Natural killer (NK) cell-based therapies are a promising modality for the treatment of liquid and solid tumors. With their intrinsic ability to mediate both cytotoxicity as well as cytokine secretion that helps mobilize other immune effector cells, NK cells are a powerful tool in the armamentarium of cancer immunotherapy (1, 2). While the initial recognition of their anti-tumor efficacy came from studies of allogeneic stem cell transplantation, numerous other studies have demonstrated their cytotoxic potential against tumor targets even in non-transplant settings (3, 4). In recent years, great progress has been made in the ex vivo activation and expansion of NK cells for clinical applications, as well as their manipulation with exogenous cytokines and genetic methods to further enhance cancer targeting (5–8). While this progress has led to the translation of several NK cell-based immunotherapies into early phase clinical trials, ongoing studies aim to overcome persistent challenges with this therapy. The purpose of this review is to provide an overview of the strategies being developed to address these challenges, including optimization of NK cell expansion, persistence, trafficking, and anti-tumor activity in the tumor microenvironment (TME).

The major goal for NK cell expansion strategies is to generate large numbers of these cells for clinical applications that retain high anti-tumor activity. With substantial NK cells, it is possible to achieve higher desirable NK effector to target ratios, perform multiple infusions, and potentially allow for the development of off-the-shelf NK cell products. There are multiple NK cell expansion strategies and sources for generating clinical grade NK cell products including peripheral blood (PB), cord blood (CB), bone marrow (BM), and human induced pluripotent stem cells (iPSC) ( Figure 1 ) (9). NK cells can be expanded ex vivo before infusion into patients, or in vivo after adoptive transfer with cytokine stimulation. They are typically isolated by CD3⁺ cell depletion followed by positive selection of CD56⁺ cells, or in some protocols by single step depletion of CD3⁺ and CD19⁺ cells using magnetic beads (CliniMACS) (10). Another approach includes differentiation of functional NK cells from enriched CD34⁺ progenitors present in CB and BM ( Table 1 ) (11, 29).

Although good manufacturing practice guidelines have been defined for NK cell products for clinical applications, the source of NK cells does result in some important differences with respect to NK cell numbers and phenotype ( Table 1 ) (15, 30). It is currently not clear if these differences impact clinical outcomes. For example, while CB has a higher proportion of NK cells than PB, the total NK numbers that can be enriched from CB are significantly lower. Similarly the proportion of less mature CD56^bright cells is higher in CB as compared to PB (11). NK cells can also be generated via differentiation of CB-derived CD34+ cells. This process yields a greater expression of CXCR4 on the differentiated cells that mediates bone marrow homing (31). Killer Ig-like receptor (KIR) expression is lower when CD34+ progenitors are differentiated into NK cells that may impact their tumor-targeting ability (20). Furthermore, while CB-derived NK cells seem to have comparable cytokine secretion in response to the in vitro target cell stimulation, they exhibit lower cytotoxicity than PB-derived NK cells (11).

Differentiation and expansion of NK cells from iPSC is another promising approach that has seen significant advancement for clinical application in recent years (13, 14). While the generation of a sufficiently large number of NK cells from iPSC requires a longer time than PB, it is particularly attractive in terms of establishing well characterized master banks, performing multiple gene edits (including CAR arming and CRISPR) and generating functional immune effector cells in large numbers for off-the-shelf use. iPSC-derived NK cells exhibit comparable cytokine-secreting and cytotoxic capacity to PB-derived sources. They are able to infiltrate tumor spheroids in vitro, interact with other effector T-cells in mouse models and are amenable to gene editing with CRISPR technology (13, 32). A phase 1 clinical trial evaluating the feasibility and efficacy of iPSC-derived NK cells is currently ongoing (NCT03841110).

The efficacy of NK cell therapy depends in part on the ability of the infused cells to persist in vivo after adoptive transfer (54). While NK cells could be expanded in large numbers ex vivo, following infusion these NK cells exhibit limited in vivo persistence and expansion (55). Efforts to expand and activate NK cells after their adoptive transfer have typically relied on the cytokines IL-2 or IL-15, the latter being somewhat favored because of its greater selectivity for NK cells (56). A super-agonist complex (ALT-803) of an IL-15 mutein bound to the sushi domain of the IL-15 receptor alpha (IL-15Rα) and fused to the immunoglobulin G1 Fc region has been generated that in addition to having a longer half-life and improved functionality over the individual cytokine (57), resulted in marked expansion and activation of NK cells in a phase I clinical trial (58).

The expansion of NK cells using feeder cell lines such as K562, Epstein-Barr lymphoblastoid cell line, and human B-lymphoblastoid cell line 721.221 that either express membrane-bound cytokines or in co-culture with exogenously administered cytokines such as IL-21 are methods developed for the generation of activated NK cells with a potential for long-term persistence in vivo (39, 41, 43–45, 59). In addition, genetic manipulation of NK cells via viral and non-viral techniques for autocrine expression of cytokines (IL-15 or IL-2) has been shown to increase the cells’ persistence and expansion post infusion (60).

Differentiation into a memory-like phenotype is another novel approach that results in normally short-lived NK cells persisting for much longer periods with significantly increased in vivo expansion and anti-tumor responses (6, 21, 61–63). Cytokine induced memory-like (CIML) NK cells generated after brief activation with cytokines IL-12, IL-15, and IL-18 exhibit this unique memory-like phenotype ( Figure 1 ). In preclinical and recent clinical settings, CIML NK cells have demonstrated massive expansion, prolonged persistence, and reduced threshold for activation in response to tumor targets (64).

Ongoing clinical studies have shown that the prolonged persistence of the adoptively transferred CIML NK cells is associated with the attainment of clinical responses (6, 21, 61–63). Upon their adoptive transfer into an immune compatible environment in the context of an allogeneic HCT in a phase 1 clinical trial, the infusion of CIML NK cells resulted in a massive expansion of the NK cell compartment that persisted for up to 6 months (26). The use of the CIML NK platform for the prevention of post-allogeneic HCT relapse of myeloid disease is concurrently being explored (65).

In contrast to hematological cancers, solid tumors are characterized by a dense, highly heterogeneous, stroma-rich TME within which crosstalk between the cancer cells and their microenvironment helps drive tumor progression, invasion, and metastasis (66, 67). The TME is composed of cellular and acellular components which often hinder the delivery of small molecules and infiltration of infused adoptive cells (68, 69). NK cell infiltration in the tumor is hindered by lack of chemotactic signaling and activation, as well as by the physical barriers in the TME due to hypo-vasculature and dense extracellular matrix (ECM) deposition (69). In addition, myeloid-derived suppressor cells (MDSCs), tumor associated macrophages (TAMs) and Tregs are often enriched within the advanced tumor sites, and play a key role in inhibiting immune effector cells such as NK and CD8⁺ T cells (66). The tumor-stromal interaction results in the release of immunosuppressive molecules including IL-10 and TGFβ, the downregulation of antigens targeted by immune effector cells and the upregulation of checkpoint blockade proteins. All these molecules shape the TME to be highly immune-suppressive, thereby hampering effective antitumor immune responses (66). To mediate optimal anti-tumor responses in the solid tumor setting, circulating NK cells need to exit the bloodstream, traffic to tumor sites that include distant metastatic sites, penetrate through the dense ECM, and finally interact with and respond to the tumor cells. Previous reports have shown that only a minority of adoptively-transferred lymphocytes traffic to the tumor sites (70, 71).Various strategies have been investigated to potentially overcome this major barrier, including utilizing chemokine networks for efficient NK cell infiltration into the TME, imparting features to NK cells to overcome the immunosuppressive TME, as well as altering the TME to create a favorable niche for the adoptively transferred immune cells (72).

Early phase NK cell clinical trials have shown favorable safety profiles and promising responses using advances in NK cell engineering, including the differentiation into memory-like phenotype or arming with CARs. However, these studies have a common feature of an initial reduction of tumor burden followed by eventual reappearance of detectable disease in the majority of treated patients (21, 60, 93). These observations can be partially attributed to the tumor escape mechanism and exhaustion of adoptive NK cells (72).

The transformation of normal cells to malignant cells requires elaborate adaptations and oncogenic changes that allow a tumor to bypass detection and elimination by immune effector cells (94). Some of these adaptations involve downregulation or shedding of ligands that activate NK cells (e.g. MICA/B), release of immune-modulatory molecules from tumor cells or supporting stromal cells (e.g. TGFβ), and epigenetic changes (e.g. histone deacetylases, HDACs) (95, 96). Escape of tumors from NK cells must be overcome for NK cell-mediated therapies to become standalone treatments without the need for subsequent consolidation with other therapies. Current strategies to overcome tumor escape include but are not limited to the reversal of NK cell anergy, the combination of NK cell therapy with antibody-mediated therapy that can engage their ADCC function, and the development of off-the-shelf NK cells that may be transduced with CARs that can target escaping tumor cells ( Figure 2 ) (72).

A lack of the KIR-HLA class I interaction (KIR-ligand mismatch) is associated with enhanced NK activity, and one strategy used by tumors to escape NK cell mediated effects involves the increased expression of HLA class I molecules on their surface (97). To overcome this, anti-KIR antibodies such as IPH2101 (previously known as 1-7F9) that bind the inhibitory KIR2DL1, -L2 and -L3 as well as the activating KIR2DS1 and -S2, and interefere with the receptors’ interaction with HLA-C have been developed (98). Lirilumab, a fully human IgG4 anti-KIR2D monoclonal antibody that recognizes the same epitope as IPH2101 and in a similar manner blocks the KIR interaction with HLA-C, has been well-tolerated in early phase clinical trials (99, 100) and has shown some clinical activity when combined with azacytidine in high risk myelodysplastic syndrome (MDS) (101). Similarly, the anti-NKG2A antibody monalizumab has been shown to increase the effector function of NK cells, and its combination with PD-1/PD-L1 blockade yielded evidence of clinical activity in early phase trials (102, 103). An alternative approach to blocking antibodies that targets the KIR-HLA class I interaction is to genetically engineer the NK cells to downregulate inhibitory signals including NKG2A and KIRs (104–107).

The antitumor response of NK cells can also be manipulated using monoclonal antibodies engaging activating receptors like CD16A, NKG2D, and NKp46 (108–110). Antibody dependent cellular cytotoxicity (ADCC) of the target cells coated with IgG molecules is mediated by NK cells through engagement of CD16 (Fc receptor) on NK cells (111). Based on a similar concept, novel NK cell engagers (NKCE), synthetic molecules composed of the fragments of monoclonal antibodies that simultaneously engage NK activating receptors and tumor antigens, have been designed (112, 113). Bispecific killer cell engages (BiKEs) containing ScFv fragments that target CD16 and a tumor antigen are one example of this, and have been used to target EpCAM on carcinomas (114), CD133 on cancer stem cells (115), and CD30 on lymphomas (116). AFM13 (CD16/CD30) is the first-in-class tetravalent bispecific molecule targeting CD30-positive lymphoma, and has shown promising antitumor responses in recent clinical trials when used as a monotherapy as well as in combination with allogenic NK cells (116, 117). This molecule belongs to the class of redirected optimized cell killing (ROCK) platform, which is equipped with high affinity towards specific epitopes on CD16, is independent of individuals’ CD16A allotype, not inhibited by serum IgG, and also prevents NK fratricide (118).

Similarly, trispecific killer engagers (TriKE) have been developed that consist of two scFv fragments connected by an IL-15 domain, enhancing NK cell-mediated anti-tumor efficacy by engaging CD16 on the NK cells and an antigen on the tumor cells. The GTB-3550 TriKE (CD16/IL-15/CD33), for example, has been shown to enhance NK cell proliferation and function, and is being evaluated in clinical trials (119). Other TriKE molecules targeting CD133 and B7-H3 have also shown promising efficacy in cell lines and mouse models (120, 121). The targeting of CD16 on NK cells by BiKEs and TriKEs may be limited by the known process of CD16 shedding (122). However, the combination of the NKCE with novel approaches that interfere with the shedding process are being investigated (113).

While engagement of CD16 has been one successful strategy of NKCE, other engagers have been designed to bridge the NKG2D activating receptor with tumor targets such as CS1 on multiple myeloma (MM) and HER2 on breast cancer cells (123, 124). Alternative NKCE designs, such as those bridging NKp30 and NKp46 with tumor antigens for example, have been reviewed recently (112). Trifunctional NKCE that co-engage two different NK activating receptors, such as CD16 and NKp46, with a tumor target antigen have also been developed, and have shown impressive in vitro activity and suppression of tumor growth in mouse models (110).

A plethora of immune-suppressive pathways in the TME combine to inhibit NK cell function. Exhausted or impaired NK cells in the TME exhibit downregulation of effector cytokines, decreased granulation, downregulation of NKG2D, a decrease in transcription factors like Eomesodermin and T-bet (125), and upregulation of inhibitory receptors such as PD-1, TIM-3, TIGIT and NKG2A (72). Expression of the latter markers on NK cells have been correlated with decreased NK cell functionality and importantly, various studies demonstrate that blockade of these receptors can increase NK cell cytotoxicity and function (72, 126).

TGF-β is one of the important cytokines that plays a critical role in the immune-suppressive TME. It limits NK cell function by inhibiting the T-bet transcription factor (SMAD3) and by targeting the mTOR pathway (127). In addition, TGF-β has been reported to contribute to the selective recruitment of non-cytotoxic CD56^bright NK cells by modulating the chemokine repertoire in tumors (127). To potentially overcome TGF-β mediated immune suppression, Otegbeye et al. combined ex vivo expanded NK cells with the TGF-β inhibitor LY2157299, resulting in a significant decrease in liver metastasis as observed in a colon cancer model (128). Another strategy by Burga et al. was to engineer a variant of the TGF-β receptor where the receptor domain was coupled with NK specific activating domains (129). The authors showed that these engineered NK cells could efficiently convert inhibitory TGF-β signals into activating signals leading to augmented cytotoxic function in a neuroblastoma model (129).

Impaired cellular metabolism in the nutrient-poor TME is another important factor responsible for the dysregulation of NK cells (130). Glycolysis is a critical metabolic process for NK cells, especially for ex vivo expanded NK cells. A study by Cong et al. showed that aberrant expression of fructose bisphosphatase 1 (FBP1) in intratumoral NK cells impaired their cytotoxic function and could be a potential target to improve NK-based therapy (131). Recently, cytokine-inducible SH2-containing protein (CISH)-knockout in iPSC-derived NK cells using the CRISPR/cas9 system exhibited improved metabolic fitness as exemplified by increased basal glycolysis and mitochondrial activity, and by increased mTOR signaling which contributed to the enhanced NK cell function (132). Overall, metabolic reprograming of the TME or adoptive NK cells is a promising strategy to prevent NK cell dysfunction.

NK cells have emerged as an attractive platform for adoptive cell immunotherapy in recent years. Key advances, including the development of efficient ex vivo expansion systems, the use of CIML NK cells for prolonged in vivo persistence and genetic manipulations strategies involving CARs have helped to catapult the field toward more effective clinical translation. However, more work needs to be done to further enhance adoptively transferred NK cells’ tumor targeting, overcome their immune suppression and exhaustion in the TME, increase their persistence in an allogeneic setting, and augment their capacity for long-term surveillance against tumor relapse. Taking advantage of recent advances in our understanding of basic NK cell biology and genetic manipulation techniques, we expect a bright future for the NK cell immunotherapy field.

MT, RR, and RS conceived of the topics to include in the review. MT and RS wrote the manuscript. MT designed the figures. RR provided critical edits to the manuscript. All authors approved of the final version.

Funding support was provided by the Michelle D. and Douglas W. Bell Fund for Stem Cell Transplant (grant# 9618342) and the Ted and Eileen Pasquarello Research Fund (grant #9616306).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.