The enrolled patients were featured by a class A liver cirrhosis (Child–Pugh) and an early stage of HCC (Barcelona Clinic Liver Cancer Stage A) diagnosed through ultrasonography/computed tomography or MRI in specific cases. All the patients were HBV- and HIV-negative. Additional information on the study population are shown in Supplementary Table S1 .

Tumor and non-tumorous specimens were obtained from 11 HCC patients infected with hepatitis C virus, 6 patients with alcohol-associated HCC, one NASH and HCC patient, and control samples from the liver resection of 7 patients with colorectal metastasis. Functional and metabolic analyses of tissue-infiltrating NK cells were conducted on 12 matched HCC–non-tumorous samples and controls. Eight paired HCC– non-tumorous samples were used to evaluate the effect of in vitro treatment on cell metabolism.

The study underwent local ethical committee approval [Comitato Etico Indipendente Area Vasta Emilia Nord (AVEN) of the AOU of Parma, Parma, Italy]. Signed informed consent for each patient was obtained for them to take part in the study.

Thawed liver- and tumor-infiltrating lymphocytes from the 11 HCC and HCV patients and from the 6 liver-resected control patients were stained with anti-CD3-PeCy7 (Biolegend, San Diego, CA, USA), anti-CD16 FITC (BD Bioscience, Franklin Lakes, NJ, USA), anti-CD56 PECF594 (BD), and then 7-amino-actinomycin D [7AAD PerCPCy5.5 vitality dye, for discriminating NK cells in normal liver-infiltrating NK cells (NLINK), tumor-infiltrating NK cells (TINK), and NK cells infiltrating hepatic tissue around HCC (LINK)]. NK cell sorting was performed with FACSAria III Cell Sorter (BD) by selecting CD3-negative, CD56-positive, CD16-positive, or CD16-negative subpopulations.

RNA was extracted from the isolated NK cells with the RNase Free DNase I Kit (Norgen Biotek), and the RNA concentration was measured using a Nanodrop spectrophotometer, in accordance with the manufacturer’s instructions. The RNA integrity was detected with a Bioanalyzer2100 system (Agilent Technologies). Transplex Whole Transcriptome Amplification (WTA2) and GenElute PCR Clean-Up kits (Sigma) were employed to amplify and then purify the total RNA, respectively, following the manufacturer’s protocol. cDNA labeling was performed by SureTag DNA Labeling kit (Agilent Technologies) and then hybridized to 60-bp oligonucleotide whole human genome arrays (Human GE 8x60K v2 SurePrint G3, Agilent Technologies), following the manufacturer’s instruction. The microarrays were scanned with an Agilent dual-laser DNA scanner. The Agilent Feature Extraction software, v.7.5, was employed with default settings to achieve normalized expression values from raw data. The expression data are available at the National Center for Biotechnology Information Gene Expression Omnibus: GSE183349.

GeneSpring software package GX 13.1 (Agilent Technologies) was employed for quality control, data normalization (75th percentile procedure), and preliminary microarray data analysis. The probe signals in at least four replicates for each condition were kept for subsequent analyses. Benjamini–Hochberg-corrected ANOVA for multiple testing [false discovery rate (FDR), ≤0.05] was employed to track differentially expressed genes (DEGs) by comparing LINK-, TINK-, and NLINK-derived samples. Student–Newman–Keuls post-hoc test was used to determine the DEGs in the three patient groups.

Median baseline transformation was applied before determining the unsupervised hierarchical clustering of the samples with Ward’s linkage and Euclidean distance. In order to reduce data dimensionality, principal component analysis (PCA) was used with orthogonal transformation to cluster sample populations.

Gene Set Enrichment Analysis (GSEA) was employed on the detected probes in order to pinpoint significant pathways enriched in downregulated and upregulated probes and to analyze the gene profiles with other studies. Molecular Signature Database C2, canonical pathways version 6, was employed for statistical analysis with the permutation type (“gene set” for fewer than seven replicates); the default settings for all other options were kept. Significantly enriched gene sets were obtained with a FDR lower than 0.25, which was determined using 1,000 permutations of gene set and Signal2Noise as metric.

For all flow cytometry analysis, after the exclusion of cell aggregates, gating was performed on lymphocytes, and dead cells were subsequently excluded (7-AAD or live/dead as appropriate); CD3, CD56, and CD16 were used to select NK cells that were identified as CD3− CD56+.

Mitochondrial membrane potential was detected on infiltrating NK cells through the JC-1 probe to measure the mitochondrial membrane potential (ThermoFisher). Anti-CD3 APCCy7 and anti-CD56 APCR700 were used to stain LINK, TINK, and NLINK, and then the cells were treated for 10 min at RT with JC-1 (2.5 μg/ml) before the FACS analysis. Finally, viability probe 7-AAD was used for cell staining, and FACS Canto II was used to acquire samples. Mitochondrial depolarization was measured in NK cells by quantifying the percentage of FL1high/FL2low cells (JC-1 staining) detected in the different samples.

The uptake of fluorescent glucose was assessed in order to define the glycolysis capacity. After washing with 1× phosphate-buffered saline (PBS) to withdraw endogenous glucose, the infiltrating lymphocytes were marked for 30 min at 37°C with the analog 2-NBDG {2-deoxy-2-[(7-nitro-2,1,3-benzoxadiazol-4-yl) amino]-D-glucose, ThermoFisher} at 40 μM in RPMI without glucose and added with 10% dialyzed fetal bovine serum, and then staining was caried out with anti-CD3, anti-CD56, and 7-AAD probes. The frequency and median fluorescence intensity (MFI) of 2-NBDG NK-positive cells were measured.

The autophagy potential was analyzed through the Cyto-ID Autophagy kit (Enzo Life Sciences, NY, USA) that detects autophagic vesicles and reveals autophagic flux in chloroquine diphosphate treatment to inhibit lysosomes and live cells (overnight at 30 µM), through a probe, which accumulate within the autophagic vacuole.

Changes in Cyto-ID expression were measured with or without chloroquine-mediated blocking of the autophagosome turnover, which hampers the fusion of autophagosome–lysosome vesicles, thus preventing their degradation.

On the following day, the NK cells were washed and resuspended in 1× PBS in order to eliminate any trace of phenol red. The Cyto-ID Autophagy kit was employed after anti-CD3 and anti-CD56 staining, according to the manufacturer’s instructions, for the final acquisition on a FACS Canto II flow cytometer. Data were expressed as Cyto-ID MFI in NK cells from different groups.

After thawing, the liver- and tumor-infiltrating NK cells (7 NLINK, 12 LINK, and 12 TINK) were stained with surface antibodies (anti-CD3-Alexa Fluor700 and anti-CD56-PerCp). Then, for anti-phosphorylated-p38 assay, Fixation/Permeabilization Solution Kit (Cytofix/Cytoperm BD) for intracellular staining was employed, in accordance with the manufacturer’s protocol and the antibody specific for the phosphorylated form of the human p38 mitogen-activated protein kinase (MAPK; pT180/pY182)PE-Cy7 (BD). Samples were acquired with a FACSCANTO II BD and analyzed with DIVA (BD) and Flowjo software. Data were expressed as phosphorylated-p38 MFI in NK cells from different groups.

IFNγ and TNFα cytokine protein expression in NK cells was assessed after 4 h of phorbol 12-myristate 13-acetate (PMA, 50 ng/ml) and ionomycin (1 µg/ml) stimulation. After 1 h, Brefeldin A (BFA, 10 µg/ml) was supplemented. Anti-CD3 PE (BD), anti-CD56 PECF594 (BD), and anti-CD16 FITC (BD) were used to stain the cells before fixing with medium A reagent and then permeabilization with medium B reagent (Nordic Mubio), according to manufacturer’s instructions. Cytokine measurements were assessed by intracellular cytokine staining (ICS) with monoclonal antibodies specific for IFNγ (PerCp-Cy5.5, Biolegend) and TNFα (APC, Biolegend) and detected by FACS Canto II. In order to evaluate the activity of cytotoxic NK cells , CD107a degranulation marker was employed after PMA and ionomycin stimuli. Following the stimulus, the NK cells were incubated for 4 h with the antibody specific for CD107a (PE-Cy7 -BD) and the Golgi inhibitor, Brefeldin A (10 µg/ml). The results were displayed as the difference between cytokine- and CD107a-positive NK cells with or without stimulus. In order to study the adhesion ability of infiltrating NK cells toward K562 target cells, K562 cells were marked with a green fluorescent probe (CFSE, carboxyfluoroscein succinimidyl ester) and NK cells with antibodies specific for CD3 (APC-Cy7) and CD56 (APC-R700). To block CFSE staining, two volumes of cold fetal bovine serum were added, followed by washing thrice with Hanks’ balanced salt solution. The NK cells have been joined to target cells in a ratio of 1:5, respectively, and then they have been mixed by gentle vortexing and centrifuged for 3 min at 4°C (at 300 rpm). The cells were incubated at different timepoints (0, 15, and 30 min) at 37°C, fixed with 1 ml of Fixation Permeabilization Concentrate and Diluent (eBioscience), and then measured on FACS Canto II (BD). The NK cells which adhered to the K562 cells were analyzed, and the results were shown as mean fluorescence intensity related to the NK lymphocytes gated on target cells.

IFN-γ and TNF-α release from NK cells was assessed after IL-12 and IL-18 (5 ng/ml) overnight (O/N) stimuli in the presence or absence of specific p38 inhibitors. After 1 h of the stimulus, 10 µg/ml of BFA was supplemented for a further 3 h of culture. Anti-CD3 PE (BD), anti-CD56 PECF594 (BD), and anti-CD16 FITC (BD) were employed to stain the NK cells before fixing them with the medium A reagent and permeabilization with medium B reagent (Nordic Mubio) according to the manufacturer’s instructions. Cytokine production measurements were assessed by ICS after staining with monoclonal antibodies for IFNγ (PerCp-Cy5.5, Biolegend) and TNFα (APC, Biolegend) and subsequently run on FACS cytometry. The cytotoxic activity of NK cells was monitored through CD107a degranulation staining. Specifically, IL-12 and IL-18 were employed to stimulate the inhibitor-treated NK cells overnight (as described above). The cells together with target cells (K562) were then incubated with BFA and CD107 antibody (PECy7, BD) for the last 4 h.

The results are displayed as a ratio (fold change) of cytokine-positive NK cells detectable in inhibitor-treated vs . untreated cells. As control, the percentage of dead cells was evaluated in each inhibitor-treated sample by FACS in order to assess the cell toxicity.

Mitochondrial membrane potential (JC-1), glucose uptake, and autophagy potential by NK cells from LINK and TINK samples were evaluated, as described above, after O/N IL-12 + IL-18 (5 ng/ml) stimulation with or without the specific p38 inhibitors.

Data were presented as the ratio between the cells in the fluorescent channel FL1high and FL2low (JC-1 staining) and MFI (2-NBDG or Cyto-ID staining) detected in inhibitor-treated vs . untreated cells (fold change). As control, the percentage of dead cells was evaluated in each inhibitor-treated sample by FACS in order to assess cell toxicity.

The compounds tested were the SB203580 p38 inhibitor, used at 0.05–1 μM (Sellekchem), and doramapimod (BIRB-796), used at a concentration of 0.05–1 μM (Selleckem).

All assays were assessed on 8-color flow cytometer (FACS Canto II, BD), and the results were analyzed with FACS Diva and Flowjo software (BD). The frequency of CD56^DIM and CD56^BRIGHT NK cells was determined on the CD3-negative CD56-positive cells by evaluating the fluorescence intensity level of CD56 marker, and each parameter expression (carried out as percentage and MFI) was measured on total CD56 positive, CD56^DIM, or CD56^BRIGHT NK cells .

Statistics was assessed with GraphPad Prism v.7 software. Analyses of variance (F test) and of normality (Kolmogorov–Smirnov test) were initially evaluated. Subsequently, Mann–Whitney U-test, paired t-test, Wilcoxon matched-pairs test, and Pearson’s correlation were used. Moreover, inhibitory drug experiments and fold change upon inhibitory treatments were statistically calculated by Wilcoxon signed rank tests (as compared to a theoretical median of 1).

All statistics were two-tailed, and significance was observed as p <0.05.

NK cells (CD3^-, CD56⁺) were purified by flow cytometric cell sorting from tumor and liver tissue samples of patients with HCC arising in HCV infection (LINK and TINK, n = 11) and from liver control tissue samples (NLINK, n = 7). The analysis of variance on data of the detected probes (19,716 genes) identified 108 DEGs in NK cells from HCCs and controls. The hierarchical clustering of DEGs is depicted in Figure 1A . PCA showed a partial segregation among groups of patients and controls ( Figure 1B ).

We then focused on the comparison of NK cells infiltrating the HCC and the liver from the same patients and interrogated the Molecular Signatures database (http://www.broadinstitute.org/gsea). GSEA was conducted to understand the NK cell expression modifications in the tumor microenvironment. Gene sets related to mitochondrial functions and ROS detoxification system, glycolytic activity, ubiquitin-proteasome-mediated degradation system, and pathways related to cell cycle and DNA damage and repair (included the stress sensor p38-related pathway) appeared to be enriched and upregulated in TINK compared to LINK, as shown in Figure 1C .

Genes encoding the electron transport chain components and many subunits of complex I (NADH dehydrogenase), II (succinate dehydrogenase), III (cytochrome c reductase), IV (cytochrome c oxidase), and V (ATP synthase) were upregulated in HCC-infiltrating NK cells . Among the genes upregulated in tumor-infiltrating NK cells were genes related to glucose metabolism, particularly the aldehyde dehydrogenase family, aldolase complex, and phosphoglycerate kinase 1 ( Figure 1D ). Another group of transcripts upregulated in TINK encodes intracellular signaling and cell cycle control genes, such as MAPK14 and MAPK11 (subunits of p38 protein complex), TP53, p73, and p21 ( Figure 1D ).

Moreover, genes encoding the key components of the immune ubiquitin-proteasome pathway, including the 19S and 11S regulatory particles, the 20S proteolytic core, and the 26S proteasome subunits, resulted as upregulated in TINK. The detailed lists of genes may be found in Supplementary Table S2 .

GSEA was also applied for the comparison between gene expression profiles of TINK and NLINK, leading to the identification of enriched upregulated genes related to the similar pathways described above in the previous comparison ( Supplementary Table S3 ), confirming that tumor-infiltrating NK cells showed a misregulation in transcriptome profiles. The NK cells from tumor and liver tissue samples were analyzed by distinguishing the total NK cells from the CD56^DIM and CD56^BRIGHT subsets. A generalized upregulation thus appears to be the predominant feature of tumor-infiltrating NK cells .

Because of similar data reported for tumor microenvironment (32), we hypothesized that HCC-infiltrating NK cells are in a condition of nutrient deprivation, which is required to meet the increased energy and biosynthetic demands for NK cell expansion and anti-tumor functions (37, 38). To evaluate the possible intra-tumor dysregulation of NK cell metabolism, we first focused on mitochondrion and glucose metabolism.

Mitochondrial membrane depolarization and glucose import were measured. As shown in Figure 2A , tumor-infiltrating NK cells displayed a defective mitochondrial functionality, indicated by increased depolarization of the mitochondria compared to the non-tumorous counterpart and the controls. On the left side is the frequency of JC-1-positive cells, while on the right side is the median fluorescence intensity, showing that tumor-infiltrating NK cells exhibited a higher content of depolarized mitochondria in all NK cell subsets (TOTAL, DIM, and BRIGHT).

Glucose uptake capacity, described by the 2-NBDG analysis, was reduced in TINK compared to NLINK and LINK cells, both in terms of the percentage of metabolically active NK cells and in terms of fluorescence intensity, confirming the existence of metabolic impairment at different levels ( Figure 2B ). CD56^BRIGHT TINK showed the worst performance in glucose uptake capacity.

Then, we tried to clarify the mechanisms responsible for glycolytic impairment and to understand the upregulation of OXPHOS gene that was functionally ineffective. To this end, we performed a leading edge on GSEA data sets in order to extract a candidate list of regulatory genes. Notably, the majority of GSEA misregulated pathways highlighted an upregulation of MAPK11 and MAPK14 (α and β subunits of p38 complex).

Indeed p38 protein is a stress sensor kinase from MAPK family that can be activated by nutrient deprivation in the microenvironment. Based on this observation, we analyzed the phosphorylation status of p38-MAPK protein in the three study groups. A higher level of phosphorylated p38, observed as median fluorescence intensity, was detected in total NK cells derived from tumor samples and was particularly enhanced in the CD56^BRIGHT subset ( Figure 3A ). This observation confirms the activation of this specific pathway as indicated by the transcriptomic analysis.

A dimension reduction by t-distributed stochastic neighbor embedding (tSNE) on flow cytometry data of phosphorylated p38 MAPK (pT180/pY182) from the LINK and TINK groups was performed to represent the relative intensity of NK-cell p38 expression. Despite the fact that the distribution of events was almost similar in the total, CD56^BRIGHT, and CD56^DIM subsets in the two groups ( Figure 3B ), the color shades representing the protein level of p38 marker showed a higher expression in CD56^BRIGHT cells from the TINK samples (red indicating upregulation and blue indicating downregulation, Figure 3C ).

In order to better understand the metabolic assessment of HCC-infiltrating NK cells , the autophagy potential was analyzed.

The results of the transcriptomic analysis and validation suggested an engulfment of ROS clearance and accumulation of damaged mitochondria ( Figures 1C, D , 2A ). In addition, p38 has been implicated in autophagy repression in different cell types, including T cells, during starvation conditions such as those occurring in the tumor microenvironment (39–41).

Similarly, we studied the lysosome degradation pathway by evaluating Cyto-ID cationic incorporation into autophagic vesicles (pre-autophagosomes, autophagosomes, and autophagolysosomes) with or without chloroquine treatment.

As presented in Figure 4A , TINK displayed a significantly lower basal Cyto-ID expression in all NK cell subsets compared to LINK and a limited one to CD56^BRIGHT compared to NLINK. After treatment with chloroquine, the reduced accumulation of autophagic vesicles was even more evident in HCC-infiltrating NK cells compared to NLINK and LINK ( Figure 4C ).

Cyto-ID expression was negatively correlated to p38 levels in TINK samples, suggesting a strong relationship between p38 and autophagy repression in TINK ( Figures 4B, D ).

In order to define the functional capacity of tumor-infiltrating NK cells, cytokine production and degranulation (CD107a expression) activity were assessed. IFN-γ production was not significantly impaired in TINK compared to LINK and NLINK ( Figure 5A ).

Conversely, TNF-α ( Figure 5B ) and CD107a ( Figure 5C ) expression was significantly downregulated in TINK compared to liver-infiltrating NK cells (LINK and NLINK) ( Figure 5D ), particularly in the CD56^BRIGHT subset.

Dysfunctional NK adhesion to target K562 cells was detected in TINK, also suggesting an impairment in immunological synapse establishment by intratumor NK cells compared to NLINK and LINK ( Figure 5E ).

We then investigated the possible relationship between cytokine production/degranulation and metabolic functions ( Supplementary Figure S1 ). IFN-y, TNF-α, and CD107a expression did not correlate to the frequency of NK cells with depolarized mitochondria and autophagy function ( Supplementary Figure S1 , plots on the left) in LINK, whereas glucose uptake turned out to be positively correlated with IFN-y production ( Supplementary Figure S1B ).

By contrast, the metabolic status of intratumor NK cells showed a strong association with cytokine production and degranulation capacity. In particular, the frequency of NK cells with depolarized mitochondria exhibited a negative correlation with respect to IFN-y, TNF-α, and CD107a expression ( Supplementary Figure S1D ) suggesting that the mitochondrial functionality in TINK impacts on cytokine production and cytotoxic function. Glucose consumption and autophagy potential were also strictly connected to the immune functions of HCC-infiltrating NK cells ( Supplementary Figures S1E, F ).

To further analyze the relationship between p38 activation and TINK dysfunction, we assessed the effect on NK metabolic profile and anti-tumoral functions of two different chemical inhibitors targeting p38 ( Figures 6A–D ).

p38 blockade strongly increased the TNF-α ( Figure 6C ) and CD107a ( Figure 6D ) expression in TINK in all the NK subsets, whereas the effect on IFN-γ production was limited ( Figure 6A ). LINK and NLINK function did not benefit from p38 blockade, demonstrating the specificity of p38 inhibition in the tumor context. Representative dot plots showing CD107a, IFN-γ, and TNF-α expression in NK cells from NLINK, LINK, and TINK in untreated vs. p38 inhibitor-treated samples are shown in Supplementary Figure S2 .

Interestingly, although IFN-y production was less affected than the other functions, TINK samples with impaired function were more affected by p38 inhibition ( Figure 6B ). Moreover, treatment with p38 inhibitors enhanced the adhesion to target cells by TINK compared to LINK, but the effect did not reach statistical significance ( Figure 6E ). Finally, p38 blockade strongly reduced the NK cells with depolarized mitochondria in the tumor, in particular in the CD56^BRIGHT subset, while p38 inhibition did not show any effect in modulating the mitochondrial membrane potential in the liver counterpart ( Figure 7A ). In addition, the NK autophagy potential was significantly enhanced by p38 inhibition, especially in the chloroquine-treated tumor-infiltrating NK cells ( Figure 7B ) in all NK cell subsets. Differently though, glucose import was not affected by treatment both in liver- and tumor-infiltrating NK cells ( Figure 7C ).