Background

Non-small cell lung cancer (NSCLC) accounts for approximately 80% of all lung malignancies and is diagnosed at an advanced stage in nearly 70% of patients. The introduction of targeted therapies and immunotherapies, involving immune checkpoint inhibitors, has improved the prognosis in some cases, yet has proved poorly effective against metastatic spread.1 These facts pose the control of metastases as the major therapeutic concern for this type of cancer.

Natural killer (NK) cells may represent a potential therapeutic tool in this context. Indeed, they are endowed with potent antitumor activity, and it has also been suggested that they can kill cancer stem cells (CSC), which are accountable for tumor initiation and dissemination.2–4 Moreover, the mechanisms driving their capability to recognize and kill tumor cells have been in large part defined5–8 providing useful hints for the development of new therapeutic strategies. Nevertheless, the issue of how to get effective and durable NK cell-based immunotherapies is still the object of intense investigation, particularly in solid tumor settings and metastases. Major challenges concern the identification of new strategies that can enhance and drive NK cell-cytotoxicity into the tumor nests, overcome tumor-induced immune suppression, and prolong NK cell activity and persistence in vivo.3 9 10

Cytokine-induced memory-like NK cells (CIML-NK) have emerged as a promising tool to address some of these concerns. CIML-NK cells can be easily induced from peripheral blood (PB)-NK cells after brief exposure to the cytokine cocktail, interleukin (IL)-12+IL-15+IL-18,11 and can recall the initial cytokine stimulus by showing enhanced responses, particularly interferon (IFN)-γ release, to either cytokines or tumor cells.12–14 Thus, although IL-2 still represents a commonly used cytokine to activate ex vivo or in vivo adoptively transferred NK cells in clinical trials, the clinical relevance of CIML-NK cells is now under examination. A first in-human study conducted in Acute myeloid leukemia (AML) patients has provided evidence that allogenic CIML-NK cells can differentiate, proliferate, and persist in vivo, showing antitumor effects in more than 50% of patients.15 By a multidimensional analytic approach, the study also characterized CIML-NK cells from patients, highlighting that this cell population is heterogeneous, with only a fraction of cells responding to tumor cells. This observation poses the issue of whether and how CIML-NK effectors could be further optimized, especially considering their possible exploitation in the context of solid tumors and metastases.

Recent studies have reported that CIML-NK cells can effectively attack melanoma, head and neck, colon, and ovarian cancer cell lines both in vitro and after administration to human tumor-bearing NSG-mouse models,16–19 suggesting that different solid tumors could be targeted by CIML-NK cells. However, no information is presently available on the possible role of these cells in controlling CSCs and metastasis development, which is crucial for a concrete step forward in the cure of solid tumors. CSCs can be delivered into the pre-metastatic niches as single cells or through multicellular three-dimensional (3D) structures that leave the primary tumor and circulate as tumor-cell clusters.20–22 Therefore, in vitro generated, or tumor-derived CSC-enriched spheroids could be a valuable platform to test enhanced NK cell functions.23 24 In this regard, therapeutic antibodies or bi/tri-specific molecules, simultaneously engaging key NK receptors and tumor-expressed epitopes represent potential tools to address enhanced NK responses to specific targets.25–27 Their potential in targeting CSC, however, remains poorly investigated. In NSCLC, CSCs are included within the CD133+ tumor cell population, which displays elevated tumorigenicity and is considered responsible for therapeutic resistance.28 29 Therefore, a recently generated tri-specific cell engager (1615133 TriKE)30 targeting CD133 could be particularly appropriate to support CIML-NK cell activity against NSCLC CSCs.

In this study, we show that CIML-NK cells target NSCLC CSCs and highlight, for the first time, the unique role of their CD56bright cell subset and of a specific TriKE in maximizing the anti-CSC effect of both healthy-derived and patient-derived CIML-NK cells.

Methods

See online supplemental material and methods for details.

Cell lines and tumor spheroids

NSCLC cell lines A549, H3122, H661, and SW900 were cultured under standard cell culture conditions and were regularly tested for Mycoplasma contamination. Tumor spheroids were generated as previously described.23 Briefly, tumor cells were cultured in DMEM/F-12 medium supplemented with B-27 without vitamin A, 20 ng/mL epidermal growth factor, 20 ng/mL basic fibroblast growth factor, 5 µg/mL heparin in ultra-low attachment 6-well or 96-well microplates.

Surface phenotypical analysis

NK and tumor cells were phenotypically characterized by flow cytometry. Details of all antibodies used are found in online supplemental table S2.

Sorting of tumor-cell or NK-cell subsets

Tumor cells or lymphocytes were incubated with viability dye for 20 min, washed, and stained with fluorescently-labeled antibodies. Viable CD133+ cancer cells or CD3−CD19−CD56bright/dim lymphocytes were sorted on an FACSymphony S6 system (BD Biosciences).

CD107a and IFN-γ intracellular staining

NK:tumor cell co-cultures (1:1 ratio) were performed for 6 hours in the presence of anti-CD107a antibody, with the addition of brefeldin A and monensin after the first hour. After 6 hours, cells were stained for viability and surface markers. Next, cells were fixed, permeabilized, and stained for intracellular markers. Where indicated, 1615133 TriKE was added to the co-cultures 10 min before the anti-CD107a antibody.30

Killing assays

Flow cytometry-based killing assays were performed by co-culturing for 24 hours NK cells with Carboxyfluorescein succinimidyl ester (CFSE)-labeled target cells at the indicated Effector (E):Target (T) ratios. After co-culture, cells were stained with viability dye (dead cell marker, DCM) and annexin V. Dead target cells were defined as CSFE+ annexin V+ and/or DCM+.

For calcein-based killing assays, target cells were labeled for 30 min with 5 µg/mL calcein, washed, and co-cultured with effector cells in 96-well U-bottom plates in assay medium (red-phenol-free RPMI, 1% FBS, 2.5 mM probenecid). After 4 hours, supernatants were collected, and fluorescence emission was measured using a plate reader.31

IncuCyte measurement of spheroid killing

Real-time spheroid killing was evaluated using the IncuCyte S5 platform (Essen Bioscience, Sartorius). NK cells were added to wells containing a single-spheroid. Tumor cell apoptosis within spheroids was evaluated by including caspase 3/7 green detection dye in the media. Images were acquired using a 4× objective lens and analyzed by IncuCyte Controller V.2020A. The means of the technical replicates (n=3) for each condition were compiled for N=3 donors.

Colony forming assay

Tumor spheroids were co-cultured with NK cells for 24 hours. Then, CD45-depleted viable tumor cells were seeded in 6-well plates (500 cells/well). Colonies were stained with 0.5% crystal violet. After image acquisition, crystal violet was dissolved (10% acetic acid) for absorbance quantification using a BioTek microplate reader.

Patient-derived xenograft and patient-derived tumor cells

For NSCLC patient-derived xenograft (PDX) establishment, small tumor fragments from surgical specimens were implanted subcutaneously (s.c.) in the flanks of SCID mice, as previously described32 and PDX were maintained by serial s.c. passages of fragments of growing tumors into mice. When indicated, tumor cells derived either from PDX or directly from patients’ surgical specimens were analyzed in vitro in functional assays.

In vivo tumorigenicity studies

A549 spheroids 6–8-week male NSG mice were housed at the University of Minnesota Animal Facility. Spheroids from A549 cells were cultured alone or co-cultured with the indicated NK cell types for 24 hours (1:1 E:T ratio). After spheroid dissociation, and NK cell removal, A549 cells were assessed for purity and vitality by flow cytometry. 105 viable A549 cells, resuspended in 100 µL phosphate-buffered saline (PBS)+Matrigel (1:1 v/v), were subcutaneously inoculated to the mouse flank (n=6 mice/group).

PDX 7–10-week female SCID mice were housed at the INT (Milan) animal facility. PDX-derived single-cell suspensions were co-cultured with the indicated NK cell types for 4 hours (1:1 E:T ratio). After NK cell removal by the use of anti-CD45-microbeads and their immunomagnetic depletion with AutoMACS (all from Miltenyi), and purity and vitality assessment, 105 viable tumor cells, resuspended in PBS+Matrigel (1:1 v/v), were subcutaneously inoculated to the mouse flank. At endpoint, mice were euthanized and lungs were collected for immunohistochemistry (IHC) analysis or processed for flow cytometry analysis.28 For in vivo evaluation of CIML-NK cell function, PDX were directly implanted into the mice. After 4 weeks 5×106 CIML-NK cells were injected intravenous and 2 days after mice were sacrificed to analyze lungs (n=5 mice/group).

ResultsIdentification of the CD56bright cell subset as the major driver of CIML-NK cell functions

Following well-established protocols, we generated CIML-NK cells from the PB of healthy donors (HD) by stimulating freshly isolated NK cells for 16 hours with IL-15+IL-12+IL-18, followed by a 7-day resting period in the presence of low-dose IL-15. Afterward, we assessed the cells by cytofluorimetry for the expression of informative markers to identify possible characterizing cell subsets and evaluate their contribution to the functional CIML-NK cell features. As a comparison, NK cells were also analyzed before stimulation, at day 0, or after parallel 7-day culture in the presence of IL-2 (IL-2-NK), still used for clinical grade preparations, or low dose IL-15, as classical control of CIML-NK cells (IL-15 control cells: IL-15(c)-NK) (online supplemental figure S1). This analysis revealed that CIML-NK cells were characterized by an important expansion of the CD56brightCD16dim (hereafter CD56bright) cell subset, which was significantly larger than that observed in IL-2-NK and IL-15(c)-NK cells (figure 1A,B), reaching in one case 60% of total CIML-NK cells. To confirm that such expansion originated from “bona fide” CD56bright cells, we sorted naïve CD56bright and CD56dimCD16bright (hereafter CD56dim) cells, fluorescently labeled the CD56bright fraction, rejoined the two sorted subsets to reconstitute original PB-NK cell population, and induced CIML differentiation (online supplemental figure S2). This experiment showed that, indeed, real (labeled) CD56bright NK cells could significantly expand after CIML differentiation (up to 20-fold size increase) (figure 1C). In additional experiments, both CD56bright and CD56dim sorted cells were labeled and induced to CIML differentiation as separated populations to evaluate proliferation. Consistently, CD56bright cells showed a significantly higher proliferation rate compared with CD56dim cells (figure 1D). In agreement with these results, the majority of CIML-NK cells expressed an NKG2A+KIR– phenotype (figure 1E,F), which typically characterizes CD56bright and a fraction of less mature CD56dim cells (online supplemental figure S3). It is noteworthy that in CIML-NK cells, the NKG2A–KIR+ cell population poorly proliferated and nearly disappeared during culture, indicating that these terminally differentiated cells minimally participate in the generation of CIML-NK cells (figure 1E–G and online supplemental figure S3). Differential proliferation of the CD56bright/CD56dim or KIR+/KIR– cell subsets could be also confirmed in CIML-NK cells induced from unsorted cells (online supplemental figure S4).

Figure 1

Characterization of freshly isolated NK (D0), IL-15(c)-NK, IL-2-NK and CIML-NK cells by flow cytometry. (A) Representative plots of CD56 versus CD16 expression showing the gating of CD56bright and CD56dim subsets. (B) Percentage of CD56bright cells in the different types of NK cells. Bars show mean±SD, and dots connected by lines represent percentages from each individual donor (n=11 donors). (C) Expansion of CD56bright cells in CIML-NK cells assessed at day 6 following cytokine priming. Prior to memory-like differentiation, naïve CD56bright and CD56dim cells were sorted, CD56bright were labeled with CellTrace Violet (CTV) and pooled with unlabeled CD56dim cells. The percentage of (CTV+) CD56bright cells was assessed by flow cytometric analysis before (day 0) and after (day 6) CIML differentiation (C, left). Representative flow cytometry plots and (C, right) graph displaying the percentage of CTV+ cells from each donor (individual dots) at days 0 and 6 (n=3 donors). (D) Proliferation of CIML-NK cells assessed at day 6 following cytokine priming. The assessment was done by flow cytometric analysis of CTV dilution on CIML derived from sorted CD56bright (red) or CD56dim (blue) cells (D, left). Representative flow cytometry plots and (D, right) violin plot summarizing proliferation data of CD56bright and CD56dim CIML-NK cells at day 6 quantified as division index (n=3 donors). (E, F) Characterization of the NK cell subsets as identified by the combined expression NKG2A and KIRs (Mix KIR) in the different indicated cell types. (E) Representative flow cytometry plots of NKG2A versus KIR expression and (F) bar graphs displaying the mean percentage of the subsets±SD (n=6 donors). (G) Proliferation of NKG2A+KIR– (red), NKG2A+ KIR+ (green), and NKG2A–KIR+ (violet) in CIML-NK generated from sorted CD56dim cells assessed by flow cytometric analysis of CTV dilution (G, left). Representative flow cytometry plots and (G, right) violin plot showing the division index in each subset at day 6 (n=3 donors). Groups were compared using a one-way repeated measures analysis of variance with post hoc Tukey test (B, F, G) and paired two-tailed t-test (C, D) (⁎p<0.05; ⁎⁎p<0.01; ⁎⁎⁎p<0.001). CIML, cytokine-induced memory-like; IL, interleukin; NK, natural killer.

We next characterized in more detail the expanded CIML CD56bright cells, to evaluate their potential contribution to the “CIML-functional traits”. Specifically, we analyzed the expression levels of major activating, inhibitory, and chemokine receptors in the different NK cell subsets (online supplemental figure S5). This analysis indicated that CD56bright NK cells undergoing CIML differentiation and expansion acquired a favorable antitumor functional profile, characterized by increased expression of certain activating receptors (including NKp44), reduced TIGIT and TIM-3 inhibitory checkpoints, and a slight increase of the chemokine receptors CXCR3 and CXCR4. To check the potential functional advantage of CIML-NK cells, we evaluated their response to NSCLC cell lines derived from tumors of the three main subtypes: adenocarcinoma (A549 and H3122), squamous cell carcinoma (SW900), large cell carcinoma (H661). Overall, CIML-NK cells showed enhanced IFN-γ production, compared with IL-15(c)-NK cells, and higher cytotoxic degranulation over both IL-15(c) and IL-2-NK cells (figure 2A,B). Moreover, the combined phenotypical and functional analysis suggested that the CD56bright subset could be the main driver of CIML-NK cell responses (online supplemental figure S6A). Consistent with this observation, in most of the effectors analyzed in functional experiments CIML-NK cells showed maximal CD56bright expansion when compared with IL-15(c)-NK or IL-2-NK cells (online supplemental figure S7). To confirm these latter findings, CIML-NK cells were generated from naïve CD56dim or CD56bright sorted cells, or from pooled populations after distinctive labeling of CD56bright cells, and assessed for their IFN-γ, degranulation, and cytotoxic response to NSCLC targets. These experiments demonstrated that, indeed, CD56bright cells were more responsive than CD56dim cells (figure 2C–E and S6B,C) and supported most of the CIML-NK cytotoxic effect against NSCLC cells (figure 2F). CD56bright cells showed higher responses also in IL-2-NK and IL-15(c)-NK cells (online supplemental figure S6A), with the remarkable difference that in these latter cases the CD56bright cell subset was generally less expanded (figure 1A,B). Moreover, at variance with IL-2-CD56bright and IL-15(c)-CD56bright cells, CIML-CD56bright NK cells showed a high granzyme B content (figure 2G), which could support their effective killing capabilities.

Figure 2

Functional characterization of CIML, IL-15(c), and IL-2-NK cells against NSCLC cell lines, evaluation of the CD56bright and CD56dim CIML subsets. (A–B) The different NK cell types were co-cultured with the indicated NSCLC cell lines (E:T ratio of 1:1) for 6 hours, and the production of IFN-γ or the expression of CD107a on NK cells (gated as CD56+ cells) was measured by flow cytometry. Data are shown as box and whisker plots with median±minimum to maximum of n=5–6 donors. (C–E) CIML-NK cells were generated from pooled labeled (CTV+) CD56bright and unlabeled (CTVneg) CD56dim naïve NK cells, and then cultured with or without the indicated cell lines for functional assessments. (C) Representative flow cytometry plots showing: (left) the gating of CTV+ (red) and CTVneg (blue) cells in day 6 CIML-NK cells, and (right) the CD107a versus IFN-γ expression on the gated subsets of CIML-NK cells in the different co-culture conditions. (D,E) Cumulative data on the frequencies of IFN-γ+ or CD107a+ cells in CD56bright (CTV+, red) and CD56dim (CTVneg, blue) subsets of unstimulated/stimulated CIML NK cells (n=3 donors). (F) Calcein-based cytotoxicity of SW900 cells on co-culture with CIML-NK cells differentiated from sorted CD56bright (red), sorted CD56dim (blue), or unsorted (black) NK cells. The graph displays the mean cytotoxicity±SD of n=4 donors at different E:T ratios performed in technical duplicates. (G) Flow cytometry analysis of granzyme B expression in the total (CD56+), CD56bright, and CD56dim NK cell populations of CIML-NK, IL15(c)-NK, and IL-2-NK cells. Bars display the mean percentage of granzyme B-positive cells of n=4 donors. Dots represent the percentage of positive cells in each individual donor. Comparisons between groups were performed using one-way ANOVA with Tukey post hoc test (A, B, G) or a two-way ANOVA with Sidak post test (D–F). Only significant values are shown (⁎p<0.05; ⁎⁎p<0.01; ⁎⁎⁎p<0.001). ANOVA, analysis of variance; CIML, cytokine-induced memory-like; CTV, CellTrace Violet; Effector, (E): Target, (T); IFN, interferon; IL, interleukin; NK, natural killer; NSCLC, non-small cell lung cancer.

Overall, these experiments indicate that CIML-NK cells have a superior functional response over IL-2-NK cells against NSCLC cells of different tumor subtypes, with the CD56bright subset serving as the main driver of CIML effector functions.

CIML-NK cells show superior ability to attack NSCLC spheroids

Tumor cells can disseminate through circulation as cellular clusters, therefore we generated spheroids from the above NSCLC cell lines following a previously described protocol23 and assessed NK cell reactivity. We co-cultured NK cells in functional assays with either spheroid-derived cells or 3D spheroids (online supplemental figure S8A,B), to dissect the functional features of dissociated tumor cells and the general effects of the organized 3D structures. Since CIML-NK cells can upregulate the expression of the IL-2Rα subunit,33 we also investigated the possible additional effect of overnight exposure to low-dose IL-2 (50 IU/mL). The CD56bright/CD56dim cell subset composition of this latter effector was similar to that of CIML-NK cells (online supplemental figure S7).

The functional response of the different NK cell preparations, evaluated as cytotoxic degranulation and IFN-γ production, significantly decreased when target cells were switched from adherent to spheroid-derived cells (online supplemental figure S9). Nevertheless, CIML-NK cells displayed superior functional responses to these targets over IL-2-NK and IL-15(c)-NK cells (figure 3A,B). Moreover, CIML-NK cells’ responsiveness was further increased by IL-2 boost, particularly in terms of IFN-γ production. NK cells were also tested in a cytotoxicity assay against A549 and SW900 cells (the lowest and highest NK cell stimulators, respectively), confirming that, indeed, CIML-NK cells could exert superior killing activity of spheroid-derived cells over IL-2-NK and IL-15(c)-NK cells (figure 3C,D).

Figure 3

Functional characterization of CIML, IL-15(c) and IL-2 NK cells against non-small cell lung cancer spheroids. (A–D) Evaluation of NK cell functional response to spheroid-derived cells. (A, B) CIML-NK, IL-15(c)-NK, IL-2-NK cells and CIML-NK cells boosted with IL-2 (CIML+IL-2) were co-cultured for 6 hours with H3122, A549, H661 or SW900 spheroid-derived cells. Afterward, the expression of IFN-γ and CD107a was measured on NK cells by flow cytometric analysis. Data are shown as box a whisker plots with median±minimum to maximum, dots represent the value of each single donor. Number of NK cell donors analyzed, n=6 (SW900, H661), n=5 (H3122), n=4 (A549). (C, D) Evaluation of NK cell cytotoxicity against spheroid-derived cells. The specific lysis of CFSE-labeled SW900 or A549 spheroid-derived cells was measured after 24 hours of co-culture with the indicated NK cell types at different E:T ratios. After co-culture, target cells were stained with dead cell marker and annexin V, and assessed by flow cytometry gating on CFSE-positive events. (E–H) Evaluation of NK cell cytotoxicity in the context of cell aggregates (spheroids). SW900 and A549 spheroids were co-cultured with the different NK cells (E:T ratio 8:1) in the presence of caspase 3–7 green dye in a 96-well plates. Wells were imaged every hour for 48 hours to quantify the caspase 3–7 activation (green signal) within the spheroids using the IncuCyte system. (E, G) Representative images of SW900 and A549 spheroids co-cultured with the different NK cell types displaying caspase activation (at time=48 hours), and quantification of caspase activation (green mean intensity) normalized to control in spheroid boundary. Graphs show mean±SEM of n=3 donors ran in triplicates. (F, H) Visualization of target-specific caspase 3–7 activity at the indicated time points of the co-culture as bar graphs showing mean+SEM. Groups were compared using a one-way (A–B) or two-way analysis of variance (C–H). (⁎p<0.05; ⁎⁎p<0.01; ⁎⁎⁎p<0.001). Carboxyfluorescein succinimidyl ester, (CFSE); CIML, cytokine-induced memory-like; Effector, (E): Target, (T); IFN, interferon; IL, interleukin; NK, natural killer.

We next assessed the effects of NK cells directly on spheroids (one spheroid/well online supplemental figure S8B—). Spheroids from A549 and SW900 cells were loaded with caspase 3–7 substrate, co-cultured with NK cells, and evaluated by real-time live cell imaging (IncuCyte). This analysis revealed that NK cells could kill tumor cells even when they are aggregated in 3D spheroids. More importantly, CIML-NK cells, especially when supplemented with IL-2, showed higher killing efficiency over IL-15(c)-NK and IL-2-NK cells. Remarkably, CIML-NK cells boosted with IL-2 showed persistent killing capabilities even after 36–48 hours of co-culture (figure 3E–H), suggesting that these effectors could minimize tumor-related exhaustion and suppression effects.

CIML-NK cells decrease the CSC content and tumorigenicity of 3D tumor cell clusters

Given their superior response to spheroids, CIML-NK cells may be effective in attacking CD133+ CSC, which have been shown to be enriched in NSCLC spheroids.23 To assess this possibility, we determined the percentage of viable CD133+ cells within A549-derived spheroids before and after co-culture with CIML-NK IL-2-NL, IL15(c)-NK cells (Control (CTRL)-spheroids, CIML-spheroids, IL-2-spheroids, IL-15(c)-spheroids, respectively) (online supplemental figure S8C). We observed that CIML-NK cells were effective at reducing the CD133+ CSC content in spheroids, and such capability of CIML-NK cells could not be maintained after the IL-2 boost (figure 4A). Remarkably, neither IL-2-NK nor IL-15(c)-NK cells could reduce the fraction of CD133+ cells within spheroids.

Figure 4

CIML-NK cells target spheroid and PDX CD133+ cells and limit tumorigenicity and tumor dissemination in vivo. (A–C) Analysis of the effect of NK cells on spheroids. A549 spheroids were co-cultured for 24 hours at a 1:1 E:T ratio with CIML-NK, IL15(c)-NK, IL-2-NK cells, or CIML-NK cells boosted with IL-2 (CIML+IL-2), or were cultured alone (CTRL). Next, spheroids were dissociated, NK cells were removed by depleting CD45+ cells (using CD45-specific microbeads) and recovered tumor cells were analyzed for viability and CD133 expression, and assayed for clonogenicity, in vitro, and tumorigenicity, in vivo. (A) Flow cytometry assessment of CD133+ percentage in the different co-cultures, reported as fold-change to control untreated spheroids. Data are shown as box and whisker plots with median±5–95 percentiles (n=3 donors). (B) In vitro clonogenicity assay of spheroid-derived cells. Viable tumor cells from dissociated spheroids were appropriately diluted and seeded in 6-well plates to follow colony formation (B, top). Image of a representative experiment. Wells were stained with crystal violet for clone evaluation (B, bottom). Bar graphs showing the mean normalized relative absorbances of the crystal violet dissolved with 10% acetic acid of n=3 donors ran in technical triplicates. (C) In vivo tumorigenicity of spheroid-derived tumor cells. Viable tumor cells from dissociated spheroids were s.c inoculated to NSG mice, and tumor formation and growth was followed over time. Bar graph shows the mean maximum tumor volume±SD reached per group (n=6 mice per group). (D–H) Analysis of the effect of NK cells on PDX (patient LT710). PDX cells were co-cultured for 4 hours with the different NK cell types or cultured alone (CTRL). Next, NK cells were removed from co-cultures by depleting CD45+ cells and tumor cells were analyzed for tumorigenicity in vivo. (D) Assessment of in vivo tumorigenicity of PDX cells from the indicated cultures. Mean group tumor growth curve+SD are shown (n=4 mice/group). (E–H) Evaluation of tumor cell dissemination to the lungs from subcutaneous PDXs. Subcutaneous tumors were generated from PDX cells from the indicated cultures. (E) Percentage of lung disseminating tumor cells (DTC) assessed by flow cytometry. DTC were identified as viable-H2K− (non-murine) cells. (F) Percentage of CD133+ cells within the lung DTCs assessed by flow cytometry. Data are shown as box and whisker plots with median±minimum to maximum values (n=4 mice per group). (G) Representative images of IHC staining to evaluate the presence of CK+metastatic nodules (at higher magnification in the insets) in lung tissue sections. (H) Graph showing the number of CK+lung metastatic foci from each mouse (dots) and the mean foci number (line) per group. (I–K) Effect of CIML-NK cells in vivo. 5×106 CIML-NK cells were inoculated intravenous in near end-point PDX-bearing mice. 48 hours after, mice were sacrificed, lungs and primary s.c tumors were dissociated, and analyzed. (I) Percentage of lung DTC assessed by flow cytometry. (J) Percentage of CD133+ DTC and (K) percentage of CD133+ tumor cells in s.c tumors evaluated by flow cytometry. Data are shown as box and whisker plots with median±minimum to maximum values (n=6 mice per group ran in technical duplicates). (L) Heatmap showing the expression levels of major activating and inhibitory NK-cell ligands measured by flow cytometry on CD133+ and CD133neg PDX cells. (H) Flow cytometry analysis of the surface expression of LFA-1 in IL-15(c), IL-2-NK, and CIML-NK cells. Bars report mean MFI±SD and dots represent MFI values of individual donors (n=3 donors). Comparisons between groups were assessed using one-way ANOVA with Dunnett’s multiple comparison test (A, C), Turkey post hoc test (B, E, F, H–K, L) or two-way ANOVA (D). Only significant values are shown (⁎p<0.05; ⁎⁎p<0.01; ⁎⁎⁎p<0.001). ANOVA, analysis of variance; CIML, cytokine-induced memory-like; Effector, (E):Target, (T); IL, interleukin; Immunohistochemistry, (IHC); Mean Fluorescence Intensity, (MFI); NK, natural killer; PDX, patient-derived xenograft.

We next analyzed whether CIML-NK cells, consistent with their effect on CD133+ cells, could impact spheroid tumorigenicity. Thus, viable tumor cells were isolated from spheroid-NK cell co-cultures and assayed for clonogenicity in vitro, or evaluated for in vivo tumorigenic potential (online supplemental figure S8C). As shown in figure 4B, tumor cells from spheroids that were co-cultured with CIML-NK cells (either boosted or not with IL-2) formed significantly fewer colonies than those derived from CTRL-spheroids, IL-2-spheroids, or IL-15(c)-spheroids. For the in vivo studies, viable A549 cells from CTRL-spheroids or sorted from co-cultures were subcutaneously inoculated in NSG mice, and in situ tumor formation was monitored. Mice inoculated with tumor cells from CIML-spheroids displayed lower tumor incidence compared with mice receiving equal numbers of tumor cells from CTRL-spheroids, IL-2-spheroids, or IL-15(c)-spheroids (online supplemental figure S10). Moreover, tumor cells from CIML-spheroids, and not those from IL-15(c)-spheroids or IL-2-spheroids, generated significantly smaller tumors compared with those induced by CTRL-spheroids (figure 4C). On the whole, these data indicate that CIML-NK cells affect clonogenicity and tumorigenicity by targeting CD133+ cells. However, IL-2-boosted CIML-NK cells, which do not affect the CD133+ cell population, may target different clonogenic tumor cell subsets, with less pronounced in vivo effects (ie, no statistical significance in the tumor volume reduction).

CIML-NK cells target disseminated CSC in PDX models

To investigate in more detail the effects of CIML-NK cells boosted or not with IL-2 on CD133+ NSCLC cells, and to extend the study to patient-derived tumor cells, we analyzed in co-culture experiments four different targets: A549 and SW900 spheroids, and cells derived from two NSCLC PDX models (PDX-LT710 and PDX-LT111) (see extended M&M section). These experiments indicated that CIML-NK cells consistently reduced the CD133+ cell fraction in all the analyzed targets, whereas IL-2-boosted CIML showed variable behavior (online supplemental figure S11).

We then focused on CIML-NK cells and asked whether they could be effective in influencing PDX capability to support dissemination driven by CSCs. We first analyzed PDX-LT710. On co-culture, CIML-NK cells, besides reducing the PDX CD133+ cell content, also showed higher cytotoxic degranulation and IFN-γ production over IL-15(c)-NK and IL-2-NK cells (online supplemental figure S12). Accordingly, after co-culture with CIML-NK cells, PDX cells displayed reduced tumorigenic capability, generating significantly smaller subcutaneous tumors (CIML-PDX) compared with those induced by PDX cells cultured alone (CTRL-PDX) or in the presenc