Chimeric antigen receptor-engineered natural killer cells for cancer immunotherapy

A CAR is a functional hybrid antigen receptor usually combining variable regions of an antibody with the constant region of a T cell receptor (TCR), grafting an immune cell with the specificity for TAA expressed on tumor cells [26]. A typical CAR is composed of a signal peptide, single-chain antibody variable fragment (scFv), and transmembrane and intracellular signaling domains. A linker combines the variable regions of heavy and light chains of an antibody molecule. A hinge, also called spacer, connects the scFv to the transmembrane domain. scFv is derived from a tumor-specific antibody and can target any antigen expressed on tumor cell surface [27]. In some CARs, a ligand of a tumor antigen is used [28, 29]. An intracellular signaling domain is derived from immunoreceptor tyrosine-based activation motifs (ITAMs) of TCR or a cytoplasmic domain of other activating receptors. Phosphorylation of ITAMs recruits adapter molecules, stimulates downstream pathways, and activates immune cells [30, 73].

The first-generation CARs contained only the CD3ζ chain, the common signal transducing subunit of TCR, or FcRγ subunits as the intracellular signaling domain [31‐33]. They were not effective in initiating T cell activation or tumor elimination as they lacked co-stimulatory signaling required for full T cell activation. To improve the immune cell response to tumor cells, second-generation CARs were created by adding one of the T cell co-stimulatory molecules such as CD28, 4-1BB (CD137), ICOS, or OX40 (CD134) in addition to CD3ζ. The third-generation CARs include three signaling domains (for example, CD3ζ and two co-stimulatory domains) [34, 35]. There is no consistent evidence suggesting increased performance in the third—compared to the second—generation CARs. The type of co-stimulatory domain used can dramatically influence effector functions mediated by the CAR. For example, 4-1BB co-activation results in development of T memory cell characteristics, whereas CD28 co-activation results in enhanced T cell activation and expansion [36].

CARs can be generated against any of the surface antigens expressed on tumor cells. However, the target should have a specific antigen density to avoid "on-target, off-tumor" effects. Studies in T cells have shown that target density between 200 and 2000 copies per cell is optimal to avoid damage to the normal cells [37]. Generation of a CAR usually starts with an scFv derived from an antibody, composed of heavy and light chains [38]. Recently, nanobodies composed only of heavy chains are being utilized for CAR NK cell production [39]. These single-chain antibodies not only possess specificity comparable to scFvs but also are much smaller in size, facilitating transduction. However, their use in CAR NK cell production is currently at the experimental stage.

Usually, a signal peptide is introduced in the front of an scFv. A linker that connects heavy and light chains usually contains the (G₄S)₃ or (G₄S)₅ signature that provides flexibility for accurate folding of connected protein domains [40, 41]. A spacer, also referred to as a hinge, usually derived from IgG, CD8α, or CD28, plays a critical role in specificity of the CAR-bearing immune cells [41]. Adjusting the length of spacer may help optimize the distance between CAR-bearing cells and target cells for an optimal immunologic synapse (IS) formation and specificity. Use of CAR T cells created using spacers with suboptimal length results in off-target effects. The downstream CAR component(s) are co-stimulatory domain(s) that can be shared between CAR NK and CAR T cells [42, 43], followed by CD3ζ. With the rapid development of DNA synthesis technology, many CAR sequences are now synthesized by a vendor if the DNA sequences encoding the scFv are known. All CAR components or a cassette are commonly cloned into a retrovirus- or lentivirus-based expression vector, which is then used to transduce primary NK cells or NK cell lines [44] (Fig. 1). In some instances, an inducible iCaspase can be incorporated as a safety switch for use in case of adverse reactions [10]. Different CAR designs and detailed engineering approaches have been reviewed elsewhere [45].

CAR T cells targeting CD19 or other B cell markers such as CD20 have achieved remarkable responses in patients with B cell malignancies including chronic lymphocytic leukemia (CLL), acute lymphoblastic leukemia (ALL), and lymphoma [26, 46‐48]. CTL019 (Kymriah, tisagenlecleucel) marketed by Novartis achieved 82.5% overall remission rate (ORR), which includes 40 complete remission (CR) and 12 CR with incomplete hematologic recovery (CRi) in 63 patients with ALL. A biologic license application for Kymriah was assessed based on B2202, a single-arm, open label, multi-center phase 2 study including 107 patients who signed consent. There were 19 patients who were not eligible due to death before the study began, unacceptable apheresis, incomplete screening before the cutoff date, or loss of eligibility for other reasons [49]. Additional twenty patients from the remaining 88 were further excluded due to adverse events, manufacturing failure, death, or pending infusion. The remaining 68 patients were included in the safety set, but five were further excluded from the efficacy set because the products were manufactured in Germany, and thus, 63 patients were included in the ORR calculation above. The ORR of 82.5% was an impressive outcome compared to less than 20% of the standard treatment, and thus, the product received an expedited approval by the US Food and Drug Administration (FDA) [36, 49, 50]. The intent-to-treatment (ITT) rate, as defined by the number of patients with ORR divided by the number of patients consented, was 48.5% (52/107). In the Yescarta study, 111 patients were enrolled, but 10 withdrew [51]. Objective response rate was 72% (51% for advanced cases). In the treatment of DLBCL, CR rates were higher in the Yescarta study compared to the Kymriah study (72% vs. 50%). It is unknown whether this difference was influenced by differences in the number of cells infused (60–600 million cells were infused if the body weight was greater than 50 kg in the Kymriah study and 2 million cells per kg were infused in the Yescarta study).

CD19-CAR T cells are the first CAR T cell-based drug approved by the FDA. Studies to develop CAR T cells for treating other tumor types including solid tumors are increasing [52‐55]. However, there have been reports of serious side effects in the clinic associated with CAR T cell therapy including cytokine release syndrome (CRS), “on-target, off-tumor” toxicity, neurologic symptoms, graft vs. host disease (GvHD) if used in an allogeneic setting, and tumor lysis syndrome. For example, 94% of the patients in the Yescarta study and 74% in the Kymriah study had CRS and at least four two patients in each study died from CRS [51]. CRS is a rapid and massive release of inflammatory cytokines such as IFN-γ, TNF-α, and IL-6 into circulation [47]. It can present with symptoms such as high fever, malaise, fatigue, myalgia, nausea, anorexia, tachycardia/hypotension, capillary leak, or disseminated intravascular coagulation (reviewed by Bonifant et al. (2016) [56, 57]. Other studies have also shown that CRS by CAR T cells can be life-threatening [56‐58]. In addition, CAR T cells may attack not only tumor but also normal cells, resulting in on-target, off-tumor toxicity. For example, HER2-specific CAR T treatment resulted in acute respiratory failure and death in a patient with metastatic colon cancer possibly due to low levels of HER2/neu expressed in the pulmonary parenchyma or vasculature [59]. Severe neurological toxicities such as acute cerebral edema resulting in brain death a few hours after initial appearance of neurological symptoms have also been reported in treatment with CD19-CAR T cells [60]. In the separate Kymriah and Yescarta studies, although, respectively, noted in 72% and 87% of the patients treated with CAR T cells, no deaths were attributed to neurotoxicity. CAR T cells obtained from allogeneic sources may also result in GvHD [61]. Disabling the endogenous TCR has the potential to avoid or lower GvHD risk. Endogenous TCR has been disabled using recent techniques such as CRISPR/Cas9, TALEN, zinc finger endonucleases, and receptor blockage using an array of proteins [61]. Some of these techniques are currently being evaluated for clinical use.

Compared to T cells, NK cells offer many advantages for cancer immunotherapy (Table 1). For example, it has been reported that allogeneic NK cells do not need complete HLA matching and cause little or no GvHD compared to allogeneic T cells [61‐63]. In fact, there is experimental data in animals showing that NK cells not only do not cause GvHD but also prevent it in some cases by inhibiting activated alloreactive T cells while retaining graft vs. tumor effects [63]. On the contrary, CAR T cells can cause GvHD even when HLA-matched as minor mismatches may still be provocative [64]. Additionally, the cytokine levels produced by NK cell infusions are moderate [65] compared to the levels seen in life-threatening cases of CRS observed in CAR T cell trials [61].

NK cells have a limited life span in circulation and do not produce memory cells except for some viruses. Therefore, a primary-NK cell-based CAR approach may not need to be modified to include a suicide gene as a safety switch [66]. A further potential advantage of using CAR NK cells is that, unlike CAR T cells, they can be derived from umbilical cord blood (UCB) or cell lines such as NK-92, both of which have the potential for use as an “off-the-shelf” therapeutic product [61].

The safety of CAR NK-92 cells was tested in acute myeloid leukemia (AML) patients in a recent “first-in-man” clinical trial targeting CD33 [67]. Although the CAR NK-92 cells did not seem significantly beneficial in that particular study, injection of approximately five billion irradiated cells per patient did not produce any significant adverse side effects, suggesting that injection with a large number of NK-92 cells may be safe. A recent clinical trial (NCT03056339) produced exciting evidence supporting the potential for production of “off-the-shelf” NK cells with significant benefits in relapsed or refractory CD19-positive lymphoma and leukemia [68]. HLA-mismatched NK cells isolated from UCB and transfected with a retrovirus-based vector expressing anti-CD19 CAR, iCaspase-9, and IL-15 were administered in 11 patients with CLL or non-Hodgkin’s lymphoma. Nine patients received haplo-mismatched CAR NK cells and 2 patients received completely mismatched CAR NK cells.  Seven patients had a complete response and one had a partial response with remission of the Richter’s transformation component. These patients had relapsed/refractory tumors and were heavily pre-treated using conventional methods before starting CAR NK cell therapy. The patients had a median of four lines of therapies before beginning the CAR NK cell treatment. Five were CLL patients who failed to respond to ibrutinib and other therapies. Six were lymphoma patients, four of whom failed to respond to autologous hematopoietic stem cell transplantation (HSCT), and two had refractory disease. All but one patient had at least two prior relapses. An objective response of eight patients (73%) was observed during a follow-up period of 13.8 months. In all eight patients, a response was seen in the first month of the CAR NK treatment. An additional patient had a complete response, but his response was not attributed to CAR NK cell therapy. It was possible to bridge five of the eight patients who responded CAR NK therapy to HSCT, rituximab, venetoclax or lenalidomide treatment after CAR NK-cell-induced remission. None of the side effects seen with CAR T cells were noted. Excluding transient myelotoxicity, there was no neurotoxicity, elevation in IL-6 levels, CRS, or GvHD. The infused CAR NK cells expanded and persisted in vivo for at least 12 months [68].

NK cell lines as well as primary NK cells from peripheral blood (PB), UCB, or induced pluripotent stem cells (iPSCs) have been used for CAR NK cell production. Several NK cell lines such as NK-92, NKG, YT, NK-YS, HANK-1, YTS, and NKL are commercially available [69]. NK-92 is the most widely used cell line and has already entered into clinical trials [70]. In comparison to PB NK cells, NK-92 cells are easier to expand and transduce and do not carry any contaminating T cells that may cause GvHD [71]. In addition, their production is inexpensive and easy to standardize, and they do not express fewer inhibitory KIRs that restrict NK effector functions [72, 73]. However, there are also several disadvantages associated with the use of NK-92 cells. For example, the cells must be irradiated before infusion into patients to prevent unrestricted cell growth, thereby limiting their capacity for in vivo expansion. Additionally, the NK-92 cell line carries Epstein–Barr virus and has an abnormal genome.

NK cells constitute only about 10% of the PB lymphocytes and thus need expansion in vitro to obtain a sufficient quantity for a therapeutic effect. Various methods to enhance expansion of NK cells isolated from PB have been detailed below in the "Challenges in expanding CAR NK cells" section. NK cells from PB, and consequently CAR NK cells produced from PB, are potent killers of tumor cells as they are fully matured [74, 75]. Another well-studied source of primary NK cells is UCB, which may offer several advantages. Approximately 15 to 30% of UCB lymphocytes are NK cells, which can result in more than NK cells found in PB [76, 77]. There are less contaminating T cells in UCB compared to PB, which reduces the risk of GvHD [78‐80]. In addition, NK cells in UCB may be less likely to undergo rejection by the patient compared to PB NK cells. However, NK cells isolated from UCB are immunologically immature and initiate only a weak immune response against target cells after being activated by stimuli such as cytokines or alloantigens [81]. Therefore, CAR NK cells derived from UCB may be less cytotoxic than those derived from PB [82]. UCB is easy to collect with no harm to the mother or child and almost never contaminated with Epstein–Barr virus or cytomegalovirus.

A recent study showed that CD19-CAR NK cells isolated from UCB transduced to express IL-15 and an iCaspase-9-based suicide gene had effective and persistent anti-tumor activity in NOD scid gamma (NSG) mice [83]. The incorporated IL-15 dramatically increased anti-tumor activity in the CD19-CAR NK cells. This study has been translated into a clinical trial with impressive outcomes [68]. Other less common sources for obtaining NK cells include pluripotent stem cells (human embryonic stem cells or iPSCs or adult stem cells (CD34+ hematopoietic stem cells) [84‐88]. Recently, Li et al. (2018) utilized NK cells derived from human iPSC with a mesothelin-CAR modification as a potential source for "off-the-shelf" CAR NK cells for anti-cancer immunotherapy [89].

The choice of targets for CARs depends on the type of tumor studied. Several proteins expressed by tumor cells are being targeted in preclinical or clinical studies. In lymphoid malignancies, the most widely targeted cell surface proteins include CD19 and CD20 [90‐97]. CAR NK cells for multiple myeloma (MM) have targeted CD138 and CS1 [44, 98]. CS1 is a cell surface glycoprotein highly and ubiquitously expressed on the surface of myeloma cells [99]. In our previous studies, we found that CS1-CAR NK cells were able to target MM cells and eradicate MM both in vitro and in vivo [44]. MM cancer stem cell-like cells (CSCs) expressed significantly higher levels of CS1 than any other cell type, making them especially vulnerable to being eradicated by CS1-CAR NK cells [100]. Thus, CS1-CAR NK cells may prevent MM relapse via targeting and eliminating MM CSCs. CD22, important in B cell trafficking, can be targeted for treatment of B-cell lymphoma [101]. CD7, important in lymphoid development and B/T cell cross talk, can be targeted in ALL, whereas CD33, a lectin inhibiting multiple cellular functions, can be targeted in AML [102].

As for solid tumors, a set of targets have been evaluated. A well-known target is HER2 that is up-regulated in several cancers including breast, colon, and ovarian cancers [103]. HER2-CAR cells using PB NK cells or NK-92 cells have been used to treat breast and ovarian cancers and neuroblastoma [104‐108]. GD2 has been targeted for treating neuroblastoma using CAR NK-92 cells [109, 110]. GD2-CAR NK-92 cells effectively lyse multi-drug resistant neuroblastoma cell lines and show significant anti-tumor activity in xenograft mouse models [110]. Additional antigens being explored as CAR NK cell targets in solid tumors include EGFR, GPA7, CD244, and EpCAM [44, 111‐114].

EGFR is one of the most frequently amplified genes in glioblastoma, the most frequent and lethal type of brain tumor [115]. Approximately 20–40% of the EGFR-amplified tumors harbor the EGFR variant III (EGFRvIII). In our previous work, we found that EGFR-CAR NK cells targeted glioblastoma with both wild-type EGFR and EGFRvIII. EGFR-CAR NK cells were strongly cytotoxic, inhibited glioblastoma growth, and prolonged survival in tumor-bearing mice [112]. Use of CAR NK cells in combination with oncolytic viruses has been tested to eliminate solid tumors. Combination of EGFR-CAR NK-92 cells with herpes simplex virus 1-based oncolytic virus (oHSV-1) results in increased efficiency in killing a breast cancer cell line engrafted into the brain [116].

Additional targets being evaluated in recent clinical trials include a prostate-specific membrane antigen (PSMA) expressed over 100 times higher in prostate tumor tissues than normal prostate and the tumor differentiation antigen mesothelin in ovarian cancer [117, 118]. CAR NK cells and CAR NKT cells have been reported to target MUC1, a glycoprotein that creates a significant barrier preventing drugs from binding to their targets on cancer cells [119]. ROBO1, important in neuronal development, is being evaluated for treatment of pancreatic cancer [120].

Although CAR NK cells and CAR T cells have opened a new era in immunotherapy, results from clinical studies suggest that improvement is still needed. Key areas that need further attention for improvement include modifying CAR structure for better efficiency and/or suitability to the TME, over-expressing activating and/or blocking inhibitory receptors on NK cells, improving cytotoxic mechanisms involving death receptors or granule-dependent pathways, and/or use of multiple compounds to increase CAR efficacy. Efforts to optimize CAR NK cell efficacy are extensive but can be categorized into the groups below:

Currently, there are 19 CAR NK clinical trials available on ClinicalTrials.Gov website (Table 4). Eleven and eight trials have targeted liquid and solid tumors, respectively. The most common targets (with number of clinical trials in parentheses) are CD19 (7), ROBO1 (3), and CD22 (2). One clinical trial has been completed, eight have been recruiting patients, one is suspended, and one is withdrawn. Five trials have not begun recruiting patients, and the status of the remaining three trials is unknown. Five trials are at early phase 1, four at phase 1, and ten are at phase 1 or 2 stages. Fourteen of the trials originated from China, three from the USA, one from Singapore, and one from Germany.

In one of those studies, a phase 1/2 trial (NCT02944162) targeting CD33 in AML patients had three patients who were administered nearly five billion irradiated CD33-CAR NK-92 cells. Although therapeutic response was limited, the patients did not show any significant adverse reactions after administration of up to five billion NK-92 cells, suggesting that injection of a large number of NK-92 cells may be safe [67]. A recent promising phase 1/2 clinical trial in the USA (NCT03056339) is being conducted to investigate the safety and efficacy of primary NK cells obtained from UCB carrying a CD19-CD28-CD3ζ-2A-iCasp9-IL15 construct in patients with relapsed/refractory B-cell ALL, CLL, and non-Hodgkin lymphoma. The CAR NK treatment was carried out in conjunction with lymphodepleting chemotherapy. Very encouraging results from this trial have recently been published [68]. As summarized above, eight of the 11 patients responded to the therapy. Importantly, NK cells persisted in patients for over 12 months possibly due to the incorporation of IL-15 in the CAR vector. A clinical trial being conducted in Germany has been investigating the safety and tolerability of NK-92 cells carrying anti-HER2 CAR in glioblastoma patients (NCT03383978).