Matched peripheral blood and tumor-free liver samples were obtained from patients at the Asklepios Hospital Barmbek (AKB) undergoing extended liver resection due to liver metastases following colorectal cancer, liver adenoma, cholangiocellular carcinoma or hemangioma (n=17), and at the University Medical Center Hamburg-Eppendorf (UKE) from patients undergoing liver transplantation (n=17) due to end-stage liver disease. Human control peripheral blood samples were obtained from 13 healthy individuals. Furthermore, residual amounts of anonymized peripheral blood samples (buffy coats) from 12 randomly-selected healthy blood donors recruited at the Institute for Transfusion Medicine at the UKE were used. All blood donors gave their general written consent to use their blood samples for scientific studies. The anonymized use of buffy coats complied with a vote by the ethics committee of the German Medical Association. Study protocols (PV4898, PV4081, and PV4780) were approved by the ethics committee of the medical association of Hamburg, and all study participants provided written informed consent. The demographics and clinical characteristics of study participants are summarized in Tables 1 , 2 .

Blood sample and liver tissue preparation were performed as previously described (43–45). Peripheral blood mononuclear cells (PBMCs) were obtained using density gradient centrifugation with Ficoll (Biochrom GmbH, Berlin, Germany). Tumor-free liver tissue was dissolved mechanically without enzymatic digestion for the extraction of hepatocytes and intrahepatic NK cells. For liver organoid generation and culture, liver tissue was mechanically and enzymatically digested and cultured as previously described (15, 46). In brief, the liver tissue was cut into small pieces and washed with DMEM (Gibco) supplemented with 10% fetal bovine serum (FBS) and 1% Penicillin/Streptomycin. The tissue pieces were then incubated at 37°C in EBSS (Gibco) with Collagenase D (Sigma-Aldrich) and DNase I (STEMCELL) until single cells were microscopically detectable. The cell solution was then filtered and washed. Single cells were seeded in 50µl droplets of BME2 (Basement Membrane Extract, Type 2, Pathclear) and organoid culture medium (46) and passaged 3 to 4 weeks after the digestion. Further passaging was performed approximately after 7 days of culture. Medium was changed every 2-3 days. PBMCs and liver-derived cells were cryopreserved, stored in liquid nitrogen and thawed for immediate experimental use.

After thawing, liver-derived cells and PBMCs were washed with phosphate buffered saline (PBS) and stained with surface antibodies and the live-dead markers Zombie NIR or Zombie Aqua (BioLegend) for 20 minutes in the dark at room temperature. All antibodies used in this study and corresponding dilutions are listed in Supplementary Table S1 . After washing with PBS, cells were either fixed immediately with 4% paraformaldehyde or fixed and permeabilized using the BD Cytofix/Cytoperm Kit (BD Bioscience), as recommended by the manufacturer, and subsequently stained with intracellular antibodies. The samples were measured using a LSRFortessa™ flow cytometer (BD Bioscience) or a Cytek Aurora (Cytek Biosciences). FlowJo™ version 10.7.1 was used to analyze the data. Gating strategies are shown in the supplementary information ( Figures S1A, B ).

Gene expression analyses and data processing were performed as previously described (23, 47, 48). To assess single-cell mRNA expression of selected genes, a Fluidigm platform consisting of C1, Juno and Biomark HD was used. Matched PBMCs and liver-derived cells from four liver resection patients and PBMCs of two healthy donors were sorted for live, CD3⁻CD14⁻CD19⁻CD56^brightCD16⁻ NK cells ( Figure S1A ) with a BD FACSAria Fusion (BD Bioscience) into a C1 Single-Cell Preamp Integrated Fluidic Circuit (IFC), 5-10 µm (Fluidigm). Cell lysing, reverse transcription, and preamplification were performed using the C1, according to company protocol (PN 100-4904 K1). 52 primers were used to quantify gene expression and are summarized in the supporting data ( Supplementary Table S2 ). Quantitative PCR was performed using the 96.96 Dynamic Array IFC for Gene Expression in the Biomark HD. All data analysis was performed with the Fluidigm Real-Time PCR Analysis software (Version 4.3.1). The Limit of Detection (LOD) was defined as a Ct value of 24 and subsequently all Ct values above 24 were set to the LOD. Gene expression levels are displayed as 2^(LOD-Ct), thus a value of 1 indicates a cell without detectable mRNA expression of the respective gene (49, 50). In the Fluidigm Real-Time PCR Software, all peak detection ranges of the melting curves were defined as the median temperature peak of all non-failed results of the corresponding primer pair +/-1.2°C. Other settings remained unchanged (Peak Sensitivity: 7; Peak Ratio Threshold: 0.8). The Ct value was set to the LOD for all reactions that were marked as failed under these settings.

Liver-derived cells were thawed in pre-warmed RPMI (Thermo Fisher Scientific) supplemented with 10% FBS (R10) (Biochrom GmbH) and 10 µg/ml DNase. Subsequently, after centrifugation and washing, cells were resuspended at a concentration of 2x10⁶ cells/mL in R10 supplemented with 1 ng/mL IL-15 and rested overnight at 37°C. Cells were washed with PBS and counted before being co-incubated with K562 cells at an effector to target cell (E:T) ratio of 10:1 for 5h at 37°C in the presence of an anti-CD107a-antibody (BV711) and Brefeldin A. Hereafter, cells were stained with surface antibodies and Zombie NIR, as well as stained intracellularly with an anti-TNF-α-antibody (PE-Cy7), as described above.

NK cell isolation was performed as recommended by the manufacturer (EasySep Human NK cell Isolation Kit, Stemcell Technologies) with fresh PBMCs from healthy donors. Isolated NK cells were stimulated for 12h with 1 ng/mL IL-15 at 37°C. Afterwards, NK cells were either (co-)incubated with or without Huh7 cells at an E:T ratio of 2:1 in direct contact or in indirect contact using a transwell insert (Sarstedt AG & Co. KG) with pore sizes of 1 µm for 6, 12, 24 and 48h. Finally, cells were stained with antibodies for flow cytometric analysis, as described above. For blocking experiments, Huh7 cells were co-incubated with an anti-PVR blocking antibody (30µg/ml) for 15 min prior to co-culture.

Hepatocyte organoids were thawed, cultured in 50 µl droplets containing reduced growth factor BME2 and culture medium, as previously described (46). Growing hepatocyte organoids were passaged once before experimental use. Fresh PBMCs from buffy coats were enriched for NK cells using the EasySep™ Human NK Cell Enrichment Kit (Stemcell Technologies) following manufacturers protocols. NK cells were resuspended in RPMI (Thermo Fisher Scientific) supplemented with 10% FBS (Biochrom GmbH) (R10) and 1% Penicillin/Streptomycin and were added to fully-grown hepatocyte organoids. After 24 and 48 hours of co-incubation, cells present in the supernatant or the droplet, respectively, were collected separately, treated with TripLE™ Express Enzyme (Life Technologies GmbH) for organoid dissociation, stained and analyzed by flow cytometry. As controls, NK cells were cultured in R10 supplemented with 1% Penicillin/Streptomycin, but no hepatocyte organoids, and subsequently equally treated with TripLE™ Express Enzyme and stained.

Supernatants from hepatocyte organoids were collected 48h after the last medium change and stored at -80°C until further analysis. The multiplex assay (Bio-Rad) was performed on a Bio-Plex 200 System. Secretion of various chemokines by hepatocyte organoids (CXCL1, CXCL5, CXCL9, CXCL10, CXCL11, CXCL12, CXCL16, CX3CL1, CCL2, CCL5, CCL19, CCL20, and CCL21) was measured.

Statistical analysis was performed using GraphPad Prism 8 and 9 (GraphPad Software, Inc., La Jolla, CA) or R (R 4.0.3 GUI 1.73; R Studio, Version 1.3.1093, lme4, ggplot2, FactoMineR and factoextra packages). Principal component analysis was generated with R using the mRNA expression of 52 genes of 440 single CD56^bright ihNK cells or pbNK cells. Wilcoxon matched-pairs signed rank test was used to determine differences between paired samples with assumed non-normal distribution. To assess normal distribution within small sample sizes, the Shapiro-Wilk test was applied and subsequently, the paired t test was used. A mixed-effects model with random intercept, considering intra-sample correlations, was used to compare the single-cell mRNA expression data from different donors (or groups). To limit the detection of false positives, the p-values were adjusted by the Benjamini and Hochberg False Discovery Rate (FDR) method between each comparison groups (cutoff of 0.05). P values <0.05 were considered statistically significant. Statistical analysis for additional data presented in the figures are provided in the respective figure legends.

Flow cytometry data were visualized using viSNE (Cytobank, Santa Clara, CA). A tool for visualization of high-dimensional data based on the Barnes-Hut implementation of the t-distributed stochastic neighbor embedding algorithm was used (51). NK cells were equally sampled to randomly select 50 000 cells for analysis. The analysis was performed using the surface expression of CD56, CXCR6, CD69, TIGIT, DNAM-1 and CD96. The following settings were adjusted: 2000 iterations, 20 perplexity, theta 0.5.

Matched peripheral blood (pb) and intrahepatic (ih) NK cells from patients undergoing liver resection due to either liver metastases (n=3) or liver adenoma (n=1) were used for targeted single-cell mRNA expression analysis in this study. Furthermore, pbNK cells from healthy individuals (n=2) of the Hamburg Healthy Cohort were used as a control. We sorted CD56^bright ihNK cells and CD56^bright pbNK cells ( Figure S1A ) and investigated the single-cell mRNA expression of 52 representative genes on a total of 440 CD56^bright NK cells. All differentially expressed genes can be found in Supplementary Table S3 . Principal component analysis (PCA) revealed divergence of the NK cells depending on their tissue origin. Thus, this unsupervised dimension reduction method showed two partly separating clusters, one for CD56^bright ihNK cells and another one for CD56^bright pbNK cells, the latter overlapping with the control group of CD56^bright pbNK cells ( Figure 1A ). The separation of the PCA-dimension-1 was mainly driven by 18 genes, including genes associated with NK cell activity (e.g. CD38, NCR1, KLRB1) and tissue-residency (e.g. EOMES, TIGIT, CXCR6) ( 16, 22, 52–54). In conclusion, distinct single-cell gene expression discriminates CD56^bright pbNK cells and ihNK cells.

We next analyzed whether expression levels of mRNA of the known liver-residency markers CD69, CXCR6, CCR5, Eomes and T-bet differed between CD56^bright NK cells derived from livers and matched PBMCs ( Figure 1B ). Compared to CD56^bright pbNK cells, CD56^bright ihNK cells showed a significantly higher mRNA expression of CD69 (p<0.0001), CXCR6 (p<0.0001), CCR5 (p<0.0001) and the transcription factor EOMES (p<0.0001), while the transcription factor T-bet (TBX21) showed significantly lower (p<0.0001) mRNA expression in CD56^bright ihNK cells. This distinct mRNA expression of liver-residency markers in CD56^bright ihNK cells is in line with previous results from studies characterizing the phenotype of CD56^bright ihNK cells (16, 18–20, 23). In addition to the mentioned differentially expressed known liver-residency markers, we identified 17 genes with a highly significant differential mRNA expression comparing single CD56^bright ihNK cells and pbNK cells ( Figure 1C ). CD56^bright ihNK cells showed significantly higher mRNA expression of CD38, FCER1G, IFNG, IL2RB, IRF8, KLRB1, RORA, S1PR5, STAT4, TGFBR2, TIGIT and ZBTB16 (all p<0.0001). Most of these genes are primarily associated with regulatory immune functions, anti-inflammatory activity and tissue residency (55–60), while others are involved in pathways of NK cell proliferation (61). In contrast, CD56^bright pbNK cells expressed significantly higher levels of CXCR3, IL7R, ITGAM, S1PR1 and SERPINB1 (all p<0.0001), genes that are mainly associated with NK cell migration and maturation (62–67). Interestingly, CD56^bright ihNK cells that expressed high mRNA levels of TIGIT also encoded mostly (63%) for CXCR6 and CD69 mRNAs, two markers of liver residency, and also for the chemokine receptor CCR5, associated with tissue homing (see Supplementary Figure S2 ). Taken together, the analysis of gene expression on single NK cells identified significant differences between CD56^bright ihNK cells and CD56^bright pbNK cells, including the higher expression of genes involved in NK cell proliferation, regulatory functions, and tissue-residency in ihNK cells.

TIGIT is a co-inhibitory receptor that inhibits human NK cell cytotoxicity and has been suggested to be critical in murine liver regeneration (26, 42). The significantly elevated mRNA expression of TIGIT in CD56^bright ihNK cells prompted us to seek confirmation for this specific target on protein level by flow cytometry. We included TIGIT’s competing receptors DNAM-1 and CD96 in the analysis. All these receptors bind to the poliovirus receptor (PVR/CD155) and facilitate either immune activation or inhibition (32, 41). We analyzed TIGIT, DNAM-1 and CD96 surface protein expression on CD56^bright ihNK cells and matched CD56^bright pbNK cells from patients undergoing liver resection or liver transplantation ( Figure 2A ), as well as peripheral blood samples from healthy control individuals ( Figure 2C ). Frequencies of CD56^bright and CD56^dim NK cell subsets varied between ihNK cells and pbNK cells. CD56^bright NK cells were more frequent in ihNK cells compared to pbNK cells or healthy control NK cells (median frequency (%) of CD56^bright NK cells: 35%; 3,4%; 6,4%, respectively), while CD56^dim NK cells were less frequent in ihNK cells compared to pbNK cells or healthy control NK cells (median frequency (%) of CD56^dim NK cells: 40,1%; 88,7%; 82,5%, respectively).

A representative viSNE analysis revealed different clusters when comparing matched CD56^bright pbNK cells and ihNK cells from the same individual ( Figure 2B ). We identified clusters with high density and overlapping expression of CD56, liver-residency markers CD69, CXCR6, TIGIT and CD96. However, those clusters showed a distinctly lower expression of DNAM-1. When analyzing both subsets in detail, bulk ihNK cells showed higher TIGIT (p=0.01) and CD96 (p=0.04) expression ( Figure 2C ). In contrast, bulk pbNK cells showed a higher DNAM-1 expression (p<0.0001), which is consistent with a previous study demonstrating significantly higher DNAM-1 expression on pbNK cells (21). Regarding the CD56^dim NK cell subset, TIGIT (p=0.8) and CD96 (p=0.4) were equally expressed on pb and ihNK cells ( Figure 2C ). However, CD56^dim pbNK cells exhibited a significantly higher expression of DNAM-1 (p=0.0003). Differences in surface protein expression between pb and ihNK cells were most pronounced in the CD56^bright NK cell compartment ( Figure 2C ). In line with the corresponding mRNA data, we observed a significantly stronger surface protein expression of the inhibitory receptor TIGIT on CD56^bright ihNK cells than in CD56^bright pbNK cells (p<0.0001). In contrast, the CD56^bright pbNK cells showed significantly higher expression of DNAM-1 (p<0.0001) and CD96 (p=0.007) compared to their intrahepatic counterpart. Taken together, ihNK cells and especially the subset of CD56^bright ihNK cells showed higher levels of TIGIT, whereas pbNK cells distinctly exhibited higher expressions of DNAM-1.

Previous data in mice and humans have suggested that TIGIT⁺ NK cells are functionally exhausted (68, 69), but the role of TIGIT in human ihNK cells is not well understood. To assess functional differences between TIGIT⁺ and TIGIT^– CD56^bright ihNK cells, we next determined tumor necrosis factor alpha (TNF-α) production and degranulation (CD107a expression) after stimulation of intrahepatic cells from patients undergoing liver transplantation (n=6) or liver metastases resection (n=2) with K562 target cells ( Figure 3 ). After stimulation, we observed significantly less degranulation of TIGIT⁺ CD56^bright ihNK cells compared to TIGIT^– CD56^bright in NK cells (p=0.02). Moreover, the TIGIT⁺ CD56^bright ihNK cells also exhibited a significantly diminished TNF-α response compared to TIGIT^– CD56^bright ihNK cells (p=0.03). Overall, TIGIT⁺ CD56^bright ihNK cells were significantly less responsive to stimulation than TIGIT^– CD56^bright ihNK cells, in line with previous findings describing reduced functionality of TIGIT⁺ NK cells (68, 69).

The above data demonstrate that CD56^bright ihNK cells express higher levels of TIGIT on both mRNA and protein levels, and that TIGIT⁺ NK cells show reduced functionality. We next determined whether the liver environment itself promoted an increased immunotolerance of ihNK cells, as the ligand for TIGIT, PVR/CD155, is expressed by hepatocytes (42). For this purpose, we co-cultured isolated pbNK cells from healthy individuals with the human hepatoma cell line Huh7, which stably expresses PVR ( Figure 4C ). To be able to differentiate between effects of direct receptor binding or soluble stimulation, NK cells were cultured either in direct contact with Huh7 cells or separated via a transwell insert ( Figure 4A ). Subsequently, the expression of TIGIT, DNAM-1 and CD96 was assessed ( Figure 4B ). Regardless of the condition, we did not observe any differences in CD96 expression up to 48 hours of co-culture. DNAM-1 expression in CD56^bright NK cells started to decline after 6 hours and continued to decrease up to 48 hours during co-culture, but only when in direct contact with the Huh7 cells (p=0.003). After 48 hours of co-incubation, DNAM-1 expression decreased significantly by almost 40% (p=0.01). However, no effect on the modulation of DNAM-1 expression was observed in the transwell experiments or in the absence of Huh7 cells, suggesting that direct cell-to-cell contact between NK cells and hepatoma cells was required to induce DNAM-1 downregulation. In contrast, TIGIT expression started to increase within 12 hours upon direct contact with Huh7 cells. This increase was even more pronounced after 24 hours of co-culture of CD56^bright NK cells in direct contact with Huh7 cells. Again, no effect on the modulation of TIGIT expression was detectable in CD56^bright NK cells cultured without Huh7 cells, and only minor changes in TIGIT expression were detected in transwell experiments. To start to elucidate the contribution of different factors to the observed TIGIT upregulation on NK cells, we compared the receptor expression of TIGIT on NK cells after direct co-incubation with Huh7 cells alone or in the presence of an anti-PVR blocking antibody ( Figure 4D ). As observed before, direct co-culture with Huh7 cells resulted in a significant upregulation of TIGIT on NK cells, which was significantly reduced, but not fully suppressed, by addition of the PVR-blocking antibody. Taken together, these data show that CD56^bright pbNK cells altered their phenotypes towards CD56^bright ihNK cells when cultured together with PVR-expressing human hepatoma cells, by downregulating activating and simultaneously upregulating inhibitory NK cell receptors that bind to PVR. These data suggest a direct impact of the liver environment on the phenotype and functional activity of ihNK cells.

To further determine whether the observed changes in pbNK cells in the presence of hepatoma cells were also observed in response to human primary hepatocytes, we co-cultured hepatocyte organoids, which stably express PVR ( Figure 5A ) and were derived from human liver tissue and were growing within BME2 droplets with pbNK cells. pbNK cells migrate towards hepatocyte organoids, as the number of NK cells entering hepatocyte organoids and the surrounding BME2 droplets was significantly higher in wells with hepatocyte organoids compared to wells only containing empty BME2 droplets after 24 hours ( Figure 5B ). To assess which chemokines attracting NK cells are secreted by hepatocyte organoids, chemokine levels in the cell-culture supernatant from hepatocyte organoids were analyzed ( Figure 5C ). Chemokine levels were detected at different concentrations ranging from mean concentrations below 100 pg/ml (CCL5, CXCL9, CCL19, CXCL12, CXCL11) to concentrations above 700 pg/ml (CCL21, CXCL10, CXCL16, CCL2, CX3CL1, CXCL1, CCL20, CXCL5) ( Figure 5C ). The results from these co-culture studies demonstrate that hepatocyte organoids secrete chemokines that attract NK cells.

To identify phenotypical changes in CD56^bright pbNK cells following migration into hepatocyte organoids, we subsequently analyzed TIGIT and DNAM-1 surface expression of CD56^bright NK cells upon co-culture with hepatocyte organoids compared to CD56^bright NK cells in medium only (controls), and furthermore differentiated those NK cells that entered hepatocyte organoids within BME2 droplets from those NK cells that remained outside of BME2 droplets. Surface expression of TIGIT and DNAM-1 on NK cells was analyzed after 24 and 48 hours of co-culture ( Figure 5D ). Both, CD56^bright NK cells that entered BME2 droplets and those that did not showed an increased TIGIT expression after co-culture with hepatocyte organoids compared to control conditions. However, the percentage of TIGIT⁺ NK cells that migrated into hepatocyte organoids was already significantly higher after 24 hours compared to control conditions without organoids, while the frequency of TIGIT⁺ NK cells outside of the BME2 droplet (supernatant) increased more slowly ( Figure 5D ). Furthermore, DNAM-1 expression on NK cells continuously decreased only on those NK cells that entered the hepatocyte organoids, reaching significantly lower levels after 48 hours of co-incubation, while no decrease in percentages of DNAM-1 expressing NK cells was observed under the other conditions ( Figure 5D ). These data suggest that TIGIT and DNAM-1 expression on NK cells is preferentially modulated by direct cell-to-cell-contact and that primary human hepatocyte organoids can be used as an in vitro model to study migration of immune cells and their interactions with tissue cells.