Single-cell multiomics identifies Tcf1 and Lef1 as key factors : initiating early thymic progenitor fate

Abstract

Bone marrow-derived multipotent hematopoietic progenitors seed the thymus and generate early thymic progenitors (ETPs). However, the factors governing ETP formation remain poorly defined. Using scRNA-seq and scATAC-seq, we dissected the heterogeneity of transcriptomic and chromatin accessibility landscapes in ETPs. While Tcf1⁻ ETPs exhibited higher proliferative capacity, Tcf1⁺ ETPs appeared to be immediate, more robust precursors to T-lineage-specified early thymocytes. Pre-thymic ablation of Tcf1 and its homolog Lef1 severely impaired ETP formation in vivo. Whereas ablating Tcf1 alone had limited impact, loss of both Tcf1 and Lef1 impaired transcriptional activation of Notch1 and Notch pathway effector molecules including Hes1 and Hhex, accompanied by aberrantly induced B-cell and myeloid gene programs. Acute deletion of both factors compromised Notch pathway, glycolysis and T-cell gene programs in emergent ETPs ex vivo. Thus, Tcf1 and Lef1 act upstream of the Notch pathway, functioning as pre-thymic initiators of ETP fate and intrathymic gatekeepers of ETP identity and T-lineage potential.

One sentence summary:

Tcf1 and Lef1 prime pre-thymic progenitors for Notch responsiveness, promoting early thymic progenitor fate and lineage stability.

INTRODUCTION

Effective production and export of T lymphocytes from the thymus are essential for immune defense against various pathogens and malignantly transformed cells. Bone marrow (BM)-derived thymus-seeding progenitors migrate to the thymus, and through interaction with thymic stromal cells, they undergo sequential processes of T-cell lineage specification and commitment (1, 2). The earliest thymocytes with T-cell lineage potential reside in the lineage-negative, CD4⁻CD8⁻ double negative (DN) compartment, with a CD44⁺CD25⁻ phenotype known as DN1 cells (3). The DN1 thymocytes are heterogenous, with c-Kit⁺ subset considered as early thymic progenitors (ETPs), which give rise to T-lineage cells with higher efficacy (4–7). ETPs remain multipotent, with the ability to generate non-T lineage cells, including B cells, dendritic cells (DCs), myeloid cells and innate lymphoid cells (ILCs) (8, 9). Such multipotency is partly sustained in c-Kit⁺ CD44⁺CD25⁺ DN2 cells but is largely relinquished in CD44⁻CD25⁺ DN3 cells, which have a rearranged Tcrb gene locus and are committed to the T-cell lineage (8).

Notch signaling is essential for establishing ETPs and promoting T-cell lineage specification and commitment (4, 10). Notch signals also play pivotal roles in suppressing the non-T lineage potentials in ETPs (11–14). For example, Hes1, an effector transcription factor downstream of Notch signaling, represses Cebpa expression in ETPs and prevents myeloid lineage deviation (15). Thymus-seeding progenitors have been considered to receive first Notch instructive signals in the thymus from delta-like ligands (DL) expressed on thymic epithelial cells (16, 17). Recent work indicates that Notch signaling occurs in a subset of BM progenitors such as lymphoid-primed multipotent progenitors and hence constitutes a pre-thymic, initiating event leading to generation of T lineage-competent thymus-seeding progenitors and ETPs (18). While the importance of Notch signals is well-appreciated, little is known about factors that control Notch signaling components and generation of ETPs.

Differentiation of ETPs to T lineage-committed DN3 cells is accompanied and guided by dynamically expressed transcription regulators (8). In-depth single cell analyses of ETPs with DN2 and DN3 thymocytes support a “multilineage priming” model where ETPs express hematopoietic stem/progenitor cell (HSPC) legacy transcription factors (such as PU.1 and Bcl11a), together with T lineage-characteristic transcription factors (such as Tcf1 and Gata3), albeit at lower levels (19, 20). Following the T lineage commitment process, HSPC legacy genes are progressively downregulated, while Tcf1 and Gata3 are more robustly induced, leading to transactivation of Bcl11b and T cell identity genes in DN2 and DN3 thymocytes (21–24). The induction of Tcf1 is mediated by active Notch signals during ETP to DN2 transition and is essential for activation of T-lineage gene program (23–26). Lef1, a homolog of Tcf1 in the HMG family of transcription factors, is induced to a higher level later during DN2 to DN3 transition and shows functional redundancy with Tcf1 in HSPCs, thymocyte maturation beyond the DN3 stage, and mature T cell responses (27–30). However, whether Tcf1 and Lef1 contribute to ETP formation before exerting regulatory roles in driving ETPs toward T lineage commitment in the thymus remains unknown. Coupled with multiomics analyses on single cell and population levels, we utilized pre-thymic gene ablation and temporally controlled targeting to delineate the functional and molecular requirements for Tcf1 and Lef1 in initializing and sustaining ETP fate.

RESULTS

Single cell transcriptomics of DN1 thymocytes distinguishes quiescent and proliferative ETPs

Previous phenotypic analyses have shown that DN1 thymocytes are heterogeneous, containing primitive c-Kit⁺ ETPs together with cells manifesting features of mature T and non-T lineage cells (7); however, their molecular heterogeneity has not been defined. Comparison of ETPs with DN2/3 thymocytes at the single cell level provided insight into transcriptional dynamics during the transition from multipotency to T-cell lineage commitment (20). Therefore, we hypothesized that comparative analysis of ETPs with non-ETP cells within the DN1 compartment might shed new light on mechanisms for ETP fate decision. To test this, we first depleted non-T lineage cells (i.e., cells expressing B220, CD19, TCRγ, NK1.1, CD11c, Mac1, Gr.1, and Ter119) from total thymocytes isolated from WT mice, and sorted Lin⁻TCRβ⁻CD8⁻CD44^hiCD25⁻ DN1 cells for single-cell RNA sequencing (scRNA-seq) (fig. S1A). We obtained 4,434 high-quality single cells from two biological replicates and resolved these into 18 clusters (Fig. 1A). Cells in clusters 1–6 expressed genes that are characteristic of ETPs, including Kit, Cd34, Flt3, Notch1 and Notch1 downstream genes such as Hes1 and Hhex (Fig. 1B, fig. S1B). These clusters also had high expression of HSPC legacy genes including Lyl1, Hoxa9, Bcl11a and Spi1 (encoding PU.1) (fig. S1C). Additionally, clusters 4–6 showed features of proliferation, including elevated expression of Mki67, Lig1 (encoding DNA ligase 1, indicating DNA replication), Aurkb, Birc5, Bulb1b, Cdk1 (encoding cyclin-dependent kinase 1), and various cyclin-encoding genes (Ccna2, Ccnb1 and Ccnb2) (Fig. 1C, fig. S1D).

Figure 1. Single-cell RNA-seq analysis of WT DN1 thymocytes resolves transcriptomic heterogeneity.

A. Uniform manifold approximation and projection (UMAP) plot of scRNA-seq data on sort-purified WT Lin⁻TCRβ⁻CD8⁻CD44^hiCD25⁻ DN1 thymocytes (pooled from two mice), with 18 clusters identified with Seurat and marked in distinct colors.

B–E. UMAP plots showing single-cell transcript levels of characteristic genes in All ETPs (B), proliferative ETPs (C), cells with nonT potentials (D), and T-ILC like cells (E). The color scale denotes log-normalized expression values with arbitrary units.

F. scRNA-seq UMAP showing functional subsets, as denoted with distinct colors.

G. Heat map showing expression of selected genes in DN1 functional subsets as defined in F, with each column corresponding to a single cell and the color scale representing z-normalized transcript levels with arbitrary units.

H. Pseudotime analysis of WT DN1 thymocytes using Monocle3, with the black line denoting the trajectory.

Cells in clusters 7–10 had higher expression of Irf8, which encodes a master regulator of DCs (31, 32) (Fig. 1D). Cluster 7 cells showed distinctive expression of Cebpa (encoding CEBPα) and Elane (20, 33, 34), and higher expression of Mpo (encoding myeloperoxidase), suggesting a myeloid-biased fate (Fig. 1D, fig. S1E). Clusters 8–10 expressed B cell lineage marker genes including Cd19, Blnk (encoding B cell linker adaptor protein), Cd79a and Cd79b (encoding Ig-α and Ig-β), but had few Pax5 transcripts (Fig. 1D, fig. S1E). In our experimental setting, non-T lineage cells were first actively depleted and then excluded in cell sorting to yield DN1 cells. Nonetheless, we cannot completely rule out the possibility that these cells were derived from residual contamination. We cautiously interpreted that clusters 7–10 comprised cells that were not fully committed but retained the potential to differentiate into non-T cells.

Cells in clusters 11–18 all expressed mature T cell characteristic genes in the following categories: 1) cell surface proteins, Thy1, Cd3d, Cd3e and Cd3g, but lacking the DN2/3 marker Il2ra (encoding CD25), 2) signaling molecules, Zap70, Prkcq (encoding PKCθ) and Lck (fig. S2A,B), and 3) transcription factors, Ets1, Tcf7 (encoding Tcf1), Lef1, and Bcl11b (Fig. 1E, fig. S2C). These clusters also expressed innate lymphoid cell (ILC)-characteristic transcription factors, with clusters 11 and 15 showing particularly high expression of Gata3 and Rorc, which promote ILC2 and ILC3 differentiation, respectively (fig. S2D). The remaining clusters expressing higher levels of Id2, Zbtb16 (encoding PLZF) and Il7r, which are associated with ILC progenitors/precursors (35, 36) (fig. S2E). Notably, NK cell-characteristic genes, such as Klf12 and Nfil3, were not detected in high abundance and/or in strong association with specific clusters (fig. S2F).

Based on these molecular features, WT DN1 thymocytes can be divided into 5 functional subsets. Clusters 1–6 cells were ETPs, with clusters 1–3 annotated as quiescent ETPs (“ETP_quiescent”) and the rest as “ETP_proliferative”; clusters 7–10 appeared to have limited T cell potential (“nonT_potential”), likely giving rise to B and myeloid lineage cells; clusters 11–18 had T cell and ILC-like features, with clusters 11 and 15 annotated as “T-ILC_Gata3^hi” and the rest as “T-ILC_Id2^hi” (Fig. 1F, fig. S2G). Each subset had distinctive marker gene expression profile (Fig. 1G, Data file S1). Pseudotime analysis with Monocle⁷⁴ projected a continuum differentiation of ETP cells, with branches toward nonT_potential, T-ILC_Gata3^hi and T-ILC_Id2^hi cells (Fig. 1H). These data provide further insight into the heterogeneity of DN1 thymocytes, with increased resolution of the subset-specific transcriptional profiles.

Single cell profiling of chromatin accessibility in DN1 cells identifies distinctive ETP molecular features

We next used single-cell assay for transposase-accessible chromatin with sequencing (scATAC-seq) to resolve WT DN1 thymocyte heterogeneity based on chromatin accessibility (ChrAcc) profiles. We obtained 6,554 high-quality single cells and resolved these into 19 clusters (Fig. 2A). Using ArchR (37), we computed gene activity scores based on ChrAcc signals flanking each gene and then projected scRNA-seq-based functional subsets to scATAC-seq clusters (Fig. 2B, fig. S2H). While nonT_potential, T-ILC_Gata3^hi and T-ILC_Id2^hi scATAC-seq clusters were distinctive from each other, the ETP_quiescent and ETP_proliferative scATAC-seq clusters appeared to be aggregated together (Fig. 2B). By the criteria of ≥2-fold difference and FDR≤0.01 by Wilcoxon test, we identified marker ChrAcc sites for scATAC-seq functional subsets, and unsupervised clustering analysis showed that marker sites in nonT_potential, T-ILC_Gata3^hi and T-ILC_Id2^hi subsets were clearly distinguishable from each other and from ETP clusters, whereas marker sites in ETP_quiescent and ETP_proliferative subsets exhibited similar ChrAcc state (Fig. 2C). Gene activity scores based on the marker ChrAcc sites showed a similar trend among the scATAC-seq functional subsets (Fig. 2D). These data indicated the ETPs were clearly distinguishable from other DN1 thymocytes based on ChrAcc activity.

Figure 2. Single-cell ATAC-seq analysis of WT DN1 thymocytes resolves heterogeneity in chromatin accessibility.

A. UMAP plot of scATAC-seq data on sort-purified WT DN1 thymocytes (pooled from three mice), with 19 clusters identified with Seurat and marked in distinct colors.

B. UMAP plot showing scATAC-seq functional subsets based on gene activity scores and projection from scRNA-seq functional subsets.

C. Heat map of marker ChrAcc sites in each scATAC-seq functional subset.

D. Heat map of gene activity score based on marker ChrAcc sites in each scATAC-seq functional subset.

E. Heat map of marker ChrAcc sites in regrouped ETP, nonT_potential and T-ILC like subsets using more stringent criteria (≥2-fold difference in one cluster over the other two clusters). The color scales denote z-normalized signal strength (C, E) or gene activity score (D) with arbitrary units.

F. Bar graph showing the odds ratio of link between subset marker gene promoters (as determined with scRNA-seq) and subset marker ChrAcc sites (as determined with scATAC-seq). P values were determined with Fisher exact tests over intersection between marker ChrAcc sites and marker gene-linked sites in each functional subset.

G. Pseudo-bulk ChrAcc tracks at select gene loci in ETP subset, with red arrows indicating ETP marker ChrAcc sites, green dotted lines denoting gene transcription start sites (TSS), and arches denoting coaccessibility of TSS and ChrAcc sites. Arches in red highlight coaccessibility of TSS and ETP marker ChrAcc sites.

We then merged ETP_quiescent and ETP_proliferative cells as ETP subset, and T-ILC_Gata3^hi and T-ILC_Id2^hi cells as T-ILC-like subset. By applying more stringent criteria by requiring ≥2-fold difference and adjusted p-value≤0.001 over the other two clusters, we identified 14,200; 16,174; 15,419 high-confidence marker ChrAcc sites in ETP, nonT_potential, and T-ILC-like subsets, respectively (Fig. 2E). Co-accessibility analysis revealed that the promoters of ETP marker genes (as defined with scRNA-seq, Data file S1) were specifically linked to ETP marker ChrAcc sites with concomitant activities (Fig. 2F, Data file S2), as exemplified at ETP-characteristic genes such as Notch1, Hes1, Cd34 and Flt3 (Fig. 2G). The same was true for the links between marker genes and marker sites for nonT_potential and T-ILC like cells (Fig. 2F, Data file S2). These analyses thus resolved DN1 thymocytes based on single-cell ChrAcc activity, which was closely linked to transcriptional output.

Tcf1 expression differentiates lineage potential of ETP and T-ILC-like cells

Transcriptomic analysis on the single cell level showed that ETPs did not uniformly express the Tcf7 transcript, with ETP_quiescent showing higher Tcf7 expression than ETP_proliferative cells (Fig. 1E). Similarly, T-ILC_Gata3^hi cells showed a trend of higher Tcf7 expression than T-ILC_Id2^hi cells (Fig. 1E). Lin⁻DN1 thymocytes can be clearly distinguished into c-Kit⁺CD3ε⁻ ETP and c-Kit⁻CD3ε⁺ T-ILC-like cells on distinct differentiation paths. Using mice bearing a Tcf7-GFP reporter allele (38, 39), both cell types could be further separated into Tcf7-GFP⁺ and Tcf7-GFP⁻ subsets (Fig. 3A). To assess whether Tcf1 expression was associated with differences in differentiation potential, we purified four thymic subsets and separately plated them on a OP9-DL1 cell monolayer, from which constitutively expressed DL1 ligand provides Notch1-activating signals (40). By tracking Lin⁻ cells derived from the thymic subsets over time, we observed that c-Kit⁺Tcf1⁺ subset-derived Lin⁻ cells mostly showed a DN2 phenotype by day 4 and progressively matured into DN3 cells on days 6 and 8 in culture (Fig. 3B,C, lower panels in B). Conversely, the c-Kit⁺Tcf1⁻ subset showed similar capacity of producing DN2 and DN3 cells but at a consistently slower pace: a substantial portion of c-Kit⁺Tcf1⁻ subset-derived Lin⁻ cells were in DN1 phenotype by day 4, and the frequency of DN3 derived from the c-Kit⁺Tcf1⁻ subset was lower than that from in the c-Kit⁺Tcf1⁺ subset on days 6 and 8 in culture (Fig. 3B,C, top panels in B). In contrast, Lin⁻ cells derived from the CD3ε⁺Tcf1⁺ and CD3ε⁺Tcf1⁻ thymic subsets were all arrested at the DN1 phenotype, with virtually no DN2 or DN3 cells produced at all time points examined (Fig. 3D). These data indicate that both c-Kit⁺Tcf1⁺ and c-Kit⁺Tcf1⁻ thymic subsets are robust ETPs, and further suggest that the c-Kit⁺Tcf1⁻ subset is more developmentally immature and requires additional time for committing to the T cell lineage under the Notch instructing signals.

Figure 3. Thymic DN1 subsets exhibit distinct capacity and kinetics in generating CD25⁺ cells ex vivo.

A. Gating strategy for identifying and sorting of thymic DN1 subsets.

B–C. Detection of DN1-DN3 cells in Lin⁻ cells generated from c-Kit⁺Tcf1⁻ and c-Kit⁺Tcf1⁺ thymic subsets seeded on OP9-DL1 monolayers. At the indicated time periods, CD11c⁻, CD11b⁻, CD19⁻ and NK1.1⁻ lymphocytes were analyzed for CD44 and CD25 expression. Representative contour plots (B) were from 2 independent experiments with 4 biological replicates examined. Cumulative data (C) are means ± s.d. in bar graphs. *, p<0.05; **, p<0.01; ***, p < 0.001 by two-tailed Student’s t-test.

D. Detection of CD44 and CD25 expression in Lin⁻ cells generated from CD3ε⁺Tcf1⁺ and CD3ε⁺Tcf1⁻ thymic subsets seeded on OP9-DL1 monolayers for 4 days.

E, F. Detection of CD11c⁺, CD11b⁺ (E), CD19⁺ or NK1.1⁺ cells (F) generated from the four thymic DN1 subsets seeded on OP9-DL1 monolayers. Representative contour plots were from 2 independent experiments with 4 biological replicates examined.

G. Cumulative data on frequency of CD11c⁺, CD11b⁺, and NK1.1⁺ cells generated from the four thymic DN1 subsets after 4 and 6 days in culture. Data are means ± s.d. Statistical significance for multiple-group comparisons was first determined with one-way ANOVA, and Tukey’s test was used as post hoc correction for indicated pairwise comparison. ***, p < 0.001; ns, not statistically significant.

To substantiate the latter point, we examined Lin⁺ cells derived from the OP9-DL1 culture. Among all thymic DN1 subsets, the c-Kit⁺Tcf1⁻ subset indeed had the strongest capacity to generate CD11c⁺ and CD11b⁺ myeloid lineage cells on day 4 in culture, albeit these non-T lineage potentials were largely suppressed after prolonged culture (Fig. 3E,G). In contrast, no CD19⁺ B-lineage cells were generated from any thymic subsets (Fig. 3F,G), likely due to suppression by the active Notch signals. Interestingly, whereas c-Kit⁺ subsets did not have meaningful contribution to NK1.1⁺ cell production, the CD3ε⁺Tcf1⁻ subset showed more robust capacity of producing NK1.1⁺ cells than the CD3ε⁺Tcf1⁺ subset (Fig. 3F,G). Consistent with higher Id2 and Zbtb16 (encoding PLZF) expression in Tcf1⁻ T-ILC-like cells (Fig. 1E,G), our data suggest that the CD3ε⁺Tcf1⁻ subset is more biased toward NK/ILC-1 fate in the presence of active Notch signals (41–43).

Tcf1 and Lef1 are required for pre-thymic ETP formation

Tcf1 and Lef1 are well-documented to be essential for ETPs to fully commit to the T cell lineage and prevent malignant transformation of early thymocytes, with both cooperative and opposing roles (23, 24, 30, 44); yet, it remains unknown if they are required for ETP generation. We crossed Vav-Cre mice to Tcf7^FL/FL and Lef1^FL/FL mice to achieve pre-thymic ablation of both genes in hematopoietic stem cells (HSCs). Using the same approach, we previously demonstrated that deleting either factor alone did not detectably perturb HSC cell frequency and numbers at homeostatic conditions, or their repopulation capacity of multi-blood lineages (excluding T cells) under regenerative stress or competitive conditions (27). However, ablating both factors with Vav-Cre modestly diminished HSC functions in tertiary recipients after serial bone marrow transplantation (27). Consistent with our previous report (27), Lin⁻Sca1⁺c-Kit⁺ (LSK) cells were increased in frequency and numbers in the BM of Tcf7^−/−Lef1^−/− mice compared to WT littermates (Fig. 4A). Cell surface detection of the SLAM family receptors in LSK cells identifies CD150⁺CD48⁻ cells as the most primitive HSCs with long-term repopulation capacity (45). SLAM HSCs showed decreased frequency among LSKs but remained elevated in numbers by ~1.5 fold in Tcf7^−/−Lef1^−/− mice (Fig. 4A). Sca1 is a known interferon-responsive gene (46, 47), and therefore it should be noted that the increased LSK frequency and SLAM HSC numbers in Tcf7^−/−Lef1^−/− mice do not necessarily reflect elevated multi-lineage potentials. Indeed, we previously observed that WT and Tcf7^−/−Lef1^−/− BM cells reconstituted multiple blood lineages with similar efficacy (except for T cells) in both primary and secondary recipients in a serial transplantation assay (27). Multipotent progenitors (MPPs), identified as Flt3⁺IL-7Rα⁻ LSK cells (48), showed a similar 1.5-fold increase in numbers in Tcf7^−/−Lef1^−/− mice (Fig. 4B). In contrast, common lymphoid progenitors (CLPs), identified as Flt3⁺IL-7Rα⁺ Lin⁻Sca1^medc-Kit^med cells (49), showed ~50% reduction in numbers in Tcf7^−/−Lef1^−/− mice (Fig. 4C). Therefore, combined Tcf1 and Lef1 deficiency expanded HSC and MPP compartments but impeded their differentiation to CLPs.

Figure 4. Pre-thymic ablation of Tcf1 and Lef1 modestly affects hematopoietic progenitors but impairs ETP formation.

A. Detection of LSKs in Lin⁻ BM cells (top) and SLAM HSCs in LSKs (bottom) in Tcf7^−/−Lef1^−/− mice and WT littermates.

B. Detection of LSKs in Lin⁻IL-7Rα⁻ BM cells (top) and Flt3⁺ MPPs in LSKs (bottom).

C. Detection of Sca1^medc-Kit^med cells in Lin⁻IL-7Rα⁺ BM cells (top) and Flt3⁺ CLPs in Sca1^medc-Kit^med cells (bottom). Contour plots in A–C are representative from 4 independent experiments, and cumulative data in bar graphs are means ± s.d. *, p<0.05; **, p<0.01; ***, p < 0.001; ns, not statistically significant by two-tailed Student’s t-test.

D. Total thymocytes numbers.

E. Analysis of DN subsets in thymocytes of WT, Tcf7^−/−, and Tcf7^−/−Lef1^−/− mice following the Lin⁻CD8⁻TCRβ⁻ gate strategy to extract DN cells.

F. Detection of c-Kit expression in DN1 and DN2 thymocytes of WT, Tcf7^−/−, and Tcf7^−/−Lef1^−/− mice. Representative contour plots are from ≥3 independent experiments, and cumulative data are means ± s.d. Statistical significance for multiple-group comparisons in D and F was first determined with one-way ANOVA, and Tukey’s test was used as post hoc correction for indicated pairwise comparison. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

Because ETPs can develop independent of CLPs (6, 50), we directly examined ETP formation in the thymii of Vav-cre⁺Tcf7^FL/FL (Tcf7^−/−) and Tcf7^−/−Lef1^−/− mice. Total thymocytes were reduced by ~20-fold in Tcf7^−/− and ~40-fold in Tcf7^−/−Lef1^−/− mice (Fig. 4D). Consistent with our previous report (30), CD8⁺ or TCRβ⁺ thymocytes were produced in Tcf7^−/− mice but rarely detected in Tcf7^−/−Lef1^−/− mice (Fig. 4E, top and middle panels), indicating severe developmental blocks in the absence of both factors. Tcf1 and Lef1 contribute to Cd4 silencing in mature CD8⁺ T cells (51) and might act similarly in DN thymocytes. We therefore adopted the Lin⁻TCRβ⁻CD8⁻ gating strategy (Fig. 4E, top) for analysis of the DN compartment. We indeed observed variegated CD4 derepression in Tcf7^−/− and Tcf7^−/−Lef1^−/− DN thymocytes (Fig. 4E, middle), which both showed a partial block at the DN1 stage and insufficient induction of CD25 expression for progression to the DN2 stage (Fig. 4E, bottom), consistent with previous observations in germline-targeted Tcf1-deficient mice (30, 52). Among the DN1 thymocytes, c-Kit⁺ ETPs were detected at similar frequencies in Tcf7^−/− mice as in WT mice, albeit the expression levels of c-Kit protein were lower in Tcf7^−/− than WT DN1 cells (Fig. 4F). Due to developmental blocks in Tcf7^−/− mice, ETP cell numbers were even higher in Tcf7^−/− than WT mice (Fig. 4F), suggesting that Tcf1 is not strictly required for generation of ETPs per se. In contrast, c-Kit⁺ ETPs were greatly diminished in frequency and numbers of Tcf7^−/−Lef1^−/− DN1 compartment (Fig. 4F).

By multi-parameter phenotypic analyses, the Thy1 and c-Kit combination provided better resolution in separating WT ETPs as c-Kit^hi Thy1^med DN1 cells expressing high levels of Flt3 and CD24 (fig. S3A,B). This approach confirmed Tcf7^−/− ETPs as a distinct subset despite diminished c-Kit expression and profound reduction of Tcf7^−/−Lef1^−/− ETPs; Tcf7^−/− and Tcf7^−/−Lef1^−/− ETPs both retained high levels of CD24 but showed modest reduction in Flt3 expression (fig. S3A,B). The WT c-Kit^lo Thy1^hi DN1 cells contained CD24^hi and CD24^lo subsets without strong Flt3 expression, while the corresponding population in Tcf7^−/− and Tcf7^−/−Lef1^−/− DN1 cells lost both CD24 and Flt3 expression (fig. S3A,B). Additionally, c-Kit^lo Thy1^lo cells constituted a minor subset in Tcf7^−/− and Tcf7^−/−Lef1^−/− DN1 cells with aberrant changes in CD24 and Flt3 expression (fig. S3A,B). Conversely, c-Kit was highly expressed in almost all WT DN2 cells, but were invariably diminished in Tcf7^−/− and Tcf7^−/−Lef1^−/− DN2 cells (Fig. 4F), consistent with an indispensable role of Tcf1 in promoting ETP to DN2 maturation, as recently demonstrated on the single cell level (19). Notably, the residual DN2 cells detected in Tcf7^−/−Lef1^−/− mice were not derived from cells that had escaped Vav-Cre-mediated excision, as evidenced by the absence of targeted Tcf7 exon 4 and Lef1 exons 7 and 8 in bulk RNA-seq of Tcf7^−/−Lef1^−/− DN2 cells (fig. S3C). Together, these observations suggest that Tcf1 and Lef1 were unexpectedly required for generation of ETPs, and such requirement was only evident after both factors were targeted.

Deletion of Tcf1 alone or in combination with Lef1 perturbs the ETP transcriptome

To determine how Tcf1 and Lef1 regulate ETP formation, we performed CITE-seq (53) on sorted Tcf7^−/−, Tcf7^−/−Lef1^−/− Lin⁻TCRβ⁻CD8⁻CD44^hiCD25⁻ DN1 thymocytes (Fig. 4D, fig. S4A). When pooling these factor-targeted (KO) DN1 cells together with WT DN1 cells (described in Fig. 1), most of Tcf7^−/− cells aggregated closely on a UMAP with WT cells, while Tcf7^−/−Lef1^−/− DN1 cells were largely distributed in distinct clusters (Fig. 5A,B). WT, Tcf7^−/−, and Tcf7^−/−Lef1^−/− DN1 cells were resolved into 23 major clusters (Fig. 5C). Based on the expression of marker genes, WT DN1 cells were detected as ETP, nonT_potential and T-ILC-like functional subsets (Fig. 5D, left), and a majority of Tcf7^−/− cells fitted into ETP clusters. In contrast, ~50% of Tcf7^−/−Lef1^−/− cells populated nonT_potential clusters, the remaining Tcf7^−/−Lef1^−/− cells, along with with Tcf7^−/−cells, formed a distinct KO-specific cluster (Fig. 5D, right). After normalizing cell numbers for each genotype, we calculated the percentage contribution of each genotype to these functional subsets. In the ETP_quiescent (clusters 1–5) and ETP_proliferative (clusters 6–8) subsets, WT and Tcf7^−/− cells showed similar contribution, constituting ~90% of ETPs, whereas the contribution of Tcf7^−/−Lef1^−/− cells was greatly diminished (Fig. 5E), consistent with flow cytometry analysis (Fig. 4E, fig. S3A). Conversely, the contribution of Tcf7^−/−Lef1^−/− cells expanded in nonT_potential cells (Fig. 5E). The KO-specific cluster, exclusively consisting of Tcf7^−/− and Tcf7^−/−Lef1^−/− cells, lacked expression of ETP-characteristic genes such as Notch1 and its downstream factors Hes1 and Hhex, along with Kit, Flt3 and Cd34 (Fig. 5F), but showed robust induction of Klrb1b, Cxcr6, Maf, Rora, and Rorc (Fig. 5G). The latter group of genes was closely linked to ILC differentiation and functions (54–57) and was detectable in the T-ILC-like subset, consistent with their co-clustering on UMAP (Fig. 5D). These single-cell transcriptomic analyses suggest that Tcf1 and Lef1 promote positive regulation of ETP-characteristic genes, while preventing diversion into nonT and T-ILC-like lineages.

Figure 5. Pre-thymic ablation of Tcf1 and Lef1 impairs ETP transcriptional program.

A–B. UMAP plots of scRNA-seq data of WT, Tcf7^−/−, and Tcf7^−/−Lef1^−/− DN1 thymocytes, all combined (A) or individually overlaid on all-cell background (B), as denoted with distinct colors. Cells from three Tcf7^−/− and three Tcf7^−/−Lef1^−/− mice were pooled for the assay.

C. scRNA-seq UMAP plot showing cluster annotation of combined WT+KO single cells. D. scRNA-seq UMAP plots showing distribution of functional subsets of WT cells only (left, KO cells shown in grey) and that of WT+KO single cells (right).

E. Bar graph showing relative proportion of WT and each type of KO cells in the indicated scRNA-seq-based functional subsets after normalized to counts of high-quality cells.

F–G. Violin plots showing transcript levels of ETP-characteristic genes (F) and unique genes repressed by Tcf1 and Lef1 (G).

H. Heat map showing the expression of differentially expressed genes between WT and Tcf7^−/− ETP_quiescent cells by the criteria of ≥1.5-fold difference and padj<0.001. Genes in red denotes those showing the same directional changes between WT and Tcf7^−/− ETP_proliferative cells (fig. S4B). The color scale denotes z-normalized transcript levels with arbitrary units.

Our phenotypic and single-cell transcriptomic analyses consistently demonstrated a requirement for both Tcf1 and Lef1 in ETP formation, while Tcf1 deficiency alone appeared to be less consequential. Because WT and Tcf7^−/− cells contributed similarly to the ETP subsets, we compared transcriptomic profiles of the two genotypes within each ETP subset and found that predominant changes were the induction of effector T cell genes in the Tcf7^−/− ETP_quiescent subset, which included early response genes such as Egr1, Nfkb1, AP1 family (Fos and Junb), and Klf family (Klf2 and Klf6), along with Gzma and Gzmb (Fig. 5H). Some of these changes were also observed in the Tcf7^−/− ETP_proliferative subset (fig. S4B). These data highlight that Tcf1-mediated repression of effector-associated genes was hardwired as early as the ETP stage (29).

Deletion of Tcf1 and Lef1 disrupts the chromatin accessibility landscape of ETPs

We next performed scATAC-seq on Lin⁻TCRβ⁻CD8⁻CD44^hiCD25⁻ DN1 thymocytes to determine how Tcf1 and Lef1 affect the ChrAcc landscape. WT and Tcf7^−/−Lef1^−/− cells were mostly distributed in different clusters on UMAP (Fig. 6A,B, fig. S4C). By projecting scRNA-seq functional subsets based on gene activity scores at ChrAcc sites, the single cells from scATAC-seq analysis of WT and Tcf7^−/−Lef1^−/− DN1 thymocytes were distributed into 4 major subsets including ETP, nonT_potential and T-ILC-like, and Tcf7^−/−Lef1^−/− KO_specific cells (Fig. 6C). Consistent with scRNA-seq and phenotypic analyses, contribution of Tcf7^−/−Lef1^−/− cells to ETPs was profoundly diminished (Fig. 6D), and Tcf7^−/−Lef1^−/− cells contributed less to the T-ILC-like subset but more to the nonT-potential subset (Fig. 6D).

Figure 6. Pre-thymic ablation of Tcf1 and Lef1 perturbs ChrAcc landscape of ETPs.

A–B. UMAP plots of scATAC-seq data of WT and Tcf7^−/−Lef1^−/− DN1 thymocytes, all combined (A) or individually overlaid on all-cell background (B), as denoted with distinct colors. Cells from four Tcf7^−/−Lef1^−/− mice were pooled for the assay.

C. scATAC-seq UMAP plot showing four functional subsets based on projection from scRNA-seq-defined subsets. Note that cells that were not projected to these four subsets were labeled in grey.

D. Bar graph showing relative proportion of WT and Tcf7^−/−Lef1^−/− cells in the indicated scATAC-seq-based functional subsets after normalized to counts of high-quality cells.

E. Detection of non-T lineage markers in DN1 thymocytes using an alternative gating strategy, where Lin⁺ cells were not excluded from the TCRβ⁻CD8⁻ DN thymocytes. The resulting DN1 and CD44^medCD25⁻ subsets were further analyzed for B220, CD19, CD11c and Mac1 expression. Representative contour plots are from ≥3 independent experiments, and cumulative data are means ± s.d. Statistical significance for multiple-group comparisons was first determined with one-way ANOVA, and Tukey’s test was used as post hoc correction for indicated pairwise comparison, *, p < 0.05; **, p < 0.01; ***, p < 0.001.

F. Bar graph summarizing overlapping rates between Tcf1 binding peaks and differential ChrAcc sites in KO_specific cluster as compared to WT ETPs.

G,H. scATAC-seq data presented as pseudo-bulk ATAC-seq tracks resulting from summation of WT ETP and KO_specific single cells at select ETP-characteristic (G) and Tcf1/Lef1-repressed genes (H), along with sequencing tracks of Tcf1 CUT&RUN in OP9-DL1-derived WT ETPs. In G, cyan bars denote ‘more closed’ ChrAcc sites in KO_specific cells, with filled bars highlighting overlap with Tcf1 binding peaks. In H, orange bars denote ‘more open’ ChrAcc sites in KO_specific cells. Green dotted line denotes gene TSS. Arches on the top of tracks denote coaccessibility of TSS and factor-dependent ChrAcc sites, with red ones highlighting coaccessibility with ETP marker ChrAcc sites (defined in Fig. 2G).

To fully assess the biological impact of ablating Tcf1 and Lef1 on suppression of non-T lineage cells in the thymus, we used an alternative flow cytometry analysis strategy where DN subsets were analyzed within TCRβ⁻CD8⁻ thymocytes, without excluding lineage-positive cells. This approach identified an unusually large population of CD25⁻CD44^med cells among Tcf7^−/− and Tcf7^−/−Lef1^−/− DN thymocytes, which contained almost exclusively B220⁺CD19⁺ cells (Fig. 6E, left) and were depleted, hence undetected in the original gating strategy for single-cell analyses (compare with Fig. 4E). A remarkably similar population was also observed upon genetic deletion of a DN-specific Notch1 enhancer (58), highlighting the essential role of Notch1-Tcf1 axis in repressing B-cell lineage potential in early thymocytes. Further analysis of DN1 cells excluding the CD25⁻CD44^med population showed that CD19⁺ B lineage-diverted cells were not evident in the narrowly gated Tcf7^−/− and Tcf7^−/−Lef1^−/− DN1 cells; while a Mac1⁺ but not CD11c⁺ population was elevated in Tcf7^−/− and Tcf7^−/−Lef1^−/− DN1 cells (Fig. 6E, right). In this context, the predisposition toward non-T lineage caused by Tcf1/Lef1 deficiency could have occurred in prethymic progenitors before seeding the thymus; nonetheless, our data support an essential role for Tcf1 and Lef1 in suppressing B and myeloid lineage potentials.

Among DN1 scATAC-seq clusters, Tcf7^−/−Lef1^−/− cells formed a KO_specific cluster besides contributing to nonT-potential and T-ILC-like subsets. Although the cells in the KO_specific cluster were not necessarily direct descendents of Tcf1/Lef1-deficient ETPs, a comparison between KO_specific and WT ETP subsets provided an opportunity to assess the net impact of prethymic deletion of Tcf1 and Lef1 on the ChrAcc landscape. This comparison identified 31,616 sites that were “more closed” and 11,343 sites that were “more open” in the absence of Tcf1 and Lef1 (fig. S4D). As previously mapped with CUT&RUN (26), Tcf1 has 44,799 binding peaks in OP9-DL1-derived ETPs. Tcf1 binding peaks overlapped with >40% “more closed” sites, but only about 10% “more open” sites in the KO_specific subset (Fig. 6F), suggesting a more dominant role of Tcf1 and Lef1 in establishing/maintaining chromatin open state in ETPs. Approximately 75% of “more closed” ChrAcc sites in KO_specific cluster were common with those in WT nonT-potential and T-ILC-like subsets when compared with WT ETPs (fig. S4E), suggesting that intact expression of Tcf1 and Lef1 causatively protects ETPs from diversion to nonT or T-ILC-like lineages. In fact, Tcf1/Lef1 deficiency resulted in “more closed” chromatin state in many sites that were co-accessible with promoters of Notch1, Hhex and Hes1 (Fig. 6G), as well as those of ETP-characteristic genes including Flt3, Cd34 and Kit (fig. S4F). In addition, all these genes had 2 or more linked “more closed” sites showing overlap with Tcf1 binding peaks. These “more closed” sites were enriched in motifs of known Tcf1 partner transcription factors, such as Runx, Ctcf and Heb in the E2 family (51, 59, 60) (fig. S4G). Conversely, among the “more open” ChrAcc sites in the KO_specific subset, about one-third overlapped with “more open” sites in WT T-ILC-like vs. WT ETP comparison (fig. S4H), as observed at the ILC-linked gene loci, including Rorc, Maf, Rora, Cxcr6 and Klrb1b (Fig. 6H, fig. S4I). The “more open” sites in the KO_specific subset had Rorγt (encoded by Rorc) and AP-1 as top enriched motifs besides Ets and Runx (fig. S4G), suggesting that Rorγt and Maf (one of the AP-1 factors) might be key mediators that promoted divergence of Tcf1/Lef1-deficient ETPs to T-ILC-like cell fates. Together, these observations highlight essential roles of Tcf1 and Lef1 in establishing ChrAcc landscape appropriate for ETPs from the prethymic stage, while restricting ChrAcc sites leading to nonT or T-ILC-like lineage deviation in the thymus.

Tcf1 and Lef1 are required for sustaining ETP lineage stability

We noted that residual ETPs, DN2 and DN3 thymocytes did persist in Tcf7^−/−Lef1^−/− thymii (Fig. 4E, 5E) (30), and these cells did not result from incomplete gene excision (fig. S3C). A previous scRNAseq analysis showed that differentiation of Tcf7^−/− DN1 to DN2/3 thymocytes followed two distinct trajectories (61), suggesting potentially independent contributions by BM- and fetal liver-derived progenitors (29). To further define the requirements for Tcf1 and Lef1 in ETP differentiation, we employed ex vivo culture of BM prethymic progenitors with OP9-DL1 cells. CLPs are heterogeneous (5, 62), and CLP numbers were diminished in Tcf7^−/−Lef1^−/− mice (Fig. 4C). To bypass these confounding factors, we seeded WT or Tcf7^−/−Lef1^−/− MPPs on OP9-DL1 monolayers (Fig. 7A). While WT MPPs generated c-Kit⁺ DN1 cells and progressively gave rise to DN2 and DN3 cells after 7–9 days culture, Tcf7^−/−Lef1^−/− cells were completely blocked at the DN1 stage, with near absence of c-Kit⁺ cells (Fig. 7B), validating severely impaired ETP formation ex vivo when Tcf1 and Lef1 were ablated in prethymic progenitors. We then investigated if Tcf1 and Lef1 remained necessary after the ETP program was initiated by instructive Notch1 signals. To this end, we seeded MPP cells from CreER⁺WT and CreER⁺Tcf7^FL/FL Lef1^FL/FL (called CreER⁺dKO hereafter) mice on OP9-DL1 monolayers, and then treated the cells with 4-hydroxytamoxifen (4-OHT) during days 3–5 in culture (Fig. 7C). On day 7 in culture, 4-OHT-treated CreER⁺dKO MPPs failed to produce DN2 cells but generated c-Kit⁺ DN1 cells, albeit c-Kit expression level was substantially lower than those derived from CreER⁺WT cells (Fig. 7D). These observations support an essential role of Tcf1 and Lef1 in maintaining ETP lineage stability, besides known requirements for these molecules in T-cell lineage specification and commitment (23, 24).

Figure 7. Acute deletion of Tcf1 and Lef1 perturbs transcriptomes of emergent ETPs.

A. Experimental design for evaluating pre-thymic deletion of Tcf1 and Lef1 on ETP formation ex vivo.

B. Detection of DN1 and DN2 formation from WT or Tcf7^−/−Lef1^−/− BM MPPs after ex vivo culture on OP9-DL1 monolayers for 7–9 days, with c-Kit expression analyzed on DN1 cells. Representative contour plots are from 2 independent experiments, and cumulative data are means ± s.d. in bar graphs in lower panels. ***, p < 0.001 by two-tailed Student’s t-test.

C. Experimental design for acute deletion of Tcf1 and Lef1 in emergent ETPs.

D. Detection of emergent ETPs from CreER⁺WT and CreER⁺dKO BM MPPs after 7-day culture on OP9-DL1 monolayers, with 4-OHT treatment during days 3–5. Representative contour plots are from 2 independent experiments, and cumulative data are means ± s.d. in bar graphs in lower panels. *, p<0.05; ***, p < 0.001 by two-tailed Student’s t-test.

E. Volcano plot showing DEGs between CreER⁺WT and CreER⁺dKO emergent ETPs sorted on day 7 of culture, by the criteria of ≥1.5 fold changes and FDR<0.05.

F, G. Heat maps showing Tcf1/Lef1-activated genes in glycolysis, Notch and TCR signaling pathways (F), and select differentially expressed transcriptional regulators (G). The color scales denote z-normalized transcript levels with arbitrary units.

H, I. Enrichment plots of hypoxia-induced (H) and -repressed gene sets (I) in comparison of CreER⁺WT and CreER⁺dKO ETP transcriptomes as determined with GSEA, with top 20 genes in the leading edge shown in heat maps. NES, normalized enrichment score; NOM P value and nominal P values were defined in GSEA.

To further define the Tcf1/Lef1-dependent molecular circuit during ETP emergence, we performed bulk RNA-seq on c-Kit⁺ DN1 cells derived from 4-OHT-treated CreER⁺WT or CreER⁺dKO MPPs on day 7 of culture, with high reproducibility (fig. S5A). By the criteria of ≥1.5 fold changes and FDR<0.05, 985 genes showed decreased expression while 1,110 genes showed increased expression in dKO compared to WT emergent ETPs (Fig. 7E, Data file S3), which we refer to as Tcf1/Lef1-activeated and -repressed genes, respectively. Functional annotation with DAVID showed that Tcf1/Lef1-activated genes included Notch1 and several genes in the Notch signaling pathway (Fig. 7F, fig. S5B), suggesting that Tcf1 and Lef1 continued to exert positive regulation on Notch1 itself and sustain responsiveness to Notch ligands in emergent ETPs. Key components in TCR signaling pathway, such as Cd3e, Cd3e, Cd3g, Lck, Itk Plcg1 and Prkcq, and many transcriptional regulators including Bcl11b, Ets1 and Gata3 depended on Tcf1/Lef1 for normal expression (Fig. 7F,G, fig. S5B,C), suggesting that Tcf1 and Lef1 primed these gene loci early on for further induction in the ETP to DN2 transition. Notably, the thymus is considered a physiologically hypoxic organ (63), and gene set enrichment analysis (GSEA) showed that hypoxia-induced genes were downregulated while hypoxia-repressed genes were upregulated in dKO cells (Fig. 7H,I), suggestive of a role of Tcf1/Lef1 in adapting emergent ETPs to the hypoxic thymic environment. Early thymocytes are metabolically more active than single positive mature thymocytes (64). Tcf1 and Lef1 appeared to favor supporting aerobic glycolysis but limited more oxygen-consuming lipid metabolism, mitochondrial activities, and oxidative phosphorylation (Fig. 7F, fig. S6A–C). Tcf1/Lef1-repressed genes included cell cycle regulators (fig. S6D,E), suggesting a role of restraining excessive proliferation of emergent ETPs. Among Tcf1/Lef1-repressed transcription regulator genes were Cepba and Cepbe, which control myeloid differentiation; Ebf1, critical for B cell lineage development; and Gata1 and Gata2, which are essential for erythropoiesis (Fig. 7G, fig. S6F); supporting continued requirements for Tcf1 and Lef1 in guarding emergent ETPs from deviating to non-T lineages. These observations demonstrate that Tcf1 and Lef1 transition into new regulatory roles in emergent ETPs to protect their lineage stability.

With regard to relative expression between Tcf1 and Lef1, scRNA-seq showed that both Tcf7 and Lef1 transcripts were detectable in the ETP_quiescent cluster (Fig. 1E, fig. S2C) with Tcf7 being more abundant. In bulk RNA-seq analysis of emergent ETPs with more robust sequencing depth, Lef1 transcripts were readily detected, albeit Tcf7 transcripts were >150-fold more abundant (fig. S6G). We therefore posit that although expressed at low levels, intact expression of Lef1 remains critical for compensating for loss of Tcf1. This is consistent with our previous studies demonstrating an intrinsic requirement for Tcf1 and Lef1 in maintaining the self-renewal of leukemic stem cells in a chronic myeloid leukemia model (28), where Tcf7 and Lef1 transcripts were both detected at low levels (fig. S6H). Notably, Tcf7^−/− ETPs did not exhibit elevated expression of Lef1 transcripts by scRNA-seq (fig. S6I), suggesting that intact expression of Lef1 is essential for preventing complete loss of ETPs in the absence of Tcf1.

We then mapped the ChrAcc landscape in CreER⁺WT and CreER⁺dKO emergent ETPs using bulk ATAC-seq, with high reproducibility (fig. S7A). By applying the criteria of ≥2 fold change and FDR<0.05, 6,539 sites were more closed in CreER⁺dKO cells, which we refer to as “Tcf1/Lef1-dependent ChrAcc sites” in emergent ETPs. These Tcf1/Lef1-dependent sites had a Tcf/Lef consensus sequence among the top enriched motifs besides Runx, Ets and E2 family transcription factors (Fig. 8A,B). In contrast, fewer sites (1,854 sites) became more open in CreER⁺dKO emergent ETPs (“Tcf1/Lef1-repressed ChrAcc sites”) and were enriched in motifs of CEBP/α and Gata1 factors, which are drivers of myeloid and erythroid lineages (Fig. 8A,B). We then empirically linked the differential ChrAcc sites with DEGs identified above if a site was detected in a DEG gene body and its 100 kb flanking sequence. Over 70% of DEGs had an altered ChrAcc state (fig. S7B), with Tcf1/Lef1-activated genes showing more frequent association with Tcf1/Lef1-dependent ChrAcc sites (Fig. 8C). As observed in group 1 Tcf1/Lef1-activated genes (Fig. 8C), the ETP characteristic Kit, Notch target gene Il2ra, and T-lineage primed kinase Itk and transcription factors including Runx1, Ets1 and Bcl11b were all associated with Tcf1/Lef1-dependent ChrAcc sites in emergent ETPs (Fig. 8D). These observations suggest that Tcf1 and Lef1 sustain chromatin open state and enhancer activity to achieve positive regulation of ETP commitment and T-lineage fate.

Figure 8. Acute deletion of Tcf1 and Lef1 perturbs ChrAcc landscape in emergent ETPs.

A. Volcano plot showing differential ChrAcc sites between CreER⁺WT and CreER⁺dKO emergent ETPs sorted on day 7 of culture, by the criteria of ≥2 fold changes and FDR<0.05.

B. Top motifs in differential ChrAcc sites in A, as determined with HOMER.

C. Heatmaps showing connection of DEGs and differential ChrAcc sites between WT and dKO emergent ETPs. Heat map on the left displays grouping of Tcf1/Lef1-activated and -repressed genes based on their connection with ‘more closed’ and ‘more open’ ChrAcc sites, with gene numbers denoted in parentheses for each group and select genes marked in different colors, where color scale denotes z-normalized transcript levels with arbitrary units. In each group of genes, the associated differential ChrAcc sites were clustered according to the changes in ChrAcc (middle panel), with site numbers denoted in parentheses for each cluster, where color scale denotes normalized signal strength with arbitrary units. In the right panel, the presence of Tcf/Lef motif in the differential ChrAcc sites was detected with Jasper and marked with a horizontal red line, where the percentages denote the frequency of motif presence in each ChrAcc site cluster.

D. ATAC-seq tracks at the indicated gene loci in CreER⁺WT and CreER⁺dKO emergent ETPs with both replicates shown. Cyan and orange open bars denote ‘more closed’ and ‘more open’ sites in dKO cells, respectively.

Conversely, Tcf1/Lef1-repressed genes did not show as pronounced of a preference for Tcf1/Lef1-dependent or -repressed ChrAcc sites (Fig. 8C, fig. S7B). HSC heritage genes such as Tal1, Lmo2, and Hoxa9, as well as non-T lineage characteristic factors including Cebpe, Gata2 and Mpo were associated with Tcf1/Lef1-repressed ChrAcc sites (fig. S7C), suggesting a role of Tcf1 and Lef1 in restraining activation of aberrant enhancers. Coupled with enrichment of Cebpa and Gata1 motifs in the Tcf1/Lef1-repressed ChrAcc sites (Fig. 8B), the induced expression of Cebp and Gata genes (Fig. 7F) likely acted as key secondary mediators in inducing chromatin open state and enhancers linked to non-T lineages. Meanwhile, the proapoptotic Bok and cell cycle regulators including Cdk14, E2f7 and E2f8 were connected with Tcf1/Lef1-dependent ChrAcc sites (fig. S7D). We posit that these Tcf1/Lef1-dependent ChrAcc sites may function as silencers in these gene contexts, similar to a Tcf1/Lef1-dependent chromatin open site that is located –24kb upstream of Prdm1 and contributes to Prdm1 gene silencing in naïve CD8⁺ T cells (65). We also noted that a fraction of DEGs was linked with both ‘more open’ and ‘more closed’ sites, as observed for the Tcf1/Lef1-activated Maml3 (encoding mastermind-like transcriptional coactivator 3 in the Notch signaling pathway) and Tcf1/Lef1-repressed Ebf1 (encoding early B cell factor 1, EBF1, a key regulator of B cell development) (Fig. 8D, fig. S7D). Motif search with JASPAR (66) showed that Tcf1/Lef1-dependent sites had a Tcf/Lef motif at higher frequency than Tcf1/Lef1-repressed sites, regardless of association with Tcf1/Lef1-activated or -repressed genes (Fig. 8C, right column), suggesting that Tcf1 and Lef1 engaged in sustaining open chromatin state more frequently in a direct regulatory manner. These observations suggest that Tcf1 and Lef1 engage both direct (motif-dependent) and indirect (secondary mediator-driven) mechanisms for governance of T cell identity.

Discussion

Through single-cell multiomic and functional analyses of WT DN1 thymocytes, we identified distinct subsets within the c-Kit⁺ ETPs, highlighting previously unappreciated heterogeneity. ETPs consisted of at least two subsets: one primitive, highly proliferative subset with low Tcf1 expression, and a more quiescent subset with high Tcf1 expression. The latter possessed stronger T-lineage bias, progressing to DN2 and DN3 cells at a faster pace while relinquishing non-T lineage potential under active Notch signaling. We propose a model where upon seeding the thymus, c-Kit⁺ hematopoietic progenitors initially undergo active proliferation to expand in number, and the subsequent contact with Notch ligands on thymic epithelial cells initiates the induction of Tcf1, priming T-lineage commitment. In fact, by acutely deleting Tcf1 and Lef1 in emergent ETPs, we established that Tcf1 induction was causatively linked to several key events, including limiting proliferation, adapting to the hypoxic thymic environment, and enhancing glycolytic metabolism. Furthermore, Tcf1 and Lef1 were instrumental in silencing HSC heritage genes and shutting down non-T lineage gene programs in emergent ETPs. Concurrently, both factors primed chromatin accessibility of T-cell program gene loci such as Runx1, Ets1 and Bcl11b in ETPs, ensuring their sustained and/or induced expression in DN2/3 thymocytes. Collectively, these findings provide a refined framework for the sequential events underlying the transition of thymic-seeding progenitors into ETPs and delineate the precise actions of Tcf1 in orchestrating this process.

Beyond its well-established roles in intra-thymic T cell development (10–14), Notch signaling has recently been shown to act in pre-thymic hematopoietic progenitors to promote ETP formation (18). Through pre-thymic targeting, we demonstrated that Tcf1 and Lef1 are essential for generating ETPs, whereas they are dispensable for HSCs and MPPs and only modestly affect CLPs. In this context, Tcf1 and Lef1 are functionally redundant, as the ETP compartment remained largely intact upon loss of Tcf1 alone, aside from aberrant induction of a few effector-associated genes. This conclusion is also supported by recent CRISPR/Cas9-mediated targeting of Tcf1 alone in HSCs, which impaired ETP differentiation into DN2 cells but did not prevent emergence of c-Kit⁺ ETPs in OP9-DL1 cultures (19). Importantly, Tcf1/Lef1 deficiency impaired expression of stem/progenitor-characteristic proteins including the receptor tyrosine kinases c-Kit and Flt3, as well as critical Notch signaling components, such as Notch1 itself, Hes1 and Hhex, the key mediators of Notch signaling outcome. A –9kb Notch1 upstream region has been identified as a DN-specific Notch1 enhancer that augments Notch signaling in ETPs (58), appears to be a high-occupancy target (HOT) region (67, 68) bound by multiple transcription factors, including Tcf1, Runx1 and PU.1 in ETPs (58). Our single-cell multiomic analyses revealed that Tcf1/Lef1 are the key factors that control the open chromatin state for the –9kb Notch1 enhancer in ETPs. Since Notch signaling strongly induces Tcf1 in DN2/3 cells (23, 24), our findings support a model in which Tcf1 and Notch1 form a feed-forward regulatory loop that reinforces T-cell fate decisions and drives T-lineage commitment.

Through pre-thymic genetic ablation in HSCs and temporally controlled deletion in emergent ETPs, our data indicate that Tcf1 and Lef1 are not only essential for establishing ETPs but also for sustaining ETP lineage stability. Both approaches revealed two critical regulatory outcomes by Tcf1 and Lef1, i.e., positive regulation of Notch signal pathways and suppression of non-T lineage programs. Nonetheless, the specific genes and ChrAcc sites affected by Tcf1/Lef1 deficiency were not necessarily identical between these models. The discrepancy may be attributed to differences in the duration of loss of Tcf1 and Lef1 proteins and associated secondary effects. In emergent ETPs treated with 4-OHT ex vivo, changes in gene expression and ChrAcc were likely limited to those sensitive to acute loss of both proteins. In contrast, Vav-Cre-mediated deletion, which resulted in chronic Tcf1/Lef1 deficiency throughout HSC to ETP differentiation process in vivo, likely induced broader and more cumulative changes in their direct and indirect target genes and regulatory elements.

In summary, our systematic analyses identify Tcf1 and Lef1 as pivotal regulators of the inception of T lineage potential at the pre-thymic stage. Although expressed at low levels, they have an indispensable role in enabling hematopoietic progenitors to respond to Notch signaling prior to seeding the thymus. Tcf1 and Lef1 continue to exert essential roles in protecting ETP lineage stability and guarding them against diversion to alternative lineages, while Tcf1 expression rises to halt active cell cycling. These findings reveal that Tcf1 and Lef1 act much earlier than previously recognized, extending beyond their roles in promoting T-cell lineage specification and commitment at the DN2/3 stages.

MATERIALS AND METHODS

Study design.

This study was designed to analyze the molecular heterogeneity of early thymic progenitors (ETPs) and identify key regulatory factors that control ETP formation and sustain ETP lineage stability. We employed complementary models including pre-thymic genetic ablation in vivo and temporally controlled target deletion through ex vivo differentiation systems. The impact was analyzed by RNA-seq and ATAC-seq on both single cell and FACS-sorted population levels coupled with phenotypic characterizations. All experiments were performed independently at least twice to ensure reproducibility from biological replicates, with specific details provided in figure legends. Data from all animals were included for reporting.

Mice.

C57BL/6J (B6) and Vav-Cre transgenic mice were from the Jackson Laboratory. CreER^T2, Tcf7-GFP reporter, Tcf7^FL/FL, and Lef1^FL/FL mice were previously described (27, 30, 38, 69). All compound mouse strains used in this work were from in-house breeding at the animal care facility of Center for Discovery and Innovation, Hackensack University Medical Center. All mice analyzed were 5–7 weeks of age, and both sexes were used without randomization or blinding. All mouse experiments were performed under protocols approved by the Institutional Animal Use and Care Committee of the Center for Discovery and Innovation, Hackensack University Medical Center.

Ex vivo models of T cell development.

OP9-DL1 cells (40) were cultured on 6-well plates in MEM alpha (ThermoFisher Scientific) supplemented with 20% heat-inactivated FBS and 1% penicillin/streptomycin in a 5% CO2 incubator at 37°C. Flt3⁺ LSK cells (MPPs) were sort-purified from the BM of the indicated genotypes and seeded onto OP9-DL1 monolayers at densities ranging from 10–50×10³ cells/well. Co-cultures were maintained in OP9-DL1 culture medium supplemented with 10 ng/ml Flt3 ligand (R&D systems) and 5 ng/ml IL-7 (Peprotech). The medium and cytokines were refreshed on day 4 and replenished every 2 days. For induced deletion of floxed Tcf7 and Lef1 genes, 4-hydroxytamoxifen (T176, Millipore Sigma) was added daily to the culture to a final concentration of 100 nM during days 3–5. The cells were harvested during days 7–9 for phenotypic and molecular analyses.

For assessment of lineage potential of DN1 subsets, DN thymocytes were first enriched by depleting Lin⁺CD8⁺ cells from total thymocytes of Tcf7-GFP reporter mice, Tcf1⁺ and Tcf1⁻ cells were then sorted from CD3ε⁺c-Kit⁻ and CD3ε⁻c-Kit⁺ TCRβ⁻DN1 thymocytes and plated on OP9-DL1 monolayers at densities ranging from 1–5×10³ cells/well for co-culture with the supplements as above. The cells were harvested on days 4, 6 and 8 for phenotypic analysis.

scRNA-seq/CITE-seq data analysis

Preprocessing and quality control.

CITE-seq data were first split into the RNA parts (gene expression) and ADT (antibody-derived tags, distinguishing cell genotypes) parts. The RNA parts were preprocessed using the 10X Genomics pipeline (Cell Ranger) v6.1.1 using the mm10 genome and default parameters. The ADT parts were preprocessed using CITE-seq-Count (https://github.com/Hoohm/CITE-seq-Count) v1.4.5 with additional parameters “-cbf 1 -cbl 16 -umif 17 -umil 26 -cells 10000 -trim 10 -T 5”. Individual cells with high quality on both RNA and ADT parts were kept for downstream analysis (RNA: ≥1000 covered genes. ADT: ≥1000 reads, ≥90% reads mapped to known ADT barcodes, and major ADT barcodes contain ≥90% mapped reads).

Dimensional reduction and annotation of functional subsets.

For the retained high-quality cells, read counts were normalized using the Seurat (v4.0.0) (70) global-scaling method in which gene expression measurements for each cell were normalized by the total expression, followed by multiplication with a scaling factor of 10,000 and log transformation. We then performed dimensionality reduction, uniform manifold approximation and projection (UMAP) visualization and cell clustering as previously described (71). Marker genes for each cluster were identified by the criteria of ≥2-fold difference and adjusted p values <0.01), with the Seurat default method.

Grouping clusters into functional subsets were based on comparisons between the cluster-specific marker genes and cell type/lineage-characteristic genes according to existing knowledge. WT DN1 cells were analyzed alone (Fig. 1) or together with various factor KO DN1 cells (Fig. 4) following the same workflow. The trajectory analysis of WT DN1 cells was performed using the monocle3 R package (v1.0.0) (72) with default parameters. In comparisons between WT ETPs and factor KO-specific subsets, subset-specific genes were identified using the Wilcoxon rank-sum test by the criteria of ≥2-fold differences and adjusted p value <0.01 (Benjamini–Hochberg).

scATAC-seq data analysis

Preprocessing and quality control.

The scATAC-seq data were preprocessed using the 10X Genomics pipeline (Cell Ranger ATAC) v1.2.0 using the mm10 genome and default parameters. Samples from various genotypes were processed separately, and individual cells with high quality (TSS enrichment ≥ 7, unique barcode number ≥ 10,000, and doublet ratio < 1.5) were retained for downstream analyses. Quality control measurements and filtering were conducted using ArchR (v.1.0.1) (37).

Dimensional reduction and cell clustering.

The scATAC-seq data from high-quality cells were analyzed with a similar workflow as previously described (73). In brief, we used a latent semantic indexing algorithm to map scATAC-seq reads from high-quality cells to 500-bp bins across the mm10 reference genome for dimensionality reduction and clustering. As input, we selected 25,000 bins that had the highest signal variance across individual cells and used the top 30 PCs for cell clustering. We then applied a clustering algorithm from the Seurat (v.4.0.0) based on a shared nearest neighbor modularity optimization to identify cell clusters.

Integration with CITE-seq and annotation of functional clusters.

The chromatin accessibility within a gene body and +/– 100kb region flanking transcriptional start sites (TSS) was used to infer gene expression via computation of a ‘Gene Activity Score’ using the default method in ArchR. The gene activity score profiles for all cells were subsequently used to generate a gene activity score matrix. The processed and annotated CITE-seq data (described above) were then integrated with the scATAC-seq gene activity score matrix using the ‘addGeneIntegrationMatrix’ function from ArchR, which identifies corresponding cells across datasets or ‘anchors’ using Seurat’s mutual nearest neighbor algorithm. The functional subset labels of the CITE-seq data were transferred to the corresponding mutual nearest neighbors in the scATAC-seq data along with their gene expression signatures. After integration, scATAC-seq cells were re-annotated using the CITE-seq-transferred labels, and these defined functional subsets were used for downstream analyses. scATAC-seq data in WT DN1 cells were analyzed alone (Fig. 2) or together with Tcf1^−/−Lef1^−/− DN1 cells (Fig. 6) following the same workflow.

Chromatin accessible site detection.

After assigning functional subset labels to the scATAC-seq data, reads from cells in the same subset were then collected as pseudo-bulk ATAC-seq data for detection of genome-wide chromatin accessible (ChrAcc) sites. ChrAcc sites were identified as pseudo-bulk ATAC-seq peaks called with MACS2 (v2.2.7.1, with additional parameters “--nomodel --extsize 100 -q 0.01 --keep-dup 1”) (74). ChrAcc sites from different subsets were then collected and merged for the downstream analyses including marker site detection and peak2gene co-accessibility.

Identification of functional subset-specific marker genes and marker ChrAcc sites.

Based on gene activity scores from scATAC-seq data, genes with significantly higher chromatin accessibility in a subset than in other subsets were defined as functional subset-specific marker genes, which were identified using the Wilcoxon rank-sum test by the criteria of ≥2-fold differences and adjusted p value≤0.01 (Benjamini–Hochberg). The z-normalized gene activity scores for the subset-specific genes were plotted as a heatmap for visualization (Fig. 2D). Functional subset-specific marker ChrAcc sites were detected based on the ChrAcc changes on the merged pseudo-bulk ATAC-seq peaks, using Wilcoxon rank-sum test by the criteria of 2-fold differences and adjusted p value ≤0.001 (Benjamini–Hochberg). The z-normalized ChrAcc signal for the subset-specific peaks were plotted as a heatmap for visualization (Fig. 2C).

In analysis of ETP, nonT_potential, and T-ILC-like subsets in WT DN1 cells alone, a high-confidence marker ChrAcc site in a target subset was required to be a marker site over each of the other two subsets (fold change≥2, adjusted p value ≤0.001). The z-normalized ChrAcc signals of the subset-specific sites were plotted as a heatmap for visualization (Fig. 2E). For identification of differential ChrAcc sites comparing Tcf1^−/−Lef1^−/− KO-specific, WT nonT_potential, or WT T-ILC-like cells with WT ETPs, the criteria of 2-fold differences and adjusted p value ≤0.001 (Benjamini–Hochberg) using Wilcoxon rank-sum test on the single-cell level, and additional criteria of ≥2-fold differences on the pseudo-bulk level were applied to ensure the robustness of ChrAcc changes.

ChrAcc site to gene linkage and co-accessibility analysis.

We leveraged the integrated scRNA-seq and scATAC-seq data to investigate the links of co-accessible regions to gene expression. We used ‘getPeak2GeneLinks’ function with default parameters in ArchR to predict candidate gene regulatory interactions. The collected peak2gene loops were then filtered by requiring “adjusted p-value < 1e–4”, “distance between peak center and gene TSS within 5kb ~ 100kb”, and “co-accessibility score (correlation coefficient) ≥0.6”.

Bulk RNA-seq and data analysis.

c-Kit⁺ DN1 cells were sorted as emergent ETPs after 7-day culture of CreER⁺WT or CreER⁺dKO MPPs on OP9-DL1 culture where 4-hydroxytamoxifen was added during days 3–5. Total RNA was extracted, and cDNA synthesis and amplification were performed using SMARTer Ultra Low Input RNA Kit (Clontech) following manufacturer’s instructions. The resulting libraries were sequenced on Illumina’s HiSeqX platform in paired-end mode with a read length of 150 nucleotides. We mapped the sequencing reads to the mouse genome mm10 and generated reads per kilobase of transcript per million mapped reads (RPKM) as previously described (75, 76). We then identified differentially expressed genes (DEGs) using DESeq2 (v.1.12.4) (77) by requiring 1) fold changes ≥ 1.5, 2) adjusted p values < 0.05, and 3) RPKMs in higher expression condition ≥1. DAVID Bioinformatics (78) was used for functional annotation of DEGs, and GSEA (79) was used to determine enrichment of C2 curated gene sets in whole transcriptomes of CreER⁺WT and CreER⁺dKO emergent ETPs.

Bulk ATAC-seq and data analysis

CreER⁺WT or CreER⁺dKO emergent ETPs were sort-purified as above, and approximately 1×10⁴ cells were used for ATAC-seq library preparation as previously described (76). The resulting libraries were sequenced on Illumina’s HiSeqX platform in paired-end mode with a read length of 150 nucleotides. We mapped the sequencing reads to the mouse genome mm10 and called ATAC-seq peaks for each genotype using MACS2 (v.2.1.2) as previously described (75, 76). The peaks identified from both genotypes were then merged as union ChrAcc sites for downstream analyses. We then applied DESeq2 (v.1.12.4) for identification of differential ChrAcc sites between the two genotypes, by the criteria of ≥2-fold changes and adjusted p values <0.05. The fold changes of differential ChrAcc sites were estimated using apeglm, and motif analysis was performed using HOMER (v.4.11) (80) with default parameters.

Statistical analysis.

Student’s t-test and one-way ANOVA were used for two or multiple group comparisons, respectively, as previously described (75, 76). Tukey’s test was used as post hoc correction. Statistical parameters, such as numbers of samples means and standard deviation are described in the figures and figure legends.

Supplementary Material

List of supplementary materials:

1. Supplementary Materials and Methods;

2. Supplementary Fig. S1 to S7;

3. Data files S1–S3;

4. Data file S4. Raw data file;

5. MDAR Reproducibility Checklist.

Data and materials availability:

Single cell RNA-seq and ATAC-seq data are deposited at the GEO under accession number GSE264410, and bulk RNA-seq and ATAC-seq data are under GSE283601. Tabulated data underlying the figures is provided in Data file S4. All custom scripts used to analyze CITE-seq and scATAC-seq are archived in Zenodo (81). All other data needed to support the conclusions of the paper are presented in the paper or the Supplementary Materials.