to kill K562 cells after 2 days of culture in a 4‐h chromium assay at an E:T ratio of 5:1 (n = 4, biological replicates). (C) Viability by flow cytometry using fixable live/dead staining after 2 days of culture (n = 7, biological replicates).

Intracellular cAMP detected in 2‐day IL‐2 or IL‐15 activated NK cells before and 5 min after the addition of PGE2 (1 µM; n = 3, biological replicates).

mRNA levels of selected PGE2 response elements as measured by qPCR in NK cells (n = 3–4, biological replicates).

Fold change comparing flow cytometry mean fluorescence intensity (MFI) values of PDE4A expression between IL‐2 and IL‐15 activated NK cells after 2 days activation with cytokines (n = 5, biological replicates).

Representative images showing PDE4A expression in NK cells activated by IL‐2 or IL‐15 for 2 days. Scale bar: 10 μm.

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01, ***P < 0.001). Each circle represents an individual experiment. Error bars are depicted as standard deviation.

Through the binding to the EP2 and EP4 receptors, PGE2 cause an increase in the production of intracellular cAMP leading to NK cell dysfunction (Holt et al, 2011). Hence, the concentration of intracellular cAMP was analyzed in IL‐2 and IL‐15 activated NK cells. Compared with IL‐15, IL‐2 activated NK cells showed significantly higher intracellular cAMP levels after short‐term exposure to PGE2 (Fig 1D). To investigate the underlying mechanisms of the reduced susceptibility of IL‐15 activated NK cells to PGE2, the expression of EP2 (PTGER2) and EP4 (PTGER4) receptors as well as the expression of cAMP hydrolyzing enzyme family of phosphodiesterases was analyzed. While the expression of PTGER2 and PTGER4 remained at similar levels in IL‐2 and IL‐15 activated NK cells, the expression of phosphodiesterases, and in particular PDE4A, was significantly higher following exposure to IL‐15 (Fig 1E). The expression of PDE4A protein was confirmed to be significantly upregulated by IL‐15 by flow cytometry (Fig 1F) and by immunofluorescence microscopy (Fig 1G). Selective inhibition of PDE4 significantly increased the levels of intracellular cAMP in IL‐15 activated NK cells (Fig EV1A). Taken together, our results show that IL‐15 activated NK cells are less susceptible to PGE2‐mediated suppression due to high expression of PDE4A accompanied by reduced intracellular levels of cAMP.

Figure EV1

Expression of intracellular proteins in IL‐2 and IL‐15 activated NK cells

Intracellular levels of cAMP in IL‐15 activated NK cells in the presence or absence of PDE4 inhibitor (PDE4i, Roflumilast) added at 0.1, 1, or 10 µM together with PGE2 1 µM.

Percentage of category 1 (pS6+pSTAT3+pSTAT5+pAKT+) NK cells under indicated conditions (n = 4, biological replicates).

Phosphorylation of S6 analyzed by flow cytometry in IL‐2, IL‐15, and IL‐15+ Torin‐1 treated NK cells (n = 4, biological replicates).

Representative images showing PDE4A expression in 2‐day IL‐15 activated NK cells in the presence or absence of Torin‐1 (1 μM). Scale bar: 10 μm.

Purified NK cells were stimulated with either IL‐2 or IL‐15 in the presence or absence of PGE2 and analyzed for pSTAT5 expression. Left: frequency of pSTAT5+ NK cells, right: median fluorescence intensity of pSTAT5 (n = 4, biological replicates).

Potential transcription factors that bind to the PDE4A region using CistromeDB.

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01). Each circle represents an individual experiment. Error bars are depicted as standard deviation.

Inhibition of mTOR reduces PDE4A expression and sensitizes NK cells to PGE2‐mediated suppression

Since PDE4 is linked with mTOR activity and that mTOR plays an essential role in the activation of NK cells (Mao et al, 2016; Li et al, 2019), subpopulations of NK cells based on the expression of intracellular signaling molecules, including phosphorylation of the S6 kinase (pS6), were analyzed with regard to susceptibility to PGE2‐mediated suppression. The frequency of pS6‐positive NK cells with elevated phosphorylation of AKT, STAT5, and STAT3 did not change in IL‐15 activated NK cells upon exposure to PGE2 (category 1, Fig 2A and B). In contrast, the frequency of this population of NK cells was significantly reduced in IL‐2 activated NK cells upon exposure to PGE2 (Fig EV1B). The frequency of pS6 single positive NK cells was reduced by 0.652 ± 0.094 in the presence of PGE2 in IL‐2 activated NK cells (P = 0.0051), while the frequency of pS6‐positive IL‐15 activated NK cells remained unchanged (Fig 2C).

Figure 2

Inhibition of mTOR reduces PDE4A expression and sensitizes NK cells to PGE2‐mediated suppression

NK cells were activated with IL‐2 or IL‐15 for 2 days in the presence or absence of PGE2 (1 µM) and analyzed for A) the levels of phosphorylated S6 (pS235/pS236), STAT3 (S727), STAT5 (Y694), Akt (S473; n = 4, biological replicates) and (B) 16 categories of phosphorylation array showed. Red box indicates category 1 (pS6+, pSTAT3+, pSTAT5+, pAKT+).

Relative changes in frequency of NK cells expressing pS6 in the presence and absence of PGE2 (n = 4, biological replicates).

Flow cytometry analysis of PDE4A in the presence or absence of Torin‐1 (1 µM; left, n = 7, biological replicates) with representative histograms (right).

NK cell‐mediated cytotoxicity of K562 cells by 4‐h chromium‐release assay of 2‐day cytokine‐activated NK cells in the presence or absence of Torin‐1 (1 µM) and PGE2 (1 µM; n = 6, biological replicates). E:T ratio = 5:1.

(F) Flow cytometry analysis of pS6 and (G) 4‐h chromium‐release assay of PGE2‐treated NK cells against K562 measured after 48 h treatment with Roflumilast (PDE4i, 1 µM; n = 3, biological replicates) E:T ratio = 5:1.

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01). Each circle represents an individual experiment. Error bars are depicted as standard deviation.

To confirm the link between mTOR and PDE4 activity, the expression of pS6 and PDE4A was analyzed in NK cells in the presence or absence of the PDE4 inhibitor Roflumilast or the mTOR inhibitor Torin‐1. Torin‐1 completely reduced the frequency of pS6‐positive NK (Fig EV1C), and the expression of PDE4A was significantly reduced in IL‐15 activated NK cells in the presence of Torin‐1 (Figs 2D and EV1D). To test whether the inhibition of mTOR would sensitize IL‐15 activated NK cells to PGE2‐mediated suppression, NK cell cytotoxicity against K562 target cells was evaluated in the presence of Torin‐1 and PGE2. Indeed, the cytolytic capacity of IL‐15 activated NK cells was significantly reduced upon exposure to PGE2 in the presence of Torin‐1 (P = 0.0246, Fig 2E). Although the frequency of NK cells positive for pSTAT5 did not differ between IL‐2 and IL‐15 stimulated NK cells, IL‐15 stimulation resulted in higher intensity staining for pSTAT5 (Fig EV1E). To further strengthen the association between IL‐15 and PDE4A, transcription factors that bind to the PDE4A region (chr19:10,416,773–10,469,631/hg38) were analyzed by using CistromeDB toolkit. This analysis revealed several transcription factors including STAT5 to bind to the PDE4A region suggesting that IL‐15 induced PDE4A expression can be regulated via JAK/STAT5 signaling (Fig EV1F).

In the presence of PGE2, inhibition of PDE4 significantly reduced the frequency of pS6‐positive IL‐15 activated NK cells (Fig 2F). In addition, inhibition of PDE4 resulted in reduced killing of K562 cell by IL‐15 NK cells (Fig 2G). Collectively, these results show a reciprocal cross‐talk between mTOR and PDE4 activity in IL‐15 activated NK cells and inhibition of either mTOR or PDE4 activity sensitizes IL‐15 activated NK cells to PGE2‐mediated suppression.

IL‐15 induces phenotypical changes and supports the formation of cell clustering

To comprehensively investigate differences between IL‐2 and IL‐15 activated NK cells, a previously published transcriptomic data set was analyzed (Mao et al, 2016). When enriched for immune system processes, 10 genes out of 244 differentially expressed genes (fold change > 1.5) was significantly enriched in “Positive regulation of leukocyte migration” pathway (Figs 3A and EV2A).When analyzed for the differential expression in biology process in particular for cell surface molecules (cell surface, GO:0009986) and cellular location, 27 genes were found to be differentially expressed (Fold change > 1.5). Among these, ICAM1 (CD54) and IL2‐RA (CD25) were significantly higher expressed by IL‐15 activated NK cells (Figs 3B and EV2B).

Figure 3

IL‐15 induces phenotypical changes and supports the formation of cell clustering

Gene ontology enrichment analysis showing immune system process (GO:0002376) pathways for genes with > 1.5 fold change expression between IL‐15 and IL‐2. The color of the nodes is determined by the significance of the enriched term assessed by two‐sided hypergeometric test with Bonferroni's step‐down method for P value correction.

Volcano plot showing genes compared after 2 days cytokine‐activated NK cells. Genes with significant changes (P value < 0.05) and > 1.5 log2 fold change between IL‐2 and IL‐15 activated NK cells are labeled in red. Genes belonging to cell surface (GO:0009986) are annotated with gene name.

Quantification of CD25+ CD54+ population changes after 2 days treatment (n = 4, biological replicates).

Representative flow cytometry graphs for CD25 and CD54 staining in IL‐2 and IL‐15 NK cells in the presence or absence of PGE2.

Quantification of the number of cell clusters. NK cell clusters were defined as cell aggregates occupying an area at least 2,000 μm2 and an eccentricity of < 0.8). (n = 4, biological replicates).

Representative phase contrast image of IL‐2 and IL‐15 activated NK cells in the presence or absence of PGE2. Scale bar 50 µm.

Chromium‐release cytotoxicity assay by 2‐day cytokine‐activated NK cells in the presence or absence of PGE2 (1 µM) and anti‐CD54 (20 µg/ml; n = 4, biological replicates), E:T ratio = 5:1.

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01, ***P < 0.001). Each circle represents an individual experiment. Error bars are depicted as standard deviation.

Figure EV2

Gene Ontology enrichment analysis and cellular location for significant changed genes

Gene Ontology enrichment on GO:0002376 for significantly changed genes (FC > 1.5, n = 244). Node colors are coded with immune related pathways. Each node is enriched with more than three genes in each respective pathway.

Overview for significant changed molecules in Fig 3B with cellular location, the interaction and involvement. Dashed lines separate layers with annotated cellular location information on the right.

Flow cytometry analysis confirmed the increased expression of CD25 and CD54 and increased frequency of CD25+/CD54+ NK cells upon activation with IL‐15 compared with activation with IL‐2. Although not significant, the frequency of CD25+/CD54+ NK cells were reduced upon exposure to PGE2 in both IL‐2 and IL‐15 activated NK cells (Fig 3C and D). Similarly, the expression of CD25 and CD54 was reduced in both IL‐2 and IL‐15 activated NK cells upon exposure to PGE2. Though the expression of CD54 was not significantly reduced in IL‐15 NK cells upon exposure to PGE2. Still, the frequency of CD25+/CD54+ NK cells and the expression of CD25 and CD54 were still significantly higher in IL‐15 activated NK cells compared with IL‐2 activated NK cells in the presence of PGE2 (Figs 3C and EV3A). No difference in the expression levels of CD56, CD57, CD16, NKG2A, DNAM‐1, and NKp46 was observed between IL‐2 and IL‐15 activated NK cells. However, the expression of the activation markers CD69 and NKG2D was significantly higher in IL‐15 compared with IL‐2 activated NK cells (Fig EV3B).

Figure EV3

NK cell phenotype upon exposure to PGE2 in the presence or absence of anti‐CD54 and Torin‐1

Mean fluorescence intensity expression of CD54 and CD25 by IL‐2 or IL‐15 activated NK cells in the presence or absence of PGE2 (n = 5, biological replicates).

Flow cytometry analysis of cell surface markers by cytokine‐activated NK cells in the presence or absence of PGE2 (1 µM; n = 5, biological replicates).

Analysis of cell cluster formation during 48 h of culture of IL‐2 and IL‐15 activated NK cells in the presence or absence of an anti‐CD54 antibody (20 µg/ml). One of three representative biological replicate experiments is shown.

Mean fluorescence intensity of CD54 and CD25 expression by flow cytometry in the presence or absence of Torin‐1 (1 µM; n = 4, biological replicates).

Expression of intracellular phosphorylated proteins by cytokine‐activated NK cells in the presence or absence of anti‐CD54 (20 µg/ml).

Gating strategy for sorting CD25+/CD54+ and CD25−/CD54− NK cells.

Representative example of PDE4A expression on CD56‐positive CD25+/CD54+ and CD25−/CD54− NK cells. See Fig 4F for summary of data presented as fold change between CD25/CD54 double‐positive and double‐negative NK cells.

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01, ***P < 0.001). Each circle in E represents an individual biological replicate experiment. The violin plots in A, B, D show the median (solid line) and the 25th and 75th percentile (dashed line). Error bars in C represent SD from five technical replicates.

Figure 4

CD25+/CD54+ NK cells display high cytolytic capacity in the presence of PGE2

NK cells were stimulated with either IL‐2 or IL‐15 in the presence or absence of PGE2 for 2 days and thereafter purified as CD25+/CD54+ and CD25−/CD54− and tested in a 51Cr‐release assay against K562. (E: T = 5:1, n = 4, biological replicates). Stars indicate significant P‐values comparing CD25+/CD54+ vs CD25−/CD54− NK cells for each condition.

NK cells were stimulated with IL‐15 and purified as CD25+/CD54+ and CD25−/CD54− and co‐cultures with K562 cells (5:1 ratio) for 6 h and analyzed for the expression of (B) CD107a, (C) IFNγ, (D) Perforin, (E) TRAIL and (F) Granzyme B were by flow cytometry (n = 4, biological replicates).

Flow cytometry analysis of levels (Left = % frequency, right = fold change of mean fluorescence intensity) of phosphorylated S6 (S235/236) in CD25+/CD54+ and CD25−/CD54− IL‐15 activated NK cells (n = 4, biological replicates).

Flow cytometry analysis of PDE4A expression in CD25+/CD54+ and CD25−/CD54− IL‐15 activated NK cells (n = 4, biological replicates).

Representative image of PDE4A expression in CD25+/CD54+ and CD25−/CD54− IL‐15 activated NK cells. Scale bar: 10 µm.

NK cells were first stimulated with IL‐15 for 2 days in the absence of PGE2, then purified as CD25+/CD54+ or CD25−/CD54− NK cells and tested in a 51Cr‐release cytotoxicity assay against K562 cells at an E:T ratio of 5:1 in the presence (filled bars) or absence (open bars) of PGE2 (1 µM; n = 3, biological replicates).

Mitochondria mass from purified CD25+/CD54+ and CD25−/CD54− NK cells measured by flow cytometry (n = 5, biological replicates).

Representative image of MitoTracker Red FM in CD25+/CD54+ NK cells and CD25−/CD54− NK cells. Scale bar: 20 μm.

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01, ***P < 0.001). Each circle represents an individual experiment. Error bars are depicted as standard deviation. Box and whiskers plots (K): box extends from 25th to 75th percentile, the center line indicates the median and whiskers indicate min and max values.

Through binding to its ligands LFA‐1, MAC‐1, and fibrinogen, CD54 plays an important role in cell‐to‐cell interactions (Huse, 2017). Therefore, differences in the ability to form immune cell cluster were analyzed between IL‐2 and IL‐15 activated NK cells. Upon activation with IL‐15, NK cells consistently formed distinct cell clusters even in the presence of PGE2, while IL‐2 activated NK cells displayed impaired cluster formation in the presence of with PGE2 (Fig 3E and F). To test if the formation of cell clusters impacts on the ability of IL‐15 NK cells to resist PGE2‐mediated suppression, CD54 was blocked to prevent the formation of cell clusters (Fig EV3C). However, blockade of CD54 did not impact on the ability of IL‐15 activated NK cells to kill K562 target cells in the absence or presence of PGE2 (Fig 3G). Thus, the formation of cell clusters does not contribute to the resistance to PGE2‐mediated suppression in IL‐15 activated NK cells.

Given that mTOR signaling is essential to render NK cells less susceptible to PGE2 suppression, the impact of mTOR on the expression of CD25 and CD54 was analyzed. In the presence of Torin‐1, a significant down‐regulation of CD54 (P = 0.0150) but not CD25 (P = 0.0964, ns) was observed in NK cells activated by IL‐15 (Fig EV3D). Blocking CD54 had no impact on the expression of pS6, pSTAT5, in IL‐2 or IL‐15 activated NK cells (Fig EV3E). Overall, our results demonstrate that IL‐15 induces the expression of CD54 and CD25 through mTOR signaling but CD54‐dependent cluster formation is not essential to resist PGE2‐mediated suppression in IL‐15 activated NK cells.

CD25+/CD54+ NK cells display superior anti‐tumor capacity

Given that the frequency of CD25+/CD54+ remained significantly higher upon exposure to PGE2 in IL‐15 activated NK cells, this population of NK cells was further examined for its anti‐tumor activity. After 2 days of cytokine activation, NK cells were sorted for CD25 and CD54 (Fig EV3F). Regardless of activation with either IL‐2 or IL‐15, CD25−/CD54− NK cells were unable to kill target cells even in the absence of PGE2 (Fig 4A). In contrast, CD25+/CD54+ purified from either IL‐2 or IL‐15 activated NK cells displayed high cytotoxic activity. Further investigation of the CD25+/CD54+ NK cells revealed higher levels of perforin, TRAIL, CD107a, and IFNy while granzyme B remained at similar levels compared with CD25−/CD54− NK cells, suggesting that the superior cytotoxicity of CD25+/CD54+ NK cells are mediated through perforin and TRAIL (Fig 4B–F).

To dissect the underlying mechanisms for the superior cytolytic activity of CD25+/CD54+ NK cells, NK cells were activated with IL‐15 and thereafter analyzed for the phosphorylation of pS6 and expression of PDE4A. CD25+/CD54+ NK cells showed significant increase in pS6 and PDE4A compared with CD25−/CD54− NK cells (Figs 4G and H, and EV3G). In addition, immunofluorescence staining revealed a substantially lower expression of PDE4A in CD25−/CD54−NK cells compared with CD25+/CD54+ NK cells (Fig 4I).

To test if CD25+/CD54+ NK cells are resistant to PGE2‐mediated suppression, NK cells were first activated by IL‐15 for 2 days, thereafter sorted as CD25+/CD54+ or CD25−/CD54− NK cells, and then exposed to PGE2 during the killing assay. CD25−/CD54− NK cells showed significantly reduced ability to kill K562 target cells regardless of PGE2 exposure or not. In contrast, CD25+/CD54+ showed significantly higher ability to kill K562 target cells even in the presence of PGE2 (Fig 4J).

Since previous studies have highlighted that mTOR activity is linked with increased metabolic activity in NK cells, the mitochondrial mass and mitochondrial membrane potential was analyzed in CD25+/CD54+ and CD25−/CD54− purified NK cells (Marcais et al, 2017; Zheng et al, 2019b). Flow cytometry and confocal imaging analysis revealed a higher mitochondrial mass (MitoTracker Green) and mitochondria membrane potential (MitoTracker Red FM) in CD25+/CD54+ NK cells compared with CD25−/CD54− NK cells (Fig 4K and L).

These results demonstrate that CD25+/CD54+ NK cells resist PGE2‐mediated suppression and have higher levels of pS6, PDE4A and increased mitochondrial activity compared with CD25−/CD54− NK cells.

PTGES affects the prognostic value of NK cells in patients with lung adenocarcinoma

To investigate any potential relationship between infiltration of NK cells and the presence of PGE2 on prognosis in patients with cancer, gene expression analysis of prostaglandin E synthase (PTGES) was performed across 33 TCGA data cohorts (Appendix Fig S1B). In lung adenocarcinoma (LUAD), the expression of PTGES was significantly higher in tumor tissue compared with normal tissue (Fig 5A). A high PTGES expression showed a significant negative correlation with overall survival and disease‐free survival in lung adenocarcinoma (Fig 5B). Using specific gene expression signature to identify NK cells, as defined by Bottcher et al (2018), we found that in contrary to PTGES expression, NK cells level was significantly lower in tumor tissue compared with normal tissue (Fig 5C). Between the different stages, stage I LUAD showed significantly higher NK cell gene expression signature compared with patients with stage III LUAD. Notably, the expression of IL‐15 was higher in patients with a high NK cell gene expression (Fig 5D).

Figure 5

PTGES affects the prognostic value of NK cells in patients with lung adenocarcinoma

Expression of PTGES in normal and tumor tissue.

Kaplan–Meier analysis of overall (left) and progression‐free (right) survival in patients with lung adenocarcinoma with high PTGES (top 10%) and low PTGES (bottom 90%) expression.

NK cell gene signature in normal and tumor tissues based on a TCGA lung adenocarcinoma cohort and quantification of NK cell gene signature in patients with different stages of lung adenocarcinoma.

IL15 gene expression analysis in high and low NK gene signature (NK GS).

Overall survival in patients with stage I (n = 269) lung adenocarcinoma with high (top 50%) and low (bottom 50%) NK cell gene expression signature expression.

Overall survival in patients with high (top 50%) PTGES expression in relation to high (top 50%) and low (bottom 50%) NK cell gene expression signature.

Data information: Panels A, C, D were analyzed using an unpaired parametric t‐test (****P < 0.0001). Log‐rank (Mantel–Cox) test was used to test for significance and log‐rank P‐values are displayed in Kaplan–Meier analysis graph. **P < 0.01, ****P < 0.0001. LUAD: Lung adenocarcinoma. HR is calculated at 95% CI. The violin plots in A, C and D show the median (solid line) and the 25th and 75th percentile (dashed line).

We next sought to investigate the prognostic value of NK cells in different disease stages in LUAD. In stage I lung adenocarcinoma, a high NK cell gene signature positively correlated with improved overall survival (Fig 5E). However, NK cell gene signature was not associated with survival rate in late stage (II–IV) lung adenocarcinoma (Fig EV4A). To investigate whether NK cell gene signature and the expression of PTGES have an impact on survival in lung adenocarcinoma, samples were divided based on high and low PTGES expression. While NK cell gene signature was not associated with improved survival in patients with low expression of PTGES (HR = 1.337), a high NK cell gene signature correlated with improved survival in patients with high PTGES expression (HR = 1.517; Figs 5F and EV4B). Taken together, these results show that PTGES is a factor for poor prognosis and it affects the prognostic value of NK cells in patients with lung adenocarcinoma. Furthermore, NK cell gene expression signature has a prognostic value in stage I lung adenocarcinoma.

Figure EV4

Survival in stage II‐IV lung adenocarcinoma patients. Tumor spheroid PGE2 concentration and NK cell infiltration

Overall survival in patients with stage II‐IV lung adenocarcinoma with high (top 50%) and low (bottom 50%) NK cell gene expression signature.

Overall survival in patients with low (bottom 50%) PTGES expression in relation to high (top 50%) and low (bottom 50%) NK cell gene expression signature.

For detection of PGE2 production by A549 spheroids, supernatants were harvested 7 days after initial seeding and PGE2 was quantified by ELISA. PGE2 levels of A549 monolayer cultured cells were below detection (39 pg/ml), (n = 3, biological replicates).

Example of mask (yellow) used to enumerate fluorescent events for NK cell infiltration, scale bar:400 μm.

Data information: In A and B, Log‐rank (Mantel–Cox) test used for significance, P‐values, and hazard ratio (HR) are displayed in each graph. Each circle represents an individual experiment. Error bars are depicted as standard deviation.

CD54‐positive NK cells infiltrate lung adenocarcinoma tumors

Given the prognostic value of NK cell infiltration in patients with PTGES high lung adenocarcinoma, we sought to confirm these results using an in vitro lung cancer model. Purified NK cells were co‐cultured with the lung cancer cell line A549 in the presence or absence of PGE2. The ability of IL‐2 activated NK cells to kill A549 cells was significantly delayed compared with IL‐15 activated NK cells. Although the presence of PGE2 reduced the killing by both IL‐2 and IL‐15 activated NK cells, the ability of IL‐15 but not IL‐2 NK cells to kill A549 cells was restored at 12 h (Fig 6A). Similar to killing of K562 cells, CD25+/CD54+ NK cells maintained their killing of A549 in the presence of PGE2 (Fig 6B). While A549 cells cultured as 3D‐spheroids produce PGE2, monolayer cultured A549 cells did not produce any detectable PGE2 (Fig EV4C). When added to A549 spheroids, CD25+/CD54+ NK cells showed increased infiltration compared with CD25−/CD54− NK cells (Figs 6C and D, and EV4D, Movies [Link], [Link]). The total number of both tumor‐infiltrating and tumor‐non‐infiltrating NK cells was similar between CD25+/CD54+ and CD25−/CD54− NK cells, confirming an equal NK cell seeding throughout the assay (Fig 6E). To confirm these results, tumor spheroids were harvested and analyzed for the presence of NK cells by flow cytometry. Indeed, CD25+/CD54+ NK cells were present at higher frequencies within the spheroids compared with CD25−/CD54− NK cells (Fig 6F and G).

Figure 6

CD54‐positive NK cells infiltrate lung adenocarcinoma tumors

Real‐time image‐based killing of A549 cells by NK cells isolated from two individual healthy donors at an E:T ratio of 10:1.

NK cells were stimulated with IL‐2 or IL‐15 in the presence or absence of PGE2 and then purified as CD25+/CD54+ or CD25−/CD54− cells and tested for cytotoxicity against A549 tumor cell in a 51Cr‐release assay. E:T ratio = 5:1 (n = 4, biological replicates). Stars indicate significant P‐values comparing CD25+/CD54+ vs CD25−/CD54−.

Representative image of NK cell infiltration (red) into A549 tumor spheroids at the time of addition of NK cells and at 20 h after the addition of NK cells, scale bar: 400 μm.

Incucyte quantification of (D) total and (E) tumor‐infiltrating CD25+/CD54+ and CD25−/CD54− NK cells. RCU, Red Calibrated Unit (n = 4, biological replicates).

Flow cytometry quantification of (F) tumor‐infiltrating and (G) non‐infiltrating NK cells. Results are depicted as total NK cells among CD45‐positive live cells. (n = 4, biological replicates).

48 h dpi zebrafish injected with A549, A549+CD25+/CD54+, or A549+CD25−/CD54− NK cells after 24 h. Fluorescent image was taken by Leica epi‐fluorescent stereomicroscope. Scale bar: 500 µm. The tumor injection area (perivitelline space) is depicted with a dashed line.

3D view of tumor injection area 24 h post‐tumor inoculation (left). Scale bar = 50 µm. The tumor injection area (perivitelline space) is depicted with a dashed line. Magnified view for tumor area (right). The A549‐td tumor cells are annotated with a purple translucent mask and NK cells are color‐coded based on the distance to the tumor area (near‐far: blue‐red). Scale bar = 10 µm.

The distance between NK cells and tumor area in CD25−/CD54− and CD25+/CD54+ NK cell treated fish (n = 5, biological replicates).

Frequency of tumor‐infiltrating NK cells in CD25−/CD54− and CD25+/CD54+ NK cell treated fish (n = 5, biological replicates).

Quantification of (L) tumor cell and (M) NK cell counts 24 h after injection of NK cells by Imaris (n = 5, biological replicates).

Frequency of tissue‐infiltrating total NK cells (CD56+/CD3−; left) and CD54‐positive NK cells (right) among CD45‐positive live cells (n = 10, biological replicates).

Data information: Data were analyzed using a paired parametric t‐test (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). Each circle represents an individual experiment. Error bars are depicted as standard deviation. The violin plots in J and K show the median (solid line) and the 25th and 75th percentile (dashed line).

To further substantiate if CD25+/CD54+ NK cells infiltrate tumors at higher frequencies compared with CD25−/CD54− NK cells site, a xenograft zebrafish model was developed. A549 tumor burden was significantly reduced in animals treated with CD25+/CD54+ NK cells. In contrast, injection of CD25−/CD54− did not significantly reduce the tumor burden (Fig 6H and L). In addition, CD25+/CD54+ NK cells infiltrated A549 tumors to a significantly higher degree compared with CD25−/CD54− NK cells (Fig 6I and K). Furthermore, the distance between the tumor and CD25+/CD54+ NK cells was significantly lower than that of the distance between the tumor to CD25−/CD54− NK cells (Fig 6J). The overall injected CD25−/CD54− NK cells remained at the same frequencies as CD25+/CD54+ NK cells (Fig 6M).

Finally, infiltration of NK cells was analyzed in a cohort of lung adenocarcinoma patients (n = 10). Comparing the center of the tumor, invasive margin, and normal adjacent tissue, NK cell frequency was significantly higher in the invasive margin and normal adjacent tissue compared with the center of the tumor. While the median frequency of CD25‐positive NK cells was 5.31%, the frequency of CD54‐positive NK cells ranged between 12.5–64.3%. The frequency of CD54‐positive NK cells was significantly higher in the central tumor region compared with the invasive margin and normal adjacent tissue (Fig 6N). Taken together, these results show that NK cells expressing CD25 and CD54 have superior capacity to kill and infiltrate lung cancer cells in the presence of PGE2 and that NK cells expressing CD54 are more abundant within central lung cancer tissue.

Discussion

Intratumoral NK cell frequency correlates with improved prognosis in several different cancers (Barry et al, 2018; Lee et al, 2019; Souza‐Fonseca‐Guimaraes et al, 2019). However, suppressive factors produced within the tumor microenvironment including PGE2 limit the activation of NK cells (Park et al, 2018). Here we show that NK cells activated by IL‐2 are suppressed by PGE2, whereas IL‐15 activated NK cells are less sensitive to suppression by PGE2. The reduced susceptibility to PGE2 was dependent on the expression of the cyclic nucleotide phosphodiesterase family member PDE4A.

The expression of PDE4 family members has been linked to tumor progression and metastatic dissemination in lung cancer and hepatocellular carcinoma (Pullamsetti et al, 2013; Peng et al, 2018). PDE4A is also expressed in human inflammatory cells including T cells and monocytes where it terminates the activity of cAMP by hydrolyzing it into AMP (Manning et al, 1999). Intracellular cAMP acts as a potent immunosuppressive signaling molecule in T and NK cells and is upregulated by multiple factors including PGE2 (Harris et al, 2002). Recently, overexpression of PDE4A in T cells was proposed as an immune checkpoint inhibitor against cAMP‐mediated immunosuppression in vitro (Schmetterer et al, 2019). However, the role of PDE4A in regulating the levels of cAMP in human NK cells is unknown. Here, we first show that IL‐15 increases the expression of PDE4A and inhibition of PDE4A increases the levels of cAMP.

Exposure to IL‐15 activates mTOR signaling in human NK cells (Marcais et al, 2014; Mao et al, 2016; Wang et al, 2018). While inhibition of PDE4D, another PDE4 member capable of hydrolyzing cAMP into AMP, can suppress mTOR activity in colorectal cancer cells, it is unknown if PDE4A cross‐talk with mTOR in human NK cells (Kim et al, 2019). A recent study showed that increased cAMP signaling activates protein kinase A (PKA) and inhibits mTORC1 (Jewell et al, 2019). Here we demonstrate that inhibition of mTOR reduces the expression of PDE4A. Likewise, inhibition of PDE4A reduces the activation of mTOR signaling. Thus, there is a reciprocal cross‐talk between mTOR and PDE4A in human NK cells.

The 5′ AMP‐activated protein kinase (AMPK) acts as a sensor of cellular energy status and is activated under conditions of low intracellular ATP. Given that AMPK can inhibit mTOR and that PDE4 was recently shown to regulate AMPK‐dependent autophagy, it is plausible that AMPK can also affect the expression of PDE4A and mTOR activity in NK cells (Gwinn et al, 2008; Zhong et al, 2019). Upon exposure to PGE2, pAMPK levels were significantly upregulated in both IL‐2 and IL‐15 stimulated NK cells. Furthermore, the expression of CD54 and frequency of CD25+/CD54+ was significantly reduced upon inhibition of AMPK using compound C. In addition, pAMPK levels were not affected upon inhibition of mTOR, whereas the expression of PDE4A and pS6 were significantly reduced upon inhibition of AMPK (Appendix Fig S2). Taken together, these results indicate AMPK does play a role to regulate NK cell expression of CD54, PDE4A, and mTOR in NK cells.

Upregulation of PDE4A was only observed upon IL‐15 stimulation but not upon stimulation with IL‐2. We previously demonstrated that IL‐15 is superior to IL‐2 to increase mTOR activity and that priming with IL‐15 increases differential translation compared with NK cells primed with IL‐2 (Mao et al, 2016). Although IL‐2 and IL‐15 both rely on the JAK/STAT signaling pathway, stimulation with IL‐15 resulted in a higher phosphorylation of STAT5. Furthermore, STAT5 was identified as a putative transcription factor binding to PDE4A thus strengthening our results that PDE4A expression is regulated via IL15/JAK/STAT5 signaling.

Compared with IL‐2, activation with IL‐15 changes the expression of several cell surface receptors. Similar to other studies, we previously showed that the expression of CD25 is significantly higher upon activation with IL‐15 compared with IL‐2 (Dybkaer et al, 2007; Bezman et al, 2012; Pradier et al, 2014; Mao et al, 2016). Our present study confirms these results and also shows that the expression of CD54 is significantly higher on NK cells upon IL‐15 stimulation compared with IL‐2 stimulation.

CD54 is an adhesion molecule expressed by hematopoietic and non‐hematopoietic cells and can be upregulated on lymphocytes following activation (Roy et al, 2001; Harjunpaa et al, 2019). Ligation of adhesion molecules is a hallmark for lymphocyte activation and an initial step for immune synapse formation to drive the cancer‐NK cell immunity cycle (Huntington et al, 2020). The expression of CD54 is well‐studied on myeloid cells but few studies have focused on studying CD54 and its role in human NK cells. Early studies have shown that IL‐15 upregulates the expression of CD54 on human cord‐blood derived NK cells and on NK cell lines (Lin & Yan, 2000; Zhang et al, 2004). NK cells expressing CD54 following IL‐15 stimulation show a higher capacity to adhere to cell culture plastics (Sun et al, 2003). A recent study by Ni et al (2020) showed that cytokine‐activated IFNγ‐producing NK cells express higher levels of several activating receptors as well as CD54. Basingab et al (2016) showed that the overexpression of CD54 on tumor cells counteracts immune‐suppression by PGE2 to restore T‐cell activity. Here, we found that IL‐15‐induced CD54 resulted in a higher capacity to form cell clusters even in the presence of PGE2. However, blocking the CD54 receptor did not change susceptibility of NK cell to PGE2‐mediated suppression. Through further analysis of NK cells expressing CD25 and CD54 following activation with IL‐15, only the population of cells expressing both receptors showed increased S6 phosphorylation and PDE4A expression, and an increased capacity to resist PGE2‐mediated suppression in vitro. We hypothesize that CD25 is essential to preserve NK cell function under PGE2‐cAMP pressure; however, we were unable to test this hypothesis due to the low frequency of CD25 single positive NK cells following activation with IL‐15.

The role of PGE2 has previously been studied in lung cancer where it can synergize with IL‐17 to form an M2 macrophage‐dominant tumor microenvironment (Liu et al, 2012). In an experimental lung cancer model, Ogawa et al (2014) found that aspirin reduces lung cancer metastasis to regional lymph nodes. From the analysis of TCGA dataset, we found a higher expression of PTGES in lung cancer tissue compared with normal tissue and that PTGES is a poor prognostic factor in lung adenocarcinoma. Furthermore, NK cell gene expression signature correlated with improved prognosis in early but not late stage lung adenocarcinoma, suggesting that NK cells play a role in the immune surveillance in early stage lung adenocarcinoma.

Despite high expression of PTGES in tumors, NK cell gene expression was still detected. To investigate if tumors with high and low NK cell gene expression differ in tumors with a high PTGES expression, we performed gene set enrichment analysis. Tumors with high PTGES and high NK cell gene expression showed enriched expression of inflammatory‐related pathways including T‐cell activation, T‐cell receptor signaling pathway, activation of innate immune response, and positive regulation of cytokine production (Appendix Fig S3). These results indicate that PTGES high tumors with high NK cell gene expression are in general more inflamed than PTGES high tumors with a low NK cell gene expression. Thus, inflammatory pathways could potentially increase the persistence of NK cell despite high levels of PTGES.

Notably, NK cell high gene expression signature correlated with improved prognosis and significant improved overall survival only in patients with PTGES high tumors. These observations might be explained by an existing NK cell‐dependent immune selective pressure in the PTGES high tumors. Zelenay et al (2015) showed that genetic ablation of COX through PTGS2 knock‐out shifts the tumor inflammatory profile, where an increase in Cd25, Ifng, and Gzmb is observed, thereby impairing the tumor growth. Here, we show that the CD25+CD54+ subpopulation of NK cells resists such immune escape mechanism by being unresponsive to suppression by PGE2.

Natural killer cells generally showed a poor infiltration in central lung cancer tissue compared with peripheral lung cancer tissue, but NK cells expressing CD54 showed a preferential infiltration into the central tumor region. Similarly, NK cells expressing CD25 and CD54 showed significantly higher ability to infiltrate lung adenocarcinoma spheroids in vitro. In addition, these NK cells showed increased ability to kill lung adenocarcinoma cells in vitro. These data suggest that the expression of CD54 is required for NK cells to infiltrate lung adenocarcinoma tissue.

The zebrafish is a recent addition to animal models of human cancer (Fior et al, 2017; Hason & Bartunek, 2019; Yan et al, 2019). With optically clear zebrafish model, visualization of single‐cell phenotypes is applicable. Here we report a model based on infusion of NK cells to monitor tumor cells growth in vivo in zebrafish. Consistent with our in vitro results, CD25+CD54+ NK cells show superior ability to infiltrate and kill lung adenocarcinoma tumors.

In summary, we investigated if modulation of the cytokine milieu represents a strategy to render NK cells less susceptible to tumor‐induced immunosuppression to increase their intratumoral persistence. This approach was recently shown by Fujii et al (2018) who demonstrated that IL‐15 protects and rescues NK cell cytotoxicity from TGFβ‐mediated immunosuppression. Here we uncover a unique mechanism of IL‐15 to activate NK cells to resist PGE2‐mediated suppression. Mechanistically, IL‐15 enriches a subset of CD25+/CD54+ NK cells with superior mTOR activity and PDE4A expression. Therefore, approaches to selectively expand CD25+CD54+ NK cells for adoptive cell therapy represent a potential therapeutic strategy for patients with tumors producing high levels of PGE2.

Materials and Methods

Human specimen

All human specimens were collected from patients with primary diagnosis of Non‐small‐cell lung cancer (NSCLC). Written informed consent was signed by all the patients for sample acquisition for research purposes. This study was conducted under the approval of the Ethics Committee of the Second Affiliated Hospital of Zhejiang University (IR2019001101). Fresh specimens (n = 10) were obtained from patients undergoing lung resection surgery. All patients were diagnosed with adenocarcinoma and received no previous cancer treatment before the resection. Pathology diagnosis was based on routine H&E staining and cytological assessments. Central tumor tissue, adjacent tumor tissue, and non‐adjacent normal lung tissue were collected within 2 h after resection and immediately processed for flow cytometry analysis (Appendix Fig S1D). The non‐adjacent normal lung tissues were taken as distant as possible from the tumor and were reviewed by a pathologist. The clinical and pathologic characteristics of all patients included in this study are summarized in Appendix Table S1.

Cell culture

Peripheral blood mononuclear cells (PBMC) were isolated from buffy coat by Ficoll gradient centrifugation (GE healthcare) and washed three times in PBS, followed by red‐cell lysis by ACK buffer (Thermo Fisher). NK cells were purified from PBMC using negative selection microbeads (NK cell isolation kit, human, Miltenyi biotech) according to the manufacturer's protocol. The purity of isolated NK cells was above 95% as assessed by flow cytometry staining for CD56 and CD3. Isolated NK cells were seeded in 48‐well flat‐bottom plate (TPP techno) in X‐vivo 20 medium (Lonza) supplemented with 10% heat‐inactivated human AB serum and 300 IU/ml of IL‐2 or 21 ng/ml IL‐15. Where indicated, NK cells were cultured in the presence of 1 μM PGE2 (Sigma Aldrich), 1 μM Torin‐1 (SelleckChem), or 1 μM PDE4 inhibitor, (Roflumilast, SelleckChem). The NSCLC cell line A549, A549‐tdtomato cell line and chronic myelogenous leukemia cell line K562 were maintained in RPMI1640 medium (Life Technologies) supplemented with 10% heat‐inactivated FBS (Life Technologies) and 1% antibiotics (penicillin, streptomycin).

Cytotoxicity assays

Natural killer cell cytotoxicity was measured by chromium‐release assay (4 h) against K562 and A549 cells at a 5‐to‐1 effector‐to‐target ratio, as previous described (Mao et al, 2016). For real‐time killing, A549 cells were labeled with IncuCyte® NucLight Rapid Red Reagent (Essen BioScience) at a dilution of 1:1,000, then seeded at 5,000 cells per well into a 96‐well flat‐bottom plate in RPMI medium supplemented with 10% FBS (Invitrogen) and cultured at 37ºC overnight to attach. NK cells were plated at 25,000 cells per well in complete RPMI medium, with or without PGE2. IncuCyte® Caspase‐3/7 Green Apoptosis Assay Reagent (Essen Bioscience) was added to each well at a dilution of 1:1,000. The assay plate was monitored, and four images were acquired per well every hour. Average numbers of dead tumor cells were determined at each time point using the IncuCyte live‐cell analysis software. Where indicated, anti‐CD54 antibody (20 μg/ml) was added to assays.

Real‐time imaging‐based cluster assay

Live content imaging and analysis was performed with an IncuCyte S3 system (Essen Bioscience, Sartorius). In brief, images of cytokine‐activated NK cells were taken with a 10× objective lens every 2 h from initial cell seeding. IncuCyte S3 software was used to automatically score and quantify cluster formation. Clusters were defined as cell aggregates occupying an area at least 2,000 μm2 and an eccentricity (non‐circularity) of < 0.8.

Quantitative RT‐PCR

Total RNA was extracted from 0.5 million cultured NK cells using the RNeasy Micro Kit (Qiagen). cDNA was synthesized according to the manufacturer's instructions of QuantiTect Reverse Transcription Kit (Qiagen). Quantitative real‐time PCR was performed using Power SYBR Green PCR Master Mix (Thermo Fisher) and detected on the CFX96 Touch Real‐Time PCR Detection System (Bio‐Rad). The expression levels of each gene were determined according to the comparative Ct method and normalized to housekeeping gene TBP. The sequences of used primers are described in the Appendix Table S2.

CistromeDB toolkit

Data type used in CistromeDB as “Transcription factor, chromatin regulator”, the interval of PDE4A (chr19:10,416,773–10,469,631/hg38) was submitted via CistromeDB (http://dbtoolkit.cistrome.org/) (Zheng et al, 2019a). Results were downloaded from the website and plotted using GraphPad Prism (GraphPad).

Flow cytometry

Detailed information of antibodies used in this study is summarized in Appendix Table S3. Briefly, 1–2 × 105 cells were harvested in Falcon® Round‐Bottom Polystyrene Tubes (BD) and washed twice with FACS buffer (PBS with 2% FBS). Next, cells were resuspended in 20 μl PBS containing the appropriate antibody cocktails for extracellular antigens and incubated at 4ºC for 30 min. Cells were acquired on a Novocyte flow cytometer (ACEA biosciences) and analyzed by Flowjo (BD). For intracellular staining, the stain‐fix‐perm procedure performed according to BD phosflow alternative protocol II as used. In brief, following staining for cell surface antigens, cells were fixed in BD phosflow fix buffer for 10 min, and wash twice with PBS. Then, cells were incubated on ice with BD phosflow perm buffer III for 30 min, followed by three times washing prior staining for intracellular antigens.

For degranulation and intracellular staining for perforin and IFNγ, NK cells were incubated with an equal amount of K562 cells for 4 h in medium containing CD107a and CD56 antibodies. Golgiplug (BD) and Golgistop (BD) were added in order to accumulate cytokine production during incubation with K562. Intracellular staining for IFN‐γ and Perforin was performed as previously described (Neo et al, 2020).

For mitochondria staining, sorted cells were stained with MitoTracker Green FM in accordance with manufacture's guidelines (Thermo Fisher). Following three washing steps with FACS buffer, cells were acquired on a Novocyte flow cytometer (ACEA biosciences) and analyzed by Flowjo (BD) and SPICE 6.0 (Roederer et al, 2011). Gating strategy for NK cells is shown in Appendix Fig S1A.

For analysis of spheroid cultures, 5,000 A549 cells were seeded in Nunclon sphera 96‐well Round (U) bottom plate (Thermo Fisher) for 7 days. NK cells were labeled with IncuCyte® NucLight Rapid Red Reagent (Essen BioScience) and added at 25,000 cells to each well. Following 24 h co‐culture, spheroids were gently resuspended in 1 ml PBS and left to sediment to the bottom of a 5‐ml Eppendorf tube. This washing step was repeated twice. After each washing step, supernatants were collected as non‐tumor‐infiltrating NK cells. Spheroids containing tumor‐infiltrating NK cells were harvested and trypsinized to obtain a single‐cell suspension and thereafter analyzed by flow cytometry. Single‐cell suspensions of spheroids and non‐tumor‐infiltrating cells were stained with Live/Dead cell marker and anti‐CD45 antibody to distinguish NK cells from tumor cells.

Gene ontology enrichment analysis

Analysis of differentially expressed genes between IL‐2 and IL‐15 primed NK cells was performed using a 1.5‐fold change cutoff from a publicly available data set (Mao et al, 2016). Comparative GO analysis among different categories of genes was performed using the ClueGo (2.5.4) plugin in Cytoscape (3.7.1). Gene IDs were submitted to query the GO‐immunology process database. The working ClueGo parameters were set as follows: Evidence code set to: All; Use Go Term Fusion; GO tree interval, level 2–5; GO term selection minimum number of genes 8; GO term connectivity threshold (Kappa score) 0.7; Significance test using Two‐sided hypergemetric test with Bonferroni step‐down method for P value correction. Gene ontology terms are presented as nodes and clustered together based on the term similarity. Node size is proportional to the P value for GO term enrichment with larger node sizes representing more significant GO term. The node color is set according to the relative enrichment of the GO term in each cluster, gray color indicates no significant difference in enrichment for the represented term. The leading GO terms were annotated with larger font text for easier visualization. For cerebral view, CluePedia (1.5.4) plugin was used. Color coded with enriched pathway with annotation for cellular location from extra cellular to nucleus.

Immunofluorescence microscopy and image analysis

For intracellular staining, cells were fixed and permeabilized using the Foxp3/Transcription Factor Staining Buffer Set in accordance with manufacture's guidelines (eBioscience). Then, cells were incubated with primary antibodies and fluorescence‐conjugated secondary antibodies. The nucleus was stained by Fluoroshield™ with DAPI (Sigma). For mitochondria staining, sorted cells were stained with MitoTracker Red FM according to the manufacture's guidelines (Thermo Fisher). Slides of stained cells ware prepared with the cytospin 2 (SHANDON). Immunofluorescent staining was analyzed using the confocal laser scanning microscope LSM 700 system (Zeiss).

Zebrafish xenografts were fixed in 4% formaldehyde and stored in methanol at −20°C. Xenografts were mounted with 80% glycerol on µ‐Dish 35 mm (ibidi) monitored by confocal laser scanning microscope LSM Airy 800 (Zeiss). Confocal image stacks were reconstructed and visualized as three dimensional (3D) volumes with Imaris software (version 9.5; Bitplane). The Imaris Spot detection algorithm was used as described by the manufacturer for semiautomatic identification and counting of fluorescently labeled NK cells and A549‐tdTomato. Main parameters were absolute thresholding, an object size of 5 μm diameter, mean fluorescent intensity cutoff is 250. Errors in the software detection results (both erroneous positive and negative objects) were corrected by manual inspection of the data sets to adjust parameters.

cAMP and PGE2 detection assays

Cytokine‐activated NK cells were stimulated with 1 μM PGE2 for 5 min and intracellular cAMP was measured by competitive immunoassay using a cAMP Parameter Assay Kit (R&D Life Sciences) according to the manufacturer's instruction. Following culture for 7 days, supernatants were harvested and assayed for PGE2 content by ELISA (R&D Life Sciences).

Cell sorting

Natural killer cells were sorted after 2 days of cytokine activation in the absence or presence of PGE2. Cells were harvested and stained with surface antibodies against CD56, CD16, CD25, CD54, and Live/Dead cell marker. Stained cells were then sorted using BD FACSAria™ Fusion.

TCGA datasets analysis

All data were retrieved from The Cancer Genome Atlas (TCGA) lung adenocarcinoma (LUAD) dataset (https://portal.gdc.cancer.gov/projects/TCGA‐LUAD). For generation of gene expression, NK cell gene signatures, normalized expression values were log2‐transformed and ranked by the mean expression value of signature genes. The following genes were used for NK cell gene signature: Natural Cytotoxicity Receptor 1 [NCR1], [NCR3], Killer Cell Lectin Like Receptor B1 [KLRB1], [CD160], Perforin 1 [PRF1] (Bottcher et al, 2018)). Overall survival analysis for PTGES and NK cell gene expression signature selected genes was performed with top and bottom expression values as indicated in respective figure legends and plotted as Kaplan–Meier curves using GraphPad Prism (GraphPad).

GEPIA 2 (http://gepia2.cancer‐pku.cn/) was used to compare PTGES gene expression across 33 TCGA databases (Tang et al, 2019). Significantly changed (P < 0.01, |log2FC|>1) cohort names were labeled in red for Tumor > Normal and green for Normal > Tumor. For pathways analysis of PTGEShiNKhi vs PTGEShiNKlow, the top 200 genes that were significantly changed (Fold changes > 2, P < 0.05) in PTGEShiNKhi compared with PTGEShiNKlow were processed via R for Gene Set Enrichment Analysis (GSEA).

Zebrafish tumor model

Zebrafish embryos were raised at 28°C under standard experimental conditions. Zebrafish embryos at the age of 24 hpf (hours post‐fertilization) were incubated in water containing 0.2 mmol/l 1‐phenyl‐2‐thio‐urea (PTU, Sigma). At 48‐hpf, zebrafish embryos were dechorionated with a pair of sharp‐tip forceps and anesthetized with 0.04 mg/ml of tricaine (MS‐222, Sigma). Anesthetized embryos were subjected for microinjection. Sorted NK cells were labeled in vitro with 2 µM of Carboxyfluorescein succinimidyl ester (CFSE) and mixed with A549‐tdTomato cells (1:1 ratio) and injected at 5 nl (approximately total 500 cells) into the perivitelline space (PVS) of each embryo by an Eppendorf microinjector (FemtoJet 5247, Eppendorf and Manipulator MM33‐Right, Märzhäuser Wetziar). Non‐filamentous borosilicate glass capillaries needles were used for injection and the injected zebrafish embryos were immediately transferred into PTU aquarium water 33°C until the end of experiment. 24 h after injection, zebrafish embryos were monitored by confocal laser scanning microscope LSM 700 system (Zeiss). Appendix Fig S1C illustrates the workflow for zebrafish injection.

Statistical analysis

Unless otherwise stated, all results were collected from multiple experiments and figures were prepared in a Prism 8.0 (GraphPad version 8). Differences between experimental groups were analyzed by paired Student’s t‐test. For TCGA dataset analysis, two‐tail unpaired t‐test was used. The difference in overall survival was tested using log‐rank tests. All results are presented as mean ± SD and represented histogram or images were selected based on the average values, P < 0.05 was considered significant (*P < 0.05; **P < 0.01; ***P < 0.005; ****P < 0.0001).

Ethics approval

This study was conducted under the approval of the Ethics Committee of the Second Affiliated Hospital of Zhejiang University (IR2019001101), China. Written informed consent was signed by all the patients for sample acquisition for research purposes. For in vivo studies, no ethical permit is required as zebrafish embryos younger than 5 days have been used only. Those embryos are excluded from the normative on animal testing by the EU directive 2010/63/EU.

Author contributions

ZC, YY, LT, and SYN, conducted experiments and analyzed data. HS, YC, JW, AKW, contributed to acquisition of data and methods development. KL, P‐JJ, YC, KW, and EA, contributed to data analysis and interpretation, and essential reagents. AL, YM, DS, LLL, conceived the project, supervised the research, and wrote the manuscript with input and editing from all authors.

Conflict of interest

The authors declare that they have no conflict of interest.

Appendix

Click here for additional data file.

Expanded View Figures PDF

Click here for additional data file.

Movie EV1

Click here for additional data file.

Movie EV2

Click here for additional data file.

Review Process File

Click here for additional data file.