Patrolling leukocytes detect mutated, early transformed cells, deem them sufficiently “non-self” and co-ordinately eliminate them, as predicted by the cancer immunosurveillance concept [22]. As cells undergo oncogenesis, neoantigens are released and captured on major histocompatibility complex (MHC)/ human leukocyte antigen (HLA) of dendritic cells (DCs) that subsequently mature and migrate to central lymphoid organs. Here, the peptide neoantigen on the DCs’ MHC is presented to the awaiting CD4⁺ or CD8⁺ T cell receptor (TCR) complex. Subsequently, binding of the CD28 co-stimulatory receptor to the DCs’ CD80/86 receptor fully activates the cytotoxic T cells which then migrate to infiltrate the tumour and kill the cells by locally releasing perforin and granzymes [23,24]. These lymphocytes successfully eliminate the genetically unstable tumour cells with intrinsically high immunogenicity [25] through a series of successive stages [26]. The T cells also effectively terminate their activation and proliferation as a means of avoiding autoimmunity, resulting in different phenotypes that either further activate Th1 immune responses or suppress via Th2-driven responses. Surface receptors such as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), programmed cell death protein 1 (PD-1) and nuclear transcription factors attenuate T cell responses, where CTLA-4 competes with CD28 for binding to CD80/86, resulting in inhibitory downstream signalling [27]. PD-1 is an immunoinhibitory receptor that stymies lymphocyte proliferation and cytokine secretion when bound to its membrane-bound or secreted ligands, PD-L1 or PD-L2, expressed by both immune and tumour cells [28]. In addition, activated T cells can express an inducible co-stimulator (iCOS), a surface receptor that is structurally and functionally similar to CD28 and enhances expression of Th2-related interleukin (IL)-10 rather than immune activating IL-2 [29]. This immunosuppressive helper T cell phenotype can be further induced by the zinc-finger transcription factor GATA3, which regulates Th2 cytokine expression [30]. Infiltrating CD8⁺ T cells in GBM have been shown to bear an exhausted phenotype, with upregulated expression of PD-1, LAG-3, TIGIT and CD39 [31,32,33].

GBM is considered an immunologically “cold tumour” due to the immunosuppressive microenvironment [33,34] and few immunogenic tumour-associated antigens (TAAs) [35]. The rapidly growing tumour alters the balance of interaction between cancer and immune cells, by outstripping its metabolic resources and shifting to glycolysis [36]. Cancer cells also recruit and alter nearby stromal cells to aid the tumour cells in avoiding immune detection and destruction [37,38]. Competition for glucose results in low pH, which creates a hostile tumour microenvironment [39,40,41] and drives a local increase in immunosuppressive stromal cells (Figure 1). These cells secrete immune inhibitory growth factors and cytokines, including vascular endothelial growth factor (VEGF), which is primarily produced by microglia, myeloid-derived suppressor cells (MDSCs) and tumour-associated macrophages (TAMs). TAMs and regulatory T cells such as the CD8⁺ CD28^- FoxP3⁺ T_reg cells also secrete immunosuppressive IL-10 [33] which downregulates interferon gamma (IFNγ) and MHC I through the JAK/STAT pathway [42]. Accumulating MDSCs are rendered in an immature state [43], secreting IL-13, which when bound to IL13Rα2, send anti-inflammatory signals to promote the alternative M2 phenotype in macrophages, causing them to lose their cytotoxic function and instead become immunosuppressive. Tumour cells aid in this immunosuppression by secreting transforming growth factor β (TGFβ) and expressing both membrane-bound and soluble PD-L1, inducing exhaustion in T and NK cells [44,45]. Moreover, the molecular heterogeneity of GBM, observed as cells with wildly dissimilar molecular genetic aberrations within the same tumour mass [16,17] allows the tumour microenvironment to evolve and adapt to the immunological selection pressure. These conditions lead to immune escape and an immune-edited microenvironment that is refractory to killing by cytotoxic lymphocytes, while promoting resistance to standard treatments. New, personalised therapies that specifically address these molecular, genetic and immunological challenges are sorely needed.

Another approach recently translated into clinical trials is chimeric antigen receptor (CAR) NK therapy. CARs are a synthetic, modified representation of the TCR complex, that contains both an antigen-binding site and a T cell costimulatory domain that allows for autonomous, MHC-independent activation of the T cells [48,49]. They consist of an extracellular single-chain variable fragment (ScFv) derived from an antibody, and an intracellular domain with signalling motifs. Over time, adjustments made to CARs have targeted their structure; where first-generation CARs contained the intracellular CD3ζ from a TCR, second-generation CARs included the costimulatory CD28 or 4-1BB endodomains fused to CD3ζ (reviewed in [50]). Originally, the only difference between CAR T and CAR NK cells was the lymphocyte that held the CAR. The CARs had the same structure in both lymphocytes, since CARs with the CD3ζ chain alone or in combination with other costimulatory domains also promoted NK cell activation [72]. However, as the field advanced, domains from adaptor molecules associated with NK receptors such as DAP10 and DAP12 have been introduced. While CAR NK cells with a DAP10 domain showed poor efficiency [73], CAR NK cells with DAP12 demonstrated enhanced cytotoxicity when compared with those with CD3ζ [74].

Cytotoxic T cells and NK cells differ in many ways. Unlike T cells, NK cells are innate lymphocytes that specialise in recognising and killing transformed- and virus-infected cells through germline-encoded receptors that yield activating or inhibiting signals when interacting with target cell surface ligands [75,76,77]. Activating receptors include the natural cytotoxicity receptors NKp30 [78], NKp44 [79] and NKp46 [80], NK group 2 D (NKG2D) receptor [81], and killer immunoglobulin-like receptors (KIRs) with short cytoplasmic domains [82,83,84]. With these receptors, educated, cytotoxic CD56^dim NK cells can become activated and kill tumour cells through a variety of mechanisms [85,86]. We have shown that NK cells that selectively express particular KIRs potently kill GBM cells and GBM stem-like cells [84,87,88]. A recent study showed enhanced killing capability of NK cells against GBM stem-like cells when inhibiting αv integrin or TGFβ [89]. Unlike CAR T cells, in the event of antigen loss by the tumour cells, CAR NK cells may still possess the ability to kill by their intrinsic activating receptors [90,91,92,93,94]. In addition, CAR NK cells are considered safer than CAR T cells since they are not known to induce immune effector cell-associated neurotoxicity syndrome (ICANS) or cytokine release syndrome (CRS) [95,96,97].

Importantly, allogeneic T cells are known to induce graft-versus-host disease (GvHD) while NK cells do not [95]. Autologous T cell therapy may not be amenable for all patients since those with significant lymphopenia due to prior anti-cancer treatments or those with a rapidly progressive disease will not be eligible due to insufficient numbers of T cells to expand into therapeutic doses of CAR T cells. This is important for GBM, as both treated and treatment-naïve GBM patients experience lymphopenia [98]. The protracted time required for CAR T cell production is also a bottleneck [97]. The first CAR NK cell approaches focused on primary peripheral blood-derived NK cells transfected with the CAR construct [99], but NK cells are more difficult to transfect than T cells [100]. To circumvent difficulties in introducing gene edits to NK cells, a recent publication has outlined a method for efficient electroporation of CRISPR-associated Cas9 ribonuclear protein (RNP) complexes in primary NK cells [101]. A similar protocol has been used to efficiently knock out the immune checkpoint receptor T-cell immunoglobulin mucin family member 3 (TIM3) in primary NK cells from healthy donors [102], PD-1 [103], and CD38 [104]. However, NK cells only represent approximately 5–15% of circulating lymphocytes in healthy adult humans [105] and require extensive ex vivo expansion. Additionally, they have a much shorter life span than T cells [106], although the persistence of CD-19-directed, IL-15 expressing CAR NK cells from cord blood were shown to survive for at least 4 weeks in mouse models [107]. In humans, multiple or higher doses of engineered NK cells might be required for a sustained treatment effect. NK cells may be obtained from various tissue sources, however, such as umbilical cord blood, induced pluripotent stem cells (iPSCs) or cell lines, such as NK-92 [108]. This characteristic provides the opportunity for “off-the-shelf” cells that can be applied to any patient, which is less readily achieved with autologous T cell therapies [100]. However, the possibility of a host-versus-graft scenario, where the implanted cells are systematically destroyed by the host immune system, must also be considered. Research into the production of reliable, “off-the-shelf” alternatives for cancer immunotherapy is currently in progress for both T and NK cells (reviewed in [109,110]). Strategies to avoid GvHD include knockout of the TCR in engineered T cells [111,112], while host attacks and graft rejection may be mitigated through knockout of MHC or application of lymphodepleting chemotherapeutics prior to engineered T/NK cell administration [113,114,115,116]. In GBM, NK cells have been shown to comprise only a small portion of immune cells in the tumour microenvironment, and to have a CD56⁺ CD16^- phenotype [33], indicating that these cells may not be capable of killing CAR-expressing cells. Another opportunity may lie in NK cells’ inherent expression of KIR2DL4, which can bind HLA G to inhibit host NK cells [117,118].

The heterogeneity of GBM creates a significant challenge for the identification of a ubiquitously expressed tumour specific target for CAR constructs. In addition, GBM cells grown in stem cell media, which most closely preserve the original tumour phenotype, tend to exhibit reduced expression of both classical and non-classical MHC class I ligands for inhibitory NK receptors [33,119]. Therefore, these conditions, if present within patient tumour microenvironments, may provide additional opportunities for receptor mediated cytotoxicity in CAR NK cells [120,121] via both KIR and natural cytotoxicity receptors against stress-induced ligands known to be expressed by GBM tumour cells [87,88,122]. Bispecific TanCAR NK cells could be used to mitigate possible tumour escape by targeting two antigens [57]. As such, the choice of either CAR T or NK cell therapy depends on the strategy utilised, as the two provide different advantages and disadvantages, as summarised in Table 1. Taken together, CAR NK cells represent potentially superior effectors for solid tumours such as GBM.

In GBM, heterogenous mutations are typically not immunogenic [15,132], which renders the tumours resistant to new generation immunotherapies, including CAR T and immune checkpoint blockade (reviewed in [133,134]). Choosing one or more suitable targets is critical for a sustained, beneficial effect. In addition to absent or limited expression in healthy tissues, the target antigen must have persistent expression on the tumour cells and present an appropriate affinity to the CAR which allows recognition of the target with low off-target toxicity [135]. An overview of immunotherapeutic targets for the treatment of GBM currently in clinical trials is shown in Table 2, which includes EGFRvIII, HER2/ErbB2, interleukin-13 receptor alpha 2 (IL13Rα2), B7H3 and NKG2D ligands. Although T cell exhaustion may be limited by targeting CTLA-4 and PD-1, chemotherapeutics with monoclonal antibodies against CTLA-4, PD-1 and PD-L1 have shown mixed and unpredictable response in GBM [136]. In a recent publication, however, Cloughesy et al. performed a randomised, early phase clinical trial using neoadjuvant pembrolizumab, a PD-1 inhibitor, which showed both overall and progression-free survival benefit to patients with recurrent GBM when applied both before and after surgery [137]. These results emphasize the importance of timing of treatment.

Cutting edge gene editing technologies [157] provide the opportunity to enhance the efficacy of existing molecular immunotherapy strategies. Introduction of clustered regularly interspaced short palindromic repeats (CRISPR) and the CRISPR-associated endonuclease Cas (typically Cas9) to design T or NK cells may improve the potency of cancer immunotherapy. CRISPR/Cas9 technology is used to knockout genes by inducing double-stranded DNA breaks (DSB) at specific sites within the genome. The Cas9 enzyme forms a complex with a 20-base guide RNA (gRNA) that is predesigned to recognize a complementary DNA target site in a gene of interest. Cas9 locates a protospacer adjacent motif (PAM) sequence directly downstream of the gRNA sequence, binds and triggers a DSB at the genomic target site [158] (Figure 3). The cell then repairs the damage through mechanisms based on the cell cycle stage and proliferative status [159,160]. The most common repair pathway is non-homologous end joining, wherein the two ends of the DNA are fused, often resulting in faulty repair, with random insertion or deletion of nucleotides causing frameshift mutations and thus a knockout of the gene. The other, less common pathway, homology-directed repair (HDR), uses a DNA repair template to repair the damaged DNA, or introduces a gene edit, as with HDR CRISPR [161]. HDR CRISPR, however, is difficult to accomplish and often leads to high numbers of unwanted insertions and deletions (indels) outside the genetic area of interest, so-called off-target effects [162,163].

The use of CRISPR/Cas9 technology in the development of cancer therapy has only recently begun to have an impact on human trials, as the ethical use of CRISPR/Cas9 to edit the human genome has been a hotly debated topic [164,165]. CRISPR/Cas9 mutagenesis screening has been used as a tool to identify REGNASE-1 as a negative regulator of antitumour responses, whose knockout improved efficiency and persistence of effector T cells in vitro [166]. In addition, CRISPR/Cas9 was used to successfully generate CBLB knockout NK cells derived from human placental stem-cells, without altering their phenotype [167], indicating possibility for use in primary lymphocytes. Preclinical research has successfully used multiplex CRISPR/Cas9 to limit universal CAR T cell exhaustion by disrupting the PD-1 gene, as well as the endogenous TCR α chain (TRAC) and β-2 microglobulin (B2M) [112]. The T cells, directed against EGFRvIII, were administered intracerebrally in GBM bearing mice, and enhanced survival. In a recent phase I human clinical trial, CRISPR/Cas9 was used on autologous T cells to knock out the endogenous TCR α and β chains as well as PD-1 in 3 patients with refractory cancer: 2 patients with multiple myeloma, and 1 with myxoid and round cell liposarcoma [111]. In addition, a transgenic TCR was transduced into the cells, targeting the cancer testis antigen NY-ESO-1. The T cells were then heterogeneously expanded and intravenously reinfused into the patients. The engineered cells persisted in the patients for up to 9 months. Stadtmauer et al. showed that at the time of infusion, the amount of Cas9 protein had diminished to a non-detectable level in the engineered T cells, had acceptable levels of off-target editing, and that the patients did not develop humoral responses to Cas9.