miR-17∼92 family clusters control iNKT cell ontogenesis via modulation of TGF-β signaling

Transcriptional Signature of iNKT Cells Lacking miRNAs.

We first characterized the changes in gene expression occurring in iNKT cells developing in the absence of miRNAs induced by the lack of Dicer (15). The absence of miRNAs results in an anticorrelated increased expression of their direct mRNA targets (16, 17), as well as in an indirect modification of the expression of additional genes connected with them in regulatory networks. Thymic iNKT cells were sorted from Dicer KO and Dicer WT mice (Fig. 1A). Because of their paucity, iNKT cells from Dicer KO mice were collected in a single sample (iNKT KO). iNKT cells from Dicer WT mice were divided in two samples containing stages 1 and 2 together (iNKT 1–2), corresponding to the maturation attained by Dicer KO iNKT cells, and stage 3 alone (iNKT 3), respectively (Fig. 1A). Total mature thymic T cells were also collected. The samples were analyzed in triplicate and 8,357 genes were expressed in at least 1 of 15 biological samples (Dataset S1). Principal component analysis (18) performed on the genes expressed in the experiment (Fig. 1B) showed that iNKT KO cells were close to iNKT 1–2 WT cells, as expected from their similar maturation stages. iNKT 3 WT cells segregated from both iNKT KO and iNKT 1–2 WT cells and from T KO and T WT cells, which grouped together, consistent with the distinct gene-expression program displayed by developing iNKT cells. Linear models for microarray data (Limma) analysis with a 5% false discovery rate (FDR) threshold identified the genes differentially expressed among the samples analyzed (Dataset S1). Based on this threshold, T KO cells differed from T WT lymphocytes only for 28 genes, confirming the normal phenotype observed in thymic T cells when Dicer was ablated from the DP stage (7, 15, 19), and were no longer considered in the study. In iNKT cells, a pairwise comparison of the samples (Fig. 1C) showed that 255 genes were differentially expressed in iNKT KO vs. iNKT 1–2 cells, whereas 2,023 and 1,362 were differentially modulated in iNKT KO vs. iNKT 3 cells and iNKT 1–2 vs. iNKT 3 cells, respectively. This analysis allowed also identified the transcripts that were statistically up- or down-regulated as a result of: (i) the absence of miRNA expression (Fig. 1D), (ii) the progressive maturation attained by iNKT cells (Fig. 1E), and (iii) the likely combination of both mechanisms (Fig. 1F). The gene-expression profile associated with the progressive maturation from iNKT 1–2 to iNKT 3 cells was in good agreement with published datasets (20, 21) (Fig. 1E and Table S1). The genes that were up-regulated in iNKT KO cells, compared with WT iNKT cells, were candidates for being direct miRNA targets (Fig. 1 D and F). A Venn diagram in Fig. 1G depicts the numbers of private and shared differentially expressed genes identified in the indicated cell comparisons.

To systematically identify mRNAs potentially modulated by miRNAs during iNKT cell development, we focused on the 255 transcripts differentially expressed between iNKT KO and iNKT 1–-2 cells. Hierarchical clustering using these 255 transcripts divided the samples into the three expected iNKT KO, iNKT 3 WT, and iNKT 1–2 WT groups (Fig. 2; gene names are available in Fig. S1). Moreover, the transcripts clustered in four distinct groups. Specifically, cluster 1 included genes with higher expression in iNKT KO cells versus all iNKT WT cells, irrespective of their developmental stage (Fig. 2A, in black). Cluster 2 (Fig. 2A, in red) comprised genes with higher expression in both iNKT KO and iNKT WT 3, compared with iNKT WT 1–2 cells. Cluster 3 (Fig. 2B, in red) included genes down-regulated in both iNKT KO and iNKT WT 3 cells compared with the iNKT WT 1–2 samples. Finally, cluster 4 (Fig. 2B, in black) contained genes selectively down-regulated in iNKT KO cells compared with both iNKT WT 1–2 and 3 samples.

To shed light on the general molecular mechanisms underlying the defect observed in iNKT cells devoid of Dicer-dependent miRNAs, we performed a Gene Ontology (GO) term enrichment analysis on genes differentially expressed between iNKT KO and iNKT 1–2 WT cells. Clustering together the biological processes enriched in a statistically significant manner through a semantic similarities analysis (Fig. S2) emphasized four main groups of GO categories that can be summarized as: (i) ion homeostasis, (ii) signaling, (iii) development, and (iv) response to stress stimuli. These GO categories comprise genes related to the metabolic control and activation of the cells, as ion channels or controllers of the redox state, as also confirmed by the statistically significant enrichment of the relative molecular function GO term (GO:0004601 peroxidase activity, with FDR = 0.00855).

Finally, we matched the 72 miRNAs that we previously reported to be expressed by thymic iNKT cells (7) with their predicted targets within the 255 transcripts differentially expressed in iNKT KO compared with iNKT 1–2 cells. Fig. 2 shows that the majority of the predicted targets of these miRNAs were up-regulated in iNKT KO cells, consistent with a role for these miRNAs in modulating gene expression in iNKT cell development. Indeed, the 150 genes up-regulated in iNKT KO cells (clusters 1 and 2) (Fig. 2A) show a significantly greater density of predicted miRNA binding sites (Fisher test, P = 5.89e-3 and 2.69e-9, respectively) compared with the 105 transcripts down-regulated in iNKT KO cells (clusters 3 and 4) (Fig. 2B).

Collectively, these data defined the changes in gene expression and the resulting defective functional processes, occurring in developing iNKT cells as a result of the lack of miRNA-dependent posttranscriptional control.

Systemic Definition of miRNAs and Target Genes Relevant for iNKT Cell Development.

To systemically define the miRNAs that were putatively involved in controlling the development of iNKT cells, we first identified the miRNAs whose predicted targets had a significantly different fold-change compared with nontarget genes, in the comparison between iNKT KO and iNKT 1–2 cells. This unbiased approach identified miR-142-3p, miR-181abcd/4262, miR-30abcdef/30abe-5p/384-5p, miR-19ab, and miR-23abc/23b-3p (Table 1). Notably, the miR-181abcd/4262 family was already shown to play a critical role in iNKT cell development (12, 13), underscoring the likelihood of the selection approach. The expression of these miRNAs was assessed by quantitative RT-PCR (qRT-PCR) in the three thymic iNKT cell subsets purified from WT mice (Fig. S3). miR-142-3p was equally expressed at all three stages. miR-181a and -b and miR-30e and -c were expressed at a higher level at stage 1 and down-regulated at stages 2 and 3, as reported previously (12, 13). miR-23b, miR30-a-5p, and miR-30b were up-regulated at stage 3. Finally miR-19a and -b, two miRNAs of the miR-17∼92 family clusters, were selectively up-regulated at stage 2, the stage at which maturation of Dicer KO iNKT cells is blocked. To identify potentially relevant targets for these miRNAs, we matched them against the transcripts that were up-regulated in Dicer KO iNKT cells (Fig. 2A) and also contained in the enriched GO categories (Fig. S2). As shown in Table 2, we obtained a restricted list of genes. We focused on Tgfbr2, one of the two subunits of the heterodimeric receptor for TGF-β, because of the known role played by this cytokine in regulating iNKT cell development. TGF-β is required for protection against apoptosis, lineage expansion, and maturation of iNKT cells (22). TGF-β signaling sustains iNKT cell expansion at stage 1, whereas it acts as a “brake” for differentiating iNKT cells that must be selectively down-modulated at stage 2 to let their maturation proceed to stage 3 (22). Tgfbr1, the other subunit of the heterodimeric TGF-β receptor, was also up-regulated in iNKT KO cells (Fig. 2A). We thus posited that the modulation of TGF-β signaling might result from the miRNA-dependent down-regulation of Tgfbr2 expression at stage 2. In the absence of miRNAs, the heterodimeric TGF-β receptor expression would remain high in stage 2 iNKT cells, resulting in unconstrained TGF-β signaling that would contribute to their defective development. To test this hypothesis, we first confirmed the up-regulation of the TGF-βRII protein on the surface of primary thymic iNKT KO compared with WT iNKT 1–2 cells (Fig. 3A). Second, we verified that TGF-βRII up-regulation was also associated with an amplified intracellular signaling, as shown by the increased phosphorylation of the TGF-β–dependent transcription factors Smad2/3 (23) in primary iNKT KO cells (Fig. 3B).

Together, these systemic analyses highlighted miRNA/mRNA pairs that might play a role in regulating iNKT cell development and suggested that a deregulated miRNA-dependent fine-tuning of TGF-β signaling, via defective TGF-βRII expression at stage 2, would be one of the mechanisms contributing to the altered iNKT cell development in Dicer KO mice.

Identification of the miRNAs Regulating TGF-βRII Expression in Developing iNKT Cells.

To identify the miRNAs potentially involved in the posttranscriptional control of TGF-βRII, we determined in the three thymic iNKT cell stages the anticorrelated expression between Tgfbr2 mRNA (Fig. 4A) and the miRNAs expressed by iNKT cells that were predicted to target this transcript (Fig. 4 B–D). The majority of the predicted miRNAs belonged to the miR-17∼92 family, made of the three paralog miR-17∼92, miR-106a∼363, and miR-106b∼25 clusters (24) (Fig. 4E). These highly redundant clusters comprise 15 miRNAs that form four “seed” families: the miR-17, miR-18, miR-19, and miR-92 families. Two sequences in the 3′UTR of Tgfbr2 mRNA are experimentally validated targets for all eight miRNAs belonging to the miR-17 and miR-19 families (25–28). The independent miR-21 and miR-23b were also predicted to target Tgfbr2 mRNA. However, only the expression of the predicted miR-17, miR-19a, miR-19b, miR-20a, miR-20b, miR-93, miR-106a, and miR-106b peaked at stage 2 and perfectly anticorrelated with Tgfbr2 mRNA expression (Fig. 4 C and D). Notably, the miR-17 seed family (miR-17, miR-20a, miR-20b, miR-93, miR-106a, and miR-106b) on the one hand, and the miR-19 seed family (miR-19a and miR-19b) on the other hand, targeted the same sequences in the 3′UTR of Tgfbr2 mRNA, respectively (Fig. 4B), suggesting a high level of redundancy in the posttranscriptional regulation (29). Consistent with the coordinated expression of the three miR-17∼92 family clusters as polycistronic precursors undergoing posttranscriptional processing, the other members of these clusters (miR-18a, miR-18b, miR-92a, miR-25, miR-363) also displayed a comparable expression peak at stage 2 in thymic iNKT cells (Fig. 4C); however, they were not predicted to target Tgfbr2 transcript.

miRNAs of the miR-17∼92 Family Clusters Regulate TGF-β Signaling in iNKT Cell Development.

To validate the role of the predicted and anticorrelated miRNAs in the control of iNKT cell development through the regulation of the Tgfbr2 transcript, we analyzed mice in which the genes encoding all of the three paralog miR-17∼92, miR-106a∼363, and miR-106b∼25 clusters were deleted (29) to overcome the targeting redundancy of the these mRNA families (29). In triple KO (TrKO) mice, the 106a∼363 and miR-106b∼25 clusters were eliminated in the germ line, and an inducible knocked-in miR-17∼92 cluster was deleted at the DP thymocyte stage by crossing with CD4-Cre transgenic mice (29). Gross thymocytes and detailed iNKT cell development in TrKO, Dicer KO and WT animals were compared. Total thymocyte numbers were reduced by 30% in TrKO mice compared with WT or Dicer KO animals, because of a reduction in DP thymocytes resulting from increased apoptosis of heat stable antigen (HSA)^high cells. However, mature single positive (SP) thymocyte numbers were normal in TrKO mice (CD4⁺ 1.2 × 10⁷ ± 1 × 10⁶, CD8⁺ 4.4 × 10⁶ ± 5 × 10⁵), and comparable to both Dicer KO (CD4⁺ 1.2 × 10⁷ ± 1 × 10⁶, CD8⁺ 4.3 × 10⁶ ± 4.5 × 10⁵) and WT (CD4⁺ 1.4 × 10⁷ ± 9 × 10⁵, CD8⁺ 4.2 × 10⁶ ± 3 × 10⁵) mice, confirming that the triple miRNA gene deletion did not perturb mature T cells (30). In contrast, both percentage and number of thymic iNKT cells in TrKO mice were markedly reduced, similar to those of Dicer KO mice (Fig. 5 A and B). This result was because of a decreased proliferation of TrKO iNKT cells compared with the WT ones, whereas cell death was comparable (Fig. S4). Of note, the residual thymic TrKO iNKT cells displayed a significant reduction of mature stage 3 cells (Fig. 5 A and C), comparable to the defect found in the absence of Dicer. TGF-βRII expression level doubled in thymic TrKO iNKT cells compared with WT ones, approaching the level of Dicer KO iNKT cells (Fig. 5D) and in line with the expected extent of miRNA-mediated repression, which rarely exceeds twofold (31). Consistent with the high target redundancy of these miRNAs, we detected the most profound iNKT cell developmental defect (Fig. S5 A and B), with the highest TGF-βRII expression level (Fig. S5C) in the TrKO mice, compared with either miR-17∼92/miR-106b∼25 or miR-106a∼363/miR-106b∼25 double KO combinations. The higher TGF-βRII expression was also similarly associated with increased signaling, as assessed by increased pSMAD2/3 levels (Fig. 5E). This result occurred in the face of normal Tgfb1 mRNA expression levels, detected by qRT-PCR in the whole thymocyte population of TrKO and Dicer KO mice and comparable to that of WT thymocytes (Fig. S6). Culturing thymocytes with TGF-β1 evoked a greater increase in pSMAD2/3 level in TrKO (+37%) compared with WT iNKT cells (+18%), consistent with their increased TGF-βRII expression (Fig. 5F). These results supported a direct role for the higher TGF-βRII expression, resulting from the lack of miRNAs, in determining the increased pSMAD2/3 levels exhibited by TrKO iNKT cells.

We next wondered whether the expression of targets of TGF-β signaling was altered in TrKO iNKT cells. The expression of several genes that are either transcriptionally repressed (Tbx21-Tbet, Gata3, Myc) (32–34) or induced (Cdkn1a-p21, Cdkn2b-p15, Bcl2l11-Bim, Sgk1, Fos, Jun) (35–39) by TGF-β signaling was assessed in stages 1, 2, and 3 WT and TrKO iNKT cells. All transcripts displayed the expected up- or down-regulation at stage 2, resulting from an increased TGF-β signaling, compared with the WT pattern (Fig. S7), with the exception of Bcl2l11-Bim.

TGF-β signaling is also implicated in the differentiation of the iNKT cell subset displaying T_H17 effector functions (iNKT17) that, together with iNKT cells displaying T_H1 (iNKT1) and T_H2 (iNKT2) effector functions, are generated in the thymus. We thus investigated the effects of deleting miRNAs in development of the three iNKT cell subsets in Dicer KO and TrKO mice compared with WT animals. Dicer KO and TrKO mice displayed a significant increase of iNKT2 cells and decrease of iNKT1 and, particularly of iNKT17 cells, which were completely absent in thymic iNKT cells from the two KO mice (Fig. 5 G and H). The down-regulation of the iNKT17 RORγt protein in Dicer KO iNKT cells, compared with WT iNKT 1–2 cells, confirmed the microarray data (Fig. 2A). Moreover, this skewing toward the iNKT2 subset was detected also in the thymus, spleen, liver, and lymph nodes of older (8-wk-old) TrKO mice (Fig. S8), suggesting that this defect is not the result of a delayed iNKT cell development but rather of a diversion of their effector maturation.

Taken together, these results identified miRNAs belonging to the miR-17 and miR-19 seed families, which are part of the miR-17∼92 family clusters, as critical regulators of TGF-βRII expression in developing iNKT cells, and supported a role for these paralog miRNAs in directing iNKT cell effector differentiation via fine-tuning of TGF-β signaling at stage 2.

Genetic Complementation Experiments Support miRNA-Dependent Modulation of TGF-β Signaling in iNKT Cell Development.

Because the above results suggested that the deletion of the three miRNA families resulted in deregulated TGF-β signaling and altered iNKT cell development, we sought to undertake a genetic complementation approach to assess whether the deletion of TGF-βRII expression from miRNA-deficient iNKT cells would restore their correct maturation. In these mice, the lack of TGF-βRII expression in developing iNKT cells would counteract the increased TGF-β–dependent signaling resulting from the lack of miRNA-mediated regulation. Given the complexity of breeding TrKO and TGF-βRII KO mice to obtain animals with four homozygous knockout genes, and the fact that TGF-βRII and TGF-β–dependent signaling were up-regulated in iNKT cells from Dicer KO mice as well, we simplified the experiment by crossing Dicer^fl/fl (Dicer KO) and TGF-βRII^fl/fl (TGF-βRII KO) mice, in which both genes are deleted at the DP stage by CD4-Cre. Thymic iNKT cells were analyzed from 2-wk-old mice before TGF-βRII KO animals develop fulminant autoimmunity (40). In line with published data (22), TGF-βRII KO mice displayed reduced percentage and number of iNKT cells (Fig. 6 A and B) and an increased frequency in iNKT stage 3 cells (Fig. 6 C and D) compared with WT animals, consistent with the role of TGF-β signaling in sustaining iNKT cell stage 1 expansion and inhibiting stage 3 maturation. Comparable reductions in iNKT cell percentage and number were also found in Dicer KO mice; however, they were accompanied by the expected maturation block at stage 2. In Dicer–TGF-βRII double KO mice, frequency and numbers of iNKT cells remained low (Fig. 6 A and B). Remarkably, however, iNKT cell maturation to stage 3 was completely restored, with a distribution throughout the stages more similar to that of WT iNKT cells than to those displayed by single TGF-βRII or Dicer KO cells (Fig. 6 C and D).

Collectively, these findings further support the concept that the expression of TGF-βRII must be specifically tempered by miRNAs at stage 2 of iNKT cell development to deliver the appropriate level of TGF-β signaling for their correct maturation.

Mice.

C57BL/6(N) (Charles River), Dicer^fl/fl and CD4-Cre Dicer^fl/fl (KO) (7, 15), miR-106a∼363^−/− miR-106b∼25^−/− CD4-Cre miR-17∼92^fl/fl mice (29) were housed in a pathogen-free environment. Procedures involving animals were approved by the Institutional Animal Care and Use Committee at the San Raffaele Scientific Institute in Italy (IACUC no. 678).

Cell Staining and Flow Cytometry.

Total thymocytes were stained with anti-CD24/HSA, CD4, CD44, NK1.1, TCR-β, CD69 (Biolegend); anti–TGF-βRII (R&D Systems); anti–p-SMAD2/3 (sc-11769) (Santa Cruz), followed by swine anti-goat IgG (Southern Biotech); anti-RORγt and anti-PLZF Abs (e-Bioscience), all in the presence of rat anti-mouse CD16/CD32 Fc Blocker 2.4G2 mAb. iNKT cells were identified by PBS-57 loaded mCD1d-PE tetramers (NIH Tetramer Facility). BrdU staining was performed using BrdU Flow Kits (BD Biosciences). Apoptosis was detected by fluorescent CaspACE FITC-VAD-FMK (Promega). Where indicated, cells were stimulated for 30 min at 37 °C with αCD3/CD28 beads (Dynabeads Mouse T-Activator CD3/CD28, Gibco) at a 1 cell:3 beads ratio ± 3 ng/mL human recombinant TGF-β1 (R&D Systems). Dead cells and doublets were excluded with DAPI staining and physical gating. Intracellular staining was performed using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience). Samples were acquired on FACSCantoII flow cytometer (BD Biosciences) and data analyzed by FlowJo software (TreeStar).

Isolation and Purification of Cells by Sorting.

Cells were labeled with mAbs and CD1d-tetramers and subsequently sorted using a MoFlo cell sorter (BeckmanCoulter). Cell debris and doublets were excluded. The purity of sorted fractions was checked by flow cytometry reanalysis.

RNA Extraction and Amplification.

Total RNA from sorted cells was isolated with the miRvana kit (Ambion) and quantified by NanoDrop spectrophotometer (Thermo Scientific). RNA quality and integrity were verified by an Agilent 2100 Bioanalyzer profile using Agilent RNA 6000 Pico or Nano kits.

Quantitative Real-Time RT-PCR.

miRNA-specific reverse transcription was performed using TaqMan MicroRNA Reverse Transcription kit (Applied Biosystems). qRT-PCR was performed using TaqMan MicroRNA Assay Mix containing PCR primers and TaqMan probes (Applied Biosystems). Expression values were normalized to snoRNA-202. For mRNA expression, RNA was amplified with a MessageAmp RNA Amplification Kit (Ambion) and reverse transcription was performed with RNA and a High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). qRT-PCR was performed in triplicate with 10–100 ng cDNA/reaction using Taqman technology (Applied Biosystems) on an ABI Prism 7900 cycler (95 °C for 15 s and 60 °C for 1 min, 40 cycles) and analyzed with SDS 2.2.1 software. HPRT was used as a housekeeping gene. The relative quantification of gene expression was determined by comparative Ct method.

Gene-Expression Profiling.

Mature HSA^low-enriched thymic iNKT cells and T cells were sorted from 4-wk-old Dicer KO mice and Dicer WT mice. iNKT cells from Dicer WT were subdivided into stage 1–2 cells (HSA^lowTCRβ⁺CD1dtet⁺NK1.1⁻) or stage 3 cells (HSA^lowTCR-β⁺CD1dtet⁺ NK1.1⁺CD44⁺), whereas Dicer KO iNKT cells were sorted in toto (HSA^lowTCR-β^highCD1dtet⁺). SP T cells, both from Dicer KO and Dicer WT mice, were sorted as pooled HSA^lowTCR-β^high CD4 and CD8 SP cells. All of the five samples were sorted in triplicate, from a pool of 20 mice per replicate. Total RNA was extracted from the 15 samples, retrotranscribed, amplified, and biotinylated with the Illumina Totalprep RNA Amplification kit (Ambion). Amplified cRNA integrity was checked by an Agilent 2100 Bioanalyzer profile with Agilent RNA 6000 Nano kit. Washing, staining, and hybridization were performed according to the standard Illumina protocol with 750 ng of cRNA of each sample hybridized onto Illumina Sentrix beadchip arrays Mouseref-8v2. Hybridization and scanning were performed on an Illumina iScan System and data were processed with GenomeStudio software (Illumina).

Statistical Analysis.

All statistical tests were conducted with GraphPad Prism. Comparisons between two groups were done with the two-tailed Student t test for unpaired samples, applying correction for unequal variances when required. Comparisons between three or more groups were done with the one-way ANOVA test plus the Tukey posttest. For all tests, a value of P < 0.05 was considered significant.

Bioinformatic Analyses.

Arrays were quantile-normalized, with no background subtraction, and expression profiles were obtained for 18,138 genes. Of these, 8,357 genes whose intensity value was significantly different from the background (detection P < 0.01) in at least one sample of the entire series were considered “expressed” and included in the analysis. Principal component analysis was performed on expressed genes using the “mixOmics” (44) R library (45). The “Limma” Bioconductor package (46) was used to extract differentially expressed genes with an FDR of 5%. A biological term enrichment analysis using the GO biological process or molecular function database was performed using the DAVID tool (47, 48), comparing the prevalence of gene annotations among differentially expressed genes in specific contrasts to their prevalence in a background defined by all genes expressed in at least one iNKT sample. To focus on GO categories specifically related to differential expression, rather than to the cellular context, we performed the same analysis comparing the list of expressed genes to a background defined by all genes represented on the array and discarded the annotation terms that were enriched in this second comparison. The GOSemSim package (49) was used to calculate semantic distances among GO terms and the results used for hierarchical clustering: each of the clusters thus obtained represents a set of closely related, enriched GO annotations. The data are deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus (50) and accessible through GEO Series accession no. GSE79901.

miRNA target predictions were downloaded from TargetScan v6.1 (51). For each sample comparison, and each miRNA family included in Targetscan, we asked whether the predicted targets of each miRNA family expressed by thymic iNKT cells and included in Targetscan had significantly different fold-change compared with all other genes. The query was limited to genes that were predicted targets of at least one miRNA. The comparison between fold-changes was performed with a Mann–Whitney u test. No explicit selection of differentially expressed genes was used in this analysis (all fold-change values are included in the analysis).