PLZF expression maps the early stages of ILC1 lineage development

Early NK1.1⁻ Stages of Development.

We used PLZF-IRES-GFPCre mice to report and fate-map PLZF expression in lymphoid compartments. We previously established that GFP faithfully reported PLZF expression in the hematopoietic compartment, with high levels of expression in ILCP and in NKT cells, but not in other lymphoid subsets (15). However, in fate-mapping experiments based on the ROSA26-fl-STOP-fl-YFP allele, we reported that, whereas a majority of mature ILCs and NKT cells were YFP⁺, as expected, ∼30% of other hematopoietic cells were also YFP⁺, including T, B, myeloid cells, and early hematopoietic cell precursors such as early HSC, LMPP, and CLP cells. This indiscriminate background mapping occurred before hematopoiesis, likely in early embryonic stem cells, because it was readily abrogated in radiation chimeras reconstituted with sorted YFP-negative Lin⁻ Sca1⁺ cKit⁺ (LSK) precursors extracted from PLZF-IRES-GFPCre⁺ ROSA26-fl-STOP-fl-YFP mice (15). In the experiments reported below, we used these chimeras, termed YFP-negative chimeras, to faithfully fate-map PLZF expression in the hematopoietic system.

To precisely track the expression of PLZF side by side with fate-mapping at different stages of lymphoid development, we compared GFP expression in straight PLZF-IRES-GFPCre⁺ mice with YFP expression in YFP-negative chimeras (Fig. 1). We zoomed in on the pre-NKP and rNKP stages, identified by their Lin⁻ IL7Rα⁺ CD27⁺ CD244⁺ CD117 (cKit)^int/low Flt3⁻ profile, with the pre-NKP expressing weak or absent CD122 (IL2Rβ), and the rNKP expressing high CD122 (19) as gated in Fig. 1A. Both pre-NKPs and rNKPs were previously shown to selectively reconstitute CD3ε⁻ NKp46⁺ NK1.1⁺ Ly49D⁺ Ly49H⁺ cNK cells in the spleen and liver of Rag^−/−Il2rg^−/− mice after in vivo transfer, but a detailed analysis of markers that can identify ILC1s, such as CD49a, DX5, TRAIL, CD127 (IL7Rα), or eomesodermin, was not reported (19). Whereas CLPs failed to express either GFP or YFP, as expected, we found that nearly half of pre-NKPs expressed high amounts of GFP (Fig. 1A), a property previously shown to be restricted to the ILCP (15). Furthermore, at the subsequent rNKP stage, the expression of GFP was down-modulated, both in frequency and in intensity, whereas YFP fate-mapping increased from 15 to 36% on average (Fig. 1A). These findings suggested that cells with the “pre-NKP” phenotype in fact included a large fraction of PLZF-expressing ILCPs, which subsequently down-regulated PLZF and began to exhibit fate-mapping as they progressed into what was previously described as the “rNKP” stage.

We therefore extended the comparison between the GFP^high fraction of pre-NKPs and the previously defined ILCPs. Gating on GFP^high cells in the entire bone marrow compartment, we found that these cells not only uniformly expressed the Lin⁻ α4β7^high CD127 (IL7Rα)^high CD117 (c-Kit)⁺ profile, as previously described for ILCPs, but also were negative for CD135 (Flt3) and positive for CD27 and CD244 (Fig. 1B), the key markers of pre-NKPs (19). In contrast, the GFP-negative fraction of pre-NKPs lacked α4β7 expression (Fig. 1C). Because the transition between pre-NKPs and rNKPs is marked by the up-regulation of CD122, we reexamined CD122 expression with a bright antibody-biotin and streptavidin-phycoerythrin combination to detect low quantities of this protein on the cell surface. Indeed, we found that ILCPs expressed variable amounts of surface CD122 that ranged from negative to intermediate and, thus, might include cells already committing to the ILC1 sublineage (Fig. 1B, Far Right).

Thus, we conclude that the previously described pre-NKP population is composed of a mixture of ILCPs (precursors to ILC1s, ILC2s, and ILC3s) and cNK precursors. These two fractions can be distinguished by their differential pattern of expression of PLZF and α4β7. The distinct progenies of these two fractions can be inferred from transfer experiments showing that the unfractionated pre-NKPs generated classical NK cells in the spleen (19), whereas the PLZF⁺ α4β7⁺ (ILCP) fraction did not (15). In retrospect, the failure to observe significant NK1.1⁻ progeny (i.e., ILC2 and ILC3) after transfer of pre-NKPs into Rag^−/−Il2rg^−/− hosts in the study by Fathman et al. (19), is consistent with their focus on spleen and liver, whereas ILC2s and ILC3s predominate in the intestinal lamina propria.

We also examined the Lin⁻ Sca1⁺ CD117 (cKit)^int/− CD135 (Flt3)⁻ IL7Rα⁺ cell, which was proposed to be an early NK precursor, termed pre–pro-NK (20). Strikingly, we found uniform but low expression of GFP contrasting with a high frequency of YFP fate-mapping, suggesting that a large fraction of these cells had, in fact, originated from ILCPs and down-regulated PLZF (Fig. S1). Because the majority of these cells also expressed CD25, ICOS, and T1/ST2, the combination of which is highly characteristic of ILC2s, we concluded that they mostly represented late developmental intermediates of ILC2s termed immature ILC2s (iILC2 or LSIG), rather than cNK precursors, consistent with their reported Id2^high profile (21). Indeed, cell transfers into Rag^−/−Il2rg^−/− hosts have demonstrated that this precursor population exclusively generates mature ILC2 in vivo (21).

Immature NKs.

We next examined the so-called immature NK (iNK) cells, which are defined by expression of NKp46 and NK1.1, but not DX5. Although these cells were previously considered to be developmental intermediates of the cNK lineage before acquisition of DX5 (2), recent reports have emphasized that a fraction of iNKs expressed a phenotype similar to that of mature ILC1s, including expression of IL7Rα and CD49a, but neither CD49b (DX5) nor eomesodermin (4, 10, 22). In one study, transfer of eomesodermin-negative iNKs into Rag^−/−Il2rg^−/− hosts reconstituted liver ILC1s but not cNKs (10). We found that nearly 80% of iNKs, defined by their CD3ε⁻ CD122⁺ NKp46⁺ NK1.1⁺ DX5⁻ profile, were YFP⁺, but these cells did not express GFP, suggesting that a large proportion had originated from ILC1 rather than from cNK precursors (Fig. 2A). We further showed that the PLZF fate-mapped iNKs (isolated from YFP-negative chimeras) gave rise exclusively to CD49a⁺ DX5⁻ (ILC1s) but not to CD49a⁻ DX5⁺ (cNKs) liver cells after transfer into Rag^−/−Il2rg^−/− recipients (Fig. 2B). In contrast, CD45-congenic CLPs (coinjected with the YFP⁺ iNKs) generated both cNKs and ILC1s in the liver (Fig. 2B). We also purified bone marrow CD3ε⁻ NKp46⁺ NK1.1⁺ DX5⁻ iNKs, irrespective of their PLZF fate-mapping, and transferred them into Rag^−/−Il2rg^−/− hosts, to test whether they contained cNK precursors. Although these cells generated a majority of ILC1s, consistent with the predominance of ILC1 precursors, we also observed a substantial fraction of cNKs (Fig. 2B). Thus, CD3ε⁻ NKp46⁺ NK1.1⁺ DX5⁻ bone marrow cells, previously termed iNKs, are a mixture of both ILC1 and cNK lineage cells, with a strong predominance of ILC1s. Upon stimulation with ionomycin and PMA, the bone marrow CD3ε⁻ NKp46⁺ NK1.1⁺ DX5⁻ cells produced significantly less IFN-γ than their liver counterpart, further supporting the notion that they represented immature ILC1s (iILC1s) (Fig. 2C).

Altogether, our analysis of the developmental stages of ILC1s and cNKs suggests a new model whereby these lineages originate from distinct precursors, distinguished by the expression of PLZF and α4β7, before undergoing a parallel sequence of progressive steps encompassing previously described populations and diverging between iNKs and iILC1s as proposed in Fig. S2.

Predominance of ILC1s Early in Ontogeny.

Tissue-resident ILC1s are reminiscent of the early waves of tissue-resident γδ T cells that populate tissues such as skin, uterus, intestine, liver, and lung in the newborn (23). These tissue-resident γδ T cells express semiinvariant TCRs and are thought to represent a first line of defense in the perinatal period, before the generation of their blood recirculating counterparts, which have a more diverse repertoire. Indeed, we found that a great majority of CD3ε⁻ NK1.1⁺ cells in both embryonic day (E)16 fetal liver and in 1-wk-old liver were YFP⁺ fate-mapped, expressed CD49a and TRAIL, and lacked eomesodermin, whereas CD49a⁻ TRAIL⁻ eomesodermin⁺ cNK cells appeared later in ontogeny between weeks 1 and 3, and were characterized by lower YFP fate-mapping (Fig. 3). Note that to study fetal and newborn hematopoiesis, these experiments used the straight PLZF-IRES-GFPCre crossed to ROSA-fl-STOP-fl-YFP mice, rather than the YFP-negative adult bone marrow chimeras. This experimental necessity explains in part the higher frequency of YFP labeling in cNKs, due to background labeling before the hematopoietic stem cell stage. Surprisingly, liver ILC1s tended to coexpress DX5 in fetal life, but not after birth, emphasizing that DX5 expression may not always be a reliable marker of cNKs. Altogether, these results therefore demonstrate that ILC1s strongly predominate over cNK cells in perinatal life, whereas cNK cells achieve greater frequency in adults.

Impact of PLZF on ILC1 Development and Maturation.

We previously showed that PLZF-deficient ILC1s were decreased fourfold compared with their wild-type counterparts in the liver of competitive bone marrow chimeras (15). By microarray analysis, we found that ILC1s isolated from the liver of PLZF-deficient mice exhibited a limited set of changes compared with those of wild-type littermates, because only 229 genes had altered expression by more than 1.4-fold (Fig. 4A). In contrast, and similar to a recent study (6), >2,000 genes exhibited differential expression between liver cNK cells and ILC1s, indicating that, as expected, the differences between cNKs and ILC1s do not solely depend on PLZF. Strikingly, however, 59% of the PLZF-dependent genes in ILC1s were also differentially expressed between ILC1s and cNKs, a highly significant overlap (P = 2.7 × 10⁻⁹³ by hypergeometric test). Because ILC1s and cNKs differ by their history of PLZF expression, these differentially expressed genes may be involved in the divergence between these two lineages. For example, PLZF-deficient ILC1s showed decreased expression of CD127 (IL7Rα), both at the mRNA level by microarray analysis and at the protein level by flow cytometry (Fig. 4 B and C). IL7Rα is expressed in the hematopoietic precursors to both ILC1s and cNKs, but its expression is characteristically maintained only in mature ILC1s (22). Although the role of IL7Rα in ILC1 biology is not fully understood, because mature ILC1s seem mostly dependent on IL15 for their maintenance (4), it may be important for optimal survival during a developmental window, before expression of the IL15 receptor. Thus, the conspicuous defect in IL7Rα expression in PLZF-deficient ILC1s might contribute in part to their decreased frequency. Other ILC1 genes that depended on PLZF and were not expressed by cNKs included Mmp9, Klrb1b, Klrk1, and Tnfrsf25, as well as Cd3g and Cd3e, which were previously reported to be expressed by fetal but not adult NK cells (24). This expression of Cd3 genes may reflect the predominance of ILC1s over cNKs in the fetus. Most of the genes that were differentially expressed between wild-type ILC1s and PLZF-deficient ILC1s as well as between wild-type ILC1s and cNKs cells showed concordant changes in the two comparisons (Fig. 4B, upper right and lower left quadrants), although there were examples of opposite variations (Fig. 4B, upper left and lower right quadrants), for example for Il18r1. Importantly, several of the key genes that were differentially expressed between cNKs and ILC1s, including Eomes, Itga1 encoding CD49a, Tnfsf10 encoding TRAIL, Sell encoding CD62L, Edg8 encoding S1P5, and Cxcr6, did not seem to be influenced by PLZF.

Mice.

C57BL/6J (stock no. 000664), B6.SJL-Ptprc^a Pepc^b/BoyJ (CD45.1; stock no. 002014), and B6.129 × 1-Gt(ROSA)26Sor^tm1(EYFP)Cos/J (stock no. 006148) mice were obtained from The Jackson Laboratory, whereas B6.Rag2^tm1Fwa II2rg^tm1Wjl (Rag^−/−Il2rg^−/−, stock no. 4111) mice were obtained from Taconic. The PLZF^GFPcre strain has been described (15). PLZF^−/− mice were a gift from P. P. Pandolfi, Beth Israel Deaconess Medical Center, Boston, and were backcrossed to C57BL/6J for at least nine generations. Unless noted otherwise, animals were 4–10 wk of age when analyzed and were compared with littermate controls obtained from crosses between heterozygous breeders. For fetal experiments, the morning a vaginal plug was observed was considered E0. Mice were housed in a specific pathogen-free environment at the University of Chicago, and experiments were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee.

Preparation of Cell Suspensions.

Spleen and liver were mechanically dissociated through 70-μm filters, and bone marrow was isolated by gently crushing femurs and tibias before filtration. Following dissociation, each liver was centrifuged at 400 × g for 5 min, resuspended in 5 mL of 40% (vol/vol) Percoll (Sigma-Aldrich), and then centrifuged at 800 × g for 10 min. The supernatant was aspirated and the cell pellet was resuspended in HBSS (Gibco) containing 0.25% BSA (Sigma-Aldrich) and 0.65 mg/L sodium azide (Sigma-Aldrich).

Flow Cytometry.

Cell suspensions were incubated with purified anti-CD16/32 (clone 93) for 10 min on ice to block Fc receptors. Fluorochrome- or biotin-labeled monoclonal antibodies (clones denoted in parentheses) against α4β7 (DATK32), B220 (RA3-6B2), CD3ε (145-2C11), CD4 (RM4-5 or GK1.5), CD8α (53-6.7), CD11b (M1/70), CD11c (N418), CD19 (6D5), CD25 (PC61), CD27 (LG.7F9), CD45.1 (A20), CD45.2 (104), CD49a (HMα1), CD122 (5H4 or TM-β1), CD160 (7H1), CD244 (2B4), cKit (2B8), DX5 (DX5), Eomes (Dan11mag) Flt3 (A2F10), Gr-1 (RB6-8C5), ICOS (C398.4A), IFNγ (XMG1.2), IL-7Rα (A7R34), Ly-6D (49-H4), NK1.1 (PK136), NKp46 (29A1.4), Sca-1 (D7), T1/ST2 (D1H9), TCRβ (H57-597), Ter-119 (TER-119), Thy1.2 (53-2.1) and TRAIL (N2B2) were purchased from BD Biosciences, BioLegend, eBioscience, or R&D Systems. To exclude dead cells, 4′,6-diamidino-2-phenylindole (DAPI; Molecular Probes) was added to all live samples. For intracellular staining, cells were fixed with 4% paraformaldehyde and permeabilized by using the Foxp3 Transcription Factor Staining Buffer Set (eBioscience). Cells were run on an LSRII (BD Biosciences) or sorted by using a FACS Aria II (BD Biosciences) and analyzed by using FlowJo software (Tree Star). Collected events were gated on DAPI⁻ lymphocytes and doublets were excluded.

Unless specified otherwise, CLPs were gated Lin⁻ Ly6D⁻ CD244⁺ CD27⁺ IL-7Rα⁺ Flt3⁺ CD122⁻ (19), pre-NKP were gated Lin⁻ Ly6D⁻ CD244⁺ CD27⁺ IL-7Rα⁺ Flt3⁻ CD122^low/neg (19), rNKP were gated Lin⁻ Ly6D⁻ CD244⁺ CD27⁺ IL-7Rα⁺ Flt3⁻ CD122^high (19), pre–pro-NK were gated Lin⁻ Sca-1⁺ cKit^low IL-7Rα⁺ Flt3⁻ (20), bone marrow DX5⁻ cells were gated Lin⁻ CD122⁺ NK1.1⁺ NKp46⁺ DX5⁻ (14), bone marrow DX5⁺ cells were gated Lin⁻ CD122⁺ NK1.1⁺ NKp46⁺ DX5⁺, liver DX5⁺ cells were gated CD3ε⁻ TCRβ⁻ NK1.1⁺ DX5⁺ CD49a⁻, and liver DX5⁻ cells were gated CD3ε⁻ TCRβ⁻ NK1.1⁺ DX5⁻ CD49a⁺. Lineage mixtures for pre-NKPs, rNKPs, and CLPs included antibodies against CD3ε, CD11b, CD19, and NK1.1; for pre–pro-NK: CD3ε, CD8α, CD11b, CD19, Gr-1, and Ter119; for bone marrow NK1.1⁺ subsets: CD3ε, CD4, CD8α, CD19, and Ter119.

Bone Marrow Chimeras.

To generate YFP⁻ chimeras, 5–10 × 10³ LSK cells were sorted from the bone marrow of PLZF^GFPcre+/− ROSA-YFP mice and injected retroorbitally into lethally irradiated (1,000 rads) CD45.1 recipients. Chimeras were analyzed 5–7 wk after reconstitution, gating on CD45.2⁺ cells to exclude residual host cells.

Isolation and Adoptive Transfer of Bone Marrow CLPs and NK1.1⁺ Subsets.

For the isolation of CLPs from CD45.1 bone marrow, lineage⁺ cells were first depleted by using an autoMACS (Miltenyi Biotec) after staining with biotin-conjugated antibodies against B220, CD3ε, CD4, CD8α, CD11b, CD11c, CD19, Gr-1, NK1.1, TCRβ, and Ter119, followed by incubation with SAv microbeads (Miltenyi Biotec). CLPs were then sorted as DAPI⁻ Lin⁻ IL-7Rα⁺ cKit^int Sca-1^int Flt3^high to greater than 95% purity by using a FACS Aria II (BD Biosciences). For the isolation of YFP⁺ NK1.1⁺ DX5⁻ cells, bone marrow from YFP⁻ chimeras (described above) was stained with APC-conjugated anti-NK1.1 antibody, bound to anti-APC microbeads (Miltenyi Biotec), subjected to double-column enrichment, and sorted as DAPI⁻ Lin⁻ CD122⁺ NK1.1⁺ NKp46⁺ DX5⁻ YFP⁺ to greater than 95% purity. For the isolation of NK1.1⁺ DX5⁻ cells, CD45.1 bone marrow was stained with APC-conjugated anti-NK1.1 antibody, bound to anti-APC microbeads, subjected to double-column enrichment, and sorted as DAPI⁻ Lin⁻ CD122⁺ NK1.1⁺ NKp46⁺ DX5⁻ to greater than 95% purity. For the NK1.1⁺ subsets, the lineage mixture antibodies used were as follows: CD3ε, CD4, CD8α, CD19, and Ter119.

Following isolation, 1,500 YFP⁺ NK1.1⁺ DX5⁻ cells mixed with 750 CD45.1-congenic CLPs, or 1,500 CD45.1 NK1.1⁺ DX5⁻ cells alone were injected retroorbitally into sublethally irradiated (400 rads) 6- to 10-wk-old CD45.2 Rag2^−/− Il2rg^−/− mice. Recipient mice were analyzed 3–4 wk after transfer.

Isolation and in Vitro Stimulation of NK1.1⁺ DX5⁻ Cells.

Following enrichment using FITC-NK1.1 and anti-FITC microbeads (Miltenyi Biotec), NK1.1⁺ DX5⁻ cells were sorted as DAPI⁻ Lin⁻ NK1.1⁺ NKp46⁺ DX5⁻ from the bone marrow or DAPI⁻ CD3ε⁻ TCRβ⁻ NK1.1⁺ NKp46⁺ DX5⁻ from the liver of WT mice (>95% purity). For the bone marrow cells, the lineage mixture antibodies used were the following: CD3ε, CD4, CD8α, CD19, TCRβ, and Ter119.

Up to 10,000 sorted cells were resuspended in 500 μL of RPMI medium 1640 (Cellgro) containing 10% FCS (Biowest) and 1× GolgiStop (BD). Samples were stimulated with phorbol 12-myristate 13-acetate (PMA; 20 ng/mL) and ionomycin (5 μg/mL) for 4 h at 37 °C. Following the stimulation, samples were fixed and permeabilized by using the Cytofix/Cytoperm kit (BD).

Microarray Analysis.

Liver lymphocytes from pools of PLZF^+/+ or PLZF^−/− mice were sorted into ILC1s (CD3ε⁻NK1.1⁺CD49a⁺DX5⁻) and cNKs (CD3ε⁻NK1.1⁺CD49a⁻DX5⁺) and stored in TRIzol (Life Technologies). RNA was isolated by using the RNeasy mini kit (Qiagen), processed, and annealed to the mouse WG-6 array (Illumina). Illumina MouseWG-6 v2.0 Expression BeadChip data were quality-checked by using the lumi package (27) and quality control metrics in the R/Bioconductor statistical software package (v3.0.0). Data were background corrected and normalized by variance-stabilizing transformation (28) followed by quantile normalization. Data were log₂ transformed and fit to linear models by using the limma package (29) for fold change comparison between groups. A fold-change cutoff of 1.4 was used to distinguish differentially expressed genes. Microarray data has been deposited with the National Center for Biotechnology Information Gene Expression Omnibus repository under the accession number GSE65898.

Statistical Analysis.

Two-tailed Student’s t test was performed by using Prism (GraphPad Software) to determine whether data differed from the expected value. *P < 0.05.