Natural Killer (NK) cells are Innate Lymphoid Cells (ILCs) playing a crucial role in the anti-tumor and antiviral immunity (1–4). They carry out their function both by influencing and supporting the activity of different immune cell types (including myeloid cells and T lymphocytes), and by directly killing the target cells (5–8). Through the release of chemoattractant factors (CCL3, CCL4, CCL5, HMGB1) and cytokines (IFN-γ, TNFα, and GM-CSF), they can favor recruitment of immune cells, including NK cells (9, 10), and promote Th1-type inflammatory responses, while through the expression of cell death receptors and, especially, the release of cytotoxic granules they exert a potent lytic activity against tumor and virally infected cells. The activation of the different NK cell functions is orchestrated by the cytokine milieu and, more specifically, it depends on the type of cellular interactions involving NK cells. The cell-to-cell contact with normal cells, certain immune cells, or altered cells can give rise to inhibitory, regulatory, or lytic immune synapses respectively (11), with different patterns of engaged NK receptors, including: inhibitory, activating, or cytokine receptors. In different species, the discrimination between normal cells and altered tumor or virally infected cells is carried out through the action of MHC-I-specific inhibitory receptors, and activating receptors recognizing stress-induced molecules or ectopically expressed antigens. In humans, major inhibitory receptors are represented by certain Killer Ig-like Receptors (KIRs) and the CD94:NKG2A heterodimer, while the most known tumor-recognizing receptors include the Natural Cytotoxicity Receptors (NCRs) (NKp46, NKp30, and NKp44), NKG2D, and DNAM-1 (12, 13). Remarkably, the coordinated engagement of some inhibitory and activating receptors, together with cytokine receptors are involved in complex regulatory cross-talks with immune cells.

This picture of NK cell function and mechanics is the result of the many studies that have been carried out in vitro over the 90s and across the millennium to dissect single molecular pathways and cellular interactions. These results were then impressively expanded with the advent of new global and in-depth cell analytical approaches, such as the “omics” and “single cell RNA-seq”. Application of these techniques led to the identification and fine molecular characterization of new functional NK cell subsets, as naïve or adaptive NK cells, or tumor-associated NK cells (14–16). Despite such important technical advances have significantly extended the power of human in vitro and ex-vivo studies, the still open issues on NK cell differentiation, circulation, homing, tissue residency, and tumor penetration cannot disregard the option of the animal studies.

Along the timeline of the NK cell studies, the utilization of murine models proved crucial for the initial evaluation of the real anti-tumor effects of NK cells and gave the first molecular hints on the role of MHC-I recognition (17–20). Additional studies provided supportive information on how NK cells differentiate, acquire and regulate their cytolytic potential (through the “licensing/arming” process), and, even more strikingly, persist in vivo as memory-like cells (21–24). These new insights are providing important hints for the selection and preparation of optimized NK cell effectors for anti-tumor immuno-therapies, with promising results in the field of hematological malignancies. On the other hand, the successful exploitation of NK cells in the cure of solid tumors is still limited by the active role of the local tumor microenvironment (TME), which hinders infiltration and effector capabilities of NK cells (25, 26). Several in vitro studies have contributed to the fine dissection of the different molecular mechanisms underlying the suppressive/escape properties of the TME. Nevertheless, the murine models are providing the crucial answers on how each of these mechanisms, and its targeting, could impact on the real anti-tumor efficacy of NK cells within the complexity of the TME. Given the importance of this issue, many efforts are being spent to generate ever more effective models to study solid tumors, moving from the full murine to the human-in-mice or humanized models, trying to overcome the not-always obvious problems related to the inter-species differences.

Murine models are well-characterized systems useful to gain insights into human biology and various pathological conditions, including cancer. Indeed, through the development of specific strains with stable and well-known genetic background, the murine system offers defined and reproducible experimental conditions without losing the complexity of living organisms.

The initial studies to identify and molecularly characterize the anti-tumor functions of NK cells in vivo were performed on mice injected with murine malignant cells. By the use of a C57BL strain and its “beige” variant, characterized by reduced NK cell activity related to the beige-J spontaneous mutation Lystbg-J, it was demonstrated the active role of NK cells in contrasting tumor growth (18, 19). Then, experiments based on the injection of H-2 mismatched or H-2 negative tumor cells in mice bearing given H-2 haplotypes provided important information on the role of the MHC-I-specific NK receptors (27, 28). Finally, the development of induced or spontaneous tumorigenesis mouse models combined with the generation of specific gene silencing (KO mice) were crucial to demonstrate the key role of certain activating NK receptors in vivo. Thus, for instance, in the model of transgenic adenocarcinoma of the mouse prostate (TRAMP) the targeting of the klrk1 gene (TRAMP-NKG2D-KO mice) resulted in higher tumor incidence and the development of larger tumors (29). Similarly, the inactivation of the cd226 gene (DNAM-1-KO mice) favoured the growth of the (3-MethylColanthrene (MCA)-induced fibrosarcoma in BALB/c mice (30). Finally, the inactivation of the Ncr1 gene affected the ability of C57BL/6 mice to control melanoma and lung carcinoma metastases, or lymphoma induced by tumor cell line injection (31, 32).

The “mouse in mouse” studies provided important hints for the characterization of the role of human NK cells in the context of the solid tumors, as many murine NK cell receptors presented human homologues, and homologies between human and mice also emerged studying the development of NK cells, the regulation of their functional properties, and the acquisition of memory-like features. However, although these models have been exceedingly important in advancing our knowledge, there are limits and functional differences that have to be accounted for, especially, considering the translation of the obtained data to the design of new immunotherapy strategies (33). In this context, it has become ever more important the development of models enabling the study of human environments in the mouse.

The growth of human tissues and cells in a different species (xenotransplant) requires evasion from the immune system to prevent rejection. Xenotransplants became possible after the discovery, in 1962, of the nu/nu spontaneous mutation (34). Nu/nu Nude mice lacked the thymus, showed T-cell deficiency and impaired B-cell functionality, but retained macrophages and NK cells. Another spontaneous mutation affecting the immune system was discovered in 1983 in C.B-17 mice. Such autosomal recessive mutation, named Prkdcscid, affects the gene encoding for the protein kinase, DNA activated, catalytic polypeptide (PRKDC) and gives rise to severe combined immunodeficiency (SCID) in the mice. SCID mice (i.e. homozygous Prkd^scid mutants) lacked both T and B cells and had reduced NK cells (35), however their phenotype was “leaky” and clones of functioning B and T cells could randomly develop in young animals (36). Altered immunity was also observed in nonobese diabetic (NOD) mice (37, 38). These mice were obtained by a combination of inbreeding and selective breeding from a progeny of an outbred Jcl:ICR mouse, and were initially selected as a tool to study autoimmune diabetes, given their genetic predisposition to the development of the disease. However, NOD mice were also characterized by defects in the innate immunity, including the NK cell compartment, and were then considered for the preparation of new immunocompromised models. In particular, by crossing C.B-17 Prkdc^scid males with NOD/ShiLtSz females, it was generated the NOD-SCID mouse strain. Transfer of the SCID mutation onto a non-obese diabetic (NOD) background eliminated the leaky phenotype of the SCID model and generated the NOD-SCID mouse strain that rapidly became the best choice for transplant studies using freshly isolated human tumour cells, selected cancer stem cells, or tumour fragments (39, 40). NOD-SCID mice lack T and B lymphocytes and display reduced NK cell function, resulting in increased xenotransplant engraftment. However, they have a limited lifespan (7-8 months), due to the frequent spontaneous development of thymic lymphomas (41). Despite their limits, nu/nu, SCID and NOD-SCID mice have represented the most used strains for xenotransplant of human tumour cell lines in the last 50 years. A great improvement in the rate of human cell engraftment was then obtained with the backcrossing of NOD-SCID mice with either truncated or deleted interleukin-2 receptor common gamma chain giving origin to the NOG (42) or NSG (43, 44) strains, respectively. Specifically, in the early 2000s the NOD.Cg-Prkdc^scidIl2rg^tm1Sug (NOG) strain was developed at the Central Institute for Experimental Animals, Japan, by backcrossing NOD/ShiJic-Prkdc^scid mice with IL2rg^tm1Sug/ShiJic mice, carrying a truncation of the intracellular signalling domain of the IL2 receptor gamma chain. The NOD-SCID gamma NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/Szj (NSG) strain, instead, a brand from the Jackson Laboratories (USA), was developed by backcrossing NOD-SCID mice with B6.129S4-IL2rg^tmWjl/J (IL2Rγ^null) carrying the knock-out of the IL2 receptor gamma chain. NSG and NOG mice lack mature T and B cells, NK cells, and complement, have defective macrophages and dendritic cells (DCs) and gave a great impulse to the development of humanized mice models. In particular, these mice represent elective recipients for the functional studies on circulating tumor cells.

Cancer tissue represents a quite complex biological system, with its peculiar microenvironment and heterogeneity of the malignant and non-malignant components (45, 46). Therefore, the strategies to generate human tumor tissues in immunocompromised mice were generally conceived with the aim of reproducing at best such original complexity. A valid approach in this field has been the setting up of Patient Derived Tumour Xenografts (PDX), i.e. the engraftment of small pieces of freshly collected tumour samples in immunocompromised mice. PDXs warrant the conservation of the original tumor components (47–49) and, for that reason, are widely used for maintenance and therapy response experiments. In one study on lung cancer PDX, it has also been shown that lymphocytes present in the transplanted tumor (tumor infiltrating lymphocytes - TIL) could partly reconstitute the immune system in the mouse. Remarkably, in these mice, the combined treatment with IL-15 and PD-1 inhibition could induce tumor regression, which was dependent on human (tumor-derived) NK and T cells (50). Therefore, the model could provide important information on the interaction of the immune system with the tumor in the autologous setting. The use of this model, however, appear to be limited by the possible variability of the TIL component in different engrafted tumor fragments. Moreover, it should be considered that murine stroma can replace human one after two passages in the mice, and a cross-talk between the human tumor component and murine tumor-associated cells can easily take place in these models.

Another approach comes from the observation that certain human tumor cell populations, especially those containing a cancer stem cell component, can generate malignant lesions, partly recapitulating the original tumor complexity, once injected into immunocompromised mice. Following this observation, several human-in-mice tumor models have been reproduced and studied through the injection of selected tumor cells in NOD-SCID or NSG mice. An interesting example is represented by the attempt to reproduce a bone metastasis of human breast cancer or NSCLC in a human bone fragment that has been subcutaneously implanted in SCID mice (51, 52). In the case of NSCLC, it has been possible to identify a subset of human NSCLC cells endowed with metastasis-initiating-cell properties. Those cells, which were marked by the CD133⁺CXCR4⁺EpCAM^- phenotype, were shown to colonize the bone implant and to generate tumor lesions.

The immunocompromised mice are optimal recipients for the generation and development of human tumours, but their altered, non-human, immune system represents an issue. Indeed, cancer is a dynamic disease, showing plasticity in its cellular and extracellular components, and its dynamics is often influenced by the interaction with the host and, signally, by the selective pressure of the immune system (46). Therefore, the lack of the human immune system in these murine models represents a limit, both for the attainment/maintenance of reliable tumor units, and for the studies focused on the development of immunotherapy strategies.

For this reason, there has been a flourishing interest in the “humanized mice models”: immunocompromised mice transplanted with hematopoietic precursors and/or lymphoid tissues capable of reproducing in the mice several types of human immune cells.

Engraftment of human lymphoid cells in SCID mice was first reported in 1988 by Mosier and co-workers, who injected intraperitoneally human peripheral blood mononuclear cells (PBMC) from EBV-positive donors and obtained “stable long-term reconstitution of a functional human immune system” (53). However, only human T and B cells could be satisfactorily maintained in these mice, while engraftment of other hematopoietic lineages was not effective. An additional problem was represented by the development in the animals of xenogeneic graft-versus-host-disease (GvHD), which usually took place 4-6 weeks after transfer of huPBMC, thus shortening the available observation time (54, 55). Nevertheless, given its relatively easy preparation, this model has been used over time, and, recently, the generation of NSG mice carrying a deletion of the MHC-I or II genes has been proposed to overcome the problem of xenogeneic GvHD (56).

Encouraging results were also obtained by transferring CD34⁺ human hematopoietic stem cells (HSC) from cord blood into new-born BALB/c Rag2^-/-γc^-/- mice by intrahepatic injection. However, the reconstitution level remained low (57). The use of NOD-SCID mice carrying the deletion of the Rag-1 or Rag-2 gene did not change significantly the outcome (58). In these models, T cell function was limited, probably due the absence of human lymphoid organs supporting T cell maturation, selection, and activation. To overcome this problem, in a new study, CD34⁺ HSC were transferred into NOD-SCID mice that were previously transplanted with human foetal liver and thymic tissue in the renal capsule (hu-bone marrow-liver-thymus - BLT) (59, 60). hu-BLT mouse model exhibits a full reconstitution of human immune cell repertoire, with T cells capable of mounting HLA-II- and HLA-I-restricted specific responses. These models were initially prepared to assess the immune response to viral infections (namely HIV), and were then adapted for studies of onco-immunology (61). Their employment for the study of NK cells in the tumor, however, appeared to be limited by the fact that BLT-mice developed relative low numbers of, poorly functional, NK cells.

The advent of the next generation murine models is now providing specific tools to study restricted human immune cell types in selected human tumors. These models are generated by using “knock in” and “knock out” strategies in immunocompromised mouse background, and can reach considerable complexity (62). Table 1 summarizes the main models that can be used for the study of human NK cells in vivo. An important issue that is generally considered in the setting of humanized mice regards murine cytokines, which may be poorly expressed in immunocompromised mice and, due to the evolutionary divergence, quite different from their human counterparts. Therefore, several models have been developed expressing human cytokines, generally in NOG or NSG background, with the aim to improve development or persistence of given human immune cells, including NK cells (Table 1).

Katano et al. established transgenic NOG mice sub-strains expressing either human IL-2 (NOG-hIL-2 Tg) or human IL-15 (NOG-hIL-15 Tg) (63, 64). In NOG-hIL-2 Tg mice, human cord blood-derived HSCs could mature and give rise to differentiated KIR⁺ NK cells expressing IL-2–activated phenotype; while in NOG-hIL-15 Tg mice, the high levels of IL-15 in the blood could support survival of NK cells from transferred human PBMC. hIL-15 has also been expressed in NSG, and the strain NSG-IL-15 Tg is now commercially available (65). Recently, to improve the development of human NK cells, Matsuda et al. created hIL-7 and hIL-15 double knock-in (hIL-7xhIL-15 KI) NSG mice (66). Compared to NSG mice, these mice showed increased ability to develop human NK cells after engraftment of human hematopoietic stem cells. A hIL-15 and human signal regulatory protein alpha (hSIRPα) double knock-in mouse on a Rag2^-/- il2rg^-/- background (SRG-15) was described to support efficient development of innate lymphoid cell subsets and NK cells, including tissue resident cells (69, 72). Even more complex knock in mice have been generated that express different combinations of human factors such as IL-3, IL-15, GM-CSF, M-CSF, SIRPα, and thrombopoietin: NSG-SGM3, MISTRG and BRGSF (see Table 1). On the whole, these latter strains support the development of human myeloid cells, which in turn can produce IL-15 and improve development of functional human NK cells (67, 68, 70, 71). Finally, a potentially interesting humanized model is also represented by the Nonirradiated NOD.B6.SCID Il2rg^-/- kit^w41/w41 (NBSGW) mouse, which was obtained by the crossing of NSG strain with the C57BL/6.-Kit^W-41J/J strain (73). This model has the interesting advantage of supporting the human HSC engraftment without the need for pre-transplant conditioning (i.e. γ-irradiation), which could negatively affect the host microenvironment for the HSC development (74). An ovarian cancer PDX has been now developed in NBSGW mice, engrafted with human cord blood-derived HSC, to study the tumor immune environment within the peritoneal cavity (75). To date, such NBSGW model has not been used yet to study human NK cells.

Some of the above-mentioned next generation humanized mice have been assayed in different xenogeneic solid tumor models to study various immune cell types, especially T cells and myeloid cells, and only rarely NK cells (Table 2). Indeed, as yet, the large majority of the studies on NK cells have been done on “traditional” NOG or NSG mice. Nevertheless, the strategy of transferring human NK cells in such immunocompromised mice proved effective to evaluate which NK cell preparation, or way of cell activation, would be more suitable to overcome the tumor escape mechanisms and to deliver efficient NK cells into the tumor niches.

As mentioned above, the human-in-mice tumor models, combined with human NK cell transfer, represent a suitable mean to assess the real effects of NK cells at the tumor site. Through these studies, it is possible to gain knowledge on spatial localization of NK cells within the tumor, and understand their impact on stroma, tumor cells and other immune cell populations. Likewise, it is possible to characterize phenotype and function of tumor-infiltrating NK cells, and, also, to gain information on their persistence. Therefore, on the whole, these models offer a platform to test possible approaches to optimize survival, function, and safety of adoptively transferred human NK cells. As described below, the animal model receiving human NK cells is generally represented by the NSG mice.

The increasingly sophisticated analytic tools for the in-vivo and ex-vivo characterization of tumors are improving clinical decision making, and also, provide means for conducting research directly in the patients. These new approaches add support to the preclinical research but do not replace its fundamental phases, which comprise the studies in vitro to dissect molecular and cellular processes, and the validation studies in vivo.

The “traditional” 2D in vitro cultures address important questions on specific biological, genetic and epigenetic features and on the direct cellular effects of drugs or cytokines, while animal models are amply used to evaluate the actual significance of the dissected processes and to test therapeutic efficacy of drugs and dose-limiting toxicity of clinical treatments. On the other hand, the animal studies are generally expensive, time-consuming, and, for years, they have been based on a limited range of models (147, 148). Furthermore, there are ethical issues, which, for the animal welfare, limit the use of animals to what is strictly necessary. The principle of the 3Rs (Replacement, Reduction, and Refinement) is presently applied in all projects involving the use of animals. In particular, the Replacement is supported by the development and use of predictive and robust models and tools, based on the latest in vitro technologies, to address important scientific questions (149).

In this context, the setting of the 3D cultures has been proposed as an important tool for the development of innovative in vitro assays and models of neoplastic cell growth with potentially high clinical relevance. By the setting of 3D models, it is tempting to partly reproduce the complexity of the TME, in terms of extra-cellular matrix, cell types, vasculature, and oxygen, nutrients and catabolites distribution. If fully realized, these 3D tumor models would be a powerful mean to study a number of tumor-related processes, to validate specific findings, and to test the efficacy/toxicity of drug treatments, replacing the employment of animals (150, 151). This goal, however, is presently far to be achieved, as it is not easy to set a reliable balance among all the different fundamental components within a single in vitro-generated structure.

Therefore, the employment of animals remains indispensable for the preclinical studies, especially considering the recent development of the next generation humanized mouse models (65, 76–79, 81–89) (Table 2). Indeed, these models can reproduce ever more accurately the complexity of the immune response within the TME, and also offer the interesting perspective of dissecting in vivo crucial frames of such complexity. In this context, these models also represent suitable tools to study the behaviour of human NK cells. In conclusion, a powerful strategy to efficiently struggle against cancer should consider the wise integration of the different available tools to gain precise data on the single malignancies, to dissect molecular and cellular pathways, and to evaluate new related therapeutic strategies (Figure 1). In this scenario, in which different strategies are connected, the animal models maintain a pivotal role.

MP, SA, MV, and PO contributed to conception and the outline of the review article. All authors contributed to the article and approved the submitted version.