Dll1 and Dll4 are conserved in bony fish and terrestrial organisms, including mammals. These Notch ligands share structural characteristics, but mammalian Dll4s do not retain the DOS motif necessary for the binding of Dll1 to Notch1 due to the substitution of Pro in the motif to Asn at the second EGF-like repeats ( Figures 1A, B ) (2, 16). On the other hand, the N-terminal MNNL region of murine Dll4 that contains a loop structure with a wide range of motion directly contributes to binding with Notch1 (22). In contrast, that of murine Dll1 loses the ability to move widely because of its rigidity due to the sequential presence of unique prolines ( Figure 1C ) (21). Therefore, Dll molecules seem to bind to Notch1 in different regions. To examine the characteristics of Dll1 and Dll4 molecules during evolution, we sorted the genes of Dll family members from the National Center for Biotechnology Information (NCBI) database according to their homologies and formed a phylogenetic tree ( Supplementary Figure 1 ). This analysis indicated that the Dll4 gene first emerged in bony fishes, while the Dll1 genes were identified from amphioxus (Branchiostoma floridae) and vertebrates, which is consistent with a previous report (9). In the Dll4 genes, the DOS motif maintained in many bony fishes (23 species, including medaka fish, Figure 1D ) is different from the one in terrestrial organisms, with some exceptions (zebrafish, arowana, and Japanese pufferfish) that have another substitution in the DOS motif (Trp to Gly, Figure 1D ). However, one of the critical residues at the interface of the MNNL region (22), His, was substituted with Asn in all bony fishes ( Figure 1D ), suggesting that the binding ability of MNNL of Dll4 in bony fishes to Notch is likely to be reduced. In contrast, the Dll1 genes are highly conserved with the DOS motif and unique Pro residues in the C–C loop of the MNNL region ( Figure 1D ).

The Dll4 gene found in coelacanth—the famous lobe-finned fish—surprisingly encodes both the DOS motif and His in the MNNL region, which is different from those in terrestrial organisms and bony fishes ( Figure 1D ). We previously showed that a murine Dll4-derived chimera with Dll1-derived first and second EGF-like repeats containing the DOS motif exhibited stronger activity to trigger Notch signaling than the original Dll4 (21). Therefore, Dll4 in coelacanths should induce a stronger Notch signal than in other species. It is understood that ray- and lobe-finned fish that evolved into tetrapods share a common ancestor, and coelacanths have shown a slow rate of molecular and morphological evolution (23). As Dll4 in coelacanths is predicted to show intermediate characteristics between those of tetrapods and bony fishes, it was estimated that the Dll4 gene first appeared in a form similar to the coelacanth one in the common ancestor of lobe- and ray-finned fishes and changed to its respective forms in terrestrial organisms and bony fishes during evolution.

The thymus, a primary lymphoid organ essential for T-cell development, emerged in jawed vertebrates approximately 500 million years ago (9). As the Dll4 gene has never been identified in cartilaginous fish (Callorhinchus milii, Figure 1D ), Dll1 is likely to function as a Notch ligand on the thymic epithelium in gnathostomes ancestors. However, it is unclear whether Dll1 can function as a Notch ligand on the thymic epithelium and support T-cell development in the thymus, and which Notch receptor actually interacts with Dll1 in thymic immigrants.

To explore the ability of Dll1 to trigger Notch signaling in the thymic epithelium, we used conditional transgenic (Tg) mice, in which one copy of the Dll1 or Dll4 gene was transcribed by the CAG promoter after Cre-dependent gene deletion of floxed GFP cDNA at the Rosa26 locus (hereafter referred to as iD1 and iD4 Tg) (21). In these Tg mice, we were able to detect the expression of GFP in EpCAM⁺ PDGFRα⁻ thymic epithelial cells derived from the fetal murine thymus ( Figure 2A ), indicating that the CAG promoter substantially transcribed the inserted gene cassette containing Dll1 or Dll4 cDNA in the Rosa26 allele. A small difference in fluorescence intensity could be due to the difference between Dll1 and Dll4 cDNA sequences because one copy of either of the cDNAs was inserted into the same site of the Rosa26 locus. After breeding Foxn1-Cre and Dll4-floxed mice, consistent with GFP expression, exogenously expressed HA-tagged Dll4 or Dll1 was detected in cytokeratin⁺ thymic epithelium ( Figure 2B ). Thus, we examined the effect of exogenous expression of Dll4 or Dll1 on T-cell development under endogenous Dll4-deficient thymic conditions in which T-cell development has been completely impaired (4, 5). The reduction in total cell numbers observed in the Dll4-deficient thymic lobe was completely reversed by the expression of exogenous Dll4 or Dll1 ( Figure 2C ). Consistent with the cell numbers, efficient T-cell development was detected with exogenous Dll4 or Dll1 without endogenous Dll4 in the thymic epithelium ( Figures 2D , E and Supplementary Figure 2 ). Thus, Dll1 can function as a Notch ligand to support T-cell development in the thymus. However, it remains unclear which Notch receptor binds to exogenous Dll1 and mediates signal transduction in T-cell progenitors in these Tg mice.

In mammals, of the four Notch receptors identified, three—Notch1, Notch2, and Notch3—have been detected in the blood cells (24). The phenotypes of conditional KO mice demonstrated that Notch1 and Notch2 mainly contribute to the development of hematopoietic cells (3, 25). Here, we confirmed the expression of Notch receptors on pre-thymic T-cell progenitors in the murine bone marrow ( Figure 3 ). As a population that includes hematopoietic stem cells, weak but detectable expression of Notch1 and high expression of Notch2 were detected in lineage marker-negative c-kit⁺Sca1⁺ (KSL) cells. During differentiation toward the lymphoid lineage, the Notch1 expression increased, but that of Notch2 decreased, and both receptors were clearly detected in the common lymphoid progenitors (lineage marker-negative c-kit^lowSca1^lowIL-7R⁺). These profiles were consistent with those obtained in a similar population of the murine fetal liver (26). We did not observe Notch3 and Notch4 expression in these progenitors. In addition, both Notch1 and Notch2 are co-expressed in the earliest stage of T-cell progenitors in the thymus (early T-cell progenitors) (16, 27, 28). These results suggest that thymic immigrants express both Notch1 and Notch2 on their surfaces and receive Notch signaling via both receptors. We have reported that signal transduction from the active intracellular fragment of Notch1 or Notch2 is sufficient for the initiation of T-cell lineage development (29) and that Notch2 complements Notch1 to mediate inductive signaling for T-cell development in pro-T stages (27). Therefore, we expected that both Notch1 and Notch2 could contribute to the T-cell development in the thymus, especially with exogenous Dll1 that preferentially stimulates Notch2-mediated signal transduction in other cell types (18, 19).

To reveal functional Notch receptors for exogenous Dll4 and Dll1 on T-cell progenitors, we prepared BM chimeras in iD1/iD4 Tg mice with Notch1- or Notch2-deficient bone marrow cells and examined their T-cell development in the thymus. Notch2-deficient BM cells were prepared from Notch2 ^f/f mice (23) with Cre/ERT2 knock-in allele in the Rosa26 locus (30) after tamoxifen administration and then transferred into irradiated iD1 or iD4 Tg mice with a Dll4-deficient background ( Supplementary Figure 3A ). Six weeks later, T-cell development in the thymus with Notch2-deficient T-cell progenitors was completely rescued by exogenous Dll4 and Dll1 and was comparable to that in WT mice ( Figures 4A, B ). In BM chimeras with Notch2-deficient BM cells, CD21^highCD23^- splenic marginal zone B (MZB) cells disappeared selectively ( Figure 4C ), which is consistent with previous reports using mice with B-cell-specific deletion of Notch2 (25) or systemic disruption of Dll1 (19). These results suggest that both Dll1 and Dll4 can trigger Notch signaling via Notch1 in the thymus.

Next, we prepared BM chimeras with Notch1-deficient BM cells. However, systemic depletion of Notch1 affects the survival of mice, and it is difficult to obtain Notch1-deficient BM cells. Thus, we performed sequential transplantation of the BM ( Supplementary Figure 3B ). First, irradiated WT host mice (CD45.2⁺) were reconstituted with BM cells derived from Notch1 ^f/f mice (31) with Cre/ERT2 knock-in allele in Rosa26 locus; then, they were treated with tamoxifen. After a week of the last tamoxifen treatment, Notch1-deficient BM cells were secondarily transferred into iD1/iD4 mice with a Dll4-deficient background. In that case, it was difficult to control the efficiency of thymopoiesis reconstitution. Therefore, we used GFP⁺ BM cells as an internal control at the first transplantation and evaluated T-cell development relative to the GFP⁺ control in secondary BM chimeras. In these experiments, the majority of Notch1-deficient BM cells differentiated into CD19⁺ B-lineage cells in the thymus under control and Dll4-deficient conditions ( Figure 5A ). Moreover, similar developmental patterns were observed in iD4 mice, indicating that Dll4 does not support Notch2-mediated T-cell development in the thymus ( Figure 5A ). In contrast, Notch1-deficient BM cells were able to differentiate into T-lineage cells, including CD4/CD8 double-positive (DP) and single-positive (SP) cells, in the thymus of iD1 mice, but not into B-lineage cells—the default phenotype in the absence of Notch signaling ( Figure 5A ). These phenotypes were also confirmed in the inguinal lymph nodes ( Supplementary Figure 4 ). However, T-cell development of Notch1-deficient cells supported by Dll1 may be less efficient than that of WT cells because, in some cases, the ratios of the number of the DP cells derived from Notch1-deficient cells were lower than those from the GFP⁺ internal control ( Figure 5B ). We observed spontaneous differentiation of Thy1.2⁺ T-lineage cells in the Dll4-deficient thymus. This phenotype would be caused by the inefficient differentiation of B cells in the sequential BM transplantation experiments. In some cases, GFP⁺ cells did not efficiently differentiate into SP cells, which might be due to the excess expression of GFP. These results suggest that Dll1 on the epithelial cells, but not Dll4, interacts with Notch2 on the immigrant cells in the thymus and retains its superiority over Dll4 for induction of T-cell development via Notch2.

After estimating the time of emergence of the two Dll molecules, we formed a phylogenetic tree of Notch1 and Notch2 genes to determine when the Notch receptors evolved to comprise multiple molecules ( Supplementary Figure 5 ). Like with the emergence of the Dll4 gene in coelacanths and bony fishes, the Notch2 gene was first recognized in cartilaginous fishes and has been passed on to coelacanths and bony fishes. On the other hand, the Notch1 genes, like the Dll1 genes, were identified in all vertebrates, including lampreys and amphioxus. Notably, the Notch1 gene in the elephant shark (Callorhinchus milii) contains shorter amino acid residues (1950 aa) than other Notch1 homologs (2531 aa, mouse; 2508 aa, coelacanth; 2437 aa, zebrafish) and 18 EGF-like repeats (typically 36 in other species) in the extracellular region. These characteristics are unlikely to be sufficient for ligand binding. In contrast, the Notch2 homolog in elephant sharks seems functional, suggesting that only the Notch2-like receptor can be expressed as a functional receptor in thymic progenitor cells in elephant sharks. Therefore, in the thymus of elephant sharks, only the interaction between Dll1 and Notch2 is expected to transduce Notch signaling essential for T-cell development. In this study, we demonstrated that Dll1 triggers Notch signaling via Notch2 to induce the development of mature T cells, and it is the only kind of Dll-mediated Notch signaling present in elephant sharks, where the thymus primordium was first observed.

We performed an evolutionary analysis using the maximum likelihood method. The evolutionary history was inferred using the maximum likelihood method and the JTT matrix-based model (41), and the tree with the highest log likelihood (−17,502.39 for Dll, −55,656.23 for Notch) is shown. The percentage of trees in which the associated taxa were clustered together is shown next to the branches. Initial trees for the heuristic search were obtained automatically by applying neighbor-join and BioNJ algorithms to a matrix of pairwise distances estimated using the JTT model and then selecting the topology with a superior log likelihood value. The tree was drawn to scale with branch lengths measured as the number of substitutions per site. This analysis involved 20 Dll and 18 Notch amino acid sequences. We obtained a total of 937 Dll and 2,848 Notch positions in the final dataset. Evolutionary analyses were conducted using MEGA11 software (42, 43).

Dll4^f/f Foxn1-cre mice have been described previously (4). These mice were bred with iD1 or iD4 transgenic mice that retained the CAG promoter-driven floxed GFP and Dll1 or Dll4 cDNA at the Rosa26 locus (21). Notch1^f/f Rosa26^CreERT2 or Notch2^f/f Rosa26^CreERT2 mice with the CD45.1 allele (16, 25, 30, 31, 44) were maintained in our laboratory. To delete the floxed genes, we administered tamoxifen (100 mg/kg) to the mice by i.p. injection four times on separate days. One week after treatment, BM cells were obtained and used as a source of Notch1- or Notch2-deficient hematopoietic cells for transplantation. GFP transgenic mice were originally established by our group (45). All mice were maintained under specific pathogen-free conditions, and the animal experiments were approved by the Animal Experimental Committee (Tokai University, Kanagawa, Japan).

For BM reconstitution experiments with Notch2-deficient BM cells, semi-lethally irradiated (6 Gy) Dll4^f/f (Cont), Dll4^f/f FoxN1-Cre (D4KO), iD4 Dll4^f/f FoxN1-Cre (iD4/D4KO), or iD1 Dll4^f/f FoxN1-Cre (iD1/D4KO) mice (CD45.2) were transplanted intravenously with BM cells from age-matched Notch2^f/f (1 × 10⁷, CD45.1, represented as WT in the figure) or Notch2^f/f Rosa26^CreERT2 mice (1 × 10⁷, CD45.1, represented as Notch2 KO), who had been administered tamoxifen 1 week before the experiments and analyzed at 6 weeks after reconstitution. For those with Notch1-deficient BM cells, semi-lethally irradiated C57BL/6 mice were transplanted with BM cells from age-matched Notch1^f/f (5 × 10⁶, CD45.1, represented as WT in the figure) or Notch1^f/f Rosa26^CreERT2 (5 × 10⁶, CD45.1, represented as Notch1 KO) mice, with GFP-transgenic (5 × 10⁶, CD45.2) mice as an internal control. After 4 weeks, the mice were administered tamoxifen, and 1 week after the treatment, BM cells were prepared for secondary transplantation into the recipient mice as described above.

For flow cytometric analysis, the following monoclonal antibodies (mAbs) and reagents were used: BV650-CD4 and PE-Cy7-CD3 (BD Biosciences, Tokyo, Japan); PE/Cy7-CD4, BV510-CD4, APC-CD8, APC/Cy7-CD8, Alexa700-CD8, FITC-CD11b, PerCp/Cy5.5-CD21, PE-CD23, PE-CD45.1, PE-CD45.2, FITC-B220, APC-B220, Pacific Blue-Thy1.2, Pacific Blue-c-Kit, APC-IL7Rα, PE-EpCAM, FITC-Gr-1, PE/Cy7-Sca-1, PE-Hamster IgG, PE-Notch1, PE-Notch2, PE-Notch3, and PE-Notch4 (BioLegend, Tokyo, Japan); and PerCp/Cy5.5-CD4, PerCp/Cy5.5-CD19, PE-Thy1.2, APC-Thy1.2, APC-PDGFRα, and FITC-TER119 (Thermo Fisher Scientific, Tokyo, Japan). PE/Cy7-CD19 and APC/Cy7-CD45.1 (Thermo Fisher Scientific). Stained cells were measured using FACSVerse (BD Biosciences) or FACSFortessa (BD Biosciences). Data were analyzed using FlowJo software (BD Biosciences).

Immunohistochemical analysis was performed as previously described (4). Tissue sections of thymus were stained with FITC-anti-pan-cytokeratin (Sigma-Aldrich, Tokyo, Japan) and rabbit anti-HA Ab (Bio-Rad, CA, US) antibodies. Visualization was performed with Alexa594-anti-rabbit IgG antibody (Thermo Fisher Scientific). The stained slides were observed with LS880 (ZEISS).

Statistical analyses, as indicated in the figure legends, were performed using GraphPad Prism software (version 5.0; GraphPad Software, Inc., San Diego, CA). Differences between the two groups were evaluated using Student’s t-test for parametric samples and Mann–Whitney U-test for non-parametric samples. Statistical significance was set at p<0.05.