Natural killer cells in cancer biology and therapy

Due to variable expression of surface markers on NK cells, it is hard to use one or two simple molecules or traditional immunohistochemistry to accurately identify this cell type and more importantly, their functional status. However, in humans, in both clinical and research settings, CD3⁻CD56⁺ cells are commonly identified as NK cells and can be further divided into the CD56^bright and CD56^dim subgroups. CD56 is not only a marker but also plays an important role in the terminal differentiation of NK cells since its blockade by monoclonal antibodies obviously inhibits the transition from CD56^bright to CD56^dim, thus limiting the cytotoxic ability [23]. Consistently, CD3⁻NK1.1⁺ and CD3⁻CD49b⁺ cells are defined as NK cells in mice. In recent studies, the notion that natural cytotoxicity receptor 46 (NKp46), belonging to natural cytotoxicity receptors (NCRs), should also be included in this panel has been proposed based on the consensus of adding more functional proteins rather than surface molecules into the classification system of NK cells [24, 25].

As the main effector cell type in innate immunity, NK cells are capable of killing tumor cells and virus-infected cells at a very early stage. Due to the lack of abundant production of receptors for distinguishing incalculable antigens in the body specifically, they rely on the “missing self” and “induced self” modes to identify target cells by maintaining a precise balance between activating co-stimulatory and inhibitory signals (mainly by functional receptors). Those interacting signals finally decide the activation and functional status of NK cells.

Activating signals include cytokine-binding receptors, integrins, killing-receptors (CD16, NKp40, NKp30 and NKp44), receptors recognizing non-self-antigens (Ly49H) and other receptors (e.g., NKp80, SLAMs, CD18, CD2 and TLR3/9) [26, 27]. In total, the activating receptors of NK cells can be divided into at least three types according to the respective ligands, including MHC-I specific, MHC-I related and MHC-I non-related receptors (Table 1) [13, 28‐43]. To emphasize, NCRs, which belong to the third group, include three molecules (NKp30, NKp44 and NKp46), and NKp30 was shown to be capable of recognizing B7-H6 expressed on tumor cells, and could be used as a novel treatment option in the future [35].

Inhibitory signals mainly comprise receptors recognizing MHC-I, such as Ly49s, NKG2A and LLT1, as well as some MHC-I non-related receptors (Table 1) [20, 44‐60]. Moreover, MHC-I specific inhibitory receptors can be generally divided into three types according to structure and function: killer cell immunoglobulin-like receptors (KIRs), killer lectin-like receptors (KLRs) and leukocyte immunoglobulin-like receptors (LILRs).

Conventional NK (cNK) cells are mainly found in peripheral blood and migrate to a specific location to exert their effects. NK cells also include tissue-resident NK (trNK) cells. The complex process of NK-cell differentiation occurs in several distinct tissues, including bone marrow, liver, thymus, spleen and lymph nodes, and may involve cell circulation at different stages of maturation among these tissues [61]. In bone marrow, blood, spleen and lungs, NK cells are fully differentiated, while that in lymph nodes and intestines are immature and precursory [62]. Single-cell transcriptome ananlysis of bone marrow and blood NK cells helps to illustrate the changes of their characteristics during development. For example, high expreesion of TIM-3, CX3CR1 and ZEB2 represents a more mature status [63]. Many hypotheses have been proposed to describe the motivation of their migration and different biological behaviors of identically originated NK cells in different tissues. The first question could be partly explained by the multi-direction differentiation induced by heterogeneous microenvironments in different tissues, or more straightforward, different phenotypes originated from similar chemokine-recruited peripheral cNK cells.

To conclude, NK cells in various tissues have diverse features, possessing different functions and forming a close relationship with other stromal cells (Fig. 1). In the lung, trNK cells show a different phenotype from that of circulating NK cells (mainly CD56^dim) and are considered to express different levels of CD16, CD49a and CD69, with CD56^dimCD16⁺ cells representing the majority of the whole NK family there [64, 65]. Of note, CD69⁺ cells are the main type of CD56^brightCD16⁻ NK cells. However, in the thymus, most NK cells are CD56^highCD16⁻CD127⁺, highly relying on GATA3 compared with the CD56⁺CD16⁺ subgroup [66]. Besides, they produce more effector molecules, including TNF-α and IFN-γ [66, 67]. Similar to the phenotypic features in humans, skin NK cells in the mouse can be generally divided into two types: CD49a⁺DX5⁻ and CD49a⁻DX5⁺ [68, 69]. Similarly, hepatic trNK cells can be classified into two groups, including CD56^brightCD16^+/− and CD56^dimCD16⁺, both lacking CD3 and CD19 [8]. In addition, CD49a⁺CD56⁺CD3⁻CD19⁻ NK cells have been identified in liver biopsies [70]. Besides, hepatic NK cells can develop memory for structurally diverse antigens, dependent on the surface molecule CXCR6 [71]. In the uterus, most NK cells are CD56^brightCD16⁻, expressing high levels of KIRs [72]. Decidual NK cells are also CD49a⁺. For skin NK cells, it is intriguing that only few CD56⁺CD16⁺ can be detected, which are common in peripheral blood [73]. Interestingly, trNK cells are distinct between subcutaneous (CD56^dim) and visceral (CD56^bright) adipose tissues, and can be generally divided into three groups according to CD49b and Eomes, showing different expression levels of CD49a (CD49b⁺Eomes⁻ subgroup) and CD69 (CD49b⁻Eomes⁺ subgroup) [74, 75].

Besides different tissue types, NK cells are also highly heterogeneous even in the same organ and the same tissue [61]. Through high-dimensional single-cell RNA-seq, Crinier et al. revealed the heterogeneity of human and mouse NK cells in spleen and blood and identified several subpopulations of NK cells, respectively [76]. As mentioned above, NK cells are considered a subgroup of ILC1s [6]. Although ILC1s are not detectable in many tissues, intra-epithelial ILC1 (ieILC1)-like cells, which highly express IFN-γ, integrins and other cytotoxic molecules similar to the ieILC1s previously described by Fuchs except for different NKp44 expression, could represent the majority of NK cells in the mucosal tissue [61, 77]. Due to their unique features, this cell type represents a subgroup of NK cells other than conventional ILCs. Unlike other trNK cells, DX5⁻CD11c^hi liver-resident NK cells participate in autoimmune cholangitis, negatively regulating immune responses, especially by inhibiting the proliferation and function of CD4⁺ T cells in vivo, which was validated by severer biliary disease in NK-depleted mice resulting from Nfil3 knockdown or treatment with neutralizing antibodies [78].

According to surface CD56 expression, NK cells can be divided into CD56^bright and CD56^dim. CD56^dim NK cells are mainly found in peripheral blood, and are always also CD16-positive, expressing high levels of KIR and LFA-1 and showing cell killing ability. CD16 is a key receptor mediating antibody-dependent cell cytotoxicity (ADCC), inducing the phosphorylation of immunoreceptor tyrosine-based activation motif (ITAM) [40, 79, 80]. According to a time-resolved single-cell assay, the cytotoxicity of NK cells is repressed through both necrosis and apoptosis. As a result, FasL/FasR interaction, perforin/granzyme release and Ca2⁺ influx are all important for NK cell function [81]. However, CD56^bright NK cells are similar to helper cells, which mainly secrete cytokines such as IFN-γ, TNF-β and GM-CSF [23]. Researchers even further subgroup these cells into the NK1 and NK2 categories, consistent with Th1 and Th2, mainly secreting IFN-γ and IL-5, respectively [82].

Besides established cytotoxic cNK cells, it has been demonstrated that NK cells could differentiate into antigen-presenting NK (AP-NK) cells [83], helper NK (NKh) cells [84] and regulatory NK (NKreg) cells, each defined by surface molecules and individual functions. A new CD8αα⁺MHC-II⁺ phenotype with professional APC capacity was considered to represent unusual AP-NK cells, recognizing and eliminating autoreactive T cells and finally killing them like cNK cells [85]. Human plasmacytoid dendritic cells (DCs) activated by the preventive vaccine FSME upregulate CD56 expression on their surface [86]; in mice, B220⁺CD11c^intNK1.1⁺ cells have antigen-presenting capacity like DC, hence their name interferon-producing killer DC [87, 88].

Invariant natural killer T cells (iNKT) constitute a subgroup of T cells expressing NK cell-markers. Activated by CD1d-presenting antigens, NKT could secrete not only Th1-type but also Th2-type cytokines to participate in immunity [89, 90]. Th1-polarized iNKT cells exhibit a tumor-depletion phenotype, and Th2-polarized iNKT cells contribute to tumor progression, similar to polarized T cells [91, 92]. Recent studies also highlighted new functional subtypes of iNKT cells. However, in recent years, due to their close relationship with innate immunity, iNKT cells are potentially defined as a special subgroup of ILCs.

Detection of aberrant cells by NK cells is determined by the intergradation of complex signals such as IL-12, IL-15 and IL-18 [93, 94], and the balance between activating and inhibitory signals interacting with MHC-I on the surface of target cells (Fig. 2). During infection and inflammation, NK cells are recruited and activated in a short period of time, proliferate quickly and contribute largely to both innate and adaptive immune responses [8, 95]. Except for their newly proven regulatory effects, NK cells were first found to directly target infected cells or foreign pathogens; therefore, deficiency in NK cells in both mice and humans results in susceptibility to many viral infections and adverse clinical outcomes, validated by clinicians and researchers.

Similar to other innate immune cells that are unable to accurately recognize target cells, NK cells rely on other stromal cells, including DC, which trans-presents IL-15 for NK cell activation [96], and MICA-expressing monocytes, which bind to Fc receptor to enhance antitumor function [97], to fully differentiate and induce effector responses, but surprisingly possess the ability to form immunological memory, termed “trained immunity”. Once considered as a hallmark of adaptive immunity, in recent years, the phenomenon of immunological memory has also been found in innate immune cells, especially in the myeloid lineage, e.g., monocytes and macrophages. In addition, mounting evidence indicates that in humans, NK cells can remember previous exposure to inflammatory microenvironment, and occurrence of similar cytokines could induce trans-differentiation from normal NK cells to memory NK (NKm) cells [98‐101]. This was evidently observed in response to viral infection in humans, prompting the development of NK cell-based vaccines to generate potent effects toward diseases [102, 103]. A large number of NKm cells are observed in the tumor microenvironment, producing high levels of IFN-γ, perforin and granzyme family molecules after re-stimulation [104]. However, concerning to tumors, dysfunction of NK and NKm cells is emerging as an indispensable and undeniable event, leading to not only proliferation of tumor cells but also the formation of distant metastases [101]. It was observed that repeated exposure of NK cells to NK receptor ligand-expressing tumor cells (e.g. NKG2D) finally results in NK cell dysfunction, and effector responses cannot be stimulated in vivo [105, 106]. These results indicate that the formation of NKm cells may not just depend on target cell recognition through surface receptors, and certain cytokines (including IL-12, IL-15 and IL-18) could be key to this process.

Though the half-life of normal NK cells is only for 1–2 weeks, NKm cells can live for 3–4 weeks [107]. This long-term effect truly helps researchers better modulate the function of NK cells in protecting against tumors, and emerging results suggest that NK cells not only rely on MHC-I recognition but also depend on many other signals, shedding light on the use of NK cells and related signaling pathways as future treatment options.

In 2000, an 11-year follow-up study of the Japanese general population, with rigorous use of various related biochemical and immunological markers, indicated that elevated cytotoxic activity of peripheral NK cells is positively associated with reduced cancer risk and vice versa, suggesting the certain importance of natural immune response toward tumors [108]. However, the specific role of NK cells remains controversial and largely depends on distinct cancer types [109]. Even in the same type of cancer, NK cells are highly heterogeneous, characterized by the abundance of surface receptors and the complexity of tumor intrinsic signaling pathways [95, 110]. In the CIBERSORT analysis, NK cells were thoroughly divided into resting and activated subtypes, each contributing to the formation of the tumor microenvironment [111]. In preclinical studies, NK cells were shown to indicate survival and thus therapeutic response in different types of cancer, as detected by immunohistochemistry, immunofluorescence or flow cytometry using different surface and functional markers (Table 2) [46, 56, 112‐129]. Although CD3⁻CD16⁺CD56⁺ cells reflect different clinical outcomes in different cancers, functional molecule-positive NK cells, including NKp30⁺ and NKp46⁺, indicate favorable survival, pointing out the fact that full activation but not only infiltration density finally determines NK cell-associated immune response.

In BC, besides the ability of total NK cells to reflect favorable survival, peritumoral abundance of NK cells also correlates with elevated pCR rate of neoadjuvant chemotherapy in large and locally advanced breast cancer [130], and vice versa. To address this, it is currently well accepted that tumors are highly heterogeneous, and even one histologic type can be separated into several molecular subtypes. BC can be thoroughly clustered into luminal, HER2-enriched and triple-negative types according to the expression of surface molecules, and both infiltration and activation status of NK cells vary by cluster, e.g., obviously elevated NKG2D in luminal tumors [131]. However, due to the complex and variable functional status of NK cells, their actual role in the TME still awaits further elucidation.

As noted above, NK cells rely on the balance between activating and inhibitory receptors to exert their killing effects, and perforin and the granzyme family of proteins are the main effector molecules. Consistent with cNK cells, as a newly-found subgroup of NK cells, activated NKT cells directly detect and kill CD1d⁺ tumor cells in several types of cancer [132, 133]. In addition, by expressing high levels of CD40, NKT cells induce DC maturation, thus activating CTL and cNK cells to enhance their anti-tumor effects [134, 135].

Apart from primary tumor proliferation, metastasis remains lethal, accounting for the majority of cancer-associated deaths, and invasion-metastasis cascade is largely attributable to the immune escape [136]. With such rapid and effective ability to target tumor cells directly and indirectly, NK cells are suppressed by tumor-derived molecules, tumor-educated stromal cells (Fig. 2) and tumor cells (Fig. 3), eventually contributing to the progression and multi-step metastatic process of cancer. For example, single-cell analyses found that in lung adenocarcinoma, CD16⁺ NK cells are hardly infiltrated and present lower granzyme B and CD57 expression compared with normal lung tissue, appearing to form NK cell-excluded TME [137].

Class IA phosphatidylinositol 3 kinases (PI3Ks) are involved in growth and survival of normal cells, and mutation of the PI3KCA isoform is commonly found in the genomic landscape of many cancers. Inhibiting abnormally activating signals of PI3KCB with a tested inhibitor called P110β in hematologic malignancies obviously enhances susceptibility of tumors to NK cell activity in vitro, probably through the regulation of MHC-I [138]. Tumor-derived prostaglandin E2 [139], extracellular adenosine [140], fragmented mitochondria in the cytoplasm of tumor-infiltrating NK cells [141] and actin cytoskeleton remodeling [142] also lead to immunosuppression and help potentially metastatic cancer cells to avoid NK cell elimination. Interestingly, in a breast cancer cell line, CDC42 or WASP knockdown does not change the activation status of NK cells but obviously increases the expression of effective granzyme B and overcomes resistance to NK cell-mediated attack [142]. Further investigation in this field might help identify a new signaling pathway or a new marker of NK cell-activation and provide new insights into NK cell-therapy. Similar to the inhibitory role of CD73 in T cell-immunity, tumors can train normal NK cells into CD73⁺ NK cells which express high levels of checkpoint molecules, including LAG-3, PD-1, and PD-L1, finally resulting in immune escape [57]. As mentioned above, Th2-polarized iNKT cells in the TME contribute to tumor progression through immunosuppressive effects [91], and continuous exposure to ligands expressed on the surface of tumor cells induces the dysfunction of NKm cells that are engaged in long-term anti-tumor immunity [106]. Moreover, absence of NKG2D is a common feature of functionally suppressed NK cells, which is achieved through many different pathways [105, 143], thus could be used as a marker to guide NK cell-related therapies [144].

In addition, NK cell-associated cytotoxicity can also be impaired by stromal cells, for example cancer-associated fibroblasts, monocytes, macrophages and other immune cells. Fibroblasts in TME suppress function of NK cells through downregulating ligands of NK cell activating receptors [145], enhancing tumor-associated macrophages enrichment [146], and producing extracelluar matrix components such as IDO and PGE2 [147]. In human gastric cancer, tumor infiltrating monoctyes/macrophages can reduce IFNγ, TNFα, and Ki-67 expression of NK cells via TGFβ1, thus impairing NK cell function [148]. Meanwhile, the interaction of PD-1⁺ NK cells and PD-L1⁺ monoctyes/macrophages in Hodgkin lymphoma results in immune evasion, which can be reversed by PD-1 blockade [149]. The monocytes from hepatocellular carcinoma express CD48, which could block 2B4 on NK cells and induce NK cell dysfunction [150].

Consistent with CD8⁺ T cells, activated PD-1⁺ NK cells are inhibited by elevated PD-L1 expression in the TME [58]. Dysfunction of NK cell following surgery has been regarded as a risk factor of metastasis and can be partly explained by disturbing the balance between activating and inhibitory signals. NK cell-mediated metastasis control was found dependent on dectin-1-mediated activation of macrophages, especially the plasma membrane tetraspan molecule MS4A4A [151]. In colorectal cancer, lipid accumulation is common for postoperative patients and facilitates the formation of metastasis by impairing NK cell function, by elevating CD36 expression [152]. Hence, NK cells can be rarely seen in metastatic melanoma and are mainly TIGIT⁻CD226⁻ which are deprived of cytotoxicity toward MHC-I-defient malignant cells [153]. Interestingly, proved by convincing genetically engineered models, it is the absence of NK cells but not CD8⁺ T cells that evidently leads to the metastatic dissemination of small cell lung cancer, pushing us to better define the pivotal role of NK cells in both initial progression and later metastasis of cancer [154].

NK cells, though belonging to innate immunity, have characteristics similar to cytotoxic CD8⁺ T cells [95]. Originally proven important in the two-signal activation model of T cells, CD28 is also necessary for optimal cytokine secretion and proliferation of NK cells both in vitro and in vivo [155]. It was shown that IL-21 is important for the maturation and activation of NK cells [156‐158]. However, data also uncovered the double-sided function of IL-21 in the development of NK cells and surprisingly, its positive role in T-cell-based immune response [159, 160], for example inducing KLRG1⁺CD8⁺ T cells during acute intracellular infection [161]. Besides IL-21, NK cells produce multiple other cytokines during their proliferation, maturation and function similar to T cells, including IL-2 [162], IL-7 [163] and IL-15 [164], validating the close relationship between these two types of effector cells in the body’s immune system.

Although sharing many similarities, compared with effector T cells, NK cells are more cytotoxic to tumors and possess lower immunogenicity [10]. As mentioned above, NK cells respond to target cells more quickly and do not need extra ligation of activating receptors [95]. On the contrary, NK cells have the ability of suppressing the function of CD8⁺ T cells via NKG2D in severe aplastic anemia [165]. It has been underlined that during infection with chronic lymphocytic choriomeningitis virus, NK cell-intrinsic FcRγ signaling could inhibit the expansion of CD8⁺ T cells [166]. Interestingly, tumor cells that develop checkpoint blockade resistance to CTLs, especially through suppressed MHC-I expression, are more vulnerable to NK cell-based immunity; thus, combination immunotherapy utilizing both NK cells and CD8⁺ T cells can constitute a future strategy in terms of tumor immune escape [167]. Besides, in tumors that lack MHC-I-related molecules, elevated amounts of HLA-E and HLA-G were observed, indicating the possibility for NK cells to harness this unclassical pathway [168]. More basic researches are urgently needed for better understanding of the complex relationship between these two major effector cells, as NK cells-based treatment is currently highly underestimated.

Similar to high blood pressure and diabetes, cancer is currently regarded as not only a process of pathogenesis but also a social issue. Lipid accumulation in the liver, abnormal glucose metabolism and irregular lifestyle all contribute to tumorigenesis and cancer progression, which eventually prompt research about the underlying mechanisms of this phenomenon. It is admitted that the metabolic competition between tumor and stromal cells largely affects the process of tumorigenesis and cancer progression. In the TME, NK cell function is impaired not only by suppressive cytokines but is also attributable to inappropriate metabolic conditions, including hypoxia, lack of nutrition and abnormal concentrations of tumor-derived products such as lactate, which induces unfavorable acidic condition, hindering the proliferation and cytokine production of CTLs as well (Fig. 2) [169]. As metabolic disorder is currently considered a hallmark of cancer, which shares a close relationship with the microenvironment, the idea of harnessing immunometabolism attracts increasing attention to improve the efficacy of NK cell-dependent anti-tumor therapy.

The normal breast tissue is surrounded by adipose tissue, and obesity is considered a potential risk factor for breast cancer, which is supported by population-based studies, together with high mental pressure, evidently affecting lipid and glucose metabolic pathways. Obesity-induced inflammation in adipose tissue could result in the recruitment of M1-polarized macrophages, neutrophils, NK cells and CD8⁺ T cells, and higher expression levels of pro-inflammatory cytokines, as well as obvious exclusion of Treg and invariant NKT (iNKT) cells [170]. In 2011, aware of the importance of tumor-promoting inflammation in the TME, Weinberg et al. included this phenomenon into the hallmarks of cancer and particularly highlighted inflammation induced by the innate immune response [1]. Thus, improved understanding of the mechanism by which metabolic activity affects the function of tumor-infiltrating stromal cells, finally resulting in cancer progression and immune escape, would provide clues for developing novel therapeutics for immunometabolic targets.

NK cell function can be altered by different components in the TME. Breast cancer metabolomics data overtly show that lipid and glucose metabolic pathways are highly activated, especially the fatty acid synthase glycolysis pathway, compared with paired peritumoral tissue.

Similar to other lymphocytes, NK cells require energy to survive, and glucose consumption is evidently increased after full activation, while the competition between NK cells and tumor cells could disturb such need. It has been shown that surface transporters, especially glucotransporter 1 (GLUT1), help NK cells utilize glucose to generate ATP and pyruvate, contributing to glycolysis and oxidative phosphorylation [171, 172]. Several studies have pointed out the importance of sufficient glucose supply for NK cell activities, including proliferative capacity, activation status, cytokine production and direct cytotoxicity [173, 174]. Glycolysis and oxidative phosphorylation contribute to maintain the cytotoxic ability of NK cells, as their inhibition highly decreases the expression levels of IFN-γ and Fas ligand [175]. NKG2D is essential for the activation of NK cells, which relies on glycolysis. Researchers have identified several pathways pertaining to the interlinked metabolic activity and NK cell function. Obesity-related inflammation is dependent on the IL-6/Stat-dependent pathway, thereby resulting in a distinct functional status of NK cells [176]. In addition, Assmann et al. highlighted that sterol regulatory element-binding proteins (SREBP) transcription factor-controlled glucose metabolism is essential for metabolic reprogramming in activated NK cells, providing new insights into this process [177]. Accordingly, SREBP inhibitors such as 27-hydroxycholesterol (27HC) are accumulated in the TME, partly affecting SREBP-related glycolysis in ER-positive BC [178]. However, as most studies only focus on GLUT1, other transporters and unclassical pathways should also be paid attention to, paving the way for deeper understanding of the complex relationship between glucose metabolism and NK cell function.

Hypercholesterolemia remains a risk factor for ER-positive BC. In 2013, 27HC, which negatively regulates SREBP, was also found by Nelson et al. to be a bridge linking hypercholesterolemia and BC [179]. It was also found that treating mice submitted to high-cholesterol feed with an inhibitor of CYP27A1, an enzyme important in 27HC biosynthesis, obviously decreases the number of metastases in mice, which reverses immune suppressive environment [180]. Therefore, using drugs designed to decrease blood cholesterol or directly inhibiting the formation of 27HC could be a potential strategy for patients with ER-positive BC. Surprisingly, a recent study suggested that high serum cholesterol and cholesterol accumulation in NK cells increase their anti-tumor ability by facilitating the formation of lipid rafts in the liver-tumor-bearing murine model [181], highlighting the heterogeneous functions of lipid metabolism in cancer.

Hypoxia is also a common feature of cancer, and often mentioned concurrently with low pH in the TME. Previous findings indicated that besides promoting tumorigenesis and cancer progression, hypoxia also stimulates NK cell formation via HIF-1α, initiating a conflict between suppressing and activating this signal [182]. Studies have also shown that low O₂ in TME harms the function of NK cells by downregulating activating signals such as NKG2D, NKp30 and CD16, thereby limiting cytokine production and cytotoxicity and resulting in metastasis [183, 184]. In addition to regulating intracellular signals directly, the hypoxic microenvironment could degrade NK cell-secreted functional molecules such as granzyme B [185], together with CTL-based immunity, which is partly rescued by IL-2 [186, 187]. Considering the pivotal roles of interleukin family members in the maintenance of NK cells, we paid more attention to these cytokine-primed metabolic pathways and found more clues under hypoxic conditions.

Besides, tumor cells can directly alter the metabolic status of NK cells. This can be positive as CD25 expression on NK cells is overexpressed after interaction with tumor cells, inducing long-term anti-tumor metabolic responses by promoting glycolysis and NK cell survival, supported by mTORC1/cMYC signaling activation. However, worsening occurs later as the harmful effects overcome the positive impact. For instance, glutamine addiction and high consumption of nutrients remain common in BC. In vitro studies showed arginine deficiency inhibits IFN-γ production by primary human NK cells [188]. Additionally, mTOR signaling within NK cells can be largely suppressed under low-arginine or glutamine conditions, which also affect IL-2-related stimulation process via cMYC [189]. Upon direct contact with tumor cells to form an immune synapse in response to local energy consumption, mitochondria of NK cells are depolarized and lose metabolic energy [190]. Inhibiting PPARα/δ or blocking the transport of lipids into mitochondria reverses NK cell metabolic incapability and restores cytotoxicity [191].

Indeed, previous studies have reported that anti-PD-L1 therapy could reshape metabolic pathways in the tumor microenvironment and re-stimulate exhausted CD8⁺ T cells for cytotoxicity [192]. Interestingly, NK cells also express the ligands of these checkpoints. Glycoengineering of NK cells enhances their killing ability toward CD22⁺ lymphoma in a CD22-dependent manner [193]. Blockade of monocarboxylate transporter 1, which regulates cell metabolism, using AZD3965 also potentiates NK cell activity [194]. Studies that focus on translating mature theories into the practical use of NK cells are promising.

As mentioned above, iNKT cells can be polarized into different properties, each possessing distinct functions. The normal breast tissue is surrounded by adipose tissue. Different from conventional T cells, iNKT cells comprise large amounts of stromal cells in adipose tissue, whose infiltration decreases apparently in high-BMI individuals [195, 196].

With invariant TCR on the surface, iNKT cells are termed innate-like T lymphocytes and act on the front line of the immunity battle against cancer [197, 198]. iNKT cells recognize glycolipid signals but not peptides via semi-invariant TCR, and are restricted to glycolipid antigens presented via CD1d-related molecules, which are MHC-like and highly enriched, especially in adipocytes and hepatocytes, linking innate and adaptive immune responses. This process can be altered by metabolic activity. GM2 is a glycosphingolipid that binds the CD1d molecule. Pereira et al. pointed out that GM2 inhibits the activation of iNKT cells in a dose-dependent manner, which might result from its competition with α-GalCer for binding CD1d [199]. Compared with T lymphocytes, iNKT cells show much higher capacity of glycolysis but reduced mitochondrial respiratory activity, resulting in particular molecular features. Under hypoxia, widespread RNA editing can be induced by mitochondrial respiratory inhibition via APOBEC3G, an endogenous RNA editing enzyme [200]. Fu et al. demonstrated that aerobic glycolysis in iNKT cells is highly increased after TCR engagement, which is essential for the production of IFN-γ [90]. This process can also be inhibited by the lack of glucose. Hence, reduced expression of IFN-γ in iNKT cells compared with the normal tissue was confirmed in several tumor types, predicting patient response to the therapy of PD-1 blockade [201‐203]. In humanized mouse models undergoing PD-1 blockade and CAR-T (with different costimulatory molecules) combination therapy, only those with Δ-CD28 CAR control tumor growth, and in vitro analysis showed that these cells exhibit elevated glycolysis, fatty acid oxidation and oxidative phosphorylation [204]. Overcoming the impaired metabolic function combined with immune checkpoint blockade would be a potential strategy in future researches and clinical practice.

However, what is currently known about iNKT cells is just the tip of the iceberg. Compared with T lymphocytes, it remains unclear how intracellular metabolic signals influence the survival and function of iNKT cells, which deserves further investigation, as this may be the next potential target of cancer immunometabolic therapy after CD8⁺T cells and NK cells.

IL-15 promotes the development and cytotoxic ability of NK cells, and several clinical trials have illustrated the safety profile of recombinant human IL-15 (rhIL-15) in multiple tumors [208, 209] as well as its agonist, ALT-803, in metastatic lung cancer and post-transplantation patients (Fig. 4) [210, 211]. In an open-label, phase Ib trial, ALT-803 showed a fantastic potential when combined with anti-PD-1 monoclonal antibody (nivolumab) without increasing the incidence of very severe grade 4 or 5 adverse events [211], which evidently could be the future way to enhance rhIL-15 treatment. In addition to soluble IL-15 in the microenvironment, it has been revealed that in the mouse, direct contact with membrane-bound IL-15 on adjacent stromal cells could induce stronger cytotoxic effects in NK cells [212]. Heterodimeric IL-15 can also increase intratumoral NK cell and CD8⁺ T cell infiltration, elevating the effective rate of current immunotherapy [213].

Apart from IL-15, other interleukins are also synergetic to this process (Table 3). IL-21 enhances tumor rejection in mice via NKG2D-dependent NK cell activity, suggesting IL-21 to be a possible target for immune escape induced by NKG2D elicitation [214]. However, IL-15-dependent expansion of resting NK cells can be suppressed by IL-21, while on the other hand adaptive immune response is enhanced [159], providing substantial insights into this complex network in clinical use. Besides, blocking CIS could promote IL-15-type cytotoxicity and thus results in increased production of IFN-γ [22]. NK cells pre-exposed to IL-12, IL-15 and IL-18 accumulate in the tumor tissue and retain their anti-tumor function both in vitro and in vivo. However, IL-15 alone does not exert such effects [104]. In addition to the inner pathway, treatment with IL-2 and IL-15 obviously enhances glycolysis and oxidative phosphorylation of NK cells, thus promoting the killing ability. In a first-in-human phase I multicenter study, NKTR-214, a novel IL-2 pathway agonist, was found promoting proliferation and activation of NK cells without expansion of Treg cells [215]. For metastatic melanoma refractory to CD8⁺ T cell cytotoxicity due to the lack of MHC-I, combination of IL-15 and TIGIT blockade shall be effective by stimulating NK cell-mediated immunity [153].

As mentioned above, cytokines (especially IL-12, IL-15 and IL-18) are critical to the formation of NKm cells. Memory-like NK cells supplemented with IL-12, IL-15 and IL-18 also show enhanced responses against acute myeloid leukemia both in vitro and in vivo [107], and are currently assessed in a first-in-human clinical trial. However, studies also pointed out that IL-12 could increase NKG2A expression and inhibit the activation of NK cells [216].

As shown previously, in parallel with CD8⁺ T cells, NK cells can also be suppressed by immune checkpoint molecules. After cetuximab treatment, PD-1⁺ NK cells are more enriched in the TME and are correlated with favorable clinical outcome in head and neck cancer patients, which was further demonstrated by in vivo experiments and a stage III/IVA clinical trial assessing neoadjuvant cetuximab (NCT01218048) [58]. With PD-1 blockade (nivolumab), cetuximab-induced NK cell activation and function are remarkably enhanced in PD-L1^high tumors.

In addition to already-well-defined PD-1 and PD-L1, NK cells with reduced amounts of T-cell immunoglobulins and ITIM domain (TIGIT) show higher levels of cytokine secretion, degranulation activity and cytotoxicity [217], and blockade of TIGIT could prevent exhaustion in NK cells [20]. A recent study also highlighted that TIGIT-depleted NK cells are highly sensitized [218] and develop resistance to MDSC-mediated immunosuppression [219]. Meanwhile, CD96, which shares the same ligand CD155 with CD226 and TIGIT, negatively controls the immune response by NK cells [220] and predicts adverse survival in human hepatocellular carcinoma [56]. Single use of CD96 antibody promotes NK-cell-induced anti-metastatic ability [221], and such effect is largely increased when combined with anti-CTLA-4, anti-PD-1 or doxorubicin chemotherapy [222].

Though prospective, unexpected biological events have been observed in a single-arm phase II study showing that intravenous infusion of 1 mg/kg IPH2101 (a human monoclonal antibody against KIRs) results in severe contraction and obvious inhibition of NK cells in myeloma patients [223]. Lirilumab, a 2nd generation antibody targeting KIR, had encouraging results in a phase I trial, which demonstrated its safety [224]; however, the subsequent phase II trial in AML patients showed no clinical effects. Combination of CIS inhibition with CTLA-4 and PD-1 blockade exerts even greater effects in reducing melanoma metastasis compared with either of these treatments administered alone; thus, CIS inhibition could offer an alternative therapeutic option for patients not responding to other immune checkpoint inhibitors [22].

As an essential receptor for the activation of NK cells, NKG2D is blocked by many ligands (e.g., MICA, MICB, and ULBP1–6) upregulated in tumor cells as a result of abnormal cellular stress in the TME. A recent study demonstrated that antibodies targeting MICA and MICB can prevent NK cell recognition and tumor cell binding, inhibiting tumor growth in fully immunocompetent mouse models as well as humanized mouse models [129]. Furthermore, combination treatment targeting soluble MIC, e.g., MICA and MICB, and PD-L1 shows better effect than monotherapy in vivo [225]. Moreover, soluble MULT1, a high affinity mouse NKG2D ligand stimulates NKG2D in distant NK cells and enhances NK cell tumor immunity [106]. Hence, a clinically used antibody, monalizumab, has been developed targeting NKG2A, an inhibitory checkpoint of NK cells, which not only promotes NK cell function in various preclinical models, as previously characrerized, but also potentiates anti-PD-1 [226] and anti-EGFR (cetuximab) therapy [227]. In addition to antibody, NKG2A^null NK cells, constructed through retroviral transduction of NKG2A blocker which inhibits de novo NKG2A expression, present increased anti-tumor activity in pre-clinical model [228].

In summary, besides the targets close to T cells, including FDA-approved anti-CTLA-4 (ipilimumab) and anti-PD-1 (nivolumab, pembrolizumab) antibodies, others designed specifically for NK cells are also under clinical trials, e.g., anti-KIR (IPH2101, lirilumab) and anti-NKG2A (monalizumab) (Table 3) (Fig. 4).

As an applicable option for enhancing autologous immunity, adoptive transfer of NK cells has been implemented to treat certain types of cancer in the past few years [229]. Previous studies of NK adoptive transfer in acute myeloid leukemia patients have presented slightly beneficial effects in controlling disease [230, 231], and a phase II clinical trial in patients with recurrent ovarian or breast cancer showed that adoptive transfer of haploidentical NK cells after lymphodepleting chemotherapy leads to a temporary benefit but its clinical value remains controversial, and is partly limited by recipient reconstitution of regulatory T cells [232]. Apart from adults, a phase I clinical trial applied autologous ex-vivo-expanded NK cells to children with recurrent medulloblastoma and ependymoma and obtained good safety and therapitic efficacy [233]. Interestingly, transfer of NK cells along with CD34⁺ hematopoietic stem cells shows no added adverse effects but potential therapy response in older patients with acute myeloid leukemia [234]. Combinational application of NK cell infusion with monoclonal antibody provides a new direction of combinational immunotherapy. In a phase I trial, activated autologous NK cell were infused into patients with HER2-positive solid tumor undergoing trastuzmab and showed prelimitary anti-tumor phenotype [235].

In 2009, Fujisaki et al. found that overexpression of telomerase reverse transcriptase could lead to over 100 additional doubling cycles in NK cells, revealing a potential way to overcome the limitation of NK cell amplification in vitro [236]. However, though the adoption of NK cells seems promising in preclinical and clinical researches, many questions exist, e.g., the undeniable fact that NK cells gain self-renewal ability following infusion. Originally designed and considered a one-time-use therapeutic process, a small part of NK cells surprisingly remain alive and proliferate in the human body for months, mediating continuous surveillance against tumor [237]. However, in addition to beating tumor cells, the risk of long-lived NK cells should not be ignored where their “brake” is lost and they might also kill normal cells, even worse, finally increasing the possibilities of NK lymphoma [95, 238]. Hence, the ability of transferred NK cells may be limited by inappropriate persistence or expansion in vitro, and biology-driven methods are commonly used to overcome this issue before adoptive transfer. A recent study indicated that NK cells pre-activated by IL-12, IL-15 and IL-18 suppress graft-versus-host disease but obviously inhibit the generation and function of CD8⁺ T cells, which could result from mutual competition of IL-2 [239], limiting its value in the field of cancer treatment. The use of mouse models to unveil the detailed features of adoptive NK cells following in vitro proliferation would actually provide deeper insights into this treatment strategy, shedding light on the future implementation of NK cell-based therapy toward cancer.

Genetic modification of immune cells by chimeric antigen receptors (CARs) to target tumor cells directly is a promising therapeutic option in cancer therapy. Kymriah (CTL019, a CAR-T product) by Novartis was approved by the FDA for treating recurrent and refractory acute lymphoblastic leukemia in 2017, and remains under investigation for other indications in several clinical studies [240‐242]. Two months later, Yescarta by Kite Pharma was approved for diffuse large B cell lymphoma [243]. Due to serious adverse effects induced by CARs, especially cytokine releasing storm (CRS) and neurotoxicity, Actemra (tocilizumab, anti-IL-6 monoclonal antibody) was then approved for CRS, with a further study also showing potential application of Anakinra (an antagonist IL-1 receptor) in such case [244, 245]. In NK cells, antibody engineering approaches optimize NK cell-mediated ADCC to tumor cells through the bispecific killer cell engager (BiKEs) or trispecific killer engager (TriKEs) antibodies (Fig. 4). BiKE connects a single-chain variable fragment (scFv) to the anti-CD16 recognition site with the scFv of a tumor specific antigen, such as CD19/CD20 for non-Hodgkin lymphomas, CD33/CD123 for acute myelogenous leukemia/AML and CD30 for Hodgkin lymphoma, to enhance NK cell recognition of tumor cells [246]. TriKE consists of a BiKE and cytokine IL-15, which boost NK cell function and survival. It was shown that CD19-CD16 BiKE engineered NK-92 cells are sufficient to overcome NK cell resistance in B-cell malignancies [247]. Meanwhile, CD16-IL15-CD3 TriKE can activate suppressed NK cells and induce NK cell-mediated control of MDS and AML [248]. The advantages of CAR-NK therapy are obvious, including higher possibility of recognizing tumors (including cytokines and apoptosis) and lower incidence of CRS compared with CAR-T (Table 3) [249, 250]. In 2018, Enli Liu et al. transduced cord blood-derived NK cells with a retroviral vector incorporating the genes for CAR-CD19, inducible caspase-9-based suicide gene (iC9) and IL-15, and demonstrated the efficacy and safety in cell lines and the murine model. In the engineered NK cells, CAR-CD19 redirected the specificity of NK cells against leukemia, IL-15 promoted NK cell proliferation, and iC9 allowed NK cells to initiate suicide after killing the target cells [251]. Since CAR-NK cells exhibit striking efficacy and limited toxicity, clinical trials assessing these CAR-NK cells have been launched. In recent phase I and II trials, iC9/CAR-CD19/IL-15 NK cells were prepared ex vivo and infused into patients with relapsed or refractory CD19-positive cancers after lymphodepleting chemotherapy. Among the 11 treated patients, 8 had an objective response, including 7 with complete remission, without major toxic effects [250]. Apart from engaging NK cell with CAR, geneic modification involves deleting surface molecules, e.g., CD38, on NK cells, which evidently elimates fratricide and enhances cytotoxic ability [252].

However, limitations should be mentioned. Due to the complex process of producing CAR-NK cells, the current procedure is too expensive and the effects on solid tumors are far from being satisfactory. CAR-iPS might be a future direction, which can grow in vitro and differentiate into CAR-NK cells in vivo to directly enhance anti-tumor immunotherapy [253]. Recently, Zhu H et al. reprogramed NK cell metabolism by depleting CISH in human iPSC-derived NK cells and obtained satisfactory persistence and anti-tumor activity in vivo, which could be a novel method to generate CAR-NK from CAR-iPS [254].

As an important immunosuppressive cytokine that promotes tumor progression, TGF-β and its pathway represent potential opportunities for anti-tumor drug development. The clinical modulation of TGF-β, which is achieved through small-molecule inhibitors and antibodies, is being investigated in a number of clinical trials [255]. As the safety and efficacy of TGF-β blockade therapy have been demonstrated, two studies independently showed that combination treatments of TGF-β blockade with anti-PD-1/PD-L1 therapies have synergistic effects on murine EMT6 breast mammary carcinoma and colorectal cancer [256]. It is admitted that TGF-β inhibits NK cell metabolism, proliferation, cytokine production, cytotoxicity and anti-metastatic functions through mTOR signaling in multiple cancers [257]. Activin-A, a member of the TGF-β superfamily, increases ILC1-like tissue residency features, reduces cytokine production, and suppresses proliferative and metabolic functions in both human and murine NK cells through an alternative SMAD2/3-related pathway with effects similar to those of conventional TGF-β pathway [258]. Therefore, targeting TGF-β, TGF-β superfamily, and their downstream pathways in NK cells may be promising treatment options that enhance the efficacy of current immunotherapies (Fig. 4). Apart from blockage of immunosuppressive cytokine, immunostimulatory agonist have been tested in clinical trials. In a recent phase I/IIa study, BMS-986156, a human glucocorticoid-induced TNF receptor-related protein agonist, appears to increase NK cell proliferation no matter applied with or without nivolumab, and has an adorable safety and efficacy profile [259].

Rapamycin, an inhibitor of the mTOR pathway, and its derivative everolimus are effective for treating breast cancer in clinical trials, including BOLERO-6 and PrE0102 studies, either alone or in combination with endocrine and chemotherapy [260, 261]. However, NK cells highly rely on PI3K-mTOR signaling pathway-dependent metabolic reprogramming to exert their anti-tumor effects. Therefore, in patients with both PI3K-mTOR pathway activation and NK cell-based microenvironment, such inhibitors should be cautiously employed.

In HER2-enriched breast cancer, besides HER2-targeted therapies, anti-GD2 also appears to be promising in preclinical studies. However, GD2-related pathways are essential for ADCC [262], also reducing the effectiveness of NK cells, which provides insights into the complex relationships among such networks to minimize off-target effects exerted by established therapies on immunotherapy.

Radiotherapy was found to kill cancer cells via inducing DNA damage, but recent studies found its ability to cause immunogenic cell death, named immunogenic radiotherapy [263]. There is a growing interest to combine immunogenic radiotherapy and immunotherapy, plus chemotherapy or target therapy. A triple-combination therapy which inhibits PD-1 and MER proto-oncogene tyrosine kinase plus radiotherapy, increases NK cells infiltration in abscopal TME [264]. Adding indoximod, an inhibitor of IDO pathway, to radiotherapy and PD-1 blockade also enhances NK cell activity and shows great clinical response [265]. Another representative drug of triple-combination is selenium-containing nanoparticles,which delivers doxorubicin to the tumor site and releases high energy rays. The rays not only kill tumor cells, promote doxorubicin release, but also produce seleninic acid which enhances NK cell function [266].