Peripheral blood mononuclear cells (PBMCs) for mouse humanization and iPSC generation were isolated from healthy donors after informed consent using Ficoll-Paque separation. Four iPSC lines (AMF70.10, NL83.01, RP84.03, and AG89.04) were generated by reprogramming the donor-derived PBMCs with the CytoTune-iPS 2.0 Sendai Reprogramming Kit (Thermo Fisher Scientific). All lines were routinely screened for Mycoplasma using the MycoAlert Detection Kit (Lonza). Luciferase gene transduction was performed as previously described [21], and pancreatic differentiation followed established protocols [41].

NK cells were isolated from freshly obtained donor PBMCs using the CD56⁺CD16⁺ NK Cell Isolation Kit (Miltenyi Biotec) and expanded for 12 days in NK MACS Medium (Miltenyi Biotec) supplemented with 5% human AB serum (Corning), 70 ng/mL IL-15, and 500 U/mL IL-2 (Peprotech).

We quantified donor NK cell-recipient iPSC compatibility using an additive score that integrates the inhibitory (L) and activating (S) KIR copy numbers with the recipient HLA-C group and Bw4 epitopes. Copy numbers were obtained from KIR genotyping (per gene), while the HLA-C group and Bw4-I80/T80 were derived from iPSC HLA typing. A positive total score (S > 0.0) indicates a net inhibitory match, a negative score (S < 0.0) indicates a net activating/missing-self-like mismatch, and a score of zero (S = 0.0) is neutral. The Bw4 weight was defined as: W BW 4 = 1.5 n I 80 + 1.0 n T 80

For C1/C1 recipients: S C 1 / C 1 = + 1 K 2 DL 2 n C 1 + 0.5 2 DL 3 n C 1 − 1 K 2 DL 1 n C 1 − 1 K 2 DS 2 n C 1 + 1 K 2 DS 1 + n C 1 + 0.5 K 2 DS 5 n C 1 + 1 K 3 DL 1 w Bw 4 − 1 K 3 DS 1 w Bw 4 − 0.1 K 2 DS 4 f n C * 04 : 01 − 0.05 K 2 DS 4 f n A 11 + 0.1 K 3 DL 2 n A 11

For C2/C2 recipients: S C 2 / C 2 = + 1 K 2 DL 1 n C 2 − 0.6 K 2 DL 2 n C 2 − 0.1 K 2 DL 3 n C 2 − 1 K 2 DS 1 n C 2 + 1 K 2 DS 2 n C 2 − 0.5 K 2 DS 5 n C 2 + 1 K 3 DL 1 w Bw 4 − 1 K 3 DS 1 w Bw 4 − 0.1 K 2 DS 4 f n C * 04 : 01 − 0.05 K 2 DS 4 f n A 11 + 0.1 K 3 DL 2 n A 11

Where “K” represents the copy number of the given KIR gene, and “n” represents the iPSC allele counts for the corresponding HLA determinant. Interaction weights were based on published KIR-HLA interactions and included a dedicated term for KIR2DS4-HLA-C*04:01 binding, along with additional terms for KIR2DS4 and KIR3DL2 in the presence of HLA-A*11:01.

Endocrine progenitor (EP) clusters were cultured in suspension at 95 rpm in their specific medium, which was supplemented with 10 ng/mL IFN-γ and 50 ng/mL TNF-α (Peprotech). NK cells were pre-activated with 1,000 U/mL IL-2 and 20 ng/mL IL-12 (Peprotech). After overnight incubation, the iPSCs or the EP clusters were stained with 250 nM Incucyte Cytotox Green Dye (Sartorius) in complete NK MACS medium. The NK cell effectors were labeled with 5 µM Cell Proliferation Dye eFluor670 (Thermo Fisher Scientific). The effector and target cells were co-cultured at a 1:1 ratio in 96-well plates, with ≥3 target-only wells included to quantify basal cell death. The plates were then placed in the IncuCyte S3 Live-Cell Analysis System and analyzed using the associated software (Sartorius).

Female NOD-scid IL2Rgammanull (NSG) mice (age: 6–8 weeks old; weight: 20–24 g) were obtained from Charles River Laboratories, Italy. Female hIL-15 NOG mice (NOD.Cg-Prkdc ^scid Il2rg ^tm1Sug Tg (CMV-IL2/IL15)1-1Jic/JicTac) (age: 6–8 weeks old; weight: 20–24 g) were purchased from Taconic Biosciences. Female mice were chosen because males more often develop spontaneous dermatitis, which could confound symptoms of graft-versus-host disease (GvHD). Additionally, female mice exhibit more consistent human immune cell engraftment. The mice were humanized via an intravenous injection of 2.5 × 10⁶ human PBMCs, with PBMC labeling using the IVISense DiR 750 Fluorescent Cell Labeling Dye (XenoLight, Revvity, Inc.) performed as needed, following the manufacturer’s instructions. After 14 days, the mice were infused with ∼800 clusters (100–120 μm in diameter) into the intermuscular space of their lower hindlimbs and were monitored for up to 4 weeks post-transplantation. All procedures were conducted under protocols approved and overseen by the Animal Care and Use Committee of the San Raffaele Scientific Institute.

D-luciferin potassium salt (PerkinElmer) was administered intraperitoneally to anesthetized mice at 150 mg/kg. The animals were imaged using the Lumina II IVIS system (PerkinElmer), acquiring both bioluminescence and fluorescence signals. Bioluminescence was quantified as maximum radiance expressed in photons/s/cm²/sr, whereas fluorescence was measured as average radiant efficiency ([photons/s/cm²/sr]/[µW/cm²]). Signal intensities within the defined Regions of Interest (ROIs) were quantified using Aura software (Spectral Instruments Imaging).

Clusters were dissociated with trypsin (Lonza), and live cells were identified using the LIVE/DEAD Fixable Violet kit (Thermo Fisher Scientific). Surface staining was performed by incubating cells with antibodies for 30 min at 4 °C in FACS buffer (DPBS with 2% FBS and 2 mM EDTA). For intracellular staining, cells were fixed with Cytofix/Cytoperm (BD Biosciences) and permeabilized with Phosflow Perm Buffer III (BD Biosciences), then incubated for 45 min at 4 °C with intracellular antibodies. For the phenotyping of humanized mice, 50 µL of blood collected from the retro-orbital plexus was stained for 30 min at 4 °C, followed by red blood cell lysis with BD FACS™ Lysing Solution prior to analysis. The following conjugated antibodies were used: anti-B2M-APC (clone 2M2); anti- HLA-A, B, C-PE (clone W6/32), anti-CD45-PE/Dazzle™594 (clone HI30); anti-CD8-BV605 (clone SK1); anti-CD154-APC (clone 24–31) (all were obtained from Biolegend); anti-OCT3/4-AF647 (clone 40/Oct3); anti-CD184-PE (clone 12G5); anti-PDX-1-AF488 (clone 658A5); anti-NKX6.1-PE (clone R11-560); anti-insulin-AF647 (clone T56-706); anti-glucagon-BV421; anti-CD3-FITC/BUV395 (clone SK7); anti-CD4-PB (clone RP-T4); anti-CD56-PE (clone NCAM1); anti-HLA-DR-BV480 (clone G46-6); a-PD-1-R718 (clone EH12.1); anti-CD38-APC (clone HIT2); anti-CD223-BUV395 (clone T47-530), anti-FoxP3-R718 (clone 259D/C7); anti-Helios-PE (clone 22F6); anti-CD107a-APC-H7 (clone H4A3) (all were obtained from BD Biosciences); anti-CD4-PE-Vio770 (clone REA623); anti-CD14-VioGreen (clone TÜK4) (both were obtained from Miltenyi Biotec); anti-CD16-SuperBright436 (clone 3G8) (obtained from Invitrogen); and anti-CD69-StarBright UltraViolet 510 (clone FN50) (obtained from BioRad). The cells were acquired on the CytoFLEX LX flow cytometer (Beckman Coulter) using CytExpert, and the data were analyzed using FlowJo v10.

An intraperitoneal glucose tolerance test (ipGTT) was performed on day 28. After a 4-h fast, the mice received a 2 g/kg intraperitoneal glucose bolus, and blood glucose was monitored at 0, 30, 60, 90, and 120 min. At the 90-min time point, blood was collected and plasma was isolated by centrifugation and subsequently analyzed for human C-peptide using a Mercodia ELISA. Absorbance was measured on a BioRad microplate reader.

Explanted grafts were fixed, processed, and paraffin-embedded. Sections were cut at 5 µm for histological analysis. Hematoxylin and eosin (H&E) staining was performed to visualize the grafts and assess their morphology. BOND™ Ready-To-Use Primary Antibody Insulin (Leica, clone 2D11-H5) was used to detect mature iPSC-derived islets. The slides were scanned using a Leica Aperio 200.

All data are presented as the mean ± SEM unless otherwise specified. Comparisons between more than two independent groups were performed using one-way ANOVA with a Tukey’s post hoc test or a Kruskal-Wallis with Dunn’s test for non-parametric data. Longitudinal datasets were analyzed using two-way ANOVA followed by Holm-Šídák’s correction. Survival analysis was conducted using the Kaplan-Meier method with the log-rank (Mantel-Cox) test. Pairwise comparisons were made using a two-tailed unpaired or paired Student’s t-test, as appropriate. All analyses were performed using GraphPad Prism v10.

First, we examined the contribution of the NKp30-CD276 and CD226-CD155 axes to missing-self recognition under KIR-HLA class I mismatch conditions using in vitro NK cytotoxicity assays. The NK cells derived from eight KIR-genotyped donors were matched or mismatched with four HLA-typed iPSC lines. Undifferentiated iPSC lines were then co-cultured with the NK cells, and NK-mediated killing events were recorded via live cell microscopy in the presence or absence of blocking mAbs (Figure 1A). The list of characteristics of both donor NK cells and iPSC lines used in these experiments, along with the matched/mismatched pairs based on KIR gene copy number and HLA haplotype, is reported in Supplementary Tables S1-S3.

As expected, KIR-HLA mismatching resulted in a significant three-to-fourfold increase in NK-mediated killing compared to matched pairs (11.1% ± 2.7% vs. 36.9% ± 6.4%; p < 0.0001). Blocking either NKp30 or CD226 with specific mAbs reduced the killing of mismatched cells (24.2% ± 4.2% with α-NKp and 18.1% ± 2.6% with α-CD226, respectively) (Figure 1B). Accordingly, blocking the activating ligands CD276 or CD155 on target cells reduced NK cell cytotoxicity by a similar amount as blocking their counterreceptors (26.7% ± 8.9% with α-CD276 and 16.5% ± 7.0% with α-CD155, respectively). Remarkably, the combination of CD276 and CD155 blockade had a synergistic effect, reducing NK-mediated killing of the mismatched iPSCs to levels comparable to those of KIR-HLA-matched pairs (11.1% ± 2.7% for matched vs. 40.9% ± 8.9% for mismatched vs. 9.6% ± 3.0% with α-CD276+α-CD155, respectively) (Figure 1B).

Previously, we demonstrated that both β2-microglobulin (B2M) and HLA class I molecules are dynamically regulated during pancreatic differentiation [21]. Consistent with prior findings, iPSC lines exhibited the highest B2M and HLA-A/B/C expression levels, while a marked downregulation was observed at the posterior foregut (PF) and EP stages. SC-islets displayed an intermediate and more heterogeneous expression profile, with a broader range of HLA class I surface levels among cells (Supplementary Figures S1A,B).

Since EP clusters show low HLA expression, potentially limiting T cell immunogenicity, we hypothesized that they might be susceptible to missing-self-recognition by NK cells. Supplementary Figures S1C–E illustrate changes in key differentiation markers at each stage. OCT4 was highly expressed in iPSCs and was rapidly lost after definitive endoderm (DE) induction, while CXCR4 and FOXA2 increased. Progression to the EP stage was associated with increased expression of the pancreatic lineage markers PDX1 and NKX6.1, along with the early endocrine markers INS and GCG (Supplementary Figure S1C). Coexpression of PDX1 and NKX6.1 in EP cells was confirmed by flow cytometry (Supplementary Figure S1D), and EP cluster morphology was documented by bright-field imaging (Supplementary Figure S1E). To evaluate the susceptibility of EP clusters to NK cell infiltration and killing in vitro, we performed live-cell imaging cytotoxicity assays under both KIR-HLA matching and mismatching conditions. In mismatched pairs, missing-self-recognition occurred, as evidenced by increased NK cell infiltration and EP cell killing. This was measured by a progressive rise in green fluorescence over time (Figures 1C,D) and increased co-localization of red-labeled NK cells with green fluorescence signals (Figure 1E). Importantly, treatment of HLA-mismatched EPs with α-CD276 and α-CD155 monoclonal antibodies significantly reduced cell death (Figures 1F,G), in addition to NK cell activation markers, including HLA-DR (68.4% ± 9.7% vs. 22.7% ± 4.2%; p < 0.0001) and the degranulation marker CD107a (LAMP-1) on activated HLA-DR⁺ cells (29.7% ± 5.1% vs. 7.5% ± 1.9%; p < 0.0001) (Figures 1H,I).

Since CD276 and CD155 are expressed on both graft cells and immune cells, we next evaluated whether antagonizing them may provide graft tolerance by modulating immune cell activity other than direct graft targeting. NSG mice demonstrated sustained engraftment of T lymphocytes but limited NK cell and monocyte reconstitution (Supplementary Figure S2); therefore, we used hIL-15 NOG immunodeficient mice, which achieve superior humanization efficiency and improved engraftment of innate immune cells (Supplementary Figure S3). To evaluate the efficacy of blocking mAbs, mice were first injected intravenously with 2.5 × 10⁶ human PBMCs and, 14 days later, transplanted with ∼800 clusters of luciferase-expressing iPSC-derived EPs into the intermuscular space of the lower hindlimbs. The mAb treatment regimen consisted of 5 intraperitoneal injections every 3 days, starting 2 days before transplantation and continuing up to 10 days post-transplantation, as outlined in Figure 2A. We tested two doses of α-CD276 (1.25 and 15 mg/kg), one dose of α-CD155 (2.5 mg/kg), and a combined treatment consisting of the highest dose of α-CD276 (15 mg/kg) plus α-CD155 (2.5 mg/kg). Humanized untreated mice rejected EP grafts within 7–10 days (Figures 2B,C). Notably, increasing doses of α-CD276 prolonged graft survival for 2 weeks (Figure 2C), but were associated with increased mortality due to an earlier onset of GvHD, as confirmed by log-rank trend analysis (p = 0.025) (Figure 2D). In contrast, treatment with α-CD155, either alone or in combination with α-CD276, ensured graft survival for up to 4 weeks at levels comparable to non-humanized mice (Figure 2C). Moreover, co-administration of α-CD155 mitigated the severe GvHD effects observed with α-CD276 alone, as mortality in the α-CD155-treated groups was comparable to that of untreated humanized mice (Figure 2D).

The kinetics of immune cell infiltration and the impact of checkpoint blockade were longitudinally investigated by in vivo imaging combining bioluminescence (to track luciferase-expressing EP grafts) and fluorescence detection of DiR-labeled human PBMCs (Figure 3A). Humanized mice were treated intraperitoneally with α-CD276 (15 mg/kg), α-CD155 (2.5 mg/kg), or both antibodies in combination at the same doses.

As early as day 4, untreated humanized mice showed progressive PBMC accumulation at the graft site, with the fluorescence signal increasing from 13.2 ± 5.0- to 66,537 ± 59,820-fold over background by day 14. Mice treated with α-CD276 alone showed delayed yet substantial infiltration (5,727 ± 3,284 fold), while α-CD155 alone more effectively reduced immune cell recruitment (208 ± 158 fold). Strikingly, the combined blockade of CD276 and CD155 limited the increase in PBMC-associated fluorescence on day 14 to ∼69-fold over the day 0 baseline. Although this signal remains higher than that observed in untransplanted (∼19-fold) and non-humanized (∼1.8-fold) mice at the same time point, the absence of further fluorescence amplification indicates that only a few immune cells reach the grafts, without evidence of sustained recruitment (Figure 3B).

No significant differences were observed between groups at early time points (day 4–10). On day 14, however, Kruskal-Wallis analysis revealed significant divergence in PBMC accumulation across experimental groups (H = 11.48, p = 0.043). Dunn’s post hoc comparisons revealed that PBMC infiltration in untreated mice (66,537 ± 59,820-fold) was significantly higher than in non-humanized (1.81 ± 0.54-fold; p < 0.001), untransplanted (19.2 ± 14.0-fold; p < 0.01), and dual mAb-treated mice (69.6 ± 49.1-fold; p < 0.05). Differences with α-CD276 (5,727 ± 3,284-fold) and α-CD155 (208 ± 158-fold) were not statistically significant after correction, although a downward trend was noted in the α-CD155 group (Figure 3B).

Next, we sought to investigate the mechanisms underlying the protective effects of α-CD276 and α-CD155 mAb treatments and their impact on immune cell activation and migration. According to in vitro experiments, we found that both α-CD276 and α-CD155 prevented the overexpression of the tissue migration marker CD69 on circulating CD56^dimCD16⁺ NK cells (Figure 4A). Moreover, α-CD276, either alone or in combination with α-CD155, dampened CD14⁺ monocyte activation and migration, as confirmed by lower levels of HLA-DR⁺ and CD40L⁺ cells compared to the untreated group (Figure 4B), supporting the hypothesis that CD276 signaling contributes to monocyte infiltration into the grafts.

However, high-dose α-CD276 alone did not prevent CD8⁺ T cell activation (Figure 3C). In contrast, α-CD155 treatment alone significantly increased the frequency of exhausted PD-1⁺CD8⁺ T cells (Figure 4C) and LAG-3⁺CD4⁺ T cells (Figure 4D) 12 days post-transplantation. The combination of α-CD276 and α-CD155 led to a slight but significant reduction in CD38⁺HLA-DR⁺CD8⁺ T cells at 12 days post-transplantation (Figure 4C) and an early increase in LAG-3 expression on total CD4⁺ T cells compared to the untreated group (Figure 4D). In line with recent studies showing that CD226 signaling negatively affects Treg stability [42, 43], blockade of its ligand CD155 resulted in a twofold increase in the percentage of total Tregs compared to the untreated mice (Figure 4E). Furthermore, evaluation of IL-2 Receptor alpha (IL-2Rα/CD25) on FoxP3⁺Helios⁺CD4⁺ Tregs revealed a significant upregulation of CD25 surface expression in the α-CD155-treated group compared to mock controls (735 ± 66.4 vs. 488 ± 53.6; p < 0.0001) (Figures 4F,G), supporting the hypothesis of an enhanced Treg suppressive ability in response to CD155 blockade, consistent with previous reports [44, 45].

To assess the maturation and function of the iPSC-derived EP grafts, we monitored plasma human c-peptide levels for 4 weeks post-transplantation. A progressive increase in basal c-peptide levels was observed in both non-humanized and mAb-treated mice, indicating functional maturation. In stark contrast, humanized mice that did not receive CD276/CD155 blockade showed only a modest increase in c-peptide levels at 2 weeks (7.27 ± 8.11 pmol), followed by a complete loss at 4 weeks, consistent with the early rejection of endocrine cells and the failure of the grafts to mature (Figure 5A).

On day 28, an ipGTT was performed to evaluate glucose responsiveness. It should be noted that the glycemic curves shown are not intended to reflect graft functionality, as glycemia is predominantly controlled by endogenous murine islets in this setting. Both non-humanized and mAb-treated mice exhibited a significant rise in human c-peptide 90 min after glucose administration (196.5 ± 18.6 vs. 40.0 ± 5.88 pmol and 191.0 ± 35.2 vs. 30.2 ± 13.4 pmol, respectively; p < 0.0001), while the untreated humanized group showed a negligible response at 90 min post-ipGTT (1.02 ± 0.79 pmol), indicating an absence of functional endocrine cells (Figure 5B).

Histological analysis showed preserved graft architecture in non-humanized and mAb-treated mice, whereas untreated mice displayed tissue disorganization. Insulin-producing cells were found in both non-humanized and mAb-treated groups, but not in the untreated group, consistent with the c-peptide data. Moreover, human CD45 staining showed marked leukocyte infiltration in the grafts of the untreated mice, but not in the mAb-treated groups (Figure 5C). These findings indicate that dual blockade of CD276 and CD155 protects the grafts from early immune rejection, enabling in vivo maturation of EP cells into functional, glucose-responsive β-like cells.