The close lineage relationship between T cells and group 2 ILCs (35) suggests that common precursors may exist for both lineages, which, depending on microenvironmental signals, differentiate into either T cells or ILC2. T cells differentiate in the thymus. Where ILC2 develop has not been established. There is evidence that these cells are generated in other locations than thymus, as suggested by the discovery of an ILC2 precursor in bone marrow (32, 41) and the presence of ILC2 in athymic FoxN1^nu/nu (nude) mice (23, 42). However, these findings do not exclude the possibility that ILC2 can be generated in the thymus as well. If common precursors of ILC2 and T cells exist, which differentiate into ILC2 in the thymus, one would expect to find differentiated ILC2 in this organ. To examine this, we analyzed single cell suspensions of freshly isolated human thymocytes by multi-color flow cytometry for markers associated with group 2 ILC. ILCs were defined as CD45⁺ cells expressing high levels of the IL7Rα chain (CD127), but lacking expression of markers specific for T cells (CD1a, CD3, TCRγδ, TCRαβ), B cells (CD19), hematopoietic stem cells (CD34), and other lineages (CD11c, CD14, CD94, CD123, FcϵR1, BDCA2), collectively termed “lineage” (Figure 1A, left). We found that an average of 0.0625% of CD45⁺ thymocytes belong to the ILC lineages. We further analyzed these IL7Rα^highlineage⁻cells for the expression of CRTH2 (Figure 1A, right), which, within the family of ILCs, is specific for group 2 ILC (9). Using this strategy, we indeed detected a distinct population of CRTH2⁺ ILC in thymic specimens, on average accounting for 5% of the lineage⁻IL7Rα^high ILC compartment (Figure 1C).

Phenotypically, thymic lineage⁻IL7Rα^highCRTH2⁺ cells are similar to group 2 ILCs identified in other human tissues in that they express CD161 (9) and cKit (CD117) (Figure 1B), as was described for a subpopulation of human ILC2 (1). Furthermore, thymic lineage⁻IL7Rα^highCRTH2⁺ cells do not express CD56 or NKp44 (Figure 1B), which are associated with conventional NK cells and ILC3 (1). These cells therefore adhere to what have been defined as the phenotypic hallmarks of human group 2 ILC (2, 9) and can thus be regarded as thymic ILC2.

The presence of ILC2 in the thymus is remarkable, because thus far, such cells have only been found in mucosal tissues in the intestine and airways, consistent with their role in innate border patrol, in blood (9), presumably reflecting their migration, and in bone marrow, the major site of hematopoiesis (8, 34). Although we did not formally test whether ILC2 differentiation takes place in the human thymus, the fact that such cells are found at relatively high numbers in the thymus is consistent with the possibility that these cells are generated in this non-inflamed organ.

Given the presence of innate group 2 lymphocytes in the human thymus, we asked whether these cells can develop directly from thymic progenitors. A recent study showed that ILC2 differentiation can be induced from murine bone marrow-derived common lymphoid and thymic progenitors by co-culture with OP9 stromal cells expressing the Notch ligand Dll1 (23). To test whether human thymic progenitors have the capacity to differentiate into ILC2, we initiated OP9 co-cultures with human CD34⁺CD1a⁻ thymocytes, which have not yet committed to the T lineage (43). Although some lineage⁻IL7Rα⁺CRTH2⁺ cells appeared after 1 week of culture with control OP9 cells, the frequency of such cells was significantly increased upon co-culture with OP9 Dll1 (Figure 2A). Under these same conditions, OP9 Dll1 cells also induced differentiation of T cells, as expected, but this process required minimally 2 weeks of co-culture (Figure 2B).

Interestingly, when OP9 Dll1 co-cultures were initiated with thymic CD34⁺CD1a⁺ progenitors, which are believed to already have committed to the T cell lineage (43), enhanced induction of lineage⁻IL7Rα⁺CRTH2⁺ cells by OP9 Dll1 was not detected (Figure 2C). This indicates that these progenitors might have lost the potential to differentiate into the ILC lineage, as has been demonstrated for murine thymic DN3 cells (23).

OP9 cells expressing Dll1 have widely been used to induce Notch dependent T cell differentiation (44). It is not immediately obvious how activation of Notch in the same progenitors would induce two distinct differentiation programs, namely ILC2 or T cells. Although Notch is clearly required in a cell intrinsic manner in thymic progenitors during T cell differentiation (45), this has not been established for ILC2. Indeed, it is conceivable that ILC2 induction by OP9 Dll1 would be an indirect effect from lateral Notch activation in the OP9 cells, resulting in production of other signals by these cells, which promote ILC2 differentiation. To test whether cell intrinsic Notch signaling in thymic progenitors induces ILC2 differentiation, we ectopically expressed the intracellular domain of NOTCH1 (NICD1) in thymic CD34⁺CD1a⁻ progenitors, thereby inducing constitutive activation of NOTCH1 in these cells, and subjected these to co-culture on control OP9 cells. NICD1 expression resulted in robust induction of an IL7Rα⁺CRTH2⁺ population, which lacks expression of T cell (CD1a, CD3, CD4, CD8, TCRαβ, TCRγδ) and other lineage markers (CD11c, CD14, CD19, CD34, CD94, CD123, FcϵR1, BDCA2) (Figure 3A). Thus, direct activation of Notch in thymic CD34⁺CD1a⁻ progenitors results in the differentiation of cells resembling ILC2 cells.

A prominent population of lineage⁻CRTH2⁺ cells was usually present after 1 week, but these cells first appeared after as little as 4 days of co-culture on OP9 cells, and persisted for up to 2 weeks (Figure 3B). NICD1 induced differentiation of lineage⁻IL7Rα⁺CRTH2⁺ cells was subject to donor-to-donor variation. However, ectopic NICD1 expression consistently induced a prominent lineage⁻IL7Rα⁺CRTH2⁺ population (Figure 3C).

Strikingly, NICD1 even robustly induced a lineage⁻IL7Rα⁺CRTH2⁺ population from T-committed CD34⁺CD1a⁺ thymic progenitors, although to a lesser degree than from uncommitted CD34⁺CD1a⁻ thymocytes (compare Figures 3C,D). This suggests that strong, sustained Notch signaling can overcome commitment to the T cell lineage in thymic progenitors and therefore challenges the lineage fidelity of these precursors. Strikingly, ectopic expression of NICD1 in CD34⁺CD1a⁻ thymocytes also resulted in a prominent lineage⁻IL7Rα⁺CRTH2⁺ population when thymic progenitors were cultured in the combined presence of IL-7 and Flt3l (Figure 3E), conditions known to promote development into the T cell lineage (46, 47).

Collectively, these data show that activation of Notch in human thymic hematopoietic progenitors activates the ILC2 differentiation program. Indeed, the connection between Notch and the ILC2 differentiation program is potent enough to dismantle the T cell differentiation program in cells already committed to the T cell lineage and make them adopt the ILC2 fate instead.

The transcription factor GATA3 has been shown to play an essential role in development and function of both murine and human ILC2 (32, 33). Overexpression of GATA3 in a population of lineage⁻CD127⁺CD117⁺Nkp44⁻CRTH2⁻ immature ILC was sufficient to drive these cells into the ILC2 lineage (33). To determine whether GATA3 is also able to elicit ILC2 differentiation from thymic progenitors, we ectopically expressed this factor from a retroviral vector in CD34⁺CD1a⁻ uncommitted progenitors and monitored ILC2 differentiation. Indeed, GATA3 expression was sufficient to induce an ILC2 phenotype (Figure 3F), further supporting the notion that thymic progenitors have the capacity to differentiate into genuine ILCs. However, NICD1 was more potent in ILC2 differentiation in direct comparison with GATA3, at least at the expression levels obtained by our retroviral expression systems (Figure 3F).

Thus far, our interpretation that Notch activation instructs thymic progenitors to differentiate into ILC2 cells has been based on the absence of lineage markers (including those defining T cells) and the presence of CRTH2, a marker supposedly not found in immature T cells. However, given the wealth of data demonstrating the role of Notch in differentiation of T cells, it is conceivable that the cells obtained here represent a previously not described subtype of immature T cells. Also, it is possible that our results reflect atypical expression of CRTH2 on T cell lineage cells, for instance as a consequence of direct transactivation of the gene encoding this marker by Notch. We therefore sought to determine whether the CRTH2⁺ cells differentiated from thymic progenitors in response to Notch are genuine group 2 ILCs.

To this end, we performed more extensive phenotyping by multi-color flow cytometry, examining the surface expression of several markers known to be associated with human ILC2. As described above (Figure 3A), NICD1⁺ (CRTH2⁺) cells express the IL7Rα chain (Figure 4A), although surface levels varied between experiments and were sometimes lower than those found on freshly isolated ILCs. This is most likely due to in vitro culture in the presence of recombinant IL-7, which results in receptor internalization (48). Indeed, such down regulation of IL7Rα has also been observed when freshly isolated mature ILC2 cells were cultured in vitro (9). NICD1 induced CRTH2⁺ cells express low levels of KLRB1 (CD161) in a bimodal distribution (Figure 4A) as has been shown for human ILC derived from other tissues (9). Additionally, NICD1 induced CRTH2⁺ cells also express CD25 (the IL-2Rα chain), CD7, and ICOS, all of which have been shown to be expressed by group 2 ILCs (1, 9). CRTH2⁺ cells differentiated in vitro by Notch activation for the most part do not express cKit (Figure 4A). The interpretation of this observation is complicated by the fact that expression of cKit by human group 2 ILC is variable. Among human fetal gut and peripheral blood ILC2, both a cKit⁺ and cKit⁻ population have been described (9). Expression of cKit, therefore, does not seem to constitute part of the core ILC2 identity. Finally, NOTCH1 induced CRTH2⁺ cells do not show surface expression of CD56 and Nkp44 (Figure 4A), associated with the NK and other ILC subsets (1). Taken together, the expression pattern of these surface markers shows that CRTH2⁺ cells derived from thymic progenitors by Notch mediated in vitro differentiation phenotypically resemble bona fide ILC2 found in the thymus and various other human tissues.

To further characterize these Notch induced thymic ILC2, we determined mRNA expression levels of lineage defining transcription factors, cytokine receptors, and signature cytokines in cells directly isolated from differentiation cultures (Figure 4B). Cells differentiated from thymic progenitors in vitro by expression of NICD1 expressed transcripts for the lineage specific cytokine receptors IL1RL1 (ST2, a subunit of the IL-33 receptor), IL17RB (a subunit of the receptor for IL-25), and CRLF2 (TSLP-R). Together with CD25 (Figure 4A), these cells thus express the critical receptors to respond to the ILC2 activating cytokines IL-2, IL-25, IL-33, and TSLP. Notably, expression of transcripts for these cytokine receptors was not detected in empty vector control cells (Figure 4B). In vitro differentiated ILC2 also expressed RORA mRNA encoding the ILC2 lineage specific transcription factor RORα. However, these cells did not express elevated levels of GATA3 compared to control cells. This may seem counter-intuitive, given the ability of GATA3 to induce ILC2 differentiation from thymic progenitors (Figure 3E) and the fact that GATA3 is a known direct target of Notch signaling (49–51). However, it should be noted that expression of GATA3 is low also in resting ILC2 derived from peripheral tissues after in vitro culture and that expression of this factor is elevated only after exposure to activating stimuli such as TSLP (33). Therefore, high constitutive expression of GATA3 does not seem to be a characteristic of resting human ILC2 cells. Most strikingly, even without exogenous activation, NICD1⁺ cells expressed transcripts for the ILC2 signature cytokines IL-13 and IL-5 (Figure 4B), whereas empty vector control cells did not, further reinforcing the interpretation that NICD1 induced CRTH2⁺ cells are fully functional group 2 ILC. Constitutive expression of IL-5 and IL-13 transcripts by unstimulated ILC2 has been observed previously in the fetal gut (9). We therefore conclude that ILC2 differentiated in vitro from thymic progenitors by activation of NOTCH1 resemble bona fide group 2 ILCs.

The fact that activation of Notch can induce differentiation of both T cells and ILC2 from the same thymic progenitors raises the question what determines which of these differentiation programs is turned on. The Notch pathway is sensitive to signal amplitude, which allows induction of discretely different responses by one and the same signaling pathway (52, 53). Given our findings that expression of NICD1 was much more potent at eliciting ILC2 differentiation than co-culture with OP9 Dll1, we reasoned that the strength of the Notch signal might be a critical factor deciding whether T cells or ILC2 are generated.

To test this hypothesis, we generated an expression construct, in which the concentration of nuclear NICD1 can be controlled in a quantitative manner. In this construct, NICD1 is N-terminally fused to a mutated Estrogen receptor ligand binding domain (mER). This mutated domain no longer binds Estrogen, but does respond to Tamoxifen (54). In the absence of this drug, mER-NICD1 is bound by heat shock proteins in the cytoplasm and hence kept transcriptionally inactive. Addition of Tamoxifen to the culture medium induces transcriptional activity of mER-NICD1 in a dose-dependent manner (Figure 5A). Some leakiness is frequently observed with these types of ER-fusion proteins (55, 56) and indeed, we also consistently observed weak but significant activity of mER-NICD1 in the absence of Tamoxifen (Mock) on two different Notch-responsive promoters (Figure 5A). Therefore, this tool enabled us to induce different levels of NOTCH1 activation, and even explore the impact of very low levels owing to the leakiness of the mER system. In comparison, expression of constitutively active NICD1 served to induce high signaling strength. Together, these constructs allowed us to examine the consequences of Notch signaling across a more than 100-fold dynamic range, as measured in reporter gene assays (compare luciferase activities in the absence of Tamoxifen with those obtained with expression of constitutively active NICD1 in Figure 5A).

To test whether Notch signal strength affects ILC2/T cell differentiation, we retrovirally expressed mER-NICD1 or constitutive NICD1 in uncommitted, CD34⁺CD1a⁻ thymic precursors, subjected these to co-culture on control OP9 cells and titrated in Tamoxifen. The activity of the mER-NICD1 and NICD1 constructs could also be controlled quantitatively in these cells, as shown by the Tamoxifen dose-dependent induction of CD7 expression (Figure 5B), which we have found to be a sensitive gauge for Notch activity in thymic precursors.

Using this system, we measured differentiation of T cells as well as group 2 ILC by flow cytometry. We chose to examine the development of these cells after 2 weeks because of the different kinetics in T cell and ILC2 differentiation. Differentiation of T cells generally requires incubation periods of at least 2 weeks (57) (Figure 2B). As shown before (Figure 3B), group 2 ILC populations emerge within several days of culture, but persist longer and could therefore be assessed here after 2 weeks in direct comparison with T cells.

As expected, neither T cell differentiation nor generation of ILC2 occurred effectively in the absence of Notch activation: considerable populations of both CD4⁺CD8α⁺ T cells (Figure 5C) and lineage⁻IL7Rα⁺CRTH2⁺ ILC2 (Figure 5D) only arose from cultures which had received a Notch signal by means of ectopic NICD1 expression, but not in empty vector transduced control cultures. With regard to dosage dependence, however, T cell and ILC2 differentiation displayed opposite requirements: even very low doses of Notch activity were sufficient to elicit development of T-lineage cells (Figure 5C). In fact, the lowest levels of NOTCH1 signaling activity, owing to the leakiness of our system in the absence of Tamoxifen (see above), gave rise to the most prominent T cell differentiation (almost 20% of CD4⁺CD8α⁺ T cells), whereas hardly any ILC2 were observed in this condition (Figure 5D). Titrating in Tamoxifen to induce higher levels of NOTCH1 signaling gradually diminished the number of T cell lineage cells obtained (Figure 5C). In contrast, stronger NOTCH1 induced more ILC2 differentiation (Figure 5D). Most strikingly, the highest level of Notch signaling, induced here by constitutively active NICD1, did not yield any T cells (Figure 5C); instead, constitutive NICD1 expression resulted in the most prominent ILC2 differentiation (Figure 5D). Taken together, these results demonstrate that by varying signal strength, one and the same Notch pathway can activate two distinct differentiation programs, with weaker signals favoring development of T cells and stronger signals inducing more efficient ILC2 differentiation.

Postnatal thymic (PNT) tissue specimens were obtained from children undergoing open heart surgery (LUMC, Leiden, The Netherlands); their use was approved by the AMC ethical committee in accordance with the declaration of Helsinki. Cell suspensions were prepared by mechanical disruption using the Stomacher 80 Biomaster (Seward). After overnight incubation at 4°C, thymocytes were isolated from a Ficoll-Hypaque (Lymphoprep; Nycomed Pharma) density gradient. Single cell suspensions were enriched for CD34⁺ cells by MACS (Miltenyi Biotec), stained with fluorescently labeled antibodies and subsequently FACS sorted on a FACS Aria (BD Bioscience) as CD34⁺CD1a⁻CD3⁻CD56⁻BDCA2⁻ or CD34⁺CD1a⁺CD3⁻CD56⁻BDCA2⁻, respectively (referred to in this study as CD34⁺CD1a⁻ and CD34⁺CD1a⁺). Purity of the sorted populations was >99%.

Staining for expression of surface proteins was performed at 4°C for 20 min. Distinction of live and dead cells was based on staining with 7-Aminoactinomycin D (7-AAD, eBiosciences) or fixable live/dead dyes (Invitrogen). Data were acquired on a LSR Fortessa flow cytometer (BD Bioscience) and analyzed using FlowJo software (TreeStar). Single cell suspensions were stained with antibodies directly labeled with Fluorescein Isothiocyanate (FITC), Phycoerythrin (PE), Phycoerythrin-Cyanine 5 (PE-Cy5), PE-Cy5.5, PE-Cy7, PerCP-Cy5.5, Allophycocyanin (APC)/Alexa Fluor 647, APC-Cy7, AF700 (all BD Bioscience, BioLegend, or MACS Miltenyi), Horizon V500 (HV500, BD Bioscience), Brilliant Violet 421 (BV421), BV711, and BV785 (all BioLegend). Antibodies specific for the following human antigens were used: CD1a, CD3, CD4, CD7, CD8, CD11c, CD14, CD19, CD25, CD34, CD45, CD56, CD94, CD117 (cKit), CD123, CD127 (IL7Rα), CD161, CD294 (CRTH2), CD303 (BDCA2), CD336 (Nkp44), CD278 (ICOS), TCRαβ, TCRγδ, and FcεR1. Anti-mouse CD90.1 (Thy1.1) -FITC, -PE, or -APC-eFluor 780 (eBioscience) were used to detect cells transduced with MSCV – IRES-Thy1.1 retroviruses.

The human NICD1-IRES-Thy1.1-MSCV construct has been described before (49). To generate the mER-NICD fusion, an N-terminal mER domain was PCR amplified using the following primers: GATCAGGAATTCCACACCATGGGAGATCCACG AAATGAA and GATCAGGATATCCACCTTCCTCTTCTTCTTGG and cloned into the EcOR1 and EcORV sites of pBluescript (pBS) to create mER-pBS. Human NICD1 lacking a translation initiation signal was PCR amplified using these primers: ATCGGAGGTTCTCGCAAGCGCCGGCGGCAGCAT and GATCAGAAGCTTGAATTCTTACTTGAAGGCCTCCGGAATG and subsequently cloned into the EcORV and HindIII sites of mER-pBS. The mER-NICD1 fusion insert was then cloned into IRES-Thy1.1-MSCV using BamH1 and Cla1.

For virus production, Phoenix GALV packaging cells were transiently transfected using FuGene HD (Promega). Virus containing supernatant was harvested 48 h after transfection, snap frozen on dry ice and stored at −80°C until use. For transduction, cells were incubated with virus supernatant in plates coated with Retronectin (30 μg/ml, Takara Biomedicals) for 6–8 h at 37°C the following day.

Sorted thymic progenitors were cultured overnight in Yssel’s medium containing 5% normal human serum, SCF (20 ng/ml), and IL-7 (10 ng/ml, both PeproTech). OP9 cells were mitotically inactivated by irradiation with 30 Grey and seeded at a density of 5 × 10³/cm² 1 day prior to initiation of co-cultures. After transduction, thymic progenitors were added to pre-seeded OP9 cells. Co-cultures were performed in MEMα (Invitrogen) with FCS (20%, FetalClone I, Hyclone) and IL-7 (5 ng/ml). In some cases, Flt3l (5 ng/ml, PeproTech) was added to the medium. Cultures were refreshed every 3–4 days. Differentiation assays for ILCs were typically analyzed after 1 week, unless stated otherwise. Cells were harvested by forceful pipetting and passed through 70 μm nylon mesh filters (Spectrum Labs). For longer culture periods, cells were transferred onto fresh stromal cells every week.

RNA was isolated using the NucleoSpin RNA II or NucleoSpin RNA XS kit (Macherey-Nagel) according to the manufacturer’s instructions. cDNA synthesis was done using the High Capacity cDNA kit (Applied Biosystems). Quantitative Real-Time PCRs were performed in an iCycler instrument using IQ SYBR Green Supermix (both BioRad). For calculation of relative expression level of the genes of interest, the ΔCt method was used. The following primers were used: ACTB (β-Actin) (Gene ID: 60) Forward: CAC CAT TGG CAA TGA GCG GTT C; Reverse: AGG TCT TTG CGG ATG TCC ACG T; CRLF2 (TSLP-R) (Gene ID: 64109) Forward: GAG TGG CAG TCC AAA CAG GAA; Reverse: ACA TCC TCC ATA GCC TTC ACC; GATA3 (Gene ID: 2625) Forward: ACC ACA ACC ACA CTC TGG AGG A; Reverse: TCG GTT TCT GGT CTG GAT GCC T; IL1RL1 (ST2) (Gene ID: 9173) Forward: ATG TTC TGG ATT GAG GCC AC; Reverse: GAC TAC ATC TTC TCC AGG TAG CAT; IL-5 (Gene ID: 3567) Forward: AGC TGC CTA CGT GTA TGC CA; Reverse: CAG GAA CAG GAA TCC TCA GA; IL-13 (Gene ID: 3596) Forward: ATT GCT CTC ACT TGC CTT GG; Reverse: GTC AGG TTG ATG CTC CAT ACC; IL17RB (Gene ID: 55540) Forward: CCA ACA CAG CAC TAT CAT CG; Reverse: ATA TGG AGT CAG CTG CAC CG; PTGDR2 (CRTH2) (Gene ID: 11251) Forward: AAT CCT GTG CTC CCT CTG TG; Reverse: ATG TAG CGG ATG CTG GTG TT; RORA (Gene ID: 6095) Forward: ACA AGC AGC GGG AGG TGA TGT; Reverse: TGA GAG TCA AAG GCA CGG C.

U2OS cells were transiently transfected using the FuGene HD transfection reagent (Promega). Cells were co-transfected with a NOTCH-responsive promoter and either NICD1-IRES-Thy1.1-MSCV, mER-NICD1-Thy1.1-MSCV, or an empty vector control. To correct for differences in transfection efficiency, the pRL-CMV control vector was co-transfected, from which Renilla luciferase is expressed constitutively. Transfections were performed in triplicate. Where applicable, 4-Hydroxy-Tamoxifen (Sigma) was added after overnight incubation to induce nuclear translocation of mER-NICD1. Cells were lysed 48 h post transfection and luciferase activity was measured using the Dual Luciferase Reporter Assay System (Promega) on a Synergy HT microplate reader (Syntek). Two different Notch-responsive reporter constructs were used, which have been described previously (76).