We began this study by examining embryonic appearance of CCL21-expressing mTECs during mouse ontogeny. To do so, we analyzed CCL21-expressing mTECs by detecting tandem dimer Tomato (tdTomato) fluorescence proteins in Ccl21a-tdTomato knock-in mice. In postnatal thymus of Ccl21a^tdTomato/+ mice, CCL21-expressing cells were detected in keratin 14 (Krt14)-expressing medullary TECs (Fig. 1A) and not in Ly51-expressing cortical TECs (Fig. 1B). Flow cytometric analysis confirmed that Ccl21a^tdTomato+ cells in the thymus are predominantly confined in EpCAM⁺ Ly51^- UEA1⁺ mTECs (Fig. 1C, D). In accordance with the medullary localization of CCL21-expressing mTECs, thymic medulla was abundant in CCL21 proteins (Fig. 1E), which contribute to attracting CCR7-expressing positively selected thymocytes from the cortex^7,10. Among mTECs, most CCL21-expressing mTECs were distinct from Aire-expressing mTECs and DCLK1⁺ thymic tuft cells (Fig. 1F). CCL21-expressing mTECs equally distributed throughout the medullary region and were enriched neither in the pericortical area, including the corticomedullary junction area, nor in the central area within the medulla (Fig. 1G, H, Fig. S1A).

Analysis of embryonic thymus showed that CCL21-expressing cells in the central region of the thymus were prominent by embryonic day 15 (E15) (Fig. 2A) and detectable as early as E13 (Fig. 2B). The localization of CCL21-expressing cells in the central region of embryonic thymus lobes and their expression of Krt14 and Foxn1 ascertained that CCL21-expressing embryonic thymus cells represented mTECs (Fig. 2A-C), rather than neighboring CCL21⁺ parathyroid gland cells (Fig. 2B), which contribute to the attraction of hematopoietic cells to the pre-vascularized embryonic thymus²⁴. CCL21 proteins were detectable in the medullary region proximal to Ccl21a^tdTomato-expressing mTECs in the embryonic thymus by E15 (Fig. 2D), indicating that Ccl21a^tdTomato-expressing E15 embryonic mTECs represent functionally relevant CCL21-expressing mTECs. On the other hand, CCL21-expressing mTECs in the embryonic thymus were largely distinct from Aire-expressing mTECs and DCLK1⁺ thymic tuft cells (Fig. 2E). Similar to the postnatal thymus, CCL21-expressing mTECs in the embryonic thymus distributed throughout the medullary region and were not enriched in the pericortical areas including the corticomedullary junctions (Fig. 2F, G, Fig. S1B). These results indicate that CCL21-expressing mTECs arise by E15 in the central region of the embryonic thymus.

To better understand whether CCL21-expressing mTECs have a developmental potential to give rise to the self-antigen-displaying mTEC subset, we engineered mice in which the P1 bacteriophage-derived Cre recombinase gene was knocked-in at Ccl21a locus (Fig. 3A). We bred Ccl21a^Cre/+ mice with mice that carried a ubiquitous CAG promoter-driven loxP-stop-loxP-EGFP (CAG-loxP-EGFP) transgene. In Ccl21a-Cre x CAG-loxP-EGFP mice, EGFP expression labeled cells that have transcribed Ccl21a previously and/or presently (Fig. 3A).

In the postnatal thymus of Ccl21a-Cre x CAG-loxP-EGFP mice, EGFP signals in the thymus distributed prominently in the medullary region (Fig. 3B). EGFP⁺ cells in the medulla were detected in Krt14⁺ mTECs (Fig. 3B), including Aire⁺ mTECs and DCLK1⁺ thymic tuft cells (Fig. 3C). Flow cytometric analysis of enzyme-digested thymus cells showed that EGFP⁺ cells in the thymus were detected exclusively in EpCAM⁺CD45^- TECs and not in EpCAM^- CD45⁺ hematopoietic cells, including thymocytes and dendritic cells, or EpCAM^-CD45^- cells, including mesenchymal fibroblasts and endothelial cells (Fig. 3D, E). Within EpCAM⁺CD45^- TECs, approximately 95% of UEA1⁺Ly51^- mTECs were EGFP⁺, indicating that a vast majority of mTECs were derived from cells that previously transcribed Ccl21a (Fig. 3F, G). Flow cytometric analysis also confirmed the detection of Aire⁺ UEA1⁺Ly51^- cells and DCLK1⁺ UEA1⁺Ly51^- cells in EpCAM⁺CD45^- EGFP⁺ TECs (Fig. 3H). Thus, self-antigen-displaying mTECs, including Aire⁺ mTECs and DCLK1⁺ thymic tuft cells, which do not presently express CCL21 (Fig. 1F, 2E), originate from Ccl21a⁺ progenitors.

In the quantitative RT-PCR analysis of Cre, Ccl21a, and other TEC-associated genes, such as Aire, Tnfrsf11a, and Psmb11, Cre was detected specifically in the Ccl21a-expressing mTEC^low subpopulation, which includes the majority of CCL21-expressing mTECs²¹, and not in the Psmb11-expressing cTECs or the Aire- and Tnfrsf11a-expressing mTEC^high subpopulation (Fig. 4A). These results verify the lineage tracing potential of Ccl21a-Cre-expressing cells to differentiate into the mTEC^high subpopulation including Aire⁺ mTECs.

In addition to mTECs, approximately 66% of UEA1^-Ly51⁺ cTECs in the postnatal thymus were EGFP⁺ in Ccl21a-Cre x CAG-loxP-EGFP mice (Fig. 3 F, G). Section analysis of the thymus verified the presence of EGFP⁺ cells in the cortex (Fig. 3B). Because cTECs do not presently express Ccl21a transcripts¹⁰ (also shown in Fig. 1B-D), these results indicate that approximately two-thirds of postnatal cTECs are derived from cells that previously transcribed Ccl21a (Fig. 3 F, G). To characterize cTECs that previously transcribed Ccl21a and those that did not, we isolated EGFP⁺ and EGFP^- cTECs from postnatal Ccl21a-Cre x CAG-loxP-EGFP mice (Fig. S2) and performed deep-sequencing transcriptomic analysis. The MA plot and the volcano plot of RNA-sequencing data demonstrated high similarity in gene expression profiles between EGFP⁺ and EGFP^- cTECs among 20,230 genes detected (Fig. 4B, C), unlike the pronounced difference in global gene expression profiles between cTECs and mTECs (Fig. S3A, B). The genes that are relevant for T cell-inducing and positive selection-inducing functions in cTECs, including Dll4, Il7, Cxcl12, Psmb11, Ctsl, Prss16, Cd83, H-2K, H-2D, and I-Aβ, were similarly abundant in EGFP⁺ and EGFP^- cTECs (Fig. S3C). These results suggest that cTECs derived from cells that either do or do not transcribe Ccl21a are functionally comparable to promote the generation and positive selection of cortical thymocytes.

Only 14 genes, including Ccl21a, showed significantly higher expression in EGFP⁺ cTECs than in EGFP^- cTECs, whereas only one gene, the parathyroid hormone-encoding Pth, showed significantly lower expression in EGFP⁺ cTECs than in EGFP^- cTECs (Fig. 4C, Fig. S3D). All the 14 genes that showed significantly higher expression in EGFP⁺ cTECs than in EGFP^- cTECs represented genes that showed higher expression in mTECs than in cTECs, and their expression, including Ccl21a expression, was much higher in mTECs than in EGFP⁺ cTECs (Fig. S3D). Interestingly, EGFP⁺ cTECs distributed unequally within the thymic cortex and were detected more frequently in the perimedullary cortical region than in the central cortical region or the subcapsular region of the cortex (Fig. 4D, E, Fig. S1C). In contrast, Ly51⁺ total cTECs did not show such an unequal distribution in the thymic cortex (Fig. S1D, E). Thus, the intrathymic proximity to the medullary region contributes to the previous expression of Ccl21a transcripts in cTECs, suggesting that approximately two-thirds of postnatal cTECs originate from Ccl21a⁺ cells generated in the thymic medulla.

Collectively, these fate-mapping experiments demonstrate that a vast majority of mTECs, including the self-antigen-displaying mTEC subset, and approximately two-thirds of cTECs, particularly cTECs in the perimedullary cortical region, are derived from progenitor cells that previously transcribed Ccl21a.

It is further interesting to note that EGFP⁺ cells within Foxn1⁺ mTECs were detected in the central region of the thymus primordium in Ccl21a-Cre x CAG-loxP-EGFP mice as early as E11 (Fig. 5A). The detection of EGFP⁺ cells occurred earlier in the ontogeny than the detection of Ccl21a^tdTomato+ mTECs by E13 (Fig. 2B) and CCL21-protein-expressing mTECs by E15 (Fig. 2D). These sequential detections suggested that low-level Ccl21a transcription initiated as early as E11 during thymus organogenesis was sufficient to produce functional Cre proteins to mediate the expression of CAG-promoter-dependent EGFP proteins, but failed to produce a detectable amount of tdTomato or CCL21 proteins. Subsequent elevation of Ccl21a transcription level resulted in the production of detectable amounts of tdTomato proteins followed by CCL21 proteins. Indeed, single-cell RNA-sequencing analysis of Ccl21a transcript counts demonstrated that Ccl21a expression levels within individual Epcam⁺ Foxn1⁺ TECs elevated during the embryogenesis from E12 to E14 (Fig. 5B). At E14, Ccl21a^low cells were detected at an equivalent frequency in β5t⁺ and β5t^- TEC populations (Fig. 5B). Ccl21a transcription in CCL21-expressing mTECs was further elevated from embryonic to postnatal development, as shown by single-cell RNA-sequencing analysis of Ccl21a transcript counts (Fig. 5B) and verified by flow cytometric analysis of Ccl21a-dependent tdTomato expression levels (Fig. 5C). Importantly, Ccl21a⁺ Foxn1⁺ mTECs detected during early thymus organogenesis were localized in the central region of the thymus primordium and were anatomically and transcriptomically distinguishable from Ccl21a⁺ Gcm2⁺ parathyroid epithelial cells (Fig. 2B, 5A, 5D). These results reveal that Ccl21a⁺ mTECs develop in the central region of the thymus primordium as early as E11 before the generation of detectably CCL21 protein-expressing mTECs in the embryonic thymus by E15.

Along with the fate-mapping detection of the developmental potential in Ccl21a⁺ cells, the early and temporal detection of Ccl21a⁺ mTECs and CCL21-expressing mTECs during embryogenesis tempted us to speculate that Ccl21a⁺ mTECs, and even CCL21-expressing mTECs, might possess mTEC progenitor activity. Previous studies reported that embryonic thymus develops RANK-expressing mTEC-restricted progenitors, which have a developmental capability to give rise to self-antigen-displaying mTECs, including Aire⁺ mTECs^25,26. RANK-expressing mTEC progenitors arise in the thymus during embryogenesis downstream of mTEC stem cells, which originate from cTEC-trait-expressing bipotent progenitors^5,25–28. We next examined whether CCL21-expressing mTECs might overlap with RANK-expressing mTEC progenitors in CCL21^tdTomato RANK^Venus embryonic thymus. Interestingly, however, we found that CCL21-expressing mTECs and RANK-expressing mTECs were largely distinct cell populations within the embryonic thymus (Fig. 6A, B), although they were closely localized within the medullary region (Fig. 6A). RNA-sequencing analysis of CCL21⁺RANK^- and CCL21^-RANK⁺ TECs highly purified from CCL21^tdTomato RANK^Venus embryonic thymus (Fig. S4A) demonstrated that CCL21-expressing mTECs were markedly different from RANK-expressing mTECs in global gene expression profiles (Fig. 6C). Gene ontology enrichment analysis ascertained the difference in global gene expression profiles in CCL21-expressing mTECs and RANK-expressing mTECs (Fig. 6D). Indeed, CCL21-expressing mTECs retained more cTEC-trait genes, including Psmb11, Cxcl12, CD83, and Ackr4, than RANK-expressing mTECs (Fig. 6E), suggesting that in comparison with RANK-expressing mTECs, CCL21-expressing mTECs are more recently derived from cTEC-trait-expressing bipotent TEC progenitors. However, similarly to RANK-expressing mTECs, CCL21-expressing mTECs expressed equivalent levels of some mTEC-trait transcripts, including Krt5 and Krt14 (Fig. 6E). Among the genes associated with TEC progenitor activity^29–34, Trp63 and Pax1 were predominant in CCL21-expressing mTECs, whereas Plet1, Krt19, Cldn3, Cldn4, and Ly6a were readily detectable in RANK-expressing mTECs (Fig. 6E). Thus, CCL21-expressing mTECs in embryonic thymus are largely distinct from RANK-expressing mTECs, whereas CCL21-expressing mTECs and RANK-expressing mTECs are proximally localized with each other within the thymus medulla.

Nonetheless, it is important to point out that fate-mapping experiments using Ccl21a-Cre x CAG-loxP-EGFP mice were unable to distinguish whether TEC progenitor activity resides in CCL21-expressing mTECs generated within the thymus or in CCL21-expressing non-mTECs including neighboring parathyroid cells.

To directly examine whether CCL21-expressing mTECs have a developmental potential to give rise to self-antigen-displaying mTECs, we finally traced the fate of CCL21-expressing mTECs within the thymus microenvironment. To do so, we employed reaggregate thymus organ culture experiments^35,36 where RelB-deficient thymic stromal cells do not generate any functional mTECs^37,38. Reaggregate thymuses were transplanted under mouse kidney capsules for reconstitution with hematopoietic cells derived from recipient mice. In this experimental condition, RelB-deficient thymic stromal cells were reaggregated with tdTomato-expressing mTECs highly purified from Ccl21a^tdTomato E17 embryos (Fig. S4B). Ccl21a^tdTomato+ mTECs had already become functional CCL21-protein-expressing mTECs by E17 (Fig. 2), and the accumulation of CD4⁺CD8^- and CD4^-CD8⁺ thymocytes in E17 thymic medullary region was defective in mice deficient in CCL21 (Fig. 7A-C), so that the reaggregate thymus experiments would address whether functionally relevant CCL21-expressing mTECs give rise to other mTEC subpopulations including Aire⁺ mTECs.

We found that Ccl21a^tdTomato-positive EpCAM⁺CD45^- mTECs isolated from Ccl21a^tdTomato/+ E17 embryos generated Aire⁺ Krt14⁺ mTECs in the thymus reaggregated with RelB-deficient thymic stromal cells (Fig. 8A). Aire⁺ Krt14⁺ mTECs also developed from an equal number of parallelly isolated Ccl21a^tdTomato-negative EpCAM⁺CD45^- TECs (Fig. 8B), which included cTEC-mTEC bipotent TEC progenitors^5,39. RelB-deficient thymic stromal reaggregates without additional TECs failed to generate Aire⁺ Krt14⁺ mTECs (Fig. 8C), whereas control organ culture using RelB-heterozygous thymus apparently developed medullary regions that contained Aire⁺ Krt14⁺ mTECs (Fig. 8D). Quantitative evaluation of Krt14⁺ medullary areas (Fig. 8E) within total thymus aggregates (Fig. 8F) demonstrated that Ccl21a⁺ mTECs have a developmental potential to give rise to Aire⁺ Krt14⁺ mTECs, equivalently to and not statistically different from Ccl21a-negative TECs, including bipotent TEC progenitors (Fig. 8G, H). Interestingly, Ccl21a⁺ mTECs were still detected in the Krt14⁺ medullary areas in reconstituted thymus reaggregated with either Ccl21a⁺ mTECs or Ccl21a-negative TECs (Fig. 8I, J), indicating that Ccl21a⁺ mTECs have a potential to maintain themselves in the thymus and that Ccl21a-negative TECs have a developmental potential to produce Ccl21a⁺ mTECs. Most importantly, these results directly indicate that functionally relevant CCL21-expressing mTECs have a developmental potential to give rise to Aire-expressing mTECs.

In contrast to CCL21-expressing mTECs isolated from E17 embryos, Ccl21a^tdTomato-positive EpCAM⁺CD45^- mTECs isolated from 4-week-old Ccl21a^tdTomato/+ mice did not exhibit any signs of Aire⁺ Krt14⁺ mTEC generation in the thymus reaggregated with RelB-deficient thymic stromal cells (Fig. S5A-G), suggesting a decline in the developmental potential of CCL21⁺ mTECs during the ontogeny from embryonic to postnatal period. However, no detection of progeny could also result from technical caveats associated with reaggregate thymus organ culture conditions that were suitable for detection of the developmental potential of embryonic but not postnatal CCL21-expressing mTECs, or differences in the lifespan and/or turnover of embryonic and postnatal CCL21-expressing cells.

Throughout the development including E17 embryos, CCL21-expressing mTECs were localized specifically in the central region of the thymus and were Krt14⁺ (Fig. 1, 2, 6, 9A). However, single-cell RNA-sequencing analysis demonstrated that the global gene expression profiles were largely different between embryonic Ccl21a^low mTECs and postnatal Ccl21a^high mTECs (Fig. 9B). Indeed, the majority of CCL21-expressing mTECs at E17 but not the postnatal period simultaneously expressed cTEC-associated protein Ly51 and mTEC-associated UEA1-binding molecules on the cell surface (Fig. 9C). Ly51 expression was lower in embryonic TECs than postnatal TECs⁴⁰ and equivalently distributed in the thymus including the medullary region defined by Krt14-expressing mTECs centrally localized in E17 thymus (Fig. 9A). These results highlight the difference in molecular expression profiles, including Ly51 expression, between E17 CCL21-expressing mTECs and postnatal CCL21-expressing mTECs.

C57BL/6 (B6) mice were obtained from The Jackson Laboratory (Bar Harbor, ME) and SLC (Shizuoka, Japan). Ccl21a-tdTomato knock-in mice¹⁰, RelB-knockout mice³⁸, RANK-Venus transgenic mice⁴⁵, β5t-Venus knock-in mice⁴⁶, CAG-loxP-stop-loxP-EGFP transgenic mice⁴⁷, and Foxn1-GFP transgenic mice⁴⁸ were described previously. All mouse experiments were performed with consent from the Animal Experimentation Committee of the University of Tokushima (T2022–50), the Animal Care and Use Committee of the National Cancer Institute (ASP 21–431, ASP 21-432, and EIB-076–3), the Birmingham Animal Welfare and Ethical Review Board and UK Home Office Animal Licences (PP7518148 and PP2990911), and the Institutional Animal Care and Use Committee of RIKEN Kobe Branch (A2001-03).

The targeting vector was prepared by subcloning Ccl21a-containing mouse genomic BAC fragments (Advanced GenoTechs) and Cre-encoding cDNA into a plasmid containing a pgk-neo cassette (Gene Bridges). The linearized targeting vector was introduced into TT2 ES cells⁴⁹. Targeted knock-in alleles were screened by genomic PCR analysis and Southern blot analysis. The Ccl21a-Cre mice (https://large.riken.jp/distribution/mutant-list.html; Accession No. CDB1131K) are available to the scientific community and were backcrossed to B6 mouse strain for more than 5 generations. The knock-in allele is detected using PCR primers as follows; Cre-F, 5’-AGGTTCGTTCACTCATGGA-3’ and Cre-R, 5′-TCGACCAGTTTAGTTACCC-3′ (product size, 235bp). Wild-type allele is detected using PCR primers as follows; WT-F, 5′-CTGGTCTCATCCTCAACTCA-3′ and WT-R, 5′-TGTAACCCTAGGATTGTAGG-3′ (product size, 1,563bp).

Fetal thymus lobes isolated from relB-KO mice at embryonic day (E) 15 were organ cultured for 5 days on sponge-supported Nucleopore filters (Whatman) placed on Dulbecco’s Modified Eagle Medium supplemented with 10% fetal bovine serum (FBS) and penicillin-streptomycin mixed solution. To eliminate endogenous hematopoietic cells from the lobes, 1.35 mM 2’-deoxyguanosine (dGuo) was added in the culture medium^35,36. dGuo-treated thymus lobes were digested with 0.25% trypsin supplemented with 0.45 mM EDTA and 0.02% DNase I for 15 min at 37℃ and reaggregated with tdTomato⁺ or tdTomato^- TECs isolated from Ccl21a^tdTomato/+ mice at E17 or 4 weeks old. After one day culture on Nucleopore filters, reaggregated thymuses were transplanted under kidney capsules of B6 mice. dGuo-treated fetal thymus lobes isolated from RelB-deficient and RelB-heterozygous mice were also transplanted under kidney capsules of B6 mice. Grafts were harvested 5 weeks post transplantation.

Paraformaldehyde-fixed frozen tissues embedded in OCT compound (Sakura Finetek) were sliced into 10-μm-thick sections. The sections were stained with anti-CCL21 (Bio-Rad), anti-Aire (eBioscience), anti-Ly51 (Biolegend), anti-Krt14 (Biolegend), anti-DCLK1 (abcam), anti-CD4 (Biolegend), anti-CD8 (Biolegend) antibodies, anti-Foxn1 antiserum⁵⁰, and UEA-1 (Vector Laboratories). Images were analyzed with a TCS SP8 confocal laser scanning microscope (Leica).

The plot profile intensity for the fluorescence image of each channel in the regions of interest (ROI) was measured using ImageJ software (https://imagej.nih.gov/ij/index.html). The fluorescence intensity at each horizontal distance with the total intensity of the vertically averaged tdTomato, EGFP, or Ly51 in the ROI was calculated as 100. The sum of tdTomato or EGFP intensities for the medullary region defined by Krt14 expression or UEA1 reactivity and for the cortical region was calculated.

For the analysis of thymic epithelial cells (TECs), minced thymuses were digested with 0.5 unit/ml Liberase (Roche) in the presence of 0.02% DNase I (Roche). Single-cell suspensions were stained with antibodies specific for EpCAM (BioLegend), CD45 (eBioscience), Ly51 (Biolegend), Podoplanin (Biolegend), I-A^b (BioLegend), and for the reactivity with UEA-1 (Vector Laboratories). For the analysis of Aire and DCLK1, surface-stained cells were fixed in 5% formaldehyde neutral buffer solution (Nacarai Tesque), permeabilized in 1x permeabilization buffer (eBioscience), and stained with anti-Aire (eBioscience) or anti-DCLK1 (abcam) antibody. For the isolation of TECs, CD45^- cells were enriched in magnetic bead conjugated anti-CD45 antibody (Miltenyi Biotec). Multicolor flow cytometry and cell sorting were performed on FACSVerse, FACSAria II, and LSRFortessa (BD Biosciences).

cDNAs were prepared by using SMART-Seq v4 Ultra Low Input RNA Kit, according to the manufacturer’s protocol (Clontech). Sequencing libraries were generated by using a Nextera XT DNA Library Prep Kit, according to the manufacturer’s protocol (Illumina). The concentration of libraries was measured by an ABI PRISM 7500 Real-time PCR system in combination with a Power SYBR Green PCR Master Mix (Thermo Fisher). Single-end sequencing of cDNA libraries with a read length of 51 was performed with HiSeq 1500 platform (Illumina). For RNA-sequencing analysis of CCL21⁺RANK^- and CCL21^-RANK⁺ TECs isolated from CCL21^tdTomato RANK^Venus embryonic E17 thymus, RNA was prepared using Qiagen RNAeasy Plus Micro Kit with genomic DNA eliminator columns. RNA-sequencing samples were sequenced on NovaSeq_SP flowcell using Illumina TruSeq Stranded Total RNA Library Prep with 2×100bp paired-end sequencing. Data were analyzed by using CLC Genomics Workbench 12.0 (Qiagen) with default parameters.

E12 embryonic pharyngeal cells, E14 embryonic CD45^-EpCAM⁺ β5t-Venus⁺ and β5t-Venus^- TECs, E16 embryonic CD45^-EpCAM⁺ β5t-Venus⁺ and β5t-Venus^- TECs isolated from β5t-Venus knock-in mice, and 4-week-old postnatal TEC-enriched thymus cells were employed for single-cell RNA sequencing. 3′ mRNA-seq library was generated as outlined in the Chromium Connect platform using 3’ v3.1 Single Cell RNA-Seq chemistry reagents from 10X Genomics. Reverse transcription and cDNA amplification were performed according to the manufacturer’s instructions. Libraries were sequenced on NextSeq 2000 and FASTQ files were generated from demultiplexes raw base call files using Cell Ranger 6.1.1 mkfastq. Alignment, filtering, barcode counting, and UMI counting were performed using Cell Ranger 6.1.1 count with mouse reference 3.0.0 (mm10 and Ensembl release 98). The resulting matrix was imported into Seurat (v4.3.0) in R v 4.3.0 to create a Seurat object. The object underwent normalization, filtering, scaling, and dimensionality reduction steps sequentially using Seurat build-in functions, with default parameters. To integrate all the filtered objects from each sample, the CCA algorithm was employed through the FindIntegrationAnchors and IntegrateData functions in Seurat. The Uniform Manifold Approximation and Projection (UMAP) plots were visualized using the DimPlot function of Seurat. The histograms were visualized using the ggplot2 package (v3.4.2).

Statistical significance was assessed using the two-tailed unpaired Student’s t test with Welch’s correction for unequal variances or one-way ANOVA with Tukey’s correction.

RNA sequencing data are deposited in The DNA Data Bank of Japan (https://www.ddbj.nig.ac.jp) BioProject database with the accession number PRJDB15439 and NCBI Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo) database with the accession number GSE239913. Single-cell RNA sequencing data are deposited in NCBI GEO with the accession numbers (deposition in progress).