Bone Marrow-Derived IFN-Producing Killer Dendritic Cells Account for the Tumoricidal Activity of Unpulsed Dendritic Cells1

The CD11cintB220+NK1.1+CD49+ subset of cells has recently been described as IFN-producing killer dendritic cells (IKDC), which share phenotypic and functional properties of dendritic cells and NK cells. Herein we show that bone marrow-derived murine dendritic cell preparations contain abundant CD11cintB220+NK1.1+CD49+ cells, the removal of which results in loss of tumoricidal activity of unpulsed dendritic cells in vivo. Moreover, following s.c. injection, as few as 5 × 103 highly pure bone marrow-derived IKDC cells are capable of shrinking small contralateral syngeneic tumors in C57BL/6 mice, but not in immunodeficient mice, implying the obligate involvement of host effector cells in tumor rejection. Our data suggest that bone marrow-derived IKDC represent a population that has powerful tumoricidal activity in vivo.

D endritic cells (DC) 3 are pivotal to the induction of innate and adaptive immune responses. In vivo DC are attracted to sites of infection or inflammation where they can take up Ag. Subsequent maturation of DC results in the upregulation of costimulatory molecules and MHC class II and their migration to sentinel lymph nodes where they present Ags to T cells and can stimulate innate responses through secretion of inflammatory cytokines. For example, NK cells and DC can exchange bidrectional activating signals in a positive feedback, referred to as cross-talk (1,2). Both myeloid DC and plasmacytoid DC promote cytoxicity and cytokine production by NK cells, while activated NK cells can reciprocate by providing immunoregulatory helper function to DC, stimulating them to produce proinflammatory cytokines that stimulate cytotoxic lymphocyte (CTL) and Th1 responses.
It has long been recognized that adoptive transfer of DC has the potential to inhibit cancer growth through stimulation of both innate and adaptive immune responses (3)(4)(5)(6). Innate responses induced by DC can result, for example, from the stimulation of tumor-reactive NK cells. Adaptive anticancer immune responses result from the presentation of tumor Ags by DC. In adoptive transfer studies stimulation of adaptive immunity has generally resulted from the priming of DC ex vivo with tumor Ags (7,8), although unprimed DC have shown effective tumoricidal activity in animal models following direct intratumoral injection (9 -11). In tumor models involving injection of DC at distant sites, unpulsed DC usually show low or no antitumor activity compared with Ag-pulsed DC (12)(13)(14)(15)(16). However, some studies have shown tumoricidal activity of unpulsed DC injected at sites distant to a tumor, attributed, at least in part, to stimulation of innate NK responses (4,(17)(18)(19).
Heightened interest of the role of DC in cross-talk of innate and adaptive responses follows the recent description of IFN-␥-producing killer dendritic cells (IKDC), which were reported to show both NK and Ag presenting cell surface phenotype and function (20 -22). IKDC were originally described as having intermediate expression of CD11c coupled with high expression of B220, NK1.1, and CD49b, and absence of Gr-1, to distinguish them from conventional NK cells (NK1.1 ϩ CD49b ϩ CD11c Ϫ B220 Ϫ ) and plasmacytoid DC (pDC; NK1.1 Ϫ CD49b Ϫ Gr-1 ϩ B220 ϩ ) (20,21). However, the distinction between IKDC, DC, and NK cells has been challenged in recent reports that have described the presence of CD11c and B220 on conventional NK cells, and failed to demonstrate presentation of Ag by IKDC, leading to the suggestion that IKDC represent an activated form of NK cells (23)(24)(25).
Herein we provide further evidence for the antitumor properties of cells with the IKDC phenotype. We show that the IKDC can be readily isolated in large numbers from conventional bone marrow DC preparations, but that remarkably low numbers of bone marrow-derived IKDC administered by s.c. injection are required for effective adoptive transfer immunotherapy of distant tumors. Both innate NK response and adaptive T cell responses are observed in mice undergoing tumor rejection following administration of unpulsed DC.
Cytotoxicity, proliferation and IFN-␥ release assays CTL analysis was performed as previously described (27) with minor modifications. Briefly, spleen cells or sorted DC or NK cell subsets were used as effectors to test specific cytolytic activity against tumor targets in a standard 4-h 51 Cr-release assay. In blocking experiments, effector cells were incubated in the presence of specific neutralizing Ab for 2 h (72 h for anti-TRAIL Ab). Spontaneous and maximum release were determined from wells containing either medium alone or lysis buffer (0.5% Triton X-100). The percentage specific cytotoxicity was calculated by the formula: (release in assay Ϫ spontaneous release)/(maximum release Ϫ spontaneous release) ϫ 100. For each E:T ratio, the results are expressed as the mean of triplicates. IFN-␥-releasing cells were quantified by cytokine-specific ELISPOT as previously described (27). Briefly, effector cells were cocultured with target cells for 72 h in the presence of 10 IU/ml of IL-2. The background level was measured in wells containing effector cells in medium only. The results are shown as the mean value obtained for triplicate wells. We assessed T cell proliferation by MTS (3-(4,5dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2Htetrazolium inner salt) assay according to the manufacturer's instructions (Promega).

Statistical analyses
We compared groups using ANOVA followed by multiple comparisons of means. Nonparametric statistical methods were used when the variables studied were not normally distributed. Wilcoxon two-sample rank-sum test was used to compare the values of continuous variables between two groups. Paired comparisons were made using a Wilcoxon paired test. pvalues were determined and considered significant when p Ͻ 0.05.

Unpulsed DC are capable of shrinking small established tumors
In initial experiments designed to assess tumor lysate-pulsed DC vaccination in alveolar rhabdomyosarcoma, we made use of murine 76-9 methylcolanthrene-induced rhabdomyosarcoma cells stably transfected with the PAX3-FKHR oncogene, referred to as 76-9:C23 cells (26). To our surprise, subcutaneously administered unpulsed bone marrow-derived DC were as effective as tumor-lysate-pulsed DC in inhibiting outgrowth of subsequently administered tumor (Fig. 1A). Furthermore, unpulsed DC had a similar effect on the growth of syngeneic B16F10 melanoma and LL/2 lung cancer cells (Fig. 1B). Unpulsed DC could also inhibit growth of small established tumors (Fig. 1C), and splenocytes from DC-vaccinated tumor-bearing mice secreted large amounts of IFN-␥ following addition of tumor cells in vitro (Table II). Furthermore, these splenocytes demonstrated a high degree of cytotoxicty against 76-9:C23 cells, and the cytotoxicity was abrogated by antagonistic anti-CD3, anti-CD8, or anti-NKG2D Abs, suggesting that unpulsed DC induced both CTL-and NK-mediated cytotoxicity (Fig. 1D). This was confirmed by analysis of tumor-infiltrating lymphocytes, which showed a large increase in NK cells and CTL (but not NKT cells) as well as other inflammatory cells in unpulsed DC-treated mice (Table III). In beige mice lacking functional NK cells, 76-9:C23 tumor growth was more rapid, but unpulsed DC prepared from wild-type mice still inhibited tumor growth (Fig. 1E), indicating that 76-9:C23 are NK targets but that host NK cells are not essential for tumor inhibition mediated by unpulsed DC. Furthermore, the role of T cells in beige mice was suggested by the high number of activated tumor-infiltrating CD4 and CD8 T cells compared with untreated mice (Fig. 1F).

Unpulsed DC cause an adaptive T cell immune response, which is dependent on the presence of tumor at a distant site in immunized mice
To confirm that unpulsed DC had induced an adaptive T cellmediated immune response, we investigated whether the activated T cells were specific for the tumor implanted in the immunized mice. Purified FACS-sorted T cells and NK cells, but not NKT cells, from unpulsed DC-treated 76-9:C23 tumor-bearing mice showed markedly enhanced in vitro cytotoxicity against 76-9:C23 targets ( Fig. 2A), but the T cells did not show significant killing of B16 or LL2 targets (Fig. 2B). Moreover, in an MTS assay these purified T cells showed enhanced proliferation response following addition of tumor lysate from 76-9: C23 but not from LL2 or B16 tumors (Fig. 2C). This proliferation correlated with increased IFN-␥ secretion following addition of 76-9:C23 tumor lysate to splenocytes from tumorbearing unpulsed DC-treated mice (Fig. 3A). T cells and NK cells were also purified from spleens of non-tumor-bearing DCtreated mice, and here the T cells showed no killing of 76-9:C23 cells, although NK cells had been similarly activated (Fig. 3B). Therefore, a mixed NK and adaptive T cell-mediated immune response occurs following unpulsed DC treatment, and the adaptive T cell component is dependent on the presence of tumor at a distant s.c. site.  , Fig. 4A). These cells represented between 2 and 10% of the viable cells in DC preparations depending on the method of enrichment (Table IV). As previously described (20,21), FACS-sorted BM-IKDC stained brightly for the NK marker CD49b but low for MHC class II, and did not express Gr-1 (Fig. 4B).
The removal of BM-IKDC from the DC bulk culture caused loss of most of the tumoricidal effect. However, add-back of the FACS-sorted BM-IKDC to the sorted negative fraction of DC (CD11cϩB220 ϩ NK1.1 Ϫ and CD11c ϩ B220 Ϫ NK1.1 Ϫ ) caused a dose-dependent increase of tumoricidal activity that exceeded that observed with unsorted DC (Fig. 4C, left panel). Furthermore, in a separate in vivo experiment, as few as 5 ϫ 10 3 purified CD11c int B220 ϩ NK1.1 ϩ CD49 ϩ cells significantly slowed down the growth of established tumors (Fig. 4C, right  panel). Therefore, BM-IKDC are necessary and sufficient for a strong tumoricidal effect in vivo, and are the principal cells involved in tumor rejection when unpulsed DC are administered at a distant site. a Splenic cells from C57BL/6 naive or mice in protocol 1, 2, 3, or 4 were mixed with 76-9 or 76-9:C23 target cells and incubated in the presence of 10 IU/ml of murine IL-2 for 72 h in an IFN-␥ ELISPOT-release assay.  Fig. 1), in tumor-bearing mice and treated with two DC vaccinations (protocol 3), or in mice challenged with tumor after two DC vaccinations (protocol 2).

In vitro killing capacity of BM-IKDC
We next investigated whether the BM-IKDC in the bulk bone marrow DC culture shared the same in vivo killing activity as the IKDC described previously (20,21). We hypothesized that the activating receptor NKG2D might be essential for killing because it was consistently present in the previously published IKDC, and it was present in the CD11c int B220 ϩ NK1.1 ϩ cells that we purified from bone marrow DC (Fig. 5A). NKG2D ligands RAE1 and H60 were present on 76-9:C23 and LL2 targets but not on B16. Moreover, IFN-␥ release by BM-IKDC following addition of these targets correlated with the presence of the cognate ligands and was blocked by anti-NKG2D blocking Abs (Fig. 5B). Unlike plasmacytoid and conventional DC (CD11c ϩ B220 ϩ NK1.1 Ϫ and CD11c ϩ B220 Ϫ NK1.1 Ϫ negative FIGURE 2. Unpulsed DC cause a specific T cell immune response dependent on the presence of tumor at a distant site. Splenocytes were prepared from unpulsed DC-treated mice 2 wk after 76-9:C23 tumor challenge. CD3 T cells (CD3 ϩ NK1.1 Ϫ ), NK cells (CD3 Ϫ NK1.1 ϩ ), and NKT cells (CD3 ϩ NK1.1 ϩ ) were FACS sorted. In these experiments, DC were matured with prostaglandin E 2 and keyhole limpet hemocyanin. A, Sorted cells were stimulated for 5 days with bone marrow-derived DC pulsed with 76-9:C23 lysate in the presence of 10 U/ml of murine IL-2, and then incubated with 51 Cr-labeled 76-9:C23 target cells. B, CD3 T cells were stimulated as described in A and incubated with 51 Cr-labeled LL/2 or B16 or 76-9:C23 tumor cell lines in a 4-h 51 Cr-release assay at E:T ratio of 50:1. C, Sorted CD3 T cells were stimulated with DC pulsed with tumor lysate prepared from LL/2 or B16 or 76-9 cell lines for 5 days in the presence of 10 U/ml of murine IL-2, and the proliferation of cells was analyzed by MTS assay. Each assay point was done in triplicate, and the result is representative of mean data Ϯ SEM obtained with six mice per group. fraction cells, respectively), purified BM-IKDC were cytotoxic for 76-9:C23, LL2, B16, and YAC1 targets, and cytotoxicty was partially blocked by NKG2D Abs only in the cell lines expressing the cognate ligands. In agreement with previous findings, cytotoxicity against all the targets was inhibited by TRAILblocking Ab (Fig. 5C). Therefore, BM-IKDC present in bulk bone marrow DC cultures share cytotoxicity properties with the previously described IKDC. To determine whether the inherent cytotoxic function of the BM-IKDC could account for tumor regression, we repeated the tumor growth assays using immunodeficient tumor-bearing host mice lacking functional NK and T cells. Here the BM-IKDC had no measurable effect on tumor growth, although parallel tumor growth assays in C57BL/6 mice showed the same growth inhibition as previously observed (Fig. 5D). Therefore, BM-IKDC are dependent on the presence of host effector cells for tumor regression following injection at a distant site.

Discussion
Previous workers have shown unpulsed immature DC to be capable of tumoricidal activity. Often these studies have involved intratumor injection, although distantly injected unpulsed DC cause some tumoricidal activity in some models (17)(18)(19) and no activity in other studies (12)(13)(14)(15)(16), presumably reflecting differences in DC subtype and/or inherent antigenicity of the model. In our C57BL/6 model, bone marrow-derived unpulsed DC were generated by short-term culture from plastic-adherent monocytes, following addition of GM-CSF and IL-4 and activation with LPS ( Fig. 1) or   FIGURE 3. T cell tumoricidal effect following DC vaccination is dependent on presence of tumor in vivo. A, Bone marrow-derived DC and splenocytes were prepared from mice that had been treated with two DC injections and challenged with 76-9:C23 tumor cells, or from naive mice, or from mice challenged only with tumor cells. CD3 T cells and NK cells and NKT cells were FACS sorted. DC were pulsed with 76-9:C23 tumor lysate for 2 h, washed, and mixed with T cells, NK cells, or NKT cells and then incubated in presence of 10 U/ml of murine IL-2 with 76-9:C23 target cells for 72 h in an ELISPOT IFN-␥-release assay. B, CD3 T or NK cells were FACS sorted from non-tumor-bearing mice treated with DC or from naive mice, and stimulated as described in Fig. 2A and incubated with 51 Cr-labeled 76-9 or YAC-1 target cell lines in a 4-h 51 Cr-release assay in the presence or absence of anti-NKG2D-blocking Ab or its isotype control. Each assay point was done in triplicate and the result is representative of mean data obtained with six mice per group. Isotype controls are open histograms. C, Left panel, 76-9:C23 tumor-bearing C56BL/6 mice were left untreated or treated, as specified in protocol 3, with 1 ϫ 10 6 sorted negative fraction (shown in A) or unsorted total DC, or negative fraction supplemented with 5% or 10% of CD11c int B220 ϩ NK1.1 ϩ CD49bϩ. Animals were culled when tumor size reached 1.2-cm diameter. Right panel, Tumor-bearing mice were left untreated or treated as per protocol 3 with 10 5 , 10 4 , or 5 ϫ 10 3 of FACS-sorted CD11c int B220 ϩ NK1.1 ϩ CD49b ϩ cells. One representative out of two experiments including six mice per group is shown. prostaglandin E 2 (Fig. 2). These unpulsed DC had tumoricidal activity against three different target murine cell lines capable of forming tumors in syngeneic C57BL/6 mice. We analyzed the mechanism of tumor rejection in one of the three targets and showed that the intratumoral lymphocyte infiltration comprised a mixture of T cells and NK cells.
Remarkably, removal of CD11c int B220 ϩ NK1.1 ϩ CD49 ϩ IKDC cells from the DC bulk culture caused virtually complete loss of the tumoricidal effect. This was unlikely to result from artifact caused by FACS sorting because add-back of the sorted BM-IKDC fraction was capable of restoring killing function to a level greater than that which had been induced by unsorted DC, and remarkably few (5 ϫ 10 3 ) BM-IKDC were capable of inducing antitumor response. The BM-IKDC population is shown to have TRAIL-and NKG2D-dependent NK-type cytotoxicity in vitro. Ab-blocking experiments on splenocytes from unpulsed DC-treated mice show induction of cytotoxic activity of both CD8-positive T cells and NK1.1-positive cells. Moreover, in the tumor regression we have observed induced by purified BM-IKDC, there is infiltration of tumors by activated T cells and accumulation of Ag-specific activated T cells in spleen, showing in vivo Ag presentation to have occurred. However, it is not clear whether the host or the adoptively transferred DC (or both) are responsible for the presentation. One model is that direct cytotoxicity by IKDC allows for recruitment of host DC, which take up tumor Ag and stimulate host Ag-specific T cells or NK cells. An alternate model is that IKDC perform both killing and presentation of Ags from the killed tumor cells. These different models need to be formally assessed, for example in animals lacking host Ag presentation function. Moreover, very high degrees of purity of FACS-sorted IKDC are needed to exclude Ag presentation by contaminating conventional DC for both in vitro and in vivo assessments of IKDC Ag presentation function. Similarly, experiments using unpulsed DC and BM-IKDC from perforin-deficient hosts would be one way to delineate whether direct target cell killing is an esssntial component of the tumoricidal effect. Our data showing absence of tumoricidal effect of IKDC in hosts lacking T cells and NK cells suggests that the BM-IKDC-mediated direct tumor cytotoxicity observed in vitro is insufficient to inhibit tumor growth in vivo following administration of BM-IKDC at a distant site.
A surprising finding is that as well as regression of established tumors, the unpulsed DC also inhibited outgrowth of subsequently challenged tumor (protocol 2 in Fig. 1) by stimulating the mixed T cell and NK cell response. This suggests that the adoptively transferred DC survive in vivo until the time of the tumor challenge 1 wk later and then are able to stimulate an mixed adaptive and innate immune response on encounter with tumor. This contention will need formal testing in experiments with marked DC, both conventional DC and IKDC.
Zitvogel and coworkers have previously demonstrated the capacity of IKDC derived from spleens of mice treated with IL2 and imatinib to induce the regression of B16 tumors in immunodeficient mice following intratumoral injection (20). In common with our findings, in the absence of host immune effector cells there was no effect on growth of a contralateral tumor. This suggests that in vivo cytotoxic effects of IKDC on tumor growth require direct inoculation into tumor. Alternatively, the lack of BM-IKDC effect on tumor growth in immunodeficient mice in our hands could be due to different migratory or functional NK activity of BM-IKDC compared with splenic IKDC from IL-2-and imatinib-treated mice.
Therefore, we have identified BM-IKDC as the primary cell type involved in the initiation of tumor growth inhibition induced by unpulsed bone marrow-derived DC adoptive transfer in a murine cancer model. However, some controversy has arisen regarding the identity and function of the IKDC population, which shares the same phenotype as the CD11c int B220 ϩ NK1.1 ϩ CD49 ϩ population described in this study (23)(24)(25)28). We have elected to call these cells BM-IKDC but stress that our data, as far as they go, do not exclude the possibility that the BM-IKDC may represent activated and highly potent NK cells capable of inducing regression of established tumor through induction of secondary immune responses, but without having Ag presentation capacity in their own right. It would be important in future experiments to perform a direct comparison in vivo of the tumor inhibitory properties of BM-IKDC and conventional NK cells. The lack of tumor regression in hosts lacking functional B cells, T cells, or NK cells confirms that BM-IKDC, although cytotoxic, rely on cells of the host immune system for the majority of their tumoricidal activity in vivo. Conventional NK cells are also known to activate host APC and so this direct comparison would provide interesting data on relative tumoricidal potency.
The relationship of IKDC to NK cells and DC has been questioned in terms of ontogeny as well as function (23)(24)(25). For example, both NK and IKDC depend primarily on IL-15 and common ␥-chain signaling for development (23,25) and both lack expression of PU.1 (23). Analysis of published gene expression profiles suggests that splenic IKDC share a transcriptional signature with NK cells but have relatively little overlap with DC subsets (29). However, it has recently been shown that IKDC arise from progenitors of the lymphocytic lineage distinct from pDC or NK progenitors (28). It will be important to determine whether BM-IKDC within bulk DC preparations also appear most closely derived from NK cells and whether BM-IKDC from IL-15 knockout mice have similar tumoricidal activity.
Further work is needed to elucidate the mechanism of tumor regression induced by BM-IKDC. We have unequivocally demonstrated cytotoxic activity of these cells in vitro; however, the a DC bone marrow-derived progenitors (10 6 ) were cultured in medium supplemented with rmGM-CSF and rmIL-4 as described in Materials and Methods. Bulk DC cultures were FACS analyzed for CD11c expression at different indicated days. The percentage of IKDC was determined using triple staining for CD11c int , B220, and NK1.1 ϩ cells before or after CD49b beads selection (days 1 and 3). The data are expressed as means Ϯ SD of five independent experiments. ability to present Ag in vitro and in vivo needs formal demonstration, especially as others have failed to demonstrate any Ag presentation function in IKDC derived from splenocytes (23). Moreover, the ability of as few as 5 ϫ 10 3 purified BM-IKDC to inhibit growth of established tumors in immunocompetent mice suggests a remarkable potency in terms of induction of antitumor immunity, as well as obvious potential clinical applications. It will be of great interest to determine whether an analogous population exists in human bone marrow-or monocyte-derived DC populations and to assess their Ag presentation and NK functions.

Disclosures
The authors have no financial conflicts of interest.