Canonical NOTCH signaling controls the early progenitor state and emergence of the medullary epithelial lineage in fetal thymus development

Thymus function depends on the epithelial compartment of the thymic stroma. Cortical thymic epithelial cells (cTECs) regulate T cell lineage commitment and positive selection, while medullary (m) TECs impose central tolerance on the T cell repertoire. During thymus organogenesis, these functionally distinct sub-lineages are thought to arise from a common thymic epithelial progenitor cell (TEPC). The mechanisms controlling cTEC and mTEC production from the common TEPC are not however understood. Here, we show that emergence of the earliest mTEC lineage-restricted progenitors requires active NOTCH signaling in progenitor TEC and that, once specified, further mTEC development is NOTCH-independent. In addition, we demonstrate that persistent NOTCH activity favors maintenance of undifferentiated TEPC at the expense of cTEC differentiation. Finally, we uncover a direct interaction between NOTCH and FOXN1, the master regulator of TEC differentiation. These data establish NOTCH as a potent regulator of TEPC and mTEC fate during fetal thymus development and are thus of high relevance to strategies aimed at generating/regenerating functional thymic tissue in vitro and in vivo.


Introduction
In the thymus, thymic epithelial cells (TECs) are the essential stromal component required for T lymphocyte development (Manley et al., 2011;Ritter and Boyd, 1993). Two functionally distinct TEC subsets, cortical (c) TECs and medullary (m) TECs, exist and are found in the cortex and the medulla of the organ respectively. Thymocytes migrate in a highly stereotypical fashion to encounter cTECs and mTECs sequentially as T cell differentiation and repertoire selection proceeds (Anderson and Takahama, 2012;Klein et al., 2014). cTECs and mTECs originate from endodermal progenitor cells (thymic epithelial progenitor cells; TEPCs), that are present in the thymic primordium during its initial generation from the third pharyngeal pouches (3PPs) (Gordon et al., 2004;Le Douarin and Jotereau, 1975;Rossi et al., 2006). Several studies have shown that, during development, both cTECs and mTECs arise from cells expressing markers associated with mature cTECs, including CD205 and β 5t (Baik et al., 2013;Ohigashi et al., 2013), while clonal analyses have shown that a bipotent TEPC can exist in vivo (Bleul et al., 2006;Rossi et al., 2006). Based on these observations, a serial progression model of TEC differentiation has been proposed (Alves et al., 2014). This suggests that fetal TEPCs, which exist as a transient population, exhibit features associated with the cTEC lineage and that additional cues are required for mTEC specification from this common TEPC. Identification of cTEC-restricted sub-lineage specific progenitor TECs in the fetal thymus has proved elusive, due to the shared expression of surface antigens between this presumptive cell type and the presumptive common TEPC (Alves et al., 2014;Baik et al., 2013;Shakib et al., 2009), although cTEC-restricted progenitors clearly exist in the postnatal thymus (Ulyanchenko et al., 2016). In contrast, the presence of mTEC-restricted progenitors has been detected from day 13.5 of embryonic development (E13.5) (Rodewald et al., 2001). In the fetal thymus, these mTEC progenitors are characterised by expression of Claudin3/4 and SSEA1 (Hamazaki et al., 2007;Sekai et al., 2014). Receptors leading to the activation of NFκB pathway, including LTβR and RANK, are known to regulate the proliferation and maturation of mTEC through crosstalk with T cells and tissue inducer cells (Boehm et al., 2003;Hikosaka et al., 2008;Rossi et al., 2007) and, recently, a hierarchy of intermediate progenitors specific for the mTEC sub-lineage has been proposed based on genetic analysis of NFκB pathway components (Akiyama et al., 2016;Baik et al., 2016). Additionally, HDAC3 has emerged as an essential regulator of mTEC differentiation (Goldfarb et al., 2016), and a role for STAT3 signaling has been demonstrated in mTEC expansion and maintenance (Lomada et al., 2016;Satoh et al., 2016). Despite these advances, the molecular mechanisms governing the emergence of the earliest cTEC-and mTEC-restricted cells in thymic organogenesis are not yet understood (Hamazaki et al., 2007).
NOTCH-signaling has been extensively studied in the context of thymocyte development (Shah and Zuniga-Pflucker, 2014), and is also implicated as a regulator of TECs. Mice lacking the Notch ligand JAGGED 2 showed reduced medullary areas (Jiang et al., 1998a;Jiang et al., 1998b), while B cells overexpressing another Notch ligand, Delta like 1 (DLL1), induced organized medullary areas in a reaggregate fetal thymic organ culture (RFTOC) system (Masuda et al., 2009). In contrast, in adult thymic epithelium NOTCH activity appeared to reside in cTECs, while its TEC-specific overexpression reduced TEC cellularity and led to an imbalance between mature and immature mTECs, suggesting that NOTCH signaling might inhibit mTEC lineage development (Goldfarb et al., 2016). Overall, these results suggest that NOTCH has complex effects in TECs, but the stage(s) at and mechanism(s) through which NOTCH influences TEC development have not yet been determined.
We have addressed the role of NOTCH signaling in early TEC differentiation using loss-and gain-of-function analyses. Our data establish, via genetic ablation of NOTCH signaling in TECs using Foxn1 Cre ;Rbpj fl/fl and Foxa2 Cre ;dnMAML mice, and via fetal thymic organ culture (FTOC) in the presence of NOTCH-inhibitor, that NOTCH signaling is required for specification of the mTEC lineage. They further demonstrate that the initial sensitivity of mTEC to NOTCH is restricted to a time-window prior to E16.5, and that NOTCH is required earlier than RANK-mediated signaling in mTEC development. Finally, they show that NOTCH signaling is permissive rather than instructive for mTEC specification, since TEC-specific overexpression of Notch Intracellular Domain (NICD) in fetal TEC dictated an undifferentiated TEPC phenotype rather than uniform adoption of mTEC characteristics. Collectively, our data establish NOTCH as a potent regulator of TEPC and mTEC fate during fetal thymus development.
Furthermore, analysis of the CBF:H2B-Venus mouse line, which reports NOTCH signaling (Nowotschin et al., 2013), indicated ongoing or recent NOTCH activity in half of E14.5 UEA1 + CD205 -mTECs compared to only a small minority of cells in the CD205 + UEA1 -'cTEC' population ( Fig. 1D). Collectively, these data show that the earliest TECs experience high levels of NOTCH signaling, while early mTECs remain competent to receive further NOTCH signals.

NOTCH signaling is required for mTEC development
We next addressed the role of NOTCH in TEC development, by crossing Foxn1 Cre mice (Gordon et al., 2007) to the Rbpj fl/fl conditional knock out mouse line (Han et al., 2002). This generated mice in which RBP-Jκ was absent from all TEC and at least some cutaneous epithelial cells, rendering these cells unable to respond to NOTCH signaling (Han et al., 2002).
The recombination efficiency of Foxn1 Cre was close to 100% in E14.5 EPCAM + TECs when tested using a silent GFP (sGFP) reporter (Gilchrist et al., 2003) (Supplementary Fig. 2), and genotyping indicated complete deletion of Rbpj in total TECs purified from 4-week-old Foxn1 Cre ; RBPJ fl/fl thymi ( Supplementary Fig. 2). Having validated the Foxn1 Cre ; RBPJ fl/fl model (called Rbpj cKO herein), we next analyzed the effect of loss of RBPJ on the postnatal thymus. This revealed a significant proportional and numerical decrease in mTECs in both male and female Rbpj cKO mice at two weeks of age ( Fig. 2A), with cTEC numbers unaffected (Fig. 2B). The decrease in mTEC numbers reflected reduced numbers of MHC Class II hi (mTEC hi ) and MHC Class II lo (mTEC lo ) TEC in males, and of mTEC hi in females (Fig. 2B). This phenotype normalized by eight weeks of age, after which a second loss of mTEC was observed . No other RBP-Jκ-dependent thymic phenotypes were observed: T cell development in the Rbpj cKO mice was not blocked at any stage, and no difference in any of the intrathymic Treg precursor or Treg populations (CD25 -FOXP3 + , CD25 + FOXP3 -, CD25 + FOXP3 + ) (Lio and Hsieh, 2008;Tai et al., 2013) was detected versus controls (Fig. 2F, Supplementary Fig. 2C). Thus, the thymic phenotype in the Rbpj cKO model appeared TEC-specific and affected mTEC but not cTEC.

Temporal requirement for NOTCH signaling in mTEC development
To determine whether the Rbpj cKO mTEC phenotype arose postnatally or during development, we then analyzed E14.5 control and Rbpj cKO thymi using markers characteristic of developing mTEC and cTEC. Fewer K14 + and UEA1 + presumptive mTEC were present in E14.5 cKO thymi than in littermate controls (Fig. 3A). This indicated that the medullary phenotype was evident by E14.5, three days after the onset of Cre expression/Rbpj deletion, establishing that NOTCH signaling is required during emergence of mTEC lineage cells and pointing to a potential role in TEC progenitors.
This phenotype was independently confirmed using  Fig. 3B). Moreover, E12.5 primordia cultured for three days in the presence of DAPT contained significantly fewer UEA1 + mTEC than controls (Fig. 3B). DAPT treatment had no effect on cTEC numbers or overall cellularity in this model (Fig. 3B). Control explants contained medulla-like foci that co-expressed K14 and UEA1 (Fig. 3C), similar to E15.5 thymic primordia ( Fig. 3A) (Rodewald et al., 2001), while a substantial reduction in K14 and UEA1 expressing cells was observed in the DAPT-treated explants (Fig. 3C). These effects could not be attributed to treatment-induced apoptosis or decreased proliferation, since the proportions of Caspase + and Ki67 + mTEC were not significantly affected at the concentration of DAPT used (20μM) (Supplementary Fig 4A, B). To examine the time-dependence of NOTCH signaling in early mTEC development, we extended these analyses by culturing E14.5 and E16.5 fetal thymi ± DAPT for three days. mTEC numbers in cultured E14.5 thymic primordia were significantly reduced in the presence of DAPT (Fig. 3D). However, in contrast, the percentage of mTECs in E16.5 thymi after three days of culture was unaffected (Fig. 3D). Collectively, NOTCH signaling regulates mTEC development during early thymus organogenesis in a restricted time window up to and including E15.5 but prior to E16.5.

NOTCH acts prior to NF-κB signaling to regulate mTEC lineage progression
The NF-κB pathway ligands (Receptor activator of nuclear factor kappa-B ligand [RANKL], lymphotoxin beta and CD40L) are potent regulators of mTEC development and thymic lymphoepithelial crosstalk (Boehm et al., 2003;Hikosaka et al., 2008). Of these only RANKL stimulates both proliferation of mTEC and upregulation of the autoimmune regulator (Aire). Recent studies have shown that the expression of RANK receptor and hence responsiveness to RANKL stimulation increases with increasing maturation of mTEC progenitors (Akiyama et al., 2016;Baik et al., 2016;Mouri et al., 2011). To map the requirement for NOTCH-relative to RANKsignaling, we cultured E15.5 FTOC in the presence of deoxyguanosine (dGuo) to deplete T cells, and then in the presence of RANKL and the presence or absence of DAPT. RANKL elicited a proportional increase in mTEC as expected, (Fig. 4A) and co-treatment with DAPT mildly attenuated this response ( Fig 4A). This suggested that NOTCH and NF-κB might independently regulate different aspects of mTEC development or could act sequentially in the same developmental pathway. To discriminate between these possibilities, we cultured E15.5 Rbpj cKO and littermate control thymi in dGuo-FTOC conditions with or without RANKL.
Consistent with the data shown in Figures 2 and 3, some mTEC progenitors arose in the Foxn1 Cre Rbpj fl/fl model. Culture of Rbpj cKO thymi in RANKL resulted in an approximately threefold proportional increase in mTEC versus unstimulated cKOs and these mTECs displayed a more mature phenotype (MHCII + ) than controls, indicating that once generated, these mTEC progenitors respond normally to RANK. Nevertheless, in RANKL-stimulated Rbpj cKO thymi the proportion of mTEC was substantially lower than that in RANKL-stimulated wild-type controls (Fig 4B), placing the requirement for NOTCH signaling developmentally upstream of that for RANK. These data, together with those in Figure 3D, demonstrate a limited window for NOTCH regulation of mTEC progenitor emergence and establish that NOTCH signaling acts at an earlier stage than NF-κB signaling to regulate the number of mTEC progenitors. They further indicate that once mTEC progenitors are specified, NOTCH is dispensable for mTEC differentiation.

NOTCH signaling is required for specification of the mTEC lineage
The above data would be consistent with NOTCH-regulation of mTEC specification, mTEC progenitor expansion, or both. The Foxn1 Cre ;Rbpj cKO model results in deletion of Rbpj from around E12.0, with subsequent loss of RBP-Jκ function depending on protein turnover and cell division time. The emergence of mTEC progenitors has however been suggested by phenotypic studies to occur independently of FOXN1, possibly at least as early as E10.5 (Hamazaki et al., 2007;Nowell et al., 2011), and therefore the presence of reduced numbers rather than total loss of mTEC progenitors in this model may reflect the relatively late timing of RBP-Jκ deletion. Furthermore, Rbpj mRNA is expressed at only very low levels in E12.5 TEC (not shown), suggesting that NOTCH-mediated effects should occur prior to this time-point. To discriminate between the above possibilities, we therefore determined the effect of blocking NOTCH signaling in TEC at or prior to mTEC and cTEC lineage divergence. For this we generated mice in which NOTCH-mediated transcription is blocked in the developing endoderm before E9.5, by crossing the Foxa2 T2AiCre line with mice carrying the inducible dominant negative Mastermind allele Rosa26 loxp-STOP-loxp-dnMAML-IRES-eGFP allele (Horn et al., 2012;Maillard et al., 2004). This Foxa2 T2AiCre ;Rosa26 loxp-STOP-loxp-dnMAML-IRES-eGFP model (referred to herein as dnMAML) provides a stronger and earlier block of NOTCH activity than that in Foxn1 Cre ;Rbpj fl/fl (i.e. Rbpj cKO) mice.
Collectively, these data unequivocally establish an essential role for that NOTCH-signaling in the normal emergence of the earliest mTEC progenitors, consistent with an obligatory role in mTEC sub-lineage specification. They data further suggest that during normal thymus development, mTEC progenitor emergence commences prior to E12.5.

Notch activity influences TEC progenitor differentiation
Based on the above data, we wished to test whether NOTCH signaling is permissive or instructive for the specification of mTEC progenitors from the putative common TEPC. We thus developed a TEC-specific NOTCH gain-of-functional model by crossing Foxn1 Cre with R26- (Murtaugh et al., 2003), to generate Foxn1 Cre ;R26-stop-NICD-IRES-eGFP mice. In this model, high but physiological levels of NICD -and thus constitutively active NOTCH signaling -are heritably induced in all Foxn1 + cells.
To test whether constitutive NICD expression actively promoted mTEC development, we analyzed TEC differentiation at E14.5, assaying progression of TEC differentiation using PLET1 and MHC Class II (MHCII) as markers of undifferentiated and differentiated cells respectively (Nowell et al., 2011). eGFP expression indicated activation of NICD in >90% of E14.5 TECs  Fig. 8). The proportion of UEA1 + expressing mTECs was unchanged in NICD thymi versus controls, and the cells binding the highest levels of UEA1 were missing ( Fig. 6A; NICD, 4.84% ± 0.21%; control 4.43% ± 0.34%) establishing that high NOTCH activity does not drive immediate universal commitment of mTEC at the expense of cTEC.
Since a rapid expansion of mTEC occurs from E14.5, we also analyzed NICD mice at E16.5.
Collectively, this establishes that overexpression of Notch promotes but does not dictate mTEC specification from the common TEPC and additionally blocks or substantially delays cTEC lineage progression.

Impact of NOTCH signaling modulation on gene expression in fetal TECs
To further interrogate the phenotype of NOTCH loss-and gain-of-function models, we analyzed the transcriptome of fetal TECs, aiming to identify mechanisms regulated by NOTCH signaling within specific TEC populations. For both Rbpj cKO and control thymi, we collected E12.5 PLET1 + TEPCs and E14.5 PLET1 + and PLET1 -TEC, while for NICD at E14.5 we analyzed only PLET1 + TEC, since most NICD TEC were PLET1 + at this timepoint ( Fig. 6A; for data, see https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE100314). A trend suggestive of downregulation of some Notch family and NOTCH target genes was indicated in the E14.5 PLET1 + Rbpj cKO versus controls (Supplementary Table 5, Supplementary Fig. 9) and confirmed by RT-qPCR ( Supplementary Fig. 10), pointing to a positive feedback loop regulating NOTCHsignaling competence. Conversely, several Notch family genes were significantly upregulated in E14.5 NICD TEC versus controls (Supplementary Table 5, Supplementary Fig. 9).
Independent signaling pathway enrichment analysis using all genes differentially expressed between the E14.5 NICD and wild-type datasets also revealed the NOTCH pathway as one of those most affected by NICD overexpression (Fig. 7A). In addition, we found significant upregulation of the EGFR pathway, known to promote the proliferation of mTEC precursors (Satoh et al., 2016), and of several collagen genes (annotated as "Inflammatory Response Pathway"), suggesting that NOTCH signaling may play a role in endowing proliferative capacity on nascent mTECs and in regulating TEPC differentiation by modifying extracellular matrix (Baghdadi et al., 2018). Neither Foxn1 nor Plet1 expression was significantly affected by loss of Rbpj (Supplementary Table 5 Principal Component Analysis (PCA) clustered the E12.5 and E14.5 PLET1 + Rbpj cKO and wild type, and E14.5 PLET1 + NICD, datasets into three groups, E14.5 NICD samples (Group 1); E14.5 PLET1 + and PLET1 -Rbpj cKO and controls (Group 2; see also Supplementary Fig. 12); and E12.5 Rbpj cKO and controls (Group 3) (Fig. 7B). The broad PCA analysis (Fig. 7B) separated the samples by developmental stage (PC1) and PLET1 level (PC2; note that PC2 is not solely PLET1), with Group 1 positioned between Group 2 and Group 3 in PC1. Overall, the PCA is consistent with E14.5 NICD TEC exhibiting at least a partial developmental delay (in keeping with conclusions from Fig. 6), or sustained NICD expression in early TEC inducing a distinct cell state that is not found/ is very rare in the early wild type fetal thymus.
Consistent with these possibilities, clustering analysis revealed differential effects of NOTCH signaling perturbation on markers associated with differentiation into the cTEC and mTEC sublineages, general TEC maturation, or the earliest TEPC state. In particular, genes associated with cTEC lineage identity (Ctsl,Dll4,Psmb11,Prss16,Krt8,Ly75) were up-regulated normally from E12.5 to E14.5 in the Rbpj cKO samples but were expressed at levels similar to E12.5 wild-type in the E14.5 NICD samples (Fig. 7C), consistent with maintained NOTCH signaling imposing a block on cTEC generation from the common TEPC/early cTEC progenitor. Foxn1 also exhibited this expression pattern (Fig. 7C), and indeed many genes in this panel are direct FOXN1 targets (Calderon and Boehm, 2012;Nowell et al., 2011;Zuklys et al., 2016). Notably, overexpression of FOXN1 led to down-regulation of a number of NOTCH family and NOTCH target genes in fetal TEC ( Fig. 7D and data not shown), suggesting that induction of FOXN1 may down-regulate NOTCH signaling in TEC during normal development in vivo. Consistent with this, our reanalysis of published FOXN1 ChIP-seq data (Zuklys et al., 2016) indicated Rbpj as a direct FOXN1 target (Fig. 7E). Moreover, Zuklys and colleagues (Zuklys et al., 2016) identified several known NOTCH targets and modulators as FOXN1 targets (Hey1, Hes6, Deltex4 and Fbxw7). The relative down-regulation of Foxn1 resulting from sustained NICD expression in early fetal TEC (Fig. 7C, Supplementary Fig. 9) thus suggests the possibility of reciprocal inhibition.
Other genes associated with both cTEC and mTEC differentiation, were unaffected or only marginally affected by the NOTCH signaling gain-or loss-of-function mutations. In contrast, markers associated with the mTEC sub-lineage (Krt5, Epcam) were strongly up-regulated in the E14.5 NICD samples compared to controls, and these genes also clustered with other genes normally strongly down-regulated from E12.5 to E14.5 (Cldn3, Cldn4, Cyr61, Plet1, Ccnd1).
Overall, we conclude that upregulation of NOTCH signaling in TEC during early thymus development at least partially blocks cTEC differentiation and promotes but does not dictate mTEC development, suggesting that NOTCH regulates not only mTEC specification but also maintenance of the fetal thymus common TEPC (Fig. 8).

Discussion
We have used conditional loss-and gain-of-function approaches together with pharmacological inhibition to investigate the role of NOTCH signaling in TEC. Our data show, based on TECspecific RBP-Jκ deletion, γ -secretase inhibition in FTOC and enforced dnMAML expression in the developing endoderm from E9.5, that NOTCH activity is essential for mTEC development.
Specifically, they establish that NOTCH signaling is required for the emergence of the mTEC sub-lineage from the putative bipotent TEC progenitor, strongly suggesting that NOTCH regulates mTEC specification, and further show that during mouse fetal thymus development, this requirement is restricted to a developmental window prior to E16.5. Additionally, they demonstrate that NOTCH signaling, whilst essential, is permissive rather than instructive for mTEC development and indicate a further role for NOTCH in regulating exit from the early bipotent TEPC state into mTEC and cTEC differentiation. These findings, summarised schematically in Fig. 8, raise several issues which are discussed below.

Timing of the NOTCH signaling-requirement
NOTCH signaling has been shown to regulate distinct events in the different developmental stages of a tissue (Hartman et al., 2010;Radtke et al., 2004;Shih et al., 2012). A recent study reported that NOTCH activity is enriched in cTECs and that repression of NOTCH by the histone deacetylase HDAC3 is important for expansion/maintenance of developing mTEC (Goldfarb et al., 2016). This study analyzed the same NOTCH overexpression line as used herein, but at the later time-points of 10 days and 6 weeks postnatal (Goldfarb et al., 2016). The conclusions of this and our own studies are entirely compatible, with the data presented herein establishing a requirement for NOTCH signaling at the earliest stages of TEC lineage divergence, and the data of Goldfarb indicating that down-regulation of NOTCH signaling is required for later stages of mTEC differentiation (Goldfarb et al., 2016). Indeed, the phenotypes observed in each may reflect the outcome of same perturbation in TEC differentiation, at separate stages. However, it is also possible that NOTCH has secondary roles in TECs subsequent to its initial role in mTEC specification, which ceases by E16.5 (Fig. 2).

Thymic crosstalk
The NFκB pathway plays a vital role in mTEC development and consequently in the establishment of central tolerance (Akiyama et al., 2005;Burkly et al., 1995;Kajiura et al., 2004).
Recent studies using transcriptomics and functional assays have led to more clarity on how the several NFκB ligands, through which thymic crosstalk occurs, function during mTEC maturation (Akiyama et al., 2016;Bichele et al., 2016;Desanti et al., 2012;Mouri et al., 2011). In particular, Akiyama and colleagues identified two separable UEA1 + mTEC progenitor stages, pro-pMECs and pMECs based on the expression of RANK, MHCII and CD24 (Akiyama et al., 2016). The transition from the more primitive pro-pMECs to pMECs depends on RELB, whereas further maturation from pMECs is TRAF6-dependent. Crucially, both pro-pMECs and pMECs respond to induction by RANKL in T cell-depleted FTOC (Akiyama et al., 2016). We initially interpreted our data on potential interplay between NOTCH and NFκB to suggest synergy between these two pathways in mTEC development, since NOTCH signaling-inhibition attenuated RANKL stimulation in E15.5 wild-type T cell-depleted FTOC (Fig. 4A). However, comparison with E15.5 Rbpj cKO FTOC indicated that NFκB activation of already-specified mTEC progenitors is unaffected by lack of NOTCH signaling-responsiveness: although the block in mTEC development was more severe in the Rbpj cKO FTOC the few mTEC that were present could be stimulated by RANK, indicating the presence of pMECs and/or pro-pMECs. The attenuation of RANKL stimulation upon DAPT treatment of E15.5 wild-type FTOC thus suggests that mTEC specification is still on-going at E15.5. However, we also observed that mTEC clusters in Rbpj cKO thymi tended to be smaller than those in controls, and therefore the possibility that in addition to regulating mTEC specification NOTCH also regulates the initial expansion of mTEC progenitors cannot be ruled out. Indeed, our data reveal EGFR signaling as a major target of NOTCH during early TEC development.
In contrast, our data show that while E10.5 3PP explants can generate UEA1 + mTECs and CD205 + cTEC/progenitors in culture, these UEA1 + mTEC do not respond to RANKL. It is thus likely that the UEA1 + cells in these explants represent an even more primitive mTEC progenitor state than the pro-pMECs. Of note is that some DAPT-treated E10.5 3PP explants produced no UEA1 + mTECs, and thus that mTEC specification can be completely suppressed in the absence of NOTCH signaling. Taken together, these results suggest that although NOTCH and NFκB are both required for mTEC development, the two pathways act sequentially but independently.

Notch regulation of mTEC progenitor emergence
The loss of mTECs in NOTCH loss-of-function models could be explained by three hypotheses: (i) NOTCH might regulate the decision of bipotent TEPCs to become mTECs. In this model, in the absence of NOTCH signaling, bipotent progenitors fail to commit to mTEC fate and over time become cTECs instead. (ii) Alternatively, high levels of NOTCH signaling might dictate that TEPCs remain bipotent, with cells that experience lower NOTCH committing to the cTEC lineage. Unlike the 'specification hypothesis', in this scenario mTECs would fail to emerge in the absence of NOTCH signaling because the bipotent TEPCs undergo premature differentiation into cTECs, exhausting the pool that retains the potential for mTEC generation. (iii) Finally, NOTCH might be required for the proliferation of specialized mTEC progenitors; in this case we would expect the perturbation to affect only mTECs and not cTECs or bipotent progenitors.
We conclude from the gain-of-function data that enhanced NOTCH activity neither switches all TECs to become mTECs, nor only affects mTECs. Instead, NOTCH activity is necessary but not sufficient for mTEC fate in the developmental timeframe investigated. Despite the caveats with established markers, the considerable shift towards a PLET1 + MHCII - (Fig. 6A, C) K5 + K8 + (Fig. 6B) phenotype suggests a more immature, TEPC-like state as the primary phenotype resulting from high NOTCH activity. Indeed, the transcriptome of E14.5 NICD TECs occupies a state that is separate from both E12.5 TEPCs and age-matched controls, whilst sharing certain features with both clusters. As development progresses from E14.5 to E16.5, many TECs do upregulate the mTEC markers UEA1 and K14, indicating that high NOTCH activity is compatible with acquisition of mTEC fate. Importantly, the NICD + UEA1 + mTECs at E16.5 display comparable maturation status to controls, whereas CD205 + cTEC/common TEPCs continue to exhibit a primitive phenotype (Fig. 6). These data suggest that once mTECs are specified, further development is independent of NOTCH signaling.
The gain-of-function results also support our hypothesis that NOTCH operates at the TEC progenitor level, whilst opposing the model that NOTCH activity only influences mTECs. It does not however rule out the specification model. Although retention of an early progenitor state seems to be the primary outcome of enforced NOTCH signaling, the proportion of mTEC in the E16.5 gain-of-function thymi is higher than controls. Several factors may be in play in this second phase. The duration of signaling has been shown to result in the temporal adaptation of sensitivity in several pathways (reviewed in (Kutejova et al., 2009)). Moreover, instead of a simple ON/OFF response, the NOTCH response may be graded, as in the case of inner ear (Petrovic et al., 2014) and pancreas development (Shih et al., 2012). mTEC specification may require higher levels of NOTCH, which could for instance be achieved by positive feedback above the levels of those imposed by the enforced NICD expression in the NICD hemizygous mice used in these experiments. Variables independent from NOTCH may also play a part. A potential candidate is FOXN1, which drives TEPCs out of the primitive undifferentiated state and into differentiation (Nowell et al., 2011), and indeed our data indicate interplay between FOXN1 expression levels and NOTCH activity (as depicted in Fig. 8). In addition to the direct interaction suggested from our analysis, FOXN1-mediated repression of NOTCH activity could be reinforced via its direct targets DLL4 and FBXW7; the former may mediate cis-inhibition of NOTCH receptors, while the latter has been shown to enhance the degradation of NICD (Carrieri and Dale, 2016;del Alamo et al., 2011). We note that the thymic phenotype of the NOTCH gain-of-function mutant reported here resembles those of the Foxn1 R/- (Nowell et al., 2011) and the Foxn1 Cre ;iTbx1 (Reeh et al., 2014) mutant mice, in which exit from the earliest TEPC compartment is also severely perturbed due to the inability to express normal levels of FOXN1.
One of the long-term goals of the field is to create fully functional thymus organoids from TECs derived from pluripotent stem cells or by direct conversion from unrelated cell types (reviewed in (Bredenkamp et al., 2015)). Understanding the duration of TEPC bipotency, lineage plasticity and NOTCH activity would improve protocols and inform strategies in this regard. Our data predict that, by manipulating the levels of NOTCH signaling TEPCs experience, it may be possible to produce more homogenous populations of TEC subsets, including TEPC. However, the complexities indicated from studies on NOTCH in other organs, together with the potential for differential effects on TEC at different stages of lineage progression, suggest that further advances in this direction will require caution and precision.

Mice
CBAxC57BL/6 F1 mice were used for isolation of fetal TEC. For timed matings, C57BL/6 females were housed with CBA males, and noon of the day of the vaginal plug was taken as Thymus dissociation: Postnatal thymi were dissociated in 1.25mg/ml collagenase D (Roche), and subsequently in 1.25mg/ml collagenase/dispase (Roche) diluted in RPMI medium (Life Technologies). 0.05mg/ml DNaseI (Lorne) was added to the buffer to minimize cell adhesion.
Fetal thymi were dissociated for 20 minutes using a PBS-based buffer consisting of 1.25mg/ml collagenase D, 1.4mg/ml hyaluronidase (Sigma) and 0.05mg/ml DNaseI. After digestion cells were spun down and digested in 1x trypsin for two minutes. Cell suspension was then filtered through 70μm cell strainer (Corning) to remove clumps.
Flow Cytometry: Adult thymi and grafted RFTOC were processed for flow cytometric sorting and analysis as previously described (Bredenkamp et al., 2014;Nowell et al., 2011). See Supplemental Experimental Procedures for detailed protocols. For analysis and sorting, adult thymic tissue was depleted of T cells using anti-CD45 MACS beads (Miltenyi Biotec); fetal tissue was not T cell depleted. Cell counts were carried out using a BioRad cell counter and slides, where required. Sorting and analysis was performed using a BD FACS Aria II and a BD LSR Fortessa respectively at the CRM, University of Edinburgh. For Rosa26NICD TEC, sorting was performed on a BD FACS Aria II at the University of Lausanne, Epalinges. Sorting protocols were identical for all cell isolation experiments. All flow cytometry data were analyzed using FlowJo Version 9.7.6 (Tree Star, Inc).  Table S2. See also Supplemental Experimental Procedures.

Fetal thymus organ culture (FTOC):
For reaggregates or thymi older than E15, FTOCs were cultured on a Millipore membrane raft floating on DMEM supplemented with 10% FCS and Lglutamine. Third pharyngeal pouches or thymic primordia younger than E15.5 were submerged and allowed to settle on thin matrigel (Corning), then cultured in N2B27 supplemented with BMP4 (Peprotech) and FGF8 (Peprotech). Where DAPT (Tocris) or deoxyguanosine (dGUO; Sigma) were used, the equivalent amount of DMSO was added to the control medium. RANKL (Peprotech) was used at 500ng/ml.  Table S3. indicated small amount of adaptor contamination and few low quality reads, therefore the raw data were trimmed with Trimmomatic (Bolger et al., 2014) using default parameters for PE reads and the cropping option specific for the Nextera PE adapters. Only paired reads that passed QC were aligned with STAR against the mouse genome assembly (GRCm28 -Ensembl 87) and the aligned reads were assigned to genes with featureCounts (Liao et al., 2014). The resulting count tables were imported to R for further normalisation and analysis.

RNA-seq
Batch effect correction was applied for the within group lane effects, however, some batch effects could not be corrected. This applied to the potential for a laboratory effect between the E14.5 NICD and all other samples, since the E14.5 NICD sample was collected at EPFL Lausanne. However, the same thymus dissociation and cell sorting protocols, and the same make and model of cell sorter, were used, and the subsequent sample processing was performed at the University of Oxford using the same protocol as for all of the other samples. To control for this, the expression levels of housekeeping genes were determined for all samples and were not biased in any particular groups ( Supplementary Fig.12B).
Differential expression analysis was performed using the LIMMA package and voom (Ritchie et al., 2015) from Bioconductor (Gentleman et al., 2004) and a threshold of FDR         Data Collection: (A-C) To obtain the E12.5 and E14.5 cKO and wild type samples, thymi were microdissected from E12.5 and E14.5 embryos generated from a Foxn1 Cre ;Rbpj FL/+ x Rbpj FL/FL cross and TECs obtained by flow cytometric cell sorting. Following genotyping, cells from three cKO and three control samples were processed for sequencing. The E12.5 and E14.5 samples were each obtained from two separate litters, on two separate days for each timepoint. To obtain the E14.5 NICD samples, thymi were microdissected from five E14.5 Foxn1Cre; R26 LSL-NICD-EGFP embryos of the same litter, TECs were obtained by flow cytometric cell sorting, and the samples processed for sequencing. (D) n=3, where each n represents TECs sorted from pooled embryos from a single litter of E17.5 iFoxn1 or wild-type embryos.