Identification of a Natural Killer Cell Receptor Allele That Prolongs Survival of Cytomegalovirus-Positive Glioblastoma Patients

One of the more devious strategies utilized by cancer is evasion of immune surveillance, our primary defense against pathogens and rogue mutated cells (1). The molecular mechanisms underlying this property of cancer cells have been the topic of intense investigation. Specifically, as a treatment strategy of human cancers, new generation immunotherapy has shown unprecedented success in some solid tumor types (2, 3). These approaches, including checkpoint-inhibiting immunotherapy, have unfortunately proven largely unsuccessful in the treatment of human glioblastoma, a highly aggressive malignant brain tumor in adults, where the standard, multimodal treatment extends survival to only 14.6 months (4, 5). The approach in most ongoing immunotherapy trials is to use genetically modified autologous T lymphocytes and dendritic cells. None propose to investigate natural killer (NK) cells as therapeutic effectors despite their ability to spontaneously detect and destroy malignant and virus-infected cells without prior sensitization or antigen specificity (6, 7). NK cells' activity is regulated by the balance of activating and inhibitory signals that are transmitted through their receptors, including killer immunoglobulin-like receptors (KIRs) upon ligation to their cognate HLA ligands (8, 9). The KIR receptors are encoded by a set of 15 genes for inhibitory and activating receptors on chromosome 19q13.4. Ligation of inhibitory KIRs (iKIRs, comprising two or three immunoglobulin domains and long cytoplasmic tails, KIR2DL or KIR3DL) to their cognate HLA ligands leads to inhibitory signaling, rendering the NK-cell tolerant to self cells (10). The KIRs are categorized into phylogenetic lineages II and III (11), where the latter is denoted by a polymorphism at position 44 in the D1 domain. Amino acid substitutions at this position confer specificity for particular HLA-C epitopes.

Activating KIRs, (aKIRs, comprising two or three immunoglobulin domains and short cytoplasmic tails, KIR2DS or KIR3DS) have high sequence homology to the corresponding iKIR and share HLA-ligand specificity, albeit with reduced binding affinity (12). It has been suggested that ligation of aKIRs to cognate HLA ligands results in NK-cell licensing (13). Despite epidemiological studies implicating aKIR–HLA–ligand combinations with disease entities, the physiological role of aKIR receptors in cancer has been elusive. KIR2DS1 and KIR2DL1 both bind class I HLA-C2 group ligands, whereas KIR2DS4*00101 was shown to bind some HLA-C1^Asp80, HLA-C2^Lys80 epitopes that are also recognized by KIR2DL2/3 (14), and HLA-A*1101, HLA-A*1102 epitopes recognized by KIR3DL2 (15, 16). However, careful analysis indicated that KIR2DS4 does have selectivity for HLA-C epitopes that is distinct from KIR2DL1 and KIR2DL2/3, possibly implying an important and unique role for KIR2DS4. KIR2DS4 is the most prevalent and evolutionary conserved lineage III KIR (11) distinguished from other lineage III KIRs by a Pro71-Val72 motif in the D1 domain, that it shares with lineage II KIR3DL2 as a result of gene conversion before separation of humans from chimpanzee ancestry 6.5 to 10 million years ago (15, 17). KIR2DS4 gene also exists as deletion mutants (KIR2DS4*00030101-2; KIR2DS4*004-9), resulting in nonfunctional protein lacking intracellular and transmembrane anchoring (18). Moreover, KIR2DS4*00101 specifically recognizes an as yet classified, non-class I HLA ligand expressed on melanoma, leading to potent killing of these targets (19). Cytokines and viral infection may break tolerance induced by ligation of iKIRs to self-HLA, invoking potent cytolysis via the activating receptors (20, 21). Hence, some activating KIRs have been associated with protection from viral infections (20, 22, 23) and induction of autoimmunity (24, 25). Intriguingly, some autoimmune diseases are associated with reduced risk of glioma (26). The common denominator being increased inflammation due to pronounced IFNγ release and augmented cytolytic responses.

However, virus infections, such as with cytomegalovirus (CMV), may render infected cells highly resistant to immunosurveillance. CMV proteins contribute to immune evasion in part by targeting NK cells (27). In addition, viral proteins downregulate class I HLA to evade antigen presentation and detection by CD8⁺ T cells (28). CMV infection alters the distribution of NK-cell receptors (28) to generate a greater abundance of terminally differentiated NK-cell subsets characterized by elevated CD57, KIRs, activating receptor CD94/NKG2C, and prolonged persistence in peripheral blood (29). Although CD94/NKG2C⁺ NK cells have been shown to be remobilized upon subsequent CMV reactivation (30, 31) and to potently kill antibody-coated cells via antibody-dependent cellular cytotoxicity (ADCC; ref. 32), these findings have largely been demonstrated only in healthy individuals.

Interestingly, greater than 90% of glioblastomas express CMV gene products (33–35), and high-grade CMV infection is associated with poor overall survival (36). These results suggest that CMV could contribute to the development of glioblastoma through evasion of immune surveillance, and in fact, treatment of CMV-positive glioblastoma patients with anti-retroviral drugs prolonged survival (37). Although most studies have focused on tumor molecular changes that impact disease progression, few have investigated immune molecular profiles associated with glioblastoma. Here, we hypothesized that CMV infection interferes with immune responsiveness and contributes to glioblastoma progression by altering the distribution of activating KIR–HLA ligand interactions, which regulate NK-cell functionality and are important for target recognition (38). Our results reveal for the first time the association of a specific KIR2DS4*00101 gene allele with protection from glioblastoma development and enhanced survival in patients who do contract glioblastoma. This novel finding is of direct clinical relevance for the identification of patients who may exhibit a less aggressive clinical course and who may potentially respond to immunotherapy.

Hundred and eight (108) glioblastoma biopsies and paired blood samples were obtained during surgical resections at the Haukeland University Hospital, Norway, between 1998 and 2015. There were n = 62 males and n = 46 females of median 61 years (range 10–81 years). Material was collected with the informed consent of patients and healthy controls. The regional ethical committee approved the study (REK vest 013.09/20879; 2014/588). Neuropathologists confirmed glioblastoma diagnosis and eligibility criteria included availability of follow-up data, less than 50% necrosis, and only biopsies obtained at primary glioblastoma diagnosis. Survival was determined as time elapsed from the date of surgery to death or last follow-up until death or August 2015. The six living patients at this last follow-up had been followed up for 10.7 months.

Tumor-derived DNA exome sequences from mixed American glioblastoma patients (n = 300; 191 males and 108 females; wherein white Caucasian patients accounting for n = 265, 96 males and 168 females that could be substratified in analyses; Supplementary Tables S1 and S2) collected into the TCGA database between 1993 and 2013 were analyzed. Frequency of KIR2DS4*00101 that codes full-length membrane anchored activating receptor (FUNC) or (KIR2DS4*0030101-2; KIR2DS4*004-9) 22 bp deletion variants (DEL) that encode nonfunctional protein was determined (detailed bioinformatics description in Supplementary Methods and in Supplementary Figs. S1, S2, and S3). Ethical approval and access to the database was obtained from the TCGA Network and The NCBI dbGaP. Survival time was determined as the time elapsed to the date of death and/or last follow-up.

Plasma and peripheral blood mononuclear cells (PBMC) were isolated from age-matched controls (n = 454 Caucasian donors; 301 females and 153 males) using standard procedures. Median age was 49 years (range 20–73 years). Blood-derived DNA exome sequences from controls (n = 2504) in the 1000 Genome (OKG) public domain database were also analyzed. Information on sex, ethnic origin, and KIR2DS4 status was obtained (Supplementary Table S3), and bioinformatics analysis is described in supplementary methods.

HLA class I (A, B, and C) locus was genotyped in glioblastoma patients (n = 108) and controls (n = 76) using sequence-specific oligonucleotide probe hybridization (ProImmune) within four-digit resolution and KIR genotyping utilized the KIR Typing Kit (Miltenyi Biotec), according to the manufacturer's protocol. The remaining controls (n = 363) were KIR and class I HLA genotyped as previously reported (39).

Tumor DNA was sodium bisulfite converted using the DNA Methylation-Gold Kit (Zymo Research) and methylation-specific PCR performed using standard protocols.

PBMCs were harvested by density gradient centrifugation in Lymphoprep using standard procedures. Thereafter, PBMCs from healthy controls (n = 56) and glioblastoma patients (n = 21) were fixed and phenotyped as listed in Supplementary Table S4. For good separation of KIR3DL2 and KIR3DL1, as well as KIR2DL2 and KIR2DL3, two-step staining was performed. PBMCs were preincubated for 20 minutes with anti-KIR2DL1 and anti-KIR3DL1 Abs before incubation with anti-KIR3DL2/1 and anti-KIR2DL2/3 Abs. Fluorescence minus one (FMO) for each channel was used and lymphocytes gated as indicated in Fig. 1B. Data acquired on LSR Fortessa (BD Biosciences) was analyzed using FlowJo, version 10 (Tree Star Inc.).

Plasma from patients (n = 27) and healthy controls (n = 91) was serotyped for CMV-specific IgG and IgM antibodies using ARCHITECT CMV ELISA and PCR-based assays (Abbott). A clinical diagnostic cut-off for seropositivity was designated as IgG (Au/mL) ≥6 and IgM (index) ≥1. The CMV antigen UL83 (pp65) was detected by qPCR on genomic DNA from paired blood and glioblastoma biopsy samples (n = 108) as previously described (40, 41). A CMV-infected patient sample was used as positive control and pp65 considered positive when cycle threshold (C_t) values were ≤34.

Patient survival was presented using the Kaplan–Meier method (42), and Cox-proportional hazards regression was used to assess significant differences adjusted for age, postoperative treatment, sex, and MGMT as possible confounders. Fisher exact test was used to compare gene frequencies between patient and healthy control cohorts. Two-way ANOVA with Bonferoni correction for multiple testing was used. Descriptive statistics were reported as mean ± SEM unless otherwise stated and two-sided P-values less than 0.05 were considered significant. Statistical analyses were performed in Stata version 13.1 (Texas) or the Graphpad PRISM 6.0 software (La Jolla, CA). Figures were made in R version 3, The R foundation.

We previously demonstrated that NK cells infiltrating patient glioblastoma tumors highly express NKG2D (43) and that subsets derived from KIR2DS4 immunogenotype donors were more potent against glioblastoma stem-like cells (44). Therefore, we hypothesized that altered distribution of NK cells' KIR–HLA ligand interactions as a result of CMV infections may influence malignant progression of glioblastoma. Specifically, we postulated that reduced frequency of the potent KIR2DS4*00101 allele may underlie this progression. Genomic analysis of all individuals demonstrated that iKIR and aKIR gene frequencies were as previously reported (39), as were frequencies of the haplotypes within glioblastoma and healthy Norwegian control groups when analyzed independently (Fig. 1A). The frequency of KIR2DS4*00101 did not differ between glioblastoma patients and controls (41.12% vs. 41.46% respectively; χ², P = 0.94; Fig. 1A). Stochastic expression of KIR genes on NK cells means they may not necessarily be expressed even if the gene is present. Thus, we confirmed surface expression of the KIR receptors on CD56⁺CD3⁻ NK cells (Fig. 1B), where KIR3DL2 and KIR2DL2/3 were highly expressed in controls than glioblastoma NK cells (two-way ANOVA, df = 1, P < 0.0001 and P < 0.01, respectively, Fig. 1C). KIR2DL3 expression appeared in negative linkage disequilibrium with KIR2DL2.

Engagement of the inhibitory KIR receptors that are licenced by self-HLA ligands during development potentiates NK-cell responses against target cells lacking these self-HLA ligands (45). Therefore, we asked whether the presence of KIRs in the context of their cognate HLA ligands differed in glioblastoma patients compared to controls. The KIR2DS4*00101 allele was present at 100% concordance with HLA-C1 ligands in glioblastoma patients but not in controls (Fisher exact test, P = 0.004; Fig. 1D; and Table 1), whereas it was reduced in the presence of HLA-C2 ligands in patients compared to controls (43.18% and 53.59%, respectively; Fisher exact test, P = 0.027; Fig. 1E; and Table 1). Finally, there was a trend for KIR3DS1 association with HLA-Bw4 in controls but not in glioblastoma patients (Fisher exact text, P = 0.079; Table 1). Taken together, these results indicated a potentially functional and specific interaction of the KIR2DS4*00101 allele with both of its ligands in glioblastoma patients but not in controls.

Next, we asked whether the strong association of KIR2DS4*00101 with HLA-C1 had impact on patient outcomes. Carrying KIR2DS4*00101 was significantly associated with improved patients' survival by 2.1 months (median 11.8 months vs. 9.7 months, HR 0.6, P = 0.034), or restricted mean survival by 3.3 months (14.9 months vs. 11.6 months; Fig. 1F and Table 2). Importantly, the survival benefit of carrying KIR2DS4*00101 was independent of other strong prognostic factors, including age (HR 1.34; P = 0.002; Fig. 1G), MGMT promoter methylation (HR 2.0, P = 0.008; Fig. 1H), and postoperative treatment (P = 0.049; Fig. 1I; Table 2).

To rule out the possibility that survival benefit observed with KIR2DS4*00101 was simply due to HLA-C1 in the absence of HLA-C2 ligation of KIR2DL1 that render autologous NK cells strongly hyporesponsive (46), we performed Cox regression survival analyses comparing associations of HLA-C1 versus HLA-C2 corrected for KIR2DS4*00101 on glioblastoma patient survival. Patients bearing HLA-C1/C1 in absence of KIR2DS4*00101 had median survival of 8.2 months, which was the poorest survival time for all groups (Fig. 1J), whereas the median survival times of glioblastoma patients HLA-C2/C2 and HLA-C1/C2 without KIR2DS4*00101 were better at 10.5 months (data not shown) and 10.9 months (HR = 0.537; 95% CI, 0.28–1.02; P = 0.056; Fig. 1J). Carrying KIR2DS4*00101 in the presence of HLA-C1/C1, however, significantly extended patient median and restricted mean survival outcomes to 13.8 and 17.4 months, respectively (HR 0.36; 95% CI, 0.18–0.73; P = 0.005; Fig. 1J) as well as in the context of HLA-C1/C2, where median and restricted mean survival extended to 14.2 and 17.2 months, respectively (HR 0.34; 95% CI, 0.16–0.70; P = 0.004; Fig. 1J). No patients were KIR2DS4*00101⁺ in the context of HLA-C2/C2.

As KIR2DS4*00101 binds strongly only to a select group of HLA-C1 (C*1601, C*0102, C*1402) and HLA-C2 epitopes (C*0501, C*0201, C*0401; refs. 15, 16), we investigated association of these specific HLA-C1/C2 epitopes with KIR2DS4*00101 in glioblastoma patients and controls from the Norwegian cohort. We also investigated the association of KIR2DS4*00101 with the weakly binding HLA epitopes (C*0304 and C*0702). 47.22% (17/36) of patients carried KIR2DS4*00101 in context of both strongly binding HLA-C1/C2 epitopes compared to 67.86% (38/56) of patients lacking the KIR2DS4*00101 allele (χ² = 3.84; OR 0.423; 95% CI, 0.17–1.027; P = 0.051; Fig. 1K). Strikingly, none of the healthy controls carried the KIR2DS4*00101 allele in context of the strongly or weakly binding HLA-C1/C2 epitopes (data not shown). Eight KIR2DS4*00101 bearing patients, and seven patients without KIR2DS4*00101 allele lacked the binding epitopes. These findings demonstrated a cancer-specific positive association of the KIR2DS4*00101 allele with selective HLA-C1/C2 binding alleles.

We next investigated the hypothesis that CMV infection in glioblastoma patients might lead to redistribution of the NK cells' KIR–HLA ligand repertoire. Intriguingly, KIR2DS4*00101 was in 100% concordance with CMV seropositivity in glioblastoma patients but not in healthy controls (Fishers exact P = 0.031; Fig. 2A). Sixty-seven percent (18/27) of glioblastoma patients were IgG seropositive compared to 50.5% (46/91) of healthy Norwegian controls (OR 11.25, exact logistic regression; P = 0.044). Two patients and controls were additionally IgM seropositive. Of the glioblastoma patients, 30.5% (33/108) were positive for pp65 DNA in tumor tissue and/or blood, whereas 12% (12/100) were positive for pp65 DNA in blood, which could be an indication of the reactivation of CMV (data not shown). Seropositivity correlated with increased presence of pp65 CMV DNA in the tumor microenvironment in only 11.1% (3/27) cases (Fig. 2B). NK cells from all glioblastoma patients were more differentiated than those from controls (Fig. 2C), as more NK cells expressing differentiation markers, CD57, CD16, NKG2D, and CD94/NKG2C, were present in glioblastoma patients who were CMV seropositive (Fig. 2C and D). Thus, there was a dual effect of cancer and CMV seropositivity on expression of these NK-cell differentiation markers and expansion of the CD94/NKG2C subset.

We hypothesized that association of seropositivity with KIR2DS4*00101 might thus be due to CMV imprinting of CD94/NKG2C⁺ NK-cell compartment. Therefore, we investigated whether KIR2DS4⁺ NK-cell subsets were associated with the CD94/NKG2C subsets acquired in response to CMV infection. In the KIR2DS4*00101⁺, CMV⁺ glioblastoma patients, KIR2DS4 was coexpressed with 35.8 ± 10.1% of NKG2C⁺ NK cells (Fig. 2E). In KIR2DS4*00101⁺, CMV⁺ controls, the percentage of coexpressing NK cells was similar (32.44 ± 9.1%; P > 0.05; Fig. 2E). CD94/NKG2C subset analysis further confirmed that elevated expression of CD16 (P < 0.01) and NKG2D (P < 0.05, two-way ANOVA, df = 1; Fig. 2F and G) was a characteristic of more NK cells in CMV⁺ glioblastoma patients compared to controls. However, no difference in the levels of CD16 was apparent in CMV⁺KIR2DS4*00101 bearing patients and controls (Fig. 2H), indicating that CD16 expression was increased because of CMV infection and not KIR2DS4*00101 status (Fig. 2H and J). In contrast, expansion of KIR3DL2/1⁺ subsets in healthy controls was irrespective of their CMV status (two-way ANOVA, df = 1, P < 0.0001; Fig. 2I and J). Taken together, these data indicate that selective expansion of CD94/NKG2C⁺/CD16⁺ NK-cell subsets was a result of CMV infection, and expansion of this subset in the context of KIR2DS4*00101 with HLA-C1/C2 was associated with improved survival in glioblastoma patients.

Given the favorable prognosis of KIR2DS4*00101 in the context of its HLA-C1/C2 ligands and CMV infection in the Norwegian glioblastoma patients, we hypothesized that the KIR2DS4*00101 allele might increase inflammation and be protective against glioblastoma development. We thus investigated the risk of ending up in the glioblastoma group as opposed to the noncancer group if the individual carries two alleles of KIR2DS4*00101 (KIR2DS4FUNC/FUNC) relative to the deletion alleles (KIR2DS4DEL/DEL). The number of mixed American glioblastoma patients possessing KIR2DS4FUNC/FUNC was significantly diminished compared to healthy multinational controls (13.33%, 40/300 vs. 18.77%, 470/2504, respectively, P = 0.029; Fig. 3A). Likewise, white Caucasian glioblastoma patients possessing KIR2DS4FUNC/FUNC was significantly diminished compared to healthy multinational controls (10.19%, 27/265 vs. 18.77%, 470/2504, respectively, P = 0.001; Fig. 3A), Multinational controls possessing KIR2DS4FUNC/FUNC allele relative to KIR2DS4DEL/DEL had 37% reduced risk of ending up in mixed glioblastoma patient group (OR 0.63; 95% CI, 0.41–0.98%; n = 2804; P = 0.0038), in contrast to 50% reduced risk of ending up in the white glioblastoma patient group (OR 0.496; 96% CI, 0.327–0.752%; n = 2769; P = 0.001; Fig. 3A).

Likewise, American Caucasian controls carrying KIR2DS4FUNC/FUNC (20.46%, 71/347) had 33% reduced risk of ending up in mixed glioblastoma patient group (OR 0.67; 96% CI, 0.47–0.96%; n = 647; P = 0.029; Fig. 3A). However, risk of American Caucasian controls ending up in the white glioblastoma patient group was substantially reduced by 53% (OR 0.466; 95% CI, 0.287–0.755%; n = 612; P = 0.002). This finding underscores the importance of controlling for ethnicity as potential confounder.

Associations with age, Karnofsky performance and treatment were also examined in order to validate our overall statistical approach and our cohorts. First, age was prognostic (HR 1.042; 95% CI, 1.03–1.06%; P < 0.001). The median survival was 10.5 months in glioblastoma patients <55 years, 9.6 months for patients between 55 and 65 years, and 5.4 months in patients >65 years (Fig. 3B). Second, Karnofsky performance status of at least 80 was associated with improved survival by 29 days (HR 1.85; 95% CI, 1.28–1.66%; P = 0.001; Fig. 3C). Third, the standard treatment of Temozolomide combined with 60 Gy ionizing radiation (IR) significantly improved median survival compared to chemotherapy only (10.9 months vs. 5.5 months, respectively; HR 2.86; 95% CI, 1.78–4.60%; P < 0.0001; Fig. 3D). The standard treatment was also superior to Temozolomide + low-dose IR (median survival, 6.8 months; HR 2.43; 95% CI, 1.09–4.11%; P = 0.018), IR alone (median survival, 6.9 months; HR 2.11; 95% CI, 1.09–4.11%; P = 0.028), unknown treatment (median survival, 3.7 months; HR 4.35; 95% CI, 3.06–6.19%; P < 0.0001), or no treatment (median survival, 2.3 months; HR 11.84; 95% CI, 5.53–25.35%; P < 0.0001; Fig. 3D). Finally, the standard treatment was not better than IR combined with other alkylating chemotherapy drugs (median survival, 10.9 months; HR 1.61; 95% CI, 0.90–2.89%; P = 0.108; Fig. 3D). These results corroborate established findings in the field, and therefore support our analysis regarding the KIR2DS4*00101 allele.

Immunotherapy is beginning to take hold as a viable option for the treatment of some solid tumor types. Unfortunately, the modality has been so far ineffective in the treatment of human glioblastoma, a clinically aggressive and psychologically devastating disease with poor treatment options. Understanding the molecular factors that drive immunogenicity in glioblastoma patients is further required for designing effective immunotherapy strategies that might be based on the specific immunobiological characteristics of each patient's tumor. Here, we have used the unique approach of characterizing large cohorts of glioblastoma patients and healthy controls based on NK-cell receptor repertoires and associated these profiles with risk of acquiring the disease and progression with endpoint patient survival. We are the first to report KIR2DS4*00101 allele as an independent prognostic marker for improved survival in glioblastoma patients, where the effect was further modified by the presence of its cognate HLA-ligands in the context of CMV. Our findings should inspire greater focus on molecular characterization of immune of cell types in solid tumors to not only devise appropriate immunotherapy strategies but importantly, to identify those patients most likely to respond. Because KIR2DS4*00101 associated with protection from developing highly aggressive glioblastoma, we identified also a prognostic factor that may inform loss of health life years. Our findings highlight a new paradigm that may advance research within solid tumors, where focus is still on subtyping tumors based on their intrinsically aberrant molecular profiles.

Indeed, few studies have investigated the potential of NK-cell therapy for glioblastoma patients (47). We have previously demonstrated that NK cells infiltrating patient glioblastoma tumors highly express activating receptor NKG2D (43) and that NK cells derived from aKIRs KIR2DS4⁺ and KIR2DS2⁺ donors were more potent against glioblastoma stem-like cells (44). Thus, the specific association of KIR2DS4*00101 with HLA-C1 in the context of CMV seropositivity in glioblastoma patients but not healthy controls has intriguing implications for immune response, although the mechanisms underlying their association remain unknown. We speculate that licensing of KIR2DS4 in the context of CMV might modulate receptor specificity and that these NK-cell subsets might bind preferentially and more strongly to HLA-C1 and HLA-C2 ligands. Indeed, nearly half of the KIR2DS4*00101 bearing CMV⁺ patients had the KIR2DS4*00101 allele in context of strongly binding HLA-C1/C2 epitopes in contrast to healthy controls, where none had KIR2DS4*00101 in context of these strong epitopes. This may indicate a cancer and/or CMV-specific association, as well as an important role HLA-C in cancer as opposed to HLA-A because no significant difference in association of KIR2DS4*00101 with HLA-A11 ligands was uncovered. Unfortunately, because diminished sample size in these subanalyses, we could not determine whether KIR2DS4*00101 association with specific HLA-C1/C2 binding epitopes further enhanced patient survival.

A possible mechanism for the appearance in patients, but not controls, of KIR2DS4*00101 NK cells in context of strongly binding HLA-C1/C2 epitopes may be through CMV peptides generated in post-childhood infections, which enhance the specificity and strength of KIR2DS4*00101 binding to HLA-C1/C2 epitopes and ultimately impact NK cell response. Indeed peptides have been shown to influence the specificity of particular KIR–HLA ligand interactions (48). Unfortunately, the associative nature of our study falls short in providing precise mechanistic insights of how the KIR2DS4*00101–HLA-C1/C2 interactions may be contributing to the immune regulation in glioblastoma patients. Indeed, a caveat that the observed effect of KIR2DS4*00101 may potentially be mediated by linked genes cannot be entirely ruled out. However, because glioblastoma patients lacking KIR2DS4*00101 exhibited poor prognosis despite frequency of strongly binding HLA-C1/C2 epitopes, the importance of KIR2DS4*00101 may be accentuated. KIR2DL2 that shares these epitopes had no significant impact on patient survival.

HLA-C2/C2 was associated with a shorter median survival of 10.5 months, whereas patients with HLA-C1/C2 had modest survival 10.9 months. In other cancer types, HLA-C2 has been associated with higher relapse rates (46, 49), and diminished progression free and overall survival. However, in the context of CMV seropositivity, HLA-C1/C2 plus KIR2DS4*00101 was associated with an extended median survival of 14.2 months. An interpretation of these findings might be that HLA-C1/C2 educated KIR2DS4 subsets in context of CMV antibodies override inhibitory effects of KIR2DL1 ligation to HLA-C2 and subsequent tolerance to self. However, our findings do not rule out the possibility of KIR2DS4*00101 recognizing similar non-class I HLA ligands in the glioblastoma tumors that were previously reported in melanoma (19). The possible significance of our findings is that the immune system of KIR2DS4*00101 glioblastoma patients might be capable of generating potent proinflammatory NK-cell subsets in response to CMV infection that effectively impede malignant progression. Indeed, the uncovered protection against glioblastoma development in the independent populations of the OKG and TCGA networks implies a causal benevolent effect of KIR2DS4*00101 allele against malignancy. Such NK-cell subsets carrying the KIR2DS4*00101 allele may promote inflammation through potent IFNγ release and cytotoxic responses. Further studies are underway to delineate the precise role of NK cells in glioblastoma patients in the context of CMV infection with a particular focus on elucidating whether NK cells from individuals bearing KIR2DS4*00101 allele display distinct functional activity towards glioblastoma cells with or without CMV infection. Furthermore, it would be intriguing to investigate whether the strong protection against glioblastoma might be apparent in other solid tumors with a strong anti-inflammatory signature and in other carefully controlled ethnic groups.

Activating KIR receptors have been associated with delayed progression of various diseases, including respiratory papillomatosis, human immune deficiency virus to AIDS, and Hepatitis C virus (20, 23, 50). A critical role for CD94/NKG2C in antiviral response was revealed in several studies performed on AIDS patients, where a natural chromosomal variation, which has a deletion of KLRC2, the gene that codes for CD94/NKG2C, was associated with increased risk of contracting HIV and more aggressive disease progression to AIDS (51). It is therefore possible that a heightened state of anti-CMV immune surveillance in carriers of KIR2DS4*00101 forestalls development and progression of tumors.

CD94/NKG2C⁺ NK-cell subsets have previously been correlated with CMV seropositivity in healthy individuals (29, 52), and these subsets have been found to be highly cytotoxic against viral targets and antibody-coated cells. Ordinarily, 10% of NK cells in peripheral blood express CD94/NKG2C. Thus, the observed three-fold increase of these NK-cell subsets in KIR2DS4*00101, CMV seropositive glioblastoma patients, and controls was a consequence of CMV imprinting. The role of CMV imprinting in stimulating development of these subsets is further supported by elevated expression of differentiation markers CD16, CD57, and NKG2D in both CMV⁺ patients and controls. Loss of CD16 and NKG2D is characteristic of NK cells from cancer patients and is a hallmark of immune escape. Changes in these receptors confer poor cytotoxicity, reduced cytokine production, and the inability to execute ADCC onto NK cells. Our finding of coexpression of CD94/NKG2C with KIR2DS4⁺ on NK-cell subsets from CMV-seropositive glioblastoma patients also highlights the role of aKIRs in response to CMV infection. The association of KIR2DS4*00101 with enhanced survival in glioblastoma patients underscores this argument.

CMV seropositivity in glioblastoma patients was high compared to healthy Norwegian controls (67% vs. 50.5%, respectively), but may be explained by a higher median age (61 years vs. 49 years; glioblastoma patients and controls, respectively) as CMV seropositivity has been correlated with age, race/ethnicity, gender, and socioeconomic status (53). However, the serostatus of our Norwegian cohort corroborates results reported from a German cohort of similar social demographics (54). Of our entire glioblastoma patients, 30.5% were pp65 positive in tumor, whereas 12% had CMV DNA in blood. The latter may be indicative of reactivation as a consequence of changes in the immune function of patients as CMV generally establishes latency after primary infection during childhood or adolescence (55). One difficulty is that there is currently no consensus regarding the best method for demonstrating CMV positivity in glioblastoma patients. Others have reported positivity for IE1 protein in 11% of tumors using immunohistochemistry and fluorescence in situ hybridization (56).

In summary, KIR2DS4⁺NKG2C⁺–HLA-C1/C2 interactions emerged as a novel, positive prognostic indicator for the glioblastoma patients undergoing standard treatment. Further work will characterize whether the CD94/NKG2C⁺ NK cells correspond to adaptive NK subsets exhibiting long-term persistence in blood and whether these subsets can effectively assist in the execution of immunotherapy targeting glioblastoma.

No potential conflicts of interest were disclosed.

Conception and design: M. Chekenya

Development of methodology: M. Dominguez–Valentin, A. Gras Navarro, A.M. Rahman, S. Kumar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Gras Navarro, E. Ulvestad, M. Lund-Johansen, B.A. Lie, P. Øyvind Enger, G. Njølstad, E. Kristoffersen, M. Chekenya

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Dominguez–Valentin, A. Gras Navarro, A.M. Rahman, S. Kumar, E. Ulvestad, V. Kristensen, P. Øyvind Enger, S.A. Lie, M. Chekenya

Writing, review, and/or revision of the manuscript: M. Dominguez–Valentin, A. Gras Navarro, A.M. Rahman, S. Kumar, E. Ulvestad, V. Kristensen, M. Lund-Johansen, B.A. Lie, G. Njølstad, E. Kristoffersen, S.A. Lie, M. Chekenya

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Retière, B.A. Lie

Study supervision: S.A. Lie, M. Chekenya

Other (Statistical analyses): S.A. Lie

Other (Provided funding for study): M. Chekenya

We are grateful to Bendik Nordanger, Ingrid Sandvik Gavlen, Varathalingam Sharanga, and Guro Gundersen for technical assistance, and neuropathologist Dr. Hrvoje Miletic for scoring viable tumor/necrosis fraction. We thank the patients, controls, and the staff at the neurosurgical unit at Haukeland University Hospital for providing tumor tissue, blood samples, and for the Brain Tumor Biobank. We thank the Norwegian Bone Marrow Donor Registry and The Blood Transfusion Bank, Bergen, for providing healthy controls. Flow cytometry was performed at the molecular imaging centre at UiB. The results presented here are in part based upon data generated by TCGA and The 1000 Genome research networks. We thank Mr. Tore Linde at the Dept. Information and Technology at UiB for secure analysis and supercomputing platform. We are grateful to Dr. Janice Nigro for manuscript editing.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.