NK cells that are activated by CXCL10 can kill dormant tumor cells that resist CTL-mediated lysis and can express B7-H1 that stimulates T cells

Tumor dormancy is a condition in which tumor cells persist in the host for a long time but do not proliferate.¹  It is a common phenomenon in cancer, and eradication of tumor dormant cells is an important goal toward the definitive cure of malignancies.

We previously reported in the DA1-3b/C3H mouse model of acute myeloid leukemia (AML) that vaccination of mice with leukemia cells that had been transduced with CD154, granulocyte-macrophage colony-stimulating factor (GM-CSF), or interleukin 12 (IL-12) genes led to significant protection against leukemia, and generated specific cytotoxic T lymphocytes (CTLs) against leukemic cells.^2-4  Despite these responses, we detected several hundred persistent leukemic cells in some long-term surviving mice. Persistent leukemic cells were more resistant to specific CTL-mediated killing because they expressed more B7-H1 and B7.1, and expression was proportional to the time they had persisted in the host.⁴ 

Eradication of tumor dormant cells in AML would require the generation of cytotoxic cells that cannot be inhibited by B7-H1 or B7.1. We observed in previous experiments that minimal residual disease (MRD) was less common in the spleen of mice vaccinated with irradiated IL-12–transduced cells than those vaccinated with cells transduced with other genes, like CD154.⁴  CXCL10 (also called interferon-inducible protein 10 [IP-10]) is an essential mediator of the antitumor effect of IL-12.^5-7  CXCL10 is a CXC chemokine structurally and functionally related to monokine induced by interferon γ (IFN-γ; Mig, CXCL9) and IFN-inducible T-cell α chemoattractant (I-TAC, CXCL11).^8-10  CXCL10, like CXCL9 and CXCL11, acts though the chemokine receptor CXCR3. These chemokines attract activated T cells and natural killer (NK) cells.^11-14  CXCL10-deficient mice also had a defect in effector T-cell generation, indicating that CXCL10 might also be essential to T-cell adaptive immunity.¹⁵  We therefore used the DA1-3b mouse model of AML to investigate if CXCL10 could induce an efficient antileukemic immune response.

Female C3H/HeJ mice, 8 to 10 weeks old, were obtained from Charles River Laboratories (Lyon, France). The leukemic DA1-3b BCR/ABL-expressing cell line has been described previously.^2,16  Mice injected intraperitoneally with 10⁴ to 10⁶ DA1-3b cells developed systemic lethal leukemia in 3 to 4 weeks, with massive infiltration of bone marrow, spleen, and peripheral blood.

pGT-IL12, pGT-CXCL10, and pGT-Zeo were purchased from Invivogen (Toulouse, France). DA1-3b cells were electroporated at 1500 μF and 250 V and further selected with Zeomycin (Invitrogen). Large batches of selected transduced cells were frozen until use. Expression of IL-12 and CXCL10 was assessed by enzyme-linked immunosorbent assay (ELISA; IL-2, OptEIA kit; CXCL10, rat IgG1 anti-mIP-10; PharMingen, San Diego, CA).

Groups of 10 C3H/Hej mice were injected intradermally with irradiated (100 Gy from a ¹³⁷C source) IL-12– or CXCL10-transduced cells (termed DA1-3b/IL12 and DA1-3b/CXCL10, respectively). DA1-3b cells transfected with the pGT-Zeo empty plasmid (DA1-3b/Zeo) were used as controls. Cells were diluted in 200 μL phosphate-buffered saline (PBS) before injection. Each mouse received 3 injections, a week apart, and immunity was challenged a week after the last injection by intraperitoneal injection of 10⁴ to 10⁶ leukemic DA1-3b cells. For therapeutic experiments, mice were first injected intraperitoneally with 10⁴ to 10⁶ leukemic DA1-3b cells, and then 3 days later received the first of 3 weekly intradermal injections of 10⁶ irradiated DA1-3b/CXCL10, DA1-3b/IL12, or DA1-3b/Zeo cells. Experiments were repeated at least 3 times. In mice that survived 1 year, persisting leukemic cells were quantified by real-time quantitative polymerase chain reaction (RQ-PCR) on gDNA extracted from all organs.⁴ 

Hybridoma cell lines GK-1.5 (rat antimouse CD4⁺) and 53-6.72 (rat antimouse CD8⁺) from American Type Culture Collection (Rockville, MD), and antimouse IFN-γ (clone XMG1.2), antimouse tumor necrosis factor α (TNF-α; clone MP3-XT3; PharMingen), and antimouse B7-H1 (clone MIH5; Clinisciences, Montrouge, France) were used as a source of monoclonal antibodies (mAbs). Groups of 11 C3H/Hej mice were vaccinated on 3 occasions, 1 week apart, by intradermal injection of 10⁶ transduced DA1-3b cells, or DA1-3b/Zeo cells as controls, irradiated at 100 Gy. For depletion, mice were then injected intraperitoneally on 3 consecutive days with an anti-CD4 or anti-CD8 (500 μg/injection) and 1 week later antileukemic immunity was challenged by intraperitoneal injection of 10⁴ wild-type DA1-3b cells. Efficacy of depletion was checked by flow cytometry on splenocytes from one mouse per group. For depletion of NK cells, mice were injected intraperitoneally with 500 μg antiasialo GM1 (Wako, Osaka, Japan) on each of 3 consecutive days and weekly after the challenge. To block IFN-γ and TNF-α, mice were injected intraperitoneally on each of 3 consecutive days with the blocking antibody or with a control isotype (500 μg/injection). To block B7-H1, mice were first injected with 10⁶ irradiated transduced DA1-3b cells. Each mouse received 3 intraperitoneal injections, a week apart and 2 days after the last injection, of 500 μg anti-B7-H1 (clone MIH5), or the isotype control mAb (rat IgG2a, λ, Clinisciences).^4,17,18 

Recruitment of immune cells induced by transduced leukemic cells was analyzed using a matrigel technique adapted from Kowalczyk et al.¹⁹  The matrigels were introduced by intradermal injection of a mixture of 0.5 mL liquefied matrigel (BD Biosciences, San Jose, CA) and 5 × 10⁶ irradiated DA1-3b/Zeo, DA1-3b/IL12, or DA1-3b/CXCL10 cells diluted in 200 μL PBS. Ten mice were killed at day 1, 2, 3, 5, 7, 10, and 15, and matrigels from the site of vaccination were harvested by skin excision. Matrigels were liquefied in 0.3 mL PBS at 4°C overnight. Isolated cells were analyzed by flow cytometry with mAbs against CD8-fluorescein isothiocyanate (FITC; clone 53-3.1), CD3-phycoerythrin (PE; clone 17A2), CD4-PE (clone H129.19), pan NK-FITC (clone DX5), B7.1-FITC (clone 1G10), B7.2-PE (clone GL1), and CD154-PE (clone MR1) from PharMingen; PD1-FITC (clone J43), FasL-PE (clone MFL3), B7-H1-PE (clone MIH5), OX40L-PE (clone RM134L), OX40-FITC (clone ACT-35), 4-1BB-PE (clone 17B5), and 4-1BBL-PE (clone TKS-1) from Clinisciences; B220-PE (clone RA3-6B2), CD19-FITC (clone 1D3), CD11c-PE (clone HL3), and CD11b (clone M1/70) from PharMingen.

Mice were injected subcutaneously with 10⁶ irradiated DA1-3b cells transduced with pGT-IL12, pGT-CXCL10, or pGT-Zeo. Spleens were removed from these mice and from naive mice, and lymphocytes were isolated with a Miltenyi-Biotec separation column and anti-CD90 (THY1.2) mAb, according to the manufacturer's recommendations (Miltenyi Biotec, Bergisch Gladbach, Germany). CD90⁺ cells were cocultured with irradiated DA1-3b cells and 20 IU murine IL-2/mL (PreproTech, Rocky Hill, NJ) at 5 mL/well in a 6-well tissue culture plate. For the CTL assay, CD8⁺ cells were isolated with the CD8a⁺ T-cell isolation kit (Miltenyi Biotec). CTL activity was measured after 15 days of coculture, using the Cytotox 96 nonradioactive assay (Promega, Madison, WI) with cultured cells from vaccinated mice as effectors, and DA1-3b cells as targets, mixed at various ratios. Major histocompatibility complex (MHC) class I restricted CTL lysis of DA1-3b cells was verified using anti-MHC I and anti-MHC II blocking antibodies.⁴  Absence of NK cell-mediated cytotoxicity was verified with the YAC-1 cell line. Specificity of CD8⁺ CTL lysis against DA1-3b cells was controlled with the EL4 cell line.

For NK cell isolation, mice were injected intradermally with 10⁶ irradiated DA1-3b cells transduced with pGT-IL12, pGT-CXCL10, or pGT-Zeo, and injections were repeated weekly 3 times. One week later mice were humanely killed and the spleen was removed. NK cells were isolated using an NK cell isolation kit according to the manufacturer's recommendations (Miltenyi Biotec). Purity of NK cell samples was analyzed by flow cytometry using anti-B220 and anti-CD122 mAbs (PharMingen). NK cell cytotoxicity was analyzed at the day of isolation using the Cytotox 96 nonradioactive cytotoxicity assay (Promega) by mixing NK cells isolated from the spleen as effectors and DA1-3b or YAC-1 cells as targets at various ratios. To inhibit perforin-mediated cytotoxicity, effector cells were incubated for 2 hours in 100 nM concanamycin A (CMA; Sigma, St Louis, MO).

For blocking B7-H1 and B7.1, 0.5 μg/10⁶ cells of anti-B7.1 (clone 1G10; PharMingen) or anti-B7-H1 (clone MIH5) and corresponding isotype control mAbs (Clinisciences) were added to the effectors cells or the targets cells as appropriate, and cells were incubated for 30 minutes at 4°C.^17,20-23 

Purified NK cells from naive mice were incubated with recombinant IL-2 (200 IU/mL), IL-3 (200 IU/mL), IL-4 (10 ng/mL), IL-5 (30 IU/mL), IL-7 (50 ng/mL), IL-10 (100 IU/mL), IL-12 (20 IU/mL), IL-13 (50 ng/mL), IL-15 (100 ng/mL), IL-18 (50 ng/mL), GM-CSF (100 ng/mL), IFN-γ (200 IU/mL), TNF-α (500 IU/mL), CXCL9 (100 ng/mL), CXCL10 (100 ng/mL), CXCL11 (200 ng/mL), CXCL12 (40 ng/mL), CCL3 (100 ng/mL), CCL5 (100 ng/mL), or CCL11 (100 ng/mL). Flow cytometry analysis of B7-H1 expression was performed after 6 and 24 hours of incubation. Human NK cells were isolated from 5 healthy volunteers who had provided their informed consent. The cell isolation was performed using an NK cell isolation kit (Miltenyi Biotec), following which cells were incubated with recombinant IL-2, IL-3, IL-4, GM-CSF, IL-12, IFN-γ, TNF-α, CXCL9, CXCL10, CXCL11, CXCL12, CCL3, CCL5, and CCL11 at the same doses used in the mouse experiments. Human CD56^brightCD16^dim and CD56^dimCD16^bright NK cells subpopulations were purified by cell sorting on a EPICS ALTRA sorter (Beckman Coulter, Marseille, France) based on mean fluorescence intensity (MFI) expression, assessed using human CD56 (clone B159; PharMingen) and CD16 mAb, incubated in vitro with recombinant CXCL10, and then analyzed for B7-H1 expression after 6 hours and 24 hours.

Mouse or human CD8⁺ or CD4⁺ T cells were washed in PBS, suspended at 10⁶ cells/mL, and labeled with 10 μM carboxyfluorescein diacetate succinimidyl ester (CFSE; Molecular Probes Europe, Leiden, The Netherlands) for 10 minutes at 37°C. Cells were again washed twice before addition to cocultures with NK cells at a 1:1 ratio and with 200 IU/mL IL-2. At the indicated times CFSE-labeled cells were collected, washed twice in PBS/1% bovine serum albumin (BSA), and fixed in PBS/1% paraformaldehyde. For 2-color fluorescence studies, CFSE-labeled cells were incubated with saturating concentrations of directly PE-conjugated antimouse CD4 or CD8 mAbs for 30 minutes at 4°C, washed twice, and subjected to flow cytometry analysis. Unlabeled and labeled CD8⁺ and CD4⁺ T cells, fixed immediately after CFSE labeling, were used as negative and positive internal controls for staining. Expression of mouse IFN-γ, TNF-α, IL-4, and IL-10, and human IFN-γ, TNF-α, IL-4, and IL-10 were assessed by intracellular staining with appropriate antibodies (PharMingen). Mouse and human B7-H1 were blocked with antimouse B7-H1 (clone MIH5) and antihuman B7-H1 (clone MIH1) antibodies, respectively. Mouse and human programmed death 1 (PD-1) were blocked with antimouse PD-1 (clone J43) and antihuman PD-1 (clone J116) antibodies, respectively.

Statistical analysis was performed using the Sigma Stat 3.0 software (SPSS Science, Chicago, IL).

DA1-3b cells transfected with pGT-CXCL10 (DA1-3b/CXCL10) or pGT-IL12 plasmid (DA1-3b/IL12) produced 3600 ± 20 pg CXCL10/10⁶ cells/mL/24 h and 2100 ± 55 pg IL12/10⁶ cells/ml/24 h, respectively. In prophylactic experiments, mice vaccinated with irradiated DA1-3b/CXCL10 cells showed better survival than those vaccinated with DA1-3b/IL12 or control DA1-3b/Zeo cells (Figure 1A-1B), even when challenged with 10⁶ wild-type DA1-3b cells (Figure 1C).

Next, we examined the therapeutic effect of DA1-3b/CXCL10 cells on established leukemia. Seventy-nine percent of mice injected with 10⁴ DA1-3b cells survived when given irradiated DA1-3b/CXCL10 cells, but only 38% of mice that received irradiated DA1-3b/IL12 cells did so (Figure 1D). Irradiated DA1-3b/CXCL10 cells increased long-term survivals when 10⁵ DA1-3b cells were injected (Figure 1E) and delayed death of mice injected with 10⁶ cells (P < .001, log-rank test; Figure 1F). Mice vaccinated with DA1-3b/CXCL10 cells that survived to the first injection of 10⁴ DA1-3b cells were rechallenged 3 months later with 10⁴ or 10⁵ leukemic cells. As shown in Figure 2A-2B, all mice were still protected from leukemia, demonstrating a long-term immunologic memory. The immunity induced by DA1-3b/CXCL10 cells was dose-dependent. Mice vaccinated with 10⁶ DA1-3b/CXCL10 cells were significantly better protected than those injected with only 10⁵ cells (P = .037, log-rank test; Figure 2C). However, vaccination with 10⁷ cells resulted in a lower survival rate (P = .005, log-rank test).

To investigate if CXCL10 could also eradicate persistent leukemic cells, we vaccinated mice with irradiated DA1-3b/CXCL10, DA1-3b/IL12, DA1-3b/CD154, or DA1-3b/Zeo cells, challenged immunity with live DA1-3b cells, and then extracted DNA of all organs of every mouse that survived 1 year. Leukemic cells were found in the spleen and liver of mice vaccinated with DA1-3b/IL12, DA1-3b/CD154, and DA1-3/Zeo, as previously described.⁴  However, no DA1-3b cells could be found in 65 mice from the DA1-3b/CXCL10 vaccine group (Table 1) and the difference with the DA1-3b/IL12 group was significant (P < .001, χ² test).

To identify the effectors of the antileukemic immunity induced by CXCL10, we vaccinated mice with irradiated DA1-3b/CXCL10 or with control DA1-3b/Zeo cells, depleted them with anti-CD4, anti-CD8, or anti-NK antibodies or control isotypes, and challenged immunity by injection of 10⁵ wild-type DA1-3b cells. As shown in Figure 3A, the protective effect of CXCL10 was completely dependent on NK cells. Depletion of CD4 or CD8 T cells partially abolished the vaccine effect (Figure 3B-C).

We also injected blocking antibodies or control isotypes against IFN-γ and TNF-α. The CXCL10 vaccine effect was completely dependent on IFN-γ but independent of TNF-α (Figure 3D-E).

To investigate which cells the leukemia vaccine attracted, we diluted 10⁶ irradiated DA1-3b/IP10, DA1-3b/IL12, or control DA1-3b/Zeo cells in matrigel and injected 100 μL intradermally into mice. Gels were excised at different times and flow cytometry analysis of cells in the resolubilized matrigel revealed a progressive infiltration by CD3⁺, CD3⁺CD4⁺, and CD3⁺CD8⁺ T cells induced in similar proportions by DA1-3b/IL12 or DA1-3b/CXCL10 cells (Figure 4A-C). However, DA1-3b/CXCL10 cells induced significantly higher infiltration of the vaccine site by NK cells than DA1-3b/IL12 cells (Figure 4D). No B220⁺CD19⁺B cells could be detected in the infiltrates. CD11c⁺CD11b⁺ dendritic cells were detected only 15 days after injection and represented 11% ± 3%, 12% ± 3%, and 16% ± 4% of cells in infiltrates from DA1-3b/Zeo, DA1-3b/IL12, and DA1-3b/CXCL10 matrigels, respectively. These slight differences were statistically significant (P = .021, Student t test).

Vaccine cells induced also a significant modification of the proportion of immune cells distant from the vaccine site. Flow cytometry analysis of the spleen of vaccinated animals showed higher proportions of CD3⁺ (P = .004) and CD3⁺CD4⁺ (P = .014) T cells in mice vaccinated with DA1-3b/IL12 cells and higher proportion of NK cells (P = .009) in mice vaccinated with DA1-3b/CXCL10 cells (Table 2).

To evaluate the relative role of CD8⁺ CTL and NK cells as effectors of the antileukemic immune response and in the control of dormant tumor cells, we isolated T cells and NK cells from mice vaccinated with DA1-3b/CXCL10, DA1-3b/IL12, control DA1-3b cells or naive mice as previously described.⁴  CD8⁺ CTLs isolated from mice vaccinated with DA1-3b/CXCL10 showed lower lytic activity against leukemic DA1-3b cells than those from the DA1-3b/IL12 group (Figure 5A). As previously reported, persistent leukemic cells isolated 1, 3, and 12 months of survival after challenge from the spleen of DA1-3b/IL12–vaccinated mice showed resistance to CD8⁺ CTL-specific lysis proportional to the time they spent in the host (Figure 5B).⁴  NK cells isolated from DA1-3b/CXCL10–vaccinated mice showed higher lytic activity against DA1-3b cells than did NK cells from DA1-3b/IL12–vaccinated mice (Figure 5C). EL4 thymoma cells, which have been reported to be relatively resistant to NK-mediated lysis, were killed by NK cells isolated from DA1-3b/CXCL10–vaccinated mice in similar proportion as DA1-3b cells and positive control YAC-1 cells (Figure 5C).²⁴  NK cells from DA1-3b/CXCL10–vaccinated mice killed persistent leukemic cells with the same efficacy whatever time these cells have spent in the host (Figure 5D). NK cell cytotoxicity was partially inhibited by CMA, indicating that the cytolytic effect was at least partially mediated by the perforin/granzyme pathway (Figure 5D). We previously showed that resistance of dormant tumor cells to CD8⁺ CTL lysis was mediated by overexpression of B7-H1 and B7.1.⁴  NK cells from vaccinated mice express PD-1, the only known receptor for B7-H1, CD28, and CTLA-4 (data not shown). To explore a possible role for B7-H1/PD-1 and B7.1/CD28 or CTL antigen 4 (CTLA4) in NK/persistent leukemic cells interaction, we added antibodies against B7-H1 or B7.1, or control isotypes, to the NK cell cytotoxicity assay. No inhibitory or activating effect could be observed (Figure 5D).

To analyze how NK cells could contribute to the antileukemic immunity observed in mice vaccinated with DA1-3b/CXCL10 cells, we analyzed the expression of several molecules that stimulate T cells. No variation of expression of B7.1, B7.2, 4-1BB, 4-1BBL, OX40, OX40L, CD154, Fas, FasL, Trail, PD1, and B7-H2 was seen between NK cells from matrigels from the different groups of vaccinated mice (see “Time course of recruitment of immune cells at the vaccine site”) or from spleen (data not shown). The LY49D-activating NK receptor was expressed at the same level in NK cells from mice vaccinated with DA1-3b/CXCL10, DA1-3b/IL12, or DA1-3b/Zeo cells. LY49A, LY49I, LY49H, and LY49F were not expressed. However, B7-H1 expression was detected in NK cells from spleen, matrigel infiltrates, and blood from mice vaccinated with DA1-3b/CXCL10 (Figure 5E). To verify if this induction of B7-H1 expression was specific to CXCL10, we incubated NK cells from naive mice with recombinant CXCL10 or with the structurally related chemokines CXCL9 and CXCL11. We observed the same effect in all (Figure 6A). CXCL12 (stromal-derived factor 1α [SDF-1α]), IFN-γ, and TNF-α were also effective. However, the 3 CC chemokines tested, CCL11 (eotaxin), CCL5 (RANTES), and CCL3 (macrophage inflammatory protein 1α [MIP-1α]) showed no effect. IL-3, IL-4, IL-6, IL-10, IL-13, IL-15, IL-18, and GM-CSF showed no or little effect. Human NK cells, incubated with recombinant human CXCL10, CXCL9, CXCL10, CXCL11, CXCL12, IFN-γ, or TNF-α, showed increased B7-H1 expression, but IL-2, IL-3, IL-4, GM-CSF, IL-12, CCL3, CCL5, and CCL11 had no effect. The purified CD56^brightCD16^dim NK cell subpopulation expressed higher levels of B7-H1 than the CD56^dimCD16^bright NK cell subset after incubation with CXCL10 (Figure 6B).

To explore the potential effect of B7-H1 overexpression, we cocultivated B7-H1⁺ NK cells isolated from mice vaccinated with DA1-3b/CXCL10 with naive CD4⁺ or CD8⁺ T cells. These B7-H1⁺ NK cells induced proliferation of CD4⁺ and CD8⁺ T cells, and this effect was inhibited by anti-B7-H1 antibody (Figure 7A). B7-H1⁺ NK cells obtained in vitro by incubation with recombinant CXCL10 produced similar results (Figure 7B). The same experiments performed with human NK cells induced proliferation of human CD4⁺ and CD8⁺ T cells (Figure 7C). Mouse B7-H1⁺ NK cells induced IFN-γ and TNF-α production in CD4⁺ and CD8⁺ T cells (Figure 7D-E), and similar results were obtained with human cells. In both humans and mice, B7-H1⁺ NK cells increased IL-10 production slightly, but did not affect IL-4 production (data not shown). Blocking PD-1, currently the only known receptor of B7-H1, did not affect the results, thus indicating that B7-H1 molecules expressed by NK cells stimulate T cells via another unknown receptor (Figure 7A-E).

To evaluate the role of B7-H1 in CXCL10-induced antileukemic immunity, we injected anti-B7-H1–blocking antibodies, or a control isotype, into mice vaccinated with irradiated DA1-3b/CXCL10, DA1-3b/IL12, or control DA1-3b/Zeo cells, and then challenged immunity by injection of 10⁵ live wild-type DA1-3b cells. Blocking B7-H1 partially abolished the prophylactic effect of DA1-3b/CXCL10 vaccine (P = .03, log-rank test), but had no significant effect on DA1-3b/Zeo and DA1-3b/IL12 groups (Figure 7F).

In the DA1-3b mouse model of AML, injection of irradiated leukemic cells transduced by the chemokine CXCL10 led to a prophylactic immunity, with long-term memory, and cured pre-established leukemia in mice. This immunity was far more efficient than immunity induced by vaccines transduced with genes for IL-12, CD154, GM-CSF, or other vaccines we tested previously.^2-4  The CXC10 vaccine cured 80% of animals, whereas the IL-12 vaccine cured only 40%. The effect of the CXCL10 vaccine was dose-dependent because injection of 10⁶ cells led to better results than 10⁵ cells, but injection of 10⁷ cells was deleterious, possibly because a too large amount of chemokine might disrupt the trafficking of immune cells. Other chemokine-transduced vaccines also induce efficient antileukemic immunity. In the A20 mouse lymphoma/leukemia model, CCL3 cells transduced with GM-CSF or IL-2 were effective, but CCL3 cells alone were not.²⁵  In C57BL/6 mice, CXCL12 transduction of C1498 leukemia also induced rejection of leukemic cells and cured pre-established leukemia.²⁶  Thus, chemokines appear as interesting candidate genes for tumor vaccine strategies.

Eradication of dormant tumor cells is another aim of immunotherapy. Long-term persistence of tumor cells in patients in complete remission is common in human cancer and explains the occurrence of relapses years or decades later. Concerning the DA1-3b mouse model of AML, we recently reported that despite an effective antileukemic response induced by CD154- or IL-12–transduced leukemic cells, several hundred leukemic cells were detected in the spleen and the liver of mice 1 year after challenge.⁴  Those cells were still able to kill naive hosts, and they resisted CD8⁺ CTL-mediated lysis. In the present study of mice immunized with DA1-3b/CXCL10 cells, we sought persistent leukemic cells in every organ of mice that survived 1 year after challenge. We never detected any leukemic cells, which indicates the antileukemic immunity both improves survival rates and eradicates dormant tumor cells.

In vivo depletion experiments showed that this antileukemic immunity was partially dependent on CD4⁺ and CD8⁺ cells, but completely dependent on NK cells. In addition, blocking IFN-γ completely inhibited the vaccine effect but blocking TNF-α did not. The immune infiltrate at the vaccine site contained no B cells, and dendritic cells were detected only 2 weeks after vaccination, with a slightly higher proportion in mice vaccinated with DA1-3b/CXCL10 cells. DA1-3b/CXCL10 cells and DA1-3b/IL12 cells attracted T cells equally, but the former were more effective in attracting NK cells. An increased percentage of NK cells was also found in the spleen of mice vaccinated with DA1-3b/CXCL10 cells, even though the percentage of T cells was higher in the DA1-3b/IL12 group. Thus, it appears that DA1-3b/CXCL10 cells induce a massive recruitment of NK cells and modifications of lymphoid cell repartition in the spleen. The chemoattractant effect of CXCL10 on NK cells has been demonstrated in vitro, and CXCL10 expressed by thymic epithelial cells attracts NK-type cells.^11,13  In our experiments, the NK cells that infiltrated the matrigel at the site of vaccination expressed the activating NK receptor LY49D, but not the inhibitory receptor LY49A. Using a rat NK cell line transfected with LY49D or LY49A, Inngjerdingen et al have shown that ligation of LY49D increased migration toward CXCL10, whereas ligation of LY49A decreased migration.¹²  Thus, CXCL10-transduced leukemia cell vaccines preferentially attract to the vaccine site NK cells that respond to this chemokine.

CD8⁺ CTLs isolated from mice vaccinated with DA1-3b/CXCL10 cells showed specific but lower lytic activity against leukemic cells than CTLs from the DA1-3b/IL12-vaccinated mice. In contrast, NK cells isolated from mice vaccinated with DA1-3b/CXCL10 showed higher lytic activity. In our previous experiments, we isolated persistent leukemic cells from the spleen of mice that survived the challenge.⁴  These cells showed resistance to CTL activity, proportional to the time they had spent in the host. However, NK cells isolated from DA1-3b/CXCL10-vaccinated mice killed these leukemic cells with equal efficacy regardless of whether the leukemic cells were isolated after 1, 3, or 12 months in the host. Thus, dormant tumor cells can escape to specific CTL activity but are killed by NK cells activated by CXCL10. We previously showed that dormant tumor cells inhibited CD8⁺ CTL-mediated lysis by overexpressing B7-H1 and B7.1.⁴  B7-H1 (also known as PD-L1) is a B7 family member and ligand for PD-1, a member of the CD28 family.²⁷  B7-H1 can interact with receptors on CTLs and promote CTL death, and thus increased expression of B7-H1 might protect persistent leukemic cells from elimination. In the present experiments, addition of anti-B7-H1– and anti-B7.1–blocking antibodies did not assist NK cells to kill dormant tumor cells overexpressing B7-H1 and B7.1. NK cells express PD-1, the only known receptor of B7-H1, but no differences of expression could be found between the different vaccine groups. Several reports have shown that anti-PD-1–blocking antibodies do not reverse the inhibitory effect of B7-H1, suggesting that there is a second receptor for B7-H1, which might mediate its inhibitory effects on T cells.^18,28  This hypothetical receptor is presumably either not expressed on NK cells or is inactive.

NK cells appear as key effectors of CXCL10-induced antileukemic immunity. However CD4⁺ and CD8⁺ T cells also play a role because depletion of these cells partially inhibited the antileukemic effect. In addition, the efficient memory immunity observed suggests that both innate and adaptive immunity control the leukemic disease. To determine if CXCL10 induced changes in NK cells that then allowed them to stimulate T cells, we analyzed costimulatory molecules in NK cells isolated from matrigel infiltrates, or from spleen, or incubated in vitro with recombinant CXCL10. Surprisingly, we found a strong expression of B7-H1 in NK cells exposed in vitro or in vivo to CXCL10. This B7-H1 expression was also obtained with the structurally related chemokines CXCL11 and CXCL9, which also bind to the CXCR3 receptor. CXCL12, which binds to the CXCR4 receptor, had the same effect. The effect was not restricted to CXC chemokines because IFN-γ and TNF-α also induced B7-H1. However, the tested CC chemokines did not modify the expression of B7-H1. IL-2 had no effect, indicating that nonspecific activation of NK cells does not induce B7-H1 expression. Blocking B7-H1 in vivo partially inhibited the vaccine effect of irradiated DA1-3b/CXCL10 but not of DA1-3b/IL12 or DA1-3b/Zeo controls. Thus, in our model, B7-H1 participates in the antileukemic immunity induced by CXCL10.

The role of B7-H1 in tumor escape from CTL activity has been described in several models, including ours. Recently, the generation of B7-H1–deficient mice has shown that B7-H1 expressed on antigen-presenting cells (APCs), on T cells, or on host tissue regulates T cells negatively.²⁹  However, in nontumoral pathogenic models, B7-H1 apparently activates T cells. In a mouse model of diabetes, B7-H1 promotes autoimmune T-cell responses.³⁰  In a model of colitis, blocking B7-H1 suppressed chronic intestinal inflammation.¹⁷  Thus, B7-H1 appears to have a dual activity. The first report on B7-H1 showed that it costimulated T-cell proliferation and induced IL-10 and IFN-γ production, whereas IL-4 remained unaffected.³¹  In our experiments, either mouse or human B7-H1⁺ NK cells stimulated the proliferation of CD4⁺ and CD8⁺ T cells and secretion of IFN-γ and TNF-α, whereas IL-10 was slightly increased and IL-4 was not modified. These effects appeared specific because blocking B7-H1 in vitro inhibited these stimulatory effects. Blocking PD-1, which is currently the only known receptor of B7-H1, had no effect on the results, thus indicating that B7-H1⁺ NK cells stimulate T cells via a second receptor, which might also mediate activating effects on T cells. Thus, it can be hypothesized that CXCL10 might also enhance T-cell response via B7-H1⁺ NK cells, in addition to enhancing NK cell cytotoxicity against leukemic cells and dormant tumor cells.

Matrigel infiltrates induced by CXCL10 were composed of both T cells and B7-H1⁺ NK cells. The proximity of these cells at the vaccine site might facilitate NK/T-cell interaction. The reason B7-H1–blocking antibodies abolish the vaccine effect of CXCL10 might be that other cells, besides NK cells, also express B7-H1. In our model, B7-H1 is also expressed on leukemic cells, but only after months of persistence.⁴  Thus, the injection of anti-B7-H1 antibodies into mice immediately after the last injection of leukemia vaccine is unlikely to interfere with long-term persistence of leukemic cells. B7-H1 also occurs naturally in a range of tissues, T cells, and APCs, and their function could be affected by anti-B7-H1.^32,33  However, the demonstration of inhibitory function of B7-H1 on T cells and APCs, using B7-H1–deficient mice, argues against a positive role for the B7-H1 molecules expressed on these cells in antitumor immunity.²⁹  Our observation that B7-H1 is expressed on NK cells after exposure to inflammatory chemokines might indicate that B7-H1⁺ NK cells might constitute a new mediator of T-cell activation. We observed in human NK cells the same induction of B7-H1 expression by in vitro exposure to CXC chemokines. More specifically, the 2 major subsets of NK cells in humans were differentially sensitive to CXCL10. The CD56^brightCD16^dim NK cells showed a stronger induction of B7-H1 expression than CD56^dimCD^16bright cells. The majority of human NK cells are phenotypically CD56^dimCD^16bright and are more highly cytotoxic than the other subset, which represent less than 10% of NK cells.³⁴  CD56^brightCD16^dim NK cells express high levels of CXCR3, the receptor of CXCL10³⁵ , and perhaps this is why they produce higher levels of B7-H1. CD56^brightCD16^dim NK cells express homing receptors and have been identified in lymph nodes and may therefore regulate T cells and dendritic cells.³⁶  Our finding that these NK cells may also express B7-H1 is another argument for a regulatory role.

In summary, we identified CXCL10 as a powerful chemokine able to induce a strong antileukemic response against both leukemic disease and long-term persistent leukemic cells via enhanced NK cell cytotoxicity and B7-H1–mediated immunity. NK cells are key effectors of the graft-versus-leukemia effect after allogeneic stem cell transplantation. Our results indicate that CXCL10-activated NK cells might also mediate an efficient autologous immune response in human AML.