Natural killer cells and T cells induce different types ofskin reactions during recall responses to haptens

Authors


Abstract

The role of T cells in contact hypersensitivity (CHS) to haptens has been well described. However, recent reports demonstrated that CHS-like reactions to experimental haptens could be induced in mice deficient in T cells and B cells, as a result of adaptive-like features of NK cells. Here, we compared hapten-specific inflammatory reactions induced by memory T cells or NK cells. Classical CHS protocols were applied to WT or T- and B-cell deficient mice. Adoptive transfers of hapten-specific T cells and NK cells were also performed. Liver NK cells from hapten-primed mice induced specific recall responses to haptens upon transfer in CD3ε-deficient mice, thus confirming the existence of “memory” NK cells in the liver. We investigated the nature of the inflammation generated in these transfer conditions and found that hapten-induced skin inflammation mediated by CD8+ T cells or “memory” NK cells are different. Indeed, ear swelling induced by memory NK cells was transient and not associated with cellular infiltrate and inflammation markers, characteristic for T-cell-mediated responses. Thus, NK cells and T cells mediate distinct forms of skin inflammation. NK cell-mediated pathogenesis does not rely on cellular infiltrate and could be involved in atypical forms of adverse drug reactions.

Introduction

Contact hypersensitivity (CHS) to haptens has long been known as the archetype of adaptive T-cell-mediated immune responses 1, 2. The ability of haptens to induce T-cell priming during epithelial contact, i.e. sensitization, relies on two distinct properties. First, through their proinflammatory properties, haptens activate skin innate immunity and deliver signals that are able to induce the recruitment, migration and maturation of cutaneous DCs. Second, through their binding to amino acid residues, they modify self-proteins and generate new antigenic determinants triggering the priming of hapten-specific T cells in the draining LNs. Then, the challenge of sensitized individuals with the same hapten induces a local recall response caused by the recruitment of specific T cells. We and others have reported that hapten-specific CD8+ T cells are the effector cells of CHS through IFN-γ production and direct cytotoxicity to keratinocytes 1–3. Initial activation of T cells induces the recruitment of a polymorphous cellular infiltrate that is characteristic of the CHS reaction. Most studies on CHS were conducted using experimental haptens, such as 2,4-dinitrofluorobenzene (DNFB), which is endowed with strong irritant properties.

In contrast to these results, more recent reports demonstrated that CHS-like reactions to DNFB could be obtained in mice deficient in T cells and B cells such as RAG-deficient mice 4, 5. One key experiment showed that DNFB-sensitized Rag2−/− mice could develop a CHS-like inflammation upon challenge, mediated by NK cells. These data were followed by other studies reporting adaptive-like properties of NK cells 6–9. “Memory” NK cells were found in the liver of primed animals. Here, we sought to compare the nature of the inflammatory reaction that causes ear swelling in WT or RAG-deficient mice, following the classical CHS protocol using the hapten DNFB. We failed to detect genuine CHS in RAG-deficient mice but found these mice to be very susceptible to hapten-mediated irritation, probably because of the lack of T-cell-mediated regulation 10. Despite the lack of CHS in RAG-deficient mice, we confirmed original findings that liver NK cells from DNFB-primed mice can mediate specific recall responses to DNFB upon transfer into recipient mice. However, the inflammatory reaction that causes ear swelling in these conditions is very different than T-cell-mediated CHS. In particular, the inflammation is transient; there is very little cellular infiltrate, very few cytotoxic cells, no IFN-γ and no TNF-α production. In fact, the inflammation seems to be dominated by few myeloid cells producing IL-6. Therefore, recall responses mediated by NK or T cells induce different pathogeneses.

Results

T- and B-cell-deficient mice do not mount a genuine CHS response

To better understand the nature of the previously described 4, 5 CHS-like skin inflammation mediated by NK cells, in the absence of T cells, we performed CHS experiments in Rag2−/− mice. Mice were sensitized by the application of DNFB on the belly skin at day 0 and were challenged on the ear skin at day 5. As expected, DNFB-sensitized control WT mice mounted an important and significant CHS response upon DNFB challenge (Fig. 1A) characterized by a massive skin infiltration by mononuclear cells and granulocytes (Fig. 1B and D) and by recruitment of CD3+ T cells to the skin (Fig. 1C), whereas naïve mice (challenged-only) did not develop any skin inflammation (Fig. 1 A–C). In contrast, in Rag2−/− mice, no difference was observed in the magnitude of the ear swelling in sensitized/challenged mice compared with challenged-only mice (Fig. 1A). Different lymphocyte-deficient mice, i.e. CD3ε−/− (T-cell-deficient mice) and Rag1−/− (T- and B-cell-deficient mice), were also tested for ear skin inflammation and showed similar results to Rag2−/− mice (Supporting Information Fig. 1). These experiments were repeated in two different animal facilities to exclude a local effect of mouse housing conditions (data not shown). Histological analysis showed a skin edema with very little cellular infiltration (Fig. 1B). Next, we stained ear skin preparations for monocyte (CD11b+Gr1low) and neutrophil (CD11b+Gr1high) markers 11. In WT mice, an increased percentage of monocyte and neutrophils in sensitized and challenged mice compared with challenged-only mice was detected (Fig. 1D). In contrast, in Rag2−/− mice there was no difference between sensitized/challenged and challenged-only mice in terms of percentage (Fig. 1D) and absolute numbers (Supporting Information Fig. 2) of monocytes and neutrophils. Moreover, the total number of monocytes plus neutrophils recovered from the ears of DNFB sensitized/challenged mice was two to three times higher for WT than for Rag2−/− mice.

Figure 1.

Lack of DNFB-specific CHS in Rag2−/− mice. (A) CHS response in C57BL/6 WT mice and Rag2−/− mice. Mice were sensitized on day 0 by painting 0.5% DNFB in acetone:olive oil (4:1) on shaved belly. On day 5, the left ears were challenged with an optimal dose of DNFB (0.15%) and the right ears were painted with vehicle. Ear thickness was measured 24 h later. CHS responses were assessed as described in the Materials and methods. Results are expressed as the mean+SEM, n=5 mice per group. Results are representative of three independent experiments. **p<0.01, unpaired two-tailed Student's t-test. (B) Histological analysis of CHS reaction in C57BL/6 WT (left) and Rag2−/− (right) mice. Four-micrometer-thick sections of ears were recovered 24 h after challenge and stained with H&E. Original magnification, 400×. (C) Flow cytometric analysis of single-cell suspensions recovered from DNFB-challenged ears 24 h after challenge. Dot plot analysis shows CD3 and CD45 staining on gated live cells (with lymphocyte morphology) in the ears of either WT (top) or Rag2−/− (bottom) mice. Results are representative of three independent experiments. (D) Flow cytometric analysis of single-cell suspensions recovered from DNFB challenged ears 24 h after challenge. Dot plot analysis shows Gr1 and CD11b staining on gated live CD45+ cells in the ears of either WT (top) or Rag2−/− (bottom) mice. Results are representative of three independent experiments.

These results indicate that T/B-deficient mice, including Rag1−/− and CD3ε−/−, are prone to nonspecific inflammation and are consistent with the previous report showing that adaptive immune cells temper initial innate responses 10. Since Rag2−/− mice appear more susceptible to the skin irritant effect of DNFB than WT mice, we next performed CHS experiments with lower doses of DNFB that were unable to induce skin inflammation in unsensitized Rag2−/− mice. In this setting, WT mice were able to mount a robust CHS upon challenge (Supporting Information Fig. 3). In contrast, Rag2−/− mice did not develop any skin inflammation. Thus, T- and B-deficient mice are unable to mount CHS responses to DNFB.

DNFB-specific “memory” NK cells mediate hapten-specific ear inflammation upon challenge

Hapten-specific “memory” NK cells have been shown to home to the liver and to mediate specific recall responses to haptens upon adoptive transfer in challenged mice. Although we did not detect CHS in unmanipulated RAG-deficient mice, we reasoned that adoptive transfer conditions may be more sensitive to reveal adaptive-like features of NK cells and study underlying inflammation. We used CD3ε−/− as T-cell-deficient recipient mice for all transfer experiments. WT mice were sensitized by the application of DNFB, oxazolone (OXA), or vehicle on the back skin at day 0, and sacrificed on day 5 to isolate liver NK cells by negative selection. NK cells were transferred into CD3ε−/− syngenic mice and subsequently challenged with a nonirritant dose of DNFB (0.15%) either at day 1 or at day 21 after transfer. We verified that the transferred NK cells survived at least 3 weeks in recipient animals (data not shown). Moreover, we measured the expression of perforin and CXCR6 by transferred NK cells. We found them to be all perforin positive and CXCR6 positive for about 20% of them (Fig. 2A), as reported previously 5. As a control condition, CD3ε−/− recipient mice develop ear swelling after the transfer of CD8+ T cells from previously DNFB or OXA-sensitized mice and challenge with the same hapten (Fig. 2B and C). After adoptive transfer of liver NK cells from DNFB-primed animals, subsequent challenge with DNFB 24 h (Fig. 2B) or 21 days (Fig. 2C) later also elicited typical CHS-mediated ear swelling. This CHS-like reaction was also observed using OXA at priming and challenge but was absent in crossed conditions (DNFB/OXA pr OXA/DNFB, Supporting Information Fig. 4). Moreover, this CHS-like reaction was not due to contaminating CD8+ T cells in the NK cell preparations. Indeed, similar CHS-like reactions were obtained when donor DNFB-sensitized mice were RAG2−/− mice that are devoid of T cells (Supporting Information Fig. 5). These results are in accordance with the previous reports 4, 5 and confirm adaptive-like features of NK cells in this adoptive transfer model. We therefore selected this model for further comparison between T-cell-mediated and NK cell-mediated hapten-specific responses.

Figure 2.

Adoptively transferred NK cells from DNFB-sensitized mice mediate hapten-specific ear inflammation. (A) Dot plot analysis shows CXCR6 (top) and perforin (bottom) expression of transferred cells from naïve (middle) or DNFB-sensitized (right) mice. (B and C) Recipient CD3ε−/− mice were i.v. transferred with 5×105 NK cells or CD8+ T cells isolated from WT DNFB- or OXA-sensitized mice, or injected with PBS (Ø). Transferred mice were challenged with DNFB on the left ears and the right ears were painted with vehicle (B) 24 h or (C) 21 days after transfer. Ear thickness was measured 24 h later. Results are expressed as the mean+SEM in ear swelling, n=3 mice per group. Results are representative of three independent experiments. *p<0.05, ***p<0.001, unpaired two-tailed Student's t-test.

NK cell-mediated hapten-specific responses do not increase with the repetition of antigenic challenges

One of the features of the T-cell-mediated CHS reaction is the gradual increase of the skin edema with the repetition of hapten challenges, presumably because repeated antigen challenges amplify the number of hapten-specific memory CD8+ T cells. We compared this parameter in T–cell- and NK cell-mediated hapten-specific responses. For this, we performed adoptive transfer of CD8+ T cells or NK cells from DNFB-sensitized mice into recipient mice and challenged them with DNFB a first time 1 day later and a second time on day 35. After the first DNFB challenge, NK and CD8+ T-cell-transferred mice showed equivalent ear swelling at 24 h and the inflammation resolved by day 5 (Fig. 3). After the second challenge, CD8+ T-cell-transferred mice developed a skin inflammation that persisted for several days in contrast to NK-transferred mice whose ear inflammation peaked 24 h after challenge and decreased the following days. These results show that hapten-specific recall responses mediated by NK and CD8+ T cells are mechanistically different and that NK cell-mediated responses do not increase with the repetition of antigenic challenges.

Figure 3.

NK cell-mediated hapten-specific responses do not increase with the repetition of antigenic challenges. Recipient CD3ε−/− mice were i.v. transferred with 5×105 NK cells (black line) or CD8+ T cells (gray line) purified from DNFB-sensitized WT mice. Transferred mice were challenged with DNFB on the left ears and the right ears were painted with vehicle 24 h after transfer. Ear thickness was measured daily during the following 4 days. Recipient mice were challenged with DNFB again 5 weeks later, and ear thickness was measured daily during the following 4 days. Results are expressed as the mean+SEM in ear swelling, n=5 mice per group.

No cellular infiltrate during NK cell-mediated hapten-specific recall responses

To further compare T-cell and NK cell-mediated hapten-specific recall responses, we analyzed the cellular infiltrate 24 h after challenge with DNFB in the adoptive transfer model (peak of the recall response). Histopathologic analysis of the ear in CD8+ T-cell-transferred mice revealed characteristic CHS lesions including tissue and blood vessel swelling and infiltration of mononuclear cells and neutrophils (Fig. 4A) compared with PBS-injected control mice. NK cell-transferred mice displayed similar swelling of ear tissue and blood vessels but displayed very little infiltration of mononuclear cells and neutrophils. To track the fate of adoptively transferred CD8+ T or NK cells, we purified them from CD45.1 mice and used CD45.2 mice as the recipient. As shown in Fig. 4B, CD8+CD45.1+CD45.2 cells are recruited to the draining LNs and in the ear upon DNFB challenge in CD8+ T-cell-transferred mice. In contrast, in NK cell-transferred mice, no CD45.1+ were detected in the draining LNs or in the skin, at the lesion site. Thus, hapten-specific NK cells induce ear swelling in response to DNFB challenge differently from CD8+ T cells, as they do not migrate in detectable numbers to the ear and do not induce mixed cellular infiltrate.

Figure 4.

Transferred NK cells do not infiltrate the lesion site 24 h after challenge. CD45.2 CD3ε−/− recipient mice were i.v. injected with PBS, 5×105 CD8+ T cells or NK cells purified from CD45.1 WT congenic mice sensitized with DNFB. (A) Histological analysis of four-micrometer-thick ear sections recovered 24 h after challenge, and stained with H&E. Blood vessels and red blood cells are indicated with red asterisks (panels of PBS transferred mice and NK cell-transferred mice). Mononuclear cells infiltration is indicated with red arrowheads (panel of CD8+ T cells transferred mice). Original magnification, 1000×. (B) Flow cytometric analysis of single-cell suspension recovered from DNFB challenged ears and draining LNs of recipient mice. Dot plot analysis showing CD45.1 and CD45.2 staining on gated live cells. Results are representative of three independent experiments.

No cytotoxicity markers in NK cell-transferred mice at the challenge site

To better understand the mechanisms of NK cell-mediated inflammation, we measured the expression of several transcripts related to inflammation or cellular infiltration in the ear of DNFB-challenged mice, comparing NK cell- and CD8+ T-cell-transferred mice by quantitative PCR (Fig. 5). In the ear of CD8+ T-cell-transferred mice, transcripts for CD8α, NKp46, CD11b, IFN-γ, Granzyme B, TNF-α, FasL, TRAIL, and IL-6 were all upregulated in comparison with control conditions. By contrast, in NK cell-transferred mice, a significant upregulation was only observed for CD11b and IL-6, thus confirming with a more sensitive technique in the absence of an important cellular infiltrate and in particular of cytotoxic NK cells. CD11b likely reflects the presence of a few monocytes/polymorphonuclear cells (PMNs) in the ear that any also produce IL-6. “Naïve” and “memory” NK cells were indeed found to be poor producers of IL-6 (data not shown). Altogether, these results show that NK cell-mediated hapten-specific recall responses and T-cell-mediated CHS to haptens are different immune reactions involving distinct cellular and molecular mechanisms.

Figure 5.

NK cells and CD8+ T cells mediate different types of immune reactions in response to DNFB challenge. Recipient CD3ε−/− mice were i.v. transferred either with PBS (Ø), 5×105 NK cells or CD8+ T cells obtained from DNFB- or OXA-sensitized WT mice. Transferred mice were challenged with DNFB on the left ears and the right ears were painted with vehicle 24 h after transfer. The expression of CD8α, NKp46, CD11b, IFN-γ, GrzB, TNF-α, Fas-L, TRAIL, and IL-6 transcripts in whole ears from vehicle-treated or DNFB-treated recipient mice was measured by quantitative RT-PCR 24 h after challenge. Data are presented as mean+SEM, n=3 mice per group and are representative of three experiments.

Discussion

Here, we show that DNFB-sensitized Rag2−/− mice are not able to mount a specific CHS response upon hapten sensitization and challenge in vivo. These results are in apparent contradiction with the initial observation that T/B-deficient mice could mount CHS responses 4, 12. Although the basis for this discrepancy remains to be determined, our results show that Rag2−/− mice develop significant inflammation upon a single contact with the hapten at a dose (0.15%) not irritant in control WT mice. This is consistent with a previous report showing that adaptive immune cells temper initial innate responses 10. The intrinsic hyperreactivity of RAG deficient mice to haptens may therefore hamper the study of adaptive-like features of NK cells. To circumvent this issue, we isolated hepatic NK cells from DNFB sensitized animals and transferred them to recipient mice that were subsequently challenged with a nonirritant dose of DNFB. In these conditions, and as previously described, “memory” NK cells induced hapten-specific recall responses, as measured by ear swelling. However, our findings show that the inflammation generated in these conditions does not correspond to the characteristic CHS response. Several lines of evidence substantiate this assertion: (i) the inflammation peaked at 24 h and rapidly decreased thereafter, (ii) the reaction was not amplified after repeated hapten challenges, (iii) NK cells were not recruited to the challenge site or to the skin draining LNs, and (iv) the inflammatory profile generated at the challenge site was devoid of cytotoxic markers i.e. Fas-L, TRAIL, and Granzyme B, which are mandatory for the development of hapten-specific skin inflammation 3. Instead, the inflammation induced by NK cells is characterized by edema with little cellular infiltration mainly composed of CD11b+ cells. How can NK cells induce hapten-specific recall response if they do not migrate to the challenge site or at least to the draining LNs? One possibility is that memory NK cells do migrate to the ear but are short lived and die rapidly upon triggering of the response. The production of IL-6 and possibly other cytokines and inflammation mediators by CD11b+ cells activated by NK cells may be sufficient to induce vascular leakage and edema. These results emphasize the limits of ear measurement as a readout of the skin reaction. Another article previously reported that upon transfer into recipient mice, a few “memory” NK cells could migrate to the challenged ear, in a similar CHS setting induced by DNFB. One possible explanation for the discrepancy between these data and our present report could be the nature of the recipient animals. In the study by O'Leary et al. 4, recipient mice were RAG−/−γc−/−, whereas in our study, we used CD3ε−/− mice. It has been demonstrated that NK cells can readily proliferate homeostatically in the former mice 13, but probably not in the latter. Therefore, the number of NK cells derived from the transferred cells is likely higher using RAG−/−γc−/− mice which may explain the reason why they could be detectable at the challenged sites, albeit at very low numbers.

In humans, haptens that cause hypersensitivities include a wide variety of drugs such as antibiotics. Cutaneous drug eruptions are quite common, with a prevalence of 2–3% in hospitalized patients 14. Drug hypersensitivities are classified as allergic or nonallergic when allergic tests are not conclusive. Most drug allergies in children are in fact nonallergic 15, which may reflect the poor efficiency of allergic tests or that allergies occur in a context that is not mimicked by the test. Alternatively, allergic and nonallergic reactions could involve different immune mechanisms based or not on adaptive immunity. In our study, we found that hapten-specific recall responses mediated by NK cells do not induce cellular infiltrates in the skin edema unlike classical T-cell-mediated CHS. Such cellular immune reactions have already been observed in many patients with adverse drug reactions 16. In fact, eight different histological patterns of cutaneous drug eruptions have been observed 16, with a cellular infiltrate composed of various mixtures of lymphocytes, eosinophils, and neutrophils, with even 5% of patients displaying neutrophils only. Our data suggest that the nature of the responding cells i.e. T cells or NK cells may be one of the elements explaining this diversity of reactions. NK cell-mediated hypersensitivities in human are therefore expected to produce poor cellular infiltrates and perhaps score as negative in allergic tests because of low numbers of responding cells and the absence of amplification upon repeated antigenic challenges.

In conclusion, in this present study, we demonstrate that hapten-specific recall responses mediated by NK cells and T cells display different characteristics in terms of cellular infiltrate and inflammatory molecules. Further studies will be required to identify the respective role of T cells and NK cells in the various forms of hapten hypersensitivity in patients.

Materials and methods

Mice and reagents

WT female C57BL/6 mice were from Charles River Laboratories (L'Arbresle, France). CD3ε−/− mice were kindly provided M. Malissen (Centre d'Immunologie, Marseille-Luminy, Marseille, France). Female Rag 2−/−, Rag 1−/−, and CD3ε−/− mice (all on C57BL/6 background) were bred “in-house” (At the “Plateau de Biologie Expérimentale de la Souris,” ENS Lyon, France). Mice were used at 5–10 weeks of age. All procedures were in accordance with the “Comité régional d'éthique pour l'expérimentation animale” guidelines on animal welfare. DNFB and OXA were purchased from Sigma Fine Chemicals. All antibodies (Abs) were from BD Biosciences.

DNFB-induced CHS

CHS responses were assessed using the mouse ear swelling test protocol as described previously 17. Mice were sensitized on day 0 by painting 0.5% DNFB in 25 μL acetone:olive oil (4:1) on shaved belly. On day 5, the left ears were challenged with an optimal dose of DNFB (0.15%) and the right ears were painted with vehicle (5 μL to each side). Ear thickness was measured 24 h later with a spring-loaded micrometer (J15; Blet). CHS responses were assessed as ear swelling calculated as follows: day 6 (treated ear thickness−control ear thickness)−day 5 (treated ear thickness−control ear thickness).

Flow cytometry analysis

LN cells, ear cells, and liver cells were stained using the following Abs: FITC anti-CD45.1, PerCP-Cy5.5 anti-CD45.2, PE-Cy7 anti-CD3, PE anti-Gr1, allophycocyanin-Cy7 anti-CD11b, and Alexa Fluor 647 anti-NKp46 (BD Biosciences); allophycocyanin anti-perforin (eBioscience) and allophycocyanin anti-CXCR6 (R&D Systems). Cell viability was assessed using LIVE/DEAD Fixable Blue Dead Cell Stain (Invitrogen Life Technologies). Samples were analyzed on a FACS LSRII cytometer (BD Biosciences, Le Pont de Claix, France) and using FlowJo 9.3.1 software (Treestar, Ashland, OR, USA).

NK- and T-cell isolation

C57BL/6 WT mice were sensitized on day 0 by painting 0.5% DNFB in 25 μL acetone:olive oil (4:1), or vehicle alone, on shaved belly. On day 5, NK cells were purified from the liver with mechanic digestion followed by passage through a 40-μm filter, and 30 min of centrifugation over a Percoll gradient (40–60%; Amersham Biosciences). Samples were further enriched for NK cells by negative selection with NK cell Isolation Kit (Miltenyi Biotec). The purity of the purified NK cells ranges from 65 to 75%. Residual fraction of CD3+CD8+ cells represents <0.1% of the total population. CD8+ T cells were purified by positive selection (Miltenyi Biotec). Purity was routinely >95%. A total of 5×105 cells (NK or CD8+ T cells)/200 μL of saline solution were i.v. transferred through the orbital vein into naïve syngenic CD3ε−/− mice. Recipient mice were challenged subsequently and ear edema was assessed the following days.

Histology

We fixed ears in a 3% formalin solution for 24 h and embedded them in paraffin using a routine 15 h cycle. The 4-mm sections were cut with a microtome and deposed onto Superfrost Plus slides. Sections were dried overnight at 37°C. The slides were dewaxed in Ottix bath and stained with H&E.

Quantitative RT-PCR

cDNA was obtained from whole ear lysates. Gene expression was assessed by RT-PCR with platinum SYBR Green kit (Invitrogen, Cergy Pontoise, France). Samples were run on an Applied Biosystems 7000 machine. Relative gene expression was determined by normalization to ubiquitin. Primer sequences used in this study are CD8a: ctt gtg cct caa act gca ag / ccg cta aag gca gtt ctc cNKp46: aca cta ctc atc aca gga ggt gtt / gtt gaa agg tca aac tcc caa tCD11b: aag gat gct ggg gag gtc / gtc ata agt gac agt gct ctg gaIFN-g: atc tgg agg aac tgg caa aa / ttc aag act tca aag agt ctg agg taGrB: gct gct cac tgt gaa gga agt / tgg gga atg cat ttt acc atTNF-a: agt gcc tct tct gcc agt tc / tag ctc cca gaa aag caa gcFas-L: acc ggt ggt att ttt cat gg / ttt aag gct ttg gtt ggt gaaTRAIL: gct cct gca ggc tgt gtc / cca att ttg gag taa ttg tcc tgIL-6: tgt tct ctg gga aat cgt gga aat g / gca agt gca tca tcg ttg ttc ata ca

Statistical analysis

Each experiment was conducted using three or five mice per group. Statistical significances were assessed using the unpaired two-tailed Student's t-test. Levels of significance are expressed as p-values (*p<0.05, **p<0.01, and ***p<0.001).

Acknowledgements

A. Hennino received support from the Société Française d'Immunologie. C. Luci received support from the Ligue Nationale Contre le Cancer. The authors thank Dr. Ulrich von Andrian and Dr. Silke Paust for their helpful scientific and technical advice.

Conflict of interest: The authors declare no financial or commercial conflict of interest.

Ancillary