Bench-to-bedside review: Natural killer cells in sepsis - guilty or not guilty?

Bacterial sepsis and septic shock are complex inflammatory disorders associated with a systemic inflammatory response syndrome. In the most severe cases of infection, an overzealous release of pro-inflammatory cytokines and inflammatory mediators by activated leukocytes, epithelial cells and endothelial cells, known as a 'cytokine storm', leads to deleterious effects such as organ dysfunction and even death. By the end of the 20th century, natural killer (NK) cells were for the first time identified as important players during sepsis. The role of this cell type was, however, double-edged, either 'angel' or 'devil' depending upon the bacterial infection model under study. Bacterial sensors (such as Toll-like receptors) have recently been shown to be expressed at the protein level in these cells. In addition, NK cells are important sources of interferon-γ and granulocyte-macrophage colony-stimulating factor, which are pro-inflammatory cytokines necessary to fight infection but can contribute to deleterious inflammation as well. Interestingly, an adaptative response occurs aimed to silence them, similar to the well-known phenomenon of endotoxin reprogramming.

Systemic infl ammatory response syndrome (SIRS) shares the initial clinical characteristics described for sepsis patients and is based on non-specifi c criteria during daily observations of patients in ICUs [1]. Th is syndrome (commonly observed in patients after major trauma, burns, and ischemia, among others) might promote sepsis occurrence. Both pathogen-associated molecular patterns (PAMPs) and damage associated molecular patterns (DAMPs), as exogenous and endogenous mediators, respectively, can similarly trigger the initial infl ammatory response. PAMPs are recognized by innate sensors termed pattern recognition receptors (for example, NOD-like receptors (NLRs), Toll-like receptors (TLRs)). Th e consequence of this pathogen sensing is the production of pro-infl ammatory mediators for the eradication of invading microorganisms, and in parallel the production of anti-infl ammatory mediators to control this response [2]. Th e pro-infl ammatory process can induce tissue damage and organ failure, while the antiinfl ammatory response involves leukocyte reprogramming, a natural phenomenon that renders leukocytes tolerant and hypo-reactive to activating signals in terms of infl ammatory contribution while maintaining their anti-infectious properties. Th e phenomenon has been claimed to possibly lead to increased risk for nosocomial infections [3].
Th e concept of natural killer (NK) cells was fi rst reported in 1971 by Miller and collaborators [4], and later named and better described by Hans Wigzell's group [5], which established that leukemia cell lines were lysed by cells with the morphology of small lymphocytes and with diff erent T-and B-cell characteristics. Soon after, it was reported that NK cells were also active against virusinfected cells [6]. NK cells are able to induce death of target cells expressing 'nonself ' antigens or inaccurate levels of major histocompatibility complex (MHC) type I molecule. Th is patrolling mechanism is controlled by a large family of killer-cell immunoglobulin-like receptors (KIRs), among others, which bind and activate or inhibit NK cell cytotoxicity [7]. In human, at least two subsets of circulating NK cells have been described, the CD3 -CD56 dim and CD3 -CD56 bright subsets. Th e CD56 dim subset displays enhanced cytotoxicity whereas CD56 bright NK cells produce greater amounts of cytokines [8].
Th e role of NK cells in bacterial innate immunity took longer to be demonstrated. In contrast to phagocytes, the activation of NK cells by PAMPs can only occur through complex crosstalk with other immune cells that creates the proper cytokine microenvironment required for NK cells responsiveness [9]. Accordingly, similarly to any other cellular or molecular participant in infectious

Abstract
Bacterial sepsis and septic shock are complex infl ammatory disorders associated with a systemic infl ammatory response syndrome. In the most severe cases of infection, an overzealous release of proinfl ammatory cytokines and infl ammatory mediators by activated leukocytes, epithelial cells and endothelial cells, known as a 'cytokine storm' , leads to deleterious eff ects such as organ dysfunction and even death. By the end of the 20th century, natural killer (NK) cells were for the fi rst time identifi ed as important players during sepsis. The role of this cell type was, however, double-edged, either 'angel' or 'devil' depending upon the bacterial infection model under study. Bacterial sensors (such as Toll-like receptors) have recently been shown to be expressed at the protein level in these cells. In addition, NK cells are important sources of interferon-γ and granulocytemacrophage colony-stimulating factor, which are pro-infl ammatory cytokines necessary to fi ght infection but can contribute to deleterious infl ammation as well. Interestingly, an adaptative response occurs aimed to silence them, similar to the well-known phenomenon of endotoxin reprogramming. diseases, NK cells can play a 'guilty' or 'not guilty' role in the deleterious infl ammatory process, depending on the circumstances and most probably the timing of the event. Th us, the same actors that contribute to fi ght infection can guiltily act in synergy, leading to acute deleterious infl ammation by producing powerful infl ammatory mediators [10]. Th is is particularly the case for interferon (IFN)-γ and granulocyte-macrophage colony-stimulating factor (GM-CSF), two pro-infl ammatory cytokines produced by NK cells [11].
Th e fact that pathogen sensors (for example, TLRs) were recently discovered to be expressed by NK cells has opened a new interest in their putative involvement in innate immune response to bacterial infections [10,12]. Recently, we have shown that both murine spleen and human blood NK cells express the bacterial sensors TLR2, TLR4 and TLR9 at the protein level and that they are responsive to their agonists in terms of IFN-γ production in the presence of accessory cytokines [13][14][15]. In contrast to phagocytes, the activation of NK cells by PAMPs often requires complex crosstalk with other immune cells, as already shown with dendritic cells, polymorphonuclear cells, and so on. Th ese accessory cells contribute to the cytokine microenvironment (for example, IL-12 and IL-18, cytokines that are strong NK activators and are produced by accessory cells as a parallel response to PAMPs) required for NK cell responsiveness [9]. Since 1984 [16], however, several studies and several lines of evidence have suggested a direct response of NK cells to PAMPs in the presence of an adequate cytokine environment, without the need for contact with accessory cells (see [10] for review). Th is reinforces the hypothesis that they can contribute to the overzealous infl ammation in sepsis. CD69 is an activation marker upregulated upon stimulation of NK cells. We recently observed that the expression of TLR2, TLR4, and the early activation marker CD69 was upregulated in NK cells of septic patients compared to those of healthy volunteers, suggesting blood NK cells are activated during the early stages of sepsis. Interestingly, the expression of CD69 was even higher for SIRS patients, who have sterile infl ammation, suggesting that CD69 might be a marker of acute infl ammation rather than infection [14].

NK cells as benefi cial actors to fi ght infection
Cytokines are key mediators required to orchestrate the anti-infectious process. Besides IFN-γ and GM-CSF, NK cells can produce a large panel of cytokines, including TNF [17], that have been shown to be protective against diff erent types of bacterial infections. It is not surprising then that many investigators have reported the benefi cial contributions of NK cells in fi ghting infections. NK cells have been shown to be protective in diff erent models, including infections with Mycobacterium avium, Shigella fl exneri, Chlamidia trachomatis, Staphylococcus aureus, Pseudomonas aeruginosa, Listeria monocytogenes, Bordetella pertussis, Legionella pneumophila, Shigella fl exneri, Salmonellae, Burkholderia pseudomallei, Mycobacterium tuberculosis, Rickettsiae, Yersinia enterocolitica, Chlamydo phyla abortus or polymicrobial sepsis (see [10] for a review). In addition, NK cells were shown to be the main IFN-γ producing cells in response to bacterial lipopolysaccharide (LPS) [18,19]. In addition, NK cells can also be the source of other anti-infectious mediators, such as the anti-microbial peptides and α-defensins [20]. Furthermore, their benefi cial contribution to protection occurs in combination with various cellular cross-talk with other immune cells that also are important in the process. Interestingly, IFN-γ production, which underpins the eff ective NK cell response to infection, also underpins deleterious NK cell-mediated infl ammation.

NK cells as a guilty participant of the overzealous infl ammation in sepsis
Deleterious roles of NK cells have been reported in numerous animal models (see [10] for a review). Particularly, the capacity of NK cells to favor the infl ammatory response, to promote tissue injury and to contribute to death has been reported after polymicrobial intraabdominal sepsis [21], Escherichia coli intraperitoneal injection [22], Streptococcus pyogenes intravenous injection [23], Ehrlichia-induced toxic shock-like syndrome [24] and in cytokine-induced SIRS [25]. Similarly to the association of the benefi cial role of NK cells with their production of IFN-γ, their guilty role is also associated with their production of IFN-γ. Indeed, this cytokine, alone or in synergy with others, can lead to organ failure and death [2]. Likewise, GM-CSF can further amplify the infl ammatory response and be deleterious [26,27], and even lead to death as shown in a human patient treated with this cytokine [28]. Other mediators can also contri bute to the deleterious eff ects of NK cells, such as granzyme M [29].
More evidence of a guilty participation was reported in a murine polytrauma model (consisting of femur fracture, hemorrhagic shock and subsequent sepsis), in which NK cell depletion resulted in 50% mortality reduction, a decrease of neutrophil infi ltration in diff erent compartments and lymphocyte apoptosis in the spleen [30]. In addition to the eff ect of PAMPs and cytokines, other mediators such as the anaphyloxin C5a can favor the infl ammatory role of NK cells by increasing their IFN-γ and TNF-α production and contributing to mortality during E. coli-induced sepsis [31].

Natural mechanisms restricting excessive NK cell-mediated infl ammation
Concomitant with the pro-infl ammatory response, a compensatory anti-infl ammatory response occurs that can lead to a syndrome associated with increased sensitivity of patients to nosocomial infections [32,33]. Th is phenomenon prominently involves the refractoriness of monocytes/macrophages to challenge with LPS or other PAMPs, an observation also known as endotoxin tolerance [34]. Similarly, in murine spleen cells after experimental polymicrobial sepsis and in blood samples from ICU patients (bacterial sepsis or SIRS), IFN-γ produc tion in response to TLR agonists was lost, resembling the tolerance already described for monocytes [13,14]. In concert, NK cell immunosuppression was also observed in blood samples from trauma patients with brain injury, where poor NK cell recruitment into a BCG-induced granuloma model was noticed [35]. In parallel to the NK cell suppressed state in sepsis, T regulatory cells (Tregs) were shown to be increased as a percentage in the peripheral blood of patients compared with healthy controls [36][37][38][39].
Tregs are immune regulatory players that can inhibit the diff erentiation, activation, proliferation, cytokine secretion or migration of several other leukocytes (for example, by secretion of anti-infl ammatory cytokines such as IL-10 and transforming growth factor (TGF)-β1) [40]. Tregs have been shown to contribute to the antiinfl ammatory compensatory process, in part by inhibiting LPS-induced activation of monocytes by a Fas/Fas ligand death mechanism [41]. After experimental sepsis in mice, we recently showed that the tolerance to PAMPs, in terms of IFN-γ production by purifi ed NK cells, was reversed by depletion of Tregs or TGF-β receptor inhibition prior to sepsis induction [13]. Accordingly, they also contribute to the obstruction of posterior tumor immunosurveillance, a classic NK cell eff ector function, in a profound immunosuppressive environment resulting from surviving severe sepsis [42].
Of note, measurements of NK cell function in humans have been made using NK cells exclusively from peripheral blood. Th us, one cannot speculate about NK cells from other tissues (for example, spleen and liver), which have not been evaluated and may behave diff erently. Nevertheless, the data available for spleen NK cells in mouse models show results very similar to human blood NK cells. Spleen NK cells need the same accessory cytokines for IFNγ and GM-CSF production and might undergo endotoxin tolerance after sepsis [13]. In both mouse spleen and human blood, TLR2 and TLR4 are intracellular in naive NK cells [13,14]. Only a diff erence in the response to LPS has been observed, where human blood NK cells were found to be more responsive than murine spleen NK cells [43].
In agreement with other reports, we also observed that the number of circulating NK CD56 + cells was signifi cantly decreased in sepsis and SIRS patients [44]. Moreover, we demonstrated for the fi rst time that both CD56 bright and CD56 dim subpopulations of CD56 + cells were reduced in both SIRS and sepsis [14]. Th is decreased cell number is the refl ection of a general lymphopenia, potentially due to the traffi cking of NK cells to sites of infection [21,45,46] or to apoptosis [47]. Interestingly, some reports suggest that NK cell percentages or counts are associated with patient outcome. In one study, CD4 + lymphocyte lymphopenia and increased levels of NK cells in patient blood were associated with a survival benefi t [48]. In contrast, other studies suggested that the increase of NK cell levels in blood was associated with early mortality [49,50]. Deeper investigations are still required to evaluate the predictive role of NK cell count for patient mortality before considering this information useful as a prognostic tool.
Regarding functional studies, NK cells were shown to have reduced cytotoxic activity in sepsis patients [51,52] and in SIRS patients following thermal and traumatic injury [53,54]. However, a recent review suggested a potential infl ammatory participation of NK cells in SIRS and early sepsis, associated with an acquired dysfunction of cellular functions at a later stage that could favor nosocomial infections and mortality [55]. It has been shown that IFN-γ production was altered in patients after elective surgery and severely impaired in patients with sepsis [56]. We also observed that the production of this cytokine was abolished ex vivo in whole blood of SIRS and sepsis patients after stimulation with accessory cytokines and TLR agonists [14]. In diff erent experimental conditions, however, other studies have made diff erent observations. Giannikopoulos and colleagues [57] showed that purifi ed NK cells from sepsis patients exhibit enhanced IFN-γ production after in vitro LPS stimulation. On the other hand, decreased IFN-γ production by purifi ed NK cells from sepsis or septic shock patients was observed when co-cultured with the K562 cell line, while cells purifi ed from SIRS patients displayed increased IFN-γ production [58]. In addition to NK cell lymphopenia and decreased IFN-γ production in the blood of ICU patients, another study observed this cellular impairment preceding cytomegalovirus (CMV) reactivation in critically ill patients [59].
Th ese complementary observations show that purifi ed NK cells from patients behave diff erently to those stimulated in whole blood, and that IFN-γ production depends on the culture conditions. Th e fact that purifi ed NK cells can be stronger producers of IFN-γ in certain conditions (during LPS [57] or K562 cell [58] stimulation) does not necessarily imply their participation in deleterious infl ammation, as they are in fact present in a suppressive serum environment containing anti-infl ammatory mediators (such as IL-10, TGF-β1, corticoids, and so on) that might inactivate their infl ammatory function [60]. Finally, decreased IFN-γ production by NK cells in whole blood [14] and enhanced IFN-γ production for purifi ed cells [57,58] together support the hypothesis that NK cells are not themselves under endotoxin tolerance (mechanism proposed for macrophages) but are rather inhibited by suppressive environmental factors (for example, TGF-β and Tregs [13]). Taken together, this evidence reveals that Tregs are key players in the suppression of leukocytes during sepsis progression and that NK cells might not be guilty in the overzealous infl ammatory reaction by being found to be suppressed during the disease.

Conclusion
Data gathered from recent reports show that peripheral NK cells are guilty of contributing to the overzealous infl ammation process during sepsis by producing proinfl ammatory cytokines. However, NK cells might be considered not guilty on the basis of the balance of signals between accessory and inhibitory cells that can suppress their pro-infl ammatory cytokine production (as suggested by Figure 1). Advances in NK cell research point to this cell population as a promising marker (cell counting and receptor expression (CD69, TLRs)) to be considered and evaluated during disease progression to predict patient outcome or provide supportive information for patient classifi cation (SIRS or sepsis).

Competing interests
The authors declare that they have no competing interests.

Figure1. Natural killer cell function during bacterial infection is modulated by the balance of factors into a complex environment.
Accessory cells (AC), such as dendritic cells, monocytes or others, provide activating signals that trigger synergistic activation of natural killer (NK) cells together with pathogen-associated molecular patterns (PAMPs) to produce cytokines. On the other hand, activated T regulatory cells (Tregs) can counterbalance this by providing suppressive cytokines (for example, transforming growth factor (TGF)-β and IL-10) to abolish NK cell activation. GM-CSF, granulocyte-macrophage colony-stimulating factor; IFN, interferon.