C57BL/6J and CD45.1⁺ mice were obtained from Jackson Laboratory, and were crossed to generate CD45.1⁺CD45.2⁺ mice. CNS9-deficient mice were generated on C57BL/6J background by using CRISPR-Cas9 system in our previous study (12). RORγt^GFP mice were generated by the laboratory of Prof. Dan R. Littman (4).

All mice were housed in specific pathogen-free (SPF) conditions and in isolated ventilated cages in the animal facility at Tsinghua University, which has been accredited by AAALAC (Association for Assessment and Accreditation of Laboratory Animal Care International). All the animal protocols used in this study has been approved by IACUC (Institutional Animal Care and Use Committee) of Tsinghua University. All mice used in this study were at 6-15 weeks, age and sex matched.

Lamina propria lymphocytes (LPLs) were isolated from small intestine or large intestine. The intestine was collected by removing mesenteric lymph nodes and Peyer’s patches and cut into 2-3 cm pieces longitudinally, then digested at 37°C for 30 mins in RPMI 1640 medium containing 5 mM EDTA, 20 mM HEPES, 1 mM DTT and Penicillin/Streptomycin. The tissues were washed twice using RPMI 1640 medium containing 2mM EDTA, 20mM HEPES and Penicillin/Streptomycin. The remaining tissues were chopped into small pieces and digested in digestion buffer (RPMI 1640 medium with 20 mM HEPES, 0.5 mg/ml collagenase D, 1 mg/ml Dispase and 2 mg/ml DNase I and Penicillin/Streptomycin) at 37°C for 30 min. Then the tissues were meshed through 100 μm cell strainer for cells suspension. LPLs were obtained from the interface of 70% and 40% Percoll after centrifuging at 2500 rpm, 25°C for 30 min.

Spleen, mesenteric and inguinal lymph nodes were taken out, ground and passed through 100 μm cell strainer (spleen needed additional lysing to remove red blood cells).

Mice were fasted for 7-8 hours and then orally gavaged with 2*10⁹ CFU C. rodentium/mouse. The body weight and fecal bacterial load were monitored. If indicated, the mice were sacrificed and intestinal LPLs were analyzed 8 days post-infection.

To measure IFN-γ expression, ILC3s were stimulated at 37°C for 4 hours with 25 ng/ml IL-12, 50 ng/ml IL-15 and 5 ng/ml IL-23 in the presence of GolgiStop™. While for measuring IL-22 and IL-17A, cells were stimulated at 37°C for 4 hours with 5 ng/ml IL-23 in the presence of GolgiStop™. As specifically indicated, IL-22 production in Figure S2B was measured with IL-12 and IL-15 stimulation.

On day 0, bone marrow cells were isolated from CNS9-deficient mice, depleted red blood cells and then cultured in 6-well plates (4×10⁶ cell for each well) in RPMI 1640 medium (20 ng/mL IL-3, 50 ng/mL IL-6 and 50 ng/mL SCF with Penicillin/Streptomycin) at 37°C in a CO₂ incubator. On day 2, polybrene was added into the cell mixture at a final concentration of 8 μg/mL. Bone marrow cells were infected with RVKM control or RVKM-RORγt retrovirus, centrifuged at 1800 rpm for 2 hours at 35°C. This infection was repeated on day 3. On day 4, the virus infected GFP⁺ bone marrow cells were sorted, and then intravenously injected into lethally irradiated (2 rounds of 5Gy irradiation, an interval of 2 hours) CD45.1⁺ recipient mice at 2 x 10⁶/mouse. Seven weeks later, the recipient mice were sacrificed and analyzed.

CNS9-deficient mice were treated with antibiotics (ampicillin: 1 g/L, vancomycin: 500 mg/L, neomycin: 1 mg/mL and metronidazole: 1 mg/mL) starting from 3 to 6 weeks old. The antibiotics were supplied in drinking water and changed every two days to keep fresh.

ILC3 (CD45^mid CD3^- CD90.2^high) cells were sorted from small intestine LPLs using flow cytometry. Single-cell RNA libraries were generated using 10X Genomics chromium according to the manufacturer’s instructions. The quality of libraries was determined by Agilent 2100 bioanalyzer and sequenced with NovaSeq 6000 PE150(Illumina).

To generate single cell feature counts, the clean data of each sample were mapped to mouse reference genome (mm10) by the command “cell ranger count” within cell ranger toolkit (version 6.0.1) provided by 10X genomics. For each sample, cells as outliers in feature count matrix (proportion of mitochondrial genes > 10% and ncount_feature < 300 or ncount_feature > 4700) were removed from further analysis. To remove impure cells, we first filtered the cells by Ptprc ⁺ Cd3e ^- Thy1 ⁺ according to sorting strategy, and then exclude non-ILC3 based on signature gene expression, including T cells (Cd3g, Cd3e), B cells (Cd79a, IgIv1, Jchain), macrophages (Cd68, Apoe, C1qb, Csf1r), endothelial cells (Fabp4, Id3, Lyve1 (14)), ILC1/NK (Ccl5, Klrd1, Klrc2, Eomes) and ILC2s (Klrg1, Il4, Il5, Il13) (15). Finally, we obtained 6860 cells and 4517 cells from WT and CNS9KO samples respectively.

The Seurat R package (version 4.0.3) was conducted for further analysis of feature count matrix. For each sample, about 2,000 genes with the highest variance based on a variance stabilizing transformation data were selected to compute a PCA dimensionality reduction. Integration of two samples was conducted by RunHarmony() function in Seurat R package. The 30 principal components (PCs) of the integrated Seurat object were used for further nonlinear dimensional reduction analysis. The integrated data were clustered using Seurat’s FindClusters() function (resolution parameters was set to 0.8), then visualized by UMAP. The different expression of the selected marker genes in each cluster were visualized by DotPlot() function.

For each sample, a.loom file with counts divided in spliced/unspliced/ambiguous was generated by velocyto.py toolkit (version 0.17). Then, the data were pre-processed by scVelo python toolkit (version 0.2.4). Genes that are detected in less than 20 counts were filtered out and 2000 genes with the highest variability were selected. After normalization and logarithmization, the first and second order moments (means and uncentered variances) were computed among nearest neighbors in PCA space under default parameters. Then, the RNA velocity was estimated by scVelo’s scv.tl.velocity() function, and visualized by scv.pl.velocity_embedding_stream() function.

Bone marrow cells from C57BL/6J, CNS9-deficient and CD45.1⁺CD45.2⁺ mice were isolated and depleted red blood cells. C57BL/6J or CNS9-deficient bone marrow cells were mixed with the ones from CD45.1⁺CD45.2⁺ mice at 1:1 ratio, and 4×10⁶ mixed bone marrow cells were injected intravenously into lethally irradiated (2 rounds of 5Gy irradiation, an interval of 2 hours) CD45.1⁺ recipient mice. The reconstituted mice were then sacrificed and analyzed 7 weeks later.

The antibodies and corresponding dilutions listed below were used for staining: CD45.2 (104, 109824, BioLegend; 104, 56-0454, eBioscience; 1:400), CD45.1 (A20, 558701, BD BioSciences; 1:400), CD3 (17A2, 48-0032-82, eBioscience; 1:400), CD3e (145-2C11, 45-0031-82, eBioscience; 1:200), CD90.2(53-2.1, 25-0902-82/17-0902-83, eBioscience; 1:400), CD4(RM4-5, 562314/557956, BD BioSciences; RM4-5, 25-0042-81, eBioscience; 1:400), CCR6 (29-2L17, 129819, BioLegend; 1:200), NKp46 (29A1.4, 137606, BioLegend; 29A1.4, 25-3351-82, eBioscience; 1:400), RORγt (Q31-378, 562607/562682, BD BioSciences; 1:400), T-bet (4B10, 644814, BioLegend: 1:400), IL-17a (eBio17B7, 48-7177-82, eBioscience; 1:200), IL-22(IL22JOP, 17-7222-82, eBioscience; 1:200), IFN-γ (XMG1.2, 557998, BD BioSciences; 1:200).

Data were analyzed using Graphpad Prism 6 or Graphpad Prism 8 and preformed as mean ± SD. Statistical significance was calculated using unpaired Student’s t test and shown as (*p < 0.05; **p < 0.01; ***p < 0.001, ****p < 0.0001).

We have previously identified CNS9, located at +5,802 to +7,963 bp of the Rorc gene, whose deficiency in mice caused a profound defect in Rorc expression and RORγt-directed IL-17A expression in Th17 cells (12). In contrast, CNS9 deficiency had little effect on development of total RORγt⁺ ILC3s (gated on CD45⁺CD3^-CD90⁺) in small intestine under steady state, as examinbed in both percentages and numbers ( Figure 1A ). In addition, under C. rodentium infection state, CNS9-deficient mice showed comparable degrees of weight loss and bacterial loads at the early stage of infection to control WT mice ( Figures S1A, B ). These data suggest that the general development and function of ILC3s were largely unaffected by CNS9 deficiency.

However, a careful examination revealed that CNS9 deficiency significantly increased expression of both CD4 and NKp46 in ILC3s in the LPLs of small intestine under steady state, resulting in a clear population of CD4⁺NKp46⁺ double positive cells ( Figure 1B ). As introduced previously, CD4 only expressed by part of CCR6⁺ ILC3s but not NKp46⁺ ILC3s, suggesting the CD4⁺NKp46⁺ ILC3 subset in CNS9-deficient mice is a distinct population compared with WT mice. In addition, the expression of CCR6 was found decreased in CNS9-deficient ILC3s ( Figure 1C ), suggesting the composition of ILC3 subsets was significantly altered in CNS9-deficient mice, despite their overall function may not be affected. Consistently, CD4⁺NKp46⁺ ILC3s were also present in the lamina propria of both small and large intestines in CNS9-deficient mice 8 days after C. rodentium infection ( Figures S1C, D ), as well as in other lymphoid tissues including mesenteric lymph nodes (mLN), inguinal lymph nodes (iLN) and spleen under steady status ( Figure S1E ), though less prominent than in gut-associated mucosal tissues.

At cytokine levels, CNS9-deficient ILC3s secreted 50% to 80% less IL-22 and IL-17A, but about 3-fold increase of IFN-γ, compared with WT ILC3s ( Figure 2A ). The CD4⁺NKp46⁺ ILC3s showed similar cytokine producing abilities to CD4^-NKp46⁺ ILC3s in CNS9-deficient mice, with reduced IL-22 and increased IFN-γ but no IL-17A expression, compared with other subsets ( Figure 2B ). Moreover, associated with increased IFN-γ secretion, an obvious population of IFN-γ and IL-22 double positive cells was detected in NKp46⁺ ILC3s in CNS9-deficient mice ( Figure 2C ). These results indicate that CNS9-deficient ILC3s gained increased ILC1-related cytokine secretion potential at the expense of reduced ILC3 effector function.

Notably, there was no defect in IL-22 production in CNS9-deficient ILC3s under stimulation with IL-12, IL-15 and IL-23 condition ( Figure S2A ), which was different from cells stimulated with IL-23 alone ( Figure 2A ). Interestingly, IL-12 and IL-15 treatment could also induce IL-22 production in ILC3s, in the absence of IL-23 ( Figure S2B ), suggesting distinct mechanisms involved in IL-23- and IL-12/15-stimulated conditions.

We further measured IL-22 production in ILC3s after C.rodentium infection, but with only a slightly decreased trend in CNS9-deficient ILC3s compared with WT cells ( Figure S2C ), indicating a compensatory mechanism or context-dependent regulation of IL-22 production in ILC3s by CNS9 as revealed in Figures 2A , S2A . In line with this, CNS9-deficient mice showed comparable weight loss and fecal bacterial loads after C. rodentium infection ( Figures S1A, B ).

Since RORγt and its isoform RORγ can be expressed by a number of cell subsets, including Th17 cells, γδT cells, NKT cells and even non-immune cells, CNS9 deficiency may affect ILC3 development through an indirect manner. To test this possibility, mixed bone marrow chimeric mice were generated by reconstituting lethally irradiated CD45.1⁺ mice with WT CD45.1⁺CD45.2⁺ bone marrow cells and CD45.2⁺ WT or CNS9-deficient bone marrow cells at a 1:1 ratio ( Figure 3A ). 7 weeks after reconstitution, a distinct group of CD4⁺NKp46⁺ ILC3s was developed from CNS9-deficient but not WT bone marrow cells ( Figure 3B ), similar to CNS9 germline knockout mice. In addition, CNS9 deficiency led to significantly increased NKp46⁺ but decreased CCR6⁺ ILC3 populations, similar to unmanipulated mice at steady state, but the overall frequencies of CD4-expressing ILC3s were largely not affected in the chimeric animals ( Figure 3B and Figure 1B ). Similar to those from CNS9-deficient mice, ILC3s developed from CNS9-deficient bone marrow cells in the chimeric mice showed reduced RORγt expression level compared with those from WT bone marrow cells ( Figure 3C ). These results thus demonstrate that CNS9 functions in a cell-intrinsic manner in constraining the development of CD4⁺NKp46⁺ ILC3 subset. However, unlike CNS9-deficient mice, CNS9-deficient bone marrow cells showed a severe defect in development towards ILC3s but not ILC1s, compared with WT bone marrow cells, suggesting their apparent disadvantage in ILC3 development under competitive microenvironments ( Figure 3D ).

ILCs, abundant at mucosal barriers, are regulated by microbiota in development, maintenance and function (15, 16). To examine whether the development of CD4⁺NKp46⁺ ILC3s in CNS9-deficient mice was regulated by intestinal microbiota, we treated 3-week old mice with antibiotics (ampicillin, vancomycin, neomycin and metronidazole) supplied in drinking water for 3 weeks to remove gut microbiota. Compared with water-treated control group, antibiotic treatment did not affect the population of CD4⁺NKp46⁺ ILC3s in CNS9-deficient mice ( Figure 3E ), suggesting their generation in CNS9-deficient mice were independent of gut-associated microbiota.

To further understand the function of CNS9, total ILC3s (gated on CD45^mid CD3^- CD90.2^high) were sorted from the LPLs of small intestine in WT and CNS9-deficient mice and analyzed by single-cell RNA sequencing. In total, 10 distinct clusters were identified after removing contaminated cells ( Figures 4A, B ). A previous study from Ido Amit lab (15) has defined a set of uniquely expressed genes in ILC1, ILC2 and ILC3, based on the RNA-seq data of intestinal ILC1s (CD45⁺ Lin^- CD127⁺ RORγt^– NKp46⁺), ILC2s (CD45⁺ Lin^- CD127⁺ RORγt^– KLRG-1⁺) and ILC3s (CD45⁺ Lin^- CD127⁺ RORγt⁺) in mice, which were used as different ILC signature genes in this study. Clusters 0, 1, 2, 7 and 9 were characterized as NKp46⁺ ILC3s due to high expression of Ncr1 (encoding NKp46) and Tbx21 (encoding T-bet). Cluster 2 also expressed ILC1-related signature genes, such as Ccl5 and Xcl1, and may represent ILC1-like ILC3s. Clusters 4, 5 and 6 had the highest Ccr6 expression, thus representing CCR6⁺ ILC3s ( Figures 4A , S3A ). Consistently, cluster 5 also had the highest expression of Ltb (encoding lymphotoxin beta), a gene involved in lymphoid tissue development. Cluster 4 had high H2-Aa expression, indicating a MHCII-dependent antigen presentation potential. Cluster 6 expressed high levels of Il17a, Il17f and Il22, representing mature or effector ILC3s. Cluster 3 expressed neither CCR6 nor NKp46 and was characterized as IL-22 producing CCR6^-NKp46^- ILC3s. Clusters 0 and 5 expressed high amount of Zfp36, a negative regulator for TNF-α and GM-CSF (17). Cluster 7 highly expressed Tnfrsf9 (encoding CD137 or 4-1BB). Cluster 8 highly expressed Ifit1, a gene involved in induction of type I interferons. Cluster 9 was identified as actively proliferating ILC3s due to high expression of Mki67 and Stmn1 ( Figure S3A ).

In the UMAP plots analysis, CNS9-deficient ILC3s contained clearly more single Cd4 or Ncr1 expressing and co-expressing cell populations, with reduced Ccr6-expressing population ( Figure 4C ). For effector cytokines secreted by ILC3s, CNS9 deficiency resulted in decreased Il22 and Il17a but increased Ifng transcription, consistent with the results obtained by flow cytometry in CNS9-deficient mice ( Figures 4C , 2A ). In addition, CNS9-deficient ILC3s contained cells with increased expression of Klrb1c (encoding NK1.1), Tbx21 (encoding T-bet), Ccl5 and Xcl1, at both population and per cell levels ( Figures 4C , S3B ), indicating CNS9-deficient ILC3s acquired an ILC1-gene expression feature, when compared with WT ILC3s. Gene set variation analysis (GSVA) further confirmed our hypothesis that CNS9 deficiency increased expression of ILC1 signature genes but decreased expression of ILC3 signature genes (15) ( Figure S3C ).

We further analyzed the relationships of cell clusters by RNA velocity. Increased frequency of cluster 2 and decreased frequency of cluster 0 ( Figure 4B ), along with the velocity direction from cluster 0 to cluster 1 and cluster 2 ( Figure 4D ), indicated a strong conversion potential from cluster 0 to cluster 2 in CNS9-deficient NKp46⁺ ILC3s. Similarly, decreased cluster 5 and increased cluster 4 represented a more flexible change from cluster 5 to cluster 4 in CCR6⁺ ILC3 subset after CNS9 deficiency ( Figures 4B, D ). Moreover, cluster 7, ending with all velocity arrows, represented as a fully differentiated cluster in WT ILC3s, but still had a transition potential towards other clusters in CNS9-deficient ILC3s ( Figure 4D ). Taken together, these results suggest that genetic ablation of CNS9 in ILC3s increased their plasticity and conversion potentials towards ILC1-like cells.

Our scRNA-seq data showed that Rorc transcription was decreased in CNS9-deficient ILC3s compared with WT ILC3s, also with a trend following velocity directions (cluster 0 -> cluster 1&2 and cluster 5 -> cluster 4) ( Figure S4A ), suggesting an internal link between Rorc expression levels with cluster variability. CNS9-deficient ILC3s showed increased Cd4 transcription ( Figure S4B ). Of note, cluster 4 had lower Rorc but higher Cd4 expression, compared with cluster 5, in both WT and CNS9-deficient ILC3s ( Figures S4A, B ), suggesting a potential negative regulation of Cd4 by RORγt in ILC3s. Consistently, CNS9-deficient ILC3s had reduced RORγt protein expression at per cell levels, as examined by flow cytometry, compared with WT ILC3s ( Figure 5A ). Specifically, the subset of CD4⁺NKp46⁺ cells in CNS9-deficient ILC3s showed the lowest RORγt expression ( Figure S4C ), indicating CNS9-directed RORγt expression plays a critical role in restricting induction of CD4⁺NKp46⁺ ILC3s, likely in a dosage-dependent manner.

To test the above possibility, we examined ILC3s in the heterozygous RORγt^GFP mice, in which the Gfp gene was inserted into the first exon of Rorc and thus disrupted ~50% of RORγt expression at per cell levels ( Figure 5B ). Similar to CNS9-deficient mice, RORγt^GFP mice contained a distinct CD4⁺NKp46⁺ ILC3 subset ( Figure 5C ), and showed increased NKp46 and IFN-γ expression in total ILC3s, but a reciprocal decrease of CCR6, IL-22 and IL-17A, though CD4 expression was normal, compared with control WT mice ( Figures 5C, D ).

To further confirm the dosage effect of RORγt protein, Rorc was retrovirally overexpressed in CNS9-deficient bone marrow cells (CD45.2⁺) in vitro, then transferred into irradiated CD45.1⁺ mice. 7 weeks after reconstitution, the mice were sacrificed and analyzed. Overall, Rorc overexpression reduced ~70% of CD4⁺NKp46⁺ ILC3s compared with control empty vector ( Figure 6A ), suggesting that decreased RORγt expression indeed contributes to the generation of CD4⁺NKp46⁺ ILC3s in CNS9-deficient mice. For unknown reasons, control retro-virus infection also promoted overall cellular RORγt expression ( Figure 6B ) and decreased the frequency of CD4⁺NKp46⁺ ILC3s compared with mice under steady state.

It has been reported previously that T-bet is critical for generation of NKp46⁺ ILC3s via inducing IFN-γ and NKp46 expression (8–10). We further confirmed our scRNA-seq data with flow cytometry, and found that CNS9-deficient ILC3s, including ILC3 subsets, had increased T-bet protein expression compared with WT cells ( Figure S5 ). Thus, this result suggests the decreased RORγt expression directly promoted T-bet expression, and further resulted in upregulation of IFN-γ and NKp46.

CNS9-deficient ILC3s showed gain of ILC1-like gene expression feature ( Figures 4C and S3B, C ), increased T-bet ( Figure S5 ) and IFN-γ expression ( Figure 2A ), suggesting that RORγt may regulate the balance between ILC3s and ILC1s, via its dosage expression. However, the percentages and cell numbers of ILC1s were only slightly increased, whereas the ratio of ILC3s to ILC1s was not significantly altered in CNS9-deficient mice, compared with WT control mice ( Figure S6 ). These suggest that CNS9-directed Rorc transcription is not involved in transition between ILC3s and ILC1s, and CD4⁺NKp46⁺ ILC3s represent a transient state fixed by reduced but not abolished RORγt expression in CNS9-deficient mice.