Distinct developmental pathways generate functionally distinct populations of natural killer cells

Early NK cell development in BM

We performed single-cell RNA sequencing (scRNA-seq) on Lin⁻CD27⁺2B4⁺CD122⁺ cells sorted from BM, that were likely to contain NK cell lineage progenitors, as CD27 and 2B4 are expressed on ALPs, EILPs and ILCPs^15,16. We also performed scRNA-seq on sorted and pooled ALPs, EILP, ILCP and Lin⁻CD25⁺Sca1⁺ ILC2P, and separately, Lin^T−NK1.1⁺CD122⁺ group 1 ILCs from BM (Fig. 1a and Extended Data Fig. 1a,b)^15,16. We defined Flt3⁺Il7r⁺ ALP, Il2rb(encoding CD122)^loNfil3⁺Tcf7⁺Id2^lo EILP, Flt3⁻Il7r⁺Tcf7⁺Zbtb16^hi ILCP, Il2ra⁺Il7r⁺Gata3⁺ ILC2P, Klrb1c⁺Eomes⁺Tbx21⁺ NK cells and Klrb1c⁺Eomes⁻Tbx21⁺ ILC1 clusters (Fig. 1b). The UMAP revealed that the ILCP cluster, which was Zbtb16^hi (encoding PLZF), overlapped with some CD122⁺ progenitor cells, which included rNKPs (Figure 1a,b and Extended Data Fig. 1b,c), suggesting overlap between PLZF^hi rNKPs and ILCPs. A cluster identified as Il2rb⁺Flt3⁻Zbtb16^lo was named rNKP’ (Fig. 1a,b).

A Il2rb⁺Flt3⁺ cluster within CD122⁺ progenitor cells formed close to the EILP cluster (Fig. 1a,b and Extended data Fig. 1b). Flt3 is expressed by ALP and EILP, but downregulated in downstream progenitors such as ILCP¹⁵, suggesting the Il2rb⁺Flt3⁺ NK cell progenitor cluster could be an early-stage precursor for NK cells that we tentatively referred to as “early NK progenitor” (ENKP). The ENKP cluster did not express Klrc1, Klrc2 and Klrc3 (encoding NKG2A, NKG2C and NKG2E) (Extended Data Fig. 1d), suggesting ENKPs were distinct from aceNKPs¹⁷.

We predicted developmental relationships between ALP, EILP, ENKP, rNKP’, ILCP, and NK cell clusters using pseudotime reconstruction. One sequence of predicted development (lineage 1) was ALP->EILP->ENKP->rNKP’->NK (Extended Data Fig. 1e), which was independent of ILCP. ILC2 development¹⁵ was detected as ALP->EILP->ILCP->ILC2P (lineage 2)(Extended Data Fig. 1e). We did not detect an ILCP-dependent pathway of NK cell or ILC1 development (Extended Data Fig. 1e), suggesting intermediate populations may be missing in our dataset, consistent with evidence indicating ILCPs leave the BM and differentiate at extramedullary sites^18,19. In lineage 1, Flt3 was highly expressed in ALP and EILP clusters, maintained in the ENKP cluster, but downregulated in rNKP’ (Extended Data Fig. 1e), whereas Il2rb was upregulated starting from ENKP cluster (Extended Data Fig. 1e). These data suggested the existence of an ILCP-independent pathway of NK cell development in BM, and identified ENKPs as a candidate precursor for NK cells.

Consistent with the Il2rb⁺Flt3⁺ phenotype of the ENKP cluster identified by scRNA-seq, flow cytometry identified a Lin⁻NK1.1⁻CD5⁻CD3ε⁻CD27⁺2B4⁺Flt3⁺CD122⁺ population in the BM of C57BL/6 mice (Fig. 1c) that we also refer to as ENKP. ENKPs represented ~0.005% of live cells in BM, similar to EILP² (Fig. 1c) and were different from rNKP⁶. Numbers of ENKPs in BM were greatly reduced in Nfil3^−/− mice, but were intact in Eomes^−/−, Tcf7^−/− and Gata3^−/− mice (Fig. 1d,e and Extended Data Fig. 2a,b). NK cell numbers in BM were significantly reduced in Nfil3^−/−, Eomes^−/− and Tcf7^−/− mice, but were intact in Gata3^−/− mice, as expected (Fig. 1d,e and Extended Data Fig. 2a,b)^2,20,21,22. Thus, among the transcription factors assessed, only Nfil3 was required for ENKP development. These results indicated ENKPs were not downstream of ILCPs, as ILCPs do not develop in Tcf7^{−/− 16} or Gata3^−/− mice^15,23. ENKPs had high expression of α4β7 and Kit, low expression of IL-7Rα and did not express Thy1.2 (Extended Data Fig. 2c). Approximately 70% of ENKPs were fate-mapped by Il7r (Extended Data Fig. 2d), similar to the ratios of Il7r-fate-mapping in ALPs and EILPs¹⁵, indicating ENKPs developed from Il7r-expressing lymphoid progenitors, as predicted for NK cells²⁴. These results suggested ENKPs could represent an early NK progenitor population.

ENKPs are NK lineage-biased progenitors

We seeded ENKPs or ILCPs on OP9 stromal cells. At day 7 of culture with the cytokines stem cell factor (SCF), Flt3L, IL-7 and IL-2, both ILCPs and ENKPs gave rise to NK1.1⁺ cells (Fig. 2a), which include NK cells and ILC1s. However, ENKPs had very limited potential for ICOS⁺ ILC2/ILC3s, in contrast to ILCPs, which efficiently gave rise to ICOS⁺ ILC2/ILC3s (Fig. 2a). Thus, ENKPs were an NK cell-biased progenitor population. We sought to determine the lineage potential of single ENKPs in vitro. To distinguish NK cells from ILC1s in vitro, we used Eomes^GFP reporter mice to identify Eomes-GFP⁺ NK cells, and CD200R to identify ILC1s, as CD200R is highly expressed in ILC1s in vivo²⁵. In pilot experiments, when we cultured mature Eomes-GFP⁻ ILC1s and Eomes-GFP⁺ NK cells on OP9 cells for 7 days, Eomes-GFP⁻ ILC1s remained CD200R^hi and Eomes-GFP⁻, whereas NK cells were Eomes-GFP⁺ (~90%) and did not express CD200R (Extended Data Fig. 3a). Among the ICOS⁺ cells, we used α4β7 to further distinguish ICOS^loα4β7⁺ ILC3s and ICOS^hiα4β7⁻ ILC2s (Fig. 2b). The majority of ENKP-derived clones (74%) contained NK1.1⁺Eomes-GFP⁺CD200R⁻ NK-lineage cells, and 54% of ENKP clones contained only NK-lineage cells, whereas most ILCP-derived clones (92%) contained only helper ILC lineages (NK1.1⁺Eomes-GFP⁻CD200R⁺ ILC1, ILC2s or ILC3s)(Fig. 2c and Extended Data Fig. 3b). Thus, compared with ILCPs, ENKPs were NK cell lineage-biased progenitors in vitro. Mature NK cells generated negligible numbers of NK cells in vitro (Extended Data Fig. 3c), indicating that results obtained with ENKP and ILCPs could not be attributed to NK cell contamination. We cultured ENKPs in vitro with and without Flt3 ligand (Flt3L) to assess whether their NK-lineage potential depended on the expression of Flt3. ENKPs cultured without Flt3L generated fewer NK1.1⁺Eomes-GFP⁺CD200R⁻ NK cells than ENKPs cultured with Flt3L, although the results did not reach statistical significance (Extended Data Fig. 3d), indicating that Flt3 signaling may affect ENKP expansion.

We transferred ENKPs or ILCPs into NOD SCID gamma (NSG) mice and assessed the frequency of donor-derived CD45.2⁺CD3⁻CD19⁻NK1.1⁺NKp46⁺DX5⁺CD49a⁻ NK cells and CD45.2⁺CD3⁻CD19⁻NK1.1⁺NKp46⁺DX5⁻CD49a⁺ ILC1s within CD45.2⁺CD3⁻CD19⁻NK1.1⁺NKp46⁺ group 1 ILCs in liver. Six weeks later, ILCPs gave rise to NK cells and ILC1s, whereas ENKPs generated primarily NK cells (Fig. 2d). When co-transferred with an equal number of ALPs as competitors, ENKPs generated NK cells more efficiently than ILCPs in spleen and liver after 6 weeks (Fig. 2e and Extended Data Fig. 3e). ALPs, but not ILCPs or ENKPs, gave rise to B cells (Fig. 2e; Extended Data Fig. 3f–h). Further, the development of both CD45.2⁺ ENKP-derived NK cells (ENKP_NK cells) and CD45.2⁺ ILCP-derived NK cells (ILCP_NK cells) but not CD45.2⁺ ILC1s, depended on Eomes in vivo in spleen (Extended Data Fig. 3i), indicating ILCP_NK cells are distinct from ILC1s^26,27. Thus, ENKPs were NK-lineage-biased progenitors which can give rise to NK cells efficiently in vivo.

Next, we used PLZF-Cre Rosa26-YFP reporter mice to assess the developmental relationship between ILCP and ENKP. To eliminate the background labeling of PLZF-YFP (as PLZF is expressed before the specification of hematopoietic lineages³) we transferred CD45.2⁺PLZF-YFP⁻Lin⁻Kit⁺ hematopoietic progenitors from PLZF-Cre Rosa26-YFP reporter mice into lethally irradiated CD45.1⁺ wild-type mice (PLZF lineage-tracing chimeras). At 12–16 weeks after transfer, NK cells and ILC1s in the liver were labeled at ratios of ~20% and ~65% (Fig 3a), respectively, as reported^3,7, while ENKPs were PLZF-YFP⁻ (Fig 3a), indicating they did not develop from ILCPs. Next, we transferred PLZF-YFP⁻ ENKPs, PLZF-YFP⁻ ILCPs or PLZF-YFP⁻ ALPs from PLZF-Cre Rosa26-YFP reporter mice into NSG recipient mice. Unlike ILCP_NK cells, of which ~70% were PLZF-YFP⁺, only ~10% of ENKP_NK cells were PLZF-YFP⁺ (Fig. 3b), suggesting that ENKP_NK cells and ILCP_NK cells had different developmental pathways. ~25% of ALP-derived NK cells (ALP_NK cells) were PLZF-YFP⁺, similar to total NK cells (Fig 3a,b), thus ALP_NK cells were likely a mixture of ENKP-derived and ILCP-derived NK cells. In summary, these results indicated that ENKPs developed independently of ILCPs, and that there were at least two pathways for NK cell development.

ENKP_NK cells and ILCP_NK cells transcriptionally differ

We adoptively transferred CD45.2⁺ ENKPs, ILCPs or ALPs into NSG mice, sorted CD45.2⁺ NK cells from the spleen of recipient mice at 6 weeks post-transfer and performed multiplexed scRNA-seq for ENKP_NK cells, ILCP_NK cells and ALP_NK cells. ENKP_NK cells and ILCP_NK cells expressed comparable levels of Eomes, Tbx21 (encoding T-bet) and Klrk1 (encoding NKG2D) (Fig. 4b and Extended Data Fig. 4b). Genes expressed at higher levels in ENKP_NK cells than in ILCP_NK cells included Ly49 receptors (Klra8 and Klra4), cytotoxic genes (Gzma, Gzmb, and Prf1) and transcription factors (Zeb2, Irf8, and Klf2) (Fig. 4a,b and Extended Data Fig. 4a–c). ILCP_NK cells had higher expression of activating receptor Cd226, chemokine receptors Ccr2 and Cxcr3, and transcription factor Tcf7 (Fig. 4a,b and Extended Data Fig. 4a–c). ENKP_NK cells were enriched for cytotoxicity pathways (Extended Data Fig. 4d), whereas ILCP_NK cells were characterized by cell-cell adhesion and regulation of IFN-γ production (Extended Data Fig. 4d). Clustering of ALP_NK cells showed two major clusters with the ENKP_NK cluster more abundant than the ILCP_NK cluster (Extended Data Fig. 4e), suggesting ALP_NK cells were a mixture of ENKP_NK cells and ILCP_NK cells, with ENKP_NK cells being more abundant.

UMAP of scRNA-seq of ENKP_NK cells, ILCP_NK cells and ALP_NK cells identified that each NK subset fell into 2 clusters, with ENKP_NK cells enriched in cluster_0 and ILCP_NK cells enriched in cluster_1 (Extended Data Fig 4f). Genes differentially expressed between cluster_0 and cluster_1 (such as Zeb2, Prf1, Klra8, Xcl1 and Cxcr3) were similar to genes differentially expressed between ENKP_NK cells and ILCP_NK cells (Extended Data Fig. 4a, g). Some genes such as Klra8 (encoding Ly49H) were differentially expressed between ENKP_NK cells and ILCP_NK cells in the same clusters (Extended Data Fig. 4h). Thus, ENKP_NK cells and ILCP_NK cells were transcriptionally different.

We used the scRNA-seq dataset of Lin^T−NK1.1⁺CD122⁺ group 1 ILCs from BM and published scRNA-seq dataset of liver group 1 ILCs from steady-state wild-type mice²⁸ to compare the transcriptomes of ENKP_NK cells, ILCP_NK cells and ILC1s. Based on the expression levels of genes which were differentially expressed between ENKP_NK cells and ILCP_NK cells (see Methods) we first identified ENKP_NK cells and ILCP_NK cells within total NK cells in the BM and liver (Extended Data Fig. 4i). Then we compared the transcriptome of both NK subsets and ILC1s in the same tissues, as transcriptome can be influenced by tissue microenvironment²⁹ and batch effects. In both liver and BM, ILCP_NK cells clustered together with ENKP_NK cells rather than with ILC1s (Extended Data Fig. 4i), indicating ILCP_NK cells were transcriptionally more similar to ENKP_NK cells than ILC1s. ILCP_NK cells expressed intermediate levels of genes differentially expressed between NK cells and ILC1s¹⁷ (see Methods)(Extended Data Fig. 4j), suggesting ILCP_NK cells may share some characteristics of both NK cells and ILC1s.

Ly49H recognizes the viral protein m157 present on cells infected by MCMV¹⁴. At 6 weeks after adoptive transfer of ENKPs or ILCPs into NSG mice, ILCP_NK cells were mostly Ly49H⁻ whereas ~55% of ENKP_NK cells expressed Ly49H (Fig. 4c). In adoptive transfer experiments with ALPs as competitors, ENKPs gave rise to Ly49H⁺ NK cells more efficiently than ILCPs (Extended Data Fig. 5a), suggesting the majority of Ly49H⁺ NK cells developed from ENKPs. ENKPs were Ly49H⁻ (Extended Data Fig. 5b), indicating Ly49H was upregulated downstream during NK cell development. Consistent with the scRNA-seq data, ~35% of ENKP_NK cells were Ly49D⁺ (encoded by Klra4) whereas ILCP_NK cells were Ly49D⁻, but comparable frequencies of ENKP_NK cells and ILCP_NK cells were Ly49E/F⁺ (encoded by Klra5/Klra6) (Extended Data Fig. 5c). ILCP_NK cells had higher frequencies (~80%) of TCF1-YFP⁺ (encoded by Tcf7^YFP) than ENKP_NK cells (~40%) (Extended Data Fig. 5d–e).

The PLZF-YFP⁺/Ly49H⁺ ratio in NK cells in spleen, liver, lung, mesenteric lymph modes (mLN), thymus, salivary glands, uterus and BM of PLZF lineage-tracing chimeras indicated an inverse correlation between PLZF-YFP labeling and Ly49H expression (Extended Data Fig. 5f–i). PLZF-YFP⁺ ratios were lower in Ly49H⁺ NK cells compared to Ly49H⁻ NK cells (Extended Data Fig. 5f). PLZF-YFP⁺ ratios in NK cells were lower in liver, spleen and lung (~18%) and higher in mLN (25%) and thymus (46%) (Extended Data Fig. 5f). PLZF-YFP⁺ ratios in NK cells in other tissues including salivary glands (6%), uterus (17%) and BM (20%) were also low (Extended Data Fig. 5h). Because ENKP_NK cells and ILCP_NK cells were marked at different ratios in NSG mice reconstituted with PLZF-YFP⁻ ENKPs or PLZF-YFP⁻ ILCPs (Fig. 3b), and assuming that only ENKP_NK cells and ILCP_NK cells contributed to the total NK cell population, we used the PLZF-YFP⁺ ratios in NK cell populations in PLZF lineage-tracing chimeras (Extended Data Fig. 5f,h) and calculated that more than 80% of NK cells in spleen, liver, lung, salivary glands, uterus and BM developed from ENKPs (see Methods).

Human NK cells have two major subsets, CD56^dimCD16⁺ (CD56^dim) and CD56^brightCD16⁻ (CD56^bright) NK cells, with the CD56^dim subset further subdivided into CD57⁻ and CD57⁺ cells and the CD56^bright NK cells more efficient in producing cytokines including IFN-γ³⁰. To compare mouse and human NK cells, we analyzed published CITE-seq and scRNA-seq data^31,32 of human blood NK cells. By surface protein expression we distinguished CD56^bright, CD56^dimCD57⁻ and CD56^dimCD57⁺ clusters (Extended Data Fig. 6a–b) present in both human cytomegalovirus (HCMV) seropositive and HCMV seronegative individuals, with an enrichment of CD56^dimCD57⁺ NK cells in HCMV⁺ samples (Fig. 4e). CD56^dimCD57⁻ and CD56^dimCD57⁺ NK cells had higher expression of genes upregulated in ENKP_NK cells, whereas CD56^bright NK cells had higher expression of genes upregulated in ILCP_NK cells (Fig. 4d). NK cells expressing KIR2DL1, KIR2DL2 and KIR2DL3 expand in HCMV infection³³, and KIR2DL1, KIR2DL2 and KIR2DL3 were expressed more highly in CD56^dim NK cells than in CD56^bright NK cells (Extended Data Fig. 6c)³⁴. CD56^dimCD57⁻ and CD56^dimCD57⁺ NK cells had higher ENKP_NK scores, whereas CD56^bright NK cells had higher ILCP_NK scores (see Methods) in both the HCMV⁺ and HCMV⁻ donors (Fig. 4f), suggesting HCMV status did not affect the transcriptional similarity between human and mouse NK subsets. We assessed the correlation between mouse and human NK clusters based on transcriptional similarity, and found mouse NK cluster_0, which was enriched in ENKP_NK cells, correlated more with human CD56^dimCD57⁻ and CD56^dimCD57⁺ NK cells, whereas mouse NK cluster_1, which was enriched in ILCP_NK cells, correlated more with human CD56^bright NK cells (Extended Data Fig. 6d). ENKP_NK cells correlated more with human CD56^dimCD57⁻ and CD56^dimCD57 NK cells, whereas ILCP_NK cells correlated more with human CD56^bright NK cells (Extended Data Fig. 6e). Thus, there is developmental heterogeneity in NK cells, that may be evolutionarily conserved between human and mouse.

ENKP_NK cells respond robustly to MCMV infection

We reconstituted NSG mice with equal numbers of ENKPs or ILCPs and intraperitoneally challenged them with MCMV 5 weeks post-transfer, then assessed NK cells in spleen and blood. At day 7 post-infection, the numbers of spleen ENKP_NK cells were approximately 150-fold higher than spleen ILCP_NK cells (Fig. 5c) whereas the numbers of spleen ENKP_NK cells were only 4-fold higher than spleen ILCP_NK cells without MCMV challenge (Fig. 5b). Over 99% of spleen ENKP_NK cells were Ly49H⁺ NK at day 7 post-MCMV infection, compared to ~20% of ILCP_NK cells (Fig. 5a). Ly49H⁺ ENKP_NK cells in blood were undetectable before infection but represented ~1.5% of total blood live cells at day 7 post-MCMV infection (Extended Data Fig. 7a–c). We further co-transferred ENKPs and ILCPs in the same NSG mice at a 1:4 ratio, based on the fact that ILCPs generated 4-fold fewer NK cells than ENKPs (Fig. 5b). Seven days post-MCMV infection, ENKP_NK cells represented ~90% of total spleen NK cells, compared to ILCP_NK cells which represented ~10% in the same environment (Extended Data Fig. 7d). In PLZF lineage-tracing chimeras, the ratio of PLZF-YFP⁺ spleen NK cells at steady-state (14.5%) decreased by day 7 post MCMV infection (8.6%)(Fig. 5d–e). Together, these observations indicated ENKP_NK cells expanded more robustly than ILCP_NK cells upon MCMV infection.

Surface markers identify ENKP_NK and ILCP_NK cells

Some transcripts encoding surface proteins were differentially expressed between ENKP_NK cells and ILCP_NK cells, and we verified protein expression of Ly49H, Ly49D, CD226, Thy1.2 and CXCR3 by flow cytometry (Fig. 4d, Extended Data Fig. 5c and 8a). We used combinations of these markers to identify an Ly49H⁺ and/or Ly49D⁺ ENKP_NK-enriched cell population (~10% PLZF-YFP⁺) and an Ly49H⁻Ly49D⁻CD226⁺Thy1.2⁺ or Ly49H⁻Ly49D⁻CD226⁺CXCR3⁺ ILCP_NK-enriched cell population (~50% PLZF-YFP⁺, similar to ~55% PLZF-YFP⁺ in ILC1s) (Fig. 6a–c). In TCF1-YFP mice, ENKP_NK-enriched cells had lower ratios of TCF1-YFP⁺ cells (~70%) than ILCP_NK-enriched cells (~95%) (Extended Data Fig. 8b). The PLZF-YFP⁺ ratios of Ly49H⁺ and/or Ly49D⁺, Ly49H⁻Ly49D⁻CD226⁺Thy1.2⁺ and Ly49H⁻Ly49D⁻CD226⁺CXCR3⁺ NK subsets remained unchanged at day 7 post-MCMV infection compared to steady-state (Extended Data Fig. 8c), indicating ENKP_NK cells and ILCP_NK cells did not substantially interconvert. The expansion of the ENKP_NK-enriched subset in chimeras and C57BL/6 mice at day 7 post-MCMV infection was similar to the population expansion of Ly49H⁺ NK cells in wild-type mice post-MCMV infection (Extended Data Fig. 8d–e)³⁵. ENKP_NK-enriched and ILCP_NK-enriched cells, but not ILC1s, were greatly reduced in Eomes^−/− mice (Extended Data Fig. 8f), but remained intact in Zbtb16^−/− mice (Extended Data Fig. 8g), indicating the developmental divergence between ENKP_NK cells and ILCP_NK cells did not depend on PLZF.

Although downregulation of CD27 and upregulation of CD11b and KLRG1 correlates with NK cell maturation^36,37, ~30% of CD27⁺CD11b⁻ NK cells remained CD11b⁻ after adoptive transfer into unirradiated wild-type mice³⁶. We found that CD27⁺CD11b⁻ cells were enriched in ILCP_NK cells compared to ENKP_NK cells after adoptive transfer of ILCPs or ENKPs in NSG mice (Extended Data Fig. 9a), and the PLZF-YFP⁺ ratio in CD27⁺CD11b⁻ NK cells (~25%) was higher than in CD27⁻CD11b⁺ NK cells (~16%) (Extended Data Fig. 9b). Similarly, ENKP_NK cells had higher expression of KLRG1 compared to ILCP_NK cells (Extended Data Fig. 9c) and the PLZF-YFP⁺ ratio in KLRG1⁻CD11b⁻ “immature” NK cells (22%) was higher than in KLRG1⁺CD11b⁺ NK cells (18%) (Extended Data Fig. 9d), indicating that previously-defined immature NK cell subsets were likely a mixture of immature NK cells and mature ILCP_NK cells. Ly49H⁺CD27⁺CD62L⁻ tissue-resident NK cells³⁸ were phenotypically similar to ILCP_NK cells (Extended Data Fig. 9c), but the PLZF-YFP⁺ ratio in Ly49H⁺CD27⁺CD62L⁻ NK cells was low (8%)(Extended Data Fig. 9e), indicating they are likely a subset of ENKP_NK cells.

We sorted ILCP_NK-enriched and ENKP_NK-enriched cells from spleen of C57BL/6 mice and stimulated them in vitro with PMA/Ionomycin for 3 hours to assess expression of IFN-γ, TNF and CD107a (Extended Data Fig. 10a). The ratios of IFN-γ⁺TNF⁻, IFN-γ⁺TNF⁺ and IFN-γ⁺TNF⁺ cells in ILCP_NK-enriched cells (20%, 3% and 2.5%, respectively) were higher than in ENKP_NK-enriched cells (10%, 0.5% and 0.5%, respectively) but the CD107a⁺ ratio was comparable between ILCP_NK-enriched cells and ENKP_NK-enriched cells (Extended Data Fig. 10a). ENKP_NK-enriched cells and ILCP_NK-enriched cells killed Yac-1 cells similarly (Extended Data Fig. 10b). In vitro differentiated ILCP_NK cells also had higher frequencies of cells expressing IFN-γ upon stimulation as compared to ENKP_NK cells (Extended Data Fig. 10c). These data indicated that ILCP_NK cells have better cytokine producing ability than ENKP_NK cells.

NK cells contribute to immunity against Salmonella infection³⁹. We assessed cell number and expression of IFN-γ in ILCP_NK-enriched cell populations and ENKP_NK-enriched cell populations in the cecum lamina propria of C57BL/6 mice at day 1, 3 and 5 post-infection with Salmonella through oral gavage (Fig. 6d–e). Although both NK cell subsets increased in number after Salmonella infection (day 3, 18-fold for ENKP_NK-enriched, 13-fold for ILCP_NK-enriched; day 5, 5.7-fold for ENKP_NK-enriched, 3.5-fold for ILCP_NK-enriched) (Fig. 6d and Extended Data Fig. 10d), ILCP_NK-enriched cells had higher expression of IFN-γ than ENKP_NK-enriched cells (Fig. 6e and Extended Data Fig. 10e), while ILC1s did not expand and had significantly lower expression of IFN-γ than ILCP_NK-enriched cells (Fig. 6e and Extended Data Fig. 10e). NK cells also help control herpes simplex virus (HSV) infection⁴⁰. We assessed cell numbers and IFN-γ expression in ILCP_NK-enriched and ENKP_NK-enriched cell populations in ear skin of C57BL/6 mice infected with HSV by intradermal injection. Both NK subsets expanded at day 3 and 7 post-infection compared to uninfected mice (day 3, 21-fold for ENKP_NK-enriched, 34-fold for ILCP_NK-enriched; day 7, 24-fold for ENKP_NK-enriched, 40-fold for ILCP_NK-enriched) (Fig. 6f and Extended Data Fig. 10f). A higher percentage of ILCP_NK-enriched cells (26% at day 3, 11% at day 7) were IFN-γ⁺ compared to ENKP_NK-enriched cells (18% at day 3, 7% at day 7)(Fig. 6g and Extended Data Fig. 10g). Although ILC1s expanded post-HSV infection (day 3, 3-fold; day 7, 11-fold), the expansion of ILCP_NK-enriched cells was more robust (Fig. 6f). Thus, we identified surface markers that distinguish ILCP_NK-enriched and ENKP_NK-enriched subsets in C57BL/6 mice, and ILCP_NK cells and ENKP_NK cells contributed to immune responses to infection, but with more robust cytokine production by ILCP_NK-enriched cells.

Mice

Tcf7^YFP, Tcf7^GFP, Zbtb16^GFPcre, Tcf7^−/−, Nfil3^−/−, Il7r^cre, Gata3^flox, Eomes^flox, Eomes^GFP mice were described previously^{2,3,16,23,24,47,48,49,50}. B6-Ly5.2 (CD45.1) and B6.129X1-Gt(ROSA)26Sor^tm1(EYFP)Cos/J (strain #: 006148) mice were obtained from Jackson Laboratory. NOD-scid IL2Rgamma^null (NSG) mice were obtained from National Cancer Institute. All mice except NSG mice are maintained on C57BL/6 background. Mice were 6–18 weeks of age and of either sex. Animal procedures were approved by relevant National Institutes of Health Animal Care and Use Committees.

Antibodies and flow cytometry

BM, spleen and liver samples were incubated with a mix of purified rat, mouse and hamster immunoglobulin G (IgG) before the addition of specific antibodies, and then stained and analyzed in FACS buffer (PBS containing 2 mM EDTA and 0.5% FBS). The following anti-mouse antibodies were used for immunostaining: 2B4 (eBio244F4, PE-Cy7, Thermo Fisher Scientific, 25-2441-82), α4β7 (DATK32, BV711, BD biosciences, 740701; DATK32, PerCP-Cy5.5, Thermo Fisher Scientific, 46-5887-82), B220 (RA3-6B2, biotin, Thermo Fisher Scientific, 13-0452-86), CD11c (N418, biotin, Thermo Fisher Scientific, 13-0114-85), CD122 (TM-b1, PE, Thermo Fisher Scientific, 12-1222-82), CD19 (eBio1D3, biotin, Thermo Fisher Scientific, 13-0193-85; 6D5, BV605, BioLegend, 115540), CD200R (OX110, APC, Thermo Fisher Scientific, 17-5201-82), CD25 (BC96, PE, Thermo Fisher Scientific, 12-0259-80), CD27 (LG.7F9, PerCP-Cy5.5, Thermo Fisher Scientific, 46-0271-82), CD3 (17A2, AF700, Thermo Fisher Scientific, 56-0032-82), CD3e (eBio500A2, biotin, Thermo Fisher Scientific, 13-0033-86; 145-2C11, BV510, BioLegend, 100353; 145-2C11, SB600, Thermo Fisher Scientific, 63-0031-82), CD4 (GK1.5, biotin, Thermo Fisher Scientific, 13-0041-82; RM4-5, SB600, Thermo Fisher Scientific, 63-0042-82), CD45.1 (A20, eF450, Thermo Fisher Scientific, 48-0453-82; A20, PE-eF610, Thermo Fisher Scientific, 61-0453-82), CD45.2 (104, eF450, Thermo Fisher Scientific, 48-0454-82; 104, PerCP-Cy5.5, BioLegend, 109828), CD49a (HMα1, APC, BioLegend, 142606), CD49b (DX5, AF700, Thermo Fisher Scientific, 56-5971-82; DX5, APC-Cy7, Thermo Fisher Scientific, 47-5971-82; DX5, PE-eF610, Thermo Fisher Scientific, 61-5971-82), CD5 (53-7.3, BV605, BD biosciences, 563194), CD8α (53-6.7, biotin, Thermo Fisher Scientific, 13-0081-85; 53-6.7, SB600, Thermo Fisher Scientific, 63-0081-82), CD8β (CT-CD8b, biotin, Thermo Fisher Scientific, MA5-17845; eBioH35-17.2, SB600, Thermo Fisher Scientific, 63-0083-82), Flt3 (A2F10.1, PE-CF594, BD biosciences, 562537), Gr-1 (RB6-8C5, biotin, Thermo Fisher Scientific, 13-5931-86), ICOS (7E.17G9, PE, Thermo Fisher Scientific, 12-9942-82), IL-7Rα (A7R34, APC, Thermo Fisher Scientific, 17-1271-82), Kit (2B8, BV785, BioLegend, 105841; 2B8, PE-Cy7, Thermo Fisher Scientific, 25-1171-82), Ly49H (3D10, PE, BioLegend, 144706; 3D10, FITC, Thermo Fisher Scientific, 11-5886-82), Ly49D (4E5, BV 711, BD biosciences, 742559), Mac-1 (M1/70, biotin, Thermo Fisher Scientific, 13-0112-86), NK1.1 (PK136, APC-Cy7, Thermo Fisher Scientific, 47-5941-82; PK136, biotin, Thermo Fisher Scientific, 13-5941-85; PK136, BV785, BioLegend, 108749), NKp46 (29A1.4, PE-Cy7, Thermo Fisher Scientific, 25-3351-82), streptavidin (eF450, Thermo Fisher Scientific, 48-4317-82; APC-Cy7, Thermo Fisher Scientific, 47-4317-82), Sca1 (D7, AF700, Thermo Fisher Scientific, 56-5981-82), TCRβ (H57-597, biotin, Thermo Fisher Scientific, 13-5961-85; H57-597, SB600, Thermo Fisher Scientific, 63-5961-82), TCRγδ (eBioGL3, biotin, Thermo Fisher Scientific, 13-5711-85; eBioGL3, SB600, Thermo Fisher Scientific, 63-5711-82), Ter119 (TER-119, biotin, Thermo Fisher Scientific, 13-5921-85), Thy1.2 (53-2.1, AF700, BioLegend, 140324; 53-2.1, BV605, BioLegend, 140318), CD226 (TX42.1, BV650, BioLegend, 133621), CXCR3 (CXCR3-173, APC, Thermo Fisher Scientific, 17-1831-82). The lineage ‘cocktail’ (Lin) is a mix of the following biotin antibodies: anti-Gr-1, NK1.1, CD19, Ter119, CD11c, CD3, CD4, CD8α, CD8β, TCRβ, TCRγδ, Ly-6D, B220, Mac-1. T lineage ‘cocktail’ (Lin^T) contains anti-CD3, CD4, CD8α, CD8β, TCRβ, TCRγδ. Non-T/NK lineage ‘cocktail’ contains anti-Gr-1, CD19, Ter119, CD11c, Ly-6D, B220.

BM progenitor definition and isolation

BM progenitors were defined and isolated by flow cytometric sort as follows. ALPs (Lin⁻Kit⁺α4β7⁻Tcf7⁻Flt3⁺IL-7Rα⁺), ILCPs (Lin⁻Kit⁺α4β7⁺Tcf7^YFP+Flt3⁻Thy1.2⁺ or Lin⁻Kit⁺α4β7⁺PD-1⁺Flt3⁻Thy1.2⁺) and ENKPs (Lin⁻CD27⁺2B4⁺Flt3⁺CD122⁺) were isolated from Tcf7^YFP mice. For PLZF lineage-tracing (Fig 3b), ILCPs were sorted as Lin⁻Kit⁺α4β7⁺Tcf7^GFP+Flt3⁻Thy1.2⁺. Before sorting of BM progenitors, bone marrow samples were stained with biotin lineage ‘cocktail’ (Lin), and then depleted by streptavidin beads (Biolegend, 480016) according to the manufacturer’s instructions.

Cell culture

BM progenitors were sorted by flow cytometric cell sorting. BM progenitors were cultured on irradiated OP9 stromal cells in α-MEM supplemented with 20% fetal bovine serum, penicillin and streptomycin. Bulk cultures were supplemented with 30 ng/ml SCF, 30 ng/ml IL-7, 30 ng/ml Flt3L and 30 ng/ml IL-2, and progenitor differentiation was examined at 7d after seeding. Single cell cultures were supplemented with 30 ng/ml SCF, 30 ng/ml IL-7, 30 ng/ml Flt3L and 10 ng/ml IL-2, and progenitor differentiation was examined at 10d after seeding. For single cell assay, positive well was defined as > 10 total ILCs, and positive lineage was defined as frequency > 10% and number > = 3. For in vitro stimulation, NK cells sorted from bulk culture were incubated with IL-12 (20 ng/ml), IL-18 (10 ng/ml) or PMA (50 ng/ml), Ionomycin (500 ng/ml) for 1 hr, then Brefeldin A was added to a final concentration of 5 ug/ml for another 2 hrs’ incubation. Cytokines were purchased from PeproTech.

Adoptive transfer

For ENKP and ILCP transfer, around 200–500 progenitors (CD45.2) were injected into NSG recipient mice (CD45.1) if not mentioned specifically. For competitive assay, around 300–500 test progenitors (CD45.2) were mixed with same numbers of competitor progenitors (CD45.1) and injected into NSG recipient mice (CD45.1). Reconstitution of lymphocytes was examined at 6-weeks post-transfer. For PLZF lineage-tracing chimeras, 3X10⁵ YFP⁻ LK (Lin⁻Kit⁺) progenitors (CD45.2) obtained by cell sorting were injected into lethally irradiated (850 rads) B6-Ly5.2 recipient mice (CD45.1), or injected into sublethally irradiated (250 rads) NSG mice. Reconstitution of lymphocytes and progenitors was examined at 12–16 weeks post-transfer.

Calculation of ENKP_NK cell and ILCP_NK cell contributions to total NK population

To calculate contribution percentage of ENKP_NK cells and ILCP_NK cells, the following equation set was used: 10%*X + 70%*Y = a%; X + Y = 100%. 10% was the observed PLZF tracing ratio of ENKP_NK cells, 70% was the observed PLZF tracing ratio of ILCP_NK cells, a% was the observed PLZF tracing ratio of total NK cells in different tissues (for example, in liver a% was 18%, Extended Fig. 5h), X was the calculated percentage of ENKP_NK cells and Y was the calculated percentage of ILCP_NK cells.

scRNA-seq and analysis for mouse samples

Cells from BM were isolated by sorting from Tcf7^YFP mice (Fig. 1). NK cells from spleen were sorted from NSG mice adoptively transferred with ALPs, ENKPs or ILCPs (Fig 4). scRNA-seq libraries were prepared using 10X Genomics Single Cell 3’ Reagent Kits v3.1, according to the manufacturer’s instructions. The obtained libraries were sequenced with a NextSeq system. Base calling was performed using RTA 3.9.2 (BM samples) or RTA 3.9.25 (spleen sample), and alignment was performed using cellranger v6.1.1 (STAR 2.7.2a). For spleen sample, hashtaging antibodies used for the analysis were mHTO01, mHTO02, and mHTO03 TotalSeqA bought from BioLegend, and demultiplexing was performed using cellranger v6.1.1 (Bcl2fastq 2.20.0). Cells with low UMI counts were determined by the 10x Genomics Cell Ranger Algorithm. Single-cell analysis was performed using Seurat (version 4.1.0). Cells with >6% mitochondrial gene content and >8000 genes (BM samples) or >4000 genes (spleen sample) were excluded from analysis. Mitochondrial content and cell-cycle related genes were regressed out. T cell contaminants were removed from BM samples. Cells were analyzed mainly by Seurat’s functions⁵¹. Cells were clustered using Seurat’s “FindClusters” function. Pseudotime analysis was performed using “Slingshot” package (version 2.2.1)⁵². Differentially expressed genes (DEGs) were identified using Seurat’s “FindMarkers” function considering those genes expressed in at least 10% of cells in the clusters with a minimum log2 fold change threshold of 0.25. Violin plots were made using Seurat’s normalized data. Average gene expression from cells of the same developmental origin was generated using the Seurat’s “AverageExpression” function. Gene ontology pathway analysis of DEGs was carried out with the “enrichGO” function. Volcano plots were generated using “ggplot2” package (version 3.4.0).

To compare transcriptomes of mouse NK subsets and ILC1s, liver ILC1 and NK cell data were downloaded from the gene expression omnibus (GSE163452²⁸) and loaded into Seurat. In addition, in house generated BM derived NK and ILC1 scRNAseq (see Figure 1a) was loaded into Seurat. For published liver data, the hashtags used to label the publicly available data were demultiplexed using the HTODemux function, following CLR normalization. Cells from Hobit-KO mice were discarded, and cells were flagged as low quality and removed if they had <= 1000 or >= 8000 reads, and > 6% mitochondrial reads. All datasets were processed according to a standard Seurat pipeline. The FindNeighbors and FindClusters functions were used to cluster the cells and identify NK cells or ILC1s. Gene set scores were calculated using ENKP_NK gene list (Zeb2, Irf8, Klf2, Gzmb, Gzma, Prf1, Itgam, Ccl5, Klra3, Klra4, Klra7, Klra8, Klra9) and ILCP_NK gene list (Tcf7, Cd160, Cd226, Cd7, Cd27, Ccr2, Cxcr3, Xcl1, Ltb, Ifng, Cd69). These scores were fed into the kmeans function, allowing cells to be clustered using these gene scores. This resulted in the identification of 2 NK cell clusters, allowing them to be annotated as ENKP_NK or ILCP_NK. For each population of NK cells and ILC1s the average gene expression was calculated using the AverageExpression function, and Euclidean distance was calculated using the dist function. The averaged data was used to examine expression of genes previously determined to be differentially expressed between NK cells and ILC1s¹⁷ that also were significantly different between NK cells and ILC1s in both liver and BM, with a fold change >= 0.25 and FDR of <= 0.05.

scRNA-seq analysis of human NK cell

The human NK scRNAseq data was obtained from GSE130430³¹ and GSE197037³² was analysed using a standard Seurat analysis. Batch correction was performed using the RunHarmony function from the “harmony” package to correct batch effects. The CITEseq data was normalized by applying a centered log ratio normalization. Clusters identified using the RNA modality were annotated based on the protein expression of CD56 and CD57. The annotation was extended using the glmnet and cv.glmnet function to all datasets. Cross species comparisons were done by identifying homologs between mouse and human using the homologene package; and by turning mouse genes to uppercase. Differentially expressed gene (DEG)s were identified by applying the “FindMarkers” function from Seurat within the populations or samples of one species. The DEG shared between species were retained and for each population or sample the counts of individual cells were summed using the “AggregateExpression” function and log transformed, generating pseudobulk data. The genes of the pseudobulk data were centered by dividing by their average gene expression. Spearman correlation was done on these matrices using the “cor” function, allowing correlation of the populations between different species. Scores were calculated by averageing the expression of gene sets identified in the mouse scRNA-seq data (ENKP_NK score: ZEB2, IRF8, KLF2, GZMB, GZMA, PRF1, ITGAM and CCL5; ILCP_NK score: TCF7, CD160, CD226, CD7, CCR2, CXCR3, XCL1, LTB, IFNG and CD69) and substracting their average by the average gene expression of 100 randomly sampled genes. Plots were generated using either the “Seurat” or “ggplot2” packages.

MCMV infection

MCMV (Smith strain) was prepared as described previously⁵³ and virus titer was 2200 PFU/ml. MCMV stock was 1:400 diluted in 1XPBS and 200 μl of MCMV diluent was intraperitoneally injected into each mouse. Blood and spleen samples were analyzed by flow cytometry 7 days after infection.

Salmonella infection and cecal immune cell isolation

Salmonella Typhimurium strain IR715, a nalidixic acid resistant strain, was used for all experiments. Bacteria was cultured in Luria broth (LB) at 37°C shaking for 18 hours. Cells were washed in PBS and adjusted to final concentration of 5×10⁸ CFU/mL in PBS. C57BL6 mice, 8–10-week-old, were gavaged with 20mg streptomycin, and fasted overnight. 24 hours later mice were gavaged with 10⁸ CFU S. Typhimurium in 200 uL PBS. Infection was confirmed by measuring CFU at 24 hours by homogenizing fecal pellets in PBS and plating on LB agar with 50μg/mL nalidixic acid. Cecal contents was also plated at takedown to quantify S. Typhimurium CFU. Mice were sacrificed at 1, 3 or 5 days post-infection to analyze gut immune cells in the cecum. Immune cells from the cecum of S. Typhimurium infected mice were isolated as previously described with some modifications⁵⁴. Briefly, ceca were extracted and contents removed into sterile PBS for CFU quantification. Ceca were then cut open longitudinally, washed and incubated in complete medium (RPMI 1640 plus 3% fetal bovine serum, 1 mM pyruvate, 0.1 mM nonessential amino acids, 2 mM glutamine, 20 mM HEPES, 100 U/ml penicillin, 0.1 mg/ml streptomycin, and 50 μM β-mercaptoethanol) with 5mM EDTA and 1mM dithiothreitol for 15 minutes at 37°C stirring. Ceca were shaken 3 times in 10mL RPMI + 2mM EDTA to remove epithelial cells, then cut into small pieces for digestion in RPMI media containing 0.5 mg/ml Collagenase Type 4 (Worthington Biochemical), 8% FBS and 50 mM CaCl₂. After a 30-minute digestion at room temperature with stirring, cells were passed through a 70μm filter and washed with 10mL complete media. Finally, cells were pelleted, resuspended in 37.5% Percoll, spun at 600g for 5 minutes, and cell pellets were resuspended in complete media for flow cytometry analysis.

Herpes simplex virus-1 (HSV-1) infection and ear digestion

Female mice were injected intradermally with 15uL of HSV-1 (1×10⁶ PFU in sterile tissue-culture-grade PBS) in the ear pinnae using a 29G needle. Control mice were injected with same amount of sterile tissue-culture-grade PBS. Immune responses were analyzed at day 3 or 7 post-infection. At day 3 or 7 post-infection, mice were euthanized and ears were cut and the ventral and dorsal sheets of ears were separated and digested in the media with DNase and Liberase (RPMI 1640 media with 2 mM L-glutamine, 1 mM sodium pyruvate and nonessential amino acids, 55 mM β-mercaptoethanol, 20 mM HEPES, 100 U/ml penicillin, 100 mg/ml streptomycin, 0.5 mg/ml DNase and 0.025 mg/ml Liberase TL purified enzyme blend) at 37°C for 1 hour and 45min. Digested ears were homogenized and filtered through 70-μm cell strainer.

In vitro cytotoxicity assay

Freshly sorted spleen NK cells were co-cultured with Yac-1 mouse lymphoma cells at 4:1, 2:1 and 1:1 ratios in IMDM media (supplemented with 20% fetal bovine serum, nonessential amino acids, L-glutamine, 2-mercaptoethanol, penicillin and streptomycin) for 4 hours. Yac-1 cells were traced by CFSE to distinguish from NK cells. Specific lysis was calculated by the formula: (a − b) / a × 100%; a is the number of Yac-1 cells without NK cell co-culture; b is the number of Yac-1 cells with NK cell co-culture.

Statistical analysis & Reproducibility

Statistical significance was performed with GraphPad Prism. Differences between groups were determined by a two-tailed unpaired or paired Student’s t test and were considered to be statistically significant when P values were less than 0.05. No statistical method was used to predetermine sample size. No data were excluded from the analyses. Animal/samples were randomly assigned to the various experimental groups. The Investigators were not blinded to allocation during experiments and outcome assessment. Data collection and analysis were not performed blind to the conditions of the experiments. No statistical methods were used to pre-determine sample sizes but our sample sizes are similar to those reported in previous publications^2,16. Data distribution was assumed to be normal but this was not formally tested.