The present study was approved by the Institutional Ethics Committee of the Instituto Alexander Fleming (IAF), and all patients enrolled provided written-informed consent. Samples were obtained from 52 patients (AJCC stages I–IV) without any other concomitant colorectal disease who underwent surgical resection of CRC at the Surgery Service of the IAF (Table 1). Inclusion criteria: written-informed consent, age ≥18 years old, and available blood sample collected at the moment of surgery. Exclusion criteria: exposure to chemotherapy and/or lack of written consent. As blood samples were limited in some cases, some determinations could not be performed for all patients. Nine of them where only assayed for TGF-β measurement, and functional assays were performed in a reduced number of samples. As controls, PB samples were obtained from healthy donors (HD) at the Hemotherapy Service of the IAF.

Peripheral blood samples from CRC patients (3–15 ml each) and HD were obtained in heparinized collection tubes. PB mononuclear cells (PBMC) were isolated by Ficoll-Paque PLUS (GE Healthcare Bio-Sciences AB) density gradient centrifugation. For xCELLigence assay, NK cells were purified by negative immune selection using the NK selection kit (Miltenyi Biotech), following company instructions. Purified NK cells (0.5–1.6 × 10⁶/ml) were cultured in RPMI 1640 medium (GIBCO Invitrogen) supplemented with IL-2 (1000 IU/ml; Miltenyi Biotech) and 10% human serum AB (Biowest) for 2 days.

The colon carcinoma cell line DLD-1 (ATCC) was maintained in Dulbecco’s modified eagle medium (DMEM, Invitrogen) supplemented with 10% heat-inactivated fetal calf serum (FCS) Natocor, 2 mM l-glutamine, 3.5 mg/ml sodium bicarbonate, 4.5 mg/ml glucose, and 1% Penicillin–Streptomycin (Invitrogen). The leukemic cell line K562 (ATCC) was maintained in RPMI 1640 supplemented with 10% FCS and 1% Penicillin–Streptomycin.

Approximately 10⁶ PBMC were cultured at 10:1 effector/target (E:T) ratio for 6 h at 37°C with K562 cells and incubated with anti-CD107a-FITC. After 1 h, protein transport inhibitor (Golgi Stop-BD) was added. Five hours later, cells were labeled in PBS for 30 min at 4°C with anti-CD56-APC and anti-CD3-PerCP, after that, cells were fixed and permeabilized (Cytofix/Cytoperm, BD Biosciences) and washed (Perm/Wash, BD Biosciences). Finally, cells were labeled in Perm/Wash buffer for 30 min at 4°C with anti-IFN-γ-PE (BD Biosciences) and then collected on a FACSCalibur flow cytometer. The results are expressed as the percentage of IFN-γ⁺ or CD107a⁺ gated on NK cells. Spontaneous basal IFN-γ secretion and degranulation were determined in absence of targets and cytokines.

DLD-1 cells were used as target and labeled with Calcein-acetyoxymethyl (Calcein-AM; Molecular Probes, Invitrogen Life Technology). The effector cells were PBMC, normalized by percentage of NK cells. The cytotoxicity assay was performed at 2.5:1 E:T ratio, in triplicate, with 1 μg/ml of cetuximab or control mAb (rituximab). Three replicate wells for spontaneous (only target cells in RPMI medium with 10% FCS) and maximum release (only target cells in medium plus 1% Triton X-100) were measured. After incubation at 37°C in 5%CO₂ for 4 h, supernatants were analyzed by fluorimetry (OPTIMA-BMG Labtech) to measure cell death (Calcein release), and the percentage of specific lysis was calculated as: (experimental fluorescence−spontaneous fluorescence)/(maximum fluorescence−spontaneous fluorescence) × 100.

Natural killer cell-mediated lysis of DLD-1 cells was assessed by xCELLigence System (Roche), a label-free real-time monitoring assay of adherent cell lysis (26). It measures electrical impedance across interdigitated micro-electrodes integrated on the bottom of tissue culture E-Plates. The electrode impedance, which is recorded as a dimensionless parameter denominated cell index (CI) value, provides quantitative information about the status of the adherent cells, including cell number, viability, and morphology. DLD-1 targets (15,000 cells) were seeded into the wells of 96 E-Plates in 100 μl of RPMI 1640 medium plus 10% human serum and their adhesion was monitored for 5 h. IL-2-activated purified NK cells were added in a volume of 50 μl/well at 1:1 E:T ratio. Cocultures were assessed by the system with a measurement of CI every 15 min for up to 300 min. Results expressed as CI, were normalized with RTCA Software (Roche, Applied Science, Mannheim, Germany) (nCI) and are expressed as % specific lysis:% lysis = (nCI (no effector) − nCI (effector))/nCi (no effector) × 100. In some cocultures, NK cells were pretreated with anti-NKp46, anti-NKp30, anti-NKG2D, or anti-DNAM-1 mAbs (R&D System).

Direct immunofluorescence was performed on PBMC or NK cells using specific mAbs and isotype-matched mAbs. Cells were washed with PBS and stained for 30 min at 4°C. NK cells, gated on CD3⁻CD56⁺ cells in the lymphocyte FSC/SSC subset were analyzed for the expression of CD16, NKp46, NKp30, NKp44, NKG2D, DNAM-1, CD161, CD94, CD8, NKG2A, CD85j, CD158a, and CD158b (PE-conjugated antibodies, BD Biosciences). mAbs and respective clones are listed in Table S1 in Supplementary Material. Cells were collected on a FACSCalibur flow cytometer (BD Biosciences). These data were analyzed using FlowJo software.

Plasma samples were tested for the presence of HLA-E and HLA-G soluble molecules by home-made ELISA tests. TGF-β content in human plasma was evaluated using a human TGF-β 1 ELISA kit (BD) following the manufacturer’s instructions.

Natural killer cells were incubated overnight in medium supplemented with rhIL-2 1000 IU/ml, rhIL-15 10 ng/ml, or lenalidomide 10 μM (LC Laboratories) prior to performing degranulation, IFN-γ release, lysis, and ADCC, as described above.

Comparisons between CRC patients and HD groups were assessed by non-parametric Mann–Whitney test when samples did not present normal distribution or parametric t-test, when samples presented normal distribution. For functional assays, we used paired t-test to compare different in vitro treatment conditions. Correlations were analyzed by Pearson (when samples presented normal distribution) or Spearman tests (in case of samples not normally distributed). The survival curves were plotted according to the Kaplan–Meier method and compared using the Log-rank test (p values <0.05 noted as *, p < 0.01 as ** and p < 0.001 as ***).

In an observational and exploratory study, 52 CRC patients were enrolled, 34 from I–II stages without lymph node invasion and 18 from stages III–IV with lymph node or metastatic dissemination. Fifty-two HD were analyzed as reference. The clinical characteristics of patients are summarized in Table 1. Phenotypic analysis from CRC patients (43/52) displayed a significant increase of PB CRC-NK cell proportions among the lymphocyte subset as compared with HD (p < 0.001) (Figure 1A, left panel). Since mean age was significantly higher in the group of CRC patients (61.4 ± 12.0 vs. 38.0 ± 13.0 in HD, p < 0.0001), a linear regression model was developed using the percentage of NK cells as the outcome variable, and the presence of cancer and age as regressive variables. The presence of cancer increases by 5.29% the amount of NK cells after adjusting by age (IC95 = 1.004–9.579, p = 0,016). Surprisingly, CRC-NK percentages were increased from stage II, reaching progressively the highest values in stage IV (linear trend ANOVA, p = 0.0122) (Figure 1A, middle panel). We also found higher absolute numbers of PB CRC-NK cells from available blood cell counts prior to surgery (25/43) (Figure 1A, right panel). CD56^dim percentages in CRC-NK cells were similar to HD-NK cells (%CD56^dim CRC-NK 93.06 ± 6.24 N = 43 vs. %CD56^dim HD-NK 91.48 ± 5.01 N = 52). We, then, studied NK cell ex vivo phenotype according to the gating strategy depicted in Figure S1 in Supplementary Material. Most of the 13 NK cell receptors analyzed showed high inter-individual variability in CRC patients. Moreover, eight of these receptors presented modified expression in CRC patients compared with donors NK cells. Furthermore, the altered phenotypic expression was evident in CRC-NK from early stages of the disease (Figure 1B). The percentages of most activating receptors, NKG2D, DNAM-1, CD161, and NKp46 were decreased from non-metastatic stages (p < 0.001, p < 0.001, p < 0.001; and p < 0.01, respectively), while NKp30 presented diminished expression only in advanced ones (p < 0.001). The inhibitory receptors, CD85j and NKG2A were augmented from stages I–II (p < 0.001), while the KIRs 2DL1/2DS1 recognized by anti-CD158a (27) was also diminished in early stages (p < 0.01). Expression of CD16, NKp44, CD94, CD158b (that encompasses KIR2DL2/2DL3/2DS2) (27), and CD8 were comparable to values observed in HD-NK cells (data not shown). Significant positive correlations were found between different activating receptors, including NKG2D, DNAM-1, NKp30, NKp46, and CD161, confirming the broadly phenotypic imbalance in CRC-NK (Figure 1C). In HD-NK cells, activating receptors displayed high and homogeneous percentages that limited correlations mainly to the NCR group and CD16 (data not shown).

Based on the association between NK cell percentages and clinical stage, we investigated the involvement of other clinical features in NK cell receptor expression, as displayed by CRC patients. NK cells from patients with poorly differentiated tumors, considered of bad prognosis, highly expressed NKG2D and DNAM-1 activating receptors (p < 0.05 in both cases) and also higher expression of CD94 (p < 0.05). This receptor, which conforms the heterodimer CD94-NKG2A or CD94-NKG2C, was also regularly expressed in CRC-NK from patients with elevated carcino-embryonic antigen (CEA) values, associated as well with poor prognosis. Moreover, NKG2A and CD8 expression were diminished (p < 0.05 and p < 0.01, respectively) in patients with positive lymphatic vascular invasion (LVI), that is a strong bad prognostic marker and taken into account in treatment determination for stage II–III patients (Table S2 in Supplementary Material).

Based on the phenotypic alterations of CRC-NK cells, we assayed the NK cell degranulation potential (CD107a surface expression) and IFN-γ production in response to challenge by K562 cells as canonical target. Freshly isolated PBMC, either incubated in medium or stimulated overnight with IL-2, were then cultivated with K562 cells for 6 h before analysis by flow cytometry (see Materials and Methods). Challenge with K562 cells led to CD107a expression in HD-NK cells, while NK cells from several patients kept their values close to spontaneous degranulation (Figure 2A, left panel). IL-2 pre-stimulated NK cells from HD and patients exhibited significantly augmented responsiveness (p < 0.001 in both cases). Still, CD107a expression levels of CRC-NK cells were lower than those achieved by HD-NK cells (p < 0.001 for both un-stimulated and stimulated NK cells) (Figure 2A, middle panel). Corresponding to phenotype alterations, diminished degranulation activity was detected in CRC patients at stages I–II showing an early immune impact on NK cells (Figure 2A, right panel).

Natural killer cells’ antitumor effects also implicate the production of cytokines, such as IFN-γ. In this regard, culture with K562 did not induce significant levels of IFN-γ production on resting NK cells (Figure 2B, left panel). IL-2-pre-stimulation enhanced significantly the levels of IFN-γ production by HD-NK and in a lower extent by CRC-NK cells (p < 0.001) (Figure 2B, left and middle panel). Once again, the low functional capacity was observed since early stages of disease (Figure 2B, right panel).

In order to evaluate NK cell cytotoxic capacity against CRC cells, we performed lysis assays of CRC cell line DLD-1 by HD and CRC-NK cells, using colorimetric calcein release test. CRC-NK also presented lower lytic capacity against DLD-1 since early stages of disease (Figure 3A), as we could anticipate from degranulation experiments (Figure 2). In addition, we observed an association between NK cell function, IFN-γ production, and lysis capacity in response to DLD-1 stimulation (p < 0.05) (Figure 3B). Even though IL-2- or IL-15-activated CRC-NK cells improved their lytic activity against CRC cells, they could still not reach HD-NK lytic ability (Figure 3C).

Correlation analyses were performed in order to establish which of the receptor abnormalities in CRC-NK play a major impact on the low-lytic activity against CRC cells. Expression levels of activating NCR NKp30 and NKp46, were positively correlated with DLD-1 cells lysis (r = 0.7364, p = 0.098; r = 0.9286, p = 0.0025; respectively) (Figure 4A). These associations were not present in HD-NK, mainly because HD exhibit homogeneous high expression of activating receptors and homogeneous low expression of the inhibitory ones (Figure 1B).

To elucidate the impact of these receptors in CRC cells cytotoxicity, we assessed the lysis of DLD-1 cells by purified IL-2-activated HD-NK cells using a dynamic assay that quantifies the CI of adherent targets. The lytic activity of NK cells provoked the detachment/apoptosis of CRC cells, as evidenced by an abrupt decay of CI after the addition of effectors (Figure 4B, upper panel). From the CI values, lysis curves were calculated. In some experiments, receptor-neutralizing mAbs, previously tested to block NKp46, NKp30, NKG2D, and DNAM-1 receptors in melanoma (28) and renal carcinoma (29) were used (Figure 4B, middle panel). Separate addition of anti-NKp46, anti-DNAM-1, or anti-NKG2D mAbs decreased CRC cell lysis; the addition of anti-NKp30 mAb induced the highest percentage of inhibition (35.70 ± 4.95%). Addition of anti-DNAM-1 and anti-NKG2D mAbs slightly synergized when combined with anti-NKp30 (Figure 4B, middle panel). These data indicate that the lysis of DLD-1 cells by IL-2-activated NK cells is mediated mainly by NKp30 (Figure 4B, lower panel) with the contribution of NKp46, NKG2D, and DNAM-1.

Several soluble molecules were associated with receptor expression modulation and the impairment of NK cell activity (30, 31). To determine whether CRC patients presented augmentation of NK cell suppressive molecules, we performed ELISA assays from plasma patients. Soluble HLA-E molecules (>5 pg/ml) were detected in 7/35 patients (mean 22.9 ± 53.4; range 0–187.4 pg/ml), and soluble HLA-G molecules (>5 ng/ml) were detected in the plasma of 6/43 patients (mean 1.8 ± 5.1; range 0–23.0 ng/ml). Non-significant differences were found as compared with plasma samples from HD (data not shown). However, the amount of total circulating TGF-β in the plasma of CRC patients, both early and late stages, was almost fivefold higher than in HD (Figure 5). Even when analyzed samples belonged to other CRC patient than those analyzed in functional assays, the increment of TGF-β production by CRC tumors could impact on NK cell phenotype and their respective functionality (32).

Taking into account that deregulation of NK cell receptor expression was associated with function detriment, we analyzed if the exhausted phenotype had a correlation with clinical outcomes in a retrospective study. Patients were followed for a median of 24.7 months and during this period 11/52 (21%) relapsed. As expected, clinical staging IV at diagnosis impacted on RFS (Figure 6A), while stage I, II, and III outcomes did not differ significantly. Considering that several receptors were underexpressed since early stages, we evaluated their expression in stages I, II, and III population (with the exclusion of stage IV patients) and correlate it with RFS. The patients were divided into groups of low and high NK cell receptor expression, using the median percentage value of each receptor as a cut-off. We found that expression of NKp46-activating receptor impacted on RFS by significant dissociation of Kaplan–Meier curves (p = 0.0261) (Figure 6B). Within stage IV, 100% (5/5) of patients progressed, independently of NKp46 status; nevertheless, it could be further analyzed if receptor expression had relevance in progression-free survival (PFS). Our data indicate that NKp46 inference warrants more investigation and validation with a larger cohort as well as longer follow-up, as it could have prognostic implications.

To identify immune strategies to restore CRC-NK functionality, we addressed different in vitro treatments, including the pre-incubation of tumor cells with cetuximab and NK cell stimulation with lenalidomide, IL-2, or IL-15. Pre-treatment of DLD-1 cells with cetuximab significantly augmented tumor cell lysis mediated by CRC-NK (Figure 7A, left panel). Still, ADCC response of CRC-NK cells reached lower levels than HD-NK, in both early and late stages (Figure 7A, middle panel), probably due to diminished molecule density of CD16 (FcγR) in CRC-NK (Figure 7A, right panel). However, addition of IL-2 or IL-15 to NK cells completely restored ADCC capacity to HD-NK levels (Figure 7B). In this respect, cetuximab combined with cytokines allowed CRC-NK to reach normal lytic activity. Pre-stimulation with IL-2 or IL-15 modulated several NK cell receptors that could contribute to enhance functionality (Figure S2 and Table S3 in Supplementary Material). Lenalidomide pre-treatment did not improve NK cell activity (data not shown). Finally, we performed a dynamic lysis assay in which low lytic activity due to DNAM-1, NKG2D, and NKp30 blockade was reestablished by cetuximab combined with IL-2 pre-stimulation (Figure 7C). Cetuximab itself did not have any effect on the CI of DLD-1 cells due to their KRAS mutational status (data not shown); the antibody acted only through the immune pathway. We observed a greater relative increase in functionality of CRC-NK than HD-NK cells in response to cytokines and mAb joint stimulus (almost fourfold higher), while the relative increase in response to independent activation with cytokines or cetuximab did not differ between CRC patients and HD (Figure S3 in Supplementary Material).