Modulation of TCR signalling components occurs prior to positive selection and lineage commitment in iNKT cells

iNKT cells play a critical role in controlling the strength and character of adaptive and innate immune responses. Their unique functional characteristics are induced by a transcriptional program initiated by positive selection mediated by CD1d expressed by CD4+CD8+ (double positive, DP) thymocytes. Here, using a novel Vα14 TCR transgenic strain bearing greatly expanded numbers of CD24hiCD44loNKT cells, we examined transcriptional events in four immature thymic iNKT cell subsets. A transcriptional regulatory network approach identified transcriptional changes in proximal components of the TCR signalling cascade in DP NKT cells. Subsequently, positive and negative selection, and lineage commitment, occurred at the transition from DP NKT to CD4 NKT. Thus, this study introduces previously unrecognised steps in early NKT cell development, and separates the events associated with modulation of the T cell signalling cascade prior to changes associated with positive selection and lineage commitment.

www.nature.com/scientificreports/ purified by agarose gel electrophoresis, agarase treated, re-purified and injected directly into NOD/Lt embryonic pronuclei at the Walter and Eliza Hall Institute microinjection unit (Melbourne, Australia). The manipulated embryos were placed in the reproductive tracts of pseudopregnant NOD/Lt recipient female mice. Transgenic mice were screened with polymerase chain reaction (PCR) (forward primer: 5′-TGT AGG CTC AGA TTC CCA ACC-3′; reverse primer: 5′-GAG GAT GGA GCT TGG GAG TCAGG-3′) and crossed onto the NOD.Nkrp1b b line to permit the use of the NK1.1 developmental marker. These studies have been reviewed and approved by the James Cook University Animal Ethics Committee. The study was carried out in compliance with the ARRIVE guidelines (https:// arriv eguid elines. org) and all methods were performed in accordance with the relevant guidelines and regulations.
Livers were harvested after being perfused with 5-10 ml PBS then minced through a 180 um wire stainless steel mesh. The dissociated material was centrifuged at 500 X g for 5 min at 4 °C. Parenchymal cells were removed over a 33.75% Percoll (GE Healthcare, NSW, Australia) gradient. Liver lymphocytes were collected after lysing red blood cells.
For surface staining, antibodies were diluted in FACS buffer. As described in Jordan et al. 24 , cells were preincubated with unconjugated anti-CD16/32 (clone 93, eBiosciences, San Diego, CA, USA) before addition of surface staining antibody cocktails. Data were acquired on a BD LSRII Fortessa (BD Bioscience, San Jose, California, USA) flow cytometer and analysed using FlowJo software (Tree Star, Inc., Ashland, OR, USA).
In vivo treatment with α-GalCer. Alpha-GalCer was prepared for injection by sonication in PBS with 0.5% Tween-20 for 2 h at 37 °C. Six-week old mice were injected either intravenously (4 μg in 200 μl) or intrathymically (2 μg in 10 μl) with α-GalCer or control vehicle. Mice were either bled four hours later for cytokine analyses or culled 40 h after injection for flow cytometric analyses.
Expression microarray hybridizations were performed using the WT Expression kit (Life Technologies, CA, USA), WT Terminal Labelling and Controls Kit (Affymetrix, CA, USA) and Affymetrix Mouse Gene_1.0ST arrays, which contain 770,317 probe sets representing an estimated 35,556 mouse transcripts. The probed arrays were washed and stained using the GeneChip Hybridization Wash and Stain Kit (Affymetrix, CA, USA) and scanned using the GeneChip Scanner 3000. Images (.dat files) were processed using GeneChip Command Console (Affymetrix, CA, USA) and CEL files imported into Partek Genomics Suite 6.6 (Partek SG, Singapore) for further analysis.
Gene co-expression network. A gene co-expression network was generated using the Affymetrix Mouse Gene_1.0ST array analyses of thymocyte subsets -a total of 28 microarrays. The Affymetrix CEL files were normalised using RMA background subtraction in Bioconductor and batch effects were removed using the nonparametric CombatR algorithm 46 . Variability of transcripts across all arrays was ranked by standard deviation 47 and the 1,929 most variable were used for network construction. Application of the WGCNA algorithm in R 48 generated a weighted gene co-expression network of 1,929 nodes (transcripts) and 10,626 edges (representing significant correlations at a p < 0.02) assigned to 12 significantly co-expressed modules.
Gene ontology analysis. Gene lists were generally split into those upregulated and those downregulated before being submitted to The Database for Annotation, Visualization and Integrated Discovery (DAVID) Bioinformatics Resources v6.7 49   Statistical analyses. Qualitative data were compared by Fisher's Exact Test or contingency table (Chi squared) analysis and quantitative data by Mann Whitney U Test as calculated by GraphPad Prism 6 (La Jolla, CA, USA). The statistical significance threshold of the microarray study comparing FACS sorted thymic cells from individual NOD.Va14 tg mice was set at a Mann-Whitney U statistic of 0 (i.e. p < 0.001, n = 7; equating to no overlap between groups). We have previously published empiric validations of microarray expression analyses of similar design by using congenic intervals to differentiate "on target" from "off target" differential expression 23

Results
Production and characterisation of NOD.Va14 tg mice. Vα14-Jα18 transgenic mice have been previously developed to aid functional and developmental studies of Type I NKT cells 22,44 . Here, we expressed a validated transgenic construct containing an NKT-associated Vα14-Jα18 TCRα chain cDNA (Fig. 1A) on the NOD mouse genetic background, which confers a partial defect in iNKT cell selection and development 5,22,35 . Purified DNA was directly injected into NOD/Lt embryonic pronuclei at the Transgenic Production Facility of the Walter and Eliza Hall Institute, tail tips were genotyped by a PCR spanning from CD4 to TCR sequence to identify transgene incorporation and germ-line transmission (Fig. 1B), and four independent lines were established. Thymic, hepatic and splenic iNKT cell numbers and subsets were assessed by flow cytometric analysis of CD1dtet + TCRβ + lymphocytes. Transgenic lines 1 and 3 had ~ 60-90-fold increases in thymic iNKT cell numbers while lines 2 and 4 had ~ threefold increases; numbers of hepatic iNKT cells were raised ~ fourfold in lines 1 and 3 and by 25-50% in lines 2 and 4; lines 1 and 3 resulted in ~ sevenfold increases in splenic iNKT cell numbers, while lines 2 and 4 had similar numbers to the non-transgenic parental strain (Fig. 1C).
iNKT cell developmental subsets in NOD.Va14 tg mice. Both the CD4 SP and DN populations of mature iNKT cells could be identified in the thymi (Fig. 2), livers and spleens (Fig. 3) of WT and NOD.Va14 tg mice.
One of the most striking characteristics of the thymic iNKT cell population in NOD.Va14 tg mice was the presence of increased numbers of DP NKT cells. Their presence in the thymi of previously produced Vα14Tg mice, on either B6 44 or NOD backgrounds 41 , had not been detected. This may be a consequence of the transgenic system used, a TCRα shuttle vector containing the Vα11 endogenous promoter and the Ig enhancer, or it may be possible that the relatively immature NKT cells were inadvertently missed from their analyses due to the detection markers, α/βTCR + and NK1.1 + , employed, or alternatively, that they were indeed not increased. Here, the very immature thymic DP NKT cells expressing a CD24 high CD44 low and NK1.1 − phenotype, were identified using TCRβ and CD1d-tetramer, along with stage-specific markers. This therefore, represents the first mouse model to produce large numbers of immature DP NKT cells. This population constituted between 35 and 65% of all thymic CD1d-tet + TCRβ + NKT cells in the various lines of NOD.Va14 tg mice, but < 5% of thymic iNKT cells in WT NOD mice. The DP NKT cells were predominantly CD24 hi CD44 lo NK1.1 − , consistent with the phenotype of the earliest identifiable iNKT cells. Another population of iNKT cells also lay within the CD8 quadrant gate of our analysis of NOD.Va14 tg mice; comparison with Fig. 2A of Gapin et al. 29 suggested that these correlate with the DP dull population previously identified as immediately post-selection iNKT cells. Consistent with this, these cells also expressed a relatively immature phenotype: CD24 hi CD44 lo NK1.1 − .
In contrast to thymic NKT cells, virtually all (> 99%) iNKT cells in the livers and spleens of NOD.Va14 tg mice were either CD4 + or DN. In both peripheral organs of NOD.Va14 tg mice, the proportions of the two major iNKT cell subsets that express the relatively mature CD24 lo CD44 hi phenotype were similar to WT, as were the proportions of CD44 hi iNKT cells of either subset that expressed NK1.1 (Fig. 3A). These peripheral iNKT cells were functional and responded to in vivo stimulation with α-GalCer by robust cytokine production (Fig. 3B). These results suggest that although expression of the Va14 transgene on this genetic background has resulted in increased numbers of mature peripheral NKT cells, the majority of immature iNKT cells fail to mature and leave the thymus.
Transcriptional analysis of immature iNKT cells in NOD.Va14 tg mice. The presence of large numbers of DP hi CD24 hi CD44 lo NK1.1 − , CD1d-tet + TCRβ + cells in the thymus and their absence in the periphery, suggested that these cells represent a very immature-possibly preselection-population of iNKT cells. This gave us the opportunity to compare the transcriptional profiles of DP CD24 hi CD44 lo NK1.1 − CD1d-tet + TCRβ + cells (i.e. very immature iNKT cells) with those of somewhat more mature iNKT cell subsets in order to identify the transcriptional transitions associated with positive selection and lineage commitment in NKT cells.
CEL files from all 28 samples were imported into Partek Genomics Suite 6.6 (Partek SG, Singapore) where a Principal Component (PC) analysis plot was generated based on the non-normalised data (as present in the Gene counts data mode). Here, samples are represented as dots and distance between each other reflects similar expression patterns of a large number of genes, while those with large gene numbers with different expression patterns are further apart. Our analysis indicated that the three largest PCs explained > 58% of variation across all our samples, and that they grouped into 4 groups. When coloured using the first categorical variable, the Cell subtype, our samples from each of the four cell types were shown to cluster together into their individual groups, and the groups were distributed evenly across PC1 (X axis, which explained > 29% of variation) in the order DP T cells → DP NKT cells → CD4 NKT cells → DN NKT cells (Fig. 4B). The transcriptional progression of these immature iNKT cell subsets therefore mirrors the postulated developmental progression previously described 51,52 . Of particular interest is the possibility that our analysis of DP NKT cells might identify the transcriptional events associated with Control Point 1 of iNKT cell development, which may be associated with positive selection 51 Fig. 4C). The network was then analysed as three progressive transitions: Transition 1) DP T cells → DP NKT cells; Transition 2) DP NKT cells → CD4 NKT cells; Transition 3) CD4 NKT cells → DN NKT cells. For each transition, uncorrected Student's t-Test p values of pair-wise comparisons of each transcript (GSE106720) were mapped onto the nodes within the network and visualised using a heat map in Cytoscape 3 53 (Fig. 4D). Highly differentially expressed (HDE) transcripts were defined as significant after Bonferroni p value correction (p < 1.5 × 10 -6 ).
Transition 1: Modulation of TCR signaling. Across transition one, 311 transcripts were identified in the network as being HDE. Of these, 116 were in the module named "Pink" (of 525 nodes) and 48 were in the Purple module (of 188 nodes; p < 0.0001 χ 2 contingency; Fig. 4D). Gene Ontology Analysis was performed on the transcripts within the Pink module in The Database for Annotation, Visualization and Integrated Discovery (DAVID) Bioinformatics Resources v6.7 49 . The top ranked Category (SP_PIR_KEYWORDS) obtained for the transcripts in the Pink module was oxidative phosphorylation with a Bonferroni-corrected p value < 4.1 × 10 -6 . Within this category were genes encoding several components of the mitochondrial electron transport chain, NADH Dehydrogenase (Nd1, Nd2, Nd4l, Nd5), Cytochrome C Oxidase (Cox1, Cox2, Cox3) and ATP Synthase 6 (Atp6), all of which were downregulated across Transition 1. This finding is consistent with the intracellular metabolic diversion of small carbon chains away from oxidative phosphorylation to fatty acid synthesis, as proposed by Warburg to be indicative of increased cell proliferation 54 .
Evidence of increased proliferation across Transition 1 was therefore sought by in vivo Bromodeoxyuridine (BrdU) uptake. NOD.Va14tg mice received three intraperitoneal (i.p.) injections of 1 mg BrdU at 12-h intervals, the mice were culled and thymocytes were examined by flow cytometry (Fig. 5A). Transition 1 was associated with a 55% increase in cells taking up BrdU in the 36-h labelling period (p < 0.0002; Mann Whitney U Test; n = 8), confirming increased proliferation.
Transcriptional evidence was sought to support the hypothesis that the increase in proliferation across Transition 1 was due to TCR signalling. The expression levels of all 35,556 transcripts across Transition 1 were compared pairwise by Mann-Whitney U (MWU) Test, and those with a U statistic of zero (i.e. no overlap between groups) shortlisted for ranking by t-Test (i.e. transcripts for which the means of the groups were separated by largest multiples of SEM were prioritised). Across this transition, 5,565 transcripts generated a Mann-Whitney U score of zero, and of these, 1,195 were HDE (p < 1.4 × 10 -6 , by t-Test). This gene list was split into transcripts upregulated (624 transcripts), and those downregulated (570), before being submitted to DAVID for functional annotation clustering.
The possibility of increased proliferation in DP NKT cells was also supported by the up-regulated expression of genes associated with the cell cycle and cell division. A large number of genes encoding proteins involved in DNA replication, components of centromeres and the E2F signalling pathway, were significantly up-regulated by immature DP NKT cells compared to immature DP T cells. E2f1, which encodes the E2F1 transcription factor, was highly significantly up-regulated across Transition 1 (p < 9.1 × 10 -11 , student's t-test; Fig. 5C). Consistent with increased E2f1 expression, E2F1-regulated genes encoding proteins critical for S phase entry 56 were also up-regulated, such as Cdc6 (p < 1.4 × 10 -9 , student's t-test) and Ccne1 (p < 8.7 × 10 -8 , student's t-test; Fig. 5C). Furthermore, consistent with E2F1 up-regulation driving cell proliferation, genes encoding proteins critical for spindle formation and chromosome segregation (such as Cenpm) 57 and DNA replication (such as Rpa2) 58 were also up-regulated (p < 3.7 × 10 -11 ; and p < 4.6 × 10 -10 respectively; Fig. 5C).
In summary, the combination of activation, proliferation, TCR tuning and allelic exclusion combine to provide compelling evidence that TCR signaling occurred across Transition 1.  www.nature.com/scientificreports/ Transition 2: Lineage Commitment. In Transition 2, from DP NKT cells-> CD4 NKT cells, 6,791 transcripts generated a Mann-Whitney U score of zero. Of these, 2,143 were HDE (student's t-test p < 1.4 × 10 -6 ), 125 of which mapped to the Weighted Gene Co-expression Network, with 33 of them in the module named "Pink" and 85 in the module named "Purple". The full HDE gene list was split into transcripts upregulated (1,216 transcripts), and those downregulated (930 transcripts), before submitting it to DAVID for annotation.
Of the upregulated HDE transcripts, amongst the most strongly differentially expressed (U stat = 0) were genes associated with the immunomodulatory and innate-like characteristics of iNKT cells, such as: Tlr1 (p < 1.1 × 10 -16 , t-Test; Fold Change (FC) 15) which encodes Toll-Like Receptor 1 and plays a fundamental role in pathogen recognition and activation of innate immunity; Nkg7 (p < 3.8 × 10 -15 ; FC 36) which encodes Natural Killer Cell Granule Protein 7, which is associated with the cell mediated cytotoxic synapse; Sema4a (p < 9.6 × 10 -10 ; FC 10), which encodes Semaphorin 4A, a type I integral membrane protein required for Th1 deviation and T-bet expression in T cells; Art2b (p < 1.5 × 10 -14 ; FC 49) which encodes ADP-ribosyltransferase 2b that mediates apoptotic deletion of T-cell subsets 59 , particularly CD4 + NKT cells 60 ; S1pr1 (p < 5 × 10 -16 ; FC 22), which encodes Sphingosine-1-Phosphate Receptor 1, which plays an important role in lymphocyte egress from lymphoid tissues; and Zbtb16 (p < 6.5 × 10 -14 ; FC 14), which encodes Promyelocytic Leukaemia Zinc Finger (PLZF), a transcription factor that drives differentiation into iNKT cells and human MR1-specific MAIT cells ( Fig. 6A and Supplementary Table 1). All of these strongly upregulated transcripts lay within the PURPLE module, and had been identified in the network analysis. These data indicate that a wide range of iNKT cell-associated surface receptors are upregulated at this transition and confirm the finding by Savage et al. 61 and Cohen et al. 62 that PLZF expression in iNKT cells is upregulated between Stage 0 and Stage 1 in development.
The strong upregulation of such a wide range of functional lymphocyte-associated, surface-expressed, integral membrane proteins suggests that Transition 2 is associated with iNKT cell lineage commitment. Particularly indicative is the upregulation of Zbtb16, which is responsible for driving the innate-like differentiation of iNKT cells 63 . To test whether the upregulation of Zbtb16 across Transition 2 was sufficient to modulate PLZF target gene expression, we examined the transcript levels of the 69 genes identified by Savage et al. 61 (Fig. 6B). In addition, members of the killer-cell lectin-like receptor group-B receptor ligands, variably known as NK1/NK1.1, were upregulated across this transition , including Klrb1a (1.8 × 10 -13 ; FC 3.63) and Klrb1c, the major target of anti-NK1.1 monoclonal antibody (PK136) binding known to identify NK cells from B6 and SLJ mice (P < 1.81 × 10 -6 ; FC 1.63).
In summary, the coordinated expression of the innate-like lymphocyte-associated transcription factor PLZF and the subsequent upregulation of a wide range of cell-surface functional receptors associated with iNKT cell immunobiology combine to provide evidence that iNKT cell lineage commitment occurred between Population 2 and Population 3, across Transition 2, and confirms previous reports by Savage et al. 61 and Cohen et al. 62 that this occurs by stage 1 in development.
Transition 2: Selection. The association of modulation or tuning of iNKT cell TCR signalling with Transition 1 and lineage commitment with Transition 2, raised the issue of the timing of iNKT cell selection, which presumably could not occur later than Transition 2. Huang et al. 64 published a custom microarray analysis of transcriptional changes during T cell selection by comparing transcripts between DP thymocytes from MHCdeficient mice (C57BL/6.B2m −/− .Ab −/− .E null ) with those from positively selecting TCR transgenic mice (5CC7, from an A b -restricted CD4 cell specific for pigeon cytochrome c; and either F5, originated from a D b restricted CD8 cell specific for influenza virus nucleoprotein; or P14, from a D b -restricted CD8 cell specific for lymphocytic choriomeningitis virus glycoprotein). We identified equivalent Affymetrix transcript cluster IDs for 32 of the 44 transcripts reported by Huang et al. as being up-regulated by T cell selection; of these, 8 had a U statistic of zero across Transition 1 (NS, Chi-square contingency table), 24 did so across Transition 2 (p < 0.0001, and 12 did so across Transition 3 (p < 0.02). At Transition 2, all but three transcripts with a U Statistic of 0 were upregulated; in total 26 of the 32 transcripts were upregulated across this transition (Fig. 6C).
These data are consistent with iNKT cell selection continuing across Transition 2. To test this, the effect of failed positive selection on iNKT cells in NOD.Va14 tg mice was examined in NOD.Va14tg.Cd1d −/− mice lacking the CD1d NKT cell selecting ligand in comparison with WT NOD mice and NOD.Va14 tg mice. In WT NOD mice, targeted deletion of CD1d resulted in the loss of almost all CD1d-tet-staining cells, both in the thymus and the periphery (Fig. 7A,C). In contrast, although NOD.Va14 tg .Cd1d −/− mice had greatly reduced numbers of peripheral iNKT cells (Fig. 7D), the numbers of thymic CD1d-tet-staining cells were more than doubled, with the  www.nature.com/scientificreports/ proportion of iNKT cells that were DP rising from ~ 60% to over 80% (p < 0.005; Mann-Whitney U Test; n = 6-10; Fig. 7B) and the vast majority (> 98%) expressing the Population 2 (DP CD24 hi CD44 lo NK1.1 − ) phenotype. With regard to absolute numbers, NOD.Va14 tg mice had 8.30 ± 0.56 × 10 6 (mean ± SEM) Population 2 iNKT cells, while NOD.Va14 tg .Cd1d −/− mice had a > twofold increase to 21.44 ± 1.87 × 10 6 cells (p < 0.005; Mann-Whitney U Test). In contrast, numbers of mature thymic (CD24 lo CD44 hi NK1.1 − ) NKT cells fell more than 20-fold from 5.99 ± 0.73 × 10 5 in NOD.Va14 tg mice to 0.29 ± 0.06 × 10 5 in NOD.Va14 tg .Cd1d −/− mice (p < 0.005; Mann-Whitney U Test). These changes are consistent with maturational arrest between these two stages in Cd1d −/− mice resulting in a backlog of DP hi CD24 hi CD44 lo pre-selection iNKT cells and a failure of the lineage to develop past the maturational step associated with positive selection.
As the CD1d-deficient mouse strain used here 43 disrupts only the CD1d1 gene, and, unlike the B6 strain who lack CD1D2 protein expression (due to a frame-shift mutation at the beginning of the fourth exon encoding the α3 domain), NOD mice express high levels of CD1D1 and CD1D2 thymic transcripts at a ratio close to 1:1 65 , it was therefore possible that CD1D2 molecules impacted iNKT cell selection and function in this strain, perhaps by presenting a different repertoire of self-antigens than CD1D1 66 . While NKT cells were undetected in the thymi of the targeted deletion NOD strain, it was possible that the presence of the transgene may have impacted their detection in NOD.Va14 tg .Cd1d −/− mice. In order to confirm that the maturational arrest of immature DP NKT cells in NOD. Va14 tg mice was a consequence of the absence of CD1d endogenous ligand, and not due to the NOD background strain, NKT cells were also examined in B6. Va14 tg .Cd1d −/− mice. Thymic NKT cells were more than quadrupled, increasing from 7.8% of total thymocytes of B6.Va14 tg mice to almost 21% in B6.Va14 tg .Cd1d −/− strain, concordant with very few peripheral NKT cells (p < 0.005; Mann-Whitney U Test; n = 5-7). In contrast, targeted deletion of CD1d in B6 WT mice, resulted in the loss of almost all NKT cells in the thymi, livers and spleens. Similar to that found in NOD.Va14 tg .Cd1d −/− mice, the increase in thymic NKT cells was largely attributed by the accumulation of immature DP NKT cells, rising from 48% of total thymocytes in B6.Va14 tg mice to over 85% in B6.Va14 tg .Cd1d −/− mice (p < 0.005; Mann-Whitney U Test; n = 5-7). Absolute numbers of thymic DP NKT cells increased more than tenfold, from 1.2 × 10 6 in B6. Va14 tg mice to 13.6 × 10 6 cells in B6. Va14 tg .Cd1d −/− mice, while mature CD24 lo CD44 hi NKT cells were severely diminished.
This accumulation of immature DP NKT cells may therefore be a consequence of early TCRα expression due to the presence of the transgene 67 and may be dependent on Vβ usage 68 . Indeed, it has been previously reported that in a few TCRβ rearranging cells, TCRα proteins are expressed so early that they mimic the pre-TCRα chain with regard to induction of cell maturation as well as allelic exclusion 69 . The NOD.Vα14 tg /1mice exhibited the same characteristic of TCR Vβ usage as NKT cells from WT mice, however, 70-80% of CD44 − NKT cells and ~ 38% of CD44 + NKT cells in the thymus of NOD. Va14 tg mice used TCR Vβ chains other than Vβ8, Vβ7 or Vβ2, exceeding by far the frequency detected in the thymi of WT NOD mice (~ 0%; data not shown). These findings suggest that the choice of TCRβ chain affects the probability that a α-GalCer/mCD1d tetramer-binding thymocyte will transition from a CD44 − to a CD44 + phenotype. These results reflect the findings of Bedel et al. 70 who constructed an unusual TCR-β chain, that when paired with the canonical Vα14-Jα18 iNKT TCR-α chain, showed increased affinity of the αβ TCR for the antigen/CD1d complex, regardless of the antigen involved. iNKT cell precursors that expressed the high-affinity TCR for "self " in these mice did not fully mature, thus providing direct in vivo evidence that iNKT cells, like their "conventional" T-cell counterparts, are subject to clonal deletion during development. Unexpectedly, they also found that a small fraction of cells with iNKT cell properties, avoided negative selection in these mice, just as in our analyses. Klibi and Benlagha 71 also reported that NKT cell migration out the thymus can occur in the absence of CD1 expression, but that it, and other key components necessary for NKT positive selection, are required for maturation in peripheral organs.
These results are consistent with the previous findings that iNKT cell positive selection and lineage commitment occur at the DP stage 29,71-73 .

Effect of enhanced negative selection on iNKT cells in NOD.Va14 tg mice. Negative selection in
the thymus is a process resulting in clonal deletion (apoptosis) of thymocytes with high avidity for the endogenous ligands of their TCR. As the T precursor frequency for any given antigen is generally low in WT animals, in vivo models of negative selection either involve administration of anti-TCR/CD3 antibodies 74 , responses to endogenous or administered superantigens that ligate whole families of TCR Vβ chains 75,76 , or TCR transgenic mice 77 . Although α-GalCer is technically not a superantigen, it is a strong NKT cell glycolipid antigen capable of activating the vast majority of Vα14-Jα18 expressing NKT cells following its presentation in the context of CD1d 78 , and is capable of causing their negative selection if administered while they transition through a susceptible developmental window before they exit the thymus 79 . Pellici et al. 79 showed that CD4 + CD1d-tet + Nk1.1 − cells (Population 3 cells in our analysis, which were the first stage of iNKT cells that they could clearly detect) were already beyond the "negative-selection" window.
The effects of negative selection on Population 2 (DP hi CD24 hi CD44 lo NK1.1 − ) NKT cells in the thymi of NOD.Va14 tg mice was examined by i.v. injection of α-GalCer, and subsequent flow cytometric analysis of numbers and subsets of iNKT cells (Fig. 8). Within 40 h of injection, total thymocytes had only slightly decreased from 1.55 ± 0.10 × 10 8 (mean ± SEM) to 1.15 ± 0.11 × 10 8 cells (p < 0.05; Mann-Whitney U Test; n = 10) while the proportion of (CD1d-tet + TCRβ + ) NKT cells reduced from 7.75% of total thymocytes in PBS-control mice to 5.4% in α-GalCer injected mice, halving the numbers of NKT cells from 11.9 ± 0.7 × 10 6 to 5.4 ± 0.2 × 10 6 cells (p < 0.0001). The majority of the reduction in thymic iNKT cells was contributed by a 60% reduction in numbers of the DP subset, which fell from 62.7 ± 0.9% to 48.4 ± 1.4% of iNKT cells (7.5 ± 0.5 × 10 6 to 3.1 ± 0.4 × 10 6 cells; p < 0.0001; Fig. 8C), consistent with this population being subject to negative selection. Although proportions of the DN and CD4 + populations of thymic iNKT cell numbers were significantly increased in α-GalCer treated mice (p < 0.0001), they showed modest reductions in terms of absolute numbers (CD4 + NS; DN p < 0.01). In contrast, www.nature.com/scientificreports/ numbers of iNKT cells in the periphery were greater in NOD.Va14 tg mice injected with α-GalCer, increasing from 1.85 ± 0.16 × 10 6 in livers of the controls to 3.50 ± 0.28 × 10 6 in those of the treated mice (p < 0.005) and from 4.65 ± 0.48 × 10 6 to 7.24 ± 0.53 × 10 6 in the spleens (p < 0.005), consistent with antigen-induced activation and proliferation (Fig. 8C). In a separate experiment, the effects of i.v. injection of α-GalCer were examined in NOD.Va14 tg mice bearing a targeted gene deficiency of CD1d (NOD.Va14 tg .Cd1d −/− mice). No significant changes in iNKT cell numbers were observed in either the thymi or the periphery, consistent with the changes observed in NOD.Va14 tg mice being dependent on α-GalCer presentation by CD1d (Supplementary Fig. 1).
Peripheral activation of large numbers of T cells by systemic administration of a polyclonal activator can cause bystander thymocyte death by eliciting a "cytokine storm" 80 . In order to minimise the effects of systemic T cell activation in our model of enhanced negative selection of immature iNKT cells, NOD.Va14 tg mice were subjected to intrathymic injection of α-GalCer and flow cytometric analysis of numbers and subsets of iNKT cells 40 h later (Fig. 8D). The effects of intrathymic injection of α-GalCer on thymic iNKT cells were similar to those of systemic α-GalCer administration. In terms of absolute numbers, thymic iNKT cell numbers fell from 3.97 ± 0.70 × 10 8 (mean ± SEM) to 2.31 ± 0.36 × 10 8 cells. The proportion of iNKT cells that were DP fell from 58.7 ± 2.5 to 18.5 ± 1.2% (p < 0.005; Mann-Whitney U Test; n = 5-7). Again, the proportions of CD4 SP and DN thymic iNKT cells increased, from 13.7 ± 0.6 to 29.5 ± 0.7% (p < 0.005) and from 23.3 ± 1.6 to 47.1 ± 0.7% (p < 0.005) respectively. The specificity of the deletion induced by intrathymic α-GalCer on developing iNKT cells, as distinct from conventional T cells, was illustrated by the relative depletion of DP hi CD24 hi CD44 lo NKT cells (> 82%) compared to DP hi CD24 hi CD44 lo conventional T cells (< 19%).
Transition 3: Differentiation. The expression levels of all 35,556 transcripts across Transition 3 (immature CD4 + NKT cells to DN NKT cells) were compared pairwise by Mann-Whitney U test, and those with a U statistic of zero (i.e. no overlap between groups) were shortlisted for ranking by Student t Test. Across this transition, 6,904 transcripts generated a Mann-Whitney U score of zero, and of these, 1,849 were highly differentially expressed (HDE; p < 1.0 × 10 -6 , by t-Test). Of the HDE transcripts, 1,270 were up-regulated in DN NKT cells compared to CD4 + NKT cells, while 579 transcripts were down-regulated. Only 47 of these HDE genes were in the network, 27 in the "Pink" module and 15 in the "Purple" module. We therefore submitted all (both up and down-regulated) genes lists to DAVID Bioinformatics Resources v6.7 49 for gene ontology analysis of functional annotation clustering.
As a generalisation, the gene ontology analysis of functional annotation clustering of the downregulated genes was unremarkable. In no case was the Bonferroni corrected Enrichment p value less than 0.001. This unexceptional finding was re-iterated when comparing our gene list with that of Cohen et al. 62 , Stage 1 genes. While we could confirm 186 of their 202 genes (Supplementary Table 1), as significantly downregulated by Stage 1, our data indicated that most were down-regulated between the DPNKT and CD4NKT stage, (6 were downregulated in Transition 1 alone, and 7 showed down-regulation but did not meet our criteria for stringent cut-off, in both Transition 1 and Transition 2). None of the genes mentioned in their gene list, however, were down-regulated between CD4 and DN stage in our dataset.
In stark contrast to this, the list of upregulated genes was dominated by those required for the production of ribosomes: The top-ranked Annotation Cluster (Enrichment Score 29.06) contained the GO term Nucleolus (GO:0005730; enrichment 5.86-fold; p < 1.54 × 10 -34 Bonferroni corrected) as well as its various parent terms (nuclear lumen, intracellular non-membrane bound organelle, intracellular organelle lumen). Similarly, the second top-ranked Annotation Cluster (Enrichment score 23.16) contained the GO term rRNA Processing (GO:0006364; enrichment 10.11-fold; p < 2.53 × 10 -19 ), as well as its various parental terms (ncRNA Processing, rRNA Metabolic Process, RNA Processing, Ribonucleoprotein Complex Biogenesis, Ribosome Biogenesis and ncRNA Metabolic Process). In a comparison with Cohen's upregulated Stage 1 gene set 62 , we confirmed 123/146 transcripts as significantly upregulated in our dataset (Supplementary Table 2; while most genes showed upregulation in Transition 2, 9 were significantly highly differentially upregulated in Transition 1, and although most remained unchanged in Transition 3, 18 of the transcripts were again highly significantly up-regulated during this transition. These included Plac8, Enpp1, St6galnac4, Slc22a3, Dhrs11, Agpat2, Sccpdh, Maf, Trf, Top1mt, Samsn1, Tgm3, Ptpn13, Zbtb10, Fcrl1, Dnmt3b, Ehd4, Ntrk3, genes associated with Metabolic Pathways, Calcium Ion Binding and Positive Regulation of Gene Expression. Nine transcripts identified as upregulated by Stage 1 in Cohen's dataset 62 , that were also up-regulated by our CD4NKT cells (Gpr114, Itgae, Ar, Appl2, St8sia6, Dse, App, Acsbg1, Adam19; genes associated with the terms Glycoprotein, Signal, Disulphite bond, Membrane, Transmembrane and Protein binding), were again significantly downregulated when transitioning to DN NKT cells.
Although these data are consistent with functional differentiation across this transition, there is no evidence of major differences in the amount of proteins being produced by the CD4 + and DN subsets of mature iNKT cell after export 52 . An alternative possibility is that the production of ribosomes reflects the maturity of the cells, and the DN population, being derived from the CD4 NKT cells is therefore likely to be, on average, more mature. iNKT1, iNKT2, iNKT17 branch point. Most 7,83 . We used these genes as identifiers as possible indicator as to branch point for the individual subsets ( Supplementary Fig. 2). Expression levels remained unchanged for most identifiers across these transitions. Tbx21 was upregulated in Transition 1 and remained consistently so throughout, while Gata3 showed high level expression in all subsets, with particular upregulation in Transition 3, confirming Cameron and Godfey's previous observations 93 . Other identifiers were upregulated, but showed little difference between our early CD4NKT and DNNKT populations; Rorc was most highly expressed in Population 1, downregulating in Transition 2 and again in Transition 3; and Bcl11b was most highly expressed in DP T cells, with slight down-regulation in Transition 1 and then remaining constant until Transition 3 when it down-regulated again, although still remaining highly expressed. All other genes were of low to medium expression, showing no difference between subsets.
In summary, these observations may be an indication that there is no subset differentiation this early in the developmental pathway, although our data is unable to provide support for either of the two theories.

Discussion
This manuscript examines the transcriptional profiles of immature thymic iNKT cell subsets in a novel Vα14 TCR transgenic strain, and reveals functional characteristics initiated by positive selection on CD4 + CD8 + (double positive, DP) thymocytes that are distinct from conventional T cells. Previously published data suggesting that the thymic DP hi CD24 hi CD44 lo NK1.1 − CD1d-tet + TCRβ + population represents iNKT cells that have not yet undergone thymic selection can be summarised as follows: (1) The population lacks expression of the developmental markers typical of mature iNKT cells, viz: NK1. 1 and CD44. (2) It expresses the CD24 marker, which is characteristic of conventional T cells prior to selection.
(3) Quantitative PCR analysis of Vα14-Jα18 encoded transcripts of sorted DP hi thymocytes found equivalent proportions in WT and Cd1d −/− mice 28 . Similarly, multiple TCR rearrangements could be detected in DP hi thymocytes of WT mice, in contrast with only a single Vα14-Jα18 rearrangement in the DP dull population of WT mice 28 .
The combination of transgenic expression of an NKT-associated Vα14-Jα18 TCRα chain cDNA on the SLAMdeficient NOD strain background 23,24,38 resulted in greatly increased numbers of immature DP iNKT cells. The behaviour of these cells under experimental conditions was consistent with them not yet having undergone selection, in as much as: (1) Cells of this phenotype were not found outside the thymus, despite very large numbers of them being found within it; (2) Their numbers were more than doubled by targeted deletion of CD1d, which is the antigen presentation glycoprotein required for iNKT cell positive selection; (3) They were depleted to about half their numbers by systemic or intrathymic injection of the strong iNKT cell glycolipid antigen α-GalCer in a CD1d-dependent manner; and (4) They showed only three of 44 transcriptional hallmarks of selection.
NOD.Va14 tg mice provided a model of iNKT cell selection in which sufficient pre-selection cells were available that the stages and processes involved could be clearly distinguished. Evidence supporting this use of the model comes from the relatively large number of transcripts differentially expressed between thymic TCRβ + DP hi CD24 hi populations that either did, or did not, bind CD1d-tet. This comparison provided evidence of successful TCR signalling in thymic DP hi CD24 hi NK1.1 − NKT cells, including TCRα allelic exclusion and Rag1 down-regulation.
iNKT cells are positively selected by ligating CD1d expressed on DP cortical thymocytes, which account for > 80% of all thymocytes. The accumulation of pre-selection DP iNKT cells in NOD.Va14 tg mice is consistent with an additional rate-limiting step or signal. One possible candidate is the SLAM-SLAM homotypic interaction between immature iNKT cells and selecting DP thymocytes. Alternatively, other co-stimulators or cytokines, such as IL7 or IL15, may be required.
Although others 62 have identified changes in gene expression between DP T and CD24 lo CD44 lo Nk1.1 − cells (Stage 1), the presence of greatly expanded numbers of very immature iNKT cells in NOD.Va14 tg mice provided an opportunity to study the very earliest identifiable stages, prior to Stage 1, of iNKT cell commitment and differentiation. Thus, while we were able to substantiate other investigators' findings, we could also determine the interval of the occurrence, and in addition, determine the most significant functional changes at each transition. In this way, we could help dissect factors associated with, and contributing to the numbers and function of this important immunoregulatory population.
Transcriptional analysis comparing immature DP T cells and immature DP iNKT cells revealed a reduction in expression of the electron transport genes, such as NADH Dehydrogenase (Nd1, Nd2, Nd4l, Nd5), Cytochrome C Oxidase (Cox1, Cox2, Cox3) and ATP Synthase 6 (Atp6) suggesting reduced mitochondrial oxidative phosphorylation. This is a characteristic of the Warburg phenomenon, in which cancerous or other rapidly proliferating cells increase aerobic glycolysis at the expense of oxidative phosphorylation 94,95 . In addition, the most strongly differentially expressed gene identified across Transition 1 encoded the transcription factor E2F1, which was highly significantly upregulated, lending support for increased proliferation in DP iNKT cells. E2F1, E2F2 and E2F3 act in a functionally redundant manner to enhance the expression of many genes required for G 1  www.nature.com/scientificreports/ regulatory function of the Rb tumour suppressor protein is mediated by its binding to E2F transcription factors, so that while overexpression of E2F1, unopposed by increased RB expression, results in increased proliferation, so does E2f1 gene deletion 97,98 . Consistent with increased E2f1 expression, E2F1-regulated genes, such as Cdc6 and Ccne1 were also significantly upregulated, as were genes encoding proteins critical for spindle formation and chromosome segregation (such as Cenpm) and DNA replication (such as Rpa2).
The role of the E2F pathway in mediating T cell proliferation has been studied in E3f1/E2f2 double knockout mice. It appears to play a role in homeostatic proliferation but not in proliferative responses to exogenous antigen, as homeostatic proliferation in E3f1/E2f2 double knockout mice is severely reduced 99 , but proliferation in response to exogenous antigen is not 99,100 . While homeostatic proliferation of naive T cells requires both IL7 signalling 101 and TCR stimulation by MHC-self peptide 102 , homeostatic proliferation of memory CD4 T cells is dependent on TCR stimulation by MHC-self peptide alone 103 . As homeostatic proliferation of all T cells, including memory CD4 T cells is impaired in E3f1/E2f2 double knockout mice, the E2F pathway must play a role in mediating T cell proliferation in the context of homeostatic proliferation 98 . The early processes of T cell selection resemble those of homeostatic proliferation, in that they are mediated by MHC-self peptide recognition and stimulate proliferation. Activation of the E2F pathway in Transition 1 of iNKT cells suggests that T cell signalling had occurred.
While Benlagha et al. 34 previously reported that DP lo CD24 hi NK1.1 − NKT cells in the thymi of B6 newborn mice were non-dividing, our functional validation by in vivo BrdU incorporation, confirmed our hypothesis that the immature thymic DP hi iNKT cells in NOD.Vα14Tg mice have increased proliferation. Transcriptional evidence of allelic exclusion of competing TCR Vα chains in immature DP iNKT cells was provided by the widespread down-regulation of non-NKT cell TCR Vα genes in Transition 1. This finding was confirmed with flow cytometry by the almost complete absence of immature DP NKT cells and mature TCR Vα14-Jα18-expressing NKT cells co-expressing other TCRα chains, such as TCR Vα2, Vα3.2 and Vα8.3 (data not shown). Downregulation of genes related to the electron transport chain, the activation of the E2F pathway, T cell signaling and the down-regulation of non-NKT associated TCR Vα genes suggests that successful TCR signaling occurs across Transition 1.
Transcriptional analysis suggests that positive selection and lineage commitment of iNKT cells occurs during the transition from immature DP NKT cells to immature CD4 NKT cells. Gene Ontogeny analysis of up-regulated HDE genes across this transition identified a wide range of functional lymphocyte-associated membrane proteins, such as Toll-like receptors, cytokine receptors, chemokine receptors, integrins and leukocyte differentiation markers. In addition, there was a predominance of genes associated with the immunomodulatory and innatelike properties of iNKT cells, such as Tlr1, Nkg7, Sema4a, Art2b, S1pr1 and Zbtb16. Commitment to the iNKT cell lineage is associated with the expression of the transcription factor, PLZF, encoded by Zbtb16 61 . While there was no significant difference in expression of Zbtb16 between immature DP T cells and immature DP NKT cells, Zbtb16 was up-regulated more than 13-fold during the subsequent transition from immature DP NKT cells to the immature CD4 NKT stage.
The coordinated expression of the innate-like lymphocyte-associated transcription factor PLZF and the subsequent up-regulation of a wide range of cell-surface functional receptors associated with iNKT cell immunobiology combine to provide evidence that iNKT cell lineage commitment occurred across Transition 2, between DP NKT and immature CD4 NKT cell stage. The occurrence of lineage commitment at Transition 2 raised the issue of the timing of the NKT cell selection event. Many of the up-regulated genes expressed by the immature CD4 NKT cells have been previously reported as associated with the positive selection of conventional T cells 64,104 consistent with iNKT cell selection also occurring across Transition 2.
Positive and negative selection of iNKT cells has been previously studied. iNKT development is markedly impaired by the absence of the antigen presenting molecule CD1d 25,27,28,43,[105][106][107] or by the early administration of α-GalCer 79,108 . In studies of WT mice, the very low numbers of DP iNKT cells prohibited examination of the effect of targeted deletion of CD1d and administration of α-GalCer on DP iNKT cells 79 .
In the absence of CD1d, immature DP iNKT cells in NOD.Va14 tg .Cd1d −/− mice did not progress to negative selection or maturation and export, resulting in accumulation in the thymus. This may be due to premature transgenic TCRα expression mimicking the pre-TCRα chain 67 and altering the frequencies of TCRβ chain usage 68 . This, however, will require further investigation. The almost complete absence of peripheral iNKT cells in NOD.Va14 tg .Cd1d −/− mice is consistent with failed positive selection of this population in the absence of CD1d. Together, these data are consistent with the hypothesis that positive selection of iNKT cells occurs between the immature DP and the CD4 single positive stages.
In summary, our transcriptional regulatory network approach of iNKT cell development mapped TCR signal modulation or "tuning" to the transition from DP T to DP NKT cells, while positive selection and lineage commitment were associated with the transition from DP NKT to CD4 NKT cells. We speculate that this early signalling event in NKT cells may constitute "validation" of an effective TCR prior to functional differentiation, and that a similar process might also occur in conventional T cells.

Data availability
The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (Dinh et al., 2021) and are accessible through GEO Series accession number GSE106720 (https:// www. ncbi. nlm. nih. gov/ geo/ query/ acc. cgi? acc= GSE10 6720).