RBPJ-dependent Notch signaling initiates the T cell program in a subset of thymus-seeding progenitors.

T cell specification and commitment require Notch signaling. Although the requirement for Notch signaling during intrathymic T cell development is known, it is still unclear whether the onset of T cell priming can occur in a prethymic niche and whether RBPJ-dependent Notch signaling has a role during this event. Here, we established an Rbpj-inducible system that allowed temporal and tissue-specific control of the responsiveness to Notch in all hematopoietic cells. Using this system, we found that Notch signaling was required before the early T cell progenitor stage in the thymus. Lymphoid-primed multipotent progenitors in the bone marrow underwent Notch signaling with Rbpj induction, which inhibited development towards the myeloid lineage in thymus-seeding progenitors. Thus, our results indicated that the onset of T cell differentiation occurred in a prethymic setting, and that Notch played an important role during this event.

T lymphopoiesis in the thymus is contingent on the homing of bone marrow (BM)-derived thymus seeding progenitors (TSPs)[1]. After TSPs enter the thymus, their interaction with thymic stromal cells results in proliferation and commitment to the T cell lineage. A key factor implicated in intrathymic T lineage decisions is Notch signaling[2]. Notch directs T cell specification and commitment[3, 4], and plays a critical role in αβ- vs γδ-lineage bifurcation[5, 6], β-selection[7, 8] and positive selection[9]. However, it is currently unclear whether Notch plays a role prior to thymic entry by initiating T cell differentiation in BM progenitors to generate T lineage competent TSPs. It is currently understood that Notch mediates T lineage commitment by dictating T versus B lineage outcomes[10, 11, 12]. However, whether TSPs first encounter Notch signals and specify to the T cell lineage before or after thymic entry remains unclear. The precise identity of adult TSPs has not been established, but potential candidates include BM-derived lineage (Lin)−Sca-1+c-Kit+Flt-3− hematopoietic stem cells (HSCs), Lin−Sca-1+c-Kit+Flt-3lo multipotent progenitors (MPPs), Lin−Sca-1+c-Kit+Flt-3hi lymphoid-primed multipotent progenitors (LMPPs)[13] and Lin−Sca-1loc-KitloFlt-3hiIL-7Rα+ common lymphoid progenitors (CLPs)[14]. Upon entry into the thymus, TSPs are referred to as early T cell progenitors (ETPs) and are found within CD4−CD8− double negative (DN)1a/b cells[15], which are defined as Lin−CD44+CD25−c-KithiCD24−/lo. ETPs efficiently develop into T cells and have limited B cell potential[15], suggesting that TSPs receive Notch instructive signals in a pre-thymic setting or immediately after thymic entry.

To further elucidate the role of Notch in this regard, here we generated an Rbpj-inducible mouse model, which renders all hematopoietic cells unresponsive to Notch signaling and also allows the establishment of their responsiveness in an inducible and temporally-regulated manner. The system reported known Notch-dependent lineage decisions in the hematopoietic system and allowed us to address the temporal and tissue-specific requirements for Notch during T cell development. We found that Notch provided a key pre-thymic signal for the development of ETPs that could generate T cells in the thymus. In addition, we found that BM LMPPs, which represent the likely candidate for adult TSPs[16], underwent Notch signaling in the BM, preventing myeloid lineage skewing within a subset of LMPPs. These findings establish a pre-thymic role for Notch in directing the generation of T lineage competent TSPs.

Results

RBPJind mice allow for controlled Notch responsiveness.

Genetic ablations of Dll1, Dll4, Jag1, Notch1, Notch2 and Rbpj result in embryonic or neonatal lethality in mice[17, 18, 19, 20, 21, 22]. To overcome these limitations and to allow the induction and temporal control of Notch responsiveness, and based on the fact that RBPJ interacts with all four Notch receptors23, we generated a mouse model that incorporated conditional deletion of Rbpj and inducible expression of a transgene encoding RBPJ. To conditionally delete Rbpj in hematopoietic cells, RBPJf/f mice11 were bred to Vav-iCre transgenic (Tg) mice24, generating RBPJf/fVav-iCre mice (Supplementary Fig. 1a). To induce Notch responsiveness in Rbpj-deficient hematopoietic cells, we generated RBPJ-HA Tg mice, in which expression of an HA-tagged RBPJ transgene is under the control of a tetracycline responsive element. Fibroblasts from these mice showed reverse tetracycline-controlled transactivator (rtTA)- and doxycycline (Dox)-dependent expression of the RBPJ-HA transgene (Supplementary Fig. 1b). ROSA26-rtTA mice, in which expression of rtTA is coupled to that of GFP upon Cre-dependent removal of a loxP-stop-loxP cassette within the ROSA26 locus25, were bred to RBPJ-HA Tg mice, generating TetonRBPJ-HA mice (Supplementary Fig. 1a). RBPJf/fVav-iCre mice were bred to TetonRBPJ-HA mice to generate RBPJf/fVav-iCreTetonRBPJ-HA mice (hereafter RBPJind), in which expression of RBPJ-HA in hematopoietic cells can be regulated through presence or absence of Dox in vivo (Supplementary Fig. 1a).

Conditional deletion of RBPJ in RBPJf/fMx-Cre mice leads to arrest of T lymphopoiesis at the DN1 stage, loss of CD4+ and CD8+ T cells and B cell accumulation in the thymus[11]. Compared to RBPJ-sufficient mice (RBPJf/+Vav-iCreTetonRBPJ-HA; hereafter RBPJCtr), the thymus of RBPJind mice not treated with Dox (hereafter RBPJind-noDox) displayed a block at the CD44+CD25− DN1 stage and a reduction or near absence of c-KithiCD24−/lo DN1a/b cells (Fig. 1a), indicating Notch-RBPJ is required for the generation or maintenance of ETPs[26]. Development of CD4 and CD8 double positive (DP) and single positive (SP) cells, as well as γδ T cells, was abrogated, along with the detection of B220+CD19+ B cells and a significant decrease in thymocyte cellularity in the thymus of RBPJind-noDox mice compared to RBPJCtr mice treated with Dox (hereafter RBPJCtr-Dox mice) (Fig. 1a,b). In RBPJind mice treated with Dox for 6 weeks (hereafter RBPJind-Dox6wk) we detected progression of DN1 cells to CD44+CD25+ DN2, CD44−CD25+ DN3 and CD44−CD25− DN4 stages, an increase in the percentage of DN1a/b cells (~4-fold), the presence of DPs, SPs and γδ T cells, a decrease in the percentage of B cells (~35-fold), as well as a significant restoration in thymocyte cellularity compared to RBPJind-noDox mice (Fig. 1a,b). RBPJind mice treated with Dox for 3 weeks and analyzed 3 weeks after stopping the Dox treatment (hereafter RBPJind-Dox3wk-noDox3wk) once again displayed a block at the DN1 stage, lacked DN1a/b cells almost entirely and lacked DPs, while CD4+ and CD8+ SPs and γδ T cells were still present (Fig. 1a). The percentage of thymic B cells was similar to that in RBPJind-noDox mice, and thymocyte cellularity was decreased compared to RBPJCtr-Dox and RBPJind-Dox6wk mice, but higher compared to RBPJind-noDox mice (Fig. 1a,b).

Figure 1.

RBPJind mice allow for controlled T cell development.

(a) Flow cytometry analysis of the thymic phenotype of RBPJCtr-Dox, RBPJind-noDox, RBPJind-Dox6wk and RBPJind-Dox3wk-noDox3wk mice. Left to right: analysis of the DN compartment (DN gated), the DN1 compartment (DN1 gated), DPs/SPs, γδ T cells (DN gated) and B cells (DN gated). DN gated: gated on CD4−CD8−. DN1 gated: gated on Lin− (CD8, CD3, NK1.1, B220, CD19, CD11b, CD11c, Gr1, Ter119) CD44+CD25−. Data are representative of three independent experiments (n=3 mice per group). (b) Total thymic cellularity of RBPJCtr-Dox, RBPJind-noDox, RBPJind-Dox6wk and RBPJind-Dox3wk-noDox3wk mice showing mean ± standard deviation (n=3 mice per group). *P<0.05, **P<0.01 (two-tailed unpaired t-test).

The organization of cytokeratin 8 (K8)+β5t+ cortical TECs (cTECs) and K5+UEA-1+ medullary TECs (mTECs)[27] was disrupted in the thymus of RBPJind-noDox mice compared to RBPJCtr-Dox mice, while RBPJind-Dox6wk mice displayed a restoration of thymic architecture (Fig. 2), indicating that T lymphopoiesis induced mature mTEC and cTEC differentiation. K5+UEA-1+ mTECs were detected along with K5+K8+ immature cTECs in RBPJind-Dox3wk-noDox3wk mice, while mature β5t+ cTECs were absent (Fig. 2), suggesting that maintenance of mature cTECs was dependent on constant supply of T cell progenitors. The thymus of RBPJind-noDox mice contained B220+ B cells that were not confined to the cortico-meduallry junction (CMJ) or the perivascular space (PVS; indicated by Tomato Lectin+ endothelial cells), as in RBPJCtr-Dox mice, but instead were dispersed throughout the thymus parenchyma (Fig. 2). Similar to RBPJCtr-Dox mice, B cells were restrained to the CMJ and PVS in RBPJind-Dox6wk mice, while in RBPJind-Dox3wk-noDox3wk mice, B cells were localized in the expanded cortical niche, similar to RBPJind-noDox mice (Fig. 2).

Figure 2.

Regulation of T lymphopoiesis in RBPJind mice induces thymic architectural changes.

Immunofluorescence analysis of K5, K8, UEA-1 and β5t (left), UEA-1, β5t and B220 (middle; 10x magnification; scale bars denote 100μm) and β5t, B220 and Tomato Lectin (right, 20x magnification; scale bars denote 50μm) in thymic sections from RBPJCtr-Dox, RBPJind-noDox, RBPJind-Dox6wk and RBPJind-Dox3wk-noDox3wk mice. The white arrows indicate cortex and medulla boundaries. Data are representative of three independent experiments (n=3 mice per group).

Sorted BM Lin−Sca-1+c-Kit+ cells (LSKs) from RBPJind-noDox mice cultured on OP9 cells expressing Delta-like-1 (OP9-DL1) for 12 days without Dox did not differentiate into CD44+/−CD25+ DN2/DN3 cells, respectively, and became CD19+ B cells, in contrast to LSKs from RBPJCtr mice (Supplementary Fig. 1c), indicating that controlled Notch responsiveness in RBPJind mouse thymus was recapitulated in vitro. RBPJind-noDox LSKs cultured with Dox for 12 days developed into DN2/DN3 cells, while B cells were not detected (Supplementary Fig. 1c), similar to RBPJCtr LSKs. Culture of RBPJind-noDox LSKs with Dox for 8 days followed by removal of Dox for 4 days led to loss of DN2/DN3 development, without the emergence of B cells (Supplementary Fig. 1c), suggesting that initial Notch responsiveness eliminated the B lineage potential. RBPJind-noDox LSKs cultured on OP9 cells with Dox for 8 days did not give rise to DN2/DN3 cells (Supplementary Fig. 1d), indicating that RBPJ-HA transgene expression did not induce T cell development in the absence of Notch ligands.

CD4+ T cells, CD8+ T cells and γδ T cells were detected in the spleens of RBPJind-Dox6wk mice, but not in in RBPJind-noDox mice (Supplementary Fig. 2a). These three T cell populations were detected in the spleen of RBPJind-Dox3wk-noDox3wk mice (Supplementary Fig. 2a), indicating that Notch was dispensable for the survival of mature T cells. B220+IgM+CD21hiCD23− marginal zone B (MZB) cells were only detected in RBPJind-Dox6wk mice (Supplementary Fig. 2a), confirming that Notch directs the survival of MZB cells[28, 29]. No significant differences in splenocyte cellularity were observed between the mouse groups analyzed (Supplementary Fig. 2b). These results indicated that the RBPJind system allowed for temporal regulation of Notch responsiveness in vivo and in vitro.

Functional ETPs are absent in the RBPJind-noDox mouse thymus.

Because the number of DN1a/b cells was markedly reduced in RBPJind-noDox mice, we investigated whether maintenance of ETPs required intrathymic Notch signals for their survival or pre-thymic Notch signals for their emergence. To address this, we analyzed the T cell developmental kinetcs in RBPJind-Dox mice. CD44+CD25+ DN2 cells were first detected at day 5 post-Dox (Fig. 3a). CD44−CD25+ DN3 cells were detected by day 7, while CD44−CD25− DN4 cells and CD4+CD8+ DPs appeared robustly by day 11 (Fig. 3a–c), reflecting expected kinetics[2]. Two weeks post-Dox, the distribution of DNs, DPs and SPs in the RBPJind-Dox mouse thymus began to resemble steady-state wild-type thymus (Fig. 3a–c).

Figure 3.

Appearance of DN2 cells in RBPJind mouse thymus is delayed upon induction of Notch responsiveness.

(a-b) Flow cytometry analysis of the day-by-day progression of (Lin− pre-gated) DN (a) and DP development (b) in the thymus of RBPJind-Dox mice after Dox treatment for 1 to 14 days (as indicated). Data are representative of three independent experiments (n=3 mice per group). (c) Percentages of DN subsets and DPs in the thymus of RBPJind-Dox mice after Dox treatment for 1 to 14 days (as indicated), showing mean ± standard deviation (n=3 mice per group).

Because it took 5 days for DN2 cells to appear in the RBPJind-Dox mouse thymus, we investigated whether RBPJind-noDox mice lacked ETPs. Sorted ETPs and LMPPs, which give rise to TSPs and thus ETPs[16], from RBPJCtr mice differentiated into CD44+CD25+ DN2 cells within 1 day and 3 days, respectively, when cultured on OP9-DL4 cells with Dox (Supplementary Fig. 3a), indicating that if ETPs were present in RBPJind-noDox mice, then DN2 cells would have appeared within day 1-3 post-Dox, and that the delay in generation of DN2 cells in RBPJind-Dox mice suggested an absence of ETPs prior to Dox treatment. ~40% of RBPJind-Dox thymocytes were RBPJ-HA+ within 4 hours post-Dox, ~80% by 8 hours, and ~100% by 24 hours (Supplementary Fig. 3b), indicating that the delayed emergence of DN2 cells in the thymi of RBPJind-Dox mice was not due to a lag in RBPJ-HA transgene expression.

To investigate whether the delayed appearance of DN2 cells in RBPJind-Dox mice was due to disrupted thymic architecture, we performed mixed BM chimeras by injecting equal numbers of CD45.2+(GFP−) wild-type and CD45.2+(GFP+, Supplemental Fig. 1a) RBPJind-noDox BM cells into lethally irradiated CD45.1+ wild-type mice. Four weeks after transfer, during which recipient thymic structure was maintained by wild-type donor T cells, recipient mice were Dox-treated and appearance of RBPJind-Dox DN2 cells assessed at day 2, 4 and 6 post-Dox initiation. CD45.2+(GFP−) wild-type and CD45.2+(GFP+) RBPJind-noDox BM cells, including LSK-Flt-3hi LMPPs, were detected at day 0, prior to Dox treatment (Fig. 4a). Recipient thymi displayed proper segregation of K8+ cTECs and K5+ mTECs at day 0, with CD45.2+(GFP−) wild-type cells showing normal development of DN2, DN3 and DN4 cells, while CD45.2+(GFP+) RBPJind-noDox cells were blocked at the DN1 stage (Fig. 4b). Notably, CD45.2+(GFP+) RBPJind-Dox DN2 cells were first detected at day 6 post-Dox, compared to CD45.2+(GFP−) wild-type DN2 cells which were present at all time-points (Fig. 4b). These results indicated that even within a normal thymic structure, RBPJind-noDox cells did not appear to give rise to ETPs.

Figure 4.

Appearance of RBPJind DN2 cells in WT thymus is delayed upon induction of Notch responsiveness.

(a) Flow cytometry analysis of BM chimerism of CD45.2+(GFP+) RBPJind-noDox donor cells and CD45.2+(GFP−) wild-type (WT) donor cells in lethally irradiated CD45.1+ WT mice 4 weeks post-irradiation and BM reconstitution and prior to start of Dox treatment (day 0). (b) Immunofluorescence analysis of host thymic architecture at day 0 (10x magnification; scale bars denote 100μm) and flow cytometry analysis of thymic chimerism of RBPJind-noDox cells and WT cells on Day 0 (top), and flow cytometry analysis of kinetics of appearance of RBPJind-Dox DN2 cells in host mice following 2, 4, and 6 days of Dox treatment (Lin− pre-gated) (bottom). Data are representative of three independent experiments (n=3 mice per group).

Emergence of functional ETPs depends on Notch signaling.

To further investigate whether Notch signaling initiates T cell differentiation pre-thymically, we examined the T lineage potential of the few Lin−CD44+CD25−c-KithiCD24−/lo DN1a/b cells in the thymi of RBPJind-noDox mice. We used stringent criteria for the isolation of c-Kithi DN1a/b cells to exclude c-Kitlo DN1c cells[15] (Supplementary Fig. 4). Sorted DN1a/b cells and BM LSKs from RBPJCtr and RBPJind-noDox mice were cultured on OP9-DL4 cells in the presence or absence of Dox. RBPJCtr LSKs and DN1a/b cells cultured on OP9-DL4 cells gave rise to CD44−CD25+ DN3 cells at day 8 and CD4+CD8+ DPs at day 14, irrespective of Dox (Fig. 5a,b). RBPJind-noDox LSKs on OP9-DL4 cells developed into DN3 cells by day 8 and DPs by day 14 only in the presence of Dox, while RBPJind-noDox DN1a/b cells on OP9-DL4 cells did not develop into T lineage cells despite provision of Dox (Fig. 5a,b). These observations indicated that the few thymic DN1a/b cells in RBPJind-noDox mice were not T cell progenitors, and that the delayed appearance of DN2 cells following Dox treatment could reflect the requiremet to recruit TSPs that had experienced Notch signals prior to thymic entry. To test this, we sorted DN1a/b cells from RBPJind mice treated with Dox for 6 days (hereafter RBPJind-Dox6d) and cultured them on OP9-DL4 cells in the presence or absece of Dox. RBPJind-Dox6d DN1a/b cells differentiated into DN3 cells at day 8 and DPs at day 14 on OP9-DL4 cells supplemented with Dox, in contrast to RBPJind-noDox DN1a/b cells which failed to do so (Fig. 5a,b). RBPJind-Dox6d LSKs differentiated into DN3 cells and DPs on OP9-DL4 cells supplemented with Dox, similar to RBPJind-noDox LSKs (Fig. 5a,b). These results indicated that thymic appearance of functional ETPs requires Notch signaling pre-thymically.

Figure 5.

Notch signaling is required for thymic appearance of T lineage competent ETPs.

(a-b) Flow cytometry analysis of DN3 and DP development from RBPJCtr, RBPJind-noDox and RBPJind-Dox6d thymic DN1a/b cells and BM LSKs cultured on OP9-DL4 with or without Dox for 8 days (a) and 14 days (b). (c) Flow cytometry analysis of DN3 development from RBPJCtr, RBPJind-noDox and RBPJind-Dox6d thymic DN1a/b cells cultured on OP9-DL4 with or without Dox for 8 days. (d) Flow cytometry analysis of B cell development from RBPJCtr, RBPJind-noDox and RBPJind-Dox6d thymic DN1a/b cells and BM LSKs cultured on OP9 for 14 days. Data are representative of three independent experiments (n=2 mice pooled for RBPJCtr and n=8 mice pooled for RBPJind-noDox and RBPJind-Dox6d for each experiment).

We also sorted thymic Lin−CD44+CD25−c-KitloCD24+ DN1c cells and cultured them on OP9-DL4 cells with or without Dox. RBPJCtr and RBPJind-Dox6d DN1c cells did not develop into DN3 cells (Fig. 5c), consistent with observations that these cells have inefficient T cell potential[15]. RBPJind-noDox DN1c cells also did not develop into DN3 cells, albeit CD44loCD25− B cells were detected in the absence of Dox (Fig. 5c). These experiments excluded the possibility that c-Kitlo TSPs with T cell potential entered the thymus of RBPJind-noDox mice in the absence of Notch responsiveness. RBPJCtr, RBPJind-noDox and RBPJind-Dox6d BM LSKs on OP9 cells developed into CD19+ B cells at day 14 (Fig. 5d), while RBPJCtr and RBPJind-Dox6d DN1a/b cells on OP9 did not develop into B cells (Fig. 5d), as expected[15]. RBPJind-noDox DN1a/b cells on OP9 also lacked B cell potential (Fig. 5d). These results indicated that the loss of T cell differentiation from RBPJind-noDox DN1a/b cells was not due to their divergence to the B cell lineage.

RBPJind-noDox DN1a/b cells have myeloid bias.

To determine whether thymic DN1a/b cells from RBPJind-noDox mice had dendritic cell (DC) potential[30], sorted BM LSKs and DN1a/b cells from RBPJCtr, RBPJind-noDox and RBPJind-Dox6d mice were cultured on OP9-DL1lo cells with Dox, which support DC differentiation. RBPJCtr DN1a/b cells differentiated into CD11c+MHC-II+ DCs at day 8 (Supplementary Fig. 5a), consistent with observations that ETPs can generate DCs[30]. RBPJind-noDox and RBPJind-Dox6d DN1a/b cells had reduced DC potentials compared to RBPJCtr DN1a/b cells (~2.5-fold reduction in percentage), while BM LSKs from the different mice developed into DCs equivalently well between each other (Supplementary Fig. 5a). RBPJind-noDox and RBPJind-Dox6d DN1a/b cells on OP9-DL1lo cells with Dox robustly developed into CD11b+MHC-II− cells by day 8, suggesting a strong myeloid potential, while RBPJCtr DN1a/b cells had a limited myeloid potential (Supplementary Fig. 5a). BM LSKs from all mice developed into myeloid cells with similar efficiency (Supplementary Fig. 5a). RBPJind-noDox and RBPJind-Dox6d DN1a/b cells generated less MHC-IIloB220+ plasmacytoid DCs and more MHC-IIhiB220−CD11b+ myeloid DCs than RBPJCtr DN1a/b cells, while BM LSK from all mice gave rise to these subsets with similar efficiency (Supplementary Fig. 5b).

To investigate the transcriptional signature of thymic DN1a/b cells, we performed RNA sequencing analysis on sorted RBPJCtr, RBPJind-noDox and RBPJind-Dox6d DN1a/b cells. Gene expression analysis between samples (≥2-fold different, P<0.05) identifyed 66 genes differentially expressed between RBPJCtr and RBPJind-noDox. RBPJCtr DN1a/b cells had high expression of Notch-target genes (Notch1, Hes1) and T lineage genes (Tcf7, Lck[31, 32]), while RBPJind-noDox DN1a/b cells had high expression of myeloid-specific genes (Mpo, Ctsg, Elane, Prtn3[33, 34]), four genes were enriched in RBPJind-Dox6d compared to RBPJCtr (Mfsd2b, Gata2, Apoe, Asph), and nine genes were enriched in RBPJind-Dox6d compared to RBPJind-noDox (Notch1, Tcf7) (Fig. 6a,b and Supplementary Tables 1–3). Genes that were highly expressed in RBPJCtr DN1a/b cells (Hes1, Tcf7) or RBPJind-noDox DN1a/b cells (Mpo, Elane) were intermediately expressed in RBPJind-Dox6d DN1a/b cells (Fig. 6a), likely due to a mix of Notch-signaled, T cell-competent ETPs and myeloid-specific progenitors in the thymus of these mice. GO analysis[35] on transcripts enriched in RBPJCtr or RBPJind-noDox DN1a/b cells compared to each other determined that pathways involving genes enriched in RBPJCtr included “T cell differentiation” and “αβ T cell differentiation”, while pathways involving genes enriched in RBPJind-noDox included “myeloid cell differentiation” and “myeloid cell homeostasis” (Fig. 6c and Supplementary Table 4). These results suggested that the few thymic DN1a/b cells from RBPJind-noDox mice lacked T cell potential, but were instead strongly biased toward the myeloid lineage.

Figure 6.

RBPJind-noDox thymic DN1a/b cells are myeloid biased.

(a) Heatmap analysis of genes enriched in DN1a/b cells sorted from RBPJCtr, RBPJind-noDox or RBPJind-Dox6d mouse thymi. (b) Genes differentially expressed between thymic DN1a/b cells of RBPJCtr, RBPJind-noDox and RBPJind-Dox6d mice. For gene expression fold-change values, see Supplementary Tables 1–3. (c) GO analysis of biological pathways involving genes enriched in RBPJCtr DN1a/b cells compared to RBPJind-noDox DN1a/b cells, or genes enriched in RBPJind-noDox DN1a/b cells compared to RBPJCtr DN1a/b cells. Full names of the biological pathways are described in Supplementary Table 4. Data are from one independent experiment, where each group was done in duplicates (n=2 mice pooled for RBPJCtr and n=8 mice pooled for RBPJind-noDox and RBPJind-Dox6d for each replicate).

BM LMPPs undergo Notch signaling in RBPJind-Dox mice.

Because RBPJind-noDox mice lacked functional ETPs, we used flow cytometry to examine whether Notch signaling affected BM progenitors with TSP potential, including HSCs, MPPs, LMPPs and CLPs (Supplementary Fig. 6a) in RBPJCtr and RBPJind-noDox mice. CD62L+ LMPPs and Ly6D− CLPs were also analyzed, as these were described to further refine a TSP population[16, 36]. We observed slight but significant decreases in the numbers of MPPs and LMPPs in RBPJind-noDox compared to RBPJCtr, while the numbers of HSCs and CLPs were not significantly different (Fig. 7a). Additionally, we observed significant decreases in the percentage and number of CD62L+ LMPPs in RBPJind-noDox compared to RBPJCtr, while the percentage and number of Ly6D− CLPs were not significantly different (Fig. 7b).

Figure 7.

BM LMPPs undergo Notch signaling.

(a) Total numbers of BM HSCs, MPPs, LMPPs and CLPs in RBPJCtr and RBPJind-noDox mice, showing mean ± standard deviation (n=7 mice per group). (b) Flow cytometry analysis of BM CD62L+ LMPPs and Ly6D− CLPs (left) and percentages and total numbers of BM CD62L+ LMPPs and Ly6D− CLPs (right) from RBPJCtr and RBPJind-noDox mice. Data are representative of four independent experiments (n=7 mice per group; left) and showing mean ± standard deviation (n=7 mice per group; right). (c) qPCR analysis of Hes1 and Notch1 gene expression by BM HSCs, MPPs, LMPPs and CLPs from RBPJCtr, RBPJind-noDox and RBPJind-Dox6d mice, showing mean ± standard deviation. Gene expression levels were normalized relative to β-actin (n=3 mice pooled for each group for each of the three independent experiments). ns, not significant (P>0.05), *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001 (two-tailed unpaired t-test).

We next examined which progenitors up-regulated the expression of Notch-target genes upon gaining Notch responsiveness by qPCR analysis. Hes1 was expressed in RBPJCtr HSCs, MPPs, LMPPs and CLPs, while its expression was low in all progenitor subsets from RBPJind-noDox BM, with LMPPs and CLPs showing significant decreases in RBPJind-noDox compared to RBPJCtr (Fig. 7c). Hes1 expression was not significantly changed in RBPJind-Dox6d HSCs, MPPs and CLPs compared to those from RBPJind-noDox BM, but its expression was significantly increased in RBPJind-Dox6d LMPPs compared to RBPJind-noDox LMPPs (Fig. 7c). Notch1 expression was simiar in all progenitor subsets between RBPJCtr, RBPJind-noDox and RBPJind-Dox6d mice (Fig. 7c). Sorted RBPJind-noDox LMPPs differentiated into CD44−CD25+ DN3 cells on OP9-DL4 cells with Dox, and did not generate more CD19+ B cells on OP9 cells compared to RBPJCtr and RBPJind-Dox6d LMPPs (Supplementary Fig. 6b). RBPJCtr and RBPJind-Dox6d LMPPs and CLPs on OP9-DL4 with Dox differentiated into CD44+CD25+ DN2 cells with similar kinetics as RBPJind-noDox LMPPs and CLPs (day 3) (Supplementary Fig. 6c), suggesting that BM Notch signals do not position LMPPs further ahead the T cell development path. Additionally, limiting dilution analysis of RBPJind-noDox CD62L+ LMPPs and CLPs on OP9-DL4 cells with Dox showed the same T cell progenitor frequencies (1/1.22-1/1.44) as the RBPJCtr counterparts (Supplementary Fig. 7a,b). These results suggested that Notch-unresponsive RBPJind-noDox LMPPs undergo Notch signaling in the BM and can effectively initiate the T cell program.

Notch signaling inhibits myeloid potential in BM TSPs.

To investigate the effect of BM Notch signals in TSPs, we performed single-cell RNA sequencing analysis on sorted LSK-Flt-3hi LMPPs from RBPJCtr (2614 cells), RBPJind-noDox (2729 cells), RBPJind-Dox3d (1268 cells) and RBPJind-Dox6d (2074 cells) mice. t-SNE analysis of combined cells from all 4 mice identified 8 clusters (cluster 1-8) (Fig. 8a), and gene expression levels were analyzed to identify transcripts enriched in each cluster (≥1.5-fold different, P<0.05) (Supplementary Tables 5–10). Clusters 1 and 2 were not enriched in any particular lineage genes and thus were likely undifferentiated, “stem-like” LMPPs (Fig. 8a). Cluster 3 showed enriched expression of common myeloid progenitor (CMP) genes, e.g., Cenpa, Ccnb2, Cdc20 and Tpx2[34] (Supplementary Table 5), Cluster 4 showed enriched expression of conventional DC genes, e.g., H2-Eb1, H2-Aa, Cd74 and H2-Ab1[34] (Supplementary Table 6). Cluster 5 showed enriched expression of genes highly expressed by RBPJCtr thymic DN1a/b cells, e.g., Mn1, Emp1, Cd33 and Smad7 (Supplementary Table 7), suggesting these cells possessed T cell progenitor function. Cluster 6 showed enriched expression of plasmacytoid DC genes, e.g., Isg15, Irf7, Ifit1 and Iigp1[37] (Supplementary Table 8). Cluster 7 showed enriched expression of CLP genes, e.g., Il7r and Rag1, but also B cell genes, e.g., Ly6d and Ebf1[38] (Supplementary Table 9). Cluster 8 showed enriched expression of granulocyte-monocyte progenitor (GMP) genes, e.g., S100a8, S100a9, Ly6g and Lyz2[34] (Supplementary Table 10).

Figure 8.

Notch signaling in BM inhibits myeloid skewing in TSPs.

(a) t-SNE analysis of 8 clusters from total BM LMPPs (n=8685 cells), analysis of Selplg, Ccr9 and Ccr7 gene expression in Sell+ LMPPs (n= 2918 cells), and analysis of Ccr9 and Ccr7 gene expression in Sell+Selplg+ LMPPs from RBPJCtr mice (n=734 cells), RBPJind-noDox mice (n=516 cells), RBPJind-Dox3d mice (n=273 cells) and RBPJind-Dox6d mice (n=498 cells). For genes enriched in clusters 3-8, see Supplementary Tables 5–10. (b) t-SNE analysis of Sell+Selplg+Ccr9+ LMPPs from RBPJCtr, RBPJind-noDox, RBPJind-Dox3d and RBPJind-Dox6d mice, and overall cluster distribution of Sell+Selplg+Ccr9+ LMPPs and Sell+Selplg+Ccr9+Ccr7+ LMPPS from RBPJCtr mice (n=24 and 4 cells, respectively), RBPJind-noDox mice (n=14 cells), RBPJind-Dox3d mice (n=6 and 2 cells, respectively) and RBPJind-Dox6d mice (n=24 and 4 cells, respectively) (c) Heatmap analysis of expression of myeloid differentiation genes in Sell+Selplg+Ccr9+ LMPPs from RBPJCtr mice (n=24 cells), RBPJind-noDox mice (n=14 cells), RBPJind-Dox3d mice (n=6 cells) and RBPJind-Dox6d mice (n=24 cells), and GO analysis of the top 10 biological pathways involving genes enriched in RBPJind-noDox Sell+Selplg+Ccr9+ LMPPs compared to RBPJCtr LMPPs. Full names of the biological pathways are described in Supplementary Table 16. For all genes differentially expressed between RBPJCtr, RBPJind-noDox and RBPJind-Dox6d Sell+Selplg+Ccr9+ LMPPs, see Supplementary Tables 13–15. Names of the biological pathways in the “leukocyte differentiation” cluster and the genes involved are described in Supplementary Table 16. Data are from one independent experiment (n=3 mice pooled for each group).

We next focused on Sell (CD62L)+ LMPPs and analyzed expression of three important thymus-homing genes: Selplg (PSGL-1), Ccr9 (CCR9) and Ccr7 (CCR7)[39, 40]. Total LMPPs and LMPPs expressing Sell, Selplg, Ccr9 or Ccr7 from RBPJCtr and RBPJind-Dox6d mice had significantly higher Hes1 expression compared to cells from RBPJind-noDox mice (Supplementary Fig. 8a and Supplementary Table 11). Selplg expression was not restricted to a particular cluster, but its expression in cluster 5 overlapped with Ccr9 expression, which was largely restricted to cluster 5 (Fig. 8a). Ccr7 expression was not restricted to a particular cluster, but its expression in cluster 5 overlapped with Selplg and Ccr9 expression, and only Sell+ LMPPs from RBPJCtr, RBPJind-Dox3d and RBPJind-Dox6d BM expressed all three genes (Fig. 8a). Of these cells, 100% RBPJCtr, 50% RBPJind-Dox3d and 75% RBPJind-Dox6d cells were located in cluster 5 (Fig. 8b and Supplementary Fig. 8b). Because the proteins encoding these genes contribute to thymus-homing and are expressed by TSPs and ETPs[39, 40, 41], these observations suggested that BM Notch signals in LMPPs induce thymus-seeding capacity. However, Sell LMPPs were still detected in RBPJind-noDox mice (Fig. 8b). Of the 35 Sell LMPPs in cluster 5, more cells were from from RBPJCtr (16) and RBPJind-Dox6d (12) mice and each formed a clear cluster, compared to fewer cells from RBPJind-noDox mice (5), which did not form a defined cluster (Fig. 8b). Of the overall distribution of Sell LMPPs, majority of RBPJCtr cells located to cluster 5 (16/24), while RBPJind-noDox cells showed decreased distribution to cluster 5 (5/14) and increased distribution to cluster 3 (3/24 versus 5/14, respectively), indicating CMP-GMP potential (Fig. 8b). RBPJind-Dox6d Sell LMPPs displayed reduced distribution to cluster 3 compared to RBPJind-noDox (1/24) and, more similar to RBPJCtr, half of RBPJind-Dox6d cells located to cluster 5 (12/24) (Fig. 8b). Distribution of RBPJind-noDox, but not RBPJind-Dox6d, Sell LMPPs to cluster 5 significantly deviated (P<0.05) from RBPJCtr (Supplementary Table 12).

Among the genes up-regulated (≥1.5-fold, P<0.05) in RBPJind-noDox Sell LMPPs compared to RBPJCtr LMPPs (Supplementary Table 13), we detected increased expression of Egr1, which was reported to bias MPPs and LMPPs toward the myeloid fate[42, 43], Klf4 and Klf2, which control differentiation of myeloid progenitors to Ly6C+ and Ly6C− monocytes, respectively[44, 45], and the AP-1 complex factors, Fos, Fosb, Jun and Junb, which control development of specific myeloid lineages[46] (Fig. 8c). Expression of these genes, including Egr1, Klf4 and Klf2, was reduced in RBPJind-Dox6d Sell LMPPs compared to RBPJind-noDox LMPPs, and as such, the transcriptional profile of RBPJind-Dox6d was more similar to RBPJCtr (Fig. 8c and Supplementary Tables 14–15). GO analysis[35] on transcripts enriched in RBPJind-noDox Sell LMPPs compred to RBPJCtr LMPPs determined that pathways involving these genes included “myeloid leukocyte differentiation” (Supplementary Table 16). Thus, Sell LMPPs were biased toward the myeloid fate in RBPJind-noDox mice, suggesting that BM Notch signaling in TSPs acts to suppress myeloid potential.

Discussion

Here we generated RBPJind mice in which hematopoietic cells could be toggled to become responsive to Notch signaling. Using this system, we found that Notch signaling in BM TSPs repressed the myeloid potential of these cells and allowed the up-regulation of a transcriptional program that could coordinate their migration to the thymus. These observations indicated that Notch plays a role in T cell differentiation prior to arrival of TSPs in the thymus.

Mice with Dll4 conditional deletion have fewer ETPs[26], but evidence for pre-thymic Notch signals in directing T lineage differentiation in TSPs was not documented. Recently, it was shown that BM osteoblasts express DLL4 and appeared to provide Notch signals for the generation of Ly6D− CLPs, which are candidate TSPs[36]. In our study, we could not detect functional ETPs in the thymus of RBPJind-noDox mice. Thus, Notch signaling appeared to be required to generate TSPs that give rise to functional ETPs. Notch signaling is thought to be the crucial determinant of T versus B cell decisions, as disruption in Notch signaling led to B cell accumulation in the thymus[10, 11]. However, more recent evidence suggests that inhibition of Notch signaling in ETPs converts them to DCs rather than B cells[30]. In our study, the few thymic DN1a/b cells in RBPJind-noDox mice had no T nor B cell potential, but displayed strong myeloid potential and expressed myeloid-specific genes. This finding is consistent with evidence that balance between Notch and PU.1 is important for T cell versus myeloid fate decisions, respectively[47, 48, 49]. Thus, Notch signaling in DN1a/b cells appears to direct T-myeloid, rather than T-B lineage decisions.

Single-cell RNA sequencing of BM LMPPs showed that Notch signals affected the expression of genes that contribute to thymus-homing of TSPs and thus appearance of ETPs in the thymus. This, together with results from the mixed BM chimera mice, suggests that Notch signaling in BM can contribute to the generation of TSPs. Further analysis indicated that RBPJind-noDox Sell LMPPs had higher expression of myeloid differentiation genes compared to RBPJCtr and RBPJind-Dox6d LMPPS, including Egr1 and Klf4. These genes may be targets of PU.1 and also represent a GMP-specific gene signature[42, 44, 50, 51]. Egr1 deficiency resulted in more ETPs in the thymus[52], and down-regulation of Klf4 was required for full T lymphopoiesis to occur in the thymus[53], suggesting that suppression of these factors may be required for the generation of ETPs capable of T cell commitment. Additionally, Sell LMPPs from RBPJind-noDox mice were biased toward the CMP cluster (cluster 3). BM endothelial cells were reported to express DLL4, which was required to prevent myeloid skewing in as early as HSC and MPP stages[54].

However, because RBPJind-noDox mice lacked LMPPs with efficient thymus-seeding capacity, we cannot exclude the possibility that thymic entry of LMPP TSPs was completely impaired in the absence of Notch responsiveness. In such a scenario, the myeloid bias of RBPJind-noDox DN1a/b cells may be due to presence of CMPs, GMPs or MEPs within the DN1a/b pool[55], and not necessarily conversion of LMPPs to CMPs. This interpretation is still consistent with the idea that lack of BM Notch signals leads to bias of myeloid-committed progenitors within the subset of cells that seed the thymus. Thus, our results indicate that BM Notch signaling inhibits early events of myeloid differentiation in TSPs, and that without these signals, TSPs become fully commited to the myeloid fate upon thymic entry. Altogether, this work revealed an important pre-thymic role for Notch in the generation of T lineage competent TSPs, such that upon thymic entry, ETP functionality is maintained and T cell development can ensue in full.

Methods

Mice.

All mice were bred and maintained in the Comparative Research Facility of the Sunnybrook Research Institute under specific pathogen-free conditions. All animal procedures were approved by the Sunnybrook Research Institute Animal Care Committee and performed in accordance with the committee’s ethical standards.

Generation of TetOS-RBPJ-HA transgenic mice.

Mouse RBPJ coding sequence (CDS)[56, 57] lacking a stop codon and carrying a Kozak consensus sequence as well as EcoRI and SnaBI restriction sites at 5’ and 3’ end, respectively, was PCR-amplified with Platinum Pfx Polymerase (Invitrogen) from a whole thymus cDNA preparation using the following set of primers: RBPJ-F 5’-ATAGCGAATTCGCCGCAACCATGGCGCCTGTTGTGACA-3’ and RBPJ-R 5’-TAATATACGTAGGACACCACGGTTGCTGT-3’. The RBPJ CDS was then cloned into the MIY-II vector using the EcoRI and SnaBI restriction sites. The hemagglutinin (HA) tag was generated by annealing complementary oligonucleotides (HA-tag F:5’-TATTATACGTAACCAGCTACCCATACGATGTTCCAGATTACGCTTGAGGATCCTGCAT-3’ and HA-tag R: 5’-ATGCAGGATCCTCAAGCGTAATCTGGAA CATCGTATGGGTAGCTGGTTACGTATAATA-3’), which incorporated a threonine-serine linker at its 5’ end as well as SnaBI and BamHI sites, and was subsequently annealed into the MIY-RBPJ construct to allow for transgene (Tg) detection and to differentiate it from the endogenously-encoded RBPJ. RBPJ-HA cassette was further subcloned into the pTetOS vector with the use of EcoRI and BamHI restriction sites. pTetOS-RBPJ-HA was digested with SalI and the cassette, which included the β-globin intron sequence and a polyA sequence, was ligated into an XhoI site of the modified insulator-containing pJC13–1 vector. Insulated TetOS-RBPJ-HA construct was linearized with SalI to remove bacterial DNA elements. Transgenic mice were generated by microinjection of the Tg construct DNA into fertilized eggs obtained from the mating of superovulated C57BL/6J females with C57BL/6J males (The Jackson Laboratory line 000664) at the University of Michigan Transgenic facility, Ann Arbor. Founders and the F1 progeny were screened by PCR for copy number of the Tg by comparing it to the Tg copy standards. Mass of Tg DNA was set as a function of number of base pairs of Tg DNA, the haploid content of a mammalian genome (3 x109 bp) and amount of tail DNA available. Copy number standards were prepared at 0.01, 0.1, 1, 10, 25 and 50 copies. Primer sequences used were: RBPJ-HA F: 5’-ATGACGGGGTCATTTACTCC-3’ and RBPJ-HA R: 5’-CAAGCGT AATCTGGAACATC-3’.

In vitro induction of RBPJ-HA expression in TetOS-RBPJ-HA founders.

Transgenic founder- and F1-derived fibroblasts were prepared by either digesting mouse tails with collagenase IV overnight or by separating two layers of ear tissue and allowing the fibroblasts to adhere to plastic-coated plates. In both cases, cells were grown in Dulbecco’s Modified Eagle Medium (DMEM) substituted with 10% fetal bovine serum (FBS), Penicillin-Streptomycin (Pen-Strep) and Hepes-Sodium Pyruvate-Gentamicin mixture. Ten million fibroblasts were transfected with 4 μg of rtTA-containing pTet-DualON vector (Clontech) using Amaxa nucleofector (Lonza, program U23) and cultured in the presence of 1 μg/ml Dox (Clontech) for 72 hours. Cells were collected and stained for HA in Western blots.

Induction of Notch responsiveness.

To induce transgenic RBPJ-HA expression in vivo, 6 to 8 week old RBPJind mice were injected with 2 mg/ml Dox (Sigma-Aldrich) intraperitoneally at time 0 and administered 1 mg/ml Dox in drinking water supplemented with 5% Splenda ad libitum, with water changed twice a week for the duration of the experiment. Mice not receiving Dox were given drinking water with Splenda alone. RBPJ-HA expression was induced in vitro by culturing cells in the presence of 1 μg/ml Dox.

Cell preparation, flow cytometry and cell sorting.

Single-cell suspensions were prepared from mouse BM, thymus and spleen. BM were crushed while thymus and spleen were mashed, then passed through cell strainers while in α-Minimum Essential Medium Eagle (αMEM) supplemented with 20% FBS and 1% Pen-Strep. Erythrocytes were lysed using BD Pharm Lyse™. Single-cell suspensions were stained with antibodies while in Hanks’ Balanced Salt Solution (HBSS) supplemented with 1% bovine serum albumin (BSA) and 2mM EDTA. Antibodies were purchased from BD Biosciences, eBiosciences, or BioLegend: CD45.1, CD45.2, MHC-II, CD44, CD25, CD24, CD117, CD4, CD8, CD3, TCRγδ, B220, CD19, IgM, CD21, CD23, NK1.1, CD11b, CD11c, Gr1, Ter119, Sca-1, Flt-3, CD127, CD62L, Ly6D. Anti-HA antibody was purchased from Roche and anti-rat secondary antibody was purchased from Jackson Immunoresearch Laboratories. Intracellular staining for HA was performed using eBiosciences Foxp3/Transcription Factor Staining Buffer Set. Flow cytometry was performed on LSR II (BD Biosciences) and data analyzed with FlowJo software version 9.9.6. Cell sorting was performed using BD FACSAria IIu.

Immunofluorescence.

Whole thymi were embedded in OCT compound (Tissue-Tek) and snapped frozen in liquid nitrogen. 6 μm tissue slices were obtained using Leica CM3050S and then fixed with 2% paraformaldehyde (Electron Microscopy Sciences) prior to staining. Tissue slices were stained with anti-cytokeratin 5 (Covance), anti-cytokeratin 8 (Troma1), Ulex Europaeus Agglutinin I (UEA-1, Vector Laboratories), anti-β5t (MBL International), anti-B220 (BD Biosciences) or Lycopersicon Esculentum (Tomato) Lectin (Vector Laboratories). Secondary antibodies (anti-rabbit and anti-rat) and streptavidin were purchased from Jackson Immunoresearch Laboratories and BD Biosciences, respectively. Tissue slices were then mounted with Dako fluorescent mounting medium prior to imaging using Zeiss Axiovert 200M.

BM chimera.

1 million whole BM cells from CD45.2+(GFP+) RBPJind mice and CD45.2+(GFP−) wild-type mice were mixed and injected into CD45.1+ wild=type hosts that were lethally irradiated at 900 rads. After irradiation and injection, hosts were left for 4 weeks to allow wild-type donor cells to maintain the host thymus. Afterwards, hosts were treated with Dox to assess appearance of DN2 cells from RBPJind donor cells.

Cell culture.

BM-derived LSKs, LMPPs and CLPs, and thymus-derived DN1a/b and DN1c cells were purified by flow cytometric cell sorting and cultured on either OP9-DL4, OP9-DL1lo or OP9 cells in αMEM media supplemented with 20% FBS (Gibco) and 1% Pen-Strep, as well as 1 ng/ml of IL-7 and 5 ng/ml of Flt-3L for T-/B- assays, and 100 ng/ml of Flt-3L for myeloid/DC assays (R&D Systems). For T-/B-/myeloid/DC assays from thymic DN1a/b and DN1c cells, ~500 cells were used for each lineage assay for each experiment, with the same numbers used for BM LSK controls. For DN2 differentiation kinetics from ETPs, LMPPs and CLPs, ~2000 cells were used for each day of assessment for each experiment. For single-cell and limiting dilution analysis assays, BM CD62L+ LMPPs and CLPs were sorted onto OP9-DL4 cells in 96 well plates at the following doses: 100 cells (12 wells), 30 cells (12 wells), 10 cells (24 wells), 3 cells (24 wells), 1 cell (48 wells), and the T cell progenitor frequency was calculated using ELDA software version 1[58].

Quantitative RT-PCR.

mRNA from sorted BM progenitors were extracted using TRIzol Reagent (ThermoFisher Scientific) and purified. cDNA synthesis and qRT-PCR reaction was performed as one-step using Luna Universal One-Step RT-qPCR Kit, and data was collected using Eppendorf Mastercycler Realplex2. Primers for the following genes were used: Notch1 (F: 5’-AGATCGACAACCGGCAATGT-3’ and R: 5’-CCCACAGCCCACAAAGAAC-3’), Hes1 (F: 5’-TCCTGACGGCCAATTTGC-3’ and R: 5’-GGAAGGTGACACTGCGTTAGG-3’), β-actin (F: 5’-GGCTCTTTTCCAGCCTTCCT-3’ and R: 5’-GTCTTTACGGATGTCAACGTCACA-3’). Normalized relative expression of Notch1 and Hes1 was determined using β-actin expression as a housekeeping gene.

RNA sequencing.

Thymic DN1a/b cells were RNA sequenced using NextSeq, and data was analyzed using R software version 3.5.1 with the package edgeR version 1[59]. RNA sequencing products were aligned to the mouse genome (GRCm38) to obtain raw read counts for genes, in which genes not expressed across all samples were removed. Gene expression was normalized between samples to account for variations in library size and sequencing depth. Unsupervised clustering of samples was done, and differential expression analysis of genes was performed after filtering for those that showed at least 2-fold changes between samples, and that were statistically significant[59].

Single-cell RNA sequencing.

BM LMPPs were RNA sequenced at single-cell resolution using 10X Genomics, and data was analyzed using R software version 3.5.1 with the package Seurat version 2.4[60]. Each sample, from a total of 2739 (RBPJCtr), 2808 (RBPJind-noDox), 1319 (RBPJind-Dox3d), 2166 (RBPJind-Dox6d) cells, was first filtered to remove cells with low gene counts that arise from aborted sequencing, and gene expression was normalized between cells. Afterwards, variable expression of genes was determined. All samples were then merged and aligned, and dimensions for t-distributed stochastic neighbor embedding (t-SNE) were calculated to identify unique cell clusters. Cell subsetting based on gene expression was done to identify and analyze TSP populations to determine their cluster location and differential gene expression between samples[60].

Statistical analysis.

The data and error bars are presented as mean ± standard deviation (SD). To determine statistical significance, a two-tailed unpaired t-test was performed using Prism software version 6. Statistical significance was determined as: ns (P>0.05), *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001. Tests of significance for RNA sequencing and single-cell RNA sequencing data was performed using empirical Bayes moderated t-statistics and Wilcoxon rank sum test, respectively, under R software version 3.5.1, where P<0.05 was considered significant.

Reporting Summary.

Further information on research design is available in the “Life Sciences Reporting Summary” linked to this article.

Data Availability

The data that support the findings of this study are available from the corresponding author upon request. Raw and processed RNA sequencing and single-cell RNA sequencing data are available at the Gene Expression Omnibus database under accession number GSE128964.