The transcription factor Aiolos restrains the activation of intestinal intraepithelial lymphocytes

Abstract

Intestinal intraepithelial lymphocytes (IELs) exhibit prompt innate-like responses to microenvironmental cues and require strict control of effector functions. Here, we showed that Aiolos, an Ikaros zinc finger family member encoded by Ikzf3, acted as a regulator of IEL activation. Ikzf3^−/− CD8αα⁺ IELs had elevated expression of NK receptors, cytotoxic enzymes, cytokines and chemokines. Single-cell RNA-seq of Ikzf3^−/− and Ikzf3^+/+ IELs showed an amplified effector machinery in Ikzf3^−/− CD8αα⁺ IELs compared to Ikzf3^+/+ counterparts. Ikzf3^−/− CD8αα⁺ IELs had increased responsiveness to IL-15, which explained a substantial part, but not all the observed phenotypes. Aiolos binding sites were close to those for the transcription factors STAT5 and RUNX, which promote IL-15 signaling and cytolytic programs, and Ikzf3 deficiency partially increased chromatin accessibility and histone acetylation in these regions. Ikzf3 deficiency in mice enhanced susceptibility to colitis, underscoring the relevance of Aiolos in regulating the effector function in IELs.

The intestinal epithelium is home to a large population of T cells, known as intraepithelial lymphocytes (IELs), as well as fewer intraepithelial innate lymphoid cells (ILCs)^1–3. IELs encompass both conventional and unconventional T cell populations (here after, referred as cIELs and uIELs respectively). cIELs include TCRαβ⁺ T cells that express CD4 (CD4⁺ cIELs) or CD8αβ heterodimers (CD8αβ⁺ cIELs), which home to the gut epithelium in response to antigenic activation in the periphery^{4, 5}. A fraction of CD4⁺ cIELs acquire cell surface expression of CD8αα and a cytolytic program, and are referred to as double positive T cells (DP IELs). uIELs encompass γδ T cells (γδ uIELs) and TCRαβ⁺ cells that express CD8αα homodimers, but lack CD4 and CD8αβ and are referred to as double negative IELs (DN uIELs)^{2, 4, 5}. Intraepithelial ILCs mainly include type 1 ILC (ILC1), which produce interferon-γ (IFN-γ) and release secretory lysosomes containing perforin and granzymes, particularly granzyme C⁶. ILC1s include a major subset of NKp46⁺NK1.1⁺ cells and a smaller subset of NKp46⁻NK1.1^lo/− cells, which express Ly49E and are highly dependent on IL-15⁷. IELs and ILC1s maintain tolerance to nutritional and commensal antigens, protect the host from infections through release of cytolytic molecules, cytokines and chemokines, and facilitate repair of damaged intestinal epithelium^{1, 4}.

uIELs recognize ligands through TCRs⁵, but also exhibit innate-like features, and display rapid and stereotypical responses to tissue alarmins and stress-induced ligands of NK receptors independent of antigen recognition⁸. This binary status of IELs, often defined as “resting yet poised to activation”^{4, 9}, is essential to maintain homeostasis and host defense without causing unwarranted inflammation and intestinal diseases. In fact, the dysregulated activation of IELs contributes to the pathogenesis of celiac disease and inflammatory bowel diseases^10–12. While mechanisms that induce activation of IELs have been extensively studied, those modulating IELs remain incompletely undefined.

The Ikaros zinc-finger (IKZF) family encompasses five transcription factors with roles in development, maturation and functions of immune cells¹³. These transcription factors form homodimerization or heterodimerization within the IKZF family members¹³. IKZF transcription factors interact with the nucleosome remodelling and deacetylase complex (NuRD), which deacetylates histones and restrains gene expression through histone deacetylase complex^{14, 15}. The IKZF family member Ikaros, encoded by Ikzf1, controls early lymphoid cell development. Ikaros-deficient mice have marked reduction of T cells, B cells and NK cells numbers and succumb to lymphoproliferative disorders^{16, 17}. Aiolos, encoded by Ikzf3, forms homodimers or heterodimers with Ikaros. Aiolos deficiency impacts NK cell maturation, B cell activation, T_H17 cell development, cytokine production by type 2 innate lymphoid cells (ILC2s) and eosinophil migration^18–23. In humans, Aiolos represses the production of IL-22, the signature cytokine of type 3 ILCs (ILC3s), and facilitates ILC3 differentiation into intestinal intraepithelial ILC1s^{24, 25}.

Here we found that Aiolos controled the expression of genes encoding NK1.1 and other NK receptors, cytotoxic mediators and IFN-γ in uIELs, in part by attenuating IL-15 signaling. Epigenetic analyses showed that Aiolos bound to regulatory regions of these genes close to STAT5 and RUNX binding sites, modifying chromatin accessibility and histone acetylation. Thus, Aiolos controlled intestinal inflammation by modulating the effector programs of uIELs through epigenetic modifications likely coordinated with STAT5 and RUNX family members.

Results

Aiolos deficiency enhances uIEL activation program

We initially compared the abundance and phenotype of intestinal intraepithelial ILC1 (live CD45⁺CD3⁻NK1.1⁺NKp46⁺ cells) in Ikzf3^−/− and Ikzf3^+/+ mice. The overall percentage of ILC1 was similar in the two groups (Fig. 1a). A more detailed analysis indicated a selective expansion of the NK1.1⁺NKp46⁻ ILC1 subset in Ikzf3^−/− compared to Ikzf3^+/+ mice, whereas the NK1.1⁺NKp46⁺ ILC1 subset was unchanged (Fig. 1a). Because the NK1.1⁺ NKp46⁻ ILC1 subset contained very few cells, it was no longer considered for subsequent analyses. The total cell count of cIELs and uIELs was similar in Ikzf3^+/+ and Ikzf3^−/− mice (Extended Data Fig. 1), but Ikzf3^−/− uIELs acquired higher expression of the NK receptor NK1.1 compared to Ikzf3^+/+ IELs (Fig. 1b). Phenotyping of IEL subsets in Ikzf3^−/− mice indicated that NK1.1 was mainly induced in CD45⁺CD3⁺TCRγδ⁺ cells (hereafter γδ uIELs unless other wise specified) and CD45⁺CD3⁺TCRγδ⁻TCRαβ⁺CD4⁻CD8β⁻CD8α⁺ cells (hereafter DN uIELs), as well as DP IELs (Fig. 1c). Intracellular staining in Ikzf3^+/+ mice indicated that uIELs and DP IELs had the highest expression of Aiolos (Fig. 1d).

Fig. 1. uIEL express Aiolos and Aiolos controls their activation.

a, Representative flow cytometry plots showing expression of NK1.1 and NKp46 along with quantification of live CD45⁺CD3⁻NK1.1⁺NKp46⁺ and live CD45⁺CD3⁻NK1.1⁺NKp46⁻ cells isolated from small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice (n=6). b, Representative flow cytometry plots showing expression of NK1.1 and CD3 on CD45⁺ cells isolated from small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice. c, Representative flow cytometry plots and quantification of NK1.1 expression on live CD45⁺CD3⁺TCRγδ⁺ cells (γδ uIELs), live CD45⁺CD3⁺TCRγδ⁻TCRαβ⁺CD4⁻CD8β⁻CD8α⁺ cells (DN uIELs), live CD45⁺CD3⁺TCRγδ⁻TCRαβ⁺CD4⁺CD8β⁻CD8α⁺ cells (DP uIELs), live CD45⁺CD3⁺TCRγδ⁻TCRαβ⁺CD4⁺CD8β⁻CD8α⁻ cells (CD4⁺ cIELs), live CD45⁺CD3⁺TCRγδ⁻TCRαβ⁺CD4⁻CD8β⁻CD8α⁺ cells (CD8αβ⁺ cIELs) from small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice (n=4). d, Expression (upper panel) and quantification (MFI, lower panel) of Aiolos in the indicated cell types as in (c) (n=4). a-d, Data are representative of two independent experiments. All samples are biological replicates. Data represent mean ± SD. DN = double negative, DP = double positive, FMO = Fluorescence Minus One. MFI=Mean Fluorescence Intensity. P values were determined by two-tailed, unpaired Student’s t-tests in a and c, and by one-way ANOVA with Tukey’s multiple comparisons test in d.

We used bulk RNA-seq to investigate broader changes in the transcriptional program of Ikzf3^−/− γδ uIELs and DN uIELs. DP IELs were not included because of their very low numbers. Ikzf3^−/− γδ uIELs and Ikzf3^−/− DN uIELs exhibited broad reprogramming, and showed upregulation of NK receptors (Klra1, Klra3, Klra9, Klra13-ps, Klrb1b, Klrb1c, Klrc1, Klrk1), granzymes (Gzmc, Gzmk, Gzmf), various cytokines and chemokines (Ifng, Ccl4, Spp1) and the antiapoptotic molecule Bcl2 compared to Ikzf3^+/+ counterparts (Fig. 2a). Ikzf3^−/− γδ uIEL and DN uIEL also exhibited increased expression of checkpoint inhibitors and intracellular modulators that are induced by activation, such as Lair, Lag3 and Socs3 compared to Ikzf3^+/+ counterparts (Fig. 2a). We concluded that Aiolos repressed the activation program of uIELs.

Fig. 2. Aiolos restrains the activation of uIELs in a cell-intrinsic manner.

a, Volcano plots comparing gene expression in γδ uIELs and DN uIELs isolated from small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice. b. Representative flow cytometry plots and quantification of NK1.1 (n=10), NKG2D (n=4), NKG2A (n=4), Ly49A (n=4), Ly49H (n=3), granzyme C (n=10) and IFN-γ (n=10) in γδ uIELs and DN uIELs isolated from small intestine epithelium of sub-lethally irradiated Rag1^−/− intravenously injected with an equal mix of bone marrow from Ikzf3^+/+ (CD45.1⁺) and Ikzf3^−/− (CD45.2⁺) mice and analyzed 6 weeks after reconstitution. Data are representative of two independent experiments. All samples are biological replicates. P values were deetrmined by two-tailed, unpaired Student’s t-test. In RNA-seq, FDR was determined by two-sided Wald test (DESeq2). Data represent mean ± SD. DN= Double Negative. FDR= False Discovery Rate.

Aiolos modulation of uIELs is cell-intrinsic

To investigate whether the effect of Aiolos deficiency in uIELs was cell-intrinsic, we co-injected bone marrow from congenically marked Ikzf3^+/+ mice (CD45.1⁺) and Ikzf3^−/− mice (CD45.2⁺) into sub-lethally irradiated Rag1^−/− mice. At week 6 post-transplant, expression of the NK receptors NK1.1, NKG2A and NKG2D, as well as the intracellular content of granzyme C and IFN-γ were higher in Ikzf3^−/− uIELs than Ikzf3^+/+ uIEL (Fig. 2b). Ly49A expression was higher in Ikzf3^−/− DN uIELs compared with Ikzf3^+/+ uIELs, but not expressed in Ikzf3^−/− and Ikzf3^+/+ γδ uIELs (Fig. 2b). Other Ly49 receptors, such as Ly49D and Ly49H, were absent in both Ikzf3^+/+ and Ikzf3^−/− uIELs (Fig. 2b). Thus, Aiolos restrained the expression of NK receptors, cytotoxic enzymes and IFN-γ in γδ uIELs and DN uIELs in a cell-intrinsic manner. Ikzf3^−/− uIELs also had enhanced expression of NKG2D (encoded by Klrk1) compared to Ikzf3^+/+ uIELs (Fig. 2b, Extended Data Fig. 2a). To test whether enhanced NKG2D expression mediated uIEL cytotoxicity, we sorted γδ and DN uIELs from Ikzf3^+/+ and Ikzf3^−/− mice and incubated them with RMA-S cells or RMA-S cells overexpressing Rae1γ, a ligand for NKG2D. Ikzf3^+/+ or Ikzf3^−/− uIELs did not lyze the RMA-S cells in a NKG2D-dependent manner (Extended Data Fig. 2b). We concluded that enhanced NKG2D expression in Ikzf3^−/− uIELs was insufficient to induce NKG2D-mediated cytotoxicity, consistent with reports that NKG2D only mediates costimulation in skin γδ T cells²⁶.

Aiolos reprograms the Tcf7⁻Il2b^hi subset of CD8αα⁺ uIELs

We sought to obtain a high-resolution view of the impact of Aiolos deficiency on individual IEL subsets by single cell RNA-sequencing (scRNA-seq). We sorted live CD45⁺ cells from the small intestine epithelium in Ikzf3^+/+ and Ikzf3^−/− mice and performed scRNA-seq. We profiled a total of 49,205 quality-controlled cells that were visualized using the uniform manifold approximation and projection (UMAP) method. Based on differentially expressed genes we annotated clusters corresponding to CD8αα⁺ uIELs, CD4⁺ cIELs, CD8αβ⁺ cIELs, NK-ILC1s, plasma cells, B cells and plasmacytoid dendritic cells (Extended Data Fig. 3a, b). Trac and Trbc1, encoding TCRα constant gene segment and TCR beta constant 1 respectively, and Tcrg-V1 and Tcrg-V7, encoding the most prevalent TCRγ chains among γδ uIELs were expressed in the same CD8αα⁺ uIEL clusters (Extended Data Fig. 3a, c), indicating a substantial overlap of γδ uIEL and DN uIEL programs, despite the expression of different TCRs.

After CD8αα⁺ uIELs computational reclustering, we could distinguish three major clusters: Tcf7⁺ uIELs, Tcf7⁻Il2rb^hi uIELs and Tcf7⁻Il2rb^lo uIELs (Fig. 3a). The profile of Tcf7⁺ uIELs was biased towards the expression of Id3, Ltb, Kit²⁷, Cd160²⁸ and Xcl1²⁹ (Fig. 3b, Extended Data Fig. 4a), indicating a functional program typical of stem-like or memory CD8⁺ T cells³⁰. Conversely, Tcf7⁻ uIELs expressed genes indicative of effector function, such as Prdm1, Gzma, Gzmb and Ifng, as well as Klrk1 and Hcst (encoding NKG2D and its adaptor DAP10 respectively³¹) (Fig. 3b). Tcf7⁻ uIELs were distinguished into Il2rb^hi and Il2rb^lo uIELs based on expression of Il2rb, which encodes CD122, the β-subunit of IL-2 receptor shared by the receptor for IL-15 and IL-2 (Fig. 3b, Extended Data Fig. 4a). Il2ra was not expressed (Extended Data Fig. 4b). Tcf7⁻Il2rb^lo uIELs preferentially expressed Il7r, Ccl3 and Ccl4, and the signaling adaptors Lat, Lat2 and Gzmk (Fig. 3b, Extended Data Fig. 4a, b). Genes preferentially expressed in Tcf7⁻Il2rb^hi uIELs included Tyrobp³², which encodes the signaling adaptor DAP12, utilized by activating NK and myeloid receptors, Gzmc, which encodes granzyme C, a marker of CD49a⁺CD103⁺ ILC1s in gut and tumors^33–35 and various genes, including Lag3, Cd200r4, Cd200r1 and Cd200r2 (Fig. 3b, Extended Data Fig. 4a, b). Conversely, Thy1 expression was relatively low in Tcf7⁻Il2rb^hi uIELs compared to Tcf7⁻Il2rb^lo or Tcf7⁺ uIELs (Fig. 3b, Extended Data Fig. 4b). Because IL-15 is known to induce DAP12³⁶ and granzyme C³³, and to downregulate Thy1³⁷, the profile of Tcf7⁻Il2rb^high uIELs was consistent with a strong IL-15 imprinting. We also identified four small CD8αα⁺ clusters: lipidolytic CD8αα⁺ uIELs expressed fatty acid metabolic genes (Fabp1, Fabp2, Rbp2); proliferating CD8αα⁺ uIELs were enriched in genes associated with cell cycling (Mki67, Tuba1b, Tubb5); IFN-stimulated CD8αα⁺ uIELs expressed IFN-stimulated genes (Ifit1, Ifit3, Isg15, Irf7, Stat1); and stressed CD8αα⁺ uIELs were enriched in stress-induced genes (Mt-Co2, Hsph1, Hst1h1c, Hspa1b) (Fig. 3b, Extended Data Fig. 4a).

Fig. 3. Aiolos impacts the Tcf7⁻Il2b^hi subset of CD8αα⁺ IEL.

a, UMAP plots of combined CD8αα⁺ uIELs isolated from Ikzf3^+/+ and Ikzf3^−/− cells (Ikzf3^+/+.1, 11,843 cells; Ikzf3^+/+.2, 14,123 cells; Ikzf3^−/−.1, 9,730 cells; Ikzf3^−/−.2, 13,509 cells). b, Representative UMAP plots depicting expression of Trac, Trbc1, Tcrg-V7, Tcf7, Id3, Kit, Cd160, Xcl1, Il2rb, Prdm1, Gzmb, Ifng, Klrk1, Hcst, Ccl3, Ccl4, Gzmk, Tyrobp, Gzmc, Thy1, Klra5. c, Separate UMAP plots of Ikzf3^+/+ CD8αα⁺ uIELs and Ikzf3^−/− CD8αα⁺ uIELs. Clusters color codes as in a. d, Cluster representation in Ikzf3^+/+ CD8αα⁺ uIELs and Ikzf3^−/− CD8αα uIELs from four biological samples (mice) analyzed. e, Volcano plots of genes differentially expressed in Ikzf3^+/+ and Ikzf3^−/− Tcf7⁻Il2rb^hi uIELs, Tcf7⁻IL2rb^lo uIELs andTcf7⁺ uIELs. FDR was determined by two-sided Wald test (DESeq2). FDR= False Discovery Rate.

Comparison of scRNA-seq data from Ikzf3^+/+ and Ikzf3^−/− CD8αα⁺ uIELs revealed that Tcf7⁻Il2rb^hi formed a distinct cluster in each genotype (Fig. 3c, d). Additionally, Tcf7⁻Il2rb^lo were less abundant in Ikzf3^−/− than in Ikzf3^+/+ mice (Fig. 3c, d), whereas other clusters were less affected by Aiolos deficiency (Fig. 3d). Comparison of Tcf7⁻Il2rb^hi uIELs from Ikzf3^−/− mice versus Ikzf3^+/+ mice indicated 431 unique differentially expressed genes (DEGs) (Fig. 3e). Upregulated genes in Ikzf3^−/− uIELs included Ccl4, Ifng, granzyme genes (Gzmc, Gzmd, Gzmf, Gzmk) and NK receptor genes (Klra1, Klra3, Klrb1b, Klrb1c) (Fig. 3e). Comparison of Tcf7⁻Il2rb^lo uIELs from Ikzf3^−/− and Ikzf3^+/+ mice showed 75 DEGs (Fig. 3e). 102 genes were differentially expressed in Ikzf3^−/− versus Ikzf3^+/+ Tcf7⁺ uIELs (Fig. 3e). In all three clusters, effector genes such as granzymes and NK receptors were upregulated in Ikzf3^−/− uIELs compared to Ikzf3^+/+ uIELs although the impact of Ikzf3 deletion was most pronounced in the Tcf7⁻Il2rb^hi cluster (Fig. 3e). In summary, Aiolos deficiency preferentially affected Tcf7⁻Il2rb^hi CD8αα⁺ uIELs, enhancing gene expression programs related to activating receptors and effector functions.

Aiolos-deficient IELs are hyperresponsive to IL-15

We analyzed intestinal CD45⁺ cells by flow cytometry to validate the protein expression of TCF1 (encoded by Tcf7) and CD122 (encoded by Il2rb) in CD8αα⁺ uIELs in Ikzf3^+/+ and Ikzf3^−/− mice. Gated CD8αα⁺ uIELs of both genotypes encompassed three cell subsets: TCF1⁺ uIELs, TCF1⁻CD122^hi uIELs, and TCF1⁻CD122^lo uIELs (Fig. 4a). The highest differential expression of NK1.1 between Ikzf3^−/− and Ikzf3^+/+ uIELs was detected on TCF1⁻CD122^hi uIELs (Fig. 4b). This observation, together with the fact that exposure to IL-15 induces expression of NK cell receptors on uIELs³⁸, indicated that loss of Aiolos impacted uIEL responsiveness to IL-15.

Fig. 4. Aiolos-deficient IELs are hyperresponsive to IL-15.

a, Representative flow cytometry plots showing CD122 and TCF1 expression in γδ uIELs and DN uIELs isolated from the small intestine epithelium of Ikzf3^+/+ mice (n=5). b, Flow cytometry plots and quantification of NK1.1⁺ γδ uIELs and NK1.1⁺ DN uIELs in the small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice (n=5). c, Flow cytometry plots and quantification of NK1.1⁺ γδ uIELs and NK1.1⁺ DN uIELs in the duodenum, jejunum and ileum of Ikzf3^+/+ and Ikzf3^−/− mice (n=5). d, Flow cytometry plots and quantification of NK1.1⁺ γδ uIELs and NK1.1⁺ DN uIELs isolated from the small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice 72 hours after in vitro culture with 10 ng/ml or 100 ng/ml of IL-15. a-d, Data are representative of two independent experiments and all samples are biological replicates. Data represent mean ± SD. P values were determined by by one-way ANOVA with Tukey’s multiple comparisons test in b and c, and by two-way ANOVA with Tukey’s multiple comparisons test in d. DN= Double Negative.

Next, we purified CD45⁺ cells from different segments of the small intestines of Ikzf3^+/+ and Ikzf3^−/− mice and analyzed differential expression of NK1.1 on gated uIELs. Expression of NK1.1 in Ikzf3^−/− mice was highest in uIELs of the duodenum, and progressively declined in uIELs of distal segments (Fig. 4c). This pattern directly correlated with the expression of mRNA for IL-15 (encoded by Il15) and IL-15Rα (encoded by Il15ra), which are required for IL-15 signaling³⁹, in the segments of the small intestine (Extended Data Fig. 5a), and suggested increased trans-presentation of IL-15 and IL-15 signaling in duodenum compared to the jejunum and ileum.

To test whether Aiolos deficiency enhanced the responsiveness to IL-15, we purified CD45⁺ cells from the whole small intestine of Ikzf3^+/+ and Ikzf3^−/− mice, and cultured these cells with IL-15. Flow cytometric analysis of γδ uIELs and DN uIELs showed that IL-15 induced the expression of NK1.1 in a dose dependent fashion, and Ikzf3^−/− uIELs expressed higher NK1.1 than Ikzf3^+/+ uIELs at all IL-15 concentrations (Fig. 4d). IL-15 induced the expression of NKG2A in Ikzf3^−/− uIELs, but not in Ikzf3^+/+ uIELs (Extended Data Fig. 5b). IL-15 also induced Ly49A in Ikzf3^−/− DN uIELs but not in Ikzf3^−/− γδ uIELs, Ikzf3^+/+ DN uIELs or Ikzf3^+/+ γδ uIELs (Extended Data Fig. 5c). IL-15 did not induce expression of NKG2D on any subset of Ikzf3^+/+ or Ikzf3^−/− uIELs (Extended Data Fig. 5d). In summary, Ikzf3^−/− uIELs acquired a heightened response to IL-15, which might account for some of the observed phenotype in vivo, although some phenotypes were probably independent of IL-15 responsiveness.

Aiolos-deficient uIELs are detrimental in colitis

We asked whether inflammatory conditions that activate IELs coordinately reduced Aiolos expression. We used intraperitoneal injection of polyinosinic:polycytidylic acid (poly-I:C) as a model of inflammation. Poly-I:C causes systemic inflammation, colitis and IEL activation^{38, 40} by inducing massive release of proinflammatory cytokines, including type I interferon (IFN) and IL-12⁴¹. At day 1 post-injection of poly-I:C in Ikzf3^+/+ mice, Aiolos expression was significantly downregulated in γδ uIELs and DN uIELs compared to uIELs from vehicle treated mice (Fig. 5a). To further define the impact of inflammatory cytokines on Aiolos expression, sorted γδ uIELs and DN uIELs were cultured with IL-2 and IL-15 to support survival in vitro and stimulated with IL-12 or IFN-β for 3 days. Both stimuli downregulated Aiolos expression in uIELs compared to the uIELs cultured with IL-2 and IL-15 alone (Fig. 5b), indicating that inflammatory conditions downregulated Aiolos expression.

Fig. 5. Inflammatory conditions control Aiolos expression.

a, Flow cytometry histograms and quantification of Aiolos protein in γδ uIELs and DN uIELs isolated from the small intestine epithelium of Ikzf3^+/+ mice (n = 7) 24 hours after (route) injeciton with either PBS or poly I:C. b, Flow cytometry histograms and quantification of Aiolos protein in γδ uIELs and DN uIELs isoalted from the small intestine of Ikzf3^+/+ mice (n=6) and cultured with either IL-2+IL-15, IL-2+IL-15+IL-12, or IL-2+IL-15+IFNβ for 72 hours prior to analysis. c-d, Representative flow cytometry plots (c) and quantification (d) of NK1.1 and granzyme C protein expression in γδ uIELs and DN uIELs isolated from the small intestine epithelium of Ikzf3^fl/fl and E8I^CreIkzf3^fl/fl mice (n=6). e, Frequency of hematochezia-free Ikzf3^fl/fl and E8I^CreIkzf3^fl/fl mice during day 0 −5 post-administration of DSS in the drinking water. f-g, Quantification of colon length (f) and representative images of H&E stained sections of the colon (g) from Ikzf3^fl/fl and E8I^CreIkzf3^fl/fl mice at day 4 post DSS administration in the drinking water (n = 9). a-g, Data are representative of two independent experiments and all samples are biological replicates. Data represent mean ± SD. The scale bar in (g) represents 100 μm. DN= Double Negative. FMO= Fluorescence Minus One. MFI= Mean Fluorescent Intensity. P values were determined by two-tailed, unpaired Student’s t-test in a, b, d and f, and log-rank test in e.

To test whether Aiolos-deficient uIELs altered host responses to local inflammation in the gut, we used a classical model of acute colitis induced by administration of dextran sulfate sodium (DSS) in drinking water. DSS damages the epithelium and induces colitis ⁴². To selectively delete Ikzf3 in CD8α⁺ T cells, we crossed Ikzf3^fl/fl mice (Extended Data Fig. 6a) with E8I^Cre mice, which express Cre recombinase under the core E8I enhancer of Cd8a gene. We validated that E8I-driven Cre activity preferentially occured in cells that express CD8α in the periphery⁴³ by crossing E8I^Cre mice with Rosa26R^tdTomato mice. E8I^Cre released tdTomato expression in γδ uIELs, DN uIELs, CD8αβ⁺ cIELs and DP IELs, but not in CD4⁺ cIELs, B cells or NK/ILC1s isolated from small intestine of E8I^CreRosa26R^tdTomato mice (Extended Data Fig. 6b), validating that E8I^Cre preferentially deleted the genes from uIELs. We then analyzed the phenotype of uIELs in E8I^Cre Ikzf3^fl/fl mice. E8I^Cre Ikzf3^fl/fl uIELs had higher expression of NK1.1 and granzyme C compared to Ikzf3^fl/fl uIELs (Fig. 5c, d), recapitulating the phenotype of Ikzf3^−/− uIELs. After Ikzf3^fl/fl and E8I^CreIkzf3^fl/fl mice were administered DSS in the drinking water, E8I^CreIkzf3^fl/fl mice exhibited more blood in stools (Fig. 5e), shorter colon length (Fig. 5f, Extended Data Fig. 6c), more prominent ulcers and higher crypt damage (Fig. 5g) than Ikzf3^fl/fl mice at day 4 of treatment, indicating that Aiolos-deficiency in CD8α⁺ cells aggravated chemically-induced colitis.

Aiolos binds regulatory regions of IEL effector genes

To elucidate the mechanisms underpinning the transcriptional reprogramming of Ikzf3^−/− uIELs, we profiled the Aiolos binding sites in γδ uIELs and DN uIELs isolated from Ikzf3^+/+ and Ikzf3^−/− mice using CUT&RUN-seq (cleavage under targets and release using nuclease sequencing). 64% and 66% of Aiolos binding sites in γδ uIELs and DN uIELs, respectively were found in promoter regions (Extended Data Fig. 7a). We identified a total of 7,284 and 3,899 Aiolos binding peaks in γδ uIELs and DN uIELs, respectively, with 3,512 of those peaks shared between the two cell types (Extended Data Fig. 7b). Pathway analysis of genes associated with Aiolos binding sites showed enrichment TCR signaling and NK cell-mediated cytotoxicity pathways (Fig. 6a). Characterization of motifs enriched in Aiolos binding sites using Homer de novo motif analysis exposed the IKZF common binding GGGAA motif (Fig. 6b)^{44, 45}, which was similar to the previously reported ETS motif⁴⁶. Comparison of genes associated with Aiolos binding sites in the CUT&RUN-seq analysis with the DEGs in the RNA-seq analysis showed that Aiolos bound 155 genes of the 343 DEGs in γδ uIELs, and 70 genes of the 396 DEGs in DN uIELs (Fig. 6c). These genes included NK cell receptors (Klrb1b, Klrb1c), cytotoxic enzymes (Gzmc), chemokines (Ccl3, Ccl4) and previously known Aiolos target genes such as Irf8^{47, 48} (Fig. 6c).

Fig. 6. Aiolos binds in close proximity of IEL effector genes and STAT5 binding regions.

a, Pathway analysis of Aiolos-bound genes derived from CUT&RUN-seq for Aiolos using γδ uIELs and DN IELs isolacted from small intestine epithelium of Ikzf3^+/+ mice. b, Homer de-novo motif prediction in genome-wide Aiolos peaks. Left, de novo motifs; middle, closest match in the HOMER database; right, P value of motif enrichment. c, Venn diagrams showing the overlap between Aiolos-bound genes detected by CUT&RUN-seq and DEGs detected by RNA-seq in γδ uIELs and DN uIELs isolated from the small intestine epithelium of Ikzf3^+/+ mice. d, Venn diagrams showing the number of overlapping peaks between Aiolos and STAT5. Aiolos peaks were derived from CUT&RUN-seq for Aiolos using γδ uIELs and DN IELs isolacted from small intestine epithelium of Ikzf3^+/+ mice. STAT5 peaks were sourced from GSE100674. e, IGV snapshots depicting Aiolos and STAT5 binding regions at selected loci. Aiolos peaks were derived from CUT&RUN-seq for Aiolos using γδ uIELs and DN IELs isolacted from small intestine epithelium of Ikzf3^+/+ mice. STAT5 peaks were sourced from GSE100674. f, Venn diagram showing the numbers of overlapping peaks between Aiolos and either RUNX1 or RUNX3. Aiolos peaks were derived from CUT&RUN-seq for Aiolos using γδ uIELs and DN IELs isolacted from small intestine epithelium of Ikzf3^+/+ mice. RUNX1 and RUNX3 peaks were sourced from GSE93755 and GSE50131 respectively. DN= Double Negative. IGV= Integrative Genomics Viewer. P values were calculated by hypergeometric enrichment calculations in a and b.

Because Aiolos restrains IL-15-driven IEL activation and IL-15 signaling triggers the activation of STAT5, we inquired whether STAT5 and Aiolos bound DNA in close proximity of each other. Comparison of genome-wide distribution of STAT5 and Aiolos to identify regions of overlap indicated that 41% and 39% of the Aiolos-binding regions were also bound by STAT5 in γδ uIELs and DN uIELs, respectively, with 2,958 overlapping peaks among 7,284 Aiolos binding peaks in γδ uIELs and 1,527 peaks among 3,899 Aiolos binding peaks in DN uIELs (Fig. 6d). These overlaps were statistically significant (Extended Data Fig. 7d). Aiolos-STAT5 overlapping regions were located in the proximity of NK receptors (Klrb1b, Klrb1c), granzyme B (Gzmb) and chemokines (Ccl3, Ccl5) (Fig. 6e). The intracellular content of phosphorylated STAT5 (p-STAT5) was not significantly different between Ikzf3^−/− and Ikzf3^+/+ γδ uIELs and DN uIELs (Extended Data Fig. 7c), suggesting Aiolos affected STAT5-driven transcription, but not STAT5 phosphorylation. Moreover, Aiolos-binding regions overlapped with the binding sites of other transcription factors and were specifically enriched in RUNX motifs (Fig. 6b). Comparison of the genome-wide distribution of Aiolos and RUNX binding sites showed that among total Aiolos binding peaks, 4,250 and 2,289 overlapped with RUNX1 in γδ uIELs and DN uIELs respectively, while 726 and 351 overlapped with RUNX3 in γδ uIELs and DN uIELs, respectively (Fig. 6f). The binding regions shared by Aiolos with either RUNX1 or RUNX3 were statistically significant (Extended Data Fig. 7d). In summary, these data suggested that Aiolos might cooperate with STAT5 and RUNX family members in regulating gene expression at certain genomic loci.

Aiolos promotes histones deacetylation of IEL effector genes

Aiolos homodimers and heterodimers with Ikaros¹⁹ interact with the NuRD complex to recruit histone deacetylases^{14, 15}. Thus, we used CUT&RUN-seq to profile H3K27ac, a histone marker associated with active transcription. In parallel, we also determined global changes of chromatin accessibility in the same uIEL subsets using ATAC-seq (assay for transposase-accessible chromatin sequencing). Principal component analysis of these epigenetic features showed a segregation of γδ uIELs and DN uIELs based on the Ikzf3 genotype (Fig. 7a), indicating Aiolos deficiency impacted the epigenetic landscape. Analysis of H3K27ac peaks showed significantly higher acetylation (5,704 higher peaks vs 72 lower peaks) in Ikzf3^−/− compared to Ikzf3^+/+ γδ uIELs (Fig. 7b). Among the differentially acetylated peaks, 1,888 were located in promoter regions (8 peaks in Ikzf3^+/+ γδ uIELs compared to 1,880 peaks in Ikzf3^−/− γδ uIELs), and 3,688 in the enhancer regions (64 peaks in Ikzf3^+/+ γδ uIELs vs. 3,624 peaks in Ikzf3^−/− γδ uIELs) (Fig. 7b). A similar trend was observed in H3K27ac peaks in Ikzf3^+/+ and Ikzf3^−/− DN uIELs, with 57 differentially acetylated peaks in the promoter regions (1 peak in Ikzf3^+/+ DN uIELs compared to 56 peaks in Ikzf3^−/− DN uIELs) and 442 peaks in the enhancer regions (36 peaks in Ikzf3^+/+ DN uIELs compared to 406 peaks in Ikzf3^−/− DN uIELs) (Fig. 7b). Differential H3K27ac peaks were more frequent in Ikzf3^−/− γδ uIELs than in Ikzf3^−/− DN uIELs (Fig. 7b), probably due to a compensatory upregulation of Helios, encoded by Ikzf2, in Ikzf3^−/− DN uIELs (Extended Data Fig. 8a) that could attenuate the effects of Aiolos-deficiency.

Fig. 7. Aiolos shapes distinct epigenetic landscapes in (what cells?).

a, Principal component analysis of the ATAC-seq peaks and H3K27ac peaks in γδ uIELs and DN uIELs isolated from the small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice. b, Total numbers of differentially H3K27ac regions in γδ uIELs and DN uIELs as in a. c, Total numbers of differentially accessible (DA) regions in γδ uIELs and DN uIELs in the small intestine epithelium in Ikzf3^+/+ and Ikzf3^−/− mice. d, IGV snapshots of representative loci depicting CUT&RUN-seq for H3K27ac and ATAC-seq in Ikzf3^+/+ and Ikzf3^−/− γδ uIELs and DN uIELs. e, Heatmap showing the transcription factor motif enrichement in differentially acetylated or differentially accessible regions in Ikzf3^+/+ and Ikzf3^−/− γδ uIELs and DN uIELs. P values were calculated by hypergeometric enrichment calculations and the log-transformed P values were z scored. f-g, Plots showing fold changes in gene expression in RNA-seq versus fold changes in H3K27ac peaks (f) or ATAC peaks (g) in the promoter and enhancer regions of γδ uIELs and DN uIELs isolated from the small intestine epithelium of Ikzf3^+/+ and Ikzf3^−/− mice. Highlighted points indicate FDR<0.05 and fold change >1.5 or <−1.5 in both RNA-seq and H3K27ac peaks/ATAC peaks. The line represents a linear regression fit to the highlighted points, with the Pearson correlation coefficient (r) indicating the strength and direction of the linear relationship between the two variables. FDR was calucated by Wald test (DEseq2). DN= Double Negative. P= promoters. E= enhancers. DEGs = Differential Expressed Genes. FDR= False Discovery Rate. FC= Fold Change. IGV= Integrative Genomics Viewer.

Next, we examined the differentially accessible chromatin regions in Ikzf3^+/+ and Ikzf3^−/− γδ uIELs and DN uIELs. A total of 2,311 peaks were differentially accessible between Ikzf3^+/+ and Ikzf3^−/− γδ uIELs (Fig. 7c), of which 437 were located in the promoter regions (28 peaks in Ikzf3^+/+ γδ uIELs vs. 409 peaks in Ikzf3^−/− γδ uIELs), and 1,874 in the enhancer regions (311 peaks in Ikzf3^+/+ γδ uIELs vs. 1,563 peaks in Ikzf3^−/− γδ uIELs) (Fig. 7c). In DN uIELs we observed 2,423 differential accessible peaks, of which 261 were in promoter regions (103 peak in Ikzf3^+/+ DN uIELs vs. 158 peaks in Ikzf3^−/− DN uIELs) and 2,162 in enhancer regions (921 peak in Ikzf3^+/+ DN uIELs vs. 1,241 peaks in Ikzf3^−/− DN uIELs) (Fig. 7c).

To explore the relationship betweens the differential H3K27ac peaks, differential accessibility regions (DAR) and Aiolos binding regions we performed an integrated analysis. Significant overlaps were observed between differential H3K27ac peaks and DAR in γδ uIELs and DN uIELs (Fig. 7d, Extended Data Fig 8b). To address whether Aiolos affected the accessibility of other transcription factors, thereby influencing gene expression indirectly, we determined which transcription factor motifs were enriched in the DAR and differential H3K27ac peaks. Transcription factor motifs including RUNX, BATF and Tbx21 were more highly represented in Ikzf3^−/− γδ uIELs and DN uIELs compared to their Ikzf3^+/+ counterparts (Fig. 7e).

Lastly, to determine how changes in the epigenetic landscape due to Aiolos deficiency contributed to differential gene expression, we annotated the genes associated with differentially accessible and acetylated peaks, and analyzed their correlations with DEGs in RNA-seq. Positive correlations were observed between DEGs and differential H3K27ac peaks both in the promoter and enhancer regions (Fig. 7f). H3K27ac peaks in Ikzf3^−/− γδ uIELs and DN uIELs correlated with genes upregulated in Ikzf3^−/− γδ uIELs and DN uIELs, including those for NK receptors (Klrb1b, Klrb1c, Klrk1), cytokines (Ccl1, Ccl3) and granzyme C (Gzmc) (Fig. 7f). Chromatin regions more accessible in Ikzf3^−/− uIELs correlated with NK receptors (Klrb1b, Klrb1c), chemokines (Ccl3) and known Aiolos-target genes, such as Irf8 (Fig. 7g), indicating that differential ATAC peaks in the promoter and enhancer regions also correlated with DEGs. Integration of Aiolos, STAT5, RUNX1 and RUNX3 binding data, DAR, and differential H3K27ac indicated that these features overlapped in multiple genes expressed in IELs, including NK receptors, chemokines and granzymes (Supplementary Table 2). Taken together, Aiolos modulated chromatin accessibility, promoted histone deacetylation and likely interfered with other transcription factors to regulate expression of IEL effector genes including NK receptors, chemokines and granzymes.

Discussion

Intestinal IELs are often described as innate-like cells that are resting, yet poised for activation, and therefore able to provide an immediate response to tissue injury. Here we showed that Aiolos had a critical role in restraining uIEL expression of NK receptors, cytotoxic mediators and IFN-γ, in part by interfering with their responsiveness to IL-15 signaling. This phenotype was particularly evident in uIELs marked by high expression of CD122, the receptor for IL-15, which also expressed high amounts of Aiolos. We also showed that inflammatory cytokines that activate IELs, such as IL-12 and type I IFN, attenuated Aiolos expression. Thus, we predict that downregulation of Aiolos in response to these cytokines might contribute to activate poised regulatory elements and unleash IEL effector programs. These observations may help explain the reported association between inflammatory bowel disease and Aiolos expression⁴⁹.

Epigenetic analyses showed that Aiolos bound the regulatory regions of genes that encode NK receptors, effector molecules and inhibitory checkpoints, reducing accessibility and H3K27 acetylation of these chromatin regions. Aiolos DNA binding sites significantly overlapped with those of STAT5, the major signaling effector of IL-15. Thus, Aiolos may attenuate STAT5-mediated transcriptional activation. This possibility is consistent with reports showing that Ikaros, a heteromeric partner of Aiolos, antagonizes the STAT5 network^{50, 51} and a report showing that Aiolos mitigates CD4 T cell effector functions by antagonizing IL-2-STAT5 signaling⁵². However, Aiolos DNA binding sites also overlapped with RUNX binding elements, suggesting that Aiolos may additionally interfere with the activity of RUNX transcription factors, such as RUNX3, which also promotes cytotoxic T cell effector programs.

Of note, despite its increased expression in Ikzf3^−/− uIELs, the NKG2D receptor did not trigger uIEL cytotoxicity. Further studies are warranted to explore the impact of Aiolos on NK receptors in uIELs beyond their expression, including their signaling pathways. The ability of Aiolos to interfere with IL-15 signaling is singularly appropriate to maintain intestinal immune homeostasis, as IL-15 is produced in high amounts in the intestinal mucosa and unrestricted IL-15 signaling can cause pathogenic intestinal inflammation as typified by celiac disease⁵³.

This study highlighted Aiolos-dependent regulatory mechanisms in uIEL function through epigenetic modification likely coordinated with STAT5 and RUNX family members, and its pivotal roles in maintaining intestinal homeostasis and preventing inflammation.

Methods

Mice

Aiolos germline deletion mice (Ikzf3^−/−) were obtained from RIKEN. Aiolos conditional knockout mice (Ikzf3^fl/fl) were generated at the Mouse Genetic Core at Washington University School of Medicine. Briefly, LoxP sites were inserting upstream of exon 4 and downstream of exon 6, to delete exons 4–6. Rag knockout mice (strain number 002216) and CD45.1 mice (strain number 002014), E8I^Cre mice (strain number 008766), Cg-Gt(Rosa)26Sor^{m14(CAG-tdTomato)Hze}/J (CAG promoter driven tdTomato expression is prevented by loxP-flanked stop cassette and this construct is inserted to locus encoding Rosa26. strain number 007914) were purchased from The Jackson Laboratory. For all experiments, mice were maintained in a specific pathogen-free facility at Washington University in St. Louis (an ambient temperature of 68–74°F, 30–70% humidity, a 12 hour light/dark cycle). All the experiments on mice were approved by and performed in accordance with the Institutional Animal Care and Use Commnittee at Washington Universtiy in St. Louis.

DSS colitis

Mice were fed with 3% DSS in the drinking water and occult blood in stool was checked by Hemoccult II Sensa kit (Fisher Scientific SK-64130). DSS was purchased from TdB Labs. Dissected full-length colons were cleaned of luminal content, open lengthwise and pinned out. Tissues for paraffin embedding were fixed in 10% neutral buffered formalin solution (MilliporeSigma) overnight at 4°C. Fixed tissues were washed three times in 70% EtOH and embedded in 2% agar (MilliporeSigma), followed by paraffin embedding, sectioning and hematoxylin and eosin staining.

Flow cytometry and antibodies

Single cell suspensions were incubated with Fc block for 15 minutes then stained with antibodies. For intra-cellular cytokine staining, BD Bioscience Fixation/Permeabilization Solution kit was used per manufacture instruction. For transcription factor staining, True-Nuclear Transcription factor staining kit (BioLegend) was used per manufacture instruction. For pSTAT5 staining, stimulated IEL were first fixed with 4% paraformaldehyde then permeabilized with 90% methanol. Data were acquired on BD FACSymphony and analyzed using FlowJo. Following antibodies are used: anti-CD3 (145–2C11 BioLegend), anti-CD4 (GK1.5 BioLegend), anti-CD8α (53–6.7 BioLegend), anti-CD8b (YTS156.7.7 BioLegend), anti-CD122 (TM-B1 BioLegend), anti-NK1.1 (PK136 BioLegend), anti-NKG2A (16A11 BioLegend), anti-NKG2D (CX5 BioLegend), anti-granzyme C (SFC1D8 BioLegend), anti-IFNg (XMG1.2 BioLegend), anti-TCRb (H57–597 BioLegend), anti-TCRd (GL3 BioLegend), anti-TCF1 (C63D9 CellSignaling), anti-Rabbit IgG (Poly4064 BioLegend), pSTAT5 (Clone 47/Stat5 (p694) BD Biosciences), Ly49D (4E5, BioLegend), Ly49A/D (12A8, eBioscience), Ly49E/F (CM4, eBioscience), Ly49H (3D10, BioLegend). For flowcytometry, all antibodies were diluted to 1 in 100.

Tissue preparation

IELs were prepared as previously described ⁵⁴. Briefly, small intestines were flushed and Peyer’s patches were removed. Then, small intestines were sliced and agitated in HBSS, HEPES, dithioerythritol (Sigma) and bovine calf serum. Supernatant were filtered through 100 micron mesh and subjected to density gradient using 40% and 70% Percoll.

Poly I:C treatment

Poly I:C was purchased from Invivogen (catalog code: tlrl-picw) and dissolved in pathogen-free phosphate-buffered saline. 20 μg/g of Poly I:C was given through intraperitoneal route. 24 hours after the treatment, IELs were harvested.

IELs in vitro culture

When cultured in vitro to analyze cytokine production and response to IL-12, IL-15 or IFNβ, IEL are purified with EasySep Mouse CD45 Positive Selection kit (STEMCELL technologies #18945). For cytokine stimulation, 96 well flat plate was coated with anti-CD3 antibody (5 μg/ml) over night, then washed three times with sterile PBS. Isolated IEL are stimulated with plate-bound CD3 for 20 hours, and last 4 hours with Brefeldin A (BD GolgiPlug). When analyzing IEL response to IL-12 (10 ng/ml), IL-15 (10 or 100 ng/ml) and IFNβ (1000 units/ml), cytokines are added to the RPMI medium and purified IEL are co-cultured for 72 hours. When analyzing pSTAT5, purified IEL were cultured with IL-15 (100 ng/ml) in RPMI medium for 15 minutes.

Mixed bone marrow chimera

Recipient Rag deficient mice underwent 6 Gy irradiation. Bone marrow was harvested from Aiolos-deficient mice and CD45.1 mice, then mixed at 1 to 1 ratio. Total 1 × 10⁶ mixed bone marrow cells were injected into sub-lethally irradiated Rag deficient mice through retro-orbital route. At week 6–8 after the injection, IEL were harvested from the recipient mice.

CUT&RUN-seq

CUT&RUN-seq was performed using CUTANA™ ChIC/CUT&RUN kit (EpiCypher #14–1048) following manufacturer’s instructions. For each procedure, 1μg of antibodies and sorted 0.5×10⁶ cells were used. Anti-H3K27 antibody (EpiCypher #13–0045), anti-Aiolos antibody (Sigma Aldrich #9D10), rabbit-anti-mouse immunoglobulin G (included in the kit, EpiCypher #13–0045) were used. Libraries are generated using NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB #E7645S) and NEBNext Multiplex Oligos for Illumina (NEB #E7600S) following the manufacture’s protocol. Total DNA integrity was determined using Agilent Bioanalyzer. 75 base paired end sequence was performed using NovaSeq6000.

ATAC-seq

50,000 cells were sorted and processed as per manufacturer’s instructions (Nextera DNA Library Preparation Kit, Illuminia) and previously described ⁵⁵. Total DNA integrity was determined using Agilent Bioanalyzer. 150 base paired end sequence was performed using NovaSeq6000.

Epigenome analysis

For STAT5, RUNX1 and RUNX3 binding analysis, published ChIP-seq data was used (GSE100674, GSE93755, GSE50131). Sequencing data from ATAC-seq, CUT&RUN-seq and ChIP-seq were mapped onto mouse genome build mm9 using Bowtie2. Duplicate reads were removed using Picard’s markDuplicates. MACS2 was used for peak calling using a p value threshold of 1 × 10⁻⁵. For CUT&RUN-seq for Aiolos, Aiolos-deficient cells were used as a control input source, and irreproducible discovery rate (IDR) peaks were identified using a threshold of 0.05. BigWig track were visualized by Integrative Genomics Viewer. Permutation test was performed using the ChIPpeakAnno library, and for each transcription factor, DAR and differential H3K27ac, we executed 1000 permutations. As the background for the permutation, we utilized the total ATAC peak counts of wild type cells. Differential accessible peaks and differential H3K27ac peaks were calculated and principal component analysis was conducted using Diffbind R package. Promoter regions were defined as peaks that overlapped a region that was +1kb to −1kb from the transcription start sites. Peak annotation was conducted using the annotatePeaks module in the HOMER package. De novo motif analysis and KEGG pathway analysis using CUT&RUN-seq for Aiolos were performed by the findMotifsGenome module and the annotatePeaks module respectively. Overlapping peaks are identified using BedTools intersect with settings -v and -wa. Graphs were generated using custom R code in Rstudio. Quality control metrics are presented in Supplemmentary Table 2.

Single cell RNA-seq

CD45⁺ cells were sorted then cDNA was prepared after the GEM generation and barcoding, followed by the GEM-RT reaction and bead cleanup steps. Purified cDNA was amplified for 10–14 cycles before being cleaned up using SPRIselect beads. Samples were then run on a Bioanalyzer to determine the cDNA concentration. GEX libraries were prepared as recommended by the 10x Genomics Chromium Single Cell V(D)J Reagent Kits (v1 Chemistry) user guide with appropriate modifications to the PCR cycles based on the calculated cDNA concentration. For sample preparation on the 10x Genomics platform, the Chromium Single Cell 5’ Library and Gel Bead Kit v1 (PN-1000006), Chromium Single Cell A Chip Kit (PN-1000152) and Chromium Single Index Kit T (PN-1000213) were used. The concentration of each library was accurately determined through qPCR utilizing the KAPA library Quantification Kit according to the manufacturer’s protocol (KAPA Biosystems/Roche) to produce cluster counts appropriate for the Illumina NovaSeq6000 instrument. Normalized libraries were sequenced on a NovaSeq6000 S4 Flow Cell using the XP workflow and a 151×10×10×151 sequencing recipe according to manufacturer protocol. A median sequencing depth of 50,000 reads/cell was targeted for each sample. Cell ranger was used to align samples to the reference mm10 genome and generate gene-cell expression matrices. Downstream analysis including normalization, scaling, and clustering was performed using Seurat package (v.4.0.3). Cells with fewer than 200 or more than 5000 detected genes, or cells with the percentage of mitochondrial reads greater than 5% were excluded from the analysis. Filtered data were scaled by library size, then long-transformed. Highly variable genes were selected using the FindVariableFeatures, and clustering was performed using FindClusters function. UMAP was used to project cells into two dimensions using 20 first principal components. Differential gene expression of Il2rb^high CD8αα or IL2rb^low CD8αα clusters from Ikzf3^+/+ and Ikzf3^−/− was analyzed using DESeq2 package.

RNA-seq

50,000 cells of Ikzf3^+/+ or Ikzf3^−/− γδ IEL and DN IEL were sorted, and RNA was extracted with an RNeasy Micro Kit (QIAGEN) following manufacture’s instructions. Total RNA integrity was determined using Agilent Bioanalyzer or 4200 Tapestation. ds-cDNA was prepared using the SMARTer Ultra Low RNA kit for Illumina Sequencing (Takara-Clontech) per manufacture’s protocol. cDNA was fragmented using a Covaris E220 sonicator using duty cycle 10, intensity 5, cycles/burst 200, time 180 seconds. cDNA was blunt ended, had an A base added to the 3’ ends, and then had Illumina sequencing adapters ligated to the ends. Ligated fragments were then amplified for 12 cycles using primers incorporating unique index tags. Fragments were sequenced on an Illumina HiSeq-3000 using single reads extending 50 bases. Sequences were aligned and gene count tables were produced using STAR aligner. DESeq2 R package was used to normalize and identify differentially expressed genes.

Microarray

A small portion of duodenum and ileum were harvested and washed with PBS. RNA was extracted with an RNeasy Mini Kit (QIAGEN) following manufacture’s instructions. Total RNA integrity was determined using Agilent Bioanalyzer. Biotinylated cDNA were prepared according to the Affymetrix WT Plus reagent kit from 100ng-500 ng total RNA. Following fragmentation and labeling, 3.5 ug of cDNA were hybridized onto the Mouse Gene 2.0 arrays and placed in GeneChip Hybridization Oven 640 for 18 hr at 45C. Microarrays were washed and stained in the Affymetrix Fluidics Station 450. GeneChips were scanned using the Affymetrix GeneChip 7G 3000 Scanner. Raw data were normalized with RMA using Affy R package, and differential expression was identified using limma R package.

Statistics and reproducibility

Statistical analysis was performed with R or Prism 9 software (GraphPad). For statistical comparisons between two groups, we used Student’s two-tailed t test. When analyzing more than 3 groups, we performed either one-way or two-way analysis of variance followed by Tukey’s multiple comparison test for pairwise group comparisons. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications^{22, 52}. Data distribution was assumed to be normal but this was not formally tested. Age- and sex-matched Ikzf3^+/+ littermates were used as controls, and mice were randomly assigned in each group. No data was excluded. Data collection and analysis were not performed blind to the conditions of the experiments.

Extended Data

Extended Data Fig. 1. Aiolos does not alter abundance of IELs.

Cell counts of γδ IEL, DN IEL, DP IEL, CD4+ IEL and CD8αβ+ IEL from small intestine in Ikzf3+/+ and Ikzf3−/− mice. Data are representative of two independent experiments (n = 4) and all samples are biological replicates. DN = double negative, DP = double positive (CD4 cytotoxic), Data represent mean ± SD. P values were determined by two-tailed, unpaired Student’s t-test.

Extended Data Fig. 2. NKG2D does not directly induce cytotoxicity in unconventional IELs.

a, Gene expression of NK receptors associated with DAP12 (Klrk1, Klrc2 and Klra8) in γδ IEL and DN IEL from small intestine in Ikzf3+/+ and Ikzf3−/− mice based on RNA-seq. b, γδ IEL and DN IEL from small intestine in Ikzf3+/+ and Ikzf3−/− mice were sorted, and co-cultured with RMA-S or RMA-S overexpressing Rae1γ. Frequency of live target cells are shown. Data showed one experiment in b and all samples are biological replicates. P values in a were calculated by Wald test (DEseq2). WT=Ikzf3 +/+; KO=Ikzf3 −/−.

Extended Data Fig. 3. scRNA-seq demonstrate heterogeneity of IELs.

a, UMAP plots of 25,966 cells of Ikzf3+/+ small intestine CD45+ IELs. b, Representative UMAP plots depicting expression of Trac, Cd8a, Tcrg-V7 and Tcrg-V1. c, Heatmap displaying top differentially expressed genes in clusters designated as CD8aa-1, CD8αα−2, CD8αα−3, CD8αα−4, CD8αα−5, CD8αα−6, CD8αβ+, CD4+, pDC, B cells and plasma cells.

Extended Data Fig. 4. CD8αα+ IELs comprise diverse groups of cells.

a, Heatmap displaying top differentially expressed genes in clusters designated as Ikzf3−/−Tcf7−Il2rbhi, Ikzf3+/+Tcf7−Il2rbhi, Tcf7−Il2rblo, Tcf7+, stressed CD8αα, lipidolytic CD8αα, proliferating CD8αα and IFN-stimulated CD8αα. b, Representative UMAP plots depicting expression of Klra4, Klra8, Klrc2, Il2ra, Cd200r1 Cd200r2, Lat, Lat2, Lag3, Cd200r4, Pdcd1, Havcr2, Lair2 and Il7r.

Extended Data Fig. 5. Ikzf3-deficiency enhances responses to IL-15.

a, Heatmap showing differentially expressed cytokine genes in the duodenum and ileum from Ikzf3+/+ mice. b-d, Flow cytometry plots and quantification of NKG2A (b), Ly49A and D (c) and NKG2D (d) on γδ IEL and DN IEL from small intestine in Ikzf3+/+ and Ikzf3−/− mice 72 hours after culturing with either 10 ng/ml or 100 ng/ml of IL-15. Data are representative of two independent experiments and all samples are biological replicates. n=8 in b, c and n=5 (Ikzf3+/+) and 6 (Ikzf3−/−) in d. Data represent mean ± SD. P values were calculated by two-way ANOVA with Tukey’s multiple comparisons test in b, c and d.

Extended Data Fig. 6. Generation and phenotype of Aiolos conditional deficient mice.

a, Diagram of the construct for conditional Ikzf3fl/fl mice showing LoxP sites flanking exon 4 and exon 6. b, Representative histograms show tdTomato expression in γδ IEL and DN IEL, DP IEL, CD4+ IEL, CD8αβ+ IEL, NK cells and B cells from small intestine in E8ICre R26RtdTomato mice. c, Representative image of colon from Ikzf3fl/fl and E8ICreIkzf3fl/fl mice at day 4 post DSS administration in the drinking water. Data are representative of two independent experiments.

Extended Data Fig. 7. Interactions of Aiolos with STAT5 and RUNX members.

a, A pie chart showing the chromosomal location of Aiolos binding regions in the indicated cell types. b, A Venn diagram depicting Aiolos binding peaks in γδ IELs and DN IELs. c, Represenative histogram plots and quantification of pSTAT5 expression in γδ IEL, DN IEL from the small intestine of Ikzf3+/+ and Ikzf3−/− mice. Cells were either not stimulated or stimulated for 15 minutes with IL-15 (100 ng/ml). d, The results of permutation test designed to analyze the significance of overlap of Aiolos with either STAT5, RUNX1 or RUNX3. Evperm: the number of permutation events, Evobs: the number of observed events (overlapped peak counts). Data are representative of two independent experiments and all samples are biological replicates (n = 4 in c). P values were calculated by unpaired, two-tailed Student’s t-test in c and Permutation test in d (ChIPpeakAnno). Data represent mean ± SD. pSTAT5 = phosphorylated STAT5, UTR = untranslated region.

Extended Data Fig. 8. Aiolos shapes a distinct epigenetic landscape.

a, Gene expression of IKZF family members in γδ IEL and DN IEL from the small intestine in Ikzf3+/+ and Ikzf3−/− mice based on RNAseq (see Fig. 2). The expression of Ikzf3 in Ikzf3−/− IEL reflects a portion of the nonfuntional transcript present in the knockout mouse. b, Venn diagrams depicting overlap of Aiolos binding regions, differential H3K27ac peaks and DAR, and the result of permutation test designed to analyze the significance of overlap of differential H3K27ac peaks with DAR in γδ IELs and DN IELs. FDR were calculated by Wald test (DESeq2) in a and P values were calculated Permutation test in b (ChIPpeakAnno). DAR = differential accessible regions.

Supplementary Material

supplemental tables

Supplementary Table 1 Comprehensive list of DEGs in the indicated cell types, along with their respective associations with DAR, differential H3K27ac, binding of Aiolos, STAT5, RUN1 or RUNX3. ‘+’ indicates positive association, whereas ‘−’ denotes no association.

Supplementary Table 2 Quality-control metrics of ATAC-seq and CUT&RUN-seq for Aiolos and H3K27ac.

Data availability

Sequencing data of RNA-seq, ATAC-seq, scRNA-seq, microarray and CUT&RUN-seq are deposited in GEO (GSE220030). For STAT5, RUNX1 and RUNX3 binding analysis, published ChIP-seq data was used (GSE100674, GSE93755, GSE50131).