Correspondence to: D. Kabelitz, Institute of Immunology, University of Kiel, Arnold-Heller-Str. 3, Bldg. 17, D-24105 Kiel, Germany. E-mail: firstname.lastname@example.org
The activating natural killer group 2 member D (NKG2D) receptor is expressed on NK cells, cytotoxic T cells and additional T cell subsets. Ligands for human NKG2D comprise two groups of MHC class I-related molecules, the MHC class I chain-related proteins A and B (MICA/B) and 6 UL16-binding proteins (ULBP1-6). While NKG2D ligands are absent from most normal cells, expression is induced upon stress and malignant transformation. In fact, most solid tumours and leukaemia/lymphomas constitutively express at least one NKG2D ligand and thereby are susceptible to NKG2D-dependent immunosurveillance. However, soluble NKG2D ligands are released from tumour cells and can down-modulate NKG2D activation as a means of tumour immune escape. In some tumour entities, levels of soluble NKG2D ligands in the serum correlate with tumour progression. NKG2D ligands can be proteolytically shed from the cell surface or liberated from the membrane by phospholipase C in the case of glycosylphosphatidylinositol (GPI)-anchored molecules. Moreover, NKG2D ligands can be secreted in exosomal microvesicles together with other tumour-derived molecules. Depending on the specific tumour/immune cell setting, these various forms of soluble and/or exosome-bound NKG2D ligands can exert multiple effects on NKG2D/NKG2D ligand interactions. In this review, we focus on the role of various proteases in the shedding of human NKG2D ligands from tumour cells and discuss the not completely unanimous reported functional implications of soluble and exosome-secreted NKG2D ligands for immunosurveillance.
The activity of natural killer (NK) cells is controlled through signals delivered by natural killer cell receptors (NKR) that include several groups of activating or inhibitory transmembrane molecules. Prominent NKR members are the killer-cell-Ig-like receptors (KIR), the natural cytotoxicity receptors (NCR), and the lectin-like receptors. Depending on their intracellular immunoreceptor tyrosine-based inhibition motif (ITIM) or their association with immunoreceptor tyrosine-based activation motif (ITAM)-expressing adaptor molecules, the 14 polymorphic HLA class I-specific KIR receptors transduce either inhibitory or activating signals . The NCR NKp46, NKp30 and NKp44 receptors also couple to ITAM-carrying adaptor molecules and thus trigger activation of NK cells; in addition, the intracellular part of NKp44 contains an ITIM, although this motif does not seem to transmit inhibitory signals [2, 3]. While the KIRs recognize HLA class I molecules that are constitutively expressed on nucleated cells (but are subject to deregulation upon viral infection or cellular transformation), NCRs recognize different sets of exogenous ligands including pathogen-derived molecules (e.g., viral proteins) and endogenous ligands expressed on tumour cells such as human leucocyte antigen-B-associated transcript 3 (BAT3), the B7 receptor family member B7-H6 or the proliferating cell nuclear antigen (PCNA) . The third group of NKR is encoded in the natural killer gene complex (NKC) on human chromosome 12 and encompasses killer-cell-lectin-like receptor (KLR) and C-type lectin receptor (CLEC) genes . Again, KLR includes both activating and inhibitory receptors. A prominent member of this family is the NK group 2 member D (NKG2D) receptor, which transmits an activating signal upon binding of stress-inducible MHC class I-related molecules . Of note, the regulation of NK cell function relies on a complex and tightly controlled interplay of a multitude of both activating and inhibitory receptors and the cooperation of distinct receptors might even allow the adjustment of NK cell activity in response to a certain stimulus [6-8]. Thus, NK cell effector function (cytokine production, cytotoxicity or ‘tolerance’) is tightly controlled by a surprising variety of receptors that recognize self and non-self ligands. Moreover, increasing evidence indicates that the responsiveness of NK cells is not precisely fixed but may adapt to environmental signals in vivo . Importantly, however, the expression of NKR is not entirely restricted to NK cells. Several NKR are also expressed on NKT cells and T cells and thus can modulate adaptive T cell responses. In this review article, we focus on the plasticity of the human NKG2D/NKG2D ligand system and discuss the current knowledge of how soluble NKG2D ligands are generated and can modulate immune responses.
Expression and function of the NKG2D receptor
NKG2D is a homodimeric type II transmembrane C-type lectin receptor that is expressed on all NK cells, most NKT cells and γδ T cells, the majority of CD8 αβ T cells and a minor proportion of CD4 αβ T cells with regulatory activity [5, 10, 11]. Upon binding of corresponding NKG2D ligands (NKG2DL), human NKG2D transmits an activating signal via the associated transmembrane adaptor protein DAP10. Of note, in mice, NKG2D associates with DAP10 and DAP12 and in contrast to DAP12, DAP10 does not contain an ITAM, but a YINM motif that is involved in costimulatory pathways . Upon phosphorylation, DAP10 recruits the p85 subunit of phosphoinositide-3-kinase (PI3-K) and Grb2-Vav1, which stimulates survival and cytotoxicity of NK cells and provides costimulatory signals to T cells (Fig. 1A) [5, 12]. However, NKG2D signalling seems not to be sufficient to elicit cytotoxicity in primary human NK cells [6, 8]. Thus, the exact mechanism underlying DAP10-YINM mediated cytotoxic granule polarization and secretion still remains to be elucidated [13, 14]. The NKG2D-DAP10 axis is regulated by cytokines. While IL-2 increases NKG2D surface expression and up-regulates DAP10 protein synthesis, opposite effects are mediated by TGF-β . Over-stimulation by extended exposure to corresponding ligands can actually result in functional impairment of T cells and NK cells, due to NKG2D-mediated Fas/Fas ligand-dependent caspase-3/-7 activation and subsequent CD3ζ chain degradation (Fig. 1B) . NKG2D triggers cytolytic effector function in killer cells and costimulates cytokine production and proliferation in T cells and NKT cells [5, 17-19]. For instance, the NKG2D-mediated costimulatory signal is essential for cytokine production of tissue-resident CD8 T cells . Moreover, activation of T cells via NKG2D has been shown to costimulate the production of proinflammatory IL-17 in human T cells under chronic inflammatory conditions and during infection [21, 22]. As tumour cells frequently overexpress ligands for NKG2D (and thereby trigger NK cell and T cell responses), the NKG2D/NKG2DL system is thought to play an important role in tumour immunosurveillance. This view is supported by studies with NKG2D-deficient mice (which develop tumours at increased rates)  and the demonstration that epithelial-mesenchymal transition (EMT), a process linked with invasiveness and metastasis of developing tumours, induces anti-tumour responses via NKG2D . All these studies underline the important role of the NKG2D receptor in controlling NK cell and T cell responses. Interestingly, functional expression of NKG2D has been recently detected also on some tumour cells. Here, it was found that NKG2D couples to PI3-K and subsequently activates the MAP kinase pathway, leading to the intriguing notion that the NKG2D/NKG2DL interaction between tumour cells may in fact contribute to tumour cell growth .
Multiple ligands for human NKG2D
The ligands for human NKG2D comprise two families of MHC class I-related cell surface molecules that are constitutively expressed on tumour cells or can be induced on non-transformed cells upon cellular stress, DNA damage and infection. Human MHC class I chain-related proteins A and B (MICA/B) are transmembrane proteins that carry three MHC class I-like domains (α1, α2, α3) and exhibit considerable allelic variation. In fact, more than 80 MICA and more than 30 MICB sequences have been described . The second family of human NKG2D ligands comprises six members of UL16-binding proteins (ULBP1-6), which have two MHC class I-like domains (α1, α2) and are either transmembrane proteins (ULBP4,5) or glycosylphosphatidylinositol (GPI)-linked molecules [26, 27]. ULBP2 has the unique feature that it can be expressed at the cell surface either as a transmembrane or a GPI-anchored protein . It is not precisely clear why there is such a multitude of ligands for just one cell surface receptor. A commonly entertained scenario relates to the view that the various ligands are differentially regulated by various stress pathways, thus enabling the NKG2D-expressing cell to adapt to different environmental conditions. Moreover, the resulting NKG2D signalling cascade is also influenced by the seemingly different affinity of the various ligands for NKG2D . The expression of human NKG2DL is regulated at the transcriptional, translational and post-translational level. This includes a role for the heat shock pathway, NF-κB and Sp family transcription factors, the DNA damage response (involving activation of ATM/ATR kinases), several microRNAs, viral or cellular oncogenes, proinflammatory signals from, for example, Toll-like receptor activation, and cytokines, notably type I and type II interferons . Undamaged, non-activated and non-transformed cells do not express NKG2DL, thus ensuring the efficacy of the NKG2D/NKG2DL system in immunosurveillance . Upon activation and maturation, however, NKG2DL expression is induced on monocytes and dendritic cells (DC) where NKG2DL contribute to T cell costimulation by mature DCs [30-32]. Moreover, low-level NKG2DL expression was also detected on antigen-activated human T cells, and elimination of activated NKG2DL-expressing T cells by NK cells was suggested to contribute to the termination of T cell immune responses .
Cell surface expression of NKG2DL is subject to epigenetic regulation. Both histone deacetylase (HDAC) inhibitors such as valproic acid and demethylating agents such as 5′-azacytidine increase the cell surface expression of MICA/B and ULBP2/3 [34-39]. Furthermore, up-regulation of cell surface NKG2DL expression is also stimulated in response to all-trans-retinoic acid [34, 38], certain cytokines such as IL-18 , or HER2/HER3 receptor tyrosine kinase signalling . Additional milieu-specific regulatory pathways have been described. In this regard, recent studies point to a pivotal role of the intestinal microflora for the regulation of NKG2DL expression on murine intestinal epithelial cells . On the other hand, a variety of regulatory signals have been identified that downregulate cell surface NKG2DL expression without inducing proteolytic shedding. These include not only several specific (partly interferon-γ-regulated) microRNAs including miR-10b, miR-34a/c and miR-520b [43-46] but also cytokines (e.g. TGF-β)  and hypoxia (via hypoxia-inducible factor-1α) . Moreover, some viruses including cytomegalovirus (CMV) and vesicular stomatitis virus (VSV) have developed strategies to downregulate cell surface NKG2DL expression, partly by intracellular protein retention and/or sequestration [49-52]. Interestingly, intracellular retention of NKG2DL can be also mediated by the tumour-expressed carcinoembryonic antigen-related cell adhesion molecule 1 (CEACAM1), resulting in decreased NKG2DL surface expression and reduced anti-tumour immunity . Moreover, MICB has a short half-life at the plasma membrane, due to cholesterol-dependent and clathrin-mediated endocytosis . Taken together, the cell surface appearance of the various human NKG2DL is thus tightly regulated by context-dependent mechanisms.
Heterogeneity of NKG2D ligand expression on human tumour cells
Most if not all human tumours constitutively express one or several NKG2DL on their cell surface, and epithelial tumour cells and leukaemia/lymphoma cells display overlapping expression patterns. MICA and/or MICB are generally expressed on epithelial tumours of various origin including breast, lung, colon, kidney, pancreas, ovary, as well as on gliomas and melanomas [47, 55-59]. Although MICA and MICB are frequently coexpressed, some tumour cell lines carry only one of them on the cell surface [55, 57]. In addition, epithelial tumours also express one or several of the ULBPs, quite often ULBP2, ULBP3 or ULBP5 [55, 57-59]. The expression of NKG2DL on leukaemic and lymphoma cell lines appears to be more variable [55, 57], but one or several of the analysed NKG2DL (MICA/B, ULBP1-3) was present on primary leukaemic cells from most patients with acute or chronic leukaemia of myeloid or lymphatic origin .
Soluble NKG2D ligands are released from tumour cells
NKG2DL on tumour cells trigger cytotoxic activity in NK cells and γδ T cells and costimulate T cell responses [17, 18, 61]. However, many tumour cells release soluble NKG2DL through alternative splicing, PI-PLC-mediated cleavage, proteolytic shedding, or via exosome secretion (Fig. 2). As a matter of fact, increased levels of soluble NKG2DL are found in the sera of cancer patients, and correlations with clinical staging and/or prognosis have been reported. Specifically, alternative splicing, proteolytic cleavage, PI-PLC-mediated release, and/or exosome secretion has been reported for MICA [62, 63], MICB [64, 65], ULBP1 , ULBP2 [67, 68], ULBP3 , ULBP4 (also termed retinoic acid early transcript 1E (RAET1E)  and ULBP5/RAET1G . The GPI-anchored NKG2DL ULBP1,2,3 (and presumably ULBP6) can be liberated from the cell surface by PI-PLC , but ULBP1,2,3 can be also released by proteolytic cleavage or exosome secretion [68, 71]. Based on inhibitor studies  and siRNA-mediated downregulation or overexpression , matrix metalloproteases (MMP) 9 and 14 have been shown to mediate MICA shedding in osteosarcomas  and some human tumour cell lines . It appears, however, that ‘A disintegrin and metalloproteases’ (ADAM) are the most important family of proteases that execute the proteolytic cleavage of cell surface NKG2DL.
‘A disintegrin and metalloproteases’ (ADAM) proteases are major NKG2D ligand sheddases
ADAMs comprise a large family of proteases characterized by their multidomain structure and diverse functional activities. While not all ADAM family members are proteolytically active, a large variety of immunologically important target proteins are substrates for the cell membrane expressed ADAMs 9,10,15,17,28 . These include not only death receptor ligands such as TNF-α and Fas ligand but also several other receptor/ligand systems, for example, EGF receptor or TGF-β signalling molecules [75, 76]. With regard to tumour cell biology, other important classes of ADAM substrates are cell adhesion molecules (CAM) such as L-selectin, L1-CAM, CD44, and E-/N-cadherin, and molecules involved in the regulation of angiogenesis [77, 78]. Thus, the tumour may benefit at multiple levels from the proteolytic activity of ADAMs that counteract effective anti-tumour immunity. Based on their tumour-promoting properties, ADAMs have raised great interest as therapeutic targets for tumour therapy. In fact, several small molecule inhibitors of (mostly) ADAM10 and 17 have performed well in preclinical studies [79, 80], and some inhibitors have entered clinical studies in selected types of cancer [81, 82].
Multiple studies have unequivocally demonstrated a role of ADAM10 and 17 in proteolytic shedding of NKG2DL from tumour cells. Shedding of MICA/B and ULBP2 was thus prevented by metalloprotease inhibitors [47, 62, 64, 65, 67, 83], and specific silencing of ADAM10 or ADAM17 in MICA or MICB overexpressing transfectants indicated that MICA can be shed by both ADAM10 and 17  while MICB was reported to be selectively shed by ADAM17 . Additional studies based on siRNA silencing identified ADAM9 as an additional sheddase for MICA in two hepatocellular carcinoma cell lines . MICA associates with the disulphide isomerase endoplasmic reticulum protein 5 (ERp5) on the surface of tumour cells, and association with functional ERp5 is actually required for proteolytic cleavage of MICA . Moreover, palmitoylation of MICA facilitates the recruitment to membrane microdomains and effective shedding . Most of the above-cited studies were carried out with transfectants overexpressing MICA or MICB. To obtain more insight into the relative contribution of ADAM10 versus ADAM17 to the shedding of MICA versus MICB in non-overexpressing tumour cells, we recently downregulated ADAM10, ADAM17, or both ADAM10/17 by siRNA silencing in several tumour cell lines of different origin . Our results revealed a substantial heterogeneity with respect to the involvement of ADAM10/17 in shedding of MICA and MICB. In some cell lines (e.g. pancreatic carcinoma Panc89), only the combined downregulation of ADAM10 and ADAM17 inhibited MICA shedding, whereas ADAM17 but not ADAM10 was identified as the major sheddase in other tumour cell lines (pancreatic carcinoma PancTu-1, breast carcinoma MDA-MB-231). With respect to MICB shedding, ADAM10 and ADAM17 were equally effective in some tumour cell lines (PancTu-1), whereas ADAM17 was clearly more active in other cell lines (Panc89, MDA-MB-231, prostate cancer PC-3). We concluded that the relative role of ADAM10 versus ADAM17 in shedding of MICA/B is a selective feature of a given tumour cell rather than reflecting different substrate specificity of ADAM10 and ADAM17 for MICA and MICB . ADAM activity is regulated at multiple levels including glycosylation, phosphorylation and interaction with accessory proteins , which may all vary in different tumour cells. Moreover, given that there are more than 80 allelic variants of MICA and more than 30 allelic variants of MICB , it is conceivable that certain allelic variants are differentially susceptible to ADAM10/17-mediated shedding.
Exosome secretion of some NKG2D ligands
In fact, certain MICA alleles including MICA*008, the most frequent variant in Caucasians, carry an altered transmembrane domain and a truncated intracellular part, which is associated with localization in detergent-resistant membrane microdomains and release in exosomal microvesicles rather than proteolytic cleavage . Among 4 analysed tumour cell lines, we detected exosome-associated MICA only in the MICA*008-expressing PC-3 cells . In addition to MICA, MICB and ULBP1/2/3 have also been reported to be secreted on exosomes purified from tumour cells , transfectants , leukaemia/lymphoma cells  and mature human DCs . Tumour-derived exosomes are loaded with a variety of cargo molecules including MHC class I/II, death receptor ligands, adhesion molecules, tumour-associated antigens and others . Therefore, the modulation of the immune response by tumour exosomes that are loaded or not with NKG2DL will be determined by the overall molecular decoration of such exosomes. The different pathways of how soluble or exosome-bound NKG2DL can be generated are schematically summarized in Fig. 2.
Functional implications of soluble NKG2D ligands
Shedding is expected to decrease the cell surface level of NKG2DL, and as a consequence, inhibition of shedding by metalloprotease inhibitors or ADAM10/17 silencing should stabilize or increase cell surface expression. In fact, up-regulated MICA, MICB and ULBP2 surface expression associated with enhanced sensitivity to NKG2D-dependent lysis was observed when tumour cells were treated with metalloprotease inhibitors [47, 64, 83, 93] or upon siRNA-mediated downregulation of ADAM9/10/17 [83, 87]. Similarly, prevention of PI-PLC-mediated release of GPI-anchored ULPBs stabilized and enhanced cell surface expression of these NKG2DL .
A number of studies have demonstrated a down-modulation of NKG2D on NK cells or T cells by soluble NKG2DL, resulting in decreased cytotoxicity. Soluble MICA (sMICA) downregulated NKG2D by facilitating its endocytosis and subsequent degradation, leading to a reduced NKG2D expression on tumour-infiltrating T cells . Similarly, sULBP1–3 were found to downregulate NKG2D on NK cells [60, 66, 68]. In some instances, sULBP1 was also reported to inhibit NK cell activity . Production of soluble MICA in hepatocellular carcinoma cells was increased by IL-1β, which was shown to stimulate ADAM9 . Importantly, however, a functional impact of soluble NKG2DL is not always observed. Thus, soluble MICB did not modulate NKG2D expression on NK cells in vitro .
Tumour exosomes contain NKG2D ligands and can modulate NKG2D expression and NK cell activity which, however, might be also affected by other molecules contained within the exosome . Exosomal ULBP3 downregulated NKG2D on primary NK cells and inhibited NK cell-mediated killing of MICA-expressing target cells . Similarly, exosomes containing MICA*008 downregulated NKG2D and inhibited NK cell cytotoxicity , and Jurkat leukaemia and Daudi lymphoma lines could be induced to secrete NKG2DL-containing exosomes upon stress . Interestingly, NKG2DL-containing exosomes can be also secreted from non-malignant cells. As an example, it was found that human placenta secretes MICA and ULBP1-5 in exosomes that downregulate NKG2D expression on NK cells, which was suggested to play a role in feto-maternal tolerance . Importantly, the effects of soluble and exosome-bound NKG2DL may differ substantially [26, 27]. In fact, rather than inhibiting NK activity, NKG2DL-containing exosomes derived from human DCs were reported to directly activate human NK cells ex vivo . A summary of the reported pathways to generate soluble forms for each NKG2DL and their functional impact is presented in Table 1.
Table 1. Overview of reported pathways to generate soluble NKG2D ligands und functional implications
Model system in vitro
NKG2D expression functional implication
C1R-MICA: human lymphoma cell line transfected with MICA*001.
HeLa-MICA: cervical cancer cell line transfected with MICA*001.
293T-MICA: human embryonic kidney cells transfected with MICA*001.
TC2-MICA: mouse prostate cancer cells transfected with MICA*001.
MyC-MICA: mouse prostate cancer cells transfected with MICA*001.
CHO-MICA: Chinese hamster ovary cells transfected with MICA*008.
Detected total MICA/B.
U373-MICB: human glioma cells transfected with MICB.
CV1-MICB: monkey kidney cells transfected with MICB.
C1R-MICB: human lymphoma cells transfected with MICB*002.
CV1-ULBP2: monkey kidney cells transfected with ULBP2.
CHO-ULBP2: Chinese hamster ovary cells transfected with ULBP2.
C1R-ULBP2: human lymphoma cell line transfected with ULBP2.
CV1-ULBP3: monkey kidney cells transfected with ULBP3.
CHO-ULBP3: Chinese hamster ovary cells transfected with ULBP3.
CHO-RAET1E2: Chinese hamster ovary cells transfected with RAET1E2 (truncated splice variant of ULBP4/RAET1E).
Cos7-RAET1G2: monkey kidney cells transfected with RAET1G2 (truncated splice variant of ULBP5/RAET1G).
In line with the suspected role of NKG2DL shedding as a tumour escape mechanism, elevated levels of soluble NKG2DL are detected in the serum of cancer patients, and correlations of soluble NKG2DL with tumour stage and/or progression have been observed . Elevated serum concentrations of soluble MICA, MICB and/or ULBP2 have been detected in patients with pancreatic adenenocarcinoma [97, 98], various gastrointestinal carcinomas , hepatocellular carcinoma , non-small-cell lung carcinoma , malignant melanoma  and various types of leukaemia [60, 102]. Serum soluble ULBP2 appears to be a useful predictive marker in some malignancies because sULBP2 levels were shown to discriminate patients at early stage of pancreatic adenocarcinoma from healthy donors  and to identify melanoma patients at risk for disease progression . Furthermore, increased sULBP2 serum levels were associated with a poorer prognosis in patients with early stage B cell chronic lymphocytic leukaemia . Interestingly, some clinically used chemotherapeutic drugs (e.g. epirubicin) inhibit MICA shedding via downregulation of ADAM10, which leads to reduced sMICA levels in vivo . In some instances, soluble NKG2DL present in the serum of cancer patients were shown to be functionally active and could down-modulate NKG2D expression and inhibit NKG2D-dependent cytotoxicity. Thus, sMICA-containing serum from pancreatic carcinoma patients inhibited NK cell and γδ T cell-mediated cytotoxicity, and this was abolished by neutralization of sMICA . Similarly, soluble NKG2DL-containing serum from patients with ovarian and prostate cancer  or various types of leukaemia  could down-modulate NKG2D expression and inhibit NKT cell killing. A direct role of the respective NKG2D ligands in these studies was suggested on the basis of neutralizing effects of anti-MIC antibodies  or NKG2D-Fc fusion proteins . While these studies suggest a direct correlation between elevated soluble NKG2D ligand levels in tumour patients and reduced NKG2D-dependent immune responses, such conclusions need to be interpreted with caution. Serum from tumour patients contains a lot of additional immunosuppressive molecules (e.g. TGF-β) and exosomes that bundle a variety of (possibly tumour-derived) molecules, which all together may impact on NK and T cell immune responses. Not in all studies could a direct effect of soluble NKG2DL on NKG2D expression and/or NK cytotoxicity be demonstrated [64, 67]. In our own studies, we could not detect reproducible effects of tumour cell supernatants that did or did not (following siRNA-mediated downregulation of ADAM10/17) contain soluble MICA/MICB on NKG2D expression or cytotoxicity of freshly isolated NK cells or leukaemic NKL cells .
The NKG2D/NKG2DL system is an important part of effective immunosurveillance. While cell surface-expressed NKG2D ligands mark tumour cells for immunological attack by NK/NKT cells and subpopulations of T cells, tumours have adopted the strategy of shedding and exosome secretion of such ligands to escape from immunosurveillance. ADAM proteases have been identified as major sheddases, and selective inhibitors might prove useful in further clinical studies to avoid accumulation of soluble NKG2D ligands.
OJ and DK are supported by the DFG-funded Collaborative Research Center 877, projects A7 and B4. This work is part of the PhD theses of GC and JB.