Plasma from 44 HNSCC patients (mean age: 65 years) and 12 age-matched healthy volunteers (mean age: 55 years) was analyzed. Patients’ demographics and tumor characteristics are summarized in Table 1. Tumor staging was assigned according to the TNM classification of the International Union Against Cancer. Thirty of the HNSCC patients were studied at primary diagnosis and 14 patients with recurrent/metastatic (relapse) disease. From the patients with primary diagnosis, blood samples of 26 patients were taken before therapeutic treatment (treatment naïve), while 4 patients had surgery and/or radiotherapy before sample collection. Ten patients with recurrent disease were under palliative therapy during sample collection, while four patients were currently not under therapy.

Two male rhesus macaques (age 6 years) from the German Primate Center breeding colony (German Primate Center, Göttingen) were used for the in vivo experiments (pilot study and apheresis). All experimental procedures were done under inhalation anesthesia. The animals were i.v. injected with sMICA*04 at 100 µg/l blood volume (blood volume corresponds to approximately 7% of body weight). Plasma volume was calculated based on individual hematocrit. For the apheresis, animals were connected to a Life18™ apheresis unit equipped with an adsorber cartridge (anti-MICA antibody covalently coupled to sepharose Cl-4B at 0.95 mg AMO1/g sepharose) via a double lumen catheter in the vena femoralis. Apheresis was started 15 min post-injection with a flow-rate of 25–30 ml/min until three complete plasma volume exchanges were reached. During apheresis, animals got heparin as anticoagulant (50 U/kg). Blood and urine samples were taken every 15 min.

Primary NK cells were purified (>95% pure) from buffy coats of healthy volunteers kindly provided by the German Red Cross Blood Service (Institute for Transfusion Medicine and Immunohematology, Medical School, Goethe University Frankfurt, Germany) with written donor approval and approval by the Ethics Committee of the Goethe University Frankfurt (Frankfurt am Main, Germany, permit #329/10). Therefore, peripheral blood mononuclear cells (PBMCs) were isolated by a density gradient with Biocoll (Biozol, Germany) followed by untouched NK cell isolation using magnetic cell separation with the human NK cell isolation kit (Miltenyi Biotec, Germany). NK cells were expanded and activated for 7 days in X-Vivo10 medium (Lonza, Switzerland) supplemented with 5% human serum (Life Technologies, USA), 1,000 IU/ml IL-2 (Promokine, Germany), and NK cell activation/expansion kit (Miltenyi Biotec, Germany). The cell lines FaDu (HTB-43, human pharynx squamous cell carcinoma cells), CAL27 (CRL-2095, human tongue squamous cell carcinoma cells), and Detroit562 (CCL-138, human pharyngeal carcinoma cells) were purchased from the American Type Culture Collection. HNSCC cell lines were cultured in MEM medium (Life Technologies, USA). SiHa (grade II, human cervix squamous cell carcinoma) cells kindly provided by Dr. Adelheid Cerwenka (DKFZ, Heidelberg, Germany) were cultured in DMEM medium (Life Technologies, USA). All media were supplemented with 10% FCS (PAA and PAN Biotech, Germany), 1% penicillin/streptomycin (Life Technologies, USA), 2 mM l-glutamine (Life Technologies, USA), and 1 mM sodium pyruvate for HNSCC cell lines (Life Technologies, USA).

The concentrations of IFNγ, TNFα, IL-1, IL-6, IL-8, IL-10, MCP-1, RANTES, MIP-1α, MIP-1β, and TGF-β1 in plasma of HNSCC patients and healthy donors were determined using the BD™ CBA Flex Set System (BD Bioscience, USA), according to manufacturer’s instructions. Samples were diluted 1:30 for TGF-β1 or 1:4 for all other sets and tested in duplicates. Samples were measured with the BD FACS Array™ and analyzed with FCAP Array Software (BD Bioscience, USA). Cytokine levels of HNSCC patient samples were normalized to healthy controls, and data are represented as x-times median control.

For analyses of sNKG2DL levels in the plasma of HNSCC patients, healthy donors, rhesus monkeys, or tumor spheroid supernatants [concentrated 10-fold using Amicon centrifugal filter units (Merckmillipore, Germany)], samples were diluted 1:2 and measured in duplicates. Soluble NKG2DLs were analyzed as described (27, 36–38). Briefly, ELISA plates (Greiner Bio-One, Austria) were coated with mouse anti-human MICA (clone AMO1), anti-human MICA/B (clone BAMO1), anti-human ULBP1 (clone AUMO5), anti-human ULBP2 (clone BUMO1), or goat anti-human ULBP3 antibodies (AF1517; R&D Systems, USA). After saturation with BSA blocking solution (Candor, Germany), the plates were incubated with soluble MICA*04 or recombinant Fc-fusion proteins of human MICB, ULBP1/2/3 (1599-MB-050, 1380-UL-050, 1298-UL-050, 1517-UL-050; R&D Systems, USA) as standards. For detection and quantification of soluble ligands, mouse anti-human MICA/B (clone BAMO3), anti-human MICB (clone BMO2), anti-human ULBP1 (clone AUMO2), or anti-human ULBP2 (MAB1298; R&D Systems, USA) antibodies were used in combination with a goat anti-mouse IgG2a horseradish peroxidase (HRP)-conjugated antibody (1080-05; Southern Biotechnologies, USA). For sULBP3, the mouse anti-human ULBP3 antibody (clone CUMO3) was used in combination with a goat anti-mouse IgG1 HRP-conjugated antibody (1070-05; Southern Biotechnologies, USA). Following visualization with 3,3′,5,5′-tetramethylbenzidine substrate (KPL, Germany), signals were measured in a microtiter plate reader (λ = 450 nm) and analyzed with Prism 6 software (GraphPad, USA). Data of tumor spheroids were calculated as mean ± SEM per 10,000 cells (per spheroid) of three independent experiments (n = 3). Data of plasma samples were calculated as mean ± SEM. Statistical analysis was performed using one-way analysis of variance (ANOVA).

Soluble NKG2DLs were depleted from plasma of HNSCC patients, healthy donors, or CAL27 cell culture supernatants (collected on d3 of culture and 10× concentrated). Samples were incubated with a cocktail of Protein A Dynabeads (Life Technologies, USA) coupled to anti-human MICA (clone AMO1), anti-human MICB (MAB1599), anti-human ULBP1 (MAB1380), anti-human ULBP2 (MAB1298), or anti-human ULBP3 (MAB1517; all R&D Systems, USA) antibodies.

For the depletion of sMICA, the anti-MICA antibody (clone AMO1) was covalently coupled to MACSxpress beads (Miltenyi Biotec, Germany). Human plasma of healthy donors supplemented with 20 ng/ml sMICA*04 or culture supernatants of C1R cells transfected with MICA*01/04/07 and MICA*08 (38) were incubated with antibody-coupled beads overnight. Beads and bound sNKG2DLs were removed from the supernatants by magnetic bead separation on a MACSxpress separator (Miltenyi Biotec, Germany) following centrifugation.

Primary human NK cells of different healthy donors were pre-incubated overnight with sNKG2DL-depleted or non-depleted plasma of HNSCC patients or healthy donors’ plasma as controls. SiHa target cells were labeled with carboxyfluorescein succinimidyl ester (CFSE; Life Technologies, USA) according to manufacturer’s instructions. Labeled SiHa target cells and pre-treated NK cells were cocultured at an E:T ratio of 15:1 for 2 h. NK cells were labeled with mouse anti-human CD45-APC antibodies (clone 5B1, Miltenyi Biotec, Germany), and live/dead cell discrimination was achieved by staining with SytoxBlue (Life Technologies, USA) prior to the measurement on a BD FACSCanto II instrument equipped with a 96-well plate high throughput sampler (HTS). Cellular event counts were analyzed with FlowJo software. The percentage of cell lysis was calculated by evaluation of viable (CFSE⁺/SytoxBlue⁻) and dead (CFSE⁺/SytoxBlue⁺) target cells. Data were normalized to mean of healthy control and are represented as mean ± SEM of triplicates of a representative experiment. Statistical significance was determined by one-way ANOVA.

Cytotoxicity assays were performed as previously described (39). In brief, CFSE-labeled HNSCC and SiHa solid tumor spheroids were co-incubated with primary human NK cells at appropriate E:T ratios (5:1) in medium without IL-2 for 48 h. For inhibition, NK cells were pre-incubated overnight with non-/sNKG2DL-depleted CAL27 SN. For endpoint analyses, single tumor spheroids were dissociated, NK cells were labeled with a mouse anti-human CD45-APC antibody (clone 5B1; Miltenyi Biotec, Germany), and live/dead cell discrimination was achieved by staining with SytoxBlue (Life Technologies, USA). Complete samples were measured on a FACS Canto II instrument equipped with a HTS and cellular event counts were analyzed with FlowJo software. The percentage of cell lysis was calculated by evaluation of viable (CFSE⁺/SytoxBlue⁻) and dead (CFSE⁺/SytoxBlue⁺) target cells. Data are represented as mean ± SEM of eight spheroids. Statistical significance was determined by one-way ANOVA.

For infiltration studies, tumor cells and CAL27 SN or sMICA*01 pre-treated NK cells were seeded 1:1 into agarose coated 96-well plates. Individual spheroids (n > 8) were collected after 48 h and embedded into OCT compound (Sakura). Cryosections (7 µm) of solid tumor spheroids were stained with rabbit anti-human cleaved caspase-3 (clone 5A1E; Cell Signaling Technology, USA) and mouse anti-human CD45 (clone 2B11 + PD7/26; DAKO, Germany) and analyzed by standard two-step polymer methods with Envision kits (DAKO, Germany). 3,3′-diaminobenzidine (DAB) and permanent red or 3-amino-9-ethylcarbazol (AEC) were used as chromogen. Counterstaining was performed using hematoxylin (DAKO, Germany) according to the manufacturer’s instructions. Phase-contrast pictures of solid spheroids were analyzed by triple measurement of independent spheroids (n = 8) for each condition by ImageJ software (40). Tumor spheroid volume was calculated based on area analysis of solid spheroids assuming a perfect sphere. The number of infiltrated NK cells was calculated as percentage of CD45⁺ NK cells from whole cell numbers in spheroid cryosections. Data are represented as mean ± SEM of eight tumor spheroid cryosections and statistical analysis was performed using unpaired Student’s two-tailed t-test.

Cryo-preserved PBMCs of healthy volunteers or HNSCC patients and corresponding disintegrated primary tumor samples were stained with a cocktail of mouse anti-human CD3-APC.Cy7 (clone HIT3a), anti-human CD4-PacificBlue (clone OKT4), anti-human CD8a-AlexaFluor700 (clone HIT8a), anti-human CD16-FITC (clone 3G8), anti-human CD56-PE.Cy7 (clone HCD56) (all, BioLegend, USA), anti-human CD45-Pe.Flour610 (clone HI30; eBioscience, USA), and anti-human CD314-PE (clone BAT221; Miltenyi Biotec, Germany). For cell viability, Aqua Dead Cells stain (Thermo Fisher Scientific, USA) was used. Cellular events were measured on a Gallios cytometer and analyzed using Kaluza software (both Beckmann Coulter, USA). Percentage of NK cell subsets and mean fluorescent intensity ratios for NKG2D expression were calculated from CD45⁺ lymphocyte fraction.

Paraffin sections of primary HNSCC tumors were stained with anti-human CD3 (clone SP7; Thermo Scientific, USA), anti-human CD8 (clone C8/144B; Dako), anti-human CD20 (clone L26; Dako), anti-human CD68 (clone PG-M1; Dako), anti-human CD56 (clone 123C3; Invitrogen), and anti-human CD163 (clone MRQ-26; Cell Marque) antibodies using the BOND-MAX automated IHC-stainer (Leica Biosystems) according to the manufacturer’s instructions in the Institute of Pathology (University of Cologne, Germany). Numbers of infiltrating cells in the invasive margin and tumor core of HNSCC sections were analyzed separately. Cell counts per high power field (HPF) at 400× magnification were assessed by an experienced pathologist in five HPF of invasive margin and tumor core of every slide, respectively, and the mean positive cell count per HPF was calculated.

Statistical analyses were performed using Prism 6 software (GraphPad, USA). Data are presented as mean ± SEM as indicated. Statistical significance was evaluated using one-way ANOVA with repeated measurements and post hoc test after Bonferroni or unpaired Student’s two-tailed t-test as indicated. Differences were considered statistically significant at p < 0.05.

Previously, we could show a correlation of high sMICA plasma levels and impaired NK cell cytotoxicity in patients with HNSCC (33, 41). Since these studies only investigated sMICA in NKG2D-dependent tumor immune escape, in the current study, we determined the levels of all the shed NKG2DLs (sMICA, sMICB, and sULBP1–3) in plasma samples of 44 HNSCC patients (patients’ characteristics are summarized in Table 1).

Significantly elevated levels of individual sNKG2DLs were detected in patients’ plasma when compared to those observed in the plasma of 12 age-matched healthy donors (Figure 1A). Moreover, the burden of sNKG2DLs in patients with advanced disease vs. healthy donors was significantly increased, whereas no significant difference between primary and relapsed HNSCC was observed (Figure 1B). Comparison of treatment naïve patients with patients currently under palliative chemo- or radiotherapy or after tumor resection showed a therapy induced reduction of sNKG2DL levels (Figure S1A in Supplementary Material). Notably, sNKG2D ligands were found to peak in the plasma of patients with advanced disease (Figure 1C), suggesting that the total level of sNKG2DLs represents an important clinical parameter (Figure S1B in Supplementary Material).

To further assess the immune status of HNSCC patients, we analyzed the plasma levels of indicative cytokines and chemokines. Elevated plasma levels could be detected for the immunosuppressive and proinflammatory factors TGF-β1, MIP-1β, RANTES, IL-8, MCP-1, and IL-6. However, compared to the plasma levels of healthy controls, HNSCC patients showed no significant differences in the levels of IFN-γ, TNF-α, IL-10, IL-1β, and MIP-1α (Figures 2A,B). Moreover, aberrant cytokine or chemokine levels were not correlated to disease stage (Figure 2A) or primary vs. relapsed HNSCC, except for TGF-β1 showing an increase between stage I and IV (2677 vs. 5358 pg/ml) (Figure 2C).

To investigate the inhibitory potency of sNKG2DLs on NK cell cytotoxicity, we performed cytotoxicity assays with primary human NK cells after pre-incubation with plasma from HNSCC patients, healthy controls, or anti-NKG2D blocking antibodies. As target cells, a human cervix carcinoma cell line (SiHa) was used whose killing is strongly dependent on NKG2D. Two-thirds of HNSCC plasma samples significantly diminished target cell killing compared to plasma of healthy donors (Figure 3A). To verify the sNKG2DL-dependency, we depleted sMICA/B and sULBP1-3 from patients’ plasma with antibody-coupled beads. The ligands were removed with efficacies between 61.9 and 93.2% (Figure 3B). Depletion of sNKG2DLs efficiently restored NK cell cytotoxicity (Figure 3C; Figure S2A in Supplementary Material), whereas depletion of healthy plasma had no effect. Thus, NK cell inhibition is mostly due to blocking and downregulation of the NKG2D receptor on NK cells (Figures S2B,C in Supplementary Material). Since we found high levels of TGF-β1, which is known to inhibit NKG2D (42, 43), we correlated the cytotoxic capacity of pre-incubated NK cells with the plasma levels of sNKG2DLs and TGF-β1. High levels of both, sNKG2DLs (>2.5 ng/ml; median: 7.6 ng/ml) and TGF-β1 (>3,000 pg/ml; median: 5,319 pg/ml) showed an additive effect on NK cell inhibition when compared to plasma with low sNKG2DLs (<2.5 ng/ml; median: 1.4 ng/ml) and TGF-β1 (<3,000 pg/ml; median: 2,045 pg/ml) levels (Figure 3D). This is in accordance with our previous results demonstrating an additive effect of sMICA and TGF-β1 on NKG2D-dependent inhibition of NK cell cytotoxicity in HNSCC (33, 41). Nevertheless, the sole depletion of sNKG2DLs was sufficient to restore NK cell cytotoxicity, assuming shedding of ligands as a key mechanism of NKG2D-dependent tumor immune escape in HNSCC. Moreover, plasma samples of one representative patient showed significantly reduced levels of sNKG2DLs and TGF-β1 after surgery (Figure 3E), which were correlated with increased NK cell cytotoxicity (Figure 3F). These findings demonstrate a direct link between tumor burden, NKG2DL shedding and TGF-β1 release and was supported by a patient with recurrent disease (Figures S2D,E in Supplementary Material).

Next, we analyzed NKG2D-dependent tumor immune escape in tumor spheroids, which resemble many features of poorly vascularized and avascular regions of solid tumors or micrometastases and allow for systematic studies of molecular parameters in a highly controlled microenvironment (39, 44, 45). Tumor spheroids were grown from HNSCC (FaDu, CAL27, and Detroit562) and SiHa (39) cell lines, which show characteristic NKG2DL expression patterns and are highly susceptible to NKG2D-dependent killing (Figures S3 and S4A,B in Supplementary Material).

To investigate the impact of sNKG2DLs on NK cell cytotoxicity, tumor spheroids were cocultured for 48 h at an E:T ratio of 5:1 with NK cells pre-incubated with a cocktail of sNKG2DLs (CAL27 SN: 27.63 ng/ml sNKG2DLs) or sNKG2DLs-depleted CAL27 supernatant (sNKG2DLs: 2.66 ng/ml) (Figures 4A,B). NK cells in basal culture medium served as controls. NK cells efficiently destroyed tumor spheroids (Figure 4A; Figure S4C in Supplementary Material). By contrast, in the presence of high levels of sNKG2DLs, NKG2D-dependent killing was reduced by 25% for FaDu, 19% for CAL27, and 46% for SiHa tumor spheroids (Figure 4A). Importantly, impaired tumor spheroid destruction and NKG2D-downregulation could be completely restored by specific depletion of sNKG2DLs (Figure 4A; Figure S4D in Supplementary Material).

Since antitumor activity of NK cells correlates with infiltration capacity into tumors (46, 47), we investigated NK cell infiltration into tumor spheroids in the absence and presence of high levels of sNKG2DLs. Untreated NK cells (kill CO) or NK cells pre-treated with CAL27 SN or purified sMICA (20 ng/ml) were mixed with HNSCC cells at a 1:1 ratio. Tumor spheroids were monitored by light microscopy for 48 h. In the absence of sNKG2DLs, NK cell killing was accompanied by massive infiltration into tumor spheroids indicated by staining of tumor spheroid cryosections with CD45-specific antibodies (Figure 4C; Figures S5A,B in Supplementary Material). Roughly 66, 38, and 65% of the cells comprising the FaDu, CAL27, or SiHa tumor spheroids, respectively, were NK cells (Figure 4D). In contrast, under inhibitory conditions (in the presence of sMICA or CAL27 SN), NK cell cytotoxicity, shown as tumor spheroid volume reduction (Figure S5C in Supplementary Material), was inhibited. This inhibition was accompanied by drastically reduced NK cell infiltration with only 26, 26, or 19% of the cells of FaDu, CAL27, or SiHa tumor spheroids being NK cells, respectively (CAL27 SN; Figure 4D). Interestingly, the same amount of sMICA alone was sufficient for NK cell inhibition, with 38, 21, and 29% of NK cell infiltration for FaDu, CAL27, and SiHa tumor spheroids, respectively, when compared to the cocktail of shed ligands in the CAL27 SN, suggesting that the total burden of NKG2DLs might be crucial for NK cell dysfunction in cancer patients. Additionally, infiltrated pre-treated NK cells were often double positive for CD45 and cleaved caspase-3 compared to untreated NK cells. Therefore, we speculate that under inhibitory conditions NK cells might proliferate less and become apoptotic. Moreover, both NK cell subtypes (CD56^bright/CD16^+/−; CD56^dim/CD16⁺) could infiltrate into tumor spheroids (Figure S5D in Supplementary Material).

To further investigate tumor infiltration in vivo, primary HNSCC tumor sections were stained for CD3⁺ and CD8⁺ T cells, CD20⁺ B cells and CD56⁺ NK cells, as well as CD68⁺/CD163⁺ macrophages (Figures 5A,B). Infiltrating cells were analyzed separately for the invasive margin (IM) or tumor core (TC). T and B cells were predominantly found in the IM (CD3 T cells: mean 71 cells/HPF; CD8 T cells: mean 50 cells/HPF; CD20 B cells: 200 cells/HPF). In contrast, only a small proportion of lymphocytes could infiltrate into the tumor core (CD3 T cells: mean 17 cells/HPF; CD8 T cells: mean 20 cells/HPF; CD20 B cells: 4 cells/HPF). No differences could be observed for CD8⁺ T cells when compared to the whole CD3⁺ T cell population. Moreover, tumor-associated macrophages were equally distributed throughout the tissue. Interestingly, only very low numbers or a complete absence of NK cells could be found in the tumor tissues (IM/TC: mean 2 cells/HPF). The low infiltration potential of the different lymphocyte populations is indicative for the insufficient tumor rejection in HNSCC patients. For further analysis, NK cells of HNSCC patients’ PBMCs and corresponding dissociated tumor samples were subjected to flow cytometry experiments. The total number of circulating NK cells in the CD45⁺ lymphocyte population of PBMCs was nearly unchanged compared to healthy controls (Figure 5C). Regarding regulatory (CD56^bright/CD16^+/−) and cytotoxic (CD56^dim/CD16⁺) NK cells, HNSCC patients showed a slight shift toward the CD56^bright NK cell subpopulation. Analysis of tumor samples confirmed the histological staining showing that only a minor proportion of tumor-associated lymphocytes were NK cells. Interestingly, the infiltrated NK cells were predominantly from the regulatory subpopulation, whereas cytotoxic NK cells were nearly absent. Moreover, analysis of NKG2D expression levels showed a small increase compared to healthy controls of maximum twofold for the CD56^bright population (Figure 5D).

Our data suggest that the cumulative sNKG2DL level is indicative for the extent of NKG2D-dependent tumor immune escape. Consequently, sNKG2DL plasma depletion by adsorption apheresis might be a clinical intervention strategy to significantly boost antitumor efficacy of autologous and, if combined with cell-based therapies, adoptively transferred or allogeneic immune cells. For proof-of-concept studies, we employed a human sMICA-specific antibody, raised by immunizing BALB/c mice with a mixture of MICA*01 and MICA*04 (38), with an epitope in the α1-helical domain (Figures S6A,B in Supplementary Material). To test its depletion efficacy, human plasma of 20 healthy donors supplemented with recombinant sMICA*04 (production and purification of sMICA*04, see Figures S6C–E in Supplementary Material) or cell culture SN containing allelic sMICA variants were incubated with antibody-coupled magnetic beads. The antibody showed a depletion efficacy of 89–97% of sMICA allelic variants frequently found in the Caucasian population (48) independent of the donor immune status (Figures 6A,B). Moreover, beside sNKG2DLs also exosomes containing NKG2DLs (MICA*08, Figure 6B) (49) could be efficiently removed.

For preclinical validation of adsorption apheresis, we used a rhesus macaque (Macaca mulatta) model. Human and rhesus MICA display a sequence identity of 79%, human and rhesus NKG2D a sequence identity of 94% (Figures S6F,G in Supplementary Material). In a pilot study, two rhesus monkeys were injected with 100 µg sMICA*04 per liter blood volume, and sMICA plasma levels were analyzed over 3 h. About 15–25% of the initially injected sMICA could be found in the plasma 15 min after injection. An exponential decrease of sMICA could be observed during the first 150 min reaching a stable plasma level with diffusion equilibrium of about 3–5 ng/ml after 3 h, which resembles the plasma levels found in cancer patients (Figure S7A in Supplementary Material). Notably, no renal clearance, proteolytic degradation of sMICA (Figures S7B,C in Supplementary Material), or immune reaction against sMICA was observed (Figure S7D in Supplementary Material). These data indicate that the decreasing sMICA plasma levels reflect equilibrium formation of sMICA’s body distribution.

For adsorption apheresis, two animals were injected with sMICA*04, and the apheresis was started 15 min post-injection until three exchanges of the animals’ plasma volume were reached. Analysis of sMICA plasma levels showed a nearly complete depletion of sMICA from the plasma after only three plasma volume exchanges with 1.44 ± 0.06 and 3.48 ± 1.79% remaining sMICA*04 when compared to the endpoints reached by body distribution without treatment (24.17 ± 0.09 and 27.98 ± 0.46% remaining sMICA*04 in the two animals, respectively) (Figures 6C,D). Notably, since the two monkeys were healthy without signs of immune challenge at the time of the experiment, the basal level of NKG2D on the monkeys’ T and NK cells was not affected by human sMICA (Figures S7E–G in Supplementary Material). In conclusion, we suggest that adsorption apheresis of sNKG2DLs from patients’ plasma might be a promising combinatorial medical approach to overcome NKG2D-dependent tumor immune escape and to improve the efficacy of (cellular) cancer immunotherapy.