Article Text
PDF
Abstract
Background Epithelial to mesenchymal transition (EMT) endows cancer cells with pro-metastatic properties, which appear most effective when cells enter an intermediate hybrid (H) state, characterized by integrated mesenchymal (M) and epithelial (E) traits. The reasons for this advantage are poorly known and, especially, it is totally unexplored whether the interplay between H-cells and NK cells could have a role. Here we characterize the pro-metastatic mechanics of non-small cell lung cancer (NSCLC) H-cells and their subset of cancer-initiating cells (CICs), dissecting crucial interactions with NK cells.
Methods Human lung cancer cell lines and sublines representative of E, M, or H states, assessed by proteomics, were analyzed in vivo for their tumor-forming and disseminating capabilities. Interactions with NK cells were investigated in vitro using migration assays, cytotoxic degranulation assays, and evaluation of CD133+ CICs modulation after coculture, and validated in vivo through NK cell neutralization assays. Correlation between EMT status, NK cell infiltration, and survival data, was evaluated in a cohort of surgically resected NSCLC cases (n=79).
Results We demonstrated that H-cells, have limited dissemination capability but show the highest potential to initiate metastases in vivo. This property was related to their ability to escape NK cell surveillance. Mechanistically, H-cells expressed low levels of NK-attracting chemokines (CXCL1 and CXCL8), generating poorly infiltrated metastases. Accordingly, proteomics and GO enrichment analysis of E, H, M cell lines showed that the related secretory processes could change during EMT.
Furthermore, H-CICs uniquely expressed high levels of the inhibitory ligand B7-H3, which protected H-CIC from NK cell-mediated clearance. In vivo neutralization assays confirmed that, indeed, the pro-metastatic properties of H-cells are poorly controlled by NK cells.
Finally, the analysis of patients revealed that detection of hybrid phenotypes associated with low NK infiltration in NSCLC clinical specimens could identify a subset of patients with poor prognosis.
Conclusions Our study demonstrates that H-cells play a central role in the metastatic spread in NSCLC. Such pro-metastatic advantage of H-cells is supported by their altered interaction with NK cells and by the critical role of B7-H3 in preserving their H-CIC component, indicating B7-H3 as a potential target in combined NK-based therapies.
- Killer Cells, Natural
- Non-Small Cell Lung Cancer
- Tumor Escape
- Receptors, Antigen
- Immunity, Innate
Data availability statement
Data are available in a public, open access repository. Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Cancer cells can reside in different phenotypic states along the axis of Epithelial-Mesenchymal Transition (EMT) with relevant implications for metastatic spread, but little is known about the mechanisms regulating interactions with immune cells.
WHAT THIS STUDY ADDS
Here we show in vitro and in vivo that a hybrid EMT state confers the highest metastatic potential to lung cancer cells through increased seeding potential and escape from Natural Killer cells clearance, exerted by lower release of attractant chemokines and by the up-regulation of B7-H3 inhibitory ligand by hybrid cancer initiating cells (CICs). In clinical specimens the combination of hybrid tumors and poor NK infiltration correlates with worst outcomes.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study highlights the relevance of integrating information on EMT states and NK infiltration for early identification of tumors with poor prognosis and provides mechanistic bases for novel potential therapeutic strategies
Background
Despite the widening of therapeutic options, non-small cell lung cancer (NSCLC) remains the most lethal cancer worldwide, mostly because of late diagnosis at metastatic phase.1 2 Therefore, targeting metastases represents an urgent need to obtain significant therapeutic advances for this disease.
The process of metastatic dissemination is driven by several complex events which comprise the premetastatic niche formation, the release of cancer-initiating cells (CICs) from the primary tumor, their seeding into the niches, and the development of metastasis.3 Several factors can influence these events and the consequent successful establishment of secondary lesions.4 Thus, the release of tumor-derived factors inducing premetastatic niches, the specific microenvironment in the primary and secondary tumors, the tumor cell heterogeneity and the efficiency of the immune surveillance represent important variables to be studied in deep as possible therapeutic targets.
CICs are characterized by stemness traits and functional plasticity5 and can be identified by specific phenotypes and by their tumor initiation ability on transfer in animals.6 Specifically, in NSCLC we have shown that the CD133+ cell subset was enriched for CIC properties.7 The epigenetic-induced plasticity of tumor cells has also recently come into play as pro-metastatic driver, influencing the CIC generation.8 9 Indeed, tumor cells can acquire pro-invasive and stemness properties through processes of phenotype switching based on epithelial to mesenchymal transition (EMT).5 8 Indeed, the acquisition of mesenchymal (M) traits by malignant cells has been associated with poor prognosis in some tumor types.10 11 In particular, in NSCLC a consistent fraction of tumor cells often displays the mesenchymal phenotype and therefore EMT has been proposed as a useful biomarker for therapeutic decision.12 On the other hand, although necessary for the dissemination of primary tumor cells, the strict requirement of a fully mesenchymal state for successful establishment of metastases is still debated.13 14 Indeed, the emerging question is to define which cells between those showing a fully mesenchymal phenotype and those residing in the intermediate “hybrid” transition stage possess higher pro-metastatic potential. We and others have indicated an advantage for the “hybrid" (H) cells in acquiring CIC properties and in establishing metastases.9 15 16 Along this line, two recent studies on breast cancer, both in a mouse model of the human tumor and in patients, highlighted that cells undergoing partial EMT were required for metastasis formation, whereas spontaneous or induced acquisition of fully mesenchymal state was accompanied by loss of tumorigenicity and inability to colonize distant organs.15 17
Nevertheless, further studies are required to fully elucidate the mechanisms conferring metastatic advantage to hybrid cells. In particular, little information is available about the interplay between cells in the hybrid status and the immune system. Some reports showed that hybrid cells, similar to mesenchymal cells, express higher level of PD-L1 compared with epithelial cells, acquiring immune-evasive properties.18 Moreover, hybrid cells, through the activation of CD73/adenosine signaling axis, can generate an immunosuppressive microenvironment enriched in M2-like macrophages and regulatory T cells able to cross-protect also epithelial tumor cells from immune attack.19
In this context, the direct interaction of hybrid cells with the antitumor immune effectors, and specifically with NK cells, represents a crucial issue, still unexplored.
NK cells are important antitumor cytotoxic effectors, whose killing activity is regulated by a complex repertoire of surface receptors, which include the activating, natural cytotoxicity receptors (NCRs) (ie, NKp46, NKp30, and NKp44), NKG2D, DNAM-1, and inhibitory, killer-cell immunoglobulin-like receptors (KIRs).20–25 Tumor cells often express ligands for activating NK receptors and low levels of HLA class I molecules (ligands for KIRs), therefore representing elective susceptible targets for NK cells. However, different escape mechanisms such as the release of suppressive factors or the expression of ligands for additional inhibitory check-point receptors characterize subsets of NK-resistant tumor cells.26–30 Indeed, such inhibitory receptors (NKG2A, TIM-3, TIGIT, PD-1) or tumor-expressed inhibitory ligands (PD-L1, PD-L2, or B7-H3) represent valuable therapeutic targets for immunotherapy. Different studies have provided data indicating that CICs and/or tumor cells that have undergone EMT can be killed by NK cells, but it has also been shown that tumor cells acquiring the mesenchymal phenotype have an increased ability to suppress NK cell function.31–33 Nevertheless, there is no information on how NK cells could functionally interact with hybrid CICs and tumor cells.
In the present study, using multiple NSCLC cell lines, including isogenic variants, in different EMT states and a murine model of lung metastasis, we investigate the traits conferring advantages to H cells in the different steps of the metastatic cascade and their unique interaction with NK cells.
Methods
Cell cultures and generation of A549 and LT73 sublines
For details on cell lines and NK cell preparation see online supplemental methods.
Supplemental material
To generate M4 sublines, 5×105 cells were plated in 300 µL of RMPI+1% FBS onto the upper chamber of 6 well-8 μm pore insert (BD Falcon) and chemoattracted in the lower chamber by RMPI+10% FBS. After 48 hours residual cells on the top of the insert membrane were removed and cells adherent to the lower face of the insert were trypsinized, collected, and plated in complete medium into 6-well plates to allow cell expansion. The same procedure was repeated for four rounds of migration to obtain M4 sublines.
To induce the A549-TGFβ mesenchymal subline, cells were treated for 7 days with human recombinant TGFβ-1 at 5 ng/mL (Peprotech).
Mass spectrometry-based proteomics and weighted gene coexpression network analysis
Samples from the indicated cell lines were lysed with 50 µL SDS in 100 mM Tris-HCl pH 8. Proteins were then isolated by the PAC method and digested with 0.7 µg trypsin and 0.3 µg LysC. Obtained peptides were analyzed by a nano-UHPLC-MS/MS system using an Ultimate 3000 RSLC coupled to an Orbitrap Q Exactice Plus mass spectrometer (Thermo Scientific Instrument). Globally, approximately 8900 proteins were identified and quantified (online supplemental table 1). Perseus software (V.1.6.15.0) was used for statistical analysis and as a graphical interface for weighted gene coexpression network analysis (WGCNA) package in R. Details in online supplemental methods.
Supplemental material
Chemotaxis assays
Polyclonal NK cell lines were plated at 2×106/mL (RPMI+10% FBS) in the upper compartment of Transwell chambers (3 µm pore size; Corning Costar) and allowed to migrate for 2 hours at 37°C. Cells migrated in the lower compartment were collected and counted using the MACSQuant Analyzer (Miltenyi Biotec). Details in online supplemental methods.
Evaluation of the effects of NK cells on tumor cells and their CIC subset
Tumor cells were seeded at 3×105 cells/2 mL in 6-well plate in complete medium and allowed to adhere overnight. The day after, 3×105 NK cells were added to tumor cells and incubated for 4 hours at 37°C 5% CO2 (with or without blocking antibodies against PD-L1 and B7-H3). NK and floating dead cells were removed and adherent cells were harvested by trypsinization, assessed for vitality by trypan blue exclusion method, and then analyzed for CD133 expression by FACS to quantify the CIC subset.
Animal studies
All in vivo experiments were performed using female SCID mice, 7–10 weeks old (Charles River). Tumorigenic assay was performed by injecting 1×105 LT73 and LT73-M4 cells into the right flank of mice with Matrigel (Corning) and monitored by a caliper two times a week.
For experimental lung metastasis assay, A549 and LT73 cell lines/sublines were injected intravenous at 1×106 and 1×105 cells, respectively and the animals were sacrificed after 2 and 3 months, respectively.
For in vivo NK cell neutralization experiments, A549 cell lines/sublines were injected at 5×105 and mice were sacrificed after 1 month. Purified rat anti-mouse IL-2Rβ antibody or rat isotype-matched control antibody (both Ultra-LEAF BioLegend) was administered once a day for 3 days at 200 µg/mouse, followed by tumor cell intravenous injection, and then once a week for 4 weeks.
Patients’ sample classification
NSCLC specimens were obtained from n=79 consenting untreated patients (clinicopathological characteristics available in online supplemental table 2) (Department of Thoracic Surgery, INT—Milan).
Supplemental material
To define the E-H-M states of the samples, E-cadherin (membrane) and SNAI2 (nuclear) expression was evaluated by immunohistochemical (IHC) as previously described in Andriani et al9 (see also online supplemental methods). The pathologist performed the histological blinded evaluation of the staining and assigned the score. Samples were stratified and classified combining the expression score of both markers (online supplemental figure 7A).
Supplemental material
The same case series were then stained by IHC for NKp46 marker. All slides were scanned and acquired with Aperio ScanscopeXT (Leica Biosystems) and quantified with ImageScope software (V.11.1.2) using an Aperio Cytoplasm Algorithm (V.2.0).
Statistics
Statistical analyses were performed using GraphPad Prism V.5.0. Statistically significant differences were determined with Student’s t-tests when comparing two groups or analysis of variance for multiple comparisons. Data are presented as mean (±SD), unless otherwise indicated. The univariate association between E/H/M status of primary tumors and clinical-pathological characteristics of NSCLC patients was calculated with χ2. Overall patient survival (OS) was calculated from the date of surgical diagnosis and that of last follow-up (censored) or patient death (event), and was analyzed using Kaplan-Meier log-rank tests. P values *≤0.05*; **≤0.01; ***≤0.001.
Results
Selection of human lung tumor cell lines and sublines with different EMT phenotypes
We have previously validated an EMT index based on the expression ratio of CDH1 and SNAI2 genes (ratio E-cadherin/SNAI2—RES index). Exploiting the RES index we defined the EMT status of two lung cancer cell lines, A549 and LT73, as E and H, respectively.9 From both cell lines we have now derived a subline (named M4) by selecting highly motile cells through sequential rounds of in vitro migration assay (figure 1A). As migration capability increases along EMT, the EMT status of the selected M4 sublines was skewed forward (see RES index of A549/A549-M4 and LT73/LT73-M4 in figure 1B). To reproduce all the transitional stages, A549 cells were also cultured in the presence of TGFβ (A549-TGFβ) to induce a fully mesenchymal phenotype. Based on the RES index, morphology, and expression of informative proteins and EMT-associated genes, the generated lines/sublines could be classified as follows: E (A549), H (A549-M4 and LT73), M (A549-TGFβ, LT73-M4) (figure 1B–E). The generated M4 sublines stably maintained EMT phenotype during in vitro culturing (online supplemental figure 1A,B).
In vitro establishment and characterization of hybrid and mesenchymal cell lines. (A) Cartoon summarizing the in vitro procedure to generate M4 cell sublines, through in vitro serial cell migration experiments. (B) Real-time PCR assessment of the ratio between CDH1 and SNAI2 gene expression (ratio between E-cadherin/SNAI2 expression (RES) index) in the indicated non-small cell lung cancer (NSCLC) cell lines and sublines. Mesenchymal cells are identified by RES<10, hybrid cells by 10≤RES≤150, epithelial cells by RES>150. (C) Representative images of A549 and LT73 sublines in bright-field (on the top) and immunofluorescence (on the bottom) for E-cadherin (red) and SNAI2 (green), identifying cells in different epithelial/mesenchymal state to confirm gene expression data evaluated by RES index. (D, E) Real-time PCR analysis for a panel of selected epithelial to mesenchymal transition genes in A549 (D) and LT73 (E) lines and sublines. Original A549 or LT73 cell lines were used as calibrators. Bars are the mean value±SD. N=4 replicates. P value was calculated by two-way analysis of variance.
Hybrid cells are less efficient than mesenchymal cells in disseminating from primary tumors but show the highest ability to found metastases
To dissect the relevance of EMT in the different phases of tumor cell dissemination and metastasis formation, we first compared LT73 and LT73-M4 cells (H and M phenotype, respectively) for their ability to spread to distant sites from tumors generated by subcutaneous injection in SCID mice. The tumorigenic assay showed that hybrid cells were more efficient than mesenchymal in generating tumors, but they poorly spread from the primary tumor to the lungs (figure 2A,B). On the other hand, mesenchymal cells disseminated more but without generating overt metastases (online supplemental figure 2A).
Hybrid cell lines show the highest ability to develop metastases. (A) Growth curve show the tumor volume of xenografts generated by subcutaneous injection of LT73 and LT73-M4 cells (1×105 cells) in the flank of SCID mice (n=4 mice/group). Treatment groups were compared by analysis of variance (ANOVA) test. (B) Evaluation of disseminated tumor cells in the lungs of mice bearing LT73 xenografts as in (A). Tumor cells were detected by FACS analysis as live (7-AADneg) mH2Kdneg cells. P value was calculated by t-test. N=4 lungs/groups in technical duplicate. (C) LT73 or LT73 M4 cells (1×05 cells) were injected in SCID mice via tail vein (intravenous). Tumor cells in the lungs were quantified 3 months postinjection by FACS analysis. N=3 independent experiments, n=3 mice/groups. P value was calculated by t-test. (D) Representative immunohistochemical (IHC) images for the expression of human cytokeratins (CKs) in lungs analyzed in C. (E–G) A549 sublines (1×106 cells) were injected intravenous in n=4 SCID mice/group. Two months postinjection, metastatic formation was assessed by FACS as percentage of human tumor cells within the lung (E) and by IHC for CKs expression in the same lungs, followed by Aperio software quantification of CKs+cells (F–G). N=4 mice/group in technical duplicate. P value was calculated by one-way ANOVA. (H–I) Assessment of CD133+ CICs evaluated by FACS in A549 (H) and LT73 (I) cell lines and sublines. N=12 independent experiments for each cell subline. Bars represent the mean value±SD. P value was calculated by one-way ANOVA. (L–M) FACS analysis of CD133+ CICs content within lung metastases induced by A549 (L) and LT73 (M) sublines. Bars represent the mean value±SD. N=4 lungs/group in technical duplicate. P value was calculated by one-way ANOVA.
Next, to test the ability of cancer cells to colonize distant organs (independent of their release from primary tumor), we performed experimental lung metastasis assays (ie, tail vein injection). In these experiments, LT73 cells (H) were more efficient than LT73-M4 (M) in colonizing lungs and generating histologically evident metastases (figure 2C,D). Indeed, M cells entered the lungs after intravenous injection, but were unable to form metastases. We also tested A549, A549-M4, and A549-TGFβ cells to compare all the EMT stages. H cells (A549-M4) showed the highest ability to invade the lungs and generate metastases while E cells (A549) showed the least (figure 2E–G).
Overall, these experiments indicate that cells in the hybrid phenotypic status, although moderately capable of leaving the primary tumor, have maximal ability to found metastases. To investigate the reason for this advantage, we evaluated proliferation capability and CICs content of cells in different EMT status. M cells exhibited reduced proliferation capabilities within the group of A549 but not within LT73 cell variants (online supplemental figure 2B,C), suggesting that proliferation could play a role in metastatic formation but it is not crucial. Regarding CICs, defined as CD133+ cells,6 31 their content was highest in M or H cells, depending on the analyzed cell line. This result also reflected the relative expression levels of stemness genes in the different cell lines/sublines (figure 2H,I and online supplemental figure 2D–F). The CICs subset, however, was expanded in the lung metastases, and in that case, it was invariably higher in lesions induced by H cells, both using A549 and LT73 cell sublines (figure 2L,M and online supplemental figure 2G).
In conclusion, we show that cells in the hybrid status induce lung metastases with a higher percentage of CICs, which suggests that CICs from H cell lines (H-CICs) could home to premetastatic sites and persist more efficiently than those from M or E cell lines (M-CICs or E-CICs).
Hybrid cells generate metastases which are poorly infiltrated by NK cells
The efficiency in generating metastases of H cells and the related persistency of their H-CIC subset may also depend on their interaction with the immune system. We focused on NK cells, the main antitumor effectors in immunocompromised SCID mice, used for our in vivo tests. Moreover, functional interactions between human tumor cells and murine NK cells (and vice versa) are well documented, and supported by cross-reactivity within several murine-human receptor/ligand pairs.31 34–36
We initially characterized the immune cell subsets within the metastatic lungs of mice injected with A549 cells in their E, H, or M transitional status. We observed a significant reduction of NK cell numbers in the lungs with H metastases, compared with those with E metastases and a trend towards reduction compared with lungs with M metastases (figure 3A). These results were confirmed by IHC analysis (figure 3B). Considering other immune cell types, only neutrophils showed a trend towards reduction in H and M compared with E metastatic lungs (online supplemental figure 3A).
Hybrid cell-derived metastases show decreased content of infiltrating NK cells. Mice were injected intravenous with A549 sublines to induce lung metastases, then metastatic lungs were analyzed to assess NK cell infiltration. (A) FACS analysis of NK cells infiltrating metastatic lungs. NK cells were identified by morphological and phenotypical gating as live cells with low SSC, expressing the CD45+DX5+ phenotype. Bars represent the mean value±SD. N=4 lungs/group in technical duplicate. P value was calculated by one-way analysis of variance (ANOVA). (B) Immunohistochemical (IHC) analysis assessing murine NKp46 NK cell marker expression in metastatic lungs. Areas within dashed boxes are shown at higher magnification. (C–D) Analysis of chemokine expression in metastatic lungs. Representative images of IHC staining for human CXCL8, CXCL12, and CXCL1 cytokines in lung metastases generated by A549 sublines (C), and quantification of chemokine-expressing metastatic cells (D); N=4 lungs/groups. P value was calculated by one-way ANOVA. (E) Supernatants (SNs) from A549 cell sublines (3×105 cells/mL) were collected and tested for their ability to induce NK cell migration in a 2-hour chemotaxis assay. Complete medium was assessed as control. The experiments were performed either in the absence (open bars) or in the presence (shaded bars) of a cocktail of anti-CXCL8, anti-CXCL1, and anti-CXCL12 neutralizing antibodies. The data are means±SEM of three independent experiments in technical triplicate. P value was evaluated by: ANOVA test for comparison among CTR and the different cell SNs, t-test for comparison between absence/presence of blocking antibodies.
We next analyzed by IHC the lung metastases for the expression of human chemokines mainly involved in NK cell and neutrophil recruitment, showing that A549 M4 ensuing tumors expressed the lowest levels of CXCL8, CXCL12, and CXCL1 (figure 3C,D).
Therefore, A549-M4 hybrid cells could generate lung metastases with the lowest expression of key attracting chemokines and, coherently, the lowest NK cell infiltrate.
To confirm in vivo data, we tested in vitro the ability of culture supernatants (SNs) derived from A549, A549-M4, or A549-TGFβ cells to induce migration of human NK cells and assessed the influence of CXCL12, CXCL8, and CXCL1 using a mixture of specific blocking monoclonal antibodies (mAbs). SNs from A549-M4 (H) or A549 TGFβ (M) induced no or poor migration, respectively, while SN from A549 (E) significantly induced NK cell migration, which was specifically prevented by mAbs (figure 3E). Gene expression analysis confirmed that A549 M4 (H) expressed the lowest levels of CXCL8 and CXCL1, A549-TGFβ (M) intermediate, and A549 (E) the highest (online supplemental figure 3B). In all the cases, in vitro cultured tumor cells minimally expressed CXCL12, implying that mechanisms triggered by the tumor microenvironment in vivo could induce the expression of CXCL12 in tumor cells (see figure 3C).
Overall, these experiments indicate that hybrid and mesenchymal cell sublines marginally attract NK cells via chemokine release.
Lung cancer cells with hybrid phenotype have low efficiency in attracting NK cells
To demonstrate that the functional features characterizing the different A549 sublines were distinctive of the E-H-M states, we extended the analysis to a panel of NSCLC cell lines of different tumor histology, which we classified by RES Index as E, H, or M (online supplemental figure 4). To further characterize the heterogeneity of cell lines in different states of EMT, we analyzed their proteomic profile (online supplemental figure 5A) and evaluated differentially expressed proteins (online supplemental table 1) using WGCNA (deposited to the ProteomeXchange Consortium, dataset identifier PXD044297. Username: reviewer_pxd044297@ebi.ac.uk, Password: KI8t8cyX). WGCNA revealed several protein clusters, which correlated with individual cell lines, E-H-M groups, or tumor histology (online supplemental figure 5B). We selected eight modules characterizing the distinct E, H, or M groups (figure 4A). The enrichment analysis of these eight WGCNA-identified protein groups indicated several pathways associated with the EMT process, confirming the validity of RES index to cluster cell lines. Additionally, this analysis highlighted terms related to secretory processes, crucial for cytokine/chemokine release (figure 4A and online supplemental figure 5C).
Proteomic and functional analysis of a panel of E, H, M cell lines. (A) Weighted gene coexpression network analysis (WGCNA) of proteins differentially expressed in a panel of non-small cell lung cancer (NSCLC) cell lines, which were classified according to ratio between E-cadherin/SNAI2 expression (RES) as E, H, and M (enclosed in the green, yellow, magenta frame, respectively). WGCNA allowed the characterization of several modules (identified by different colors) encompassing proteins with similar coexpression profiles. Eight modules correlating with E-H-M groups were selected. The correlation of the selected modules with cell lines and E-H-M cell line groups is shown. Red and blue color notes positive and negative correlation with gene expression, respectively. Statistical significance is reported: *p<0.05, **<0.01 ***<0.001, ****<0.00001. On the right, for each module, it is represented the number of terms that could be related to epithelial to mesenchymal transition (EMT) or secretion as highlighted by the functional enrichment analysis for gene ontology biological processes (GOBP). (B) Evaluation of CXCL8/CXCL1/CXCL12 gene transcripts by real-time PCR in a panel of NSCLC cell lines classified as E (n=4), H (n=4), or M (n=7) according to RES index. Whisker-box graphs represent the distribution of 2ˆ−ΔCT gene expression values. P value was calculated by one-way analysis of variance (ANOVA). (C) Chemoattractant effect of culture supernatant (SN) from cell lines classified as E (n=4), H (n=4), or M (n=7) according to RES index. SNs from 16h cell cultures (3×106 cells/mL) were assessed for their ability to induce NK cell migration as in (A). The data are means±SEM of three independent experiments. P value was calculated by one-way ANOVA. (D) Evaluation of chemokine content and chemoattractant properties of SNs from E, H, or M cell lines stimulated with phorbol 12-myristate 13 acetate+ionomycin. Each symbol represents SN from a different tumor cell line and indicates its chemokine content (pg/mL) and the mean number of induced migrated NK cells (n=3±SEM) (green: epithelial tumor cell lines; orange: hybrid; red: mesenchymal). Horizontal black line indicates the mean number of migrated NK cells in response to control medium (n=3). The linear regression blue lines estimate the correlation between the ability to attract NK cells and the production of CXCL8, CXCL1, or CXCL8+CXCL1.
The cell lines were then assessed for expression and release of CXCL8, CXCL1, and CXCL12. RT-PCR analysis, revealed that hybrid cell lines displayed the lowest levels of CXCL1 and CXCL8 gene expression while epithelial cell lines showed the highest. CXCL12 was minimally expressed in all instances (figure 4B). Coherently, hybrid cell lines also showed the lowest ability to attract NK cells, as their SNs were ineffective in migration assays (figure 4C).
We next analyzed whether the distinctive features across different EMT states could also be maintained after cell stimulation, as it occurs in the tumor microenvironment. SNs of cells stimulated with phorbol 12-myristate 13 acetate+ionomycin showed correlation between their content of CXCL8 and CXCL1 and the ability to induce NK cell migration, with the maximum score observed for CXCL8 (figure 4D). Remarkably, hybrid cell lines maintained the lowest chemokine production and NK cell chemotaxis induction.
CICs with hybrid phenotype escape from human NK cell surveillance
The above data indicate that as an escape strategy, hybrid and partly mesenchymal cancer cells limit their ability to support NK cell infiltration into tumor niches. However, tumor cells may also encounter NK cells in the circulation, before metastasis formation. Therefore, their capability to limit the NK-cell killing activity and IFN-γ production on cell-to-cell interaction could further influence the pro-metastatic properties. To evaluate this aspect, tumor sublines tested in the mouse models were assessed for the surface expression of adhesion and NK-receptor ligands and for their ability to induce cytotoxic degranulation of (human) NK cells. Mesenchymal sublines (A549-TGFβ and LT73-M4) exhibited the lowest expression of the NCR-Ligands. Additional changes in ligand expression were specific for each cell line but did not appear to parallel EMT (figure 5A,B and online supplemental figure 6A,B). Also, the ability to induce cytotoxic degranulation and IFN-γ production by NK cells was reduced in the M transitional state of A549 and not in that of LT73 (figure 5C,D, online supplemental figure 6D–E and online supplemental figure 7A,B). To assess whether tumor cells could induce inhibitory immune check-point receptor (ICI) expression on NK cells, as escape mechanism, we evaluated by flow cytometry the expression of PD-1, TIM-3, and TIGIT on NK cells before and after the degranulation assay. No statistically significant changes in ICI expression were shown by NK cells after coculture (online supplemental figure 7C,D), even though a trend to decrease could be observed for TIGIT expression, which however was similarly induced on NK cells independently by tested tumor sublines. Thus, overall, the hybrid cells did not show reduced NK cell-stimulating capabilities, compared with mesenchymal or epithelial cells, suggesting that their advantage in forming metastases is independent of this functional feature.
NK cells have similar killing activity against bulk tumor cell populations of E, H, M cell lines but are not effective against hybrid cancer-initiating cells (H-CICs). (A–B) FACS analysis of NK receptor-ligand expression in A549 and LT73 lines and sublines. The ligands for the different NK receptors are grouped as indicated above the panels (NCR-Ls, NKG2D-Ls, DNAM1-Ls, KIR-Ls, and the adhesion molecule ICAM-1). The NCR-ligands were analyzed as the expression of B7-H6 (ligand for NKp30) and as the staining effect of the NKp46-Fc or NKp44-Fc chimeric molecules, since tumor cells may express multiple and, in part, still unknown ligands for NKp46 and NKp44. Bars represent mean fluorescence intensity (mfi) mean±SEM of at least five independent experiments. P values were evaluated by analysis of variance (ANOVA) test. (C–D) Evaluation of the surface expression of CD107a cytotoxic degranulation marker on NK cells after they have been cocultured with A549 or LT73 cell lines or sublines. Whisker-box graphs represent the distribution of seven independent experiments for A549 sublines and five independent experiments for LT73 sublines. (E) Assessment of NK cell killing effects on A549 and LT73 cell lines or sublines after coculture with NK cells. Data are expressed as the ratio between the number of live tumor cells (evaluated by trypan blue count) 4 hours after culture in presence/absence of NK cells. Graphs represent the data distribution of N=6 independent experiments for both A549 and LT73 sublines. (F) FACS analysis of CD133+ cells in A549 and LT73 cell lines and their respective sublines after 4 hours coculture with NK cells. Data are expressed as fold change of the CD133+ cell percentage relative to cancer cell lines cultured alone. N=6 independent experiments for each subline. P value was calculated by one-way ANOVA. (G) Assessment of NK cell killing effects on E, H, and M cell lines as described in (E). Graphs represent the data distribution of n=4 E cell lines, n=4 H cell lines, n=7 M cell lines. P values were calculated by one-way ANOVA. (H) FACS evaluation of CD133+ cells modulation in E, H, and M cell lines before and after coculture with NK cells, as described in (F). Graphs represent the data distribution of n=4 E cell lines, n=4 H cell lines, n=7 M cell lines. P value was calculated by one-way ANOVA. (I) The heatmap reports the median values of the ratio of NK ligand expression in CIC versus non-CIC subsets, as assessed by FACS analyses in the E-H-M cell lines that were tested in (G–H). (L) FACS evaluation of the modulation of CD133+ cell percentages in A549 E, H, and M sublines after coculture with NK cells, alone or in presence of neutralizing antibody against PD-L1 and B7-H3.
In the experimental metastasis assay, hybrid cells induced lung metastases with the highest proportion of CICs (figure 2L,M), which might indicate an inherent resistance of H-CICs to the NK cell attack. To evaluate this possibility, A549 and LT73 sublines were cocultured with NK cells and the effect on CICs compartment was assessed. NK cells showed similar killing activity towards E, H, and M cells in both A549 and LT73 variants, however the CIC subset was increased in the H cells (A549-M4 and LT73), while it was unaffected or even slightly reduced in E (A549) and M cells (A549-TGFβ and LT73-M4) (figure 5E,F). This result indicated that CICs could be more resistant to NK cell attack than their non-CIC counterpart in H cell lines. We confirmed this finding in the panel of E, H, and M cell lines (online supplemental figure 4), proving that NK cells did not show preferential killing of cells belonging to a specific group, but, however, H cell lines significantly increased their CIC content after coculture, demonstrating the peculiar resistance of H-CICs to NK cells (figure 5G,H). We then analyzed by flow cytometry the expression of NK-receptor ligands and cellular adhesion molecules on CICs and non-CICs and noticed differences between the two cell subsets, that are distinctive for E, H, and M cell lines. Specifically, in the H cell lines, CICs showed the highest increase of B7-H3 and PD-L1 inhibitory ligands compared with non-CICs, while in the M cell lines, CICs showed the highest increase of some of the ligands involved in the triggering of NK cell function (namely, ICAM1 and CD48) (figure 5I). Co-cultures of A549 sublines and NK cells in presence of blocking antibodies against inhibitory ligands showed that B7-H3 blockade could significantly prevent the increase of CIC subset in H cells (figure 5L), while it had no direct effects on CICs (online supplemental figure 8). Therefore, resistance of H-CICs to NK cells appears to be mainly due to the expression of B7-H3.
Hybrid cells display pro-metastatic capabilities which are poorly controlled by NK cells in vivo
Based on the above results, we next assessed whether H cells could effectively develop metastases independent from the NK cell surveillance. We performed metastatic assays by injecting A549 sublines in SCID mice pretreated with anti-IL-2Rβ blocking antibodies, to neutralize NK cells, or with control IgG antibodies. The assays were set at a short timespan to avoid excessive metastasis outgrowth in NK-depleted mice and to maximize the possibility to observe differences among the injected sublines (figure 6A–C). The NK cell blockade resulted in a dramatic expansion of the lung metastasis formation in mice injected with M cells, while it had moderate effects in the mice injected with E or H cells (figure 6A–C). Further analyses revealed that, among control mice, those injected with H cells exhibited maximal content of CICs in their metastatic lungs, which, however, was not influenced by the NK cell neutralization (figure 6D). By contrast, control mice injected with M cells showed lower content of CICs, which was greatly increased after NK cell depletion. Finally, E-cell injection resulted in little CIC content in the lungs independent of the absence/presence of NK cells in the mice (figure 6D). Therefore, in line with the data obtained in vitro (figure 5F), murine NK cells may differently interact with M-CICs and H-CICs, crucially influencing the metastatic spread in vivo.
Effect of in vivo NK cells neutralization on metastatic outgrowth induced by E, H, M, cells and cancer-initiating cell (CIC) subset expansion. (A) SCID mice were pretreated with the anti-mouse IL-2Rβ neutralizing antibody to inhibit NK cells or with isotype-matched control antibody and then injected intravenous with 5×105 cells of the indicated A549 sublines. Histological analysis of lungs and immunohistochemical (IHC) staining for cytokeratins (CKs) was performed in lung tissue 1 month after injection. N=4 mice/group. (B) Relative quantification of CKs+lung metastatic cells. Bars represent the mean fold-change±SD of CK+cells in lungs of mice treated with IL-2Rβ neutralizing antibody compared with control, for A549, A549 M4, and A549 TGFβ sublines. P value was calculated by one-way analysis of variance (ANOVA). (C) The percentages of lung metastatic cells in the lungs analyzed in (B) was also evaluated by FACS. Metastatic cells were identified as cells negative for the mouse mH2Kd marker. Graphs represent the distribution of the data. N=4 mice/group in technical duplicate. P values were calculated by t-test, comparing groups treated with neutralizing antibody versus control IgGs for each E-H-M subline. (D) CD133+ CICs content in metastasis was evaluated by FACS analysis. Bars represent the mean value±SD. N=4 mice/group in technical duplicate. P value was calculated by t-test, comparing groups treated with neutralizing antibodies versus IgGs for each E-H-M subline.
Primary hybrid tumors with lowest NK cell infiltrate are associated with worst prognosis
To validate in a clinical setting our observation for an increased aggressiveness of hybrid tumors (H-tumors), we classified a retrospective case series of 79 surgically resected non-treated primary NSCLC tumors (online supplemental table 2) as epithelial (n=38), hybrid (n=25), mesenchymal (n=16) according to IHC staining for membranous E-cadherin and SNAI2, as previously reported9 (online supplemental figure 9A).
This analysis showed that patients with H-tumors experienced the worst prognosis compared with those with epithelial or mesenchymal tumors (figure 7A), confirming our previous data on a smaller cohort of NSCLC patients. We next investigated by IHC the NK cell infiltrate in tumor specimens and, in line with our in vitro and in vivo data, we confirmed a significantly lower presence of NK cells in H-tumors compared with E-tumors, whereas M-tumors showed an intermediate phenotype (figure 7B,C).
Hybrid state and low NK cell infiltrate in non-small cell lung cancer (NSCLC) classifies patients with worst outcome. (A) Kaplan-Meier analysis of overall survival (OS) for patients stratified by phenotypic tumor state as E (n=38), H (n=25), M (n=16) as defined by immunohistochemical (IHC) staining for E-cadherin and SNAI2. P values were calculated using the log-rank test and are indicated and statistical significance was set at p<0.05. (B) Representative images of IHC for the NK-specific NKp46 marker in the same case series of primary NSCLC tumors, evaluated in (A) classified according to phenotypic status. Areas within boxes are shown at higher magnification. (C) Quantification of the percentage of NK cells (NKp46+cells) within E, H, and M tumors. NKp46 positivity was established as 2+ and 3+ intensity by Aperio software (setting 1+ intensity staining as background). Whisker-box graphs represent the distribution of the percentage of NKp46+ tumor infiltrating cells within the three groups of patients. Comparison among groups were performed by one-way analysis of variance. Patients were also classified on the basis of the tumor NK cell infiltration: NKlow scoring<3% and NKhigh≥3%. The discriminating 3% percentage represents the mean NKp46+cell% of the total analyzed tumors. (D,E) Kaplan-Meier curves for OS of patients with high tumor NK cell infiltration (NKhigh≥3%) (D) or low tumor NK cell infiltration (NKlow<3%) (E) stratified by phenotypic state as E, H, M. Reported p values are calculated by log-rank (Mantel-Cox) test.
To further evaluate the role of NK cells in the progression of the disease, we tried to dissect the possible prognostic value of the NK cell infiltrate according to the different tumor phenotypes. Tumors were then stratified as NKlow or NKhigh, based on the median percentage of NK cell infiltrate as evaluated by IHC in the whole patient cohort (see the Methods section). Considering all phenotypic groups, patients with E, H, or M tumors and a NKhigh infiltrate showed a comparable OS, suggesting that NK cells might locally control the tumor spread, regardless the tumor phenotype (figure 7D). Conversely, patients with NKlow H-tumors exhibited a significant shorter survival compared with those with NKlow E-tumors and NKlow M-tumors (figure 7E). This observation is consistent with our in vitro and in vivo data (see figures 5F,H and 6), as it suggests that in case of poor NK cell infiltration, NK cells can still eliminate disseminating M-CICs but not H-CICs, with negative consequences on patients with NKlow H-tumors. On the other hand, the lower malignancy of NKlow E-tumors is in line with the limited CICs content of E tumor cells (see figure 2H,L).
In conclusion, we confirmed in NSCLC clinical setting that tumors characterized by a hybrid EMT profile are more aggressive than E-tumors and M-tumors, and that this behavior could be partially explained by their peculiar interaction with NK cells. More importantly, tumors characterized by H phenotype combined with low NK-infiltration identify a patient cohort with worse prognosis.
Discussion
The phenotype switching along the EMT process has been proposed several years ago as a crucial event enabling tumor cells to leave primary lesions and seed metastases.37 However, despite several reported studies, the characterization of how the transitional status of malignant cells could effectively influence the tumor progression still remains an open issue, with controversial data.13 14 38 In this context, the idea that cells possessing both epithelial and mesenchymal traits, that is, expressing a hybrid phenotype, can have the highest chance to successfully establish metastasis is rising to prominence.8 9 15 17 39 40 Our study corroborates previous data demonstrating the higher metastatic potential of hybrid cells and provides a new mechanism to elucidate this property, crucially regulated by specific interactions of hybrid CICs with NK cells. Our study, focused on NSCLC, has been designed to evaluate the pro-metastatic potential related to the different EMT states and the underlying critical mechanisms, to identify possible therapeutic targets. We previously defined in NSCLC the ratio between E-cadherin/SNAI2 expression (RES) as reliable index to assess the phenotype shifts within the EMT in a function-predictable fashion.9
Now, performing in vitro multiple cycles of cell migration and selection we have generated different human NSCLC cell line variants, representing prototypic E, H, and M stages of EMT, as defined by RES index. These variants stably maintained their phenotypic status over in vitro passages, without the need for external stimuli or transfection of EMT-master regulator genes, thus representing a robust tool to dissect the contribution of different EMT states in dictating metastatic success. By evaluating these cell sublines in vitro and in animal models, and studying additional E, H, M cell lines we could define the role of hybrid cancer cells in metastasis establishment, pointing out the crucial aspect of their interaction with NK cells.
RES could oversimplify the complexity of the different EMT status and other possible mesenchymal markers could be taken into consideration for the definition of the H and M phenotype. For example, the cell line H1975, assigned as M according to RES (online supplemental figure 4), had been previously classified as hybrid cell line using different distinctive markers and different analytical approach.41 The elusive phenotype of these cells is confirmed by the WGCNA (figure 4A), where H1975 shows a peculiar profile, which cannot be clearly associated with any group. Thus, classifying the complexity of the “EMT phenotype” could be challenging for certain tumor cells, however, our data encourage the use of RES index for a rapid, and in most cases reliable, measurement of plasticity, associated with the CIC phenotype.
Our results from in vivo experiments, using cell sublines, indicate that H cells are less efficient than M cells in leaving primary tumor, but, once entered the circulation, they are more prone to generate metastases. This effect could be partly related to the superior ability of H cells to deliver or maintain high numbers of CICs at the metastatic sites and to their modest production of NK-attracting chemokines, resulting in limited NK cell infiltration of their metastases in the animal model. This latter observation is in line with previous studies showing that human chemokines, released by cancer cells, can attract and influence murine immune cells in immunocompromised mice models.35 36 The peculiar features characterizing A549/LT73 cell sublines are confirmed in a panel of E, H, and M cell lines, subjected to proteomic and functional analyses. In the WGCNA, the protein modules correlating the E-H-M groups are enriched in terms related to epithelial traits, cytoskeletal functions, and vesicle transport, supporting the robustness of the RES index for dissecting EMT states, and suggesting modulation of the secretory processes along EMT. The in vitro experiments confirm that, indeed, H cell lines and sublines do not attract human NK cells. Interestingly, M cells show chemoattractant ability only slightly higher than that of H cells, and, accordingly, partly higher NK cell recruitment in their metastases (figures 3A–C and 4C). Such, limited, differences in NK cell infiltration may not fully explain the ability of NK cells in controlling M-metastases and not H-metastases in vivo (figure 6), suggesting that H and M cells could differently interact with NK cells, conceivably also before metastasis formation. In this context, the increased resistance to NK cell activity of H-CIC proves crucial, as demonstrated by the inability of both human and murine NK cells to control this H-CIC subset in coculture experiments and in vivo (figures 5F,H and 6D). These findings are consistent with the expression pattern of the ligands for activating/inhibitory check-point receptors in CICs and non-CICs from E, H, or M cell lines. M-CICs, which are controlled by NK cells, show increased expression of CD48, ligand of the CD244 activating receptor, and ICAM1, crucial for the lytic immunological synapse formation.29 Instead, H-CICs show increased expression of PD-L1 and B7-H3 inhibitory ligands. In particular, B7-H3 appears to be the primary mediator of H-CIC resistance to NK killing activity, as anti-B7-H3 blocking mAbs could considerably enhance the ability of NK cells to attack the fraction H-CICs within H cells (figure 5L). The in vivo experiments indicate a functional cross-reactivity between the human and murine systems. This is not surprising, as murine homologs do exist for several human NK-receptor:ligand pairs,28 42 and their cross-interaction has been suggested and demonstrated in some studies.31 34 In particular, B7-H3 has been recently described to play a role in the CIC-immune cell interaction both in humans and in mice.43
We have previously demonstrated that cells residing in the hybrid status exhibit maximal ability to acquire CIC properties in response to tumor microenvironment stimuli, providing an initial explanation for their key role in the process of metastatic spread.9 Here, we update the picture and indicate the interaction with NK cells as an important mechanistic element underlying the pro-metastatic advantage of H-tumor cells. Our findings integrate with the very current topic on the multifaceted functional relationships connecting NK cells, tumor cell plasticity, EMT, and the metastatic progression. From a broader point of view, the process of EMT may be incremented by different immune cells, including myeloid, T, and NK cells. In addition, the acquisition of the M phenotype is associated with increased escape properties from T and NK cells.33 44 45 Hybrid phenotype has also been reported to be associated with high PD-L1 expression and generation of immunosuppressive tumor microenvironment through the production of adenosine.18 19 On the other hand, the role of NK cells in targeting CICs and limiting metastasis formation has been suggested in different studies.46–48 As well, it has been shown that NK cells can control metastases by killing tumor cells undergoing complete EMT, even though cells retaining some epithelial traits could be more resistant.31 32 Nevertheless, the specific relationship of cells expressing H phenotype with different immune cell types remains poorly investigated. Here, for the first time, we provide evidence of a differential susceptibility to NK cells among CICs according to their EMT status, and demonstrate that these differences are crucial for the metastatic development in NSCLC. Indeed, the highest pro-metastatic properties of H-cells appear to be related, at least in part, to the highest escape properties of their H-CIC component, and B7-H3, protecting H-CIC from NK cells, emerges as potential key therapeutic target.
Our clinical data indicate a central role for NK cells in the connection between hybrid cells and tumor progression. Indeed, we could identify a subgroup of patients with specific primary tumor features (ie, low NK cell infiltration combined with prevalent hybrid EMT status) who experience worse prognosis (figure 7E and online supplemental figure 9B). Remarkably, besides its prognostic value, this information gives important hints on which patients (those bearing M tumors) could benefit more from NK-based therapy, and which could not (those bearing H-tumors). This is a crucial point, especially considering the increasingly high number of studies aimed at defining efficient NK-based immunotherapy strategies in oncology, and specifically in the treatment of metastatic patients.46 49–51 Targeting B7-H3 has been recently proposed as additional mean to redirect or unleash both NK and T cell antitumor effector functions, considering that B7-H3 could be expressed by consistent fractions of malignant cells in different tumor types. However, its crucial role in supporting H-CIC escape properties opens new promising perspectives to apply B7-H3-based therapies to control metastatic spread in selected NSCLC patient cohorts.52–54
Data availability statement
Data are available in a public, open access repository. Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants and was approved by Independent Ethics Committee of the Istituto Nazionale Tumori (Milano) (INT 25/01/2012-0001497). This is a retrospective study and does not require informed consent. All patient information was deidentified and not be shared with outside parties.
Acknowledgments
We thank Professor Claudia Cantoni for the kind gift of NKp46-Fc and NKp44-Fc chimeric molecules. We thank Dr Eloisa Jantus Lewintre for kindly providing us NSCLC cell lines.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
MV and GB contributed equally.
Contributors GB and MV conceived, designed, supervised the study, analyzed data, and wrote the manuscript. MP conceived and performed in vitro experiment with NK cells, analyzed data, wrote the manuscript. GC, PO performed IHC and analyzed data, contributed to the manuscript preparation. FM, CG, MB generated M4 sublines and performed in vitro experiments. FA performed analysis on NSCLC case series and contributed to the manuscript preparation. IR supervised the study, contributed to the manuscript preparation. MMoro, GT performed in vivo experiments. UP selected case series and collected clinical information. AP, CL performed proteomic analysis. MMilione performed pathological evaluation of human specimens. GS supervised the study and contributed to the manuscript preparation. LR supervised the study and wrote the manuscript. All authors reviewed and approved the manuscript.
Funding The study was supported by grants from Italian Ministry of Health (RF-2018-12366714 to MV and GB, RF-2016-02362946 to LR), Fondazione Regionale per la Ricerca Biomedica (Regione Lombardia) (1731093 to GB), the Italian Association for Cancer Research (AIRC IG 25023 to MV, IG 21431 to LR, 23244 to GS), Italian Ministry of Health (5x1000-2018) granted to IRCCS Ospedale Policlinico San Martino). Italian Ministry of Health Ricerca Corrente” funds to Fondazione IRCCS Istituto Nazionale dei Tumori.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.