PD-1 mediates functional exhaustion of activated NK cells in patients with Kaposi sarcoma

Programmed Death-1 (PD-1), an inhibitory receptor expressed by activated lymphocytes, is involved in regulating T- and B-cell responses. PD-1 and its ligands are exploited by a variety of cancers to facilitate tumor escape through PD-1-mediated functional exhaustion of effector T cells. Here, we report that PD-1 is upregulated on Natural Killer (NK) cells from patients with Kaposi sarcoma (KS). PD-1 was expressed in a sub-population of activated, mature CD56dimCD16pos NK cells with otherwise normal expression of NK surface receptors. PD-1pos NK cells from KS patients were hyporesponsive ex vivo following direct triggering of NKp30, NKp46 or CD16 activating receptors, or short stimulation with NK cell targets. PD-1pos NK cells failed to degranulate and release IFNγ, but exogenous IL-2 or IL-15 restored this defect. That PD-1 contributed to NK cell functional impairment and was not simply a marker of dysfunctional NK cells was confirmed in PD-1-transduced NKL cells. In vitro, PD-1 was induced at the surface of healthy control NK cells upon prolonged contact with cells expressing activating ligands, i.e. a condition mimicking persistent stimulation by tumor cells. Thus, PD-1 appears to plays a critical role in mediating NK cell exhaustion. The existence of this negative checkpoint fine-tuning NK activation highlights the possibility that manipulation of the PD-1 pathway may be a strategy for circumventing tumor escape not only from the T cell-, but also the NK-cell mediated immune surveillance.


INTRODUCTION
Natural Killer (NK) cells are cytotoxic innate lymphoid cells that are involved in the elimination of infected cells and tumors. Reactivity of NK cells toward target cells is tightly regulated by the integration of signals from activating and inhibitory cell surface receptors, which ultimately determine the magnitude of NK cellmediated cytotoxicity and cytokine production [1][2][3][4]. NK cell activating receptors recognize stress-induced molecules at the surface of target cells, while inhibitory receptors mostly recognize major histocompatibility complex (MHC) class I molecules that contribute to host tolerance of NK cells. Target cells with low or no expression of MHC class I molecules do not provide inhibitory signals and therefore, become highly sensitive to NK cell-mediated elimination. The potential of NK cells to attack self cells is limited by tolerance mechanisms that set the triggering threshold of individual NK cells in order to prevent reactivity against self [5][6][7]. Moreover, tumors and infected cells develop various escape strategies to bypass NK-cell mediated surveillance [8][9][10].
Kaposi sarcoma (KS) is a multifocal proliferation of lymphatic endothelium-derived spindle cells infected with the oncogenic virus, herpesvirus-8 (HHV-8) also called Kaposi-associated herpes virus (KSHV). Large leucocyte infiltrates are usually present in KS lesions, suggesting that active immune microenvironment is counterbalanced by immune inhibitory signals that prevent tumor elimination. We previously reported substantial alterations of NK cell functions in patients with KS, and identified distinct immunoevasion mechanisms that allowed HHV8 to escape NK cell responses and promote tumor cell growth [34]. Expression of PD-1 on NK cells was reported in different clinical settings [35][36][37][38][39][40], but its consequences on NK cell functions in tumor patients have not been evaluated in depth. Here, we show that PD-1 is expressed in a fraction of mature CD56 dim NK cells and associated with strongly altered functional capacities in KS patients. We demonstrate that PD-1 is induced in vitro on control NK cells upon sustained stimulation through activating ligands, and mediates inhibition of NK-cell degranulation and cytokine production. These data thus show that, as in T cells, expression of PD-1 on NK cells induces functional exhaustion, and support PD-1 as an immune checkpoint that controls NK cell activation upon chronic stimulation. An important implication of the present study is the possibility that therapeutic PD-1 blockade may be a strategy for circumventing tumor escape not only from the T cell-mediated, but also the NK cell-mediated immune surveillance.

PD-1 is expressed on a fraction of CD56 dim NK cells in KS patients
We found that a subset of NK cells from KS patients expressed PD-1 (mean frequency, 4.0% ± SEM 0.8% of NK cells vs. 0.5% ± 0.08% in age-matched healthy controls, P < 0.0001) ( Figure 1A, 1B). PD-1 pos cells were exclusively detected among the CD56 dim population, and not in CD56 bright NK cells ( Figure 1A). Elevated PD-1 levels were confirmed by qRT-PCR on sorted PD-1 pos versus PD-1 neg NK cells ( Figure 1C).
To determine if the expression of PD-1 on NK cells was related to the HHV8-related tumor process or to the presence of HHV8 infection alone, we analyzed HHV8 asymptomatic carriers. We found PD-1 pos NK cells in HHV8 asymptomatic carriers, although at two times lower frequency than in KS patients (2.0% ± 0.5% of NK cells, P = 0.01 compared to healthy controls; P = 0.02 compared to KS patients) ( Figure 1B). Since HHV8 infection frequently occurs in the context of HIV co-infection, we subgrouped KS patients and HHV8 asymptomatic carriers according to the presence or absence of HIV co-infection (Table 1). Yet, it must be noted that all HIVpositive subjects in our study were HIV-aviremic under antiretroviral treatment (ART). In both KS patients and HHV8 asymptomatic carriers, PD-1 expression was not different in HIV-positive and HIV-negative subjects ( Figure 1D). We also analyzed a series of HHV8-negative, HIV-positive patients (ART-treated, HIV aviremic) and found PD-1 pos NK cells at a frequency comparable to that in HHV8 asymptomatic carriers (mean 2.1% ± 0.8%, P = 0.01 compared to healthy controls) ( Figure 1E). Expression of PD-1 on CD56 bright NK cells was previously reported in patients with chronic hepatitis C [38]. We analyzed a series of HCV chronically infected patients and found a very small proportion of PD-1 pos NK cells (mean 1.3% ± 0.2%, P ns compared to controls) ( Figure 1E). Like in HHV8-or HIV-positive subjects, all PD-1 pos NK cells from HCV-infected patients were found in the CD56 dim population, and not in CD56 bright NK cells. Moreover, PD-1 expression was not related to active HCV replication, as it was comparable in treatment-naïve patients and in patients with sustained response to IFNα/ribavirin-treatment (data not shown).
Notably, the proportion of PD-1 pos NK cells showed a considerable inter-individual heterogeneity, sometimes accounting for more than 20% of NK cells. This proportion was stable over the time, as observed by repeated analysis over 5-years follow-up in some patients (data not shown), indicating that the presence of PD-1 pos NK cells was not related to acute intercurrent events. No association was found between the frequency of PD-1 pos NK cells and age, gender, geographic origin, duration or severity of the underlying disease (active versus stable KS (data not shown). Expression of PD-1 on HIVspecific T cells was reported to positively correlate with plasma viral load in treatment naïve HIV-infected patients [41]. However, we found no correlation between PD-1 expression on NK cells and detectable HHV8 viremia, and observed only a weak correlation between the percentages of PD-1 pos NK cells and PD-1 pos CD8 T cells (r = 0.28, P = 0.01). Recently, PD-1 pos NK cells were observed in healthy individuals seropositive for cytomegalovirus [40]. However, we found no association between PD-1 expression on NK cells and the presence of CMV-specific IgG in patients (mean PD-1 pos NK cells, 4% in CMVpositive and 3.8% in CMV-negative patients, p = 0.65). Unfortunately, the CMV serological status of our healthy controls was not available. We also tested if expression of PD-1 on NK cells might be related to another blood-born persistent viral infection in which elevated proportions of PD-1 pos T cells have been described, such as Epstein Barr virus (EBV) infection. The frequency of PD-1 pos NK cells was not correlated with EBV viral load (r = −0.06, P = 0.8), making it our hypothesis unlikely. Elevated levels of circulating microbial products were shown to be responsible for the upregulation of PD-1 on monocytes in HIV patients [42]. Thus, we measured LPS and soluble CD14 levels in patients' sera as indicators of microbial translocation, but found no correlation with PD-1 expression on NK cells (r = −0.15, P = 0.44 and r = 0.14, P = 0.42, respectively; data not shown).
Taken together, these results show that PD-1 was expressed in a fraction of CD56 dim CD16 + NK cells in KS patients, and to a lesser extent in patients with chronic viral infections (HHV8, HIV or HCV), without clear relationship between the proportion of PD-1 pos NK cells and clinical or biological characteristics of the underlying disease (duration, activity, response to treatment, viral load, associated infection).

Expression of PD-1 on NK cells is associated with functional hyporesponsiveness
Persistent expression of PD-1 on T cells renders them dysfunctional or exhausted. To assess the functional relevance of PD-1 expression on NK cells, we compared the ex vivo functional capacity of PD-1 pos and PD-1 neg NK cells from KS patients. Responsiveness was assayed by degranulation (cell surface expression of CD107a) and IFNγ production following 5-hour stimulation by HLA class I-negative K562 target cells, or direct triggering of NKp30 or NKp46 activating receptors with plate-bound agonist antibodies. Since PD-1 expression was confined to the CD56 dim subset of NK cells, comparison of PD-1 neg and PD-1 pos NK cells was restricted to the CD56 dim population to rule out functional differences between CD56 bright and CD56 dim NK cell subsets. At steady state, a fraction of PD-1 pos NK cells spontaneously expressed  CD107a (5.9% of PD-1 pos vs 1.3% of PD-1 neg CD56 dim NK cells, P = 0.007). However, following stimulation, PD-1 pos NK cells responded much less efficiently than their PD-1 neg counterpart in all conditions of stimulation. This was reflected by a lower increase of CD107a expression after stimulation by K562 targets or by plate-bound NKp30 or NKp46 antibodies (Figure 2A). Similarly, IFNγ production was much lower in PD-1 pos than in PD-1 neg NK cells ( Figure 2B). Even upon strong stimulation with PMA plus ionomycin, pharmacological activators that bypass early receptor signaling events, PD-1 pos NK cells responded poorly in both the degranulation and IFNγ production assays. Altogether, these results indicate that PD-1 expression on NK cells is associated with their functional hyporesponsiveness.

PD-1 expression inhibits NK-cell effector functions
Binding of PD-1 to one of its ligands results in T lymphocyte functional inhibition. To more directly evaluate the involvement of PD-1/PD-1 ligand interactions in NK functional defect, we analyzed CD107a expression on PD-1 neg and PD-1 pos CD56 dim NK cells from KS patients using redirected degranulation assays with anti-CD16 antibody bound to P815 or P815.PD-L1 cells ( Figure 2C). As expected, PD-1 neg cells strongly degranulated following CD16 triggering, and this response was comparable whatever P815 cells expressed PD-L1 or not. Similar to what was observed with plate-bound NKp30 or NKp46 antibodies, PD-1 pos NK cells were hyporesponsive to stimulation with CD16 antibody-coated P815 cells. However, the presence of PD-L1 on P815 cells did not further inhibit PD-1 pos NK cell degranulation. We thus wondered if PD-L1 expressed on NK cells themselves or on other mononuclear cells might be a source of ligands for PD-1 pos NK cells. In healthy controls, PD-L1 was not observed on peripheral T or NK cells. Conversely in KS patients, PD-L1 was detected on T cells and PD-1 pos CD56 dim NK cells, but not on CD56 bright or PD-1 neg CD56 dim NK cells ( Figure 2D). This suggests that PD-1/ PD-L1 homotypic (NK cell-NK cell) or heterotypic (NK cell-T cell) cell interactions may occur and inhibit NK cell effector functions, in agreement with previous observations in T cells [43,44].
To better understand if the impaired functions of PD-1 pos NK cells resulted from the expression of PD-1 itself, or whether the two phenomena were indirectly linked, we transduced the NKL cell line with a PD-1 lentiviral construct. Stable PD-1 expression was obtained and maintained up to 45 days ( Figure 3A). Expression of PD-1 on transduced NKL cells did not influence the overall expression of classical NK cell differentiation, maturation or activation markers (data not shown). A weak expression of PD-L1 was observed on NKL.PD-1 cells, but not significantly more than on NKL.control cells ( Figure 3B). We next analyzed if PD-1 expression rendered NKL cells hyporesponsive, as observed above for PD-1 pos NK cells in KS patients. NKL cells were cultured in the absence of IL-2 for 24 h prior to 5-hour stimulation with K562 cells (1:1 effector:cell ratio), after which CD107 expression was assessed. K562 cell stimulation did not induce significant degranulation in either NKL.control or NKL.PD-1 cells, in agreement with the previously reported low cytotoxicity of NKL cells against K562 targets [45]. Following stimulation by PMA plus ionomycin however, CD107a expression was strongly decreased in NKL.PD-1 compared to NKL.control cells. Importantly, if NKL.PD-1 cells were cultured in the presence of 100 U/ml of IL-2 before stimulation with PMA plus ionomycin, CD107a levels were strictly comparable to those in NKL.control cells ( Figure 3C).
We thus wondered if exogenous cytokines could also restore functional abilities of PD-1 pos NK cells from KS patients. NK cells were preincubated overnight in the absence or presence of 100 U/ml IL-2 or 50 ng/ml IL-15 prior to stimulation with plate-bound NKp30 or NKp46 antibodies, and processed as above. Both cytokines rescued the capacity of PD-1 pos NK cells to degranulate and produce IFNγ to levels comparable to those of PD-1 neg cells ( Figure 3D).
Collectively, these results indicate that PD-1 expression directly contributes to NK cell functional impairment. However, functional abilities of PD-1 pos NK cells can be rescued by cytokine treatment.

PD-1 is expressed on in vivo activated, mature NK cells
NK cells are heterogeneous with respect to expression of several receptors often corresponding to NK maturation, differentiation and functional stages [46]. Expression of PD-1 was restricted to the mature CD56 dim CD16 + NK cell subset, and was never observed on the more immature CD56 bright CD16 -NK cell subset ( Figure 4A). To determine if hyporesponsiveness of PD-1 pos NK cells was associated with a particular phenotype, we compared the expression patterns of a large panel of receptors in PD-1 pos and PD-1 neg CD56 dim NK cells from KS patients ( Figure 4B). There were no significant differences in expression of HLA class Ispecific inhibitory receptors (monitored by a cocktail of antibodies to Killer cell Immunoglobulin-like Receptors (KIR) KIR2DL1/S1, KIR2DL2/L3/S2, KIR3DL1/S1, KIR2DS4), or in expression of activating receptors such as NKp30, NKp46, NKG2D, NKG2A, NKG2C, NKp80, CD161 and DNAM-1.
PD-1 is expressed on activated T cells and functionally exhausted memory T cells. As shown above in degranulation assays, PD-1 pos NK cells from KS patients exhibited a basal expression of CD107a in the absence of in vitro stimulation (mean 5.2% in PD-1 pos vs. 2.0% in PD-1 neg NK cells, P = 0.007), suggesting that they were activated in vivo. Furthermore, PD-1 pos NK cells showed higher expression levels of the CD69 and CD25 activation markers than their PD-1 neg counterpart. They also expressed higher levels of NKp44, a natural cytotoxicity receptor (NCR) induced on NK cell surface upon activation ( Figure 4C).
In mice, NK cells undergoing homeostatic proliferation after adoptive transfer become exhausted [47]. To determine if expression of activation markers in PD-1 pos NK cells mirrored recent in vivo proliferation, we analyzed NK cells from KS patients for the presence of proliferation and maturation markers ( Figure 4D). Expression of Ki67, a nuclear protein associated with cell-cycle progression and generally used as a marker of recently divided cells, was not significantly different in PD-1 pos and PD-1 neg NK cells. At contrast with the NKG2A -KIR + CD57 + fully mature PD-1 pos NK cell population recently described in CMV-seropositive healthy controls [40], we found a similar proportion of PD-1 pos and PD-1 neg NK cells expressing CD57, a marker of CD56 dim late differentiation suggested to mark memory-like NK cells that have been expanded in response to infection [48]. We also analyzed expression of KLRG1, a marker of post-mature NK cells having undergone proliferation [49,50]. Because commercial antibodies did not provide reliable KLRG1 staining on NK cells, we quantified KLRG1 mRNA levels in sorted PD-1 pos and PD-1 neg NK cells, and found no difference between the two populations. Finally, we observed that PD-1 expression was associated with a higher sensitivity of NK cells to spontaneous apoptosis, estimated by annexin V surface exposure ( Figure 4D).
Altogether, our results indicate that PD-1 is expressed in a sub-population of NK cells showing markers of activation and apoptosis sensitivity, but is independent of their terminal maturation and proliferation state.

PD-1 expression is induced by sustained stimulation via NK cell activating receptors
The molecular mechanisms governing PD-1 expression are best defined in antitumor CD8 T cells. A link to persistent antigen exposure has been suggested [51]. Recently, soluble factors produced in the tumor microenvironment have also been involved in PD-1 induction [52]. We sought to understand the mechanisms contributing to PD-1 expression in NK cells. Healthy control PBMCs were cultured for 1 to 6 days with cytokines known to modulate NK cell proliferation, activation and/or effector functions, after which the cell surface expression of PD-1 was evaluated in parallel on NK and T cells. A very weak induction of PD-1 was observed following exposure to 100 U/ml IL-2, but at levels much lower than on T cells. IL-15, IL-12, IL-18, IFNα, IL-21, or IFNγ did not significantly induce PD-1 expression on NK cells ( Figure 5A and data not shown).
The higher frequency of PD-1 pos NK cells in KS patients compared to asymptomatic HHV8 carriers suggested that factor(s) associated with tumor development or progression could be involved in the induction of PD-1 expression. We assessed expression of PD-1 following stimulation of healthy control NK cells with pro-or anti-inflammatory molecules, growth factors or angiogenic factors previously reported to play a role in KS pathogenesis, such as IL-1β, IL-6, IL-10, TGFβ, VEGF or prostaglandin E2 (PGE2). None of these factors induced PD-1 expression on NK cells (data not shown). We also determined if pathogen-associated molecules might play a role in PD-1 induction, by stimulating control PBMCs with agonists to TLR3 (poly(I:C)), TLR4 (LPS), TLR7/8 (R848) or TLR9 (CpG ODN 2006). No PD-1 induction was observed. Lastly, incubating control NK cells with the serum of KS patients exhibiting high frequency of PD-1 pos NK cells did not demonstrate any effect, suggesting that KS-associated soluble factors were not involved in PD-1 induction (data not shown).
Prolonged antigen-specific stimulation of the TCR maintains PD-1 expression on T cells. We wondered if sustained triggering of NK cell activating receptors could induce PD-1. We stimulated control NK cells for 72 hours with plate-bound antibodies to activating receptors or their respective isotype control, in the presence of IL-15 to maintain NK cell viability. As shown in Figure 5B, triggering of NKp30 or NKp46 reproducibly induced PD-1 expression on control NK cells. Importantly, induction of PD-1 upon stimulation through NKp30 or NKp46 did not extend to other activating receptors, as triggering with plate-bound CD16 antibody had no effect. Moreover, engagement of the NKG2D receptor by immobilized recombinant MICA molecules did not induce PD-1 (data not shown).
That sustained triggering of NKp30 could induce expression of PD-1 was confirmed using P815 cells stably transfected with the NKp30 ligand, B7-H6 [53]. As shown in Figure 5C, PD-1 was induced following 72 h stimulation of control NK cells with P815.B7-H6 but not with untransfected P815 cells, thus recapitulating the results with plate-bound NKp30 antibody. Taken together, these results indicate that sustained stimulation through the NKp30 or NKp46 activating receptors can lead to the induction of PD-1 expression on CD56 dim NK cells.

Relationships between tumor cells and PD-1 expression on NK cells
Tumorigenesis can be accompanied by cell surface upregulation of NK cell activating ligands together with downregulation of MHC class I expression. The loss of MHC class I-specific inhibitory signals favors NK cell activation, and renders tumor cells sensitive to elimination by NK cells. Recent data in mice indicate that MHC class I-deficient tumor cells can escape from the immune response by inducing NK cell anergy, which can be reversed by exogenous cytokines [54]. We thus wondered if PD-1 was upregulated on NK cells following sustained stimulation cells expressing activating receptors in the absence of MHC class I inhibitory signals. Indeed, PD-1 was reproducibly induced on healthy control NK cells following 72 hours of stimulation with K562 cells, which express several ligands of activating receptors including B7-H6 [55] ( Figure 6A). In HHV8-infected cells undergoing lytic replication, MHC class I expression is downregulated due to the effect of the viral K3 and K5 ubiquitin ligases expressed during the early lytic cycle [56,57]. However HHV8 infection is predominantly latent within KS lesions [58]. As a model of HHV8-infected cells relevant to KS, we previously generated KS-derived cells persistently infected by a recombinant HHV8, which showed the same latency program as KS spindle cells [34]. Compared to uninfected cells, these HHV8-infected KS cells exhibited decreased but yet strong residual expression of MHC class I molecules, together with weak expression of NKp30 ligand [34]. We therefore, stimulated control NK cells up to 72 hours with HHV8-infected or uninfected KS-derived cells. No PD-1 induction was observed ( Figure 6B), suggesting that PD-1 is best induced upon NK encounter with target cells expressing several activating ligands. Moreover, culturing NK cells with cell-free supernatants of HHV8-infected or uninfected KS cells did not demonstrate any effect (data not shown), indicating that soluble factors released by KS cells are likely not involved in PD-1 induction on NK cells.
Finally, we aimed to determine the potential relationship between PD-1 pos NK cells and PD-L1 expression within KS lesions. We previously reported that CD56-positive cells were very scarce in KS biopsy samples, suggesting that NK cells did not reach tumor lesions or could not survive in the tumor microenvironment [34]. As a consequence, it was not possible to detect intratumoral PD-1-positive NK cells. We therefore, analyzed PD-L1 expression within KS lesions ( Figure 6C), and observed a focal staining of spindle-shaped tumor cells. Importantly, PD-L1 was predominantly expressed in inflammatory cell clusters in the tumor environment. This suggests that, like in other malignancies, PD-L1 expressed at the surface of KS tumor cells or inflammatory cells can mediate apoptosis of PD-1-positive cells.

DISCUSSION
PD-1 has recently emerged as a key inhibitory checkpoint involved in T cell exhaustion and tumor escape [25][26][27][28][29]. Manipulation of the PD-1 pathway is an area of great interest as a strategy for circumventing tumor escape from T cell-mediated responses [30][31][32][33]. Here, we show that PD-1 may also represent an inhibitory checkpoint involved in NK cell exhaustion in patients with Kaposi sarcoma, and to a lower extent in patients with some chronic viral infections.
The presence of PD-1 pos NK cells was already reported in various clinical settings [35][36][37][38][39][40], although there were sometimes notable differences between the studies. In particular, PD-1 expression was reported on large expansions of CD56 bright NK cells in patients with chronic HCV infection, particularly in treatment non-responders [38]. At contrast, we found that PD-1 expression in HCV chronically infected patients was scarce, and always confined to the CD56 dim NK cell population, in agreement with others [35]. Whether such discrepancies are related to usage of distinct PD-1 mAbs or staining strategy remains to be determined. Indeed, different anti-human PD-1 clones have been shown to detect high and/or intermediate PD-1-positive cells [59], and the clone we used targeted cells with high rather than intermediate PD-1 expression. Large expansions of PD-1 pos NK cells were also reported in patients with renal cell carcinoma, multiple myeloma, EBV-associated lymphoproliferative disorder and ovarian carcinoma [35][36][37]40]. Therefore, PD-1 may represent an inhibitory checkpoint expressed on NK cells in various cancers.
We found that PD-1 pos NK cells were hyporesponsive to short stimulation by activating NK cell receptors or by NK-cell sensitive K562 target cells. Importantly, this NK cell functional defect was not related to abnormal expression of activating or inhibitory receptors, was not associated with markers of proliferation or senescence, and was rescued by exogenous IL-2 or IL-15. Transducing PD-1 in NKL cells also induced their hyporesponsiveness. Thus, it is likely that PD-1 impedes NK-cell responses by directly inhibiting effector functions rather than by altering expression of other receptors. PD-1-mediated hyporesponsiveness may be due to alterations in intracellular signaling components, or to alterations in the spatial organization of membrane receptors, as described in mouse hyporesponsive NK cells [60]. Alternatively, PD-1 may act by upregulating genes in exhausted NK cells that impair their function. In HIV-specific CD8 T cells, PD-1 upregulated BATF, a transcription factor in the AP-1 family, and enforced expression of BATF was sufficient to impair T cell proliferation and cytokine secretion [61]. Whether PD-1 coordinately upregulates a program of genes in NK cells is currently under investigation by analyzing gene expression profiles from PD-1 pos cells.
We aimed at understanding the mechanisms involved in the chronic expression of PD-1 on NK cells. In mice, PD-1 can be induced on NK cells by various inflammatory factors including IL-2, IL-18, IFNγ, TLR3 or TLR9 agonists [62]. Among several cytokines and inflammatory molecules tested, we found that only high dose IL-2 induced a very weak expression of PD-1 on control NK cells. PD-1 expression is also rapidly enhanced on mouse   [63,64]. In KS patients however, PD-1 expression on NK cells was stable over several years, making it unlikely that it was related to an acute phenomenon. Moreover, we found no evidence for viral reactivation in patients with PD-1 pos NK cell expansions and found no correlation with the CMV serological status, at contrast with the fully mature NKG2A -KIR + CD57 + PD-1 pos NK cell population recently described in CMV-seropositive healthy individuals [40]. None of the immunosuppressive soluble factors potentially produced in the tumor microenvironment induced expression of PD-1 on NK cells. In particular, we observed no effect of VEGF-A, a key angiogenesispromoting factor secreted by Kaposi sarcoma and other tumor cells, and recently shown to enhance expression of PD-1 and other inhibitory checkpoints on intra-tumoral CD8 T cells [52].
The fact that PD-1 pos NK cells from KS patients exhibited some markers of activation ex vivo suggested us that PD-1 might be induced upon contact with (tumor) cells expressing a restricted set of activating ligands. Indeed, in vitro triggering of NKp30 or NKp46 receptors, but not of CD16 or NKG2D, significantly induced PD-1 on control NK cells. PD-1 was also expressed following sustained stimulation with murine cells overexpressing B7-H6, or with HLA class I-negative tumor cells expressing various activating ligands. Within KS lesions, HHV8 infection is predominantly latent and tumor cells express MHC class I molecules [34]. Indeed, in vitroderived KS cell lines were unable to induce PD-1 on NK cells. It may be possible that only the small fraction of HHV8-infected cells that undergo reactivation of the lytic cycle, i.e. those exhibiting MHC class I loss together with expression of activating ligands, is able to induce PD-1 on NK cells. Whether such cells are present in low amount within the tumor and/or can be found outside the tumor, such as within tumor-draining lymph nodes or distant lymph nodes in HHV8-infected individuals remains to be determined.
Our results are reminiscent of a recent report in mice showing that NK cell anergy is induced as a result of persistent stimulation by MHC class I-deficient tumor cells in the absence of an activating inflammatory environment [54]. Whether mouse anergic NK cells expressed PD-1 was not determined. However, similar to PD-1 pos NK cells from KS patients, they were anergic to stimulation through cell surface activating receptors, but not to stimulation by cytokines. We therefore, propose that PD-1 represents a mediator of functional exhaustion in NK cells that have been subjected to prolonged exposure to cells expressing activating ligands, such as tumor cells. By analogy with CD8 cells, PD-1 would characterize a subset of NK cells that adapt to persistent stimulation by upregulating the PD-1 inhibitory receptor and turning off their responsiveness to stimulating conditions. PD-1 expression may serve to limit the activity of NK cells during antitumor immune response, thereby protecting tumor cells from NK-cell mediated elimination. Releasing the PD-1 immune checkpoint in NK cells might lead to NKcell proliferation, intratumoral infiltration and increased effector functions, as shown for CD8 T cells [65].

Subjects
KS patients and HHV8-infected individuals were described previously [34] (Table 1). Briefly, there were 34 patients with a history of KS (12 HIV-related and 22 classical KS) and 25 asymptomatic HHV8 carriers (15 HIV+ and 10 HIV-subjects). The HIV patient group consisted of 14 HHV8-negative ART-treated HIV patients. Importantly, all HIV+ patients in the different groups were aviremic upon antiretroviral treatment. The HCV patient group (ANRS HC-EP28 study) consisted of 41 chronically infected patients including 32 viremic (naive of treatment, or who discontinued treatment at least 6 months before study) and 9 aviremic subjects (sustained viral response following treatment). Healthy controls were 36 blood donor volunteers (HIV-, HCV-and HHV8-negative). The study was performed in accordance with the Declaration of Helsinki and French legislation, and received approval of the Saint-Louis and Pitié-Salpêtrière Hospital Ethical Committees (P040105 and A00822-37, respectively). All participants provided written informed consent.

Flow cytometry
Ex vivo surface and intracellular staining was performed immediately after peripheral blood mononuclear cell (PBMC) isolation (Lymphoprep; Abcys). Cells were first incubated for 20 min at 4°C in the dark with FITCconjugated or APC-conjugated anti-PD-1 antibody (MIH4, BD Biosciences, 1:50) or the respective isotype control. Dead cells were excluded with 7-Amino Actinomycin (7-AAD, Via-Probe, BD Biosciences). PE-Annexin V (BD Biosciences) staining was performed according to the manufacturer's instructions. For intracellular detection of Ki-67 and perforin, cells were fixed in 1% formaldehyde, permeabilized with 0.2% saponin and stained with PEconjugated anti-Ki67 (B56, BD Biosciences, 1:10) or FITC-conjugated anti-perforin (dG9, BD Biosciences, 1:10). Samples were run using BD FACSCalibur flow cytometer in the first part of the study (quantification of PD-1 pos cells in patients and controls). Thereafter, all analysis was performed on BD LSR Fortessa. A total of at least 100,000 events in a live gate was collected. Gates were defined through isotype and fluorescence minus one (FMO) stains. The gating strategy was forward scatter versus side scatter, and NK cells were defined as CD3 -CD56+ CD16+/-lymphocytes. Data were analyzed using FlowJo software.
Alternatively, freshly purified control NK cells were stimulated for 24-72 hours in flat-bottomed, high-proteinbinding plates (Thermo Fisher Scientific) coated with 10 μg/ml of NKp30, NKp46, or CD16 antibodies or their respective isotype control, or 10 μg/ml of recombinant soluble MICA molecules [66], in the presence of IL-15 (50 ng/ml) to maintain NK cell viability.

PD-1 lentivirus generation and NKL cell transduction
Human PD-1 cDNA was PCR amplified from a pEYFP-N1-PD1 construct [67]. The PCR product was cloned into the lentiviral plasmid pDestDH1 containing the puromycin resistance gene using the Gateway technology (Invitrogen). The lentiviral plasmid pcDH1 containing the puromycin resistance gene was used as control vector (both plasmids were kindly provided by Philippe Benaroch and François-Xavier Gobert, Institut Curie, Paris, France). Replication-defective lentivirus particles pseudotyped with the envelope G protein of vesicular stomatitis virus G were produced in HEK 293T cells cotransfected with a mix of plasmids: psPAX2, pMD2.G (supplied by D. Trono; now Addgene plasmids 12260 and 12259, respectively) and either pDestDH1 PD-1 or pcDH1 plasmid. Supernatants were collected 48 h later and concentrated by ultracentrifugation (120,000 ×g for 90 min).
For degranulation experiments, untransfected or transfected NKL cells were cultured in the absence of IL-2 for 24 h prior to stimulation with K562 target cells (effector:target 1:1) or PMA plus ionomycin, after which CD107a surface expression was determined.

Statistical analysis
All statistical tests were performed with Prism software (GraphPad). Comparisons between two groups were performed using the Wilcoxon or Mann-Whitney t tests for paired and unpaired groups, respectively. Comparisons between more than two groups were performed using repeated measures or one-way ANOVA for paired and unpaired groups, respectively, followed by Tukey's multiple comparison test. Two-sided p values less than 0.05 were considered significant.