Peripheral blood mononuclear cells (PBMC) were isolated by Lymphoprep (Axis-Shield) gradient. For seahorse analysis, NK cells were purified using the EasySep negative selection purification kit (STEMCELL) as per the manufacturer’s instructions. Unless stated otherwise, 5×10⁶ cells/mL PBMC or purified NK cells were incubated at 37°C for 18 hours in RPMI V.1640 GlutaMAX medium (Life Technologies/Invitrogen) supplemented with 10% fetal calf serum (FCS), 1% penicillin/streptomycin (Invitrogen). Cells were stimulated with interleukin (IL)2 (500 IU/mL; National Cancer Institute) or IL12 (30 ng/mL; Miltenyi Biotec) and IL15 (100 ng/mL; National Cancer Institute). Where indicated, cells were cultured with or without IgG1 isotype control (5 or 10 µg/mL), anti-TGFβ MAb (5 µg/mL, 1D11, BioTechne/R&D Systems), anti-GARP MAb (10 µg/mL, Sophie Lucas lab, Belgium) or recombinant human TGFβ (10 µg/mL, Bio-Techne).

Blood samples were obtained from normal healthy donors (all female, mean age 45 years) or from patients with breast cancer (all female, mean age 58 years) from whom informed written consent had been obtained. Patients had metastatic breast cancer with mixed hormone status (table 1). They had not received conventional chemotherapy within the last year. Some patients were on one or more of the following therapies: fulvestrant, zometa, letrozole, trastuzumab, palbociclib, pertuzumab, xgeva, tamoxifen, decapeptyl and goserelin.

Cells were stained for 30 min at 4 °C with saturating concentrations of titered Abs CD56 (HCD56/NCAM16.2), CD3 (SK7/UCHT1), granzyme B (GB11), IFNγ (B27), CD71 (M-A172), CD69 (L78), CD98 (UM7F8), NKp44 (p44-8.1), TRAIL (RiK-2), pSMAD2/3 (O72-670), CD25 (M-A251) (eBioscience or BD Pharmingen); S6 ribosomal protein phosphorylated on serine 235/6 (pS6), and eukaryotic translation initiation factor 4E-binding protein 1 (4E-BP1) phosphorylated on Thr37/46 (236B4, Cell Signaling Technology), LAP (TW4-2F8) and GARP (7B11). A viability dye was included in every panel (LIVE/DEAD Near-IR, Bio Sciences). Cells were prepared, stained and analyzed as previously described.21

Mitochondrial membrane potential (MMP) was measured via staining of cells for 30 min with tetramethylrhodamine methyl ester (TMRM, 100 nM—Thermo Fisher), oligomycin (2 µM) and carbonyl cyanide p-trifluoro-methoxyphenyl hydrazone (FCCP, 2 µM) were used as positive and negative controls, respectively. Mitochondrial mass was measured via staining of cells for 30 min with MitoTracker Red (100nM—Thermo Fisher) or MitoTracker Green (100nM—Thermo Fisher). Adenosine triphosphate (ATP) synthase analysis was performed by detecting the expression of the ATP5B subunit of the (3D5, Abcam) via intracellular flow cytometry staining. Mitochondrial superoxide levels were measured via staining of cells for 15 min with MitoSOX (Thermo Fisher). Rotenone (20 µM) and MitoTEMPO (2.5 µM) were used as positive and negative controls, respectively.

Determination of oxygen consumption rate (OCR) representing OXPHOS or extracellular acidification rate (ECAR) indicating glycolysis was detected by XFp extracellular flux analyzer (Agilent Technologies). NK cells were purified from PBMC using the EasySep negative selection purification kit (STEMCELL) as per the manufacturer’s instructions; purity was routinely >90% CD56⁺CD3⁻ NK cells. Cells were stimulated for 18 hours in the presence or absence of IL2. In order to adhere NK cells to the bottom of the seahorse plate, cell plates were coated with Cell-Tak (6 µg/mL). The Cell-Tak was diluted in sodium bicarbonate (0.1M) with 0.15% (v/v) NaOH (1M) and added to the bottom of each well (25 µl). It was left at room temperature (RT) for a minimum of 20 min, removed from the plate and each well was washed two times with sterile ddH2O. NK cells were washed two times in GlutaMAX seahorse media supplemented with glucose (1M), adjusted to pH 7.4. NK cells were added to each well (2.5×10⁵ cells, 180 µl), while seahorse media were used in the blank wells. The cell plate was centrifuged at 200g for 3 min with no brake, and then placed in a non-CO₂ incubator for 30 min prior to metabolic analysis. During the assay, the following inhibitors were added in order—oligomycin (2 µM), FCCP (0.5 µM), rotenone (100 nM)+antimycin A (4 µM) and 2-deoxyglucose (30 mM).

Purified NK cells (8×10⁵ cells and >90% purity) were stained using Mitospy CMX Ros (250 nM, Biolegend) for 30 min at 37°C and fixed in 2% paraformaldehyde (PFA, Sigma) for 15 min at RT, prior to nuclear staining with 4',6-diamidino-2-phenylindole, dihydrochloride (300 nM, Thermo Fischer Scientific) for 5 min at room temperature. NK cells were mounted using Mowiol (Sigma) and imaged on a Leica SP8 inverted motorized microscope equipped with a ×63/1.4 N.A. oil objective and 405 nm diode and Leica white laser lines. Z-stacks at 0.2 µm increments were captured using an HyD detector in conjunction with Leica LAS X acquisition software.

Image analysis was performed on the maximum intensity projections of the z-stack. NK cell mitochondrial morphology was assessed by blind scoring of a mixed healthy and patient dataset (total number of cells=84). Six volunteers blindly scored whether morphology was predominantly fragmented or tubular, and if mitochondria were predominantly single or networked, as per an ordinal scale provided in online supplemental figure 1. The mode of each cell score was collated and the per cent morphology score was calculated for each individual donor as follows: (number of cells with each score divided by the total number of donor cells)×100. The per cent morphology scores were then averaged in healthy and patient groups and compared using a two-way analysis of variance (ANOVA).

Mitochondrial fluorescence and morphology were also quantified in Fiji (http://fiji.sc/Fiji) by delineating the mitochondria as regions of interest (ROI) and measuring their integrated density or circularity, respectively, using the following formulae: Integrated density: (raw integrated density of ROI)×(area of ROI). Circularity=4π (area of ROI÷〖perimeter〗² of ROI). A circularity value of 1.0 indicates a perfect circle, while values approaching 0.0 indicates an elongated polygon.22

ELISA kit was purchased from Invivogen. Plates were coated with capture antibody at 4°C overnight. Wells were blocked with ELISA coating buffer at RT for 1 hour. To activate latent TGFb, all samples were acid activated with 1M HCl for 10 min at RT. 1M NaOH was then used to neutralize the reaction. Samples/standards were added to wells in the appropriate dilutions and incubated at RT for 2 hour. Biotinylated detection antibody was added and the plate was incubated for 1 hour. Horseradish peroxidase (HRP)-conjugated streptavidin was added and incubated for 30 min at RT. Tetramethylbenzidine was added to each well. The reaction was stopped using 1M H₃PO₄. Optical density (OD) values were determined by measuring absorbance values at 450 nm using a SpectraMax Microplate Reader (Molecular Devices).

K562 cells were counted and washed thoroughly out of media two times using phosphate buffered saline (PBS). The cells were resuspended at a concentration of 2×10⁶ cells/mL in PBS+20 µM calcein acetoxymethyl (AM) dye. Following 30 min incubation, labeled K562 cells were thoroughly washed in wash buffer (20% FCS in PBS) three times after which they were resuspended at a concentration of 0.3×10⁶ cells/mL and allowed to rest for 1 hour at 37°C. PBMC were counted, washed and plated in 96 well plates according to the Effector to Target (E:T) ratios of 10:1, 20:1 and 40:1, in triplicate. K562 cells were washed again following the 1 hour rest and resuspended at 0.3×106 cells/mL in RPMI and plated in with the NK cells. K562s were plated on their own without NK cells, to measure spontaneous release. Additionally, K562 cells alone were plated in the presence of 8 µl of 10% Triton X to measure maximum release of calcein AM.

GraphPad Prism V.8.00 (GraphPad Software) was used for statistical analysis. Normality was determined using the D’Agostino-Pearson omnibus test. Data were then analyzed using a Student’s t-test when two data sets were being compared, or a one-way/two-way ANOVA test when more than two data sets were being compared.

We first defined NK cell functional responses and subset distribution in our cohort of patients with metastatic breast cancer compared with healthy controls (see table 1 for details of the patient cohorts). The overall frequency of NK cells in patient versus control PBMC was similar, though there was a decrease in the frequency of CD56^bright NK cells in patient NK cells (figure 1A). Cytokines are commonly used to activate NK cells. While both IL2 and IL12/IL15 upregulated the expression of the activation marker CD69 equivalently (figure 1B), significantly fewer patient NK cells expressed CD25 in response to either IL2 or IL12/15 compared with healthy control NK cells (figure 1C). While TNFα production was similar, patient NK cells had a clear defect in the production of IFNγ in both CD56^dim and CD56^bright NK cell subsets in response to IL12/IL15 stimulation when compared with healthy control cells (figure 1D–F). In terms of cytotoxicity, granzyme B levels of expression were similar but cytokine-induced TRAIL expression was significantly lower on CD56^dim NK cells from patients compared with controls (figure 1G,H). In terms of direct killing, NK cells from healthy donors were able kill standard K562 target cells efficiently as expected (representative cytokine-induced killing over a range of effector:target ratios are shown in figure 1I) demonstrating that their killing machinery is working well. However, in stark contrast, NK cells from patients failed to induce even minimal killing (figure 1I,J), Overall, profound defects in both cytotoxicity and cytokine production were found in circulating NK cells from patients with metastatic breast cancer.

As we have previously demonstrated that metabolic reprogramming is required for NK cell effector functions, we investigated whether metabolism was dysregulated in circulating NK cells from patients with metastatic breast cancer. Direct ex vivo analysis, without any experimental manipulation, showed evidence of mitochondrial alterations in patient NK cells with increased mitochondrial mass alongside elevated levels of mitochondrial reactive oxygen species production and increased MMP compared with healthy controls (figure 2A–D). It was striking that while an increase in mitochondrial mass correlated with increased MMP in healthy controls, as might be predicted, this was not the case for mitochondria for patient NK cells where pathology disrupted this relationship, as evidenced by R² correlation values (figure 2E).

We also performed parallel confocal microscopy analysis that revealed differences in mitochondrial structure; while mitochondria from healthy controls were generally more fused with elongated morphologies, those from patients had disrupted structures with more punctate mitochondria, a morphology indicative of decreased mitochondrial fusion (figure 3A,B). This was supported by quantitative analysis where circularity scores (n=1 for perfect circle associated with extreme fragmented phenotype) and length of individual mitochondria were significantly higher from patient compared with healthy control NK cell mitochondria (figure 3C). Furthermore, patient NK cells had increased numbers of mitochondria per cell, and integrated density values (akin to mean flourescence intensity intensity, MFI) were also significantly increased in patient NK cells (figure 3D) suggesting increased MMP, mitochondrial mass or increases in both parameters, further supporting the data generated in figure 2.

These metabolic defects observed directly ex vivo argue that patient NK cells would have impaired metabolic responses to subsequent cytokine stimulation. Indeed, patient NK cells had impaired upregulation of metabolic machinery in response to cytokine stimulation including the nutrient receptors CD71 and CD98, and the ATP synthase subunit ATP5b (figure 4A,B). Additionally, while cytokine-induced basal glycolysis levels were similar, patient NK cells had significantly reduced glycolytic capacity and reduced rates of flux through the OXPHOS pathway compared with NK cells from healthy controls (figure 4C–F). Concurrent with these overall decreased metabolic rates, IL2 stimulated patient NK cells also had reduced levels of mammalian target of rapamycin complex 1 (mTORC1) activity, an important metabolic regulator in NK cells, measured as levels of phosphorylated 4E-BP1 and S6 ribosomal protein (figure 4G,H).

Ourselves and others have previously published that TGFβ inhibits cytokine-induced metabolic reprogramming in human NK cells.19 23 We therefore hypothesized that TGFβ may contribute to the dysfunction observed in circulating NK cells from patients, particularly as it contributes to the immunosuppressive environment found in many cancer types.24 It is often found at higher circulating levels in cancer although in human breast cancer, the literature is conflicting.25–27 To investigate if TGFβ might be a mechanism leading to dysfunction in metastatic breast cancer, we first measured TGFβ levels in the serum of patients and healthy controls and found no significant difference (figure 5A). Given that there was no apparent systemic increase in TGFβ, it was considered important to investigate for any evidence of intrinsic TGFβ signaling in NK cells in patients. Using a well-described anti-phospho-SMAD2/3 to demonstrate canonical TGFβ signaling, we found that patient NK cells trended towards higher constitutive levels of pSMAD2/3 in both CD56^dim and CD56^bright NK cells (figure 5B,C). A similar pattern of expression was found for LAP, a protein which binds TGFβ homodimers intracellularly to form the TGFβ small latency complex. As LAP is cotranscribed at the N-terminal region of the TGFβ gene product, expression of LAP indicates active TGFβ transcription. Thus, while not reaching statistical significance, there was a trend towards increased expression of LAP on NK cells from patients compared with controls, for both CD56^dim and CD56^bright subsets, suggesting that these NK cells intrinsically produce both TGFβ and LAP proteins (figure 5D). We next considered whether there might be locally produced TGFβ that could impact on NK cells metabolism and function. It is described that T-regs produce TGFβ that acts in an autocrine manner through a mechanism involving the sequestering of latent TGFβ (as small latency complex) at the cell surface by the transmembrane protein, GARP.20 28 Although there were a few exceptions, most healthy donors expressed low levels of GARP on the surface of their NK cells (with no difference between levels on CD56^dim or CD56^bright subsets) (figure 5H). However, it was striking that patients with metastatic breast cancer expressed significantly higher levels of GARP on both CD56^dim and CD56^bright NK cell subsets compared with controls (figure 5E). Cytokines could increase expression of GARP, particularly on CD56^bright cells (online supplemental figure 2). In general, LAP and GARP expression mirrored each other as can be seen in representative plots in figure 5F. Expression patterns ranged from a subset of cells expressing both LAP and GARP, a general overall shift in the population, to a dramatic upregulation with almost all NK cells expressing both. In order to see if higher GARP expression was associated with intrinsic TGFβ signaling, we stratified patients according to ‘high’ or ‘low’ GARP expression and looked at pSMAD2/3 phosphorylation in the two groups. As can be seen in figure 5G, patients with higher GARP expression on their NK cells had statistically significantly higher TGFβ signaling, in both CD56^dim and CD56^bright subsets. In combination, the overexpression of TGFβ machinery and higher intrinsic TGFβ signaling in NK cells from patients compared with controls suggest that GARP may serve as a novel mechanism for autocrine inhibition of NK cell metabolism and function during cancer.

Our data support that NK cell metabolism and function are severely impacted during metastatic breast cancer and that locally produced TGFβ could potentially drive these defects in patient NK cells. Indeed, neutralizing TGFβ with anti-TGFβ antibodies had a clear restorative impact on patient NK cells in vitro. First, neutralizing TGFβ rescued both CD25 expression and mTORC1 activity in cytokine stimulated patient NK cells (figure 6A,B). This finding was supported by increased expression of the transferrin receptor CD71, a downstream target of mTORC1 signaling (figure 6C). Neutralizing TGFβ also significantly restored features of dysregulated mitochondrial metabolism in NK cells from patients, including the reduction of mitochondrial mass in patient NK cells (figure 6D). Blocking TGFβ also increased OXPHOS and maximal respiration in patient NK cells stimulated with IL2 (figure 6E,F). Overall, there was a shift in the OCR/ECAR ratio indicating an increased reliance on oxidative glucose metabolism compared with glycolysis (figure 1G). It is important to note that this latter experiment was performed using purified NK cells, further supporting endogenous TGFβ from NK cells as a contributing factor. Finally, while blocking TGFβ had minimal impact on IFNγ production by healthy NK cells (see online supplemental figure 3), it significantly increased its production in patient NK cells (figure 6H). To our knowledge, this is the first demonstration that neutralizing TGFβ can significantly contribute to restoring some of the profound metabolic defects found in peripheral blood NK cells of cancer patients.

In order to support data from purified NK cells which demonstrate that NK cell-derived TGFβ contributes to the defects observed and to further investigate GARP expression by NK cells as a potential mechanism explaining this, we obtained a well-characterized monoclonal antibody which specifically blocks the activities of GARP when bound to LAP.29 Experiments were performed on purified NK cells that were stimulated by cytokine in the presence of either an anti-GARP or an isotype control antibody. For a range of activation (CD25), signaling (mTORC1), nutrient receptor (CD71, CD98) and cytotoxicity markers (granzyme B, TRAIL), inclusion of anti-GARP restored these activities in dysregulated patient NK cells. Direct measurement of metabolic flux clearly shows that anti-GARP allowed dysfunctional NK cells from patients with cancer to reengage in OXPHOS. As anti-GARP did not restore glycolysis (if anything, it decreased already low levels of glycolysis, data not shown), its inclusion effectively switched these NK cells to OXPHOS as the primary pathway of glucose metabolism as indicated by the OCR/ECAR ratio (figure 7E). In terms of IFNγ production, results were more variable. The frequency of CD56^dim but not CD56^bright cells producing IFNγ increased significantly in the presence of anti-GARP (figure 7F). Analysis identified that increased IFNγ expression correlated with GARP expression (figure 7G,H), supporting that GARP contributes to NK cell dysfunction and that blocking its activity restores function in patient NK cells. This research demonstrates the potential in targeting metabolic pathways as a route towards restoring functional responses in ‘exhausted’ NK cells from patients with cancer.