T-bet is a crucial factor for the development of ILC1 and NKp46⁺ ILC3. However, its continued role in the ongoing peripheral maintenance of these cells has not been addressed. To test whether T-bet is also required for the peripheral maintenance of T-bet⁺ ILC, we created a model allowing for the induced depletion of Tbx21 in adult mice, using the Rosa26-Cre-Ert2 system (22, 23). In this model, injections of tamoxifen in the peritoneal cavity are employed temporally to promote the translocation of Cre-Ert2 into the nucleus. Throughout this study ILC were defined as live CD45⁺ Lin^- CD127⁺ leukocytes ( Supplementary Figure 1A ). Colonic lamina propria (cLP) ILC1 were present in untreated Tbx21^fl/fl x Rosa26-Cre-Ert^+/- (Tbx21 ^Δ) and Tbx21^fl/fl x Rosa26-Cre-Ert2^-/- (Tbx21^fl ) mice ( Supplementary Figure 1B ). Tamoxifen was injected into Tbx21 ^Δ and Tbx21^fl mice each day for 5 consecutive days ( Supplementary Figure 1C ). Partial depletion of cLP NKp46⁺ NK1.1⁺ ILC occurred in Tbx21 ^Δ mice as early as 1 week after the first injection and strongly reduced numbers of cLP and small intestinal lamina propria (SI LP) NKp46⁺ NK1.1⁺ ILC could be detected 3 weeks after the first injection ( Figures 1A, B ; Supplementary Figure 1D ). NKp46⁺ NK1.1⁺ ILC are a heterogenous population of all ILC1 with a smaller subpopulation of NK1.1⁺ ILC3 (24). Furthermore, in contrast to colonic and SI lamina propria ILC1, colonic and SI intraepithelial lymphocyte (IEL) ILC1 were less affected by induced depletion of T-bet indicating that this transcription factor is not critical for their maintenance ( Figures 1C, D ). Colonic IEL ILC1 had a reduced cellularity and expression of NKp46 and NK1.1 upon induced depletion of T-bet ( Figures 1D–F ). These observations were not reflected in SI IEL ILC1, as in these cells only NK1.1 expression was partially dependent on T-bet ( Figures 1D–F ). While ILC1 were depleted in the colonic and SI lamina propria of treated Tbx21 ^Δ mice, NK cells defined as CD49b⁺ NK1.1⁺ CD226⁺ CD3^- leukocytes were virtually absent at these sites in Tbx21 ^Δ and Tbx21^fl mice ( Supplementary Figure 2 ).

We sought to confirm that the cLP ILC1 depletion observed also resulted in loss of IFNγ-expressing CD127⁺ ILC, which are predominately ILC1. In order to test this, we infected Tbx21 ^Δ and Tbx21^fl mice with Citrobacter rodentium upon induced deletion of T-bet to establish an inflammatory condition. Induced depletion of T-bet did not cause a profound difference in body weight change over the course of infection ( Supplementary Figure 3 ). However, Tbx21 ^Δ mice had a substantial loss of IFNγ⁺ ILC supporting the notion that induced depletion of T-bet caused a loss of ILC1. In addition, CD127⁺ ILC did display reduced expression of IL-13, but no altered functionality in terms of IL-17A production upon induced depletion of T-bet ( Figures 1G, H ).

Further analysis of ILC3 and ILC2 upon induced depletion of T-bet revealed that in contrast to ILC1, the population sizes of these subsets were not significantly altered ( Figures 2A–F ). For ILC2 analyses, KLRG1 was used as a key marker of intestinal ILC2 in line with recent publications (18, 21, 25). KLRG1 as a marker for intestinal ILC2 has an advantage to GATA3 as intestinal ILC3 have a low expression of GATA3 and the expression of this transcription factor is variable among the ILC2 population (26–28). KLRG1^hi intestinal ILC as gated in this study require GATA3 for post-developmental maintenance supporting the notion these cells are ILC2 (5). The expression of RORγt and CD127 in all cLP NKp46-negative ILC3 subsets was unchanged by induced depletion of T-bet, and the expression of RORγt and CD127 were not altered in cLP NKp46⁺ ILC3 ( Figures 2G, H ). However, cLP NKp46⁺ ILC3 had a lower expression of NKp46, indicating induced depletion of T-bet had some limited effect on these cells ( Figure 2I ). Overall, there appeared to be no effect of induced depletion of T-bet on cLP NKp46^- ILC3 and ILC2 and a limited effect for NKp46⁺ ILC3 in terms of NKp46 expression levels.

To further probe a role for T-bet in cLP ILC1 maintenance, these cells were analysed one week prior to the time point at which ILC1 numbers were profoundly reduced in this model ( Figure 1A ). Two weeks after the first injection of tamoxifen, cLP NKp46⁺ NK1.1⁺ ILC were still detectable in Tbx21 ^Δ mice, but their frequency was already significantly reduced ( Figures 3A, B ). Further analysis revealed that the counts of RORγt^- NKp46⁺ cLP ILC1 per colon did not reach a statistically significant loss in comparison to Tbx21^fl cLP ILC1, indicating a gradual decline of ILC1 upon induced depletion of T-bet ( Figures 3C, D ). Although still present, cLP Tbx21 ^Δ ILC1 had a reduced surface expression of CD122, NKp46 and NK1.1 ( Figure 3E ). Critically, Tbx21 ^Δ ILC1 and NKp46⁺ ILC3 showed a significant reduction in T-bet expression, demonstrating T-bet expression was indeed altered at this time point ( Figure 3F ). cLP NKp46⁺ NK1.1⁺ ILC were present in Stat1^-/- or Stat4^-/- mice, and the percentage of these cells among total CD127⁺ ILC did not alter due to deficiency of either gene ( Figures 3G–I ). STAT4 deficiency also did not alter T-bet expression in cLP NKp46⁺ NK1.1⁺ ILC ( Supplementary Figures 4A, B ). The absence of STAT1 or STAT4 did cause significantly altered IFNγ production in CD127⁺ cLP ILC, indicating that STAT1 and STAT4 are critical factors for ILC1 functionality, but redundant for ILC1 maintenance ( Figure 3J ).

Many immune cells including T cells, ILC, NK cells and myeloid cells can express IFNγ (29). Hence, we sought to establish an overview of IFNγ-producing cLP cells in naïve mice and mice with DSS-elicited colitis (30). Rorc ^GFP mice experienced significant body weight loss upon exposure to 3% DSS in the drinking water compared to controls receiving fresh water (FW) only ( Supplementary Figure 5A ). IFNγ-producing cLP CD45⁺ leukocytes and ILC1 were detected upon DSS treatment (day 8 of our protocol), but also in naïve mice ( Supplementary Figures 5B, C ). We next analysed known IFNγ-producing cLP cell types, including CD103^- T cells and CD103⁺ resident memory T cells, ILC1, NKp46⁺ ILC3 and NK cells and found populations of all of these cell types to express IFNγ in both FW and DSS treated mice ( Supplementary Figures 5D–F ). Additional IFNγ producing cells not further characterised were also observed in both treatment groups ( Supplementary Figures 5D–F ). The frequency of IFNγ-producing ILC1 and NKp46⁺ ILC3 among CD45⁺ leukocytes was greater in control mice in comparison to DSS treated mice ( Supplementary Figures 5E, F ). In contrast, the frequency of resident memory IFNγ⁺ CD103⁺ T cells increased upon DSS treatment ( Supplementary Figures 5E, F ). While ILC1 and ILC3 were present in FW and DSS-treated mice, we only detected a substantial abundance of CD11b⁺ Eomes⁺ NK cells in DSS-treated mice, but not in naïve mice ( Supplementary Figures 5G–I ). The overall frequency of CD103^- and resident memory CD103⁺ T cells was similar between naïve and DSS-treated mice ( Supplementary Figures 5H, I ). These data indicate that ILC1 and NKp46⁺ ILC3 are relevant sources of IFNγ in mice at homeostatic baseline and may therefore become important contributors to the onset of colitis.

To assess the functional effect of T-bet depletion and consequent ILC1 loss, we again utilized the DSS model of colitis. Exposure to 3% DSS in the drinking water caused less severe weight loss in T-bet-deficient (Tbx21^-/- ) mice in comparison to matched wild type C57BL/6 mice ( Supplementary Figure 6A ). BALB/c and BALB/c-background Tbx21^-/- mice were more resistant to DSS-induced colitis as these mice did not experience weight loss in this model ( Supplementary Figure 6B ). To demonstrate a role of T-bet⁺ innate leukocytes in DSS-induced colitis, BALB/c-background Rag2 ^-/- and non-colitic Rag2 ^-/-xTbx21 ^-/- (TRnUC) mice described previously (19) were exposed to 3% or 5% DSS in the drinking water for 5 days ( Supplementary Figures 6C, D ). While 3% DSS was not sufficient to trigger weight loss independent of the T-bet genotype, 5% DSS induced colitis in those mice, however the phenotype was less severe in T-bet deficient mice ( Supplementary Figures 6C, D ) In contrast, BALB/c-background Rag2 ^-/-xγc ^-/- and Rag2 ^-/-xγc ^-/-xTbx21 ^-/- exposed to 5% DSS were resistant to colitis and only experienced weight loss after 7-8 days, which was independent of T-bet expression ( Supplementary Figure 6E ). Rag2 ^-/-xγc ^-/- mice lack mature ILC (31), thus indicating that these cells drive colitis in the absence of an adaptive immune response.

To further investigate whether the induced depletion of T-bet also affected the ability of T cells to drive a type 1 immune response, leukocytes from MLN and spleen were isolated from Tbx21^fl and Tbx21 ^Δ mice upon induced depletion of T-bet in vivo ( Supplementary Figures 7A–C ). Strikingly, the expression of T-bet in either bulk, naïve or effector CD4⁺ and CD8⁺ T cells was unaltered upon induced depletion of T-bet. Exposure to DSS of tamoxifen-pretreated Tbx21^fl and Tbx21 ^Δ mice also did not reveal a significant difference of T-bet expression in either CD4⁺ or CD8⁺ splenic T cells ( Supplementary Figures 7D, E ). Naïve CD4⁺ T cells were isolated from Tbx21^fl and Tbx21 ^Δ MLN and spleen 3 weeks after the first injection of tamoxifen in order to perform an in vitro polarization assay. The ability of MLN and splenic naïve CD4⁺ T cells to express T-bet and IFNγ was significantly, but modestly reduced in Tbx21 ^Δ T cells in comparison to Tbx21^fl T cells ( Supplementary Figures 8A–D ). This suggested that T-bet depletion in naive CD4⁺ T cells or T cell precursors was less effective in this mouse model, and T cells escaping gene deletion may occupy tissue niches in vivo. These data and the lack of alteration in the ILC2 and ILC3 numbers ( Figure 2 ), suggests that T-bet expression is maintained in many cells in Tbx21 ^Δ mice upon temporally induced depletion of the gene encoding this transcription factor. In order to test the relevance of this in an established disease model dependent on intact type 2 immunity, we infected Tbx21 ^Δ and Tbx21^fl with Trichinella spiralis upon induced depletion of T-bet. We have reported previously that mice with a germline depletion of Tbx21 are more resistant to an infection with this parasite due to a promoted type 2 immune response (18, 32). However, in the model of induced depletion of T-bet, Tbx21 ^Δmice were as resistant to the infection as Tbx21^fl mice with non-significant change in worm burden, crypt and villus length, and muscle thickness ( Supplementary Figures 8E, F ).

ILC1 have been associated with detrimental outcome of colitis in mice and humans (13). To analyse the role of ILC1 in colitis in immunocompetent mice more closely, we tested the course of colitis induced by treatment with DSS in Tbx21 ^Δ and Tbx21^fl mice upon induced T-bet depletion. Similar to Tbx21 ^-/- mice, Tbx21 ^Δ mice showed a milder course of colitis with less weight loss and a lower diarrhoea score than Tbx21^fl mice while mice receiving fresh water did not develop disease ( Figures 4A, B ). Significant change in body weight at day 8 and diarrhoea score at days 8 and 9 correlated to a significant reduction of IFNγ production in CD127⁺ cLP ILC at day 7 ( Figures 4C, D ). In contrast to IFNγ, IL-17A and IL-13 production was not different between Tbx21^fl and Tbx21 ^Δ CD127⁺ cLP ILC ( Figures 4C, D ). This indicated again that IFNγ⁺ ILC (being predominately ILC1) were depleted in Tbx21 ^Δ mice, while the frequency of Tbx21 ^Δ IL-17A⁺ and IL-13⁺ ILC was unaltered upon induced depletion of T-bet. IFNγ can be produced by an abundance of other innate and adaptive leukocytes including NK cells, T cells or neutrophils. However, in contrast to ILC, we did not observe an alteration to IFNγ production in non-ILC CD90⁺ or CD90^- leukocytes in the same DSS-treated Tbx21 ^Δ mice at the same time point ( Figure 4E ). A direct comparison of IFNγ expression in lineage marker (Lin)-positive and -negative cLP leukocytes in these mice confirmed that its expression was lower in Lin^- cells from Tbx21 ^Δ mice in comparison to Tbx21^fl mice while IFNγ production was unaltered in Lin⁺ leukocytes at day 7 ( Figure 4F ). Hence, these correlative data indicate that ILC1 play a detrimental role during DSS-induced colitis of Rag-sufficient mice.

Targeting of ILC1 to cure colitis may be an attractive therapeutic route, but it would be of relevance to establish whether depletion of cLP ILC1 is reversible. We noticed that Tbx21 ^Δ mice exposed to DSS for 5 days 21 days after the initial tamoxifen injection still had fewer cLP NKp46⁺ NK1.1⁺ ILC at day 3 after withdrawal of DSS in comparison to Tbx21^fl mice ( Supplementary Figure 9A ). However, at this time point, treatment with DSS appeared to have triggered this cell population to re-emerge modestly, as the cellularity was significantly enhanced in comparison to Tbx21 ^Δ mice receiving pure fresh water. Exposure to DSS also enhanced the frequency of Tbx21^fl cLP NKp46⁺ NK1.1⁺ ILC among CD127⁺ ILC. Interestingly, cLP NKp46⁺ NK1.1⁺ ILC in Tbx21 ^Δ mice were proliferating (or had proliferated recently) at a similar rate to Tbx21^fl cLP NKp46⁺ NK1.1⁺ ILC at day 3 of recovery from DSS- and in FW-treated control mice ( Supplementary Figure 9B ). Hence, it appeared that upon DSS exposure, the ILC1 population may be re-populated from BM-derived cells rather than as a result of enhanced proliferation of the small number of remaining tissue resident ILC1. This view was further corroborated when we observed that NKp46⁺ NK1.1⁺ ILC were still detectable in the bone marrow upon induced depletion of T-bet ( Supplementary Figure 9C ). However, the recovered ILC1 cellularity in DSS-treated mice was very minimal and did not reverse their prior loss to a great extent.

C57BL/6 WT and BALB/c WT, Tbx21^-/- (C57BL/6 and BALB/c) and BALB/c Rag2 ^-/- (all Charles River) mice were sourced commercially. BALB/c Rag2 ^-/-xγc^-/-, Rag2 ^-/-xγc^-/-xTbx21 ^-/- and C57BL/6 T-bet^fl/fl mice were previously generated by our group. C57BL/6 Rosa26-Cre-ERT2 transgenic mice were a kind gift from Dr Thomas Ludwig (Columbia University) and were created using the Cre-ERT2 construct generated by Pierre Chambon (University of Strasbourg). T-bet^fl/fl mice were crossed with Rosa26-Cre-ERT2 mice to generate Tbx2^fl/fl Rosa26-Cre-Ert2^+/- (Tbx21 ^Δ) and Tbx2^fl/fl Rosa26-Cre-Ert2^-/- (Tbx21^fl ) mice. Tbx21 ^Δ and Tbx21^fl mice used for experiments were littermates. A colony of colitis-free TRnUC mice was generated as described previously (19). Stat1 ^-/- (B6.129P2-Stat1tm1Dlv; 65) and Stat4 ^-/- (C57BL/6J-Stat4em3Adiuj/J, purchased from The Jackson Laboratory) were housed under specific pathogen-free conditions at the Institute of Animal Breeding and Genetics in Vienna according to Federation of European Laboratory Animal Science Associations (FELASA) guidelines. C57BL/6N and C57BL/6J mice were purchased from Janvier Labs and used as control mice for Stat1 ^-/- and Stat4 ^-/- mice, respectively. C57BL/6 Rorc ^GFP mice were a kind gift of Dr Gérard Eberl (Institut Pasteur, Paris). Age- and sex-matched mice aged 6-12 weeks were used for experiments.

A tamoxifen stock solution was prepared by dissolving 1g of tamoxifen (free base: MP biomedicals, LLC) in 18 ml 100% ethanol by heating the suspension briefly at 37°C. Prior to injections this solution was diluted at 1:4.56 in sunflower oil and pre-warmed to 37°C. In vivo depletion mice was induced by intraperitoneal injections of tamoxifen on five consecutive days with one injection per day (1 mg per injection). Three weeks after the first injection of tamoxifen tissues were isolated or the mice were infected or treated with DSS as outlined below. In separate experiments tissues were harvested one or two weeks after the first injection of tamoxifen.

cLP and SI LP leukocytes were isolated using a published method (66). Briefly, the epithelium of colons or Peyer’s Patch-free SI was removed by incubation in HBSS lacking Mg²⁺ or Ca²⁺ (Invitrogen) supplemented with EDTA and HEPES. The tissue was further digested in HBSS lacking Mg²⁺ or Ca²⁺ supplemented with 2% foetal calf serum (FCS Gold, PAA Laboratories), 0.5 mg/ml collagenase D, 10 μg/ml DNase I and 1.5 mg/ml dispase II (all Roche). The LP lymphocyte-enriched population was harvested from a 40%-80% Percoll (GE Healthcare) gradient interface. For IEL leukocyte isolations, colons and Peyer’s patch-free SI were cut longitudinally, washed and incubated twice in Ca²⁺ Mg²⁺ free HBSS supplemented with 2 mM DTT and 5 mM EDTA for 20 minutes at 37°C and 100 rpm shaking. After each incubation the tissues were vortexed for 10 seconds and the filtered supernatants were subject to Percoll gradient purification prior to further analyses.

Flow cytometry was performed using a standard protocol. Fc receptor blocking was carried out with anti-CD16/32 specific antibodies. For ILC analyses a lineage cocktail of antibodies specific for CD3, CD45R, CD19, CD11b, TER-119, Gr-1, CD5 and FcϵRI was used. For a complete list of the antibodies utilized see Table 1 . A FoxP3 staining kit (ebioscience) was used for intracellular staining of transcription factors and cytokines. In case of cytokine analysis, cells were pre-stimulated with 100 ng/ml PMA and 2 µM ionomycin in the presence of 6 µM monensin for 3-4 hours at 37°C 5% CO₂ prior to flow cytometry analysis. Samples were acquired using an LSRFortessa™ cell analyser (Becton Dickinson, USA) or a Cytoflex LX™ for the data on Stat1^-/- and Stat4^-/- mice. All the data were analysed using FlowJo software (Tree Star, USA).

Splenocytes and MLN leukocytes were sorted for naïve CD4⁺ cells (live CD4⁺ CD25^- CD44^lo CD62L⁺) using a BD FACSAria cell sorter (BD Biosciences). 1x10⁶ cells were plated per well of a 48 well plate pre-coated with 2 µg/ml anti-CD3 and anti-CD28 antibodies for 4 hours at 37°C 5% CO₂. Plates intended to be used for Th17 polarisation were not pre-coated with anti-CD28 antibodies. The sorted cells were cultured in complete RPMI supplemented with either 20 ng/ml rmIL-2 (Th0), 20 ng/ml rhIL-2, 20 ng/ml rmIL-12 and 5 µg/ml anti-IL4 antibody (Th1), 20 ng/ml rhIL-2, 20 ng/ml rmIL-4, 10 µg/ml and anti-IFNγ antibody (Th2) or 1 ng/ml rhTGF-β, 50 ng/ml rmIL-6, 10 ng/ml rmIL-23, 20 µg/ml anti-IFNγ antibody, 10 µg/ml anti-IL-4 antibody and 5 µg/ml anti-CD28 antibody (Th17). After 48 hours in culture, the cells were re-stimulated with the same cytokine and antibody mix. Another 24 hours later the cells were washed and cultured into uncoated 24 well plates in complete RPMI supplemented with the same cytokine and antibody mix. On day 4 of culture, cells were re-stimulated with 100 ng/ml PMA and 2 µM ionomycin in the presence of 6 µM monensin for 4 hours prior to FACS analysis.

Colitis was induced by adding 3% or 5% DSS (36-50 KDa, MP Biomedicals, Ontario, USA) to the drinking water for 5-10 days as indicated. Mice were sacrificed 7-10 days after the beginning of the treatment with DSS. Weight and clinical abnormalities were monitored on a daily basis.

Infection with C. rodentium (DSB100) was induced via oral gauvage at a dose of 5x10⁸ CFU per mouse. Prior to the infections, the bacteria were grown in LB Broth to 0.7 OD₆₀₀ and washed into sterile cold PBS.

Mice were infected with T. spiralis using a published method (18).

DSS-treated or infected mice were sex- and age-matched and not co-housed with respective control mice.

Tissues were fixed in 10% paraformaldehyde and then embedded in paraffin blocks. Microscopically-visible damage in colitis experiments were evaluated by a pathologist observer (KH), who was blinded to the experimental groups.

Results are expressed as mean ± SEM. Data were analysed using Student’s t-test or Mann-Whitney U test, as appropriate, using GraphPad Prism 5.0 (GraphPad Inc., USA). ns: non-significant; *p < 0.05; **p< 0.01; ***p<0.001; ****p<0.0001.