Blood samples were collected from 33 healthy individuals and 25 cirrhosis patients. Cirrhosis was determined by clinical judgment at the discretion of the treating physicians based on a combination of biochemical parameters, clinical signs, and radiologic and ultrasonic laboratory tests. Age- and sex-matched healthy individuals were enrolled as controls. The study protocol was approved by the Institutional Review Board of Tsinghua University. Written informed consent was obtained from each subject.

Rag2^−/−γc^−/− mice were purchased from the Jackson Lab and maintained in an SPF animal facility at the Laboratory Animal Research Center, Tsinghua University. All of the animal experiments were approved by the Institutional Animal Care and Use Committee of Tsinghua University.

The human LX-2 cell line (purchased from Cancer Research Institute, Xiangya Medical School, Central South University, Changsha) and the Huh 7 cell line (purchased from Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences, Shanghai, China) were cultured in uncoated dish under DMEM containing 10% fetal bovine serum (FBS) in a 5% CO₂ atmosphere at 37°C.

BPH-series compounds were synthesized according to reported procedures, and were characterized by ¹H NMR, ³¹P NMR, and HR-MS. The full characterization of BPH-1236, a representative lipophilic bisphosphonate, is given as follows. ¹H NMR (400 MHz, D₂O) δ: 8.45 (s, 1 H), 7.28 (s, 1 H), 7.20 (s, 1 H), 4.45 (t, J = 9.6 Hz, 2 H), 4.00 (t, J = 7.2 Hz, 2 H), 1.67 (m, 2 H), 1.08 (m, 14 H), 0.62 (t, J = 7.2 Hz, 3 H). ³¹P NMR (162 MHz, D₂O) δ: 15.20.

Peripheral blood mononuclear cells (PBMCs) were isolated from whole blood samples of healthy donors using a standard Ficoll-Paque gradient centrifugation process. For large-scale cultures, PBMCs were cultured in RPMI 1640 medium supplemented with 10% FBS, 1% Penicillin-Streptomycin, 150 U/mL human rIL-2 (PeproTech), 2 mM l-glutamine, 50 µM β-mercaptoethanol, 1% MEM non-essential amino acids, and 5 µM zoledronate (Zometa; AvaChem Scientific) or 1 µM BPH-1236. The cultures were maintained at a cell density of 2 × 10⁶ cells/mL. For the expansion assay, PBMCs were cultured in 96-well round plates at a density of 1 × 10⁵ cells/well in culture medium with different concentrations of zoledronate or BPH-1236 with/without simvastatin. Fresh medium containing human rIL-2 (150 U/mL), but without test compounds, was added every 2–3 days. Cells were harvested on day 9–14, and the frequency and phenotype of Vγ9Vδ2 T cells were evaluated by flow cytometry. Vγ9Vδ2 T cells (purity >90%) were used in further experiments or were stored in liquid nitrogen. In some experiments, expanded Vγ9Vδ2 T cells were further purified by using Anti-TCR γδ MicroBead Kits (Miltenyi Biotec), according to the manufacturer’s instructions. These experiments were carried out with the donor’s consent and were approved by the Institutional Review Board of Tsinghua University.

The expanded cells were washed and suspended in 100 µL buffer containing PBS (pH 7.2), bovine serum albumin (0.5%), and EDTA (2 mM). Vγ9Vδ2 T cells were labeled using PE or FITC-conjugated anti-human TCR Vδ2 antibody and APC or FITC-conjugated anti-human CD3 antibody (Miltenyi Biotec). For phenotype analysis, the expanded cells were additionally labeled with Vioblue-conjugated anti-human CD27 and PE-conjugated anti-human CD45RA (Miltenyi Biotec). The staining step was done at 4°C for 30 min; samples were washed with 1 mL PBS and were then analyzed with a FACSAria SORP (BD) flow cytometer. Data were analyzed using the FlowJo program.

CytoTox 96^® Non-Radioactive Cytotoxicity Assay kits (Promega), which are based on the colorimetric detection of the released enzyme lactate dehydrogenase (LDH), was used to determine specific cytotoxicity. Vγ9Vδ2 T cells (Effector, E) were co-cultured with LX-2 cells (Target, T), which were pretreated with N-BPs for 4 h at specific E/T ratios for 4 h (indicated in figure captions). In some cases, after 4 h co-cultured, the number and area of Vγ9Vδ2 T cells clusters were determined using IncuCyte (Essen BioScience) image analysis software. Three different locations per well were imaged with a 10 × objective lens. Clusters were defined as cell aggregates occupying an area at least 300 µm² and were displayed as the number and area of clusters per well. In some experiments, Transwell^® units (pore size 0.4 µm, Corning) were used to separate Vγ9Vδ2 T cells from LX-2 cells. In some cases, the neutralization antibodies anti-NKG2D (10 µg/mL, BD), anti-FasL (10 µg/mL, Biolegend), anti-TRAIL (10 µg/mL, Biolegend), anti-TCR γδ (10 µg/mL, Biolegend), and their relevant isotype controls, were individually added in the co-cultures to block the NKG2D-, FasL-, TRAIL-, and TCR γδ- mediated pathways. To block perforin and granzyme B pathways, the perforin inhibitor Concanamycin A (CMA, Selleck) (1 µg/mL) and granzyme B inactivator BCL-2 (1 µg/mL, R&D Systems) were used (15).

Atomic force microscopy-based single-cell force spectroscopy (AFM-SCFS) was performed with a JPK CellHesion unit as previously described (16). Briefly, LX-2 cells treated with or without 5 µM BPH-1236 were cultured on an uncoated glass substrate that was placed in an AFM-compatible environmental chamber to maintain 37°C and 5% CO₂. A Cell-Tak (Sigma-Aldrich) coated cantilever was used to glue individual Vγ9Vδ2 T cells and glued cells were used immediately to interact with adherent LX-2 cells. In each approach-retract probing cycle, the AFM cantilever carrying the Vγ9Vδ2 T cell was pressed to create contact with an LX-2 cell with a setpoint of 0.5 nN for a period of 2 s. Then, the cantilever was retracted from the cell of interest until detachment was achieved. The probing process was repeated until a minimum of 10 force curves were collected for each Vγ9Vδ2 T cell-LX-2 cell pair. Five independent pairs of Vγ9Vδ2 T cell-LX-2 cell were tested. The force curves were analyzed using JPK image processing software (17).

5(6)-Carboxyfluorescein N-hydroxysuccinimidyl ester (CFSE; an esterase substrate that fluoresces when hydrolyzed, eBioscience) labeled LX-2 cells (green) were pretreated with BPH-1236 (5 μM) for 4 h and then co-incubated with expanded Vγ9Vδ2 T cells labeled with LysoTracker Red (red, Life Technologies) for 1 h. Visualization of perforin/granzyme release at the single-cell level was performed using time-lapse confocal microscopy (Nikon A1Rsi). Data were analyzed using NIS Viewer.

In some cases, Vγ9Vδ2 T cells were co-incubated with LX-2 cells (pretreated with BPH-1236 for 4 h) in the presence of PI (Beyotime, China) and Hoechst 33258 (Life Technologies). Visualization of PI diffusion at the single-cell level was performed using Spinning Disk with FV1200 microscope (PerkinElmer). Images were acquired about every 88 s, with the overlay of the Hoechst 33258 (blue)/PI (red)/Brightfield was shown. Data were analyzed using Volocity (PerkinElmer).

LX-2/luciferase (Luc) cells and Huh 7/Luc cells were established by the stable transfection of the firefly luciferase gene (pPLV-Luc-GFP; kind gift from Professor Yanan Du, Tsinghua University) using linear polyethylenimines (PEI; Polysciences) reagent.

Huh 7/Luc cells were cultured in 96 round bottom plates with culture medium or LX-2 conditioned medium (CM) (72 h culture supernatant from LX-2 cells) for 72 h, the number of viable cells was determined by measuring luciferase activity after adding luciferin (China Cellcyto) by using an IVIS imaging system (PerkinElmer).

Huh 7/Luc cells were seeded on 96-well plates at 2 × 10⁴ cells per well and incubated overnight. Cells were “wounded” using a cell wound maker (ESSEN Bioscience) and cultured with/without LX-2 cells conditioned medium (CM). To measure cell migration, microscopic photographs were taken at 0 and 48 h after injury using an Operetta CLS High-Content Analysis System (PerkinElmer).

DiR (1,1′-dioctadecyltetramethyl indotricarbocyanine iodide; GeneCopoeia) labeled Vγ9Vδ2 T cells (10 × 10⁶ per mouse) were i.v. injected into Rag2^−/−γc^−/− mice (6–8 weeks old), and the distribution of Vγ9Vδ2 T cells was evaluated by measuring DiR fluorescence using an IVIS imaging system (PerkinElmer).

The liver orthotopic model was established using a liver submucosal injection method. Liver submucosal injection accessed by midline laparotomy using aseptic technique under a stereomicroscope. Specifically, 1 × 10⁶ LX-2/Luc cells or 1 × 10⁶ Huh 7/Luc cells (suspended in 20 µL Opti-MEM) were injected into liver submucosal of Rag2^−/−γc^−/− mice of 6–8 weeks age under anesthesia using isoflurane. About 7 days later, when discernible and orthotopic LX-2 cells xenografts (or Huh 7 tumors) formed, mice were divided into groups and treated with Vγ9Vδ2 T cells (10 × 10⁶/mouse, i.v.) with or without BPH-1236 (1 mg/kg, i.v.) (BPH-1236 was administrated 4 h before Vγ9Vδ2 T cells injection). The treatments were given at day 7 (after transplantation) in LX-2/Luc cells orthotopic mice, and at day 7 as well as day 15 in Huh 7/Luc cells orthotopic mice. The tumor burden of Huh 7/Luc cells and LX-2/Luc cells xenografts in livers were evaluated by measuring luciferase activity after luciferin injection by using an IVIS imaging system (PerkinElmer). The liver and tumors were harvested and photographed at day 48 after transplantation from Huh 7/Luc cells orthotopic Rag2^−/−γc^−/− mice. The tumor sizes were measured with a digital caliper and tumor volume (cm³) was calculated as V = long diameter × short diameter² × 0.5.

1 × 10⁶ Huh 7/Luc cells and/or 1 × 10⁶ LX-2/Luc cells (suspended in 50 µL Opti-MEM) were injected into spleens of 6–8 weeks old Rag2^−/−γc^−/− mice under isoflurane anesthesia. The tumors in livers and spleens were harvested at day 43. Survival of mice was monitored and recorded.

The two-tailed Student’s t-tests were used for comparing the significance of differences between groups. Statistical analyses and graphing were performed using GraphPad Prism 5 software. All values are presented as mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001.

We determined the frequencies of peripheral circulating Vγ9Vδ2 T cells in healthy donors (n = 33) and in cirrhosis patients (n = 25; n = 22 hepatitis B-progressed cirrhosis, n = 3 alcoholic cirrhosis). The number of peripheral Vγ9Vδ2 T cells in cirrhosis patients was reduced by ~1.8 × in cirrhosis patients compared to healthy controls (Figure 1A), suggesting the possibility of a protective role for this subset of T cells in the development of liver cirrhosis. We next tested the hypothesis that Vγ9Vδ2 T cells might be cytotoxic to aHSCs. We used LX-2 cells, a widely accepted human hepatic stellate cell line that is activated when cultured in uncoated plastic dishes and that has been extensively used in studies of the progression of human hepatic fibrogenesis (18, 19). We evaluated cytotoxicity by monitoring the release of an LDH marker (20) and found that aHSCs were killed by ex vivo-expanded Vγ9Vδ2 T cells from healthy donors. These Vγ9Vδ2 T cells killed 18% of aHSCs at an effector:target (E:T) ratio of 5:1, and 35% of aHSCs at an E:T ratio of 30:1 (Figure 1B).

Vγ9Vδ2 T cells are currently being used in adoptive immunotherapies to target a broad range of cancers including leukemia (21), melanoma (22), colon carcinoma (23), and breast cancer (24), among others (25–27). Vγ9Vδ2 T cells recognize tumor cells by sensing their increased accumulation of the phosphorylated metabolite isopentenyl pyrophosphate (IPP). We speculated that activated Vγ9Vδ2 T cells might be able to recognize aHSCs in a similar manner. Biochemically, IPP is a substrate of the farnesyl diphosphate synthase (FPPS) enzyme, and inhibition of FPPS by drugs such as nitrogen-containing bisphosphonates is known to result in elevated IPP levels (Figure 2A). IPP can bind to and induce a conformational change in the transmembrane protein butyrophilin 3 A1 (BTN3A1) (28–32), which is recognized by the Vγ9Vδ2 T cell receptor, ultimately resulting in target cell killing. In theory, assuming that a similar mechanism is involved in the killing of aHSCs by Vγ9Vδ2 T cells, the increase in IPP levels resulting from bisphosphonate treatment should increase the susceptibility of aHSCs to cytolysis by Vγ9Vδ2 T cells.

We used zoledronate, a potent bisphosphonate drug in current clinical use (33), to test this hypothesis by monitoring the ability of Vγ9Vδ2 T cells to kill zoledronate-treated aHSCs (LX-2 cells). At an E:T ratio of 10:1, LX-2 cells pretreated with 5 µM zoledronate were far more susceptible to killing by Vγ9Vδ2 T cells than were untreated control cells, and zoledronate alone had essentially no effect on LX-2 cells viability (Figure 2B). Consistently, zoledronate-pretreated cells were more extensively surrounded by clustered Vγ9Vδ2 T cells than were the untreated cells (Figures 2C,D). Together, these results show that human Vγ9Vδ2 T cells can kill aHSCs and demonstrate that the efficiency of this killing can be enhanced by treating aHSCs with the bisphosphonate drug, zoledronate.

Zoledronate and other bisphosphonates were initially developed for the treatment of bone diseases such as osteoporosis, Paget’s disease, and hypercalcemia due to malignancy. Chemically, they are extremely polar, and are rapidly removed from blood circulation via binding to bone (34). While these features are desirable for bone-targeting drugs, they represent challenges for their use in other indications. For example, the poor pharmacokinetic properties of zoledronate (bone affinity and short circulation) are associated with risks of osteonecrosis of the jaw and renal impairment. In previous work, we generated a new class of drug leads called lipophilic bisphosphonates for the treatment of both KRAS-driven lung adenocarcinomas (35) and malaria (36). These compounds have almost no affinity for bone (37), and remain in circulation much longer than do conventional bisphosphonates (35). Building from our discovery that zoledronate can enhance the killing of aHSCs by Vγ9Vδ2 T cells, we next explored the possibility that the lipophilic bisphosphonates might have translational potential as lead compounds for treating liver disease.

We tested the ability of 10 lipophilic analogs of zoledronate (structures shown in Figure 3A) with various alkyl chain lengths to potentiate the effects of Vγ9Vδ2 T cells killing of aHSCs (LX-2 cells). These alkyl chains do not interfere with the enzyme binding but increase the lipophilicity of the molecule. Most of these bisphosphonates are potent FPPS inhibitors (35) and as expected they increased the susceptibility of aHSCs to cytolysis by Vγ9Vδ2 T cells, and BPH-1236 was the most potent compound tested (Figure 3B). Then we ask the question whether BPH-1236 functions by increasing the levels of IPP, a danger signal recognized by Vγ9Vδ2 T cells.

In preclinical or clinical settings, zoledronate stimulates PBMCs to expand Vγ9Vδ2 T cells by increasing the IPP levels inside PBMC. We found that pulsing BPH-1236 onto PBMC also efficiently stimulated a major expansion of the number of Vγ9Vδ2 T cells, with a 10-fold lower EC₅₀ than did zoledronate (Figure 4A), and an increased population of effector memory cells (Figure S1 in Supplementary Material). The dual biological effects of BPH-1236 (sensitizing aHSCs to Vγ9Vδ2 T cells killing and ex vivo Vγ9Vδ2 T cells stimulation) are consistent with its FPPS inhibitor function in increasing IPP levels. To further confirm this, we used simvastatin, an HMG-CoA reductase inhibitor that inhibits IPP production, to rescue these effects. As expected, treatment with a combination of BPH-1236 plus simvastatin greatly diminished aHSCs killing (Figure 4B) and ex vivo Vγ9Vδ2 T cells stimulation by BPH-1236 (Figure 4C). Thus, clearly, BPH-1236 functions by increasing IPP levels in aHSCs, making them more susceptible to Vγ9Vδ2 T cells recognition and killing.

We next sought to investigate in more detail the cell–cell recognition and killing of aHSCs by Vγ9Vδ2 T cells. It is reported that IPP inside the target cell will bind to the intracellular part of BTN3A1, inducing an extra-cellular conformation change that is recognized by TCR γδ (28–32). We deduced that a cell-to-cell contact is needed between aHSCs and Vγ9Vδ2 T cells for recognition and cytolysis. Using a Transwell^® apparatus, we found that killing of LX-2 cells by Vγ9Vδ2 T cells occurred only under cell-to-cell contact conditions (Figure 5A). To quantify this cell–cell interaction, we used a technology called AFM-SCFS, a method that enables direct measurement of the adhesion force between individual pairs of interacting cells in vitro (Figure 5B) (16, 38, 39). We glued Vγ9Vδ2 T cells to the tip of a flat cantilever and used it to approach LX-2 cells placed on a glass substrate. The binding forces were measured using a cyclical approach-retract method. In the retraction phase, an average force of 280 ± 10 piconewtons was required for complete detachment (Figure 5C). However, pretreatment of the LX-2 cells with BPH-1236 increased the force (Figure 5C; Figure S2A in Supplementary Material) or the work (Figures S2A,B in Supplementary Material) required to detach cells by a factor of two. This BPH-1236 mediated increase in the adhesion strength between LX-2 cell and Vγ9Vδ2 T cell is consistent with our observation that BPH-1236 treatment enhances the ability of Vγ9Vδ2 T cells to kill aHSCs.

We next investigated the signaling cascades evoked by aHSCs-Vγ9Vδ2 T cells recognition. Vγ9Vδ2 T cells kill cancerous cells in a process that involves the TCR γδ and NKG2D receptors, with engagement of Fas-FasL and TRAIL-DR5 (40). Here, we used antibodies to separately block the TCR γδ or NKG2D receptors. This reduced the susceptibility of aHSCs to killing by Vγ9Vδ2 T cells by 30% (a-TCR γδ) and 22% (a-NKG2D) (Figure 6A). Similar modest reductions in susceptibility were observed when we blocked FasL (20%) or TRAIL (33%). However, by far the largest effect on cytolytic activity (~87% inhibition) was observed following the addition of the V-H⁺-ATPase inhibitor, CMA. This compound blocks release of perforin, and hence inhibits perforin/granzyme B lytic activity, although addition of the granzyme B inhibitor BCL-2 alone had a smaller effect (34% inhibition) (Figure 6A).

To visualize the effects of the perforin and granzyme proteins in the killing of BPH-1236 pretreated LX-2 cells by Vγ9Vδ2 T cells, we used LysoTracker Red to trace acidic lytic granules (red) in Vγ9Vδ2 T cells, and CFSE (green) to monitor LX-2 cells. Time-lapse confocal microscopy revealed that lytic granules rapidly converged to the Vγ9Vδ2 T cell/LX-2 cell immune synapse. These granules were then released into the LX-2 cells, resulting in their lysis into many membrane-bound bodies, and cell apoptosis (Figures 6B,C; Video S1 in Supplementary Material). Consistent with the target cell blebbing and lysis, we observed an increase in PI fluorescence over time in the Vγ9Vδ2 T cells hit LX-2 cells when compared with non-hit LX-2 cells (Figures 6D,E). Thus, aHSCs that are pretreated with the lipophilic bisphosphonate BPH-1236 are killed primarily by perforin/granzyme lytic granules released from Vγ9Vδ2 T cells.

We then investigated the translational potential of Vγ9Vδ2 T cells for the treatment of aHSCs-mediated diseases. Vγ9Vδ2 T cells target molecules or cell types that are only found in humans and other primates (41). Specifically, stellate cells from mouse liver lack the BTN3A1 gene and are not recognized by Vγ9Vδ2 T cells (41), thus the direct anti-fibrotic efficacy of Vγ9Vδ2 T cells can’t be evaluated in conventional mouse models. We thus chose to test whether Vγ9Vδ2 T cells can kill human aHSCs in immunodeficient mice. It is known that the liver is one of the main organs in which the “homing-in” of adoptively transferred Vγ9Vδ2 T cells occurs (42), implying that liver diseases may be amenable to treatment with Vγ9Vδ2 T cells. In a Rag2^−/−γc^−/− immune-deficient mouse model (15), we observed that ex vivo-expanded Vγ9Vδ2 T cells (tail vein injection, stained with a non-diffusing cytoplasmic membrane probe XenoLight DiR) trafficked predominantly to mice livers (Figure S3 in the Supplementary Material). We next assessed the in vivo cytotoxicity of Vγ9Vδ2 T cells against aHSCs in an orthotopic mouse model in which LX-2/Luc cells (luciferase-tagged LX-2 cells) were injected into the tunica serosa of the livers of Rag2^−/−γc^−/− mice. One week after injection, mice were treated with BPH-1236 (1 mg/kg), followed by the adoptive transfer of 1 × 10⁷ Vγ9Vδ2 T cells (>90% purity). BPH-1236 treatment greatly enhanced the in vivo killing efficacy of Vγ9Vδ2 T cells against aHSCs (Figures 7A,B). Our results with this orthotopic model thus clearly suggested the potential for using Vγ9Vδ2 T cells in combination with a lipophilic bisphosphonate to treat aHSCs driving liver diseases (e.g., liver fibrosis, cirrhosis, and even HCC).

Activated hepatic stellate cells have been reported to promote HCC tumorigenicity (43). We also found that LX-2 cells (aHSCs) conditioned medium (CM) (72 h culture supernatant from LX-2 cells) stimulated human Huh 7 cell (an HCC cell line) growth in vitro and increased Huh 7 cell migration (Figures S4A,B in Supplementary Material). We then used an intra-splenic injection model (in vivo) and found that Rag2^−/−γc^−/− mice co-injected with Huh 7 cells and LX-2 cells developed more severe liver metastases than did the Huh 7 cells-alone control group, in addition to having lower survival rates (Figures S4C,D in Supplementary Material). We found that BPH-1236 enhanced the killing effect of Vγ9Vδ2 T cells against Huh 7 cells in vitro (Figure 8A), as seen with aHSCs. This is not unexpected since cancerous cells have been reported as the main target cells of Vγ9Vδ2 T cells (27). The combination of ex vivo expanded Vγ9Vδ2 T cells with BPH-1236 efficiently also shrunk orthotopic HCC tumor burden in Rag2^−/−γc^−/− mice (Figures 8B–E). It is thus clear that Vγ9Vδ2 T cell adoptive transfer with a lipophilic bisphosphonate holds promise as a strategy for fibrosis–cirrhosis associated HCC treatment, since such a strategy killed not only aHSCs (Figure 7), the driving force of HCC, but also tumors (Figure 8) directly.