AD-like disease can be elicited in the skin of mice with repeated application of the topical vitamin D analog calcipotriol (MC903) (Li et al., 2006). Although it has been previously demonstrated that MC903 can induce AD-like disease in RAG-deficient mice that lack T and B cells, in part via ILC2 activation (Salimi et al., 2013; Kim et al., 2013), the relative contributions of ILC2s and the adaptive lymphocyte compartment have not been rigorously evaluated. We hypothesized that the presence of Th2 cells, in addition to ILC2s, would lead to enhanced AD-like disease in an additive fashion. In testing this, we evaluated both RAG1-sufficient wild-type (WT) mice and RAG1-deficient Rag1^-/- mice in the setting of AD-like disease (Figure 1A). Unexpectedly, we observed that Rag1^-/- mice developed increased ear skin thickness (Figure 1B) and increased absolute numbers and proportion of ILC2s in the skin-draining lymph nodes (sdLNs) compared to control WT mice (Figure 1C and D; Figure 1—figure supplement 1A and B). Furthermore, a larger proportion of ILC2s from Rag1^-/- mice exhibited production of both IL-5 (Figure 1E; Figure 1—figure supplement 1C) and IL-13 (Figure 1F; Figure 1—figure supplement 1C) following ex vivo stimulation and intracellular cytokine staining. Our findings indicated that RAG1 deficiency results in paradoxically worse AD-like disease in association with enhanced ILC2 expansion and activation.

To determine whether this phenomenon was specific to AD-like pathological conditions, we next examined the sdLNs in Rag1^-/- and lymphocyte-sufficient Rag1^+/- littermate control mice in the absence of disease (Figure 1G). We found that the absolute number and frequency of ILC2s was increased at steady state in Rag1^-/- sdLNs (Figure 1H and I) and that a higher proportion of these ILC2s produced both IL-5 (Figure 1J) and IL-13 (Figure 1K) compared to WT controls. The RAG recombinase requires both RAG1 and RAG2 components to successfully rearrange a functional antigen receptor in adaptive lymphocytes (Liu et al., 2022). Thus, to test whether our findings are specific to RAG1, or related to the function of the overall RAG complex, we similarly examined the steady-state profile of ILC2s in Rag2^-/- mice (Figure 1—figure supplement 2A). Deficiency of RAG2 led to an expansion of ILC2s in the sdLNs (Figure 1—figure supplement 2B and C) and increased proportions of ILC2s expressing IL-5 (Figure 1—figure supplement 2D) and IL-13 (Figure 1—figure supplement 2E) similar to Rag1^-/- mice. Collectively, these findings suggest that the RAG recombinase modulates ILC2 function at steady state and during type 2 inflammation. However, whether the hyperactive ILC2 phenotype is due to a cell-intrinsic process or simply due to the absence of T and B cells was unclear.

Given the importance of the adaptive lymphocyte compartment in shaping the secondary lymphoid organ repertoire, we next wanted to examine whether the presence of adaptive lymphocytes could restore ILC2 homeostasis in RAG-deficient mice. To test this, we created splenocyte chimera mice by reconstituting both Rag1^-/- and control WT mice with splenocytes containing T and B cells from WT donor mice (Figure 2A). We first assessed the overall level of immune reconstitution in the recipient mice and found fully restored proportions of CD4⁺ (Figure 2—figure supplement 1A) and CD8⁺ (Figure 2—figure supplement 1B) T cells in the spleens of recipient Rag1^-/- mice, although B cell numbers remained significantly lower than in WT mice (Figure 2—figure supplement 1C). Upon induction of AD-like disease, we found that the Rag1^-/- mice still exhibited increased ear skin thickness (Figure 2B), enhanced expansion of ILC2s (Figure 2C and D), and increased proportions of ILC2s expressing IL-5 (Figure 2E) and IL-13 (Figure 2F) in the sdLNs. Interestingly, we found significantly higher proportions of eosinophils in the spleens of Rag1^-/- recipient mice (Figure 2—figure supplement 1D), possibly reflecting the increased IL-5 production we observed from ILC2s. These findings indicate that the mere introduction of exogenous T and B cells is not sufficient to suppress ILC2 dysregulation in the setting of RAG deficiency.

To further test whether this phenotype is mediated by cell-intrinsic RAG expression, we next generated mixed bone marrow (BM) chimeras. We harvested BM from congenic CD90.1⁺ WT and CD90.2⁺ Rag1^-/- donor mice on a CD45.2⁺ background and reconstituted sub-lethally irradiated CD45.1⁺ congenic WT recipients with a 50:50 mixture of WT:Rag1^-/- BM (Figure 2G). After confirming the reconstitution of donor immune cells in the sdLN (Figure 2—figure supplement 2A and D), we examined the frequency and activity of ILC2s in the sdLNs based on whether they originated from WT (CD90.1⁺) or Rag1^-/- (CD90.2⁺) donors (Figure 2—figure supplement 2B and C). Strikingly, of the total donor ILC2s, the majority were derived from Rag1^-/- donors (Figure 2H; Figure 2—figure supplement 2F and G). This was not due to differences in overall donor reconstitution, since measuring all Lin^- cells revealed WT donor cells outnumbered those from Rag1^-/- donors (Figure 2—figure supplement 2E). Of the total IL-5- (Figure 2I; Figure 2—figure supplement 2H, I and K) and IL-13-expressing (Figure 2J; Figure 2—figure supplement 2H, J and L) ILCs, the majority were also derived from Rag1^-/- donors. Taken together, these data suggest that cell-intrinsic RAG activity in ILC2s may limit their capacity to expand and become activated.

In contrast to resting naïve T cells, ILC2s resemble activated Th2 cells at a steady state based on their transcriptomic and epigenomic profiles (Shih et al., 2016; Van Dyken et al., 2016). While both T cells and ILC2s exhibit historical RAG expression (Yang et al., 2011), they do not actively express the protein in their mature state (Turka et al., 1991). Taken together, these findings provoke the hypothesis that ILC2s are regulated by RAG early in development to imprint alterations that influence their activity as mature cells. To distinguish ILC2s as either having a history of RAG expression or not, we utilized a RAG lineage tracing system, whereby a Rag1^Cre mouse was crossed to a reporter mouse expressing tandem dimer red fluorescent protein (tdRFP) in a Cre-dependent manner from the Rosa26 locus (Figure 3A; Karo et al., 2014; Welner et al., 2009). This system allowed us to compare RAG-experienced (RAG^exp) and RAG-naïve (RAG^naïve) lymphoid cells, including ILC2s, simultaneously originating from the same immunocompetent host, thus removing confounders inherent in knockout and chimera experiments. Analysis of sdLN from the reporter mice revealed that nearly all CD4⁺ T cells, CD8⁺ T cells, and B220⁺ B cells expressed tdRFP (positive history of Cre expression from the Rag1 locus), consistent with the known requirement of RAG expression for their development (Figure 3B–D and G). We also examined NK cells, since certain subsets of NK cells are known to express RAG during their development (Karo et al., 2014), and we observed that roughly 60% of NK cells were tdRFP⁺, similar to previous findings (Figure 3E and G; Karo et al., 2014). In the ILC2 population of the sdLN, around 50% were tdRFP⁺ (Figure 3F and G), similar to proportions of RAG fate-mapped ILC2s previously observed in the fat (Karo et al., 2014) and lung (Yang et al., 2011; Karo et al., 2014). These findings demonstrate that there are heterogeneous populations of ILC2s marked by differential tdRFP⁺ fate mapping. Importantly, this provided us with the possibility to profile these different ILC2 subsets based on Rag1^Cre-activated expression of tdRFP.

Transient RAG expression early in lymphoid development leads to well-characterized, durable effects on B and T cell development and function mainly through successful genomic rearrangement of antigen receptors. Yet, our data indicate that RAG expression also imprints phenotypic changes on ILC2s, which can develop independently of functional antigen receptors, provoking the hypothesis that RAG expression may affect broader epigenomic and transcriptional programs. Furthermore, our data indicate that the impact of RAG on ILC2 function has implications for AD-like skin inflammation, suggesting a persistent effect that modulates phenotypes of type 2 inflammation.

To test these hypotheses, we performed combined single-nuclei RNA sequencing (snRNA-seq) and ATAC sequencing (snATAC-seq) of sdLN cells (Figure 4) from RAG fate-mapped mice at steady state and in the setting of AD-like disease (Figure 4—figure supplement 1A and B). Because fluorescence-activated cell sorting (FACS) can cause physical stress, cell loss, and contamination, which can introduce unwanted perturbations in target cells, instead we utilized gentle initial negative selection by magnetic activated cell sorting (MACS) to remove most B and T cells and monocytes prior to sequencing (Figure 4—figure supplement 1A). This allowed us to enrich for innate immune cell populations prior to sequencing. Further, we used the gene encoding tdRFP as a barcode to differentiate between RAG^exp (RAG fate map-positive) and RAG^naïve (RAG fate map-negative) ILC2s at the single-cell level (Figure 5A). The multiomic data was analyzed using recently developed pipelines in Cell Ranger, Seurat (Stuart et al., 2019; Hao et al., 2021; Butler et al., 2018), and Signac (Stuart et al., 2021), and sequenced cells were further filtered computationally to enrich for ILCs, as in previous studies (see methods) (Ghaedi et al., 2020).

In addition to gene expression (GEX) information derived from snRNA-seq (Figure 4A), we calculated a ‘gene activity’ (GA) score based on chromatin accessibility at gene loci (Figure 4B) from the corresponding snATAC-seq dataset (Stuart et al., 2021). Clustering the cells with each data subset alone and in combination using weighted nearest neighbor (WNN) analysis, we identified six clusters (Figure 4C) that demonstrated consistent differences in cellular markers based on both metrics of GEX and GA (Figure 4D and E; Supplementary file 1, Table S1). Additionally, top markers for each cluster clearly differentiated each cell type (Figure 4F). Despite successful ILC2 enrichment via MACS depletion for lineage markers and computational filtering (see methods), our data set included non-ILC2 populations determined by gene expression to be T cells, dendritic cells, B cells, and NK cells (Figure 4C and F; Supplementary file 1, Table S1), which allowed for broader multidimensional comparisons while studying this ILC2-enriched data set.

To further complement the GEX and GA assays, we utilized another method of detecting cell-specific marker genes, whereby chromatin regions that are differentially accessible (DA), or open, in each cluster could be linked by their physical proximity to specific genes (Figure 4G; Supplementary file 1, Table S2; see methods). Comparing the top 100 GEX, GA, and DA markers in the ILC2 population, we identified a multiomic ILC2 signature of 235 unique genes (Figure 4H; Supplementary file 1, Table S3, S4). While there was some overlap between each respective set, the multiomic approach enabled more extensive identification of an ILC2 gene program than through either snRNA-seq or snATAC-seq alone. The analysis revealed a variety of canonical ILC2-associated genes specific to the ILC2 cluster (Figure 4I, Figure 4—figure supplement 2, Figure 4—figure supplement 3) including the ILC2-activating cell surface receptors Icos (Maazi et al., 2015; Paclik et al., 2015), Il2ra (Roediger et al., 2013; Roediger et al., 2015), Il18r1 (Ghaedi et al., 2020; Zeis et al., 2020; Ricardo-Gonzalez et al., 2018), Nmur (Tsou et al., 2022; Jarick et al., 2022; Cardoso et al., 2017; Klose et al., 2017; Wallrapp et al., 2017; Figure 4—figure supplement 2A), and Il1rl1 (encoding the receptor for IL-33) (Imai et al., 2013; Salimi et al., 2013; Mjösberg et al., 2011; Hung et al., 2013), as well as transcription factors such as Gata3 (Moro et al., 2010; Mjösberg et al., 2012; Hoyler et al., 2012; Klein Wolterink et al., 2013; Pokrovskii et al., 2019), Bcl11b (Califano et al., 2015; Walker et al., 2015; Yu et al., 2015; Hosokawa et al., 2020), Maf (Moro et al., 2010; Björklund et al., 2016; Trabanelli et al., 2022), Ets1 (Zook et al., 2016), and Rora (Ghaedi et al., 2020; Hoyler et al., 2012; Wong et al., 2012; Halim et al., 2012; Ferreira et al., 2021), all previously shown to be important for ILC2 development and/or function.

Expression of some secreted proteins can be difficult to capture in droplet-based snRNA-seq experiments due to low transcript levels and relatively shallow sequencing depth. With the addition of the complementary GA and DA assays from snATAC-seq, our analysis identified Il5 (Figure 4I, Figure 4—figure supplement 2B), a canonical ILC2 cytokine, in the DA assay, while in the GA assay, we found Bmp7 (Miyajima et al., 2020; Chen et al., 2021), which has been shown to be secreted by ILC2s to influence browning of adipose tissue. Additionally, we identified the secreted chemokine Ccl1 as an ILC2 marker gene (Xu et al., 2019; Vivier et al., 2016; Schneider et al., 2019; Bielecki et al., 2021), which along with its cognate receptor Ccr8 (also an ILC2 marker in our analysis) (Ricardo-Gonzalez et al., 2018; Vivier et al., 2016; Sun et al., 2022), participates in a feed-forward circuit to drive ILC2 recruitment and expansion (Knipfer et al., 2019). Thus, our findings demonstrate how genetic barcoding, combining transcriptomic and epigenomic analyses, and cross-validation across many published studies can yield new insights while providing internal control measures to elevate the rigor, robustness, and confidence of identifying gene signatures of rare populations such as ILC2s at the single-cell level.

As noted above, barcoding the ILC2s afforded the opportunity to transcriptionally and epigenomically profile ILC2s under identical developmental conditions by dividing the ILC2 cluster into RAG^exp and RAG^naïve populations (Figure 5A). We hypothesized that RAG^exp ILC2s would have a distinct transcriptional profile compared to ILC2s without any history of RAG expression. To test this, we calculated differentially expressed genes (DEGs) for the ILC2 cluster by RAG fate-map status. Genes with higher expression in RAG^exp cells relative to RAG^naïve cells had positive fold change values, and vice versa, with genes relatively increased in RAG^naïve cells having negative values (Figure 5—figure supplement 1A, Supplementary file 1, Table S5). Using gene set enrichment analysis (GSEA) Subramanian et al., 2005; Mootha et al., 2003 on the ranked list of DEGs, we found that gene sets generally representing lymphocyte activation and differentiation were suppressed in RAG^exp ILC2s compared to RAG^naïve ILC2s (Figure 5—figure supplement 1B and C, Supplementary file 1, Table S6), consistent with our previous observations that ILC2s are expanded and more activated in RAG-deficient mice relative to WT mice.

We next employed newly described methodologies (Stuart et al., 2021; Ma et al., 2020) that quantify associations between open chromatin peaks and the expression of nearby genes to describe the functional regulomes of both RAG^exp and RAG^naïve ILC2s (Figure 5B). In this analysis, each ATAC peak can be linked to multiple genes, and each gene to multiple peaks, generating a list of ‘gene-to-peak links’ or GPLs (see methods). For each gene, we interpreted the number of corresponding GPLs as a quantitative representation of the regulome activity for that gene. Considering RAG^exp and RAG^naïve cells as two separate populations, we generated two lists of GPLs (Supplementary file 1, Table S7) defining functional regulomes for each population. We focused our analysis on the functional regulomes of ILC2s by filtering the GPL lists for the 235 unique ILC2 genes identified in our multiomic gene set (Figure 4H; Supplementary file 1, Table S8), then calculated the difference in GPLs between RAG^exp and RAG^naïve cells for each gene and sorted them; genes displaying greater numbers of GPLs in the RAG^naïve population are at the top, and genes with more GPLs in the RAG^exp population are at the bottom (Figure 5C; Supplementary file 1, Table S9).

We found that the ILC2 marker genes segregating toward the top of this list, corresponding to enhanced epigenomic activity in the RAG^naïve cells, tended to be genes previously identified to play positive roles in the development, expansion, and activation of lymphoid cells. These included transcriptional regulators such as Tox (Constantinides et al., 2014; Seehus et al., 2015; Aliahmad et al., 2010; Aliahmad and Kaye, 2008), Rora (Ghaedi et al., 2020; Hoyler et al., 2012; Wong et al., 2012; Halim et al., 2012; Ferreira et al., 2021), Maf (Moro et al., 2010; Björklund et al., 2016; Trabanelli et al., 2022), and Gata3 (Moro et al., 2010; Mjösberg et al., 2012; Hoyler et al., 2012; Klein Wolterink et al., 2013; Figure 5—figure supplement 2A), which are involved in early differentiation of both ILCs and lymphocytes. Indeed, epigenetic activation of the Gata3 locus is recognized to play a critical role in the development of both ILC2s (Kasal et al., 2021) and Th2 cells (Wei et al., 2010; Onodera et al., 2010). Additionally, surface receptors known to drive ILC2 activation upon stimulation including Il18r1 (Ghaedi et al., 2020; Zeis et al., 2020; Ricardo-Gonzalez et al., 2018), Il1rl1 (Neill et al., 2010; Ghaedi et al., 2020; Zeis et al., 2020; Barlow et al., 2013; Molofsky et al., 2015), and Icos (Maazi et al., 2015; Paclik et al., 2015) had increased functional regulome activity in RAG^naïve ILC2s. In contrast, genes with more GPLs in the RAG^exp ILC2s tended to be associated with suppressive functions. For example, Ndfip1 (Figure 5—figure supplement 2B) encodes a regulatory protein that enhances the activity of the ubiquitin ligase ITCH to negatively regulate inflammation (Oliver et al., 2006; Altin et al., 2014) and has been associated with asthma risk in GWAS studies (Ferreira et al., 2011). Dusp1 partially mediates glucocorticoid effects through its ability to negatively regulate inflammation (Chi et al., 2006; Clark et al., 2008), is associated with eczema by GWAS (Grosche et al., 2021), and has recently been shown to mark an anti-inflammatory set of ILCs (Bielecki et al., 2021). Last, Asxl1 encodes a tumor suppressor that inhibits clonal hematopoiesis through its epigenomic regulatory effects in both mice and humans (Xie et al., 2014; Genovese et al., 2014; Siddhartha et al., 2014; Nagase et al., 2018). Collectively, our GPL analysis stratifies the ILC2 gene signature based on RAG experience, where genes associated with ILC2 expansion and activation are poised in RAG^naïve cells, while genes associated with suppressive effects are poised in RAG^exp cells.

We expanded our multiomic analysis to infer information about transcription factor (TF) activity from open chromatin regions in our snATAC data. We used the chromVAR package (Schep et al., 2017), which finds known TF binding motifs in open chromatin regions in each cell, to identify TF motifs enriched in each cell cluster (Figure 5—figure supplement 3A and B, Supplementary file 1, Table S10). The enriched motifs were consistent with the known functional roles of associated TFs in each cell type. For example, in the NK cell cluster, we found enriched motifs recognized by the TFs EOMES and T-bet (encoded by Eomes and Tbx21, respectively), which are critical for the development of NK cells (Bando and Colonna, 2016). A limitation of this analysis is that while TF motif accessibility can be inferred from open chromatin in snATAC data, which TFs are bound to the identified accessible sites is not known. We reasoned that complementary gene expression information from our multiomic data could mitigate this limitation in part by comparing the accessibility of TF binding motifs to expression levels of corresponding TFs (Figure 5—figure supplement 3C). Indeed, motifs for both RORα and RORγ (encoded by Rora and Rorg, respectively), which share a common DNA binding 5'-AGGTCA-3' half site, have similar calculated accessibilities in both the ILC2 and NK cell clusters. Yet only Rora is expressed at appreciable levels, and only in ILC2s, consistent with its critical role in ILC2 development (Wong et al., 2012; Halim et al., 2012). In contrast, ILC2 development is not dependent on Rorg expression, and neither RORα nor RORγ plays a major role in NK cells. Taken together, this analysis confirms the known role of Rora in ILC2s and highlights how matched multiomic chromatin accessibility and gene expression data can clarify ambiguities inherent in TF enrichment analyses.

The broad effects of RAG expression on ILC2 transcriptional regulomes we observed (Figure 5C) led us to hypothesize that distinct cohorts of TFs may contribute to the differences observed between RAG^naïve and RAG^exp ILC2s. To test this hypothesis, we analyzed the open chromatin regions in GPLs unique to each RAG fatemapped ILC2 population using the FindMotifs function in Signac (Stuart et al., 2021), which returns a ranked list of enriched motifs corrected for the background presence of each motif in all cells. In both RAG^naïve and RAG^exp ILC2s, we identified enriched TF motifs (Figure 5—figure supplement 3D, Supplementary file 1, Table S11) that are GC-rich regions recognized by a large family of C2H2 zinc finger TFs, particularly the Krüppel-like factors (KLFs), which are well-established as key regulators of lymphocyte development (Hart et al., 2012; Cao et al., 2010). Given the strong sequence similarities of the identified TF motifs, we turned to the matched gene expression data to clarify which TFs may be available to engage the accessible binding sites. Of the eleven unique TFs identified, only six were detected in the gene expression assay (Figure 5—figure supplement 3E). We observed much higher expression of Klf2, Klf6, and Klf12 in Rag^exp ILC2s compared to RAG^naïve ILC2s (Figure 5—figure supplement 3E). Notably, all three of these TFs have been associated with reduced cellular proliferation and/or activation (Weinreich et al., 2009; Godin-Heymann et al., 2016; Narla et al., 2001). Klf2 expression plays a key role in T cell quiescence (Kuo et al., 1997; Buckley et al., 2001), and both Klf2 and Klf6 were recently identified as markers of ‘quiescent-like’ skin resident ILCs (Bielecki et al., 2021). In contrast, although detected in a smaller fraction of cells in our data, we found Klf7 expression was higher in RAG^naïve ILC2s compared to Rag^exp ILC2s (Figure 5—figure supplement 3E). Increased expression of Klf7 has been shown to enhance the survival of early thymocytes and is a predictor of poor outcomes in acute lymphoblastic leukemia (Schuettpelz et al., 2012). Collectively, these findings link the relatively activated or suppressed epigenomic and transcriptomic states of RAG^naïve and Rag^exp ILC2s, respectively, to distinct cohorts of homeostatic TFs.

Although our GPL and TF analyses revealed a suppressive effect of RAG expression on ILC2 gene programs, we did not account for the additional variable of disease state in the initial analysis. To test whether RAG expression promotes a suppressive epigenomic program in ILC2s that is durable in the setting of inflammation, we first recalculated GPLs after splitting our dataset by both history of RAG expression (naïve vs. experienced) and disease (steady-state vs. AD-like disease) to yield four lists of GPLs (Figure 6A; Supplementary file 1, Table S12). When we examined the intersection, or overlap, of peaks from ILC2 GPLs (Supplementary file 1, Table S13), several notable patterns emerged (Figure 6B). First, the largest set of peaks was shared by all RAG^naïve cells (gray bar), regardless of disease state, with the next largest peak sets belonging to either steady state or AD-like disease in the RAG^naïve cells. Second, there was a large set of peaks shared by all RAG^exp cells (red bar). Third, the intersections corresponding to disease states (steady state – yellow, AD-like disease – dark red), had relatively few unique peaks. These findings suggest that early exposure to RAG expression plays a larger role in modulating the epigenomic signature of the ILC2 gene program than exposure to disease. To confirm that the patterns we observed represent a specific effect of RAG expression on the ILC2 gene program, we performed the same analysis on GPL peaks for all genes in the dataset (Figure 6—figure supplement 1). In contrast to the ILC2 gene set, the majority of GPL peaks for all genes was shared among all cell populations, consistent with epigenomic regulation of most genes being minimally affected by either RAG expression or AD-like disease. Last, in the ILC2 gene set analysis, we noted a set of poised peaks shared by all RAG^naive cells and RAG^exp cells in the setting of AD-like disease, but not with RAG^exp cells at steady state (blue bar, Figure 6B). We reasoned that this condition might capture some genomic loci that are suppressed by a history of RAG expression at steady state but are induced during inflammation.

Thus, we next quantified and sorted these GPLs to generate a list of genes with the most peaks ‘induced’ during AD-like disease (Figure 6C, Supplementary file 1, Table S14). Among the identified genes, we selected Rora (Figure 6D) and Ccr6 (Figure 6E) to examine more closely for evidence of epigenomic activation in AD-like disease, given the role of these genes in ILC2 expansion (Wong et al., 2012; Halim et al., 2012) and homing to sites of inflammation (Ricardo-Gonzalez et al., 2018; Kobayashi et al., 2019), respectively. For both genes, we observed more widespread open chromatin over the genomic region in the RAG^naïve cells compared to the RAG^exp cells, but this difference was partially abolished by increased open chromatin in AD-like disease in the RAG^exp cells. Taken together, our analysis reveals that a history of RAG expression selectively modulates the activity of ILC2 gene programs across both steady state and during AD-like inflammation, while some programs are more evident at steady state given the uniquely poised nature of ILC2s.

Our functional data demonstrate a role for RAG expression in regulating ILC2 development and activation, including limiting proportions of IL5⁺ and IL-13⁺ ILC2s at steady state and in AD-like disease. Prior work identified epigenomic priming in ILC2s early in development at the Th2 locus (comprised of the Il4, Il13, Rad50, and Il5 gene loci) to enable rapid transcriptional responses during inflammation (Shih et al., 2016). Thus, we hypothesized that RAG promotes the functional observations in ILC2s by suppressing the establishment of an active regulome at the Th2 locus. To test this hypothesis, we analyzed the Th2 locus in our multiomic data in greater detail. Using a similar strategy to our analysis of ILC2 marker genes, we calculated the number of GPLs in the RAG^exp and RAG^naïve cells, respectively, for the genes in the Th2 locus. (Figure 7A, Supplementary file 1, Table S7). We found many GPLs associated with the four Th2 locus genes, including significant crosstalk between these genes, similar to previous observations (Figure 7A; Lee et al., 2003; Loots et al., 2000; Fields et al., 2004). Importantly, we identified fewer GPLs in the RAG^exp cells, particularly for the Il5 and Il13 loci (Figure 7A). As in our analysis of the ILC2 marker GPLs (Figure 5), we quantified the differences based on RAG fate mapping and found that all genes in this locus had increased GPLs in RAG^naïve cells relative to RAG^exp cells (Figure 7B; Supplementary file 1, Table S15). We also applied the same analysis strategy that identified TFs potentially mediating observed differences between RAG^exp and RAG^naïve ILC2s (Figure 5—figure supplement 3D and E) specifically to the four Th2 locus genes. Given the limited size of genomic regions (and thus open chromatin peaks) analyzed in the Th2 locus compared to all ILC2 genes, we found overall fewer enriched motifs. Strikingly, significant enrichment of TF motifs was only present in unique peaks from RAG^naïve ILC2s, while no TF motifs met the cutoff in RAG^exp cells (Figure 7—figure supplement 1A, Supplementary file 1, Table S16). These motifs primarily contained the canonical 5'-(A/T)GATA(A/G)–3' binding site recognized by the GATA family of zinc finger TFs (Lentjes et al., 2016). When we compared enriched motifs in open chromatin to gene expression of the corresponding TFs, only Gata3 was expressed at appreciable levels (Figure 7—figure supplement 1B). Critically, Gata3 expression was higher in RAG^naïve compared to RAG^exp cells, consistent with our previous analyses of the ILC2 gene regulomes (Figure 5C). Collectively, our data confirm the established role of GATA3 in mediating activation of the Th2 locus (Mjösberg et al., 2012) and are consistent with a role for RAG expression in suppressing the type 2 regulome at the Th2 locus.

We next considered the additional effects of AD-like inflammation on the Th2 epigenomic regulome using the same approach we used to analyze the ILC2 gene set in Figure 6. Again, we found the largest set of peaks was shared by the RAG^naïve cells, regardless of disease state, with the next largest peak sets belonging to either steady state or AD-like disease in the RAG^naïve cells (Figure 7C). Furthermore, there was a large proportion of peaks shared by both RAG^exp cells, consistent with a major contribution of a history of RAG expression to epigenomic modulation of the Th2 regulome. To quantify the potential effect of AD-like inflammation on reversing RAG-mediated suppression of Th2 locus genes, we mapped the 14 peaks shared by RAG^naive cells and RAG^exp cells in the setting of AD-like inflammation (i.e. only suppressed in RAG^naive cells at steady state) back to the Th2 genes via their respective GPLs (Figure 7C - blue bar, Supplementary file 1, Table S17). Interestingly, Il13, which was not identified as a top ILC2 marker in our earlier analyses, had the highest number of linked peaks associated with potential induction in AD-like disease (Figure 7D). When we examined the Il13 locus in the ILC2 cluster more closely, we found more widespread open chromatin in the RAG^naïve cells compared to the RAG^exp cells (Figure 7E). However, in the AD-like disease sample, the RAG^exp cells displayed increased open chromatin relative to the steady state, consistent with induction in the setting of inflammation, like our earlier findings for ILC2 genes such as Ccr6 (Figure 6E). Taken together, our functional data and multiomic analyses demonstrate a role for RAG expression in modulating genes critical for ILC2 development and function, including the key type 2 cytokines expressed from the Th2 locus.

Wild-type (WT) C57BL/6J and WT congenic strains (CD90.1⁺, CD45.1⁺), Rag1^-/-, and Rag2^-/- mice were initially purchased from the Jackson Laboratory and bred in-house. The RAG fate-mapping strain Rag1^Cre::Rosa26^LSL-tdRFP was originally created in the lab of Paul Kincade (Welner et al., 2009) and bred in-house. All mice were housed in specific-pathogen-free conditions in an environmentally controlled animal faculty with a 12 hr light-dark cycle and given unrestricted access to food and water at Icahn School of Medicine at Mount Sinai or Washington University School of Medicine in St. Louis. All animal protocols and experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at Icahn School of Medicine at Mount Sinai (protocol number 202200000163) or Washington University School of Medicine in St. Louis (protocol numbers 20–0017 and 23–0130).

Experiments were performed on independent cohorts of male and female mice. The sample size for animal experiments was chosen based on previous data generated in the laboratory. For induction of AD-like disease, 8- to 12-wk-old mice were treated with 2 nmol calcipotriol (MC903, Tocris Bioscience) in 10 μL of 100% ethanol (EtOH) vehicle, or vehicle alone, on the bilateral ear skin daily for 7–10 days. Body weight and ear thickness were measured daily with a digital scale and analog caliper by the same investigator. For tissue harvest, animals were euthanized by CO₂ inhalation.

Cervical skin draining lymph nodes (sdLN) were removed from the mice and immediately homogenized manually through a 100 μm cell strainer (Fisher Scientific) into a 50 mL tube with the end of a plunger from a 3 mL syringe. The strainer was washed with wash medium (2% vol/vol FBS/PBS) and the strained cells were centrifuged at 400 g for 5 min at 4 °C. Lymph node cell samples were stained with Zombie NIR viability dye (Biolegend; 1:500) to exclude dead cells, followed by Fc-receptor blocking and cell-surface staining with specific antibodies. The cells were analyzed using either LSR Fortessa (BD) or Cytek Aurora (CYTEK) flow cytometers. Data was obtained using either FACSDiva (BD) or SpectroFlo (CYTEK) software and was further analyzed using FlowJo.

After tissue harvest, ILC stimulations were performed by incubating 0.5–1×10⁶ cells for 4 hr at 37 °C in stimulation media (DMEM with 5% fetal bovine serum, 1% penicillin/streptomycin, 2 mM L-glutamine, 50 ng/mL Phorbol 12-myristate 13-acetate (PMA), 100 ng/mL ionomycin, 5 ug/mL Brefeldin A (BFA), 2 uM monensin). After stimulation, cells were washed in wash medium, fixed, and stained for surface and intracellular markers as described for unstimulated cells.

Spleens were harvested from donor WT B6 mice and immediately homogenized manually through a 100 μm cell strainer (Fisher Scientific) into a 50 mL tube with the end of a plunger from a 3 mL syringe. The strainer was washed with wash medium (2% vol/vol FBS/PBS) and the strained cells were centrifuged at 400 g for 5 min at 4 °C followed by treatment with RBC lysis buffer for 2 min and two wash steps using two volumes of wash medium. Cells were counted, and 5 million splenocytes were injected intraperitoneally into each recipient mouse. Experiments were performed 4 wk following splenocyte add-back to allow immune reconstitution.

Recipient mice were provided with antibiotic water, consisting of 5 mL of Sulfatrim (sulfamethoxazole/trimethoprim) added into 200 mL of drinking water, for 1 wk starting from the day prior to irradiation (day –1). On day 0, recipient mice were irradiated with 950 cGy using the X-RAD 320 (Precision X-Ray). BM was harvested from donor mice femurs and tibias and treated with RBC lysis buffer (Sigma-Aldrich) for 2 min. BM cells were transferred into a 15 mL conical tube through a 70 μm cell strainer (Fisher Scientific) and the cell strainer and cells were washed with 2% (vol/vol) FBS/PBS. The concentration of living cells was determined using a Cellometer Auto 2000 (Nexcelom Bioscience) with ViaStain AOPI Staining Solution (Nexcelom Bioscience). Recipient mice received the same number of cells, at 1×10⁷ live bone marrow cells per mouse, through retroorbital injection within 24 hr after irradiation. Recipients were given 8 wk for immune reconstitution after BM transplantation before experimental use.

Rag1^Cre::Rosa26^LSL-tdRFP mice were treated with 2 nmol calcipotriol (MC903, Tocris Bioscience) in 10 μL of 100% ethanol (EtOH) vehicle, or vehicle alone, on the bilateral ear skin daily for 7 days to induce AD-like inflammation. The next day, cervical sdLN were harvested and immediately homogenized manually through a 100 μm cell strainer (Fisher Scientific) into a 50 mL tube with the end of a plunger from a 3 mL syringe. The strainer was washed with wash medium (2% vol/vol FBS/PBS) and the strained cells were centrifuged at 400 g for 5 min at 4 °C. Next, cells were incubated with biotinylated antibodies (anti-mouse CD3e, CD19, CD11b; 1:300; Biolegend) in 100 μL of wash buffer for 20 min at 4 °C, followed by two washes in two volumes of wash buffer. Next, no more than 10⁷ cells were incubated with Streptavidin MicroBeads (Miltenyi) in 500 μL separation buffer (0.5% w/v BSA in PBS; BSA and PBS from Sigma) at 4 °C for 20 min, then added to LD columns (Miltenyi) pre-equilibrated with separation buffer and loaded in a QuadroMACS Separator (Miltenyi) for negative cell selection. Remaining cells were eluted in 1 mL separation buffer and cells were centrifuged at 400 g for 5 min at 4 °C, followed by resuspension in freezing buffer (10% DMSO, Invitrogen; 20% FBS in DMEM, Sigma) and slow freezing to –80 °C in a CoolCell LX (Corning) device.

Cryopreserved sdLN cells were processed as recommended by the 10X Genomics DemonstratedProtocol_NucleiIsolation_ATAC_GEX_Sequencing_RevC_(CG000365) instructions for primary cells without any modification to the protocol. Briefly, cells were thawed in a 37 °C water bath followed by dilution into media (RPMI +15% FBS, Sigma) and centrifugation at 400 g for 5 min at 4 °C. For each final sample (EtOH vehicle- or MC903-treated), cells were pooled from samples from three individual mice. Cells were resuspended in PBS +0.04% BSA (Sigma) and passed through a 40 μm Flomi strainer (Bel-art) followed by determination of cell concentration using the using Cellometer Auto 2000 (Nexcelom Bioscience) with ViaStain AOPI Staining Solution (Nexcelom Bioscience). Cells were centrifuged for 5 min at 4 °C and the supernatant was removed. Lysis Buffer (Tris HCl base with 0.1% Tween-20, 0.1% NP-40, 0.01% digitonin, 1 mM DTT, and 1 U/μL Protectors RNase inhibitor, Sigma; full recipe in 10X Genomics protocol) was added, cells mixed by pipetting 10 times, and incubated on ice for 3 min. Nuclei from lysed cells were centrifuged at 400 g for 5 min at 4 °C and washed in 1 mL Wash Buffer (Lysis Buffer, but without NP-40 or digitonin). The wash step was repeated two more times. Nuclei concentration was determined as for cell concentration using the Cellometer and ViaStain solution. The AOPI staining indicated 97–99% lysis efficiency of the cells. We manually confirmed nuclei count using a Bright-Line hemacytometer (Hausser Scientific). Nuclei were centrifuged at 400 g for 5 min and resuspended in a volume of 1 X Nuclei Buffer (10 X Genomics) to yield roughly 4000 nuclei/μL. We then immediately proceeded to the 10X Chromium Next GEM Single Cell Multiome ATAC +Gene Expression pipeline.

Multiome 3v3.1 GEX and ATAC libraries were prepared as recommended by 10X Genomics protocol Chromium_NextGEM_Multiome_ATAC_GEX_User_Guide_RevD (CG000338). For sample preparation on the 10X Genomics platform, the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Bundle, 16 rxns PN-1000283, Chromium Next GEM Chip J Single Cell Kit, 48 rxns PN-1000234, Single Index Kit N Set A, 96 rxns PN-1000212 (ATAC), Dual Index Kit TT Set A, 96 rxns PN-1000215 (3v3.1 GEX), 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. GEX libraries were pooled and run over 0.05 of a NovaSeq6000 S4 flow cell using the XP workflow and running a 28×10×10 ×150 sequencing recipe in accordance with the manufacturer’s protocol. Target coverage was 500 M reads per sample. ATAC libraries were pooled and run over 0.167 of a NovaSeq6000 S1 flow cell using the XP workflow and running a 51×8×16×51 sequencing recipe in accordance with the manufacturer’s protocol. Target coverage was 250 M reads per sample.

The cellranger-arc-2.0.0 (10X Genomics) pipeline was used to generate FASTQ files, gene expression matrices, and ATAC fragment tables for each sample, followed by aggregation using the aggr function. Default settings were utilized, with the exception that we incorporated a custom reference with the sequence for tdRFP (see Supplementary file 1, tdRFP sequence) added to the default mouse reference sequence provided by cellranger (refdata-cellranger-arc-mm10-2020-A-2.0.0). Correction for ambient RNA was performed using SoupX (Young and Behjati, 2020) with clustering information provided by the default cellranger outputs. Doublets were removed using Scrublet (Wolock et al., 2019) with default settings.

Corrected data was then processed using Signac (Stuart et al., 2021) and Seurat (Stuart et al., 2019; Hao et al., 2021; Butler et al., 2018). ATAC-seq peaks were identified using MACS2 (Zhang et al., 2008) through the CallPeaks function in Signac. Per-cell quality control metrics were computed using the TSSEnrichment and NucleosomeSignal functions, and cells retained with a nucleosome signal score <1.5, TSS enrichment score >1, total RNA counts <15,000 and>1000, total ATAC counts <75,000 and>100, percent mitochondrial reads <5%, and percent ribosomal genes detected <10%. After these filtering steps, 10,304 cells remained. Cells were further filtered by their expression of lineage-defining markers similar to the negative selection step during sample processing. Cells with detectable transcripts for Cd3d, Cd3e, Cd3g, Cd4, Cd19, Cd8a, and Itgam were removed. This left 2034 remaining cells for further analysis.

The SCTransform function of Seurat was used to normalize RNA counts. We performed integration of the two samples using the RNA assay to correct for batch effects and treatment in the initial clustering using the default parameters for the functions SelectIntegrationFeatures, FindIntegrationAnchors, and IntegrateData. The integrated data was used for PCA (25 dimensions) and UMAP reduction for the RNA assay alone. With default parameters in Signac, we used TFIDF to normalize ATAC peaks and latent semantic indexing (LSI) to reduce the dimensionality of the ATAC data. We constructed a UMAP of the ATAC data alone using the LSI reduction (dimensions 2–25). To construct a joint graph and UMAP using equal weighting from the RNA and ATAC assays, we used the FindMultiModalNeighbors function of Seurat/Signac using default parameters (RNA dimensions 1–25, ATAC dimensions 2–25). We used a resolution of 0.1 to identify clusters with the FindClusters function in Seurat/Signac. Cell types were assigned based on the manual curation of marker genes. Initially, seven clusters were identified, but two highly similar lymphocyte clusters were merged for a total of 6 cell types.

The inferred Gene Activity (GA) assay from the ATAC-seq data was calculated using default parameters of the GeneActivity function in Signac. FindAllMarkers was used to identify top markers by cluster for both RNA gene expression data (GEX) and GA, with setting adjustments including min.pct=0.20 and logfc.threshold=0.25. The differentially accessible (DA) open chromatin assay was calculated in Signac with the FindMarkers function on the ATAC-seq peaks assay (called using MACS2 as above). The differential test used was ‘LR’ (logistical regression, as suggested for snRNA-seq Ntranos et al., 2019). The total number of ATAC fragments was used as a latent variable to mitigate the effect of differential sequencing depth. Given the sparsity of the data, the min.pct parameter was set to 0.02. After identifying the top differentially accessible peaks for each cluster, the gene closest to each peak was determined using the ClosestFeature function in Signac. Results were filtered for genes within 10⁵ base pairs of the corresponding peak. The filtered gene lists were used for the ‘DA’ assay as markers of each cluster (top 25) and an expanded list for the ILC2 cluster (top 100). Venn diagrams were calculated using BioVenn/BioVennR (Hulsen et al., 2008).

Gene set enrichment analysis was performed and visualized using ClusterProfiler (Wu et al., 2021). For GSEA on steady-state ILC2 DEGs between fate mapped states, we opted to use more permissive filtering parameters instead of default parameters. We created the ranked list of DEGs using the FindMarkers function in Seurat with min.pct=0.1 and logfc.threshold=0.1. The DEG list from the GEX assay was used to generate the GSEA results. The DEG list from the GA assay did not yield any significant GSEA results. The ClusterProfiler function gseGO was used to analyze the ranked DEG list using the parameters minGSSize = 50, maxGSSize = 500, p-valueCutoff = 0.05.

A motif matrix was constructed from the ATAC data Granges using the ‘CORE’ collection and ‘vertebrates’ taxonomy group from the JASPAR2022 position weight matrix set and the mm10 reference genome. Per cell transcription factor motif activity was calculated with chromVar (Schep et al., 2017) using the motif matrix and MACS2 called peaks. Transcription factor motifs were identified in differentially accessible chromatin using the FindMotifs function in Signac.

The correlation coefficients, or gene-to-peak links (GPLs), between gene expression and accessibility of each peak, were calculated for all peaks within 10⁶ base pairs of the transcription start sites for all detected genes using the LinkPeaks function of Signac with min.cells=2. GPLs were filtered by the gene for the curated ILC2 and Th2 gene sets. Since multiple genes can be linked to one peak by GPL analysis, finding intersections of GPLs in set analysis would result in counting some epigenomic regions multiple times. Thus, for set analysis, we eliminated GPLs with redundant peaks. Then, we used each list of non-redundant peaks as input sets to generate UpSet plots and lists of intersecting peaks between states (RAG1 fate map positive or negative; AD-like disease or steady state) using UpSetR (Conway et al., 2017). Coverage plots of the single cell multiomic data, including open chromatin, peaks, and links (GPLs), were plotted using the CoveragePlot function in Signac.