To examine the reconstitution of iNKT cells in haplo-BMT with and without GvHD, CD3⁺ T cell-depleted BM cells (5x10 (6)) from C57BL/6 mice were adoptively transferred into lethally irradiated F1 (C57BL/6 x BALB/c) mice with or without 2x10 (6) B6-derived splenic cells (BM + Spl (GvHD) versus BM only) ( Figures S1A–C ), ( 37 ). At 9- and 13-week post-BMT, the cell numbers, frequencies of iNKT cells (TCRβ⁺PBS57:CD1d-tetramer (CD1d-tet)⁺), and iNKT/T ratios in the thymus, spleen, and liver were significantly lower in mice with GvHD than those with BM only ( Figures 1A ; S1D ), agreeing well with the previous results of impaired iNKT cell reconstitution in allo-HSCT patients with GvHD (2, 16–19). Except for a slight decrease in Qa2 expression, the maturation of peripheral iNKT cells, measured by the expression of surface markers (CD44, NK1.1, Ly6C, Ly49C/I) and cytotoxicity-related molecules (CD107a, Granzyme B), production of cytokines (IL-4 and IFN-γ), exhibited no substantial differences between mice with and without GvHD ( Figures 1B–D ; S1E, F ). Nrp1 is expressed in recent thymic iNKT emigrants but not mature iNKT cells (39). We found an increased relative percentage of Nrp1^int/hiQa2^- iNKT cells but a substantially decreased number of both Nrp1^int/hiQa2^- and Nrp1^int/loQa2⁺ iNKT cells in the spleen and liver of GvHD mice ( Figures 1E, F ). The comparison of Ki67⁺ iNKT cells showed enhanced proliferation in the thymus but not the periphery of GvHD mice ( Figure 1G ). No difference was found in the levels of iNKT cell apoptosis/survival (AnnexinV⁺7-AAD^-, AnnexinV⁺7-AAD⁺, or Bcl2⁺ cells) between the two groups ( Figure 1H ). As more than 95% of iNKT cells in mice with GvHD were of donor origin (CD45.1⁺) ( Figure S1G ), the reduced iNKT cells in the thymus and decreased recent thymic iNKT emigrants in the periphery suggest that de novo generation of thymic iNKT cells is impaired in GvHD mice.

To determine the underlying cause of iNKT cell defects in the thymus, we first examined which iNKT cell stage or effector subset is affected. Compared to BMT mice with BM only, those with GvHD showed a marked loss of ST2/ST3 cells and a relative accumulation of ST0/ST1 cells ( Figures 2A ; S1H ). We further observed a significant decrease in the cell number of the thymic NKTp subset (CCR7⁺ or S1P1⁺ cells) at various stages ( Figure 2B ), agreeing well with the reduced Nrp1⁺ thymic iNKT emigrants in the periphery of GvHD mice. We also stained the cells with antibodies specific for transcription factors and surface markers related to iNKT effector subsets. Consistent with ST3 cell reduction, the mice with GvHD had a marked reduction of PLZF^loRORγt^- or PLZF^loT-bet⁺ NKT1 cells ( Figures 2C, D ). The percentages of PLZF^hiRORγt^- or PLZF^hiGATA3⁺ NKT2 cells were increased while the cell numbers were similar in GvHD mice relative to the controls, leading to a substantial decrease in the ratio of NKT1/NKT2 cells, in particular at 13-week after haplo-BMT ( Figures 2C–E ). No significant difference was found in PLZF^intRORγt⁺ NKT17 cells in GvHD mice ( Figure 2C ). The expression of surface markers of NKT2 (CD4, IL-17RB) and NKT1 (CD122, CXCR3) cells showed a similar pattern of changes in GvHD mice ( Figure 2F ). Together, these results indicate a profound impairment of NKTp and NKT1 generation in the thymus of haplo-BMT mice with GvHD.

Thymic NKT1 cells have been reported to derive from NKTp and NKT2-like cells at developmental stages of ST1 and ST2. The intermediate NKT1 precursors, however, were more clearly defined within ST3 cells (31). As GvHD thymi showed a substantial decrease in NKT1/ST3 cells but not NKT2 and NKT17 cells, we determined to define early NKT1 precursors at ST1 or ST2 and investigate whether these precursors were affected in GvHD mice. We first explored NKT1 precursors in post-selected TCRβ⁺CD1d-tet⁺CD24^-NK1.1^- iNKT cells (ST1 and ST2) in C57BL/6 mice without BMT by scRNA-seq ( Figure 3A ). A total of 4491 cells passed the quality control and 13 clusters were identified by unsupervised clustering analysis based on top 20 principal components ( Figures 3B ; S2A, B ). In line with the enrichment of CD24^- iNKT cells, the transcript of Cd24a was only found in a few scattered cells in cluster C12 and cycling cluster C11 ( Figures S2C, D ). The cells in cluster C12 were assigned as NKT0 as they displayed enriched expression of genes associated with post-selected ST0 cells, including lymphocyte activation (Egr1, Sox4), metabolism (Ldhb, Uqcrq), and trafficking (Ccr7) ( Figure 3C ) (40–42). High activity of transcription factors Yy1 and Sox4 (30, 40–44) were also found in C12 by transcription factor regulatory network analysis ( Figure 3D ). The cells in clusters C0, C1, C6, and C9 were categorized as NKT2 or NKT2-like cells as they displayed NKT2 signature with cluster C0 having the highest level of NKT2 related genes (Zbtb16, Gata3, Pdcd1, Tox, Id3, Il4 Il17rb, Slamf6, and Izumol1r) and highest activity of transcription factors Gata3 and Myb ( Figures 3C, D ). We also assigned the clusters C2, C3, and C4 as NKT1 for their high expression of NKT1 signature (Tbx21, Cxcr3, Xcl1, Il2rb, Slamf7, killer cell lectin type receptors, cytotoxicity-related genes) and high activity of transcription factors Tbx21 and Stat4 ( Figures 3C, D ). Among these three clusters, C2 showed the highest expression of NKT1 markers as well as cell retention marker Cd69, suggesting that C2 cells are more closely related to NK1.1⁺CD69⁺ mature NKT1 cells.

Based on the published signature genes of various iNKT subsets (31–33) and transcription factor regulatory network analysis, other clusters were classified as cycling clusters (C5, C8, and C11), NKT17 (C7), and Stat1⁺ NKT ( Figures 3C, D ; S2C, D ). Notably, relatively high levels of Gata3, Id3, Pdcd1, and Il13 transcription were found in cycling cell clusters ( Figure 3C ), supporting the previous finding of high proliferation capacity in NKT2 cells (45). The Stat1⁺ NKT cluster showed differential expression of NKT1 (Tbx21, Il2rb, Cxcr3) and NKT2 (Gata3, Izumo1r) signature genes, and retention marker Cd69 ( Figure 3C ), suggesting that this cluster may contain cells at the developmental intermediate stage that have lost some trafficking capability. We further found that C1, C12, cycling clusters, and to a lesser extent, C0, expressed NKTp marker Ccr7 while C1, C5, and C12 showed a high level of S1pr1, suggesting that these clusters are enriched with NKTp cells.

We applied Pearson correlation to assess the similarities between clusters. As shown in Figure 3E , NKT2 clusters (C0 and C1) formed a distinct group and showed a high degree of similarity with cycling clusters (C5, C8, C11). NKT1 clusters (C2, C3, and C4), NKT2 clusters (C6, C9), NKT17 cluster (C7), and Stat1⁺ NKT cluster (C10) formed another group with a higher degree of similarity with cluster C1. We further performed RNA velocity analysis ( Figure 3F ) and found that clusters C2 and C0 represented well-differentiated NKT1 and NKT2 effector subpopulations, respectively. The NKT1 clusters (C3, C4), NKT2 clusters (C6, C9), and Stat1⁺ NKT cluster (C10) all had precursor cells for NKT1 cluster C2, suggesting that these clusters may represent distinct but intermediate precursor stages during NKT1 differentiation ( Figure 3F ). Notably, a small fraction of cells in NKT2 cluster C1 appeared to be precursors for C3, C6, and C10 while the rest of the cells had precursors for NKT2 cluster C0 ( Figure 3F ). Collectively, the scRNA-seq data suggest that C1 may be a critical Ccr7 ⁺ NKTp cluster that can either emigrate from the thymus (S1pr1) or differentiate into NKT1 (C2) and NKT2 (C0) effector clusters within the thymus. C6, C9, C10, C3, and C4 may represent intermediate clusters gradually downregulating the NKT2 signature while upregulating the NKT1 signature.

We next analyzed scRNA-seq data of thymic CD24^-NK1.1^- iNKT cells obtained from haplo-BMT mice with and without GvHD ( Figure 4A ). Compared to those with BM only, the mice with GvHD had markedly reduced proportions of cells in NKT1 cluster C2 and intermediate precursor clusters C3, C4, C6, C9 ( Figures 4A, B ). The percentages of NKT2 (C0, C1), NKT0 (C12), and cycling clusters (C5, C8, C11), however, were increased in mice with GvHD. We also found that mice without GvHD had enrichment of TCR clonal expansion in NKT1 and their intermediate precursor clusters (clusters C2, C3, C4, and C6) while those with GvHD only had small or single-cell clones ( Figure 4C ). Accordingly, higher clonal diversity was found in mice without GvHD ( Figure 4D ). However, the comparison of TCR Vβ usage and CDR3 length of Vβ chain showed no difference between the two groups ( Figures 4E , S2E ).

We further compared the transcriptome of each cluster and found that NKT1 and precursor clusters (C2, C3, C4, C6, C9, C10) in GvHD thymi had upregulation of NKT2-signature genes (Zbtb16, Pdcd1, Izumo1r, Tox, Il4), chemokine receptor (Ccr7), interferon-responsive genes (Irf7, Stat1), and down-regulation of NKT1 signature genes (Tbx21, Gzmb, Nkg7, Stat4, Xcl1) ( Figure 4F ). Increased expression of genes related to glucose metabolisms (Pkm, Pgk1) was also observed in NKT1 precursor clusters C4 and C6 in GvHD mice. GSEA analysis of clusters C3, C4, and C6 further showed that iNKT cells in GvHD mice had down-regulation of NK-mediated cytotoxicity, inflammatory response, and upregulation of oxidative phosphorylation, citric acid TCA cycle & respiratory electron transport, metabolism of amino acids & derivatives ( Figure S2F ).

To assess whether NKT1 precursor subpopulations within ST1 and ST2 could be identified by flow cytometry, we first analyzed scRNA-seq data ( Figure 3B ) again to create a pseudo-time ordering of iNKT cell transcriptomes. As shown in Figure 5A , Il17rb and Cxcr3 were upregulated along with Gata3 and Tbx21, respectively. The clusters at intermediate developmental stages, including C1, C6, C9, and C10 showed co-expression of Il17rb and Cxcr3 ( Figure 5B ). We thus characterized CD24^-NK1.1^- iNKT cells with different expression statuses of IL-17RB, CXCR3, and CD44. CD138⁺ NKT17 cells were excluded in the gating strategy and PLZF^hiRORγt^- NKT2, PLZF^loT-bet⁺ NKT1, CD24⁺CD44^lo ST0, NK1.1⁺CD44^hi ST3 cells were used in the comparison ( Figures 5C, D ). Different from the mRNA expression data ( Figure 5B ), the flow cytometry analysis showed that 46.64 ± 6.56% and 26.40 ± 2.91% of the cells were IL-17RB⁺ and CXCR3⁺, respectively, while the cells with IL-17RB⁺CXCR3⁺ were only 11.06 ± 0.51% ( Figure 5C ). However, when including more molecules for phenotypic analysis, similarities were found between the subpopulations analyzed by flow cytometry and clusters by scRNA-seq. For instance, P2 (IL-17RB⁺CXCR3^-CD44^lo/int) subpopulation corresponded to clusters C0 and C1 in scRNA-seq, ST1 cells (46), or NKTp/NKT2 subsets as the cells in this subpopulation had TCR^hiPLZF^hiT-bet^loPD-1^hiCCR7^hi phenotype, were larger in cell size with more proliferating cells and high level of glucose uptake (measured by 2-NBDG) ( Figures 5D, E ). By contrast, P4 (IL-17RB^-CXCR3⁺CD44^hi) subpopulation was very close to cluster C2 in scRNA-seq or NKT1 subset/ST3 as the cells in this subpopulation exhibited TCR^loPLZF^loT-bet^hiCD122^hiPD-1^loCCR7^lo phenotype and had smaller cell size, lower level of glucose uptake and proliferation. Compared to P2, P1 (IL-17RB^-CXCR3^-CD44^lo, 4.99 ± 1.01%) subpopulation had a slightly lower level of PLZF, PD-1, CD44 expression, and less glucose uptake, implicating that P1 and P2 subpopulations may correspond to C1 and C0 in scRNA-seq analysis with a precursor-progeny relationship ( Figures 5D, E ). Compared to P4, P3 (IL-17RB⁺CXCR3^intCD44^int/hi) and P5 (IL-17RB^-CXCR3^-CD44^int/hi (P5, 8.92 ± 0.34%) subpopulations showed lower levels of T-bet, CD122, CD44, CD69, higher levels of TCR, PD-1, CCR7, and cell proliferation ( Figures 5D, E ). Thus, the cells in P3 and P5 may correspond to the intermediate precursor clusters such as C3, C4, C6, and C10 in scRNA-seq ( Figures 5C, D ).

As NKT2 cells are the main subset in the thymus of 3-week-old mice while NKT1 cells with ST3 phenotype appear much later (21, 31), we tested the evolvement of these precursor subpopulations with age. In 2-week-old thymi, around 60% of iNKT cells were T-bet^-PLZF^hi/int (NKT2 and NKT17) and 40% were T-bet⁺PLZF^lo (NKT1) cells. In 8-week-old thymi, however, the percentages of T-bet^-PLZF^hi/int were less than 20% while T-bet⁺PLZF^lo cells reached more than 70% ( Figure 5F ). Similar changes were found in ST1 and ST3 cells in adult mice ( Figure 5F ). The frequency of ST2 cells was first increased in mice at 3-week of age but then decreased with age. In line with the changes of effector iNKT subsets or ST1/ST3 cells, the prevalence of P2 cells showed a substantial decrease while that of P4 cells showed an enrichment as mice aged ( Figure 5F ). No major alterations were found in the relative proportions of P1 and P5 subpopulations while that of P3 showed a first increased then decreased trend with age ( Figure 5F ). Together, these data from C57BL/6 mice suggest that P3/P5 and P4 subpopulations within ST1/ST2 are enriched with intermediate precursors for NK1.1⁺ NKT1 cells.

In haplo-BMT mice with GvHD, we observed higher percentages of P1/P2 and substantially lower percentages of P4/P5 subpopulations relative to those with BM only ( Figure 6A ). No difference was found in the relative proportions of CD138⁺ NKT17 subset or P3 subpopulation between the two BMT groups ( Figure 6B ). Notably, the upregulation of T-bet in the few differentiated NKT1 cells (ST3) and their precursor subpopulations (P3, P4, P5) disappeared in GvHD mice ( Figure 6B ). A failure of CD69 upregulation was also observed in P4/P5 subpopulations in GvHD mice ( Figure 6B ). PLZF expression was down-regulated in P4/P5 subpopulations in mice with BM only but was much milder in those with GvHD ( Figure 2B ). We also found higher expression of CD28, PD-1, higher levels of cell proliferation and glucose uptake in P4 and/or P5 subpopulations in GvHD mice ( Figure 6B ). These results suggest that thymic iNKT cells in GvHD mice had impaired NKT1 differentiation at CD44^int/hiNK1.1^- precursor stage with failure to efficiently down-regulate NKT2 signature and upregulate NKT1 signature.

The signals from TCR/CD1d, SLAM-SAP, and CD28/CD80/CD86 promote the upregulation of Egr2, PLZF, PD-1, and facilitate the expansion of iNKT cells (20–23, 36, 47). As thymic iNKT cells in GvHD mice also showed increased expression of Egr2 in ST2 and ST3 subsets that contain NKT1 and their precursors ( Figure 7A ), we assessed whether thymic microenvironment that provides TCR and costimulatory signals is altered in GvHD mice. DP thymocytes are critical for the positive selection of iNKT cells and maturation of CD44^hi NK1.1^- thymocytes (36, 48). Compared to the mice with BM only, those with GvHD showed similar levels of CD1d, SLAMF1, SLAMF6, and surface-bound CD80 (22) in DP thymocytes ( Figures 7B, C ). However, the levels of CD28 expression and surface-bound CD86 in DP thymocytes were significantly higher in GvHD than in BM-only groups ( Figures 7B, C ). Macrophages are the predominant cell type that provides TCR signaling required for NKT2 differentiation (49). We found an increased prevalence of macrophages with upregulated CD1d and CD80 expression in GvHD thymi ( Figure 7D ). These results suggest that GvHD mice have stronger TCR and co-stimulatory signaling in GvHD thymi that may be associated with the increased proliferation, glucose uptake, Egr2, PLZF and PD-1 expression in NK1.1^- NKT1 precursors or cells with ST2 phenotype.

Cytokines IL-15 and IL-25 provided by medullary thymic epithelial cells (mTECs) are critical for the differentiation of effector NKT1 and NKT2 cells, respectively (50). As shown in Figure 7E , the percentage of mTECs, the expressions of Il15 and its transpresenting receptor Il15ra (51) were comparable between the two groups ( Figure 7E ). The expression of CD122, a component of the receptor for IL-15, was lower in ST2 and ST3 cells in GvHD mice relative to the mice with BM only ( Figure 7E ). The binding of IL-15 receptor to IL-15 leads to the phosphorylation of STAT5 (52). Significantly higher levels of phosphorylated STAT5 were found in ST2 and ST3 cells in GvHD mice ( Figure 7E ). We also determined the transcription of Il25 in the thymus and found it higher in mice with GvHD. The expression of IL-25 receptor IL-17RB in iNKT cells was comparable between the two groups ( Figure 7E ). These results together suggest that the cytokine signaling for NKT1 and NKT2 differentiation was not impaired in GvHD thymi.

C57BL/6 (CD45.1 and CD45.2) and BALB/c mice were purchased from Peking University. The animals were kept in a specific pathogen-free facility at Peking University (Beijing, China) and were housed at constant temperature (22-23˚C) with a 12/12h-light/dark cycle and optimal humidity. All animals were allowed to standard lab chow and water ad libitum. The experimental procedures for the use and care of the animals were approved by the Ethics Committee of Peking University (LA2018042). We confirm that all experiments were performed in accordance with relevant guidelines and regulations.

F1 (H-2^b/d, 8-10-week old) mice obtained from the breeding of C57BL/6J (B6, H-2^b) and BALB/c (H-2^d) mice were given lethal total body irradiation (10 Gy) as a split dose. T cell-depleted bone marrow cells (5 x 10 (6)) with or without splenic cells (2 x 10 (6)) collected from B6 mice (6-8-week old) were suspended in 200 μl PBS and intravenously injected into F1 recipient mice 4-6 hours after irradiation (62). T cell depletion was performed by magnetic-bead separation using MicroBeads and the autoMACS system (Miltenyi Biotec, Auburn, CA) with negative selection for the CD3 surface antigen. In this model, haplo-BMT recipients with splenic cells developed reproducible GvHD (BM + Spl) as assessed by survival, clinical score while those without splenic cells did not (BM only). GVHD scores were determined by the analysis of recipients’ weight, activity, skin, fur ruffling, and posture as described previously (63).

Statistical analysis was done using GraphPad Prism 8 software (San Diego, CA). Weekly average weight change and GvHD scores were evaluated by two-way analysis of variance (ANOVA) with Sidak’s post-hoc comparisons. For all other statistical analysis in this work, we used unpaired or two-tailed paired Student’s t-test to evaluate metrics between the study group and the control group. A p value of < 0.05 was considered statistically significant, with more significant values denoted by the number of symbols * < 0.05, ** < 0.01, *** < 0.001, and **** < 0.0001.

Other materials and methods can be found in supplemental materials