Not all PBMC samples from healthy donors could be used to effectively expand γδT cells using ZOL.24 In the current study, 98 PBMC samples from 43 healthy donors were tested independently. Some of the donors had multiple samples collected at different time points during a period of 2 years. Consistent with published data, the purity of expanded γδT cells was shown to be one of the important indicators for γδT cell function.25 26 The purity of the expanded γδT cells by ZOL in all of the 43 donors is displayed in figure 1A. PBMCs from 17 donors were tested three times or more independently at different time points and the purity of expanded γδT cells from the same donor yielded similar results (online supplemental figure S1A). PBMCs from 13 donors were expanded by both ZOL and anti-γδT TCR antibody. ZOL expanded γδT cells significantly better than that of anti-γδT TCR antibody (figure 1B); however, both expansion methods showed a similar trend (online supplemental figure S1B). Anti-TCRγδ Ab expanded both Vδ2 and Vδ1 cells and the ratio of Vδ2:Vδ1 is 50%–250%, while ZOL preferentially expanded Vδ2 cells and the ratio of Vδ2:Vδ1 is 173%–3500% (online supplemental figure S1C).

These donors were separated into two groups, good expansion group (γδT cell purity ≥50%) and poor expansion group (γδT cell purity <50%), as previously reported.27 Absolute cell number, the fold change of expansion and the viability of expanded γδT cells in the two groups are illustrated in figure 1C. γδT cell expansion fold ranged from 200-fold to 5000-fold in the good expanders and 60-fold to 600-fold in the poor expanders. Purer γδT cells were accompanied by higher absolute γδT cell number and cell viability at day 10 after expansion (figure 1C). A more detailed analysis of γδT cell expansion showed that the difference between good and poor expanders appeared 6 days after expansion (figure 1D,F). Age (figure 1G), gender (online supplemental figure S1D) and HLA type (online supplemental table 1) of the donors were not significantly associated with γδT cell expansion capacity.

It has been shown that the immune phenotype of γδT cells may predict its function.28 29 To study the phenotypic difference in γδT cell in healthy donors, we examined CD27 and CD45RA expression on γδT cell after 10 days of expansion. These receptors distinguish effector memory (EM, CD45RA−CD27−) cells from naive (N, CD45RA+CD27+), central memory (CM, CD45RA−CD27+) or terminal-differentiated effector memory (TDEM, CD45RA+CD27−) cells (figure 2A). The proportion of the effector memory and central memory γδT cells were significantly higher in the good expanders compared with that in the poor expanders (figure 2B). We then examined the expression of immune checkpoint proteins in the expanded γδT cells (online supplemental figure S2). γδT cells in the good expanders showed significantly lower expression of exhaustion markers, PD-1, CTLA-4 and Eomes (figure 2C), but higher expression of effector cytokine, IFN-γ, Granzyme B (figure 2D) and costimulatory molecule of CD86 (figure 2E), compared with those in the poor expanders. There were no significant changes in T-bet and CD80 expression.

To study the migratory ability of the expanded γδT cells to tumor cells, we performed the transwell migration assays. Significantly enhanced migration of γδT cells toward A2058, A375 and UACC903 melanoma cells was observed in the good expansion group compared with that in the poor expansion group after 8 hours of incubation (figure 3A). Expanded γδT cells were mixed with tumor cells at the effector:target (E:T) ratio of 5:1. γδT cells in the good expansion group more effectively killed tumor cells than those in the poor expansion group (figure 3B). We further explored the migration and cytotoxic effector function of the expanded γδT cells using 3D melanoma spheroid assays. We examined spheroid integrity by monitoring their spheroid size and γδT cell infiltration by flow cytometry. A2058 and UACC903 melanoma spheroids were co-cultured with γδT cells for 48 hours. Melanoma spheroids had greater γδT cell infiltration and were significantly smaller when cultured with expanded γδT cells from good expanders than those from poor expanders (figure 3C,D). These data suggest that γδT cell expansion capacity is correlated with their anti-tumor capacity.

We measured the basal levels of γδT cells in PBMC in 66 samples from 31 donors using flow cytometry (figure 4A). There was a significant correlation between basal Vδ2 T cells and γδT expansion capacity (figure 4B). Better expanders had a significantly higher basal Vδ2 T cells in the PBMCs (figure 4C). Using logistic regression and LASSO models, we calculated a baseline Vδ2 T cell cut-off of 0.82% to separate good expanders from poor expanders (figure 4D,E).

We then studied whether immune phenotypes of basal γδT cells in PBMC correlate with their expansion capacity. The results demonstrated more central memory (CD27+CD45RA−) and fewer terminally differentiated γδT cells (CD27−CD45RA+) in the Vδ2 T cells ≥0.82% group than those in the <0.82% group (figure 5A,B). Central memory cells have better proliferation capacity, while terminally differentiated cells display immediate effector functions with poorer proliferation capacity.13 In addition, PD-1, CTLA-4, Eomes and T-bet levels were significantly higher in the Vδ2 T cells <0.82% group than those in the ≥0.82% group (figure 5C). Vδ2 T cells with high PD-1 and CTLA-4 expression suppressed Vδ2 T cell proliferative responses and cytotoxic potential.30 Interestingly, γδT cells expressed less IFN-γ but more Granzyme B in the <0.82% group (figure 5D). Moreover, the activation marker CD69 and the cytotoxicity markers NKG2D and CD107a were also significantly higher in the <0.82% group. Inversely, the proliferation marker Ki67 was significantly lower in the <0.82% group, while other activation markers such as CD80, CD86 and CD277 showed no significant difference (online supplemental figure S3). Altogether, the γδT cells in the <0.82% group express markers of exhaustion and cytotoxicity, while those in the ≥0.82% group show memory-like phenotypes.

To study whether other cell components in PBMC impacted γδT cell expansion, we examined other immune cells in PBMC from 25 healthy individuals. In the good expansion group, we found that Vδ2 T cells comprised a larger percentage of all γδT cells. The Vδ1 T cells in different individuals varied but no significant difference was found in these two groups; however, the ratio of Vδ2:Vδ1 T cells was significantly different in these two groups (figure 5E). We measured CD3+, CD4+, CD14+, CD19+ and CD11b+ cell population in the PBMC (online supplemental figure S4A). CD4+ cell proportion in the ≥0.82% group was higher. However, CD3+, CD14+, CD11b+ and CD19+ cells showed no significant difference in these two groups (online supplemental figure S4B–D).

To study whether different culture methods influenced γδT cell expansion, we used ZOL or anti-γδ TCR antibody to expanded γδT cells from PBMC (n=6). Our data demonstrated that ZOL induced better expansion of γδT cells than anti-γδ TCR antibody. Nevertheless, similar to ZOL-induced γδT cell expansion, anti-γδ TCR antibody induced significantly better γδT cell expansion in the basal Vδ2 T cells ≥0.82% group than that in the <0.82% group (figure 6A,B). We also measured immune checkpoint protein expression on the anti-γδ TCR antibody-expanded γδT cells, and found that the Vδ2 T cells in the ≥0.82% group expressed lower PD-1 and CTLA-4 and higher Granzyme B than those in the <0.82% group (figure 6C), consistent with ZOL-expanded cells. The cytotoxicity of tumor spheroid assays in UACC903 and A2058 melanoma cells showed comparable results, regardless of whether ZOL or anti-γδ TCR antibody expansion was used (figure 6D,E and online supplemental figure S5). These data suggest that interindividual γδT expansion capacity is not related to expansion methods.

Since PD-1 and CTLA-4 proteins are highly expressed in the γδT cells in the poor expanders, we tried to rescue these cells by adding anti-PD-1 (50 µg/mL) or anti-CTLA-4 (50 µg/mL) antibody in the culture medium during the expansion period. We examined γδT cell number, cell purity, CD107a, Granzyme B and IFN-γ expression in the γδT cells after expansion. However, we did not see any increase of γδT cell expansion nor improved cytokine expression after the treatment (online supplemental figure S6).

To predict expansion capacity of γδT cells ex vivo, we developed two logistic regression models. The first predictive model is based solely on the Baseline γδT cell concentration in PMBC, from which each individual sample has a score calculated as:

Vδ2 Index Score=−8.23+10.25×Initial_γδT%

Using this model, a sensitivity of 97.78% and specificity of 90.48% is reached at cut point of 0.82% of the initial γδT%, with area under the curve (AUC)=0.968 (figure 7A,B). In the second comprehensive predictive model, we included the initial γδT cell level and surface marker expression of γδT cells. Each individual sample has a score from the established logistic regression model:

Comprehensivevδ2IndexScore=ln(P1−P)=−20.57+24.05×Initial−vδ2T%+0.04×PD1−0.08×CTLA4−0.07×Emoes+0.04×T−bet−0.22×INF−γ+0.02×GranzB+0.48×CD80−0.27×CD8

The probability P that an individual’s γδT cells can be expanded ≥50% was calculated as

P=eScore1+eScore

The optimal sensitivity of 95.56% and specificity of 95.24% is reached at cut point 0.74 of the Score, corresponding to P=0.678, with AUC=0.989 (online supplemental figure S7A). There is no significant difference of AUC between these two models (Delong’s test p value=0.3047).

Patient-derived cancer organoids retain the heterogeneity of original tumors and provide an opportunity for in vitro testing of human γδT cell therapy since good in vivo models are not available. To study whether our findings may predict effects of γδT cells on patient-derived melanoma organoids, we screened PBMC samples and used the 10 most recent samples. Vδ2 Index Score and Comprehensive Vδ2 Index Score were calculated and shown in figure 7C and online supplemental figure S7B, respectively. Fresh human melanoma tissues were obtained and processed to establish patient-derived melanoma organoids as previously described.31 Melanoma organoids grew well over time and few cells were positive for propidium iodide (PI, dead cells, red fluorescence) staining, which stains dead cells. Following γδT cell expansion, we co-cultured the expanded γδT with the organoids for 24 hours, and then cells were stained with acridine orange (AO; live cells, green) and PI (dead cells, red). Vδ2 Index Score ≥0 group exhibited significantly more cell death than the Vδ2 Index Score <0 group (figure 7D). The cytotoxicity assay showed similar results (figure 7E). FACS analysis of Annexin V (AV) and 7-amino actinomycin D (7-AAD) stained cells showed a greater number of apoptotic cells in the Vδ2 Index Score ≥0 group than that in the Vδ2 Index Score <0 group (figure 7F,G). These data support that the Vδ2 Index Score may be used to predict γδT anti-tumor function.

Peripheral blood samples were collected from apheresis circuit of 43 healthy blood donors (23 men, 20 women, age range 21–55 years) through Human Immunology Core at the University of Pennsylvania with informed consent and IRB approval (protocol number 707906).

PBMCs were plated at 2×10⁶ cells per well in 24-well flat-bottomed plates at 37℃ in a humidified atmosphere of 5% CO₂. The standard culture medium contains RPMI 1640 (Gibco), containing 10% FBS (Gibco), 2 mM L-glutamine (Gibco), 100 U/mL penicillin (Gibco), streptomycin (Gibco) and 200 IU/mL IL-2 (PeproTech, Rocky Hill, NJ).42 Zoledronic acid (5 µM; Sigma-Aldrich, St. Louis, MO, USA) was added into the culture medium at the first day and refreshed with medium containing IL-2 every 2 days. In some cases, 24-well plates were pre-coated with 0.5 µg/mL anti-γδ TCR Ab (Immunotech; Beckman Coulter, Fullerton, CA) and then PBMCs were added to the anti-γδ TCR Ab-coated wells and cultured in the standard culture medium mentioned previously. The viability and cell number of γδT cells were evaluated by the trypan blue staining assay, and the percentage of γδT cells counted by flow cytometry.23

Fluorochrome-conjugated antibodies were from eBioscience or BioLegend: CD3 (OKT3), CD4 (OKT4), CD14 (63D3), CD11b (ICRF44), CD19 (HIB19), CD27 (LG.3A10), CD45RA (HI100), CD107a (H4A3), TCR γ/δ (B1), Vδ2 (B6), Vδ1 (TS8.2), Granzyme B (QA16A02), IFN-γ (B27), PD-1 (EH12.1), CTLA4 (BNI3), T-bet (4B10), Eomes (WD1928), CD80 (2D10) and CD86 (IT2.2). For surface markers, expanded Vγ9Vδ2 T cells or PBMCs were stained with diluted fluorochrome-conjugated antibodies on ice and washed with FACS buffer; and for intracellular biomarkers, cells were permeabilized and stained with antibodies diluted in permeabilization buffer (BioLegend, San Diego, CA). Samples were acquired using a LSR-A flow cytometer (BD Biosciences, Baltimore, MD) and analyzed using FlowJo software (Tree Star, Ashland, OR).

The CytoTox 96 Non-Radioactive Cytotoxicity Assay (Promega, Madison, WI) was applied to measure the cytotoxicity of expanded Vγ9Vδ2 T cells. In this respect, A375 (2×10⁴ cells/well), A2058 (2×10⁴ cells/well) and 903 (2×10⁴ cells/well) melanoma cells were used as target cells and seeded into a 96-round well plate, and expanded Vγ9Vδ2 T cells (1×10⁵ cells/well) that served as effector cells were directly added to individual wells and incubated overnight. The supernatants were collected and the lactate dehydrogenase activity was detected. Controls for spontaneous lactate dehydrogenase release from effector and target cells, the target maximum release, as well as the culture medium background were measured simultaneously.13

Chemotaxis assays of expanded Vγ9Vδ2 T cells were performed using 24-well transwell plates with 5 µm pore polycarbonate membrane inserts (Costar; Corning, NY). Briefly, expanded Vγ9Vδ2 T cells (2×10⁵ cells/well) suspended in 200 µl RPMI 1640 containing 0.5% FBS were placed in the upper chamber, and RPMI 1640 (600 µl) containing 10% FBS was added to the lower chamber of the transwell. The cells that migrated into the lower chambers and those left in the upper chambers were collected and counted separately after 8-hour incubation at 37℃, 5% CO₂.

The 96-well plate was pre-coated with 50 µL 1.5% agarose (LONZA, Rockland, ME). Melanoma cells (A2058 and UACC 903 cells) were labeled with CellTracker Orange CMRA Dye (Thermo Fisher Scientific, Waltham, MA) and seeded at 2×10⁴ cells/well and allowed to form spheroids over 48 hours. Before starting the co-culture, PBMCs were stained with CFSE (Thermo Fisher Scientific) according to manufacturer procedure. Infiltrated spheroids were analyzed at 48 hours using a confocal microscope (Leica, Wetzlar, Germany). Spheroids were divided into single cells and CFSE+ cells infiltrating the spheroids were counted by flow cytometry.

Human melanoma samples were received from the Hospital of the University of Pennsylvania. Fresh tumor specimens were washed with PBS (containing 2% penicillin/streptomycin (P/S)) and minced into pieces of 1 mm. Fragments were enzymatically digested in DMEM (4.5 mmol/L glucose, 2% FBS, 100 U/mL collagenase type IV (200 U/mL); Sigma-Aldrich) and DNase I (50 U/mL; Sigma-Aldrich) for 20 min at 37℃. Samples were diluted in 25 mL media and pelleted, and resuspended in fresh DMEM and filtered over 100 µm strainers. Red blood cells were removed using RBC lysis buffer (Biolegend). Matrigel (Corning) and complete DMEM/F12 medium were mixed at 1:2 ratio and pre-seeded on 24-well plates to solidification. The cell mixture was seeded on the Matrigel layer and cultured for up to 10 days, and the medium was changed every 3 days. The culture medium was ADMEM/F12 composed of 1% P/S, 50% Wnt3a, RSPO1, Noggin-conditioned media (L-WRN, ATCC), nicotinamide (10 mM; Sigma), N-acetylcysteine (1 mM; Sigma), B-27 without vitamin A (1×; Invitrogen), A83-01 (0.5 mM; Tocris), EGF (50 ng/mL; Invitrogen), FGF (50 ng/mL; Peprotech) and Forskolin (10 ng/mL; Tocris).

Live/dead staining dual labeling was performed using ViaStain AOPI Staining Solution (Nexcelom, CS2-0106) following γδT cells co-cultured with organoids for 48 hours. The images were captured on Zeiss LSM 710 (Carl Zeiss, Thornwood, NY) after cell incubation with the dyes for 20 min at room temperature in the dark. Images were then analyzed using ZEN software.

Apoptosis of cells in the organoids was measured after γδT cell and organoid co-culture for 48 hours. Organoids were washed with cold PBS to remove the Matrigel and dissociated to single cells by trypsin (Gibco). Cells were washed twice with cold PBS and resuspended in 100 µL Annexin V Binding Buffer (FITC Annexin V Apoptosis Detection Kit; Biolegend). Then 5 µL of AV-FITC and 5 µL of 7-AAD solution were added and incubated for 15 min at room temperature in the dark. FACS was performed to measure AV and 7-AAD, yielding AV+/7-AAD− early apoptotic cells and AV+/7-AAD+late apoptotic/necrotic cells.

The cytotoxicity assays of γδT cells against organoids were also performed by the CytoTox 96 Non-Radioactive Cytotoxicity Assay. The lactate dehydrogenase activity released by organoids was detected and analyzed.

Data are analyzed by Prism V.8.0 (GraphPad Software, San Diego, CA) and R software. The significant differences between two groups were analyzed by Mann-Whitney U non-parametric test for unpaired samples and several groups were assessed by one-way ANOVA with multiple comparisons. Correlation analyses were performed with the non-parametric Spearman rank order test. P values <0.05 were considered significant difference (*p<0.05; **p<0.01). Logistic regression models were applied to explore if initial γδT cell proportion, each surface marker expression of γδT cell or their combination would be useful to predict whether γδT cells can be well expanded ≥50% in an individual blood sample. Receiver operating characteristic (ROC) curves were generated based on the value of initial γδT cell proportion or index obtained from multivariate models. AUC of each model’s ROC with the optimal sensitivity and specificity were calculated. Delong’s test was applied to determine the difference on AUC between regression models. A total of 66 samples have complete data. Due to such small sample size, complex multivariate predictive models were not considered for the data analysis.