T-bet and RORα control lymph node formation by regulating embryonic innate lymphoid cell differentiation

The generation of lymphoid tissues during embryogenesis relies on group 3 innate lymphoid cells (ILC3) displaying lymphoid tissue inducer (LTi) activity and expressing the master transcription factor RORγt. Accordingly, RORγt-deficient mice lack ILC3 and lymphoid structures, including lymph nodes (LN). Whereas T-bet affects differentiation and functions of ILC3 postnatally, the role of T-bet in regulating fetal ILC3 and LN formation remains completely unknown. Using multiple mouse models and single-cell analyses of fetal ILCs and ILC progenitors (ILCP), here we identify a key role for T-bet during embryogenesis and show that its deficiency rescues LN formation in RORγt-deficient mice. Mechanistically, T-bet deletion skews the differentiation fate of fetal ILCs and promotes the accumulation of PLZFhi ILCP expressing central LTi molecules in a RORα-dependent fashion. Our data unveil an unexpected role for T-bet and RORα during embryonic ILC function and highlight that RORγt is crucial in counteracting the suppressive effects of T-bet.


Introduction
Innate lymphoid cells (ILCs) comprising group 1 ILCs (ILC1 and NK cells), ILC2 and ILC3 lack rearranged antigen receptors and play an important role in regulating tissue homeostasis and mediating protective responses against pathogens, especially at mucosal surfaces. ILCs develop from the common lymphoid progenitor (CLP) under the influence of IL-2Rγc signalling and their commitment towards the ILC lineage is marked by the up-regulation of a set of genes, including the transcription factor (TF) inhibitor of DNAbinding 2 (Id2), within the ILC progenitor (ILCP) 1 . Among ILC3, α 4 β 7 + CD4 +/lymphoid tissue inducer (LTi) cells are present early in life in fetal liver (FL), spleen, intestine and lymph node (LN) anlage 2,3 , where they interact with mesenchymal stromal cells and guide development of lymphoid structures during embryogenesis 4,5 . LN formation relies on coordinated expression of lymphotoxin (LT) α/β, IL-7 receptor (IL7-R) 6 , Receptor Activator of NF-κB Ligand (RANKL, also known as TRANCE or TNFSF11) 7,8 and CXCR5 [9][10][11] , which are required for LTi cell attraction, survival and function in LN anlage. CD4 +/-LTi cells express the master transcription factor (TF) RORγt, which plays a non-redundant role for their development and thus for the generation of lymphoid tissues, as shown by lack of LN and Peyer's patches (PP) in RORγt-deficient mice 12,13 . Conversely, the pool of ILC3 expressing RORγt and emerging after birth is preferentially confined to the intestine, with few cells present in LN and spleen. Adult intestinal ILC3 are more diversified and comprise of a subset of CD4 +/-CCR6+ LTi-like cells and of CCR6 -CD4 -ILC3 expressing the natural cytotoxicity receptor NKp46. Generation of NKp46 + ILC3 requires up-regulation of ILC1-specifying TF T-bet (encoded by Tbx21) and occurs after birth, in conjunction RORγt, along with IL-22 and IFN-γ, and a tightly regulated T-bet/RORγt balance Impacts on the frequency as well as the "NK-ness" of NKp46 + ILC3, as shown in Rorc(gt) +/-, Tbx21 -/+ and Tbx21 -/mice 15,6,20 .
Whereas the antagonistic role of T-bet and RORγt in regulating the balance between ILC3 subsets and ILC1 after birth has been partially elucidated, the interplay between these two TFs during embryonic development, as well as its possible effects on the differentiation of fetal ILCs and on LN formation, has not been investigated and remains completely unknown.
Here, we report that lack of T-bet rescues LN formation in the absence of the ILC3 master TF RORγt. LN formation in mice deficient for both RORγt and T-bet is associated with the accumulation in fetal intestine and LN anlage of a RORα-dependent PLZF hi ILCP population expressing central LTi molecules. These cells are maintained in adult mice, where they produce IL-22 and promote expression of antimicrobial peptides in gut epithelium. Altogether, our data unravel a previously unrecognized role for T-bet and RORα in regulating fetal ILCP differentiation and LN development during embryonic life.
This sorting strategy enabled enrichment of the full spectrum of ILC subsets, including T-bet + ILC1, characterized by low CD127, and high CD122 expression (Extended Data fig.  1a). For simultaneous characterization of RNA and protein, we integrated oligo-conjugated antibodies into the single-cell sequencing workflow, also known as CITE-seq (Cellular Indexing of Transcriptomes and Epitopes by sequencing), which allowed us to validate ILC lineage identities with surface protein expression. After quality control, normalization and filtering out a dendritic cell-like population (displaying transcripts for Cd209a, Cd86 and Cfs1r, Extended Data fig. 1b), we analysed 1268 remaining cells with differentially regulated signatures, leading to the identification of eight distinct clusters ( Fig. 1a and Extended Data fig. 1c). The common ILC marker Id2 along with expression of Ets1, Rora, Ahr, Cxcr6 and Arg1 were detected in the majority of cells belonging to clusters 1-3 and 5-7, assigning them to the ILC lineage ( Fig. 1 b,c). Robust gene signatures, including transcripts for Gata3, Tbx21 and Rorc, enabled identification of three major superclusters: ILC2 (cluster 1 and 7), ILC1/NK (hereafter referred to as ILC1, cluster 3 and 5) and ILC3/LTi (hereafter referred to as ILC3, cluster 2 and 6). Identity of the inferred superclusters was also confirmed by the distribution of surface markers detected by oligo-conjugated antibodies, ST2, NK1.1 and CD4 (Fig. 1b). Top 50 differentially expressed genes between these clusters included Bcl11b, Il1rl1, Il17rb, Icos, Hey1, Il9r and Il4 for ILC2; Ccl5, Klrk1, Ncr1, Cxcr3, Xcl1 and Ifng for ILC1; Batf3, Batf, Tox2, Il1r1, Il23r, Nrp1, Lta and Il22 for ILC3 ( Fig. 1c and Extended Data fig. 1c) [21][22][23] . Expression of the NK cell TF Eomes was mostly negative and confined to a small subset within Tbx21 + cells, supporting the notion that clusters 3 and 5 largely comprise of ILC1 ( Fig. 1c and Extended Data fig. 1d). Transcripts for Il7r were highest in ILC3 and lowest in ILC1, which conversely showcased the most prominent expression of Il2rb transcripts, as also observed on protein level (Fig 1b,cand Extended Data fig. 1a). Within all ILC superclusters, we further detected clusters of cells (5-7) enriched in gene transcripts associated to active cell cycle, such as Mki67, Top2a and Tubb4, highlighting a substantial fraction of proliferating ILC3, ILC2 and ILC1 within the embryonic intestine ( Fig. 1a and Extended Data fig. 1c). Besides identification of classical ILC lineages, two additional clusters (4 and 8) were detected: cluster 8 had little or no expression of most ILC-related genes such as Id2, but expressed Il7r along with abundant transcripts of Flt3, Cd34, Bcl11a and Tcf3 (encoding for E2A) ( Fig. 1c and Extended Data fig. 1c), indicative of lymphocyte differentiation from hematopoietic stem cells. Moreover, in this cluster we observed few cells expressing recombination activating genes Rag1/2, Early B cell Factor 1 (Ebf1) and Nfil3 (Fig. 1c), the latter previously shown to promote early NK/ILC lineage commitment by regulating Id2 [24][25][26][27] . Altogether, this signature was compatible with the one described for common lymphocyte progenitors (CLP) 28 . Cluster 4 shared signatures with ILC precursors (ILCP), formerly identified in FL and adult bone marrow (BM) 17,[29][30][31][32][33] . This signature included surface expression of integrin α 4 β 7 and c-Kit, along with abundant transcripts for Il7r and Zbtb16 (encoding for PLZF) (Fig. 1c,d). Accordingly, surface PD-1, which identifies PLZF-expressing ILCP in the BM 32,34 , could be detected within this population and was associated with Zbtb16 transcripts ( Figure 1D). Therefore, clusters 8 and 4 are hereafter referred to as CLP and ILCP, respectively. Transition from CLP to ILCP was marked by loss of CD34, Bcl11a and Flt3 and up-regulation not only of Zbtb16, but also of Tcf7, Tox, Ikzf2, Runx3, Maf, Bcl2, Cd7, Id2, Ets1, Rora, AhR, Cxcr6 and Gata3, along with a conversion from Notch1 to Notch2 expression (Fig. 1c), as previously reported in the BM and FL 35 . Cells within the ILCP cluster largely co-expressed Zbtb16 with Tcf7, Tox and Gata3 lo , while few cells showed co-expression of Zbtb16 with Cxcr6, Tbx21, Gata3 hi or Rorc transcripts (Fig. 1e), suggesting ongoing transition from the ILCP towards the different ILC lineages. Along this line, expression of some ILC1, ILC2 or ILC3-specifying genes such as Cxcr3, Klrk1, Xcl1, Il17rb, Icos, Il4, Batf3, Batf, Nrp1 and Il1r1 could be accordingly observed within the ILCP cluster ( Fig. 1c and Extended Data fig. 1c). To model continuous developmental trajectories, we regressed out cell-cycle related genes within the proliferating clusters (5,6,7), to map these cells onto their respective ILC superclusters and applied two distinct computational models of single-cell trajectory inference, Slingshot ( Fig.  1f) and partition-based graph abstraction (PAGA) (Extended Data fig. 1e). Both methods inferred comparable trajectories with continuous cell transitions from CLP to ILCP, further branching into ILC1, ILC2 and ILC3 ( Fig. 1f and Extended Data fig. 1e). Interestingly, Slingshot analysis highlighted an early branching of the ILCP towards the ILC2 cluster, while ILC1 and ILC3 retained a common trajectory and transitional states before bifurcating into separate lineages (Fig. 1f).
Altogether, these data report transcriptional signatures of ILC lineages in the embryonic SI and identify a spectrum of progenitor populations within the isolated tissue that resemble those previously described in FL or adult BM, suggesting the intestine as an active ILC differentiation niche during embryonic development.

Fetal intestinal ILC3 are heterogeneous
In order to assess whether ILC3 from fetal SI comprise heterogeneous subsets similar to their adult counterparts, we next analysed the ILC3 supercluster in more detail ( Figure  2). Using unsupervised clustering, we identified three subclusters with defined sets of differentially regulated genes, displaying analogous Rorc expression (Fig. 2a,b). Among the top 30 genes characterizing subcluster 1, we found Tnfsf11 (encoding RANKL or TRANCE), Pdcd1 and transcripts for cytokines or chemokines such as Il22, Lif 36 , Csf2 (encoding GM-CSF) and the neutrophil chemoattractant Cxcl2 (Fig. 2c,dand Extended Data fig. 2a). In addition, Cxcr5 transcripts were also enriched in subcluster 1 (Fig. 2d). Subcluster 2 showed high expression of Il7r, Ltb, Ccr6, Batf and Cd7 ( Fig. 2c fig. 2a,b). Importantly, we also detected cells expressing transcripts for Tbx21 and Ncr1 (Fig. 2e). Notably, expression of Lta transcripts and of surface proteins CD4 and NRP-1 were comparable in subcluster 1 and 2, although also present at low levels in subcluster 3 (Fig. 2d). These data suggest that generation of specified ILC3 subsets already happens embryonically and point out that a small subset of T-bet + ILC3 is present prenatally.

RORγt + T-bet + CD4 + ILC3 transiently emerge prenatally
In order to validate our scRNA-seq data by flow cytometry and to understand when Tbet + ILC3 emerge during embryonic development, we took advantage of T-bet-ZsGreen reporter mice and performed a kinetic analysis of T-bet and RORγt expression by LinLD -CD45 + CD122 + and/or CD127 + cells derived from the developing small intestine (SI) or FL. T-bet-ZsGreen reporter signal was validated by intracellular staining of T-bet, showing full co-expression of T-bet and ZsGreen fluorescent proteins (Extended Data fig. 3a). Both ILC3 and ILC1 were present in the SI of E14.5 mice, with ILC3 persistently dominating over ILC1 throughout embryonic development until 4 weeks of age ( Fig. 3a and Extended Data fig. 3b). Eomes was only expressed on a small subset of T-bet + cells ( Fig. 3 b and Extended Data fig. 3c and e), in line with E18.5 scRNA-seq data. Although percentage of LinLD-CD45 + CD122 + and/or CD127 + cells was lower in FL compared to SI ( Fig. 3a and Extended Data fig. 3d), ILC3 and ILC1 could be observed at similar frequency in FL at E14.5, with percentage of ILC3 dropping around E16.5, concomitant with ILC1 increase (Extended Data fig. 3b). By zooming Into RORγt+ ILC3 we observed that, besides CD4 + T-betcells, around 25% of E14.5 intestinal or FL ILC3 expressed T-bet, and a significant fraction co-expressed CD4 ( Fig. 3a and Extended Data fig. 3d). A similar distribution of ILC3 subsets according to T-bet and CD4 was also observed in the E14.5 mesenteric LN (mLN) anlage (Extended Data fig. 3f). Further characterization of intestinal and liver ILC3 subsets showed high surface expression of CD127, NRP-1 and CXCR5 in the CD4 + T-betsubset, with intermediate levels in CD4 + T-bet + and lowest expression in CD4 -T-bet + cells. All ILC3 subsets expressed high levels of α 4 β 7 ; conversely, CD90 was enriched in T-bet + compared to T-bet -ILC3 ( Fig. 3b and Extended Data fig. 3e). Over time, the percentage of CD4 -T-bet + ILC3 increased gradually in the intestine, while CD4 + T-betand CD4 + T-bet + ILC3 decreased from E14.5 to 4 weeks of age (Fig. 3, c and d). Notably, regardless of T-bet expression, NKp46 was not detected on the surface of ILC3 during embryonic development and appeared only in newborn mice (Fig. 3c), in line with previous findings 14 . CCR6 was selectively expressed by a small fraction of intestinal CD4 + T-bet -ILC3 cells by E14.5 (Fig. 3b), while its expression increased until E18.5, when CCR6 + CD4 -ILC3 first appeared (Extended Data fig. 3g). Over time, there was a reduction in the frequency of CCR6-CD4 + ILC3, which were virtually absent postnatally, when CCR6 + CD4 +/became the predominant ILC3 populations (Extended Data fig. 3g).
Our data showed that CD4 + Tbet + cells represented a consistent proportion of intestinal ILC3 at E14.5, in contrast to 4-week-old mice. As their frequency was already strongly reduced at birth, we analysed newborn fate map Rorc(gt) cre × R26R eYFP (RORγt-FM) or Tbx21 cre × R26R eYFP (T-bet -FM) reporter mice to test whether CD4 + T-bet + ILC3 might have undergone lineage reprogramming already during embryonic life. Indeed, a population of "ex-ILC3" which has switched off expression of RORγt, likely in response to inflammatory signals such as IL-12, has been previously described in the adult intestine 15,17,[41][42][43] .
Among T-bet + RORγtcells, we found a population of NK1.1 + RORγt-FM + ILC1 in both liver and intestine, showing that ex-ILC3 are not only generated postnatally in response to environmental or inflammatory signals, but are already present at birth ( Fig. 3e and Extended Data fig. 3h). Reciprocally, we also detected a small but consistent fraction of T-bet -FM + cells among RORγt + CD4 + ILC3 in the intestine of newborn mice (Fig. 3f), suggesting a history of T-bet expression in these cells.
In summary, we found that T-bet is expressed by subsets of intestinal ILC3 already during embryonic development and identified a population of T-bet + CD4 + ILC3, which progressively decrease over time, revealing previously unappreciated prenatal ILC plasticity.

Lack of T-bet rescues formation of LN in RORγt-deficient mice
Following up our finding of T-bet expression in fetal ILC3, we next investigated whether T-bet affected embryonic ILC3 functions, namely LN organogenesis, by analysing Tbx21 -/-× Rorc(gt) GFP/wt mice. Rorc(gt) GFP/wt reporter mice and Rorc(gt) GFP/GFP mice (lacking RORγt) were used as control. Whereas RORγt deficiency in Rorc(gt) GFP/GFP mice led to impairment of peripheral LN formation (Fig. 4, a and b), as previously reported 12,13 , LN generation in adult Tbx21 -/-× Rorc(gt) GFP/wt mice was largely normal (Extended Data fig. 4, a -c) 16 . It is well established that T-bet and RORγt are co-expressed within adult ILC3 where they cross-regulate each other 15,16,20 . in order to investigate the relevance of both TFs for fetal ILC3 development, we generated double knockout mice. Surprisingly by IgD + naive B cells. GCs displayed a discrete segregation of dark and light zone, as shown by staining follicular dendritic cells in the latter one with CD35 ( Fig. 4c). Flow cytometric analysis of mLN from Rorc(gt) GFP/GFP × Tbx21 -/mice further identified similar numbers of CXCR5 + PD1 + T follicular helper (Tfh) cells expressing ICOS and Bcl-6 as well as IgD -/lo CD38 -Fas + GL7 + GC B cells, as compared to control mice, pinpointing functional competence of mLN in these mice (Fig. 4d). In order to rule out a possible role for T or B cells in driving LN development in the absence of T-bet and RORγt, we analysed Rorc(gt) GFP/GFP × Tbx21 -/-× Rag2 -/mice and observed mLN development in all mice analysed (Fig. 4, e and f). in sum, our data show that in the presence of RORγt, T-bet does not influence embryonic LTi cell functions. in the absence of RORγt however, T-bet inhibits LN formation in a T cell-and B cell-independent manner as evidenced by rescue of LN in mice double deficient for RORγt and T-bet.

ILCP accumulate due to impaired ILC1/3 differentiation
To investigate whether LN organogenesis in Rorc(gt) GFP/GFP × Tbx21 -/mice was associated with the existence of an ILC population that retained LTi cell functions, we performed single-cell RNA sequencing on LinLD -CD45 + CD122 + and/or CD127 + ILCs Isolated from E18.5 SI of Rorc(gt) GFP/GFP × Tbx21 -/-(DKO), as compared to Rorc(gt) GFP/GFP (RKO) or Rorc(gt) GFP/wt (reporter) mice. Combined analysis of equal numbers of cells from the three mouse strains revealed the presence of overlapping ILC clusters with comparable gene expression signatures in RKO and DKO as in reporter mice, although with clearly different distributions ( Fig. 1a and 5a-b). As expected, ILC3 populations were drastically diminished in both RKO and DKO mice due to RORγt deficiency. Lack of ILC3 was paralleled by a conspicuous increase of ILC2 in both RKO and DKO strains and of ILC1/NK selectively in RKO, highlighting competitive cross-regulation between the different ILC lineages (  fig. 5d), as also seen in reporter mice. Accordingly, PLZF hi PD1 hi ILCP were detected by flow cytometry among LinLD -CD127 + and/or CD122 + cells derived from E18.5 SI and mLN anlagen and their frequency and absolute number were increased in DKO and RKO, as compared to reporter mice ( Fig. 5d and e). ILCP accumulation in DKO mice was not associated to increased proliferation, as Ki67 expression was comparable in ILCP from the three strains (Extended Data fig. 5e and f).
These data suggest that under unperturbed conditions, there is continuous ILCP differentiation towards the mature ILC lineages, as shown in reporter mice; on the other hand, in the absence of RORγt and T-bet, ILCP cannot further differentiate towards ILC1 or ILC3 lineages and therefore accumulate at this differentiation stage.

ILCP of DKO mice are enriched in cells with LTi signatures
We next examined how deficiency of RORγt and T-bet affects ILCP transcriptional profile and differentiation commitment by analysing the enrichment score of DKO, RKO and reporter ILCP for ILCP, ILC1, ILC2 or ILC3 gene modules (Fig. 6a). We found that RKO ILCP were enriched in ILC1 and ILC2, but not in ILC3 module (Fig. 6a), as expected from RORγt deficiency. Accordingly, ILCP cells expressing Tbx21 transcripts were more abundant in RKO than of reporter mice (Extended Data fig. 6a), suggesting that in the absence of RORγt counterbalance, ILCP maturation is skewed towards ILC1 lineage. Conversely, in DKO mice additional lack of T-bet resulted in an increased frequency of ILCP with a higher enrichment score for ILC3 (Fig. 6b). Interestingly, DKO ILCP with high enrichment for ILC3 genes exhibited low enrichment score for the ILCP gene module ( Fig.  6a and 6b), suggesting that lack of T-bet facilitates progression of ILCP towards the ILC3 lineage. We then analysed which among the top 100 genes defining the ILC3 core module were upregulated in ILCP from DKO mice (Extended Data fig. 6b). DKO ILCP displayed no or only little expression of Il1r1, Il17re, Nrp1 or Ccr6, similar to their ILCP counterparts from reporter or RKO strains (Extended Data fig. 6b and c). In contrast, besides Gpr183 and Batf3 and Bcl2 (Fig. 6c), an increased fraction of DKO ILCP expressed transcripts for genes associated with LTi functions, namely Lta, Ltb, Cxcr5 and Tnfsf11 (encoding RANKL) (Fig.   6d). Of note, although lack of ILC3 advocated for a loss of RORγt-driven functions in both RKO and DKO mice (Fig. 5, a and b), residual Rorc transcripts could still be observed within the ILCP from RKO and DKO strains (Extended Data fig. 6b and d).
To validate the data obtained by scRNA-seq, we performed flow cytometry of LinLD -CD127 + and/or CD122 + cells derived from E18.5 SI and mLN anlagen. A population of CXCR6 + cells expressing surface LTi functional molecules, namely RANKL, CXCR5 and the heterotrimer LTα 1 β 2 , was present in the SI and mLN of DKO mice at comparable higher frequencies than in reporter mice where CXCR6 + RANKL + CXCR5 + LTα 1 β 2 + cells largely comprised of ILC3 ( Fig. 6e and f). Besides significant differences in frequency, the MFI of RANKL or CXCR5 was slightly increased in DKO versus RKO ILCP (Extended Data fig. 6e). In line with detection of Rorc transcripts by scRNA-seq, a consistent GFP signal indicating Rorc promoter activity was found in CXCR6 + cells from DKO and at lower frequencies in RKO mice (Extended Data fig. 6f). In support of the scRNA-seq data, DKO CXCR6 + cells expressed high levels of PLZF, confirming their ILCP identity (Fig. 6g). Thus, both scRNA-seq and flow cytometric analysis identified an ILCP population with LTi signatures in SI and mLN anlage of DKO mice. The frequency of CXCR5 + RANKL + ILCs derived from FL of DKO was also higher as compared to RKO or reporter mice, although generally lower than in SI and mLN anlage ( Fig. 6e and Extended Data fig. 6g). This shows that skewed ILCP differentiation towards acquisition of LTi signatures in DKO mice seems to occur already in the FL (Extended Data fig. 6g). To validate this concept, we performed in vitro differentiation assays culturing FL progenitors derived from the three strains on OP9 feeder cells in the presence of SCF and IL-7. After one week culture, all ILC lineages could be generated in vitro using progenitors from reporter mice ( Fig. 6h and Extended Data fig.  6h). in line with the data observed ex vivo, ILC2 differentiation was enhanced in cultures from RKO and DKO mice, while generation of NK1.1 + cells (marking both ILC1 and NK cells) was reduced in DKO as compared to RKO mice (Extended Data fig. 6h). Importantly, cultures from DKO, but not from RKO, progenitors enabled the generation of a population of RANKL + CXCR5 + cells, which co-expressed PLZF (Fig. 6h).
Altogether, these data imply that lack of RORγt prevents the development of fully mature ILC3; however, in the absence of T-bet, it still enables the differentiation of ILCP with selected LTi features.

PLZF hi CXCR6 + GFP + cells persist in adult DKO mice
The identification of ILCP with LTi features in embryos of RORγt and T-bet double deficient mice prompted us to investigate whether such a population would persist into adulthood. PLZF hi CXCR6 + cells were present in the mLN of 4 week-old DKO (Fig. 7a). In line with the data of the E18.5 SI, a GFP signal could be detected in mLN and SI of adult DKO, as well as in SI of adult RKO mice ( Fig. 7b and Extended Data fig. 7a). Interestingly, in contrast to GFP lo cells (present in both RKO and DKO mice), GFP hi ILCs were absent in the SI of RKO, but detected in similar frequencies in DKO and reporter mice and partially expressed CD4, but not NKp46 (Fig. 7b) (Fig. 7c). Importantly, Il22, but not Il17a, transcripts were restored to similar amounts in DKO CD4 + GFP + cells as in NKp46 + ILC3 from reporter mice ( Fig. 7c and e). Along this line, IL-22, but not IL-17A, protein was detected in DKO CD4 + GFP + cells and was sufficient to rescue expression of Reg3g and Reg3b encoding antimicrobial peptides in the intestinal epithelium (Fig. 7f).
These data show that in the absence of T-bet and RORγt, PLZF hi ILCP with LTi signatures and residual Rorc promoter activity develop during embryonic life and persist into adulthood. Based on these data, we next asked whether transfer of DKO fetal progenitors could give rise to such a population in adult mice. To this end, we adoptively transferred FL progenitors from reporter, RKO or DKO CD90.1 + CD45.2 + mice into irradiated CD90.2 + CD45.1 + Rag2 -/-Il2rg -/hosts. Irradiated recipient mice were treated bi-weekly with anti-CD90.2 antibody in order to deplete residual radio-resistant host ILCs [44][45][46] and analysed 4 weeks after reconstitution (Fig. 7g). In line with the phenotype observed ex vivo in adult DKO mice, only DKO progenitors efficiently reconstituted a population of GFP hi cells at similar frequency to the one observed in reporter mice (Fig. 7h).
Collectively, we have shown that deficiency of T-bet and RORγt enables the generation and accumulation of embryonic PLZF hi ILCP, which can persist into adulthood. Lack of T-bet facilitates ILCP expression of molecules conferring LTi function and IL-22, thereby rescuing LN formation during embryogenesis and promoting the production of antimicrobial peptides in the intestinal epithelium postnatally.

RORα promotes LN development in DKO mice
As we observed residual Rorc transcriptional activity as well as GFP lo/hi expression in both RKO and DKO ILCs, respectively, we further validated transcript and protein expression of RORγt in Rorc(gt) GFP/GFP and Rorc(gt) GFP/GFP × Tbx21 -/mice, in which the ATG in front of Rorc exon 1γt is replaced by GFP gene insertion disrupting the transcriptional start site (TSS) (Fig. 8a). We confirmed absence of exon 1γt to 3 junction transcripts, while residual transcriptional activity of exon 5 to 6 junction could be detected in intestinal ILCs from both RKO and DKO mice (Extended Data fig. 8a and b), along with expression of a protein stained by commonly used anti-RORγ(t) antibodies (Extended Data fig. 8c). We also confirmed that Rorc exons 1 and 2 (encoding RORγ) were not transcribed in ILCs from DKO mice (Extended Data fig. 8a). Despite these surprising findings, thymic disruption of T cell development typically associated to RORγt deficiency 12,47,48 was observed in RKO, and could not be rescued in DKO mice (Extended Data fig. 8d), indicating that the hypomorphic RORγt protein is not functional either in RKO or in DKO mice.
To rule out that LN formation observed in DKO mice was due to residual function of this truncated protein, we validated our findings by using alternative models of RORγt deficiency (Fig. 8a). To this aim, we crossed Rorc(gt) Δex4  PLZF hi ILCP were enriched in mLN of both additional mouse strains lacking RORγt and T-bet (Fig. 8d,e), reminiscent of our observations in DKO versus reporter mice. Thereby, our data indicate that, in the absence of T-bet, LN development and accumulation of PLZF hi ILCP with LTi signatures occurs in the absence of RORγt TSS ATG, DBD, hinge and LBD regions, demonstrating that acquisition of LTi functions, but not full ILC3 maturation or thymic T cell development, can happen independently of RORγt.
Finally, to assess whether other transcription factors could promote acquisition of LTi signatures and LN development in the absence of RORγt and Tbet, we analysed the role of RORα, which cooperatively regulates Th17 genes together with RORγt 49-51 . RORa was highly expressed in ILCP from all mouse strains ( Figure S8F) and its additional deletion, as shown by crossing Il7r cre/wt × Rorc(gt) fl/fl × Tbx21 fl/fl mice with Rora fl/fl mice, resulted in complete lack of LN ( Figure 8F), as well as of intestinal PLZF hi ILCP (Fig. 8g).

Discussion
ILCs are among the first lymphocytes to appear during ontogeny to seed organs and tissues. While adult intestinal ILCs have been investigated also at a single cell level 23,52 , our scRNA-seq analysis of LinLD -CD127 + and/or CD122 + cells from fetal intestine enabled us identifying the entire spectrum of mature ILCs, including ILC3, ILC2 as well as a consistent pool of ILC1 residing at this site, and characterizing their global transcriptional signatures. Besides mature ILC subsets, the embryonic intestine also holds a range of hematopoietic progenitors, with signatures compatible with the ones observed in CLP and ILCP derived from FL or adult BM. By using Arg1 reporter mice, Locksley and colleagues have previously described a population of Arg1 + ILCP in fetal intestine 53 . As the ILCP cluster emerged in our scRNA-seq data set is highly enriched in Arg1 transcripts, these progenitors might largely overlap. CLP and ILCP populations have also been recently described in peripheral LN anlagen at E13.5 and E14.5 54 , suggesting that FL CLP and ILCP might seed these sites and differentiate in situ during embryonic development.
Remarkably, we detected certain heterogeneity within the transcriptome of embryonic E18.5 intestinal ILC3, which does not follow the canonical separation between LTi 0 and LTi 4 , as CD4 expression does not resolve clustering of different fetal ILC3 subsets. Most strikingly, we observed the appearance of an unappreciated T-bet + ILC3 subset before birth. It was previously shown that NKp46 + T-bet + ILC3 emerge postnatally with colonization of the intestine by microbiota 14 . While our data confirm the prenatal absence of ILC3 expressing NKp46 protein, our T-bet analyses evidently demonstrate that ILC3 expression of T-bet transcripts and protein starts already as early as E14.5 during embryonic development. Interestingly, we additionally found a T-bet + RORγt + CD4 + population displaying some phenotypic similarities with classical LTi cells, which is virtually absent in the adult intestine. Together with the observed asynchronous expression of CCR6 and CD4 on fetal ILC3 populations, these data indicate that the embryonic intestine already holds remarkable ILC3 heterogeneity, but showcases a different spectrum of ILC3 subsets as compared to adult intestine, possibly suggesting diverse waves of ILC3 differentiation. Contrary to the prevailing view, our data indicate that acquisition of T-bet and generation of plastic transcriptional states by loss of RORγt or T-bet can occur already prenatally and might result from cellular interactions or local cytokine milieus in the developing tissue.
With the intention to delineate the role of T-bet in ILC3 lineage specification during embryogenesis, we observed that lack of T-bet restored development of peripheral LN, typically absent in RORγt-deficient mice. This phenotype could be observed in three independent mouse models, in which the Rorc locus was differentially disrupted. Notably, in Rorc(gt) GFP/GFP × Tbx21 -/and to a lower frequency in Rorc(gt) GFP/GFP mice (the latter being the most commonly used model of RORγt deletion for immunological studies in the last fifteen years), we observed the presence of intestinal ILCs displaying residual low to high expression of GFP as well as of a protein reacting with commercially available antibodies, whose specific epitope however has not been characterized. Although the significance and function of this protein remains to be determined, in our study we could exclude that it plays a role in LN restoration, as the same phenotype could also be observed in II7r cre/wt x Rorc(gt) fl/fl × Tbx21 fl/fl mice. In these mice, all functionally relevant regions for RORγt activity, namely the DBD, hinge and LBD, are deleted in Il7r + cells and no residual protein could be detected. Notably, LN formation in all double knockout strains was not associated with the appearance of ILC3, but rather with the accumulation of an ILCP population expressing PLZF. Numbers of PLZF hi PD1 hi ILCP were increased in DKO as compared to reporter mice, in which ILCP might continuously differentiate towards the mature ILC lineages. Accumulation of PLZF hi ILCP in the absence of RORγt and T-bet was not linked to increased proliferation, but might possibly be sustained by the anti-apoptotic molecule Bcl2, whose transcript expression was significantly elevated in ILCP from DKO mice.
By using Zbtb16 GFPcre FM mice, It was previously shown that PLZF transiently marks a FL and BM ILCP, which has lost LTi potential, while mature ILCs are largely negative for PLZF 33 . Our data confirm high enrichment of Zbtb16 transcripts also in ILCP from embryonic intestine and mLN anlage 55

Tissue dissociation
Single cell suspensions from intestinal lamina propria of adult animals were obtained as described before 67 . In short, intestines were isolated and PPs were removed. Tissue was cleaned, longitudinally opened and chopped into pieces of 2 cm, followed by two rounds à 15 min at 37°C of epithelial cell dissociation using Hank's balanced salt solution without calcium and magnesium (HBSS-/-, Gibco) supplemented with 2% FCS, 10 mM HEPES buffer pH 7,5 (Sigma), 1 mM DTT (ThermoFisher) and 5 mM EDTA (Sigma). After washing with HBSS-/-, tissues were digested using a lamina propria dissociation kit (Miltenyi) according to the manufacturer's instructions. Lymphocytes were further enriched on a 40%/80% Percoll (GE Healthcare) gradient.
Small intestines from fetal mice were isolated from embryos under a dissecting microscope. Surrounding mesenteric tissue was removed under further magnification and cut open longitudinally. Tissue was incubated for 30 min at 37°C under agitation in HBSS containing calcium and magnesium (HBSS + / + ), 2% FCS, 10 mM HEPES, 0.2 U/mL Liberase DL (Sigma) and 50 μg/mL DNasel (Sigma). Leftover tissue was dissociated using a 22G needle and cell suspension was passed through a 70 μm strainer and washed in PBS/BSA. E14.5 and E16.5 time points were pooled samples; E18.5 intestines were treated separately.
Fetal liver was isolated under a dissecting microscope and a single cell suspension was generated by vigorously pipetting using a 1000 μL pipette. Cell suspension was filtered through a 70 μm strainer, washed in PBS/BSA followed by erythrocyte lysis.
Embryonic lymph node anlagen from the mesenteric region were dissected under a microscope or isolated from adult animals and passed through a 70 μm strainer.

Flow cytometry
Cells were stained using the following antibodies: Flow cytometry was performed using a LSR Fortessa and BD FACSymphony flow cytometer (BD Biosciences). Data were further analysed with FlowJo Software v10 (Flow Jo). All flow cytometry analysis was performed according to the Guidelines for the use of flow cytometry and cell sorting in immunological studies 69 .

in vitro cell culture
Single-cell suspensions from E14.5 fetal liver were enriched for CD45 + cells using magnetic activated cell sorting, followed by staining with fluorochrome-conjugated antibodies and sorted on BD FACSAria II (BD Bioscience) as live Lin-CD45 + Flt3 -CD127 + α 4 β 7 + PD1cells. Sorted lymphocytes were subsequently resuspended in IMDM (Gibco BRL Life Technologies) supplemented with 10% FCS, 100 U/mL penicillin, 0.1 mg/mL streptomycin, 50 μM β-mercaptoethanol (Gibco BRL Life Technologies) and each 20 ng/mL recombinant mouse IL-7 and SCF. Cells were plated onto 70% confluent OP9 stromal cells in round bottom 96-well plates and medium was renewed every three days by replacing half of the media. Cells were analysed by flow cytometry after 5-8 days of culture.

Lymph node enumeration
Percentages of mice with LN were calculated for each LN type using the total number of mice with the respective LN found in a given strain relative to the maximal number of mice with LNs of that type found in Rorc(gt) GFP/wt or Il7r cre control mice.

Immunofluorescence staining
Lymph nodes were fixed for 4 h in 4% paraformaldehyde and dehydrated via a sucrose gradient for 24 h, embedded in O.C.T. compound and frozen in 2-Methylbutane within ethanol rinsed dry ice bath. Frozen lymph nodes were cryotome-cut into 7 μm sections and placed on microscope slides in pairs. After thawing and PBS-rehydration of slides, primary staining was carried out after 20 min blocking with 10% goat serum in PBS/ BSA-0.1%Tween20 (P-B-T). DAPI, anti-igD-AF594, PNA-bio and rat-anti-mouse-CD35 were incubated for 1 h in P-B-T. Secondary staining was performed after washing in P-B-T by 30 min incubation with SA-A555 and anti-rat-IgG-A647 in P-B-T. Samples were finally washed in PBS and covered with mounting medium and cover slip and before image acquisition was performed on Zeiss LSM 880 confocal microscope.

Quantitative real-time PCR
mRNA was isolated from small intestinal intraepithelial cells by using NucleoSpin® RNA (Macherey-Nagel) and following the manufacturer's protocol. RNA integrity was assessed on a BioAnalyzer (Agilent) and RNA samples with a RIN score ≥ 7 were further transcribed into first strand cDNA by using Reverse Transcription Reagents (Applied Biosystems). qPCR reactions were assayed in triplicates per sample by using a StepOne Plus real-

RNA sequencing and data analysis
RNA of cells was extracted using RNeasy Plus Micro Kit (Qiagen) from sorted small intestinal lamina propria cells using the following markers LinLD -CD45 + CD3-RORγt-GFP + expressing either NKp46 or CD4. RNA quality was assessed by an Agilent 2100 Bioanalyzer. SMART-SeqII (ultra-low input RNA) libraries were prepared using Nextera XT DNA sample preparation kit (Illumina). Sequencing was performed on an Illumina HiSeq4000 platform, PE100. For heatmaps, normalized RNA-seq data z-score transformed were plotted using ggplot2 R package. RNA-seq datasets analysed are publicly available in Gene Expression Omnibus repository with the accession number GSE161439.

10x Genomics Chromium single-cell RNA-sequencing
Fetal small intestinal lymphocytes were isolated as described above and sorted as LinLD -CD45 + CD122 + and/or CD127 + . A total of 7 embryos from one pregnant Rorc(gt) GFP/wt female, 18 embryos from two pregnant Rorc(gt) GFP/GFP females and 19 embryos from three pregnant Rorc(gt) GFP/wt × Tbx21 -/females were used. Samples were furthermore labelled with antibody-derived tags against specific extra-cellular targets and genotypes hash-tagged with TotalSeq™-A antibodies (BioLegend) following the manufacturer's protocol. A total of 8000 cells were loaded in the Chromium™ Controller for partitioning single cells into nanoliter-scale Gel Bead-In-Emulsions (GEMs). Single Cell 3' reagent kit v3.1 was used for reverse transcription, cDNA amplification and library construction of gene expression libraries (10x Genomics) according to the manufacturer's instructions. TotalSeq™-A libraries were prepared following the protocol for 10x Single Cell 3' Reagent Kit v3.1 provided by BioLegend, including primer sequences and reagent specifications. Concentrations of all libraries were quantified using a QubitTM 2.0 Fluorometer (ThermoFisher) and quality was tested on a 2100 Bioanalyzer with High Sensitivity DNA kit (Agilent). Sequencing was performed using the NextSeq500/550 HighOutput Kit v2.5 (150 cycles) on a NextSeq500 sequencer (Illumina). scRNA-seq datasets analysed are publicly available in Gene Expression Omnibus repository with the accession number GSE161441.

scRNA-seq data processing
Demultiplexing and alignment to the mm10 reference transcriptome was performed with Cell Ranger v3.0.2 70 . Antibody-and hashtag-barcodes were counted with CITE-seq-Count 71 and normalized by centred log ratio transformation with Seurat v3.0.2 72 , which was also used for further data processing. Genotypes were assigned based on normalized hashtag counts. Cells were filtered by number of transcripts, genes, percent mitochondrial genes and for cells present within transcriptome, antibody and hashtag libraries. Normalized expression values were generated with scran v1.16.0 73 , followed by scaling and principal component (PC) analysis. PC loadings and explained variance were visualized to determine PCs used for dimensionality reduction by uniform manifold approximation (UMAP) 74 and Seurat clustering. Wilcoxon test was applied for differential gene expression analysis of genes expressed by ≥10% of cells in a cluster and with a log fold change of ≥0.25. Lineage structures on UMAPs were analysed with Slingshot v 1.4.0 75 , giving the starting (CLP) and end populations (ILC1/2/3) as input. Partition-based graph abstraction (PAGA) using the Fruchterman & Reingold algorithm was performed within Scanpy v1.4.6 76 after import of preprocessed Seurat objects into Python with anndata2ri 77 . scRNA-seq data analysis was performed with R v4.0.0 and Python v3.7.6.

Statistical analysis
No statistical method was used to pre-determine sample sizes and no randomization was performed. Data collection and analysis were not performed blind to the conditions of the experiments. Kruskal-Wallis test with Dunn's multiple comparison correction was employed for statistical analysis of datasets comparing three groups. Two-tailed Mann-Whitney test was performed for statistical analysis of datasets where two groups were compared. Statistical analyses were performed with Prism 5.04 (GraphPad Software) using a confidence level of 0.95, and P values >0.05 were considered not significant, P ≤ 0.05 were considered as *, P ≤ 0.01 were considered as **, P ≤ 0.001 were considered as ***, P ≤ 0.0001 were considered as ****         Viable Lin(CD19, CD3, CD5, F4/80, FceRIa, Gr-1) -CD45 + cells expressing IL-7 receptor (CD127) and/or the IL-2 receptor subunit beta (CD122) isolated from the small intestine (SI) of E18.5 Rorc(gt) GFP/GFP (RKO) and Rorc(gt) GFP/GFP × Tbx21 -/-(DKO) embryos were sort-purified by flow cytometry, and a single-cell expression library was generated using 10x Genomics. a, UMAP dimensional reduction projection identifies eight distinct clusters.