The linear model has traditionally been the prevailing perspective on NK cell development, and recent insights into ILC subsets have contributed to a more nuanced understanding of this developmental trajectory. ( Figure 3 ) According to the linear model, hematopoietic stem cells (HSCs) differentiate into common lymphoid progenitors (CLPs). CLPs have the potential to commit to various ILC progenitor lineages, including B progenitor cells (pre-B), T progenitor cells (pre-T), NK progenitor cells (NKP), and others (24–26). CLPs subsequently differentiate through an intermediate stage of pre-NK progenitors (pre-NKPs) into NKPs (27, 28). The acquisition of the common β-chain of CD122 and IL-15 by NKPs represents a critical milestone in NK cell differentiation. The expression of CD122 signifies the irreversible commitment of NK cells from CLPs, while IL-15, produced by hematopoietic cells, monocytes, and dendritic cells, is thought to facilitate NK cell development in the bone marrow (29). Following this, NKP cells differentiate into immature NK (iNK) cells, which are characterized by increased expression of IL-1R1 and the emergence of CD314 (NKG2D), CD335 (NKp46), CD337 (NKp30), and CD161 (the human homolog of NK1.1 in mouse). The subsequent transitional stage is marked by the appearance of CD56^bright NK cells, which display high levels of CD56 and peak expression of NKG2D, NKp46, NKp30, and CD161. NK cells at this stage can be subdivided into phase 4a NK cells, which express more IL-22, and phase 4b NK cells, which express higher levels of transcription factors T-BET and EOMES and are capable of producing interferon-γ and thus cytotoxicity (30). CD56^bright NK cells eventually transition into CD56^dim NK cells, distinguished by increased CD16 expression, decreased CD56 expression, and the expression of various CD158 (KIR) subtypes (24, 26). CD56^dimCD16⁺ NK cells can be further classified into subpopulations based on CD57 expression, including CD56^dimCD16⁺ NK cells and CD56^dimCD16⁺CD57⁺ NK cells, with the CD56^dimCD16⁺CD57⁺ subset being considered the terminally differentiated subset, generated by prolonged stimulation from antigens or inflammatory signals (26, 31, 32). As the human NK cell developmental system has been refined, the developmental process of NK cells in mouse models has also been better characterized, aligning with this linear model. A recent study proposed a four-stage developmental model for the further progression of NKP cells according to the expression of NKp46. The NKP population sequentially matures into the CD122⁺NK1.1⁻CD49b⁺NKp46⁻ iNK-a population, the CD122⁺NK1.1⁺CD49b^−/+NKp46⁻ iNK-b population, and finally, the CD122⁺NK1.1⁺CD49b⁺NKp46⁺ mNK population. These four NK cell populations display distinct phenotypic differences in cell surface markers, transcription factors, and effector molecules (33).

Notably, ILC cells share characteristic markers with NK cells, and the homology between them has been significantly enhanced through exploration, which has been pivotal in elucidating the developmental trajectory of NK cells and constructing the developmental lineage. It is recognized that the precursors of NK cells may diverge from those of ILC1s, ILC2s, and ILC3s. According to the development and functions of ILCs, including the production of cytokines and the expression profile of transcription factors, ILCs can be divided into five major subsets, including NK cells, ILC1s, ILC2s, ILC3s, and LTis (lymphoid tissue inducers) (34, 35). Common innate lymphoid progenitors (CILPs) are considered the intermediate stage between CLPs and the development of ILCs. CILPs can differentiate into NKPs or common helper innate lymphoid progenitors (CHILPs), which subsequently produce lymphoid tissue inducer progenitors (LTiPs) and innate lymphoid cell precursors (ILCPs). LTiPs differentiate into LTis, while ILCPs further differentiate into ILC1s, ILC2s, or ILC3s (34, 36), and NKP cells develop into mature NK cells along the well-known linear model. ( Figure 3 ) However, this model cannot explain all experimental analytical conclusions, and many studies have supplemented this known developmental model, with many unknown directions yet to be explored. A study using global transcriptomics, flow cytometry analysis, and genetically engineered mouse models found that NK cells in the tumor microenvironment can depend on TGF-β to convert into intermediate ILC1 groups and ILC1 groups (37). Furthermore, emerging evidence suggests that previously identified ILCP cells retain the capacity to generate NK cells (38–41), indicating that the differentiation and commitment process within the ILC lineage is more intricate than previously thought. Additionally, Chen L et al. found that the CD34−CD117⁺ ILCP population can express CD56, produce NK cells and ILC3s, but not ILC2s. In vitro molecular and functional analysis and lineage differentiation analysis have proven that the final step of ILC2 development is independent and non-overlapping with NK cells and ILC3s (40). It is worth noting that NK cells and ILC1 together form group 1 ILCs, whose development and function rely on the T-box transcription factor T-bet and are characterized by the production of interferon-γ (IFN-γ) (34). An accurate understanding of the developmental relationships between NK cells and other cells within the ILC lineage, as well as their interactions, is essential for constructing a comprehensive NK cell developmental lineage.

Additionally, many questions remain unresolved. For example, whether there are still undefined subgroups in the development process of NK cells, from which stage NK cells separate from T/B cells and become independently developed individual cells, and the developmental process between ILCs and NK cells has not yet been precisely defined.

Deciphering the cellular state transitions behind immune adaptation is fundamental to advancing biology. Through the application of single-cell sequencing, researchers are able to track the fate of individual cells, revealing the genetic changes as cells transition from a single state to new cell states and types, which helps to understand the development process of multicellular organisms. Although single-cell sequencing has not been extensively applied to the study of NK cell development, its research achievements are still encouraging, as it is expected to construct a complete developmental system from a single-cell perspective and to deeply understand the ontogeny of NK cells. Meanwhile, scRNA technology plays a key role in exploring ILC heterogeneity (35, 42), and it is expected to provide clues for the origin and understanding of NK cells and ILCs.

scRNA-seq technology can further elucidate the reliability of the linear model, explore the continuity of the developmental process, discover new subpopulations, and define the developmental process based on gene expression patterns. Yang et al. (43) used scRNA-seq technology to study the heterogeneity of NK cells in the peripheral blood and bone marrow of healthy individuals, and transcriptomic and pseudotime analysis confirmed the progression from CD56^bright NK to CD56^dim NK. A study on liver cancer also identified the continuous process of transformation from CD56^bright NK cells to CD56^dim NK cells through single-cell analysis (44). Furthermore, Yang et al. (43) identified a transitional NK cell population that bridges CD56^bright NK cells and CD56^dim NK cells, characterized by intermediate expression of marker genes associated with either CD56^bright or CD56^dimCD57⁺ NK cells. More recently, Melsen JE et al. (45) discovered the CD56^dimGZMK⁺ subpopulation, which is considered an intermediate stage between CD56^bright and CD56^dim NK cells. This subpopulation exhibits shared characteristics with both CD56^bright and CD56^dimGZMK^- NK cells and displays higher expression of SELL (CD62L) and lower expression of GZMB, PRF1, FCGR3A (CD16), and FGFBP2 in comparison to CD56^dimGZMK^- cells.

Furthermore, single-cell sequencing has provided insights into the point of divergence of NK cells from the lymphoid lineage. A study on the single-cell transcriptomic atlas of human blood cells revealed through pseudotime analysis that memory B cells and NK cells are distributed on two different branches, which better reflects the distinct trajectories of differentiation between B cells and NK cell populations. They also found that NK cells and T cells lack continuous transcriptional transitions from progenitors to differentiated cells and inferred that these cells may deviate from T cell development during the macro-DN(Q) stage of gene expression acquisition (46). Another study, utilizing single-cell RNA sequencing and adaptive immune receptor (AIR) sequencing (scVDJ-seq), has employed TRBJ frequency to create a V(D)J feature space, as all T, ILC, and NK cells express TRBJ. The inferred trajectories suggest that ILC/NK cells diverge from T cell development at the DN (early) and DN (Q) stages, while also indicating that T cells and ILC/NK cells may share a common initial developmental stage before diverging prior to the production of productive TRB, TRG, and TRD (47). This provides better evidence for the separation of NK cells from the lymphoid lineage. Meanwhile, scRNA-seq research reveals a more continuous transcriptional differentiation landscape, although lacking classical defined precursor stages, it contributes to a more systematic understanding of NK cell development (48–50).

Notably, the relationship between the ILC and NK cell developmental lineages and the heterogeneity between them has been better authenticated under single-cell sequencing, and these studies are expected to provide clues for the origin and understanding of NK cells and ILCs. A study identified common ILC1/NK cell progenitors and referred to them as aceNKPs. Single-cell RNA sequencing of the descendants of aceNKPs in the lung confirmed that aceNKPs are not only progenitors of NK cells but also of ILC1 cells. It was found that the descendants of aceNKPs could not be separated according to the traditional dichotomy between ILC1 and NK cells, and no clear demarcation of the transcriptional profiles between the ILC1 and NK cell lineages was observed, indicating that these closely related group 1 innate lymphocytes are indistinguishable. Furthermore, a single-cell study on tissue-resident NK cells (trNK) in the skin found that trNK cells retained the expression of the transcription factor ectodermal protein EOMES and did not acquire Hobit expression, indicating that trNK cells do not transition into ILC1, and ILC1 do not convert into trNK cells (51), challenging the notion that NK cells can readily convert into ILC1. Moreover, the heterogeneity between NK cells and ILCs has been a hot topic in recent years, and a comprehensive understanding of this heterogeneity is crucial for studying the origin of both and for the identification and selection of cells in single-cell sequencing. As note, ILC cells share typical markers with NK cells, and in humans, NK cells (c-kit^-) are typically distinguished from ILC3 (c-kit⁺) (52). In mice, the distinction between NK cells and ILC1s is often based on the expression of EOMES, with NK cells (CD49a^- CD49b⁺) and ILC1s (CD49a⁺ CD49b^-) being differentiated by the differential expression of integrins CD49a and CD49b (53). Lopes et al. (35) used single-cell RNA sequencing and CITE-seq to explore the heterogeneity between inter-tissue NK and ILC1 and defined syndecan-4 as a marker to distinguish mouse ILC1 from NK cells. They also found that the expression of EOMES, GZMA, IRF8, JAK1, NKG7, PLEK, PRF1, and ZEB2 defines NK cells, while IL7R, LTB, and RGS1 distinguish ILC1s from NK cells in both mice and humans. McFarland et al. (42) employed scRNA sequencing to clarify the gene signatures of mouse ILC1^-NK cells obtained from tissue, tumors, and circulation, identifying unique phenotypic markers, transcription factors, and metabolic characteristics that distinguish tissue NK cells from circulating NK cells and ILC1. While these studies provide important clues for the investigation of NK cell development, it is acknowledged that the process of NK cells and ILCs developing from progenitors has not been thoroughly investigated. Therefore, it is imperative to further deepen our understanding of the origin of NK cells, as this knowledge will inform ongoing research into NK cell heterogeneity.

Previous studies have indicated that during differentiation, both immature and mature NK cells migrate from bone marrow parenchyma to sinusoids and subsequently enter the bloodstream (59, 60). The circulation of blood between lymphoid and non-lymphoid organs facilitates their entry into secondary lymphoid tissues and various other tissues at different stages of development, differentiation, and activation (54, 61, 62). Some specialized subsets are capable of returning to specific niches within the BM to fulfill functions such as control and surveillance of malignant cells (59). However, more detailed research has revealed the presence of trNK cell populations in certain tissues, which are believed to establish residency early in life without the need for microbial stimulation during development and are capable of self-maintenance (63). Moreover, due to the functional and residential similarities between ILC1 and NK cells, ILC1 found in organs has traditionally been considered a subtype of trNK cells (64). Despite sharing overlapping transcriptomic regions and dependence on IL-15, the developmental pathways of these two cell types are quite distinct (36). NK cells derive strictly from NKP cells, which are committed to producing NK cells and do not generate other ILC subsets. It is evident that the trNK cell population exhibits distinct developmental requirements from cNK cells and ILC1 and is considered unique in ontogeny (65).

Accurate knowledge of the origin of tissue-resident NK cells is crucial for understanding the roles of NK cells in various tissues and their physiological functions, and it forms the basis for further insights into the functions of NK cells under pathological conditions such as inflammation and tumors. Studies have suggested that trNK cells may originate from tissue-resident progenitor cells rather than following the bone marrow-bloodstream pathway. It has been observed that NK progenitor and immature cells mature in secondary lymphoid tissues such as spleen, lymph nodes, and tonsils (24, 25, 66). These progenitor and immature NK cells can also be inoculated into the lymph nodes and intestine, where they reside and act as reservoirs for the maintenance of NK cell populations (55). Furthermore, a study utilizing 12-color flow cytometry identified pre-NKPs, which represents an intermediate developmental stage between the upstream CLP cells and the downstream NKP cells and was previously considered the initial stage of commitment to the NK lineage. Unlike other transplant recipients, intrathymic injection of these progenitors did not produce NK cells, suggesting that thymic NK cells have a distinct developmental origin (28). Some studies have also provided evidence that DN1 thymic progenitor cells can differentiate into thymic NK cells, which are distinct from cNK cells (67). These findings indicate the presence of NK cell progenitors in the human thymus that can differentiate and develop independently to form tissue-resident NK cells within the thymus. Additionally, in mouse lymph nodes, certain NK cell subsets are thought to originate from NKP cells in the thymus, while others are thymus-independent NKP cells originated from lymph nodes, indicating the presence of NK progenitors in lymph nodes (68). The advent of single-cell sequencing technology, with its high-resolution capabilities, has further characterized the developmental process, providing compelling evidence for the existence of trNK progenitor cells. Recent studies combining single-cell transcriptomics, flow cytometry, and TCRD rearrangement data have demonstrated that non-T cell lineages, including NK cells, can develop in the human thymus (69), confirming the presence of NK cell progenitors in tissues capable of developing into trNK cells. In 2018, Eric Vivier et al. identified four subgroups in the human spleen at single-cell resolution, among which hNK_Sp3 and hNK_Sp4 cells did not resemble any blood NK subgroups and appeared to be spleen-specific (70). Three years later, they used the same technology for single-cell sequencing of the bone marrow and found that spleen NK 0 (hNK_Sp3) cells could also differentiate into spleen CD56^dim NK cells and CD56^bright NK cells. This contradicts the notion that NK cells develop exclusively in the bone marrow, suggesting that some precursors of NK cells are enriched in extramedullary tissues (71). Moreover, researchers have identified CD117⁺ ILC clusters lacking mature ILC markers in the colon and lung through single-cell sequencing, which, unlike circulating ILCs, exhibit a transcriptome consistent with tissue residency and may represent resident precursors of mature ILCs in these tissues (35). These studies confirm that the development of NK cells not only originates from the bone marrow but also involves some extramedullary tissues that harbor progenitors of the NK cell lineage, which differentiate into trNK cells with a distinct developmental process from cNK cells and present unique growth requirements.

Notably, in specific situations such as infection and tumors, some studies have also observed that cNK cells can be recruited to tissues to fulfill various roles. Recent investigations have revealed that cNK cells are redistributed and persistently located at previously virus- and bacteria-infected skin sites through a mechanism that promotes tissue retention, and they differentiate into TCF1^hiCD69^hi NK cells that are transcriptionally similar to CD56^brightTCF1^hi NK cells in human tissue (51, 65). Utilizing single-cell sequencing technology, Dean I et al. (72) discovered in mouse solid tumors that cNK cells recruited from the circulation can transform into CD49a⁺ NK cells that adapt to the tumor environment, establishing a tumor-resident state and expressing markers of tissue residency such as CD49a and CD69. These cells also exhibit effector functions, including the killing of cancer cells and the promotion of the recruitment and activation of dendritic cells (DCs). Another study, employing scRNA-seq, found that circulating NK cells are recruited to the salivary gland in a CX3CR1^-dependent manner, where they develop into NKRM (tissue-resident memory-like NK) cells. These long-lived tissue-resident cells prevent autoimmunity by eliminating CD4⁺ T cells in a TRAIL-dependent manner, providing sentinel protection for the host and playing a crucial role in maintaining immune homeostasis, limiting immune responses, and preventing autoimmune damage (73).

By elucidating the origin and developmental pathways of trNK cells, we can ascertain that the tissue-resident nature of these cells may arise from pre-existing tissue-resident NK cell precursors within the tissue, thereby becoming a core element of NK cell heterogeneity. Concurrently, our research offers researchers valuable insights. For instance, in recent years, the application of NK cell therapy has expanded significantly. Beyond the challenge of poor targeting affecting NK cell efficacy, insufficient infiltration of NK cells into target tissues remains a substantial obstacle that must be addressed. Our research opens up new directions for researchers, as NK cell precursors can be used as research objects to regulate the tissue-resident capacity and infiltration level.

As previously discussed, growing evidence supports the notion that the tissue microenvironment influences the expression profile of NK cells. We can discern the role of the tissue environment in the development of trNK cells by comparing the environmental differences between trNK cells and cNK cells, as well as by examining the variations in trNK cells across different tissues.

Tissue-resident trNK cells and conventional cNK cells may originate from distinct precursor cells and exhibit significant disparities in their differentiation, development, and transcriptional profiles. Unique transcriptional signatures indicate that the cellular environment is a key factor in generating NK cell heterogeneity within the same developmental stage (58, 74, 75). A consistent feature among tissue-resident NK cells across various tissues is their expression of tissue-resident markers and a less mature phenotype compared to cNK cells, with their development being less reliant on the transcription factor NFIL3 (76). In both trNK cells and cNK cells, the distinct cytokine milieus influence their activation profiles. IL-21R is highly expressed in trNK cells, while IL-18R is predominant in cNK cells, and their respective ligands, IL-21 and IL-18, primarily facilitate the activation of trNK and cNK (77). Notably, cNK cells are absent in NFIL3-deficient mice, whereas trNK cells persist in the liver, skin, and uterus of these mice, suggesting differential requirements for transcription factors during development (78). The different tissue environments also result in distinct cytokine expression profiles between trNK cells and cNK cells. For instance, liver trNK cells exhibit higher expression of TNF-α and GM-CSF compared to cNK cells (78). A recent study on liver cancer demonstrated that the local immunosuppressive microenvironment during the chronic liver injury phase affects peripheral NK cells but does not impact tissue-resident NK cell subsets (79). This suggests that different NK cell types exhibit varying tolerances to the immune microenvironment.

Additionally, significant variations exist among NK cells across different tissues, which are evident in the differential distribution of NK cell subsets, as well as in the disparities in phenotype, function, and transcription factor requirements of NK cells in various tissues. These differences reflect the impact of the local microenvironment on the acquisition of tissue-specific traits by trNK cells (76). Studies have demonstrated that the abundance of NK cells varies across different tissues. The prevalence of CD56^dimCD16⁺ NK cells expressing CD57, a marker of terminal differentiation, differs between locations, with higher frequencies observed in tissues rich in CD56^dimCD16⁺ NK cells such as blood, spleen, bone marrow and lung, and lower frequencies in locations like lymph nodes and intestines (55). In humans, CD56^brightCD16− NK cells exhibit tissue-resident traits in multiple locations, while CD56^dimCD16⁺ NK cells display characteristics of circulating cells in blood, spleen and bone marrow but can adopt a resident phenotype in lymph nodes and mucosal tissues. This distribution suggests that NK cells play a role in controlling viral infection and tumor progression in tissues accessible to circulation, while intestinal and secondary lymphoid sites may be less conducive to NK cell-mediated immune surveillance (55). Moreover, the expression profile of NK cells differs significantly across tissues. A recent study on mouse utilizing scRNA-seq and flow cytometry identified gene expression patterns in NK1.1⁺NKp46⁺ cells across various tissues, revealing substantial differences in the proportion of marker-positive cells in blood, spleen, inguinal lymph nodes, CD49α− liver, intestinal epithelial lymphocytes, intestinal lamina propria lymphocytes, liver, salivary gland, and uterus (42). In blood and other tissue sites, there is a bias towards the expression of GZMB in CD56^dim cells. Compared to CD56^dim NK cells in lymph nodes and intestines, the frequency of GZMB⁺ in CD56^dim NK cells in the lungs is significantly higher (55). Similarly, the transcriptional characteristics, phenotype, and function of ILCs are inherently influenced by their tissue microenvironment (80). Furthermore, the expression of transcription factors in trNK cells varies between species. For instance, trNK cells in livers of human and pigs express high levels of EOMES and low levels of TBox transcription factor 21 (TBX21), while trNK cells in mouse livers express high levels of TBX21 and low levels of EOMES (81). Tissue-resident NK cells exhibit significant differences in gene expression compared to conventional NK cells, and there is also substantial heterogeneity in receptor expression. ( Table 1 ) These findings highlight the role of the tissue microenvironment in shaping the distinct expressions and heterogeneity of NK cells.

scRNA-seq enables the detection of heterogeneity between cells, allowing for a more detailed classification of NK cells that is not achievable with bulk sequencing methods. ( Table 3 ) Recent studies have utilized scRNA-seq to analyze the pan-cancer landscape of NK cells. Tang F et al. identified five subgroups of CD56^brightCD16^low NK cells and nine subgroups of CD56^dimCD16^high NK cells and compared them with adjacent non-tumor tissues. The study revealed reduced expression of the cytokines CCL3 and CCL4 in the c3-CCL3 NK cells of tumor tissues and decreased expression of XCL1 and XCL2 in the c5-CREM NK cells, indicating their functional transition within tumors (100). Xu L et al. characterized the heterogeneity of NK cells in the breast TME under single-cell analysis, categorizing NK-0 to NK-5 as “memory-like” NK cells, NK cells with reduced IFN-γ production, CD56^dim bone marrow NK cells, tissue-resident NK cells, activated NK cells involved in direct anti-tumor responses, and NK cell activity inhibitors (101). Song G et al. sequenced NK cells in HBV/HCV-related hepatocellular carcinoma and non-tumor liver tissue, and divided them into six subgroups (NK_c1, 2, 3, 4, 5, 6). Based on differential gene expression, they defined an terminal NK subgroup (NK_c1), an exhausted NK subgroup (NK_c4), and two undefined NK subgroups (NK_c2 and NK_c6) and noted that NK_c5 was absent in the non-HBV population and played an additional antiviral role (44). Liu H et al. used single-cell sequencing to compare NK cells in liver cancer with NK cells in normal liver tissue and found that the L3-NK-HLA and L4-LrNK-FCGR3A subgroups were absent in tumor tissues, while L2 and L5 subgroups were significantly increased. They observed high expression of inhibitory receptor genes TIGIT, NKG2A, TIM3, and CD96 in L1-NK-CD56^bright cell, TIM3 and CD96 in L2-NK-CD56^dim cells, and KLRC1 and TIGIT in L5-LrNK-XCL1 cells. Furthermore, common tumor-related genes (RHOB, TALDO1, TKT, and HLA-DPA1) are overexpressed in L5-LrNK-XCL1 cells, exhibiting inhibitory anti-tumor activity (87). A study re-clustered all NK cells from 40 tumor samples and adjacent normal tissues in non-small cell lung cancer (NSCLC) patients into 12 subgroups (c0-c11), showing differential expression of cytotoxic markers and varying enrichment in lung squamous cell carcinoma(LUSC) and lung adenocarcinoma(LUAD) (88). Another study conducted cluster analysis on NK cells in triple-negative breast cancer (TNBC), HER2⁺ and luminal-like breast cancer and divided them into six groups. They defined two cytotoxic NK cell subgroups and two exhausted NK cell subgroups as well as two NKT cell groups. The study found that TNBC had a higher proportion of cytotoxic NK cells and an increased expression level of “cytotoxic” marker genes (GZMA, GZMB, and CST7) in NK_c03_Cytotoxic_GZMB while HER2⁺ and luminal-like breast cancer lost this feature, which provided new insights for NK cell immunotherapy (89). Additionally, de Andrade LF et al. sequenced NK cell population in melanoma metastases and defined seven NK cell subgroups. They then categorized them into proliferating NK cells, high cytotoxic NK cells and low cytotoxic NK cells based on their expression (90).

Additionally, single-cell sequencing technology can also identify new subgroups, thereby laying the foundation for a deeper understanding of tumor microenvironment changes. As previously mentioned, Tang F et al. (100) identified a c6-DNAJB1 subgroup (TaNK cells) specifically enriched in tumors. This subgroup exhibited significantly higher expression of transcripts such as KLF6 and EGR3, which are associated with the inhibition of cytotoxic function and inferred to represent a terminal state based on RNA velocity analysis. A high abundance of TaNK cells in tumors is associated with poor prognosis in cancer patients. Ding S et al. (89) observed a unique subgroup of SOCS3 ^highCD11b^-CD27^- immature NK cells that were found only in TNBC samples. These cells exhibited reduced cytotoxic granule characteristics and were responsible for activating cancer cells through Wnt signaling pathway in mice, thereby promoting tumor progression. Another study revealed that peripheral circulating NK cells differentiate along two distinct pathways within the tumor microenvironment, leading to different final states. One of these pathways results in a phenotype similar to intraepithelial ILC1s, which represents the NK cell phenotype with the highest anti-tumor activity (91). Furthermore, Lin Q et al. (92) discovered a CD8⁺ NK cell population specific to the TME of nasopharyngeal carcinoma, providing new insights into the complexity of the immune landscape in nasopharyngeal carcinoma.

Single-cell RNA sequencing has the capability to generate single-cell level maps of TME, thereby providing the most comprehensive composition and transcriptomic information of immune cell populations. By utilizing scRNA-seq, researchers can summarize and analyze the differences between tumor and normal tissue, the diversity of TME maps across different subtypes of the same tumor, and the variations within different regions of tumors and at various stages of tumor progression.

Xiao J et al. (102) utilized spatial and single-cell transcriptomics to uncover the tumor heterogeneity of colorectal cancer, revealing that the proportion of NK cells and B cells in tumors was reduced by approximately 0.3-0.4 times compared to normal tissue. A study on melanoma showed that cell clusters with shared key genes and similar transcriptional characteristics had significantly different proportions in tumors. Tumor-infiltrating NK cells exhibited significantly higher expression of FOS and JUN compared to blood NK cells, and their expression of the KLRC1 gene (encoding the inhibitory receptor NKG2A protein) also elevated (90). In lung cancer, the c5 cell cluster enriched in lung cancer tissues expressed resting NK cell markers XCL1, KLRC1 and AREG, while the c9 cell cluster expressing CX3CR1 and NKG7 was abundant in normal tissue (88). In lung adenocarcinoma lesions, NK cells were the most abundant immune cell lineage compared to non-affected lung tissue (nLung), with the CD16⁺ NK cell subset showing the most significant reduction. NK cells infiltrating tumor lesions expressed higher levels of CXCR3, which is necessary for NK cell infiltration, while lower levels of GZMB and CD57, and reduced IFN-γ production, indicated their decreased cytolytic activity (99). Furthermore, overall exhaustion of NK cells was observed in neuroblastoma and lung adenocarcinoma metastasis, indicating that the tumor immune response shifts towards immunosuppression (94, 95).

Additionally, related studies have also focused on the differences in TME maps between different subtypes of the same tumor category. Studies have demonstrated significant differences in the distribution and expression of NK cells in various types of lung cancer, including LUAD and LUSC. Specifically, c0 and c1 NK cell clusters were abundant in LUAD tissues, while c2 and c4 NK cell clusters were enriched in LUSC tissues (88). De Zuani M et al. also observed reduced cytotoxicity of NK cells in LUAD and LUSC, with significant differences in the co-expression of various immune checkpoint inhibitors (96). A study on human lung adenocarcinoma analyzed two distinct microenvironmental patterns formed by the transcriptomes of heterogeneous cancer cells. The inert N3MC microenvironment, compared to the immunologically activated CP2E microenvironment, contained normal-like myofibroblasts, non-inflammatory monocyte-derived macrophages, NK cells, bone marrow-derived dendritic cells, and conventional T cells, and was associated with a favorable prognosis (83). A study on mouse gliomas revealed increased infiltration of CD4⁺ T cells, CD8⁺ T cells and NK cells in low-grade gliomas (LGGs), while high-grade gliomas (HGGs) lacked this infiltration (98). Conversely, the abundance of neutrophils, T cells, and NK cells in samples from human high-grade gliomas showed an increasing trend compared to low-grade gliomas (103), suggesting that this trend may be related to the staging and malignancy of tumor progression and is independent of species differences. Ding S et al. (104) analyzed tumor samples from patients with TNBC and non-TNBC, finding that HER2⁺ and luminal-like breast cancer exhibited significantly lower levels of cytotoxic NK cells compared to TNBC, indicating a pro-tumor environment of NK cell exhaustion, and suggesting that these patients may benefit more from NK-based immunotherapy.

Furthermore, the enrichment and expression of NK cell subpopulations vary across different spatial areas within tumor tissue. A study examining the immune microenvironment of cervical squamous cell carcinoma (CSCC) revealed that different tumor areas exhibited varying degrees of cell infiltration. High-metabolic tumor areas displayed stronger signals for CD56⁺ NK cells and immature dendritic cells, while low-metabolic tumor areas showed stronger signals for eosinophils, immature B cells, and Treg cells (97). A single-cell study on nasopharyngeal carcinoma by Chen YP et al. (105) demonstrated that immune cells, particularly CD57⁺ NK cells, CD20⁺ B cells and IL3RA⁺ plasmacytoid DCs (pDCs), were more abundant in the tumor stroma compared to tumor epithelial nests.

What’s more, the level of NK cell infiltration and expression is dynamic and changes with tumor progression, recurrence, and metastasis. Studies have shown that chromosomal instability (CIN) in the precancerous stage can activate atypical NF-κB signaling, leading to anti-tumor immunity by NK cells. However, when CIN exceeds a certain threshold, the functional capacity of NK cells is compromised, and their effector capabilities are diminished compared to tumors with low CIN, thus creating a permissive microenvironment for tumor growth (104). Zhu J et al. characterized the cellular map of lung adenocarcinoma invasion trajectories through integrated analysis of single-cell RNA sequencing and spatial transcriptomics. The study revealed that cancer cells, NK cells, and mast cells were co-located in the cancer area in adenocarcinoma in situ (AIS), and NK cells and B cells increased during the progression from AIS to minimally invasive adenocarcinoma (MIA), enriching the study of cell ecology and spatial niches involved in the dynamic and sequential processes from AIS to MIA and subsequent invasive adenocarcinoma (IAC) (106). Tumor recurrence and metastasis are associated with shifts in the TME. In the study of chordomas, it was observed that NKT and NK cells had higher scores in recurrent chordomas compared to primary chordomas, and NK cells exhibited higher exhaustion scores in primary chordomas, providing a more comprehensive understanding of the microenvironment of primary and recurrent chordomas (107). Deng JY et al. analyzed the transcriptomic characteristics of metastatic NSCLC and found that the lung lesion group was enriched in NK cells and NK-mediated cytotoxicity and T-NK cell receptor signaling pathways, with NK cells, helper T cells, IFN-γ-related T cells, and proliferating T cells exhibiting higher cytotoxic activity than that of liver metastasis group (108). Similarly, the number and proportion of immune cells with tumor immune function, such as CD8⁺ T cells, NK cells, and monocytes, were significantly reduced in prostate cancer lymph node metastases compared to primary prostate cancer lesions (109). Gene mutation subtypes also influence cellular infiltration and transcriptional characteristics. BRAF mutant melanomas showed a significant reduction in the infiltration of CD8⁺ T cells and macrophages compared to BRAF wild-type melanomas, but an increase in the infiltration of NK cells, B cells, and NKT cells (110).

The dynamic and heterogeneous nature of the tumor microenvironment contributes to insufficient infiltration and diminished efficacy of NK cell immunotherapy, which is a significant factor in tumor progression. Understanding the interactions between NK cells and other cells in the TME provides new insights for optimizing NK cell immunotherapy. Single-cell sequencing technology presents complex interactions of NK cells in a higher dimension, providing crucial insights for the selection of NK cell immune checkpoints and the development of drug intervention strategies.

Firstly, interactions between NK cells and tumor cells can inhibit the cytotoxic function of NK cells, enabling tumor cells to evade immune surveillance and promoting tumor recurrence and metastasis. ( Figure 5 ) NKG2A, an inhibitory receptor on NK cells, acts as an immune checkpoint receptor. Jiang H et al. (111) utilized single-cell sequencing technology to reveal that NK cells exhibited the highest expression of NKG2A, with over half of NK cells in TME expressing KLRC1. Malignant cells interact with NK cells through the HLA-E-NKG2A pathway, which presents new opportunities for the clinical treatment of gastric cancer. In the study of pancreatic ductal adenocarcinoma circulating tumor cells (CTCs), Liu X et al. also demonstrated that CTCs interact with NK cells via the immune checkpoint molecule HLA-E-NKG2A, protecting CTCs from NK-mediated immune surveillance and thereby promoting the metastasis of pancreatic ductal adenocarcinoma (112). CD39, a transmembrane protein encoded by the ENTPD1 gene, in synergy with CD73, can convert ATP and ADP into the immunosuppressive adenosine through a cascade reaction. Liu L et al. (113) found that the expression of CD39 and LAG3 of NK cells was significantly upregulated in most tumor samples, creating an immunosuppressive state in the TME. Cao G et al. (114) conducted cell interaction analysis to investigate the potential mechanisms underlying T/NK cell exhaustion and identified poliovirus receptor (PVR)-like protein signaling as the primary co-inhibitory interaction between NK cells (CD96 and TIGIT) and malignant cells (PVR/CD155 and NECTIN1). Recent single-cell RNA sequencing data suggests that high expression of platelet-related genes in circulating tumor cells CTC directly upregulates CD155 expression on tumor cells through platelet adhesion. This upregulation protects CTCs from NK cell killing through the interaction between CD155 and the immune receptor TIGIT, thereby promoting the metastatic dissemination of tumor cells (115). The functional state of NK cells in the TME is significantly suppressed, while unsuppressed NK cells can exert cytotoxic effects and resist tumors. A study using scRNA-seq to analyze various immune cells within tumors found that NK cells in tumor sites expressed cytotoxic molecules such as GZMA, PRF1, and chemokines XCL2, CCL5, CCL3, and CCL4, indicating the potential for an antitumor response. Additionally, NK cells also expressed several inhibitory and co-stimulatory molecules, including CD96, TNFRSF18, and KIR2DL4, representing targets for regulating their function (116).

NK cells can also interact with immune cells, forming a complex immune network that together constructs an immunosuppressive tumor microenvironment. The study of NK cell-immune cell interactions provides foundational information for deepening the understanding of the immune network and immunotherapy. Research has found a strong negative correlation between the frequency of STAB1⁺ Mφ/AI Mφ and T/NK cells, further validating that Mφ acquire an immunosuppressive phenotype and inhibit NK cell infiltration in non-small cell lung cancer (96). Using single-cell transcriptomic analysis, Kim HJ et al. (117) found that CD169⁺ macrophages in human and mouse gliomas produce pro-inflammatory chemokines which can lead to the accumulation of NK cells, and NK cell-derived IFN-γ is crucial for the accumulation of CD169⁺ macrophages derived from blood monocytes in gliomas. In a single-cell sequencing study of primary tumors and adjacent non-tumor samples from different organs or tissues and metastases, Jiang H et al. (111) found that when macrophages and NK cells were co-cultured, treatment with an anti-NKG2A antibody inhibited HLA-KLRC1 (NKG2A) signaling and the expression of IFN-γ was significantly improved. Additionally, researchers showed through scRNA-seq of glioblastoma tumor-infiltrating immune cells that the loss of Dicer in macrophages reduced the proportion of cluster 3 of glioma-associated macrophage (GAM) and reprogrammed the remaining GAM to a pro-inflammatory activation state, interacting positively with tumor-infiltrating T cells and NK cells through TNF paracrine signaling, creating a pro-inflammatory immune microenvironment for tumor cell elimination (112). A study compared receptor-ligand pairs in LUAD and LUSC to evaluate the interaction strength between immune cell subsets in the tumor microenvironment and found that the SPP1 pathway showed greater interaction between SPP1-Mφ clusters and NK cells in LUSC than in LUAD, and further analysis revealed different pathways in LUSC, such as TIGIT and cell proliferation-related LCK, which can enhance the interaction between NK cells and SELENOP-Mφ together with SPP1-Mφ clusters (88). Similarly, the interactions between NK cells and other immune cells, including cancer-associated fibroblasts (CAF) and dendritic cells, have been thoroughly investigated. A study deconvoluted spatial transcriptomics at the single-cell level and performed functional analysis to generate a high-resolution integrated map of breast cancer. The study found that co-culture of FAP⁺ CAF clusters with NK cells reduced the levels of perforin and GZMB in NK cells and significantly increased the proportion of CD56^high NK cells and their surface NKG2A levels, thereby reducing the cytotoxicity of NK cells (118). Additionally, Ou Z et al. used single-nucleus RNA sequencing (snRNA-seq) and spatial enhanced resolution omics sequencing to study the immune microenvironment of CSCC and found that the number of B cells, T cells, NK cells, neutrophils, DCs and Th1 cells was significantly reduced in myCAF⁺ tumors, proving that myCAF may act as a physical barrier to prevent the infiltration of pro-immune cells (97). What’s more, mature cDCs (LAMP3⁺ DCs) which are recently characterized showed the strongest interaction with CD56^dim NK cells. Tumor-infiltrating LAMP3⁺ DCs showed lower IL-15 expression compared to those in non-tumor tissues, and NK cells physically close to LAMP3⁺ DCs expressed lower levels of GZMB, suggesting that LAMP3⁺ DCs have a compromised activation effect on CD56^dim NK cells in TME (100).

Immune checkpoint blockade (ICB) can disrupt the immune surveillance of cancer cells and has significantly transformed the landscape of cancer treatment (131). A research conducted single-cell RNA and T cell receptor sequencing on PBMCs from 60 patients with stage IV non-small cell lung cancer and divided NK cells into CD56^high NK cells (c1), CD56^low NK cells (c2), and CD56^low NK_IFNG cells (c3). The results revealed that after immune checkpoint inhibitor (ICI) treatment, the proportion of the c1 subgroup increased, and in the durable clinical benefit (DCB) group after ICI treatment, the proportion of the c3 subgroup also increased (132). Another study on esophageal squamous cell carcinoma analyzed the dynamic changes in the tumor immune microenvironment and found that immune checkpoint blockade enhanced the activation and cytotoxicity of NK cells by increasing the frequency of CD56^dimCD16⁺ NK cells (133). Shang J et al. (134)analyzed single-cell transcriptomic data from patients with non-small cell lung cancer treated with anti-PD-1and they observed six different NK cell clusters, among which immature NK cells were enriched in the responder group and expressed a series of marker genes associated with anti-PD-1 response and were related to antigen processing, Th1 activation, Th17 cell activation and other immune regulatory processes. Additionally, scRNA sequencing also demonstrated that microbial-derived interferon gene stimulators (STING) agonists can regulate macrophage polarization, induce IFN-I production from intratumoral monocytes, and facilitate crosstalk between NK cells and DCs. The microbiome triggers the IFN-I-NK cell-DC axis within the tumor, improving the efficacy of ICB (135).

The combination of ICB therapy with other treatment modalities has yielded promising results. Relatlimab combined with nivolumab (anti-LAG-3 and anti-PD-1) has been approved by the FDA as a first-line treatment for stage III/IV melanoma. Utilizing single-cell RNA and TCR sequencing in conjunction with other multi-omics analyses, researchers found that adaptive NK cells were enriched in responders after treatment and underwent significant transcriptomic changes, leading to an activated phenotype (136). Single-cell sequencing compared the transcriptomes of approximately 92,000 single cells from three pre-treatment and twelve post-treatment NSCLC patients who received neoadjuvant PD-1 blockade combined with chemotherapy and researchers concluded that the neoadjuvant PD-1 blockade combined with chemotherapy therapy can promote the expansion and activation of cytotoxic T cells and CD16⁺ NK cells (137). Additionally, a study on a new type of high-grade serous ovarian cancer (HGSC) compared samples treated with a bispecific anti-PD-1/PD-L1 antibody to a control group treated with monospecific anti-PD-1 or anti-PD-L1. The results demonstrated that the bispecific antibody induced superior cellular state changes in T cells and NK cells, uniquely prompting NK cells to transition from a quiescent phenotype to an active and cytotoxic phenotype. This transition enhanced the efficacy of immunotherapy for ovarian cancer (138). Another study identified 207 liver cancer NK cell marker genes based on single-cell sequencing, which are primarily involved in immune functions and found that the degree of immune cell infiltration and function in the low-risk group were higher than those in the high-risk group, indicating that the low-risk group is more suitable for ICI and anti-PD-1 treatment (139).

Some single-cell sequencing studies have shown that NK cells in the TME after chemotherapy show increased infiltration and enhanced killing efficacy, indicating good anti-tumor effects. A study conducted whole exome sequencing, large-scale RNA sequencing, and single-cell transcriptomics on paired pretreatment and treatment samples from gastric cancer patients and explained the remodeling of the TME after platinum chemotherapy, characterized by the recruitment of NK cells, the reduction of tumor-associated macrophages, the repolarization of M1 macrophages, and the increase in effector T cell infiltration (140). The scramblase Xkr8 plays a key role in promoting tumor immunosuppression and the use of Xkr8 small interfering RNA (siRNA) targeting Xkr8 combined with the chemotherapeutic prodrug FuOXP in the TME of in situ pancreatic tumors caused an increase in the infiltration of proliferative NK cells and activated macrophages which could enhance the antitumor immune response and showing good antitumor effects (141). Additionally, a study based on high-dimensional cell counting deeply analyzed the effects and mechanisms of 2,5-dimethyl-celecoxib (DMC) on the infiltration of immune cells in liver cancer. The study found that after using DMC, NK cells and T cells in the TME showed strong anti-tumor activity and effectively inhibited the exhaustion of NK cells and T cells, thereby inhibiting the growth of liver cancer cells and improving the prognosis of mice (142).

The application of certain chemical agents has also been observed to affect NK cells through single-cell sequencing technology, providing insights into clinical efficacy and guiding treatment strategies. Wu S et al. concluded from single-cell sequencing in a mouse B16F10 melanoma model that docosahexaenoic acid (DHA) can enhance the effector function and oxidative phosphorylation of NK cells, effectively inhibiting the growth of B16F10 melanoma in the lungs (143). Yi Q et al. analyzed pancreatic cancer liver metastasis (PCLM) models with or without α-galactosylceramide (αGC) treatment, characterizing the overall changes in immune cells in the TME after αGC treatment. They found that the cytotoxic activity of invariant natural killer T cells (iNKT) and NK cells increased (119). Li CY et al. studied the immune mechanisms of the anti-tumor effects of low-dose nitric oxide donors in three ways, among which scRNA sequencing showed that low-dose nitric oxide donors increased the number of CD8⁺ T cells and NK cells while reducing central memory T cells (144). Through single-cell RNA sequencing, Liu L et al. found that sodium polyoxotungstate inhibited the increase of NK cells within tumors. They also performed basic analysis of NK cell phenotypes through flow cytometry, revealing that treatment with CD39i significantly increased the proportion of mature CD27⁺CD11b⁺ and CD27^-CD11b⁺ NK cells, confirming that CD39 is a potential target for immunotherapy in bladder cancer (113).

Additionally, some treatment strategies using oncolytic viruses have led the way in biological treatment concepts. For example, in a mouse model treated with the oncolytic virus OH2, Wu Q et al. (120) found that OH2 could enhance the cytotoxic capabilities of peripheral CD8⁺ T cells and mature NK cells, effectively activate the systemic immune response and induce a sustained anti-tumor immune response.

Non-pharmacological interventions, such as radiation and microwave ablation therapy, have also been better explored for their efficacy through single-cell RNA sequencing. Jiao JZ et al. investigated the impact of non-invasive radiofrequency radiation (RFR) exposure on the tumor immunosuppressive microenvironment. ScRNA-seq analysis of infiltrating immune cells in lung metastatic melanoma revealed that RFR exposure could increase NK cell subsets, enhance the effector functions of tumor-infiltrating CD8⁺ T cells and NK cells, induce anti-tumor remodeling of TME and thereby inhibit tumor progression (121). Another study conducted scRNA-seq on PBMCs before and after microwave ablation (MWA) for breast cancer. The study found that XCL2⁺ NK cells specifically enriched after MWA were associated with genes related to cytokine production pathways, lymphocyte differentiation pathways and T cell activation pathways and presented higher chemokine function. While in GZMB⁺ NK cells, IFITM1, MYOM2, JUN, IFITM3, CCL4, and RBM25 were upregulated, indicating higher cytotoxic activity (145).