2.1 Cell culture

LN-18, T98G, A172, and U87MG cells (ATCC) were cultured in DMEM (Lonza) supplemented with 10% heat-inactivated FBS (Sigma-Aldrich), 1 mM sodium pyruvate, 2 mM L-glutamine, 100 U/mL penicillin, and 10 μg/mL streptomycin at 37°C and 5% CO₂. Buffy-coats from healthy donors were collected from Centro Comunitario de Sangre y Tejidos de Asturias following the Declaration of Helsinki. Peripheral blood mononuclear cells (PBMCs) from healthy donors were isolated by ficoll (Biowest) density gradient centrifugation and cultured in supplemented RPMI 1640 (Lonza).

2.2 Evaluation of human leukocyte antigen expression

Surface levels of HLA-I molecules were determined on glioblastoma cell lines using the following antibodies: anti-HLA-(A,B,C)-PE (clone W6/32), anti-HLA-E-PE (clone 3D12), anti-HLA-F-PE (clone 3D11/HLA-F), and anti-HLA-G-PE (clone 87G) (all from Biolegend). PE mouse IgG₁ (clone MOPC-21) or PE mouse IgG_2a (clone MOPC-173) antibodies (both from Biolegend) were employed as isotype controls. 7-AAD (Immunostep) was used to discriminate viable cells. Cells were analyzed in a Cytoflex S flow cytometer with CytExpert 2.3 software (Beckman Coulter). For the analysis of surface expression, raw MFI values were corrected by subtracting isotype values and, where indicated, normalized to the control condition as a fold induction.

2.3 Immunohistochemistry

Tissue specimens from resected glioblastoma tumors were analyzed by immunohistochemistry employing the EnVision FLEX Mini Kit and Dako Autostainer system (Agilent) following the manufacturer’s instructions. In brief, paraffin-embedded tissues (3 μm) were deparaffinized and rehydrated and epitope retrieval was performed by heat induction (HIER) at 95°C for 20 min and pH = 9 in the PT Link, Pre-Treatment Module. Endogenous peroxidase activity was blocked with EnVision FLEX Peroxidase-Blocking Reagent for 5 min. Afterwards, sections were incubated with Protein Block, Serum-Free buffer for 60 min. Tissue sections were stained with the following polyclonal antibodies: anti-human HLA-A (1:400, PA5-116980), anti-human HLA-B (1:200, PA5-35345), anti-human HLA-C (1:2000, PA5-79367), anti-human ILT2 (1:1000, PA5-98738) (all from Invitrogen), or anti-human HLA-E (1:200, HPA031454, Merck) polyclonal antibodies and counterstained with hematoxylin. Signals were detected using diaminobenzidine chromogen as substrate in EnVision FLEX HRP (Agilent). Isotype controls were processed by omitting the primary antibody. Each specimen was individually reviewed and expression intensity was scored using a scale from 0 to 3. Sample classification was performed as follows: 0 as negative staining, 1 as dim, and 2–3 as positive.

2.4 Immune cell-glioblastoma cocultures

PBMCs from healthy donors were cocultured with glioblastoma cell lines at a 5:1 ratio for 24 h. Afterwards, cells were stained with anti-CD3-FITC, anti-CD56-APC (Cytognos), anti-CD8-APC-C750 (Immunostep), and anti-ILT2-PE (clone GHI/75, Biolegend) and analyzed by flow cytometry. PE mouse IgG_2b antibodies (clone 27–35, Biolegend) were employed as isotype control.

2.5 Intracellular staining

Production of interferon-gamma (IFN-γ) and granzyme B was assessed as described elsewhere.²⁴ In short, PBMCs from healthy donors were treated with 10 μg/mL anti-ILT2 blocking monoclonal antibody (mAb) (clone HP-F1, kindly provided by Miguel López-Botet, Universitat Pompeu Fabra) or control IgG (Biolegend) for 72 h. Afterwards, PBMCs were cocultured with glioblastoma cell lines in a 5:1 ratio for 4 h. For IFN-γ detection, cocultures were plated in the presence of 50 nM PMA and 1 μg/mL ionomycin. Following incubation, cells were stained with anti-CD3-FITC, anti-CD56-APC, and anti-CD8-APC-C750 and anti-IFN-γ-PE (clone 4S.B3; Biolegend) or anti-granzyme B-PE (clone QA16A02; Biolegend).

2.6 Cytotoxicity

NK cell-mediated tumor elimination was studied by calcein-AM staining as previously described.²⁵ Briefly, glioblastoma cells were incubated with 10 μM calcein-AM (Biolegend) for 30 min at 37°C. Following staining, tumor cells were cocultured with PBMCs, treated as designated for each experiment, at 10:1, 25:1 and 50:1 (effector: target) ratio for 4 h. Calcein release was measured on a Synergy H1 microplate reader (Biotek). For the indicated experiments, LN-18 cells were pretreated with temozolomide (100 μM, Selleckchem) for 72 h.

2.7 Determination of ILT2/HLA-I binding

First, glioblastoma cells were incubated with anti-HLA-(A,B,C) (clone W6/32; Biolegend), anti-HLA-E (clone 3D12HLA-E, ThermoFisher Scientific), or anti-HLA-G (clone 87G; Biolegend) blocking antibodies. After washing, cells were incubated with 20 μg/mL human ILT2 protein, Fc tag (AcroBiosystems) and then stained with PE-conjugated goat anti-human IgG Fc secondary antibodies (ThermoFisher Scientific). All incubations were performed for 45 min at 4°C. Samples were analyzed in a Cytoflex S flow cytometer with CytExpert 2.3 software (Beckman Coulter).

2.8 In silico data analysis

Expression levels of the genes of interest were analyzed in glioblastoma tumors and normal tissue from The Cancer Genome Atlas (TCGA) database using ClickGene (http://www.clickgenome.org/).²⁶ Survival of glioblastoma patients stratified by mRNA levels of selected genes was studied via TIMER2.0 platform (http://timer.cistrome.org/).²⁷ The protein–protein interaction network was analyzed for HLA-E using STRING (https://string-db.org/).²⁸ LILRB1 mRNA analyses were performed employing transcriptome data available on the Gene Expression Omnibus (GEO) repository (dataset GSE4412²⁹) via ShinyGeo (https://gdancik.shinyapps.io/shinyGEO/).³⁰ Correlations between mRNA expression levels of distinct genes were determined employing the cBioportal platform (http://cbioportal.org/).³¹ TISIDB repository (http://cis.hku.hk/TISIDB/) was employed to evaluate the correlation between gene expression and glioblastoma-infiltrating immune subsets.³²

2.9 Statistics

Sample distribution was determined using the Shapiro–Wilk test. The relationship between continuous and categorical prognostic variables was evaluated by independent samples Student’s t-test, and paired samples Student’s t-test was applied for intra-group comparisons. For mRNA expression comparisons, adjusted p-values [p(adj)] were calculated applying the false discovery rate (FDR) method. In all cases, differences were considered statistically significant for p-values: p < 0.05 (*), p < 0.01 (**), and p < 0.001 (***). Statistical analyses were performed using GraphPad Prism 8 software (GraphPad Software). A heatmap representing Spearman’s correlation coefficients was built with Morpheus software (https://software.broadinstitute.org/morpheus).

3.1 Classical and non-classical HLA-I molecules are highly expressed in glioblastoma

To evaluate the relevance of classical and non-classical HLA-I molecules in glioblastoma, we first interrogated publicly available genomic data from the TCGA database. Greater mRNA expression of HLA-A, -B, and -C was detected in glioblastoma samples compared to normal tissue (p(adj) = 6e-07, 6e-07 and 8e-07, respectively) (Figure 1A). HLA-E and, to a lesser extent, HLA-F mRNA expression was also significantly higher in glioblastoma (p(adj) = 1.44e-05 and 9.3e-05, respectively),^{33, 34} whereas no significant difference in HLA-G levels (p(adj) = 0.256) was found (Figure 1B). In line with this, strong expression of HLA-(A,B,C) was detected on glioblastoma cell lines (Figure 1C). These tumor cells also exhibited remarkable levels of HLA-E (Figure 1D).

The impact of HLA-I molecules on the survival of patients with glioblastoma was assessed by a Kaplan–Meier survival analysis via TIMER2.0. As depicted in Figure 2A, mRNA expression of HLA-I, including classical and non-classical molecules, did not affect survival in glioblastoma patients. Nonetheless, lower mRNA levels of each of these molecules significantly correlated with improved overall survival in patients with low-grade glioma (Figure 2B).

Given the high expression of HLA-A, -B, -C, and -E observed in glioblastoma, as well as their central role in antitumor immunity, we further studied the expression of these proteins in patients with glioblastoma by immunohistochemistry (n = 40). HLA-A was vastly expressed on tumor cells, with 36 (90%) patients showing positive staining (Figure 3A). In contrast, strong HLA-B staining was only detected in 6 (15%) patients, although diffuse expression was found in 12 (30%) patients (Figure 3B). A total of 24 (60%) patients positively stained for HLA-C, with 10 (25%) of them displaying weak staining (Figure 3C). HLA-E staining was detected in 29 (72.5%) patients, 22 (55%) of which exhibited diffuse positive expression and 7 (17.5%) showed weak patchy expression on tumor cells (Figure 3D). Since HLA-E is also expressed on immune cells, we further performed sequential staining of CD45, glial fibrillary acidic protein (GFAP), and HLA-E on a glioblastoma case. HLA-E staining was observed on tumor cells (Figure S1A), although it is worth mentioning that certain CD45+ cells within the tissue expressed this protein as well (Figure S1B). Tumor cell localization within tissue samples was evaluated by GFAP staining (Figure S2).

3.2 ILT2 is highly expressed in glioblastoma and correlates to immune infiltration

HLA-E, along with classical HLA-I proteins, is part of a complex network of interacting proteins tightly related to immune modulation and acts as ligand for an array of immune receptors, including the inhibitory checkpoints NKG2A and ILT2 (Figure S3). Thus, we analyzed the mRNA expression of KLRC1 (gene coding for NKG2A) and LILRB1 (gene coding for ILT2) in glioblastoma from the TCGA database. KLRC1 was barely expressed on either condition studied and no different expression was found (p(adj) = 0.761). However, glioblastoma exhibited elevated levels of LILRB1 compared to the control (p(adj) = 0.002) (Figure 4A), suggesting that ILT2 might play a role in this malignancy. ILT2 and, to a lesser degree, NKG2A surface expression was detected on glioblastoma cell lines as well (Figure S4). Interestingly, LILRB2-4 expression was increased in glioblastoma samples compared to normal tissue (p(adj) = 6e-05 for LILRB2, 1.44e-05 for LILRB3, and 1.6e-05 for LILRB4), whereas no change was detected in LILRB5 levels (p(adj) = 0.758) (Figure 4A).

In addition to the correlation between HLA-I molecules, in silico analysis revealed that LILRB1 correlated to HLA-E in glioblastoma (Spearman’s coefficient of 0.687) (Figure 4B). Similar to HLA-I molecules, Kaplan–Meier analysis revealed that LILRB1 expression significantly affected the overall survival of patients with low-grade glioma (Figure 4C). A similar trend was observed in glioblastoma, although it did not reach significance (p = 0.0554) (Figure 4C). Still, the expression levels of this receptor are increased in grade IV gliomas compared to grade III tumors (p = 0.0002), as shown by in silico analysis of previously published data from the GEO database (dataset GSE4412²⁹) (Figure S5A). Of note, high LILRB2 and LILRB3 levels negatively influence overall survival in patients with glioblastoma compared to other LILRB receptors, although nearly all of them influence survival in low-grade glioma (Figure S5B).

Overexpression of ILT2 on immune cells is frequently linked to an exhausted phenotype that leads to immune evasion in cancer. Therefore, we next assessed whether ILT2 might be expressed by immune cells in glioblastoma. LILRB1 expression was highly correlated to infiltrating NK cell numbers in glioblastoma, and CD8⁺ T cells displayed lower correlation (Spearman’s coefficient of 0.751 vs 0.508, respectively) (Figure 4D). To validate these findings, expression of ILT2, as well as CD57 (as a marker for NK cells), was assessed by immunohistochemistry. Although mild expression of ILT2 was observed on tumor cells, immunostaining of this protein was primarily detected on NK cells (Figure 4E), supporting the results obtained from in silico analysis.

3.3 ILT2 is upregulated on immune cells upon coculture with glioblastoma

To gain further insight into the dysregulation of ILT2 on infiltrating immune cells in glioblastoma, PBMCs from healthy donors were cocultured with glioblastoma cell lines, and surface expression of ILT2 was evaluated by flow cytometry (Figure 5A). Coculture with tumor cells triggered an upregulation of ILT2 on NK cells (Figure 5B). CD8⁺ T cells expressed higher levels of ILT2 upon coculture with LN-18 and U87MG cells as well (Figure 5B). Noticeably, no changes on ILT2 expression were detected for T98G cells. Glioblastoma-conditioned media failed to modulate ILT2 surface expression on healthy immune cells (data not shown), suggesting that this effect might be mediated by cell-to-cell contact. We next investigated whether glioblastoma cells affected the antitumor function of NK cells. As depicted in Figure 5C, PBMCs previously exposed to LN-18 cells exhibited reduced cytotoxicity compared to the control, pointing out that glioblastoma cells hinder NK cell antitumor activity.

3.4 ILT2 blockade partially restores antitumor immune responses against glioblastoma

To assess whether ILT2 takes part in the glioblastoma-induced suppression of NK cell function, PBMCs from healthy donors were treated with anti-ILT2 blocking mAb, and cytokine production in response to glioblastoma cells was assessed by flow cytometry (Figure 6A). ILT2 blockade induced IFN-γ production by NK cells (Figure 6B). Lower, yet significant, IFN-γ increase was observed for CD8⁺ T cells (Figure 6C). Granzyme B levels were significantly higher in anti-ILT2-treated NK cells exposed to LN-18 cells (Figure 6D). In contrast, ILT2 blockade did not affect granzyme B production by CD8⁺ T cells (Figure 6E). Treatment with anti-ILT2 blocking mAb also boosted NK cell-mediated lysis of glioblastoma cells (Figure 7A).

Next, we sought to elucidate whether ILT2-mediated increased cytotoxicity is related to ILT2/HLA-I interactions. In this context, HLA-E blockade significantly decreased ILT2 binding to glioblastoma cells, in a similar extent to HLA-G blockade (Figure 7B). siRNA-mediated HLA-E knockdown failed to neutralize the increase in cytotoxicity mediated by ILT2 blockade (Figure S6A,B). In the same line, no changes were detected in NK cell-mediated elimination after blocking HLA-E on glioblastoma cells (Figure S6C). Of note, NKG2A blockade did not enhance tumor cell elimination (Figure S7), which agrees with the results detected upon HLA-E blockade, because NKG2A stands as a key receptor for HLA-E. Despite partially decreasing ILT2 binding (Figure 7B), HLA-G blockade did not enhance glioblastoma cell killing either (Figure S8A). Notably, ILT2 binding was impaired in a greater fashion upon blockade of classical HLA-I ligands (Figure 7B), which significantly boosted the cytotoxicity of glioblastoma cells (Figure S8B), suggesting that ILT2 takes part in a complex signaling network. It is worth mentioning that blocking classical HLA-I molecules did not affect surface expression of their non-classical counterparts (Figure S9), ensuring the specificity of these results. Additionally, ILT2 blockade partly rescued the cytotoxic capacity of PBMCs previously exposed to LN-18 cells, thus counteracting, up to a certain point, the immunosuppressive pressure brought about by the tumor (Figure 7C).

Temozolomide stands as the first-line chemotherapy approved for postoperative treatment in patients with glioblastoma. To evaluate the effect of combining ILT2 blockade with this chemotherapeutic drug on NK cell-mediated cytotoxicity, LN-18 cells were exposed to temozolomide and then cocultured with PBMCs from healthy donors previously treated with anti-ILT2 antibodies. As illustrated in Figure 7D, not only did temozolomide treatment sensitized glioblastoma cells to immune-mediated elimination, but it significantly increased tumor lysis by anti-ILT2-treated immune cells. In agreement with previous studies, discreet MICA upregulation was observed on temozolomide-treated LN-18 cells (Figure S10A). Yet, NKG2D blockade did not counteract the increase in tumor cell elimination brought about by this chemotherapeutic agent (data not shown). It is worth mentioning that temozolomide treatment also led to downregulation of non-classical HLA-I molecules on glioblastoma cells (Figure S10B). In general, these findings reveal that a combination of ILT2 blockade with clinically available therapies might prove beneficial in reestablishing the immune function in patients with glioblastoma.