γδ T cells are a potent source of innate IL-17A and IFN-γ, and they acquire the capacity to produce these cytokines within the thymus. However, the precise stages and required signals that guide this differentiation are unclear. Here we show that the CD24low CD44high effector γδ T cells of the adult thymus are segregated into two lineages by the mutually exclusive expression of CCR6 and NK1.1. Only CCR6+ γδ T cells produced IL-17A, while NK1.1+ γδ T cells were efficient producers of IFN-γ but not of IL-17A. Their effector phenotype correlated with loss of CCR9 expression, particularly among the NK1.1+ γδ T cells. Accordingly, both γδ T-cell subsets were rare in gut-associated lymphoid tissues, but abundant in peripheral lymphoid tissues. There, they provided IL-17A and IFN-γ in response to TCR-specific and TCR-independent stimuli. IL-12 and IL-18 induced IFN-γ and IL-23 induced IL-17A production by NK1.1+ or CCR6+ γδ T cells, respectively. Importantly, we show that CCR6+ γδ T cells are more responsive to TCR stimulation than their NK1.1+ counterparts. In conclusion, our findings support the hypothesis that CCR6+ IL-17A-producing γδ T cells derive from less TCR-dependent selection events than IFN-γ-producing NK1.1+ γδ T cells.
IL-17A and IFN-γ are generally regarded as pro-inflammatory effector cytokines that can be produced by Th cells but also by innate lymphocytes such as NK cells, NKT cells and γδ T cells. While macrophage activation is supposed to be the main role of IFN-γ, the induction of granulopoiesis is ascribed as a key biological function of IL-17A 1. It is currently emerging that γδ T cells are a potent source of IL-17A in the early phases of immune responses (reviewed in 2). γδ T cells constitute a large fraction of all IL-17A-producing cells in healthy mice and humans and are able to secrete IL-17A much more rapidly than CD4+ Th17 cells 3–5. These observations led to the concept that γδ T cells are important players in a transitional response between innate and adaptive immune reactions 6. The production of innate IL-17A by γδ T cells appears to be essential in situations where an effective defence against extra-cellular bacteria or fungi relies on the fast mobilization of neutrophils 4, 6–9. Moreover, IL-17A-producing γδ T cells have been described to play important roles in immunopathologic diseases such as collagen-induced arthritis 10, experimental pulmonary fibrosis 11, and in experimental autoimmune encephalitis 12, 13.
In contrast to CD4+ Th cells that can develop into Th17 cells after encounter of specific cognate TCR Ag 14–16, it is not clear which stimuli induce IL-17A production by γδ T cells in vivo. This essentially results from a lack of information about physiological γδ TCR ligands. However, the current literature suggests that the decision whether a γδ T cell will produce IL-17A is linked to thymic development. Jensen et al. introduced a concept that thymic TCR ligation determines the differentiation of γδ T cells into Ag-experienced IFN-γ-producing or Ag-naive IL-17A-producing γδ T cells 17. In concurrence, further reports showed that IL-17A-producing γδ T cells do not express CD122 and CD27, while IFN-γ-producing γδ T cells stain positive for these markers 18, 19. In addition, SCART2 was identified as a positive determinant of thymic and peripheral γδ T cells with the capacity to secrete IL-17A 20. Collectively, these studies support the hypothesis that certain γδ T-cell subsets acquire effector functions during thymic differentiation, which may be interpreted as a type of TCR-specific selection 21, 22.
Here, we departed from the analysis of thymic CD24lowCD44high γδ T cells, which display an “effector phenotype” and were previously described as “cluster B” γδ T cells 23. We found that cluster B comprised both IFN-γ-producing and IL-17A-producing γδ T cells that were identified by the expression of the mutually exclusive markers NK1.1 and CCR6, respectively. Importantly, we revealed a surprising discrepancy in the stimulation requirements of γδ T cells for IFN-γ versus IL-17A secretion. TCR ligation led to prompt IL-17A production, but much less IFN-γ was produced in response to TCR stimuli. Intense IFN-γ production could be best induced in NK1.1+ γδ T cells through stimulation with IL-12 and IL-18, while IL-23 preferentially stimulated IL-17A production of CCR6+ γδ T cells. Moreover, NK1.1 expression negatively correlated with the expression of CCR9. Expression of CCR9 was more heterogeneous in CCR6+ IL-17A-producing γδ T cells. However, both NK1.1+ and CCR6+ effector γδ T cells subsets were enriched in peripheral tissues but relatively sparse in gut-associated lymphoid tissues. Together, our results suggest that CCR6 is a valuable marker for IL-17A-producing γδ T cells, which may be best denoted as “antigen-naive effector γδ T cells.”
CCR6 and NK1.1 define two γδ T-cell effector subsets in the adult thymus
In contrast to the rigid thymic selection of αβ T cells, TCR selection of γδ T cells is not required beyond a γδ TCR quality control checkpoint for successful pairing and surface expression of the TCR 23, 24. However, 16–20% of the γδ T cells found in an adult murine thymus display an effector phenotype characterized by low expression of CD24 and high expression of CD44 and were termed cluster B 23. Here, we report that cluster B consists to >80% of two populations, which were defined by the mutually exclusive expression of NK1.1 or CCR6 (Fig. 1A). Additional analysis revealed that these two markers segregated cluster B cells further into CD44very high CD24very low CCR6+ and CD44high CD24low NK1.1+ populations (Fig. 1B, left panel). In addition, CCR6+ γδ T cells did not show surface expression of CD122 or CD27, whereas NK1.1+ γδ T cells were CD122+CD27+ (Fig. 1B, right panels). The latter molecules have recently been introduced to classify either IFN-γ-producing effector γδ T cells, which are CD122+CD27+, or IL17A-producing effector γδ T cells, which are CD122–CD27–17–19. Accordingly, we analyzed the potential of NK1.1+ and CCR6+ γδ T cells to secrete these different effector cytokines upon stimulation with PMA/ionomycin (Fig. 1C). NK1.1+ thymic γδ T cells were unable to produce IL-17A (Fig. 1C, upper left), while a clear correlation of CCR6 expression to the production of IL-17A was observed (Fig. 1C, lower left). Only 5–6% of all thymic γδ T cells were CCR6+, but those constituted approximately 75% of all thymic IL-17A-producing γδ T cells. This finding is in compliance with the notion that Th17 γδ T cells are for the most part CCR6+25–27. Conversely, CCR6+ thymic γδ T cells did not produce IFN-γ (Fig. 1C, lower middle panel), while a clear correlation of NK1.1 expression to IFN-γ production was detected (Fig. 1C, upper middle panel). Here, the approximately 10% NK1.1+ of all thymic γδ T cells accounted for 40–50% of all thymic IFN-γ-producing γδ T cells. In conclusion, we demonstrate that cluster B consists mainly of either IFN-γ or IL-17A-producing cells, which can be positively identified by the expression of either NK1.1 or CCR6, respectively.
CCR6+ and NK1.1+ effector γδ T cells are found in the circulation and peripheral lymphoid organs
Based on these observations, we next compared the distribution of CCR6+ and NK1.1+ γδ T cells within peripheral and gut-associated lymphoid tissue (GALT). For both effector populations, we discovered a preferential localization to peripheral lymphoid tissues (Fig. 2A and B). The highest frequency of NK1.1+ γδ T cells among all γδ T cells was detected within the blood (Fig. 2A). In contrast, the maximum proportion of CCR6+ γδ T cells among all γδ T cells was observed in peripheral lymph nodes (pLN) (Fig. 2B). No CCR6+ γδ T cells and very few NK1.1+ γδ T cells were detected in Peyer's patches or within the intestinal intraepithelial lymphocyte (IEL) population (Fig. 2A and B). In the mesenteric LN (mLN) and lamina propria, the frequency of CCR6+ γδ T cells was surprisingly low, i.e. in the range of 1% of all γδ T cells (Fig. 2B).
To determine whether NK1.1+ and CCR6+ γδ T cells were also sparse in GALT due to an intrinsic lack of gut-homing capability, we next examined the expression of the gut-homing-associated CCR9 (Fig. 2C) 28. As expected, most GALT-derived γδ T cells expressed CCR9, whereas the lowest proportion of CCR9+ γδ T cells was found in blood and spleen (Fig. 2C). However, CCR9 expression was not excluded from CCR6+ γδ T cells as demonstrated for LN and thymus γδ T cells in Fig. 2D. In fact, the few CCR6+ γδ T cells found within the LPL compartment were mostly co-expressing CCR9 (data not shown). Intriguingly, CCR6+ but not NK1.1+ γδ T cells showed CCR9 expression in the thymus (Fig. 2D). In line with a recent study that used T22 tetramers to conclude that downregulation of thymic CCR9 levels is a consequence of γδ TCR engagement 29, these findings may suggest that CCR6+ effector γδ T cells may be less “antigen-experienced” than NK1.1+ γδ T cells. Finally, to test whether local inflammation would lead to an accumulation of CCR6+ γδ T cells, we analyzed the draining LN of mice that were immunized subcutaneously with Freund's adjuvant. We observed a transient increase in the proportion of CCR6+ γδ T cells (Fig. 2E), which points to a rapid recruitment of CCR6+ γδ T cells through CCL20 expression in inflamed LN.
Taken together, these results suggest that the observed preferential distribution of CCR6+ and NK1.1+ γδ T cells in peripheral lymphoid tissues may be in part due to a lack of the gut-homing receptor CCR9. NK1.1+ γδ T cells were uniformly CCR9–. However, up to 25% of CCR6+ γδ T cells in pLN were CCR9+ and thus absence of CCR9 cannot account for the observed absence of IL-17A-producing CCR6+ γδ T cells from the intestinal epithelium and Peyer's patches. Via interactions with its ligand CCL20, a contribution of the CCR6 expressed on these IL-17A-producing γδ T cells to their localization in peripheral tissues is nevertheless conceivable.
Different stimulation requirements for IL-17A or IFN-γ production by effector γδ T cells
It is assumed that an immediate production of cytokines in response to innate stimuli or after recognition of self-Ag is a principal role of γδ T cells 30. However, the signals that actually start pro-inflammatory IL-17A or IFN-γ production by effector γδ T cells are not fully understood. One mode to induce cytokine production by innate lymphocytes is by exposure to other cytokines. For γδ T cells it is known that they can produce IL-17 in response to IL-23 7, and IFN-γ in response to cooperative activation with IL-12 and IL-18 31. Here, we employed these protocols to analyze the effector cytokine production of NK1.1+ and CCR6+ γδ T cells. Figure 3A demonstrates that IL-17A was produced by approximately 60% of CCR6+ effector γδ T cells when incubated for 15 h with IL-23, whereas NK1.1+ γδ T cells did not respond to this cytokine. Of note, the addition of recombinant IL-2 partially inhibited IL-23-induced IL-17 production (data not shown), reminiscent of its role in the induction of Th17 cells 32. Likewise, IL-12 and IL-18 induced IFN-γ production in more than 80% of the NK1.1+ effector γδ T cells, while CCR6+ γδ T cells were unresponsive to these stimuli (Fig. 3A). In future studies, it will be interesting to analyze the cytokine receptor pattern of sorted NK1.1+ and CCR6+ γδ T cells. At the same time, we observed that CD4+ Th cells from the same cultures responded only marginally (less than 2% of CD4+ T cells) to IL-12 and IL-18 or IL-23 (data not shown). In conclusion, these findings confirm that IL-12/IL-18 or IL-23 signals could directly activate peripheral γδ T cells from naive mice without prior sensitization. Notably, NK1.1+ and CCR6+ expression identified the most effective IFN-γ- and IL-17A-producing γδ T cells, respectively. Our results are supported by our previous data, which proposed that NK1.1+ γδ T cells with very low levels of γδ TCR on their cell surface (NK-like γδ T cells 33) have common characteristics and functional overlap with NK cells.
Next, we compared the cytokine production of NK1.1– and NK1.1+ γδ T cells in response to TCR-specific and unspecific stimulation. When cultured for 4 h on plate-bound purified anti-γδ TCR mAb (GL3), only a relatively small fraction of the NK1.1– and NK1.1+ γδ T cells readily produced IFN-γ compared with the approximately 55% of the NK1.1+ and 25% of the NK1.1– γδ T cells that responded to unspecific PMA/ionomycin activation (Fig. 3B and C). For IL-17A production, the situation was strikingly different. When stimulated with PMA/ionomycin, the production of IL-17A was confined to the NK1.1– γδ T cells (Fig. 3D). These cells were again mainly CCR6+ (data not shown). Importantly, a 3- to 4-h TCR stimulation was sufficient to induce production of IL-17A in approximately 2% of all γδ T cells from pooled spleen and LN, which represents half of the cells that were maximally induced with PMA/ionomycin stimulation (Fig. 3D and E). These findings further underline the classification that only NK1.1– γδ T cells can make IL-17A whilst NK1.1+ γδ T cells are determined to produce IFN-γ. Importantly, we reveal a differential sensitivity of IL-17A-producing γδ T cells to TCR-stimulation as compared to IFN-γ-producing γδ T cells.
CCR6+ IL-17A-producing effector γδ T cells are frequent in pLN
Although it has been described that γδ T cells are potent IL-17A-producing cells 3, it is not yet generally appreciated that they are the main producers of this cytokine in LN of naive WT mice. IL-17A staining gave the highest fluorescence intensity in γδ T cells (Fig. 4A). This suggests that γδ T cells produce higher amounts of IL-17A than other T cells (Fig. 4A). The absolute numbers and frequencies of γδ T cells among all IL-17A+ lymphocytes were more variable. In C57BL/6 Tcrd-H2BeGFP reporter mice, we observed that after PMA/ionomycin stimulation typically 35–60% of all IL-17A+ cells in WT thymus, spleen and pLN were γδ T cells (Fig. 4C) according to H2BGFP expression. Similar frequencies were found in naive C57BL/6 mice with GL3 anti-γδ TCR staining (data not shown). In thymus, spleen and pLN (inguinal, brachial, axilar and superficial cervical), the large majority of IL-17A+ cells including γδ T cells were CCR6+ (Fig. 4C). Although it was reported that the highest frequencies of IL-17A+ γδ T cells should reside in the mLN 34, we consistently found only low proportions of γδ T cells among the IL-17A+ cells of mLN (Fig. 4B). However, these results show that γδ T cells are abundant in steady-state pLN where they constitute a substantial proportion of all IL-17A+ cells.
In this study, we report that the mutually exclusive expression of CCR6 and NK1.1 corresponds to the production of IL-17A and IFN-γ by effector γδ T cells, respectively. CCR6+ and NK1.1+ γδ T cells were present throughout peripheral lymphoid tissues. The discovery of CCR6+ expression also on IL-17A-producing γδ T cells may be of clinical relevance because CCR6 and its ligand CCL20 have recently been implicated in the migration of pathogenic Th17 to the CNS in experimental autoimmune encephalitis 35. In the thymus, CCR6+ and NK1.1+ γδ T-cell populations together made up >85% of what has been described as the CD24lowCD44high “cluster B” of thymic γδ T cells 23. Although sharing a CD24lowCD44highCD62Llow effector phenotype, the CCR6+ and NK1.1+ γδ T-cell populations disclosed contrasting effector concepts characterized by the production of either the Th17-type cytokine IL-17A or the prototype Th1-cytokine IFN-γ. For NK1.1+ γδ T cells, Pereira and others have pioneered the field and established that in C57BL/6 mice, NK1.1+ γδ T cells have a Vγ1Vδ6.3 biased TCR repertoire and preferentially accumulate in the liver 36–39. However, thymic NK1.1+ γδ T cells also comprise the CD122+ population of Ag-experienced γδ T cells 17. Our findings are consistent with previous classifications because NK1.1+, but not CCR6+ γδ T cells, were expressing CD122 and CD27, both of which have been introduced as negative markers for IL-17A-producing γδ T cells 17, 19.
Regarding a possible thymic development of these populations, the group of Chien has recently reported that thymic encounter of specific TCR ligands induced a downregulation of CCR9 on developing γδ T cells 29. Along this line, we could clearly show that NK1.1+ γδ T cells were already in the thymus all CCR9–. Therefore, our observation that a large fraction of thymic and peripheral IL-17A-producing CCR6+ γδ T cells showed expression of CCR9 supports the idea that CCR6+ γδ T cells were simultaneously showing a CD44very highCD62Llow effector phenotype but were indeed Ag-naive. Furthermore, we demonstrate different stimulation requirements for CCR6+ and NK1.1+ γδ T cells. Given their dichotomy of effector cytokine production after PMA/ionomycin stimulation, it was not surprising that CCR6+ γδ T cells were highly responsive to stimulation with IL-23 7, and NK1.1+ γδ T cells were most strongly activated by a combination of IL-12 and IL-18 31. γδ T cells quickly downregulate their TCR to undetectable levels when incubated with TCR-specific mAb 40. However, the nuclear reporter fluorescence in Tcrd-H2BeGFP mice enabled us to unambiguously detect γδ T cells in TCR-stimulated bulk cultures and to stain for intracellular cytokines at the same time. We could thus show that IFN-γ production by NK1.1+ γδ T cells was much less efficient through TCR-stimulation than via cytokine or PMA/ionomycin incubation. In contrast, CCR6+ γδ T cells readily produced IL-17A after 4 h of cultivation on plate-bound anti-TCR mAb. One may therefore speculate that NK1.1+ or CD122+ γδ T cells are refractory to TCR stimulation because they have previously encountered specific Ag 21. In this context, there are intriguing parallels to the differentiation of αβ NKT cells. Recent studies have identified a population of IL-17A-producing αβ NKT cells that constitutively express IL-23R 41–43. Interestingly, IL-17A production of αβ NKT cells was again confined to the NK1.1– fraction while NK1.1+ αβ NKT cells readily produced IFN-γ. The molecular mechanisms that govern the differentiation of T cells into either IFN-γ-producing and less TCR-dependent or IL-17A-producing and more TCR-responsive innate effectors are currently under active investigation. However, it is likely that immature γδ T cells require specific stimuli to develop into IL-17A secreting effector cells. Such instructions could be encoded in TCR signals of differential strength. Kisielow et al. elegantly showed that SCART2, a marker for thymic IL-17A-producing γδ T cells, is expressed in deviated DN thymocytes of αβ TCR-transgenic mice, but again downregulated in negatively selecting male HY mice 20. Likewise, Kreslavsky et al. have recently shown that a strong TCR-signal can induce the transcription factor PLZF which is required for the developmental progress from immature IL-17A-producing into IFN-γ and IL-4-producing innate effector γδ T cells 44.
During the submission process of this manuscript, two groups have reported important additional information about IL-17A-producing γδ T cells 45, 46. Veldhoen and colleagues found that CCR6+ γδ T cells expressed aryl hydrocarbon receptor, TLR1, TLR2 as well as dectin-1 and showed that engagement of these receptors could directly initiate the secretion of IL-17A and, in the case of aryl hydrocarbon receptor, of IL-22 45. The other report showed that γδ T cells express IL-23R, but concluded in contrast to our results and previous reports that the cooperative stimulation with both IL-1β and IL-23 is necessary for innate IL-17A production from γδ T cells 46. Therefore, it will be important to determine the degree of synergism of the aforementioned stimuli with γδ TCR signaling for activation of IL-17A-producing γδ T cells.
In summary, our findings support the perspective that IL-17A-producing CCR6+ γδ T cells are a distinct population of innate effector lymphocytes, but more immature than NK1.1+ γδ T cells. While NK1.1+ γδ T cells revealed a large functional and phenotypic overlap with CCR9–CD122+ cells, which were previously characterized as Ag-experienced, we showed here that CCR6+ γδ T cells expressed very high levels of CD44, but did not necessarily downregulate CCR9 and were more responsive to TCR stimulation. It remains to conclude that among all γδ T cells, which are collectively designated as innate lymphocytes 47, some γδ T cells, namely the CCR6+ and NK1.1+ populations, are “more innate than others.”
Materials and methods
Six- to ten-week-old mice were used throughout the whole study. C57BL/6-Tcrd-H2BeGFP mice have been described 23. C57BL/6 mice were purchased from Charles-River (Germany). All mice were bred and housed under specific pathogen free conditions in individually ventilated cages at the Hanover Medical School animal facility. All animal experiments were carried out according to institutional guidelines approved by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit.
Flow cytometry and intracellular cytokine staining
Ab directed against TCRβ (clone H57-597), CD4 (clone GK1.5), NK1.1 (clone PK136), B220 (clone RA3-GB2), CD44 (clone IM7), IL-17A (clone ebio17B7) were purchased from ebiosciences. Ab against CD196 (CCR6) (clone 140706) and CD27 (clone LG.3A10) were obtained from BD Pharmingen. Anti-TCRγδ (clone GL3), anti-CD62L (clone MEL-14), anti-FcR Ab (clone 2.4G2), rat anti-mouse CCR9 (clone 7E7-1-1, described in 48) were produced with rat hybridoma cell lines. Anti-TCRγδ clone GL4 was a gift from Leo Lefrancois. The anti-IFN-γ (clone XMG1.2) Ab was purchased from Caltag and anti-CD122 (clone TM-β1) was from Biolegend. Rat serum was obtained from AbD Serotec and mouse serum from Invitrogen. IEL and lamina propria lymphocytes were obtained using Percoll gradients as described 49. Thymocytes were isolated from C57BL/6-Tcrd-H2BeGFP mice by complement lysis (Low-Tox®-M Rabbit Complement, Cedarlane Laboratories) with anti-CD4 IgM (M31), anti-CD8 IgM (RL1.72) Ab and sedimentation on Lympholite M (Cedarlane Laboratories) as described 50. For measurements of intracellular cytokines, T cells were stimulated with 50 ng/mL Phorbol-12-myristate-13-acetate (PMA, Calbiochem), 2 μg/mL ionomycin (Invitrogen) or anti-γδ TCR clone GL3 mAb coated on cell-culture plates for 1 h, followed by a further 3 h incubation in the presence of 1 μg/mL brefeldin A (Sigma). Cells were fixed using a Cytofix/Cytoperm™ Kit (BD Bioscience) as described in the suppliers' manual. Flow cytometry data was acquired by gating on the lymphocyte population and γδ T cells were identified according to their GFPhigh expression as described 23. FACS analysis was performed using a LSRII flow cytometer (BD Biosciences) and data were collected with the FACSDiva software (BD Biosciences). Post-acquisition analyses and compensation were conducted using FlowJo software 8.8.7 (Treestar).
Isolated cells were routinely cultured in RPMI 1640 (Gibco) supplemented with 10% FCS (PAA Laboratories), 1% Glutamine (Gibco) and 1% penicillin/streptomycin (Gibco).
1×106 lymphocytes (in 100 μL) were incubated with 5% mouse serum for 5 min. Rat anti-mouse CCR9 supernatant (100 μL) was added and the mix was incubated at 4°C for 45 min. Cells were washed with FACS buffer (PBS, 3% FCS) and biotin-SP conjugated AffiniPure mouse anti-rat IgG, F(ab′)2 fragment (Jackson Immuno Research) was added for 30 min at 4°C. After another wash step, 5% rat serum was added to the cells for 5 min. Finally, specific staining was revealed with Streptavidin-Phycoerythrin (SAV-PE, Jackson Immuno Research) for 10–15 min at 4°C, and further conjugated mAb were added.
In vitro culture assays
Tissue culture plates were coated with 10 μg/mL anti-TCR γδ (GL3), anti-CD3 (2C11) and anti-TCRβ (H57-597) Ab and incubated overnight at 4°C. Spleen, LN and mLN cells were incubated on Ab coated or uncoated plates, or were solely stimulated with PMA/ionomycin for 1 h before they were treated with brefeldin A for additional 3 h. rmIL-12 (Peprotech), rmIL-18 (MBL International) and rmIL-23 (Peprotech) were applied to the in vitro cultures at 10 ng/mL. Human rIL-2 was obtained through the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH: from Dr. Maurice Gately, Hoffmann – La Roche. IL-2 was applied at 1000 IU/mL. Spleen and LN cells were treated with PMA/ionomycin or anti-CD3 (clone 2C11) and brefeldin A (added after 1 h). Secretion of IL-17A and IFN-γ after stimulation for 4 h was analyzed as described in the Flow cytometry and intracellular cytokine staining section.
For immunization of mice with 200 μg Freund's adjuvant, C57BL/6-Tcrd-H2BeGFP mice were injected into both flanks subcutaneously. The right and left inguinal LN, spleen and mesenteric LN were removed at different time-points post immunization and stained with the following Ab: anti-CD196 (CCR6), anti-NK1.1, anti-TCRβ, anti-B220. Cells that stained positive for TCRβ or B220 were excluded from data analysis.
Data shown are either representatives of multiple experiments or display the combined data of all experiments as indicated in the figure legends. Bar graphs show mean values; error bars show SD.
We thank Bernard Malissen for advice, Andreas Krueger for critically reading the manuscript and Mathias Herberg for animal care. This work was supported by grants from the Deutsche Forschungsgemeinschaft PR727/2-1 and SFB621-A14.
Conflict of interest: The authors declare no financial or commercial conflict of interest.