Recognition and Killing of Human and Murine Pancreatic β Cells by the NK Receptor NKp46

Type 1 diabetes (T1D) is currently an incurable disease that is mainly treated by daily injections of insulin. In rare situations, when metabolic instability and severe hypoglycemia persist, human islet transplantation is considered as possible treatment (1) (2). However, such treatment is still not very effective, and in almost all cases, >90% of the islets are rejected within a short period of time (2, 3). The reasons for this rejection are largely unknown.

Early reports suggest that in addition to T cells, NK cells are involved in the pathogenesis of T1D (4–6). The NK cell cytotoxicity is complex and regulated by both inhibitory receptors, which recognize mainly MHC class I proteins, and activating receptors (7–10). The most prominent NK inhibitory receptors are those that recognize the HLA-C proteins, as all of the HLA-C proteins can be divided into two groups, C1 and C2, that are recognized by killer Ig-like receptor (KIR) 2DL2 and KIR2DL1, respectively (11). Another prominent inhibitory receptor is leukocyte Ig-like receptor-1 (LIR1), which recognizes a broad range of MHC class I proteins (11, 12). The NK activating receptors recognize pathogen-derived, stress-induced, tumor-derived, and even self-ligands (13–16). Prominent among the killer receptors is NKp46 (NCR1 in mice), as it is expressed exclusively by NK and NK-like cells (8, 17, 18). The only ligands identified so far for NKp46 are viral hemagglutinins (HA) (19, 20), and the recognition of viral HA by NKp46 is mediated mainly via α2,6-linked sialic acid residues carried by NKp46 and largely relies on the highly conserved sugar-carrying residue of NKp46, Thr²²⁵ (19).

Recently, we showed that NKp46 recognizes an unknown ligand expressed by pancreatic β cells (21). We showed in murine models that NK cells kill pancreatic β cells in an NKp46-dependent manner and demonstrated that T1D development could be prevented by blocking NKp46 activity using active immunizations (21).

In this study, we extensively and thoroughly analyze the interactions between NKp46 and its unknown ligand expressed by both human and murine β cells. We demonstrate that human β cells, which are intended to be used for transplantation, express the unknown ligand of NKp46 and are killed in an NKp46-dependent manner. We also describe the molecular mechanisms controlling the NKp46 interaction with its mouse and human β cell ligands and identify the binding site on NKp46 that is involved in these interactions.

All experiments were performed in a specific pathogen-free unit of the Hadassah Medical School (Ein-Kerem, Jerusalem) in accordance with the guidelines of the ethical committee. The double-transgenic Insulin-rtTA;TET-DTA mice and the conditional ablation of pancreatic β cells were generated as previously described (22). All fusion proteins used in this study—the NCR1-Ig, NKp46-Ig, NKp46D2-Ig, NKp46D1-Ig, KIR2DL1-Ig, KIR2DL2-Ig, LIR1-Ig, NKp46-T225V-Ig, NKp46-T125V-Ig, and NKp46-N216V-Ig—were generated in COS-7 cells and purified by affinity chromatography using a protein G column, as previously described (20). Human islets intended to be used for transplantations were derived from healthy donors. The anti-NKp46 and control anti-carcinoembryonic Ag (CEA) sera were generated as previously described (21).

Paraffin-embedded sections of pancreatic tissues were prepared from female NOD mice at various weeks of age (including the embryonic period) from 8–12-wk-old BALB/C mice and from double-transgenic Insulin-rtTA;TET-DTA mice. Archival human pancreatic tissue sections were obtained from autopsies (male and females) ranging in age from 17 wk of gestation to adulthood. All fetuses, infants, and adults died as a result of diseases not related to the pancreas, and all autopsies were performed due to medical reasons according to accepted procedures. The immunohistochemical and immunofluorescence stainings were performed as previously described (21).

Isolation of murine β cells was performed as previously described using a pool of β cells derived from 8–10-wk-old BALB/C mice (n > 8/experiment) (21). For FACS staining, human and murine β cells (75,000/well) were incubated on ice with the appropriate fusion protein (5 μg/well) and the biotinylated anti-mouse GLUT-2 mAb (BAM1440, clone 205115; R&D Systems), or with the guinea pig anti-insulin Ab for 2 h. Cells were then incubated with fluorochrome-conjugated secondary Abs for 30 min and analyzed by FACS using CellQuest software.

Human NK cells were isolated from peripheral blood using the Human NK Cell Isolation Kit and the autoMACS instrument (Miltenyi Biotec) according to the manufacturer’s instructions. Purified NK cells from peripheral blood were plated at 1 cell/well in 96 U-well plates in RPMI 1640 medium supplemented with 10% human serum, 50 U/ml recombinant human IL-2 (Boehringer Mannheim, Indianapolis, IN), and 1 ng/ml PHA for 2 wk. Irradiated feeder cells (2.5 × 10⁴ allogeneic PBMCs from two donors and 5 × 10³ RPMI 8866 cell line in each well) were added. Proliferating clones were verified to be positive for CD56 and negative for CD3 markers and were subsequently combined and grown for an additional week to generate bulk polyclonal NK cell lines. Purity of NK cell lines was routinely verified to be >99%. For the in vitro cytotoxicity assay, human islets were labeled with [³⁵S]methionine, for 24 h at 37°C, separated with Cell Dissociation Solution Non-enzymatic 1× (Sigma, C5914), and incubated for 5 h at 37°C with effector NK cells at various E:T ratios as previously described (20). The CD107a assays were performed as previously described (21).

For detection of NK cells in situ, pancreases were dissected out from eight 9- to 10-wk-old female nondiabetic NOD and BALB/C mice, immersed in fixative solution for 3 h, and rinsed for 24 h in PBS containing 30% sucrose. The tissue was frozen, cryostat sectioned into 14-μm thickness, and thaw-mounted onto glass slides. The pancreatic β cells were detected by polyclonal guinea pig anti-mouse insulin (Abcam). The tyramide signal amplification kit (PerkinElmer) was used to visualize NKp46 by polyclonal goat Abs against mouse NKp46 (catalog number AF2225; R&D Systems) as previously described (23).

To test whether primary human islets that are used for human transplantation express an unknown ligand for NKp46, we isolated β cells from such islets and stained them with the intact NKp46-Ig, with its membrane proximal domain, which also contains the stalk region and is named NKp46D2-Ig, and with its membrane distal domain, named NKp46D1-Ig. A schematic representation of NKp46 is presented in Supplemental Fig. 1. As can be seen in Fig. 1, isolated human β cells that express insulin (detected by staining with polyclonal anti-insulin Abs; Fig. 1A) also express a putative ligand for NKp46, and the NKp46 binding is confined to the membrane proximal domain D2 (Fig. 1B).

We also investigated whether the isolated human β cells are recognized by the KIR2DL1-Ig, KIR2DL2-Ig, and LIR1-Ig inhibitory receptor fusion proteins. Surprisingly, whereas LIR1-Ig binding was detected, little or no staining was observed with either the KIR2DL-1-Ig or the KIR2DL2-Ig (Fig. 1B). Because similar results were observed with β cells obtained from four separate donors and because all fusion proteins stained their corresponding MHC ligands on various 721.221 transfectants (12) (Supplemental Fig. 2), we concluded that the HLA-C is either not expressed on these human β cells or that its expression levels are too low to be detected by the appropriate NK inhibitory receptors. Currently, there is no specific mAb that recognizes HLA-C, and we are therefore unable to differentiate between these two options. We also stained the β cells derived from various donors for the expression of the NKG2D ligands MICA, MICB, ULBP1, ULBP2, and ULBP3 and observed no staining (data not shown).

Because we observed that human β cells express a ligand for NKp46 and because they cannot be recognized by the HLA-C–binding NK inhibitory receptors, we wondered whether they will be killed by NK cells in an NKp46-dependent manner. We therefore initially used the same primary human β cells, described above, in CD107a degranulation assays. As can be seen in Fig. 1, incubation of NK cells with human β cells leads to NK cell degranulation (Fig. 1C–E). Importantly, as can be seen in Fig. 1E, the NK cell degranulation was NKp46 dependent, as it was reduced upon preincubation of bulk NK cells with anti-NKp46 sera (produced by injecting mice with NKp46-Ig) (21) (Supplemental Fig. 3) as compared with preincubation of the NK cells with control sera (produced by injecting mice with CEA-Ig) (21) (Supplemental Fig. 3). Similar results were obtained with human β cells derived from two donors.

However, NK cell degranulation does not always correlate with actual killing of the target cells. Therefore, to make sure that human β cells are indeed killed in an NKp46-dependent manner, we took advantage of the fact that human β cells are metabolically active and labeled them with [³⁵S]methionine for 24 h. We then isolated human NK cells from the PBLs of various donors, grew them in culture, preincubated them with anti-NKp46 sera and control sera (as described above), and then incubated them with the labeled β cells to determine whether human β cells are killed in an NK-46–dependent manner. As can be seen in Fig. 1F, efficient killing of human β cells was noted, and such killing was NKp46-dependent, as it could be blocked by using the anti-NKp46 sera mentioned above (Fig. 1F). Thus, primary β cells, which are used for transplantation, express an unknown NKp46 ligand and are killed in an NKp46-dependent manner.

The human β cell distribution in the pancreas undergoes progressive changes during development and in the uterus; around the age of 6 mo, the β cells are organized into islets (24). Therefore, we next studied whether the expression of the NKp46 ligand is associated with the organization of the β cells into islet structures.

For this purpose, we stained human pancreases derived from fetuses, infants, and adults with anti-insulin p-Ab and NKp46D2-Ig (we used the D2 domain of NKp46 for this purpose because it gives a better staining as compared with the intact NKp46-Ig). As can be seen in Fig. 2, the expression of the NKp46 ligand is already detected during the embryonic period, almost concomitantly with the detection of insulin (Fig. 2A). From that point on, the expression of the NKp46 ligand is consistent, and its expression is increased with age concomitantly with the increase in β cell mass (Fig. 2A, 2B). As presented in Fig. 2A and 2B, the NKp46 staining is very similar to the insulin staining. However, a closer examination of the staining with higher magnification revealed that, indeed, the insulin staining merged with the NKp46D2-Ig staining (Fig. 2C, left panels, red staining alone could hardly be detected). However, it is also shown that the NKp46D2-Ig staining is not identical to the insulin staining, and areas inside the cells can be observed that are free from insulin and are stained with NKp46D2-Ig (observe the green D2 speckles, presumably corresponding to distinct cellular vesicles that do not contain insulin [red], in Fig. 2C). Nevertheless, to make sure that the costaining observed with NKp46D2-Ig and insulin is not an artifact that occurred due to Ab cross-reactivity, we repeated the staining with the same Abs and the control NKp46D1-Ig, as can be seen in Fig. 2C, right panels; in this case, only insulin and DAPI staining is detected (no green or yellow colors). In addition, we verified that the secondary Abs used in this assay recognize only their appropriate primary Abs (data not shown).

Working with primary human β cells is very difficult, not only because they are hard to manipulate in vitro, but also because of the limited supply. Thus, it is quite essential to determine the murine β cell’s ligand properties. In this regard, we have recently shown that murine β cells express the unknown ligand for NKp46 and that this ligand can be recognized by both the human and murine NKp46 (21). Therefore, we next investigated whether the D2 domain (including the stalk) of the human NKp46 is also involved in the recognition of mouse β cells by using two modalities: FACS and immunohistochemistry. For immunohistochemistry, pancreatic BALB/C paraffin sections were stained with the NKp46D1-Ig and NKp46D2-Ig fusion proteins, and for the FACS staining, β cells derived from BALB/C mice were stained with a combination of fusion proteins and anti-mouse GLUT-2 mAb. As can be seen in Fig. 3A and 3B, in both assays, β cell staining was obtained only with the usage of the NKp46D2-Ig, indicating that similar to human β cells, this portion of the NKp46 receptor contains the binding site for the murine β cell ligand. Similar results were obtained with β cells derived from C57BL/6 and nondiabetic NOD mice (n > 8/experiment).

In our recent publication (21) and in this study, we have demonstrated that the unknown NKp46 ligand is expressed by all β cells derived from healthy humans, healthy mice, and prediabetic NOD mice. We therefore wondered why diabetes does not develop in every individual and hypothesized that NK cells are not found in the vicinity of healthy pancreatic islets. To investigate this, we used the anti-NKp46 mAb in immunofluorescence staining of cryosections derived from female BALB/C or NOD mice pancreases at the prediabetic stage. Confirming our hypothesis, NK cells were detected in situ, in close proximity with β cells only in the pathological NOD mice pancreases and only rarely observed in the exocrine pancreatic tissue of the normal mice (examples are shown in Fig. 3C and 3D).

The results obtained so far suggest that human and mice β cells express a similar ligand for NKp46. We therefore tested whether the expression pattern of the mouse NKp46 ligand is similar to that of the human β cells, hoping that such an investigation would provide us with essential clues concerning the identity of the NKp46 ligand. For this purpose, we performed immunofluorescence staining for the expression of insulin and the unknown NKp46 ligand in pancreases obtained from NOD mice during the embryonic phase, immediately after birth, and up to the late stage of insulitis. As can be seen in Fig. 4A, a few insulin-producing cells are present in the embryonic pancreas as early as day 14.5 of embryonic development, but the organization into islets is only seen at around day 18.5. Interestingly, and in contrast to the human (Fig. 2), in the mouse, little or no expression of the NKp46 ligand was observed during the embryonic period (Fig. 4A). However, following birth, the expression of the NKp46 ligand merged with that of insulin (Fig. 4B, 4C). Similar to the human staining, a closer examination of the staining with higher magnification revealed that the NKp46D2-Ig staining of the mouse β cells is not identical to the insulin staining, and areas inside the cells can be observed that are either free from insulin or free from the NKp46-ligand (Fig. 4D). Again, we verified that the staining is specific and observed no staining with NKp46D1-Ig (Fig. 4D, right panel). In addition, we verified that the secondary Abs used in this assay recognize only their appropriate primary Abs (data not shown). Thus, because the human and mouse NKp46 ligands seem to be similar (see above) and because the NKp46 staining does not always correlate with the insulin staining, we concluded that insulin is not the NKp46 ligand. Indeed, in separate experiments, purified insulin was not recognized by NKp46 (not shown).

To gain further information concerning the identity of the NKp46 ligand, we investigated whether its expression is stable (i.e., whether under certain conditions the expression of the ligand is altered). Because we observed that ligand expression persists on β cells that survived the immune cell attack in NOD mice (21), we tested whether its expression would be maintained following cytotoxic destruction of the islets and subsequent β cell regeneration. For this purpose, we used a combined transgenic mouse in which, upon the administration of doxycycline, rtTA induces the expression of diphtheria toxin A, causing β cell apoptosis and diabetes (22). As can be seen in Fig. 5, following treatment with doxycycline for 1 wk, a severe loss of β cells was noted (Fig. 5, upper left panel, detected by anti-insulin staining). Interestingly, following such ablation, the remaining few β cells that still express insulin also express the NKp46 ligand, and the expression of the NKp46 ligand merged almost completely, but not entirely, with that of insulin (Fig. 5, upper center and right panels). Furthermore, when the mice were left to recuperate for several months, the expression of the NKp46 ligand was detected almost entirely on all insulin-producing cells (Fig. 5, bottom panels). These experiments led us to conclude that once the NKp46 ligand is expressed, its expression is stable. Thus, although these experiments did not provide us with further information concerning the identity of the ligand, they suggest that almost at any time point, β cells are at risk for being attacked by NK cells through NKp46-mediated activation.

The NKp46 protein displays two putative O-linked glycosylations, on Thr¹²⁵ and on Thr²²⁵, and one N-linked glycosylation on Asn²¹⁶ (Supplemental Fig. 1) (19, 25). Because we have previously demonstrated that Thr²²⁵ plays a critical role in the recognition of the influenza virus HA (19), we wondered whether the glycosylated residues of NKp46 are involved in its binding to β cells. We initially treated the NKp46-Ig fusion protein with neuraminidase, which cleaves sialic acid residues, and observed that it did not affect the binding to β cells (data not shown). As previously reported (19), it severely abrogated the binding of NKp46-Ig to influenza-infected cells (data not shown). Nevertheless, we continued our research and investigated whether the glycosylated residues of NKp46 are involved in its binding to the β cell ligand in a sialic acid-independent manner. To test this, we mutated each of the three glycosylated residues of the NKp46 and replaced them with Valin, cloned them in frame with Ig, and produced the corresponding fusion proteins NKp46T225V, NKp46T125V, and NKp46N216V. The various fusion proteins were then used to stain murine and human β cells. As can be seen in Fig. 6A, whereas the mutation in Thr²²⁵ did not significantly influence the NKp46 recognition of murine β cells, the two other mutations, at Thr¹²⁵ and at Asn²¹⁶, completely abolished the NKp46-Ig binding to murine β cells, using two modalities: immunohistochemistry (Fig. 6A, upper panels) and FACS staining (Fig. 6A, lower panels). The residual staining observed with the NKp46N216V-Ig using FACS is probably nonspecific, as no staining was detected with this fusion protein in the immunohistochemistry experiments. Importantly, a similar staining pattern was observed with human β cells, and mutating the 216 or the 125 residues completely abolished NKp46 binding (Fig. 6B). To verify that the Thr¹²⁵ and the Asn²¹⁶ mutations did not somehow disrupt the NKp46 structure, thus interfering with binding to its ligand, we stained influenza-infected cells with the same fusion proteins. Importantly, mutating the Thr¹²⁵ or the Asn²¹⁶ residues did not affect binding to the infected 721.221 cells, whereas as previously reported (19), the mutation in the Thr²²⁵ residue severely reduced the binding (Fig. 6C). Thus, whereas the sialic acid residues of Thr²²⁵ are critical for the binding of NKp46 to HA (this study and Ref. 19), the glycosylated residues Asn²¹⁶ and Thr¹²⁵ are critically involved in the binding of NKp46 to its human and mouse β cell ligand in a sialic acid-independent manner.

Unfortunately, despite decades of research, T1D, which affects millions of people around the world, is still an incurable disease. To date, the only treatment that releases T1D patients from their insulin dependency is allogeneic transplantation of islets obtained from deceased donors. This treatment, however, still suffers from many difficulties, and, in addition to scarcity of pancreatic islets, one of the major obstacles is the loss of a large portion of the transplanted islets within the first weeks of transplantation, even while using immunosuppressive agents (2, 3). The reasons accounting for the transplanted islets’ rejection are largely unknown.

In this study, we show that human β cells that are used for transplantation express an unknown ligand for NKp46, do not express the NKG2D ligands, and are not recognized by the NK inhibitory receptors KIR2DL1 and KIR2DL2, which together recognize the entire spectrum of all HLA-C proteins. This makes the human β cells vulnerable to NK cell attack, and, indeed, human NK cells degranulate, and β cells are killed in an NKp46-dependent manner. These results are in agreement with our recent publication, in which we showed that NKp46 is involved in the killing of murine β cells and that pathogenic murine NK cells extracted from the pancreases of NOD mice are degranulated (21). The important implication of this research is that blocking of NKp46 function during diabetes development and following human β cell transplantation might help prevent T1D development and islet graft rejection.

We have demonstrated, both in this study and previously (21), that β cells of healthy mice and humans express an unknown ligand for NKp46, and we show in this study in situ that under normal conditions in mice, NK cells are not found in the vicinity of the endocrine portion of the pancreas and accumulate in the pancreatic islets only upon disease induction. Despite years of investigations, the mechanisms controlling T cell accumulation in the pancreas are still largely unknown, and it is also unknown why NK cells accumulate in the diseased pancreas. Future research, aimed at uncovering the mechanisms controlling NK and T cell accumulation during diabetes, might lead to the development of new treatment modalities for T1D.

We also analyzed in this paper, in depth, the interactions between NKp46 and its β cell ligand and determined that the D2 domain, which also includes the stalk region, contains the binding site for the unknown ligand expressed by both human and mouse β cells. Interestingly, this part of NKp46 also binds to the viral HA and to unknown tumor ligands (19). Thus, the D1 domain seems to be dispensable for binding. Indeed, several isoforms of NKp46 were identified, some of them completely lacking the D1 domain (25).

We also show that the expression of the NKp46 ligand is not detected in the murine embryonic β cells, whereas in humans, the ligand is detected at the embryonic stage, concomitantly with the detection of insulin-producing cells. One possible explanation for the differences in ligand expression between human and murine β cells is that the NKp46 ligand is somehow associated with functional maturation of the β cells, which occurs relatively earlier in the human pregnancy (24). We show that once the NKp46 ligand appears on the β cells, its expression is stable, as it is expressed by the β cells that survive the attack by NK and T cells (21), and it is still expressed following ablation of the majority of β cells and during β cell regeneration. Thus, at any time point following the expression of the NKp46 ligand, β cells are in risk of being attacked by NK cells via the activation of NKp46.

It is still unclear why β cells of both humans and mice express a ligand that can potentially harm them. It is possible that the NKp46 ligands in the pancreas resemble foreign ligands recognized by NKp46 and that NK cells attack β cells unintentionally because of this molecular mimicry. Indeed, infectious diseases have been suggested as potential triggers for many autoimmune diseases, and a link between viruses and type 1 diabetes has been demonstrated (26, 27).

Finally, we show that two glycosylated residues of NKp46, Thr¹²⁵ and Asn²¹⁶, are critical for its β cell recognition. These observations, together with our previous ones (19), suggest that NKp46 interacts with at least three different ligands by using different means of recognition. Although Thr²²⁵ is critical for the NKp46 interaction with HA and some tumor cells (19), the two other residues, Thr¹²⁵ and Asn²¹⁶, but not Thr²²⁵, are important for the NKp46 recognition of its human and murine β cell ligand.

The identification of the NKp46 ligands is a very difficult task. One of the problems that has so far prevented identification the NKp46 ligands is, as shown in this study and previously (19, 21, 28, 29), ligand diversity. Still, the identification of the β cell ligand for NKp46 is of particular importance, as it can possibly lead to the development of novel therapy modalities for T1D.

The authors have no financial conflicts of interest.