Longitudinal analysis of invariant natural killer T cell activation reveals a cMAF-associated transcriptional state of NKT10 cells

Innate T cells, including CD1d-restricted invariant natural killer T (iNKT) cells, are characterized by their rapid activation in response to non-peptide antigens, such as lipids. While the transcriptional profiles of naive, effector, and memory adaptive T cells have been well studied, less is known about the transcriptional regulation of different iNKT cell activation states. Here, using single-cell RNA-sequencing, we performed longitudinal profiling of activated murine iNKT cells, generating a transcriptomic atlas of iNKT cell activation states. We found that transcriptional signatures of activation are highly conserved among heterogeneous iNKT cell populations, including NKT1, NKT2, and NKT17 subsets, and human iNKT cells. Strikingly, we found that regulatory iNKT cells, such as adipose iNKT cells, undergo blunted activation and display constitutive enrichment of memory-like cMAF+ and KLRG1+ populations. Moreover, we identify a conserved cMAF-associated transcriptional network among NKT10 cells, providing novel insights into the biology of regulatory and antigen-experienced iNKT cells.


Introduction
Activation of T cells following recognition of cognate antigen is essential for mounting effective immune responses against pathogens and tumors 1 .
Typically, in the case of MHC-restricted adaptive CD4 + and CD8 + T cells, this requires extensive transcriptional remodeling over several days to facilitate proliferation and differentiation of naive T cells into clonal effector populations that traffic to sites of infection or tissue damage 2 .
Transcriptional and metabolic remodeling is also needed to generate memory T cells that can be rapidly reactivated following secondary antigen encounter during reinfection 2 . Innate T cells, including CD1drestricted invariant natural killer T (iNKT) cells, contrast and complement this paradigm by exiting thymic development as poised 'effector-memorylike' cells already capable of mounting potent cytokine responses within minutes of activation. This allows iNKT cells to rapidly transactivate other immune populations and orchestrate immune responses 3,4 . Activation also induces iNKT cell proliferation, generating an expanded pool of effector cells within 72 hours, most of which subsequently undergo apoptosis as the expanded iNKT cell pool contracts within 7 days [5][6][7] . However, some iNKT cells persist after the immune response subsides [6][7][8] , and there is evidence that antigen challenge induces long term changes in the iNKT cell repertoire analogous to memory T cell differentiation. For example, several studies have demonstrated that activation of iNKT cells with -⍺ galalctosylceramide (⍺GalCer), a potent glycolipid antigen, induces the emergence of novel KLRG1 + and Follicular Helper iNKT (NKT FH ) cell populations that are greatly enriched after 3-7 days, and still detectable >30 days after ⍺GalCer challenge [8][9][10][11] . However, our knowledge of the transcriptional programs underpinning iNKT cell activation remains limited, and there are also relatively few transcriptional resources available for studying activated iNKT cells, especially compared to adaptive T cells 12 .
Analysis and interpretation of iNKT cell biology is also challenging because iNKT cells exhibit heterogeneity, including NKT1, NKT2 and NKT17 subsets that broadly mirror CD4 + Th1, Th2 and Th17 cells 13 . Past studies of NKT1, NKT2 and NKT17 subsets largely focused on iNKT cell thymic development or steady state phenotype in the absence of activation [13][14][15][16] , and less is known about iNKT cell subsets after activation.
Using parabiosis models, we and others have also shown that iNKT cells are predominantly tissue resident 17 , and that this can strongly influence their biology 18 . For example, iNKT cells resident in adipose tissue exhibit an unusual regulatory phenotype characterized by increased KLRG1 expression, reduced expression of the transcription factor promyelocytic leukemia zinc-finger (PLZF), and increased production of IL-10 through an IRE1a-XBP1s-E4BP4 axis, enabling these cells to suppress inflammation and promote metabolic homeostasis 17,18 . Interestingly, Sag et al. (2014) demonstrated that IL-10 + iNKT (NKT10) cells emerge in other organs such as the spleen after repeated antigen challenge 19 , indicating that TCR stimulation can induce a regulatory phenotype, and that NKT10 cells can potentially be considered a memory-like population. However, the relationship between NKT10 cells and other memory-like populations, such as KLRG1 + and NKTFH cells, remains unclear. Furthermore, it is unknown whether similar factors regulate NKT10 cells present in adipose tissue versus those induced after antigen challenge.
To characterize transcriptional remodeling in activated iNKT cells while also considering subset and tissue-associated heterogeneity, we performed single cell RNA-Sequencing (scRNA-Seq) of 48,813 murine iNKT cells from spleen and adipose tissue at steady state and 4 hours, 72 hours and 4 weeks after in vivo stimulation with ⍺GalCer, as well as after repeated ⍺GalCer challenge. We also reanalyzed published human and murine data to generate a transcriptomic atlas of iNKT cell activation states. We found that activation induces rapid and extensive transcriptional remodeling in iNKT cells, and that a common transcriptional framework underpins the activation of diverse iNKT cell populations. However, regulatory iNKT cell populations demonstrate largely blunted activation in response to ⍺GalCer, and display enrichment of memory-like KLRG1 + and cMAF + iNKT cell subsets expressing a T regulatory type 1 (Tr1) cell gene signature. We also show that cMAF + iNKT cells are enriched for NKT10 cells, and express a gene signature similar to NKT FH cells. Overall, this study provides novel insights into longitudinal transcriptional remodeling in activated iNKT cells and the phenotype of regulatory iNKT cells, while also generating a novel transcriptomic resource for interrogation of iNKT cell biology.

Results iNKT cells undergo rapid and extensive transcriptional remodeling in response to ⍺GalCer
To investigate transcriptional remodeling in activated iNKT cells we performed 10x scRNA-Seq of whole murine adipose and splenic iNKT cells 4 hours, 72 hours and 4 weeks after in vivo stimulation with ⍺GalCer, and reanalyzed our published scRNA-Seq of steady state murine adipose and splenic iNKT cells 18 (GSE142845, Figure 1A). We first analyzed our steady state, 4 hour and 72 hour splenic iNKT cell data. After quality control measures we obtained 16,701 splenic iNKT cells, including >4000 cells per activation state. After performing UMAP we observed minimal overlap between iNKT cells from different activation states ( Figure 1B), indicating that iNKT cells undergo rapid and extensive transcriptional remodeling during early activation. Using gene expression analysis (Table   S1) we found that steady state iNKT cells displayed enrichment of NKT1 and NKT17 cell markers such as Il2rb, Klrb1c, Rorc, and Il7r ( Figure   1C) 13 , but following activation iNKT cells rapidly downregulated these genes within 4 hours, and upregulated expression of T cell activation markers and cytokines, including Il2ra, Irf4, Nr4a1, Pdcd1, Ifng, Il4 and Il17a ( Figure 1C). This was accompanied by increased expression of Zbtb16 (PLZF) and the PLZF regulon genes Icos and Cd40lg ( Figure 1C), consistent with published data demonstrating that PLZF is required for the innate response of iNKT cells to antigen 20 . Activated iNKT cells also downregulated expression of the transcription factor Id2 ( Figure 1C), which plays an essential role in normal iNKT cell activation 21 , expression of activation and cytokine genes was   greatly reduced, and we identified enrichment of genes associated with   proliferation, stem-like T cells, and NKT2 or Stage 2 iNKT cells, including Mki67, Slamf6, Tcf7 and Ccr7 ( Figure 1C) 25,26 . We observed that some 72 hour cells displayed enrichment of T FH and NKT FH markers, including Cxcr5, Il21 and Maf 9,11,12,27 , and memory-like iNKT cell markers, such as Itga4 and Klrg1 8 ( Figure 1C, Figure S1), corresponding with previous studies documenting the appearance of NKT FH and KLRG1 + iNKT cells after ⍺GalCer challenge [8][9][10]28 . We also found increased expression of genes associated with the KLF2 regulon, including Klf2 and S1pr1 ( Figure   1C). KLF2 is known to induce T cell thymic egress and trafficking through secondary lymphoid organs 29  and Gene Ontology Consortium 31 databases (Table S2) and scored our data. We found 4 hour activated cells upregulated glycolysis, amino acid metabolism, polyamine synthesis and fatty acid synthesis signatures, whereas oxidative signatures were downregulated compared to steady state iNKT cells ( Figure 1D). This suggests that, despite being poised at steady state for cytokine production, activated iNKT cells, like adaptive T cells, switch on aerobic glycolysis and upregulate biosynthethic pathways to fuel cytokine production, growth and proliferation 22,32,33 . Our data is also consistent with recent work identifying glucose as an important fuel for iNKT cell effector function 34,35 . Interestingly, we found that 72 hour activated cells engage oxidative signatures while maintaining elevated expression of glycolytic genes ( Figure 1D), indicating that the metabolic requirements of iNKT cells change across different activation states. We also observed reduced expression of polyamine synthesis and amino acid metabolism signatures 72 hours post-⍺GalCer, suggesting that those pathways are coupled to early iNKT cell activation and cytokine production, while oxidative metabolism may be coupled to proliferation when the iNKT cell pool expands 4-10 fold in vivo by 72 hours 6,7 .
Having profiled transcriptional remodeling in activated murine iNKT cells, we wondered whether similar remodeling occurs in human iNKT cells. To investigate human iNKT cell activation we reanalyzed published scRNA-Seq data of human iNKT cells isolated from peripheral blood mononuclear cells (PBMCs) and stimulated ex vivo with phorbol 12-myristate 13acetate (PMA) and Ionomycin (GSE128243) 36 . Following quality control measures we obtained 13,957 cells, and we found that human iNKT cells also undergo rapid and extensive transcriptional remodeling after activation ( Figure 1E, Figure 1F). Furthermore, activated human iNKT cells recapitulated the metabolic reprogramming observed in activated murine iNKT cells, displaying upregulated glycolytic, amino acid metabolism and polyamine synthesis signatures, and reduced expression of oxidative signatures ( Figure 1F). Thus, transcriptional signatures of iNKT cell activation are conserved across species.

Oxidative Phosphorylation differentiates functional responses to αGalCer in NKT2 and NKT17 cells versus NKT1 cells
We next asked whether iNKT cell subsets expressed different transcriptional signatures after activation. We performed subclustering of murine splenic iNKT cells at steady state and 4 hours post-⍺GalCer, and identified clusters corresponding to NKT1, NKT2 and NKT17 cells ( Figure   2A) using the published marker genes Tbx21, Zbtb16, Rorc, Ifng, Il4 and iNKT cell subsets. We also identified genes specifically enriched in one or more subsets, such as Gzmb and Ccl4 in NKT1 cells ( Figure 2C; Table   S3). Strikingly, we found that NKT2 and NKT17 cells, but not NKT1 cells, shared expression of many genes, including Lif, Lta, Cd274 and Ncoa7 ( Figure 2C). Activated NKT2 and NKT17 cells also demonstrated increased whole transcriptome correlation compared to activated NKT1 cells ( Figure 2D), indicating that activated NKT2 and NKT17 cells are transcriptionally similar compared to NKT1 cells.
To investigate the shared transcriptional signatures of NKT2 and NKT17 cells, we performed GSEA 37,38 comparing activated NKT2 and NKT17 cells versus activated NKT1 cells using the KEGG pathway database 30 .
We identified enrichment of Oxidative Phosphorylation ( Figure 2E), suggesting that NKT2 and NKT17 cells use oxidative metabolism more than NKT1 cells. To validate this result we first measured mitochondrial mass and membrane potential in thymic CD44 + NKT1, NKT2 and NKT17 cells, and we found that NKT2 and NKT17 cells had significantly increased mitochondrial mass and membrane potential compared to NKT1 cells ( Figure 2F and 2G). We next investigated whether NKT2 and NKT17 cells were more functionally dependent on oxidative metabolism than NKT1 cells by stimulating iNKT cells ex vivo with PMA and Ionomycin for 4 hours in the presence or absence of oligomycin, to inhibit oxidative phosphorylation 39 . Treatment with oligomycin globally reduced cytokine production across all iNKT cell subsets, however, we found that production of IL-4, IL-13 and IL-17A was almost completely ablated compared to production of IFNγ ( Figure 2H), demonstrating that oxidative metabolism is essential for NKT2 and NKT17 cytokines but less so for NKT1 cytokines. and Gzmb ( Figure 3C), which are markers typically associated with Tr1 cells, a heterogeneous population of regulatory T cells that do not express FOXP3 [40][41][42][43][44] . We found that adipose iNKT cells do not express FOXP3 and instead express the transcription factor E4BP4 for IL-10 production 17 .
Other genes enriched among adipose iNKT cells included Il21r, the adenosine receptor Adora2a, and the exhaustion marker Tox ( Figure 3C).
In summary, activation with ⍺GalCer induces differential transcriptional remodeling in adipose versus splenic iNKT cells, and the peak of the regulatory response in adipose iNKT cells is delayed compared to the rapid cytokine burst in splenic iNKT cells.
Notably  Figure 4C). This indicates that all ⍺ adipose iNKT cell subsets respond to ⍺GalCer but no single subset (e.g. NKT1 or NKT17) was uniquely hyporesponsive vis a vis the spleen.
Interestingly, we identified increased oxidative gene expression in adipose NKT17 cells versus adipose NKT1 cells ( Figure 4C), similar to our finding in splenic iNKT cells ( Figure 2). We have previously shown that γδ17 cells also display enrichment of oxidative metabolism 39 , suggesting that this is a conserved feature of innate T cells that produce IL-17.

Analysis of cytokine production among adipose iNKT cells revealed that
Il10 was only expressed by NKT1 cells ( Figure 4B). Since cytokine pos adipose NKT1 cells (Cluster 6) lacked or had downregulated expression of Klrb1c (NK1.1) by 4 hours post-⍺GalCer ( Figure 4B), we could not stratify cytokine production in adipose NKT1 cells using Klrb1c. Therefore, we performed unbiased fine clustering of cytokine pos adipose NKT1 cells.
We identified one population of cells co-expressing Ifng, Il4 and Il2, and a second population expressing Il10 ( Figure 4D), suggesting that adipose

Chronic activation of splenic iNKT cells induces an adipose-like phenotype and the emergence of Tr1 iNKT cells
Since adipose iNKT cells displayed blunted activation after ⍺GalCer, and enrichment of Tr1 cell markers, we wondered whether these were conserved features of regulatory iNKT cell biology. To answer this question we repeatedly activated splenic iNKT cells, which induces IL-10 production 19 . We sequenced 5,433 ⍺GalCer activated splenic iNKT cells,  Table S6), indicating that prior exposure of splenic iNKT cells to antigen does not completely reproduce the phenotype of adipose iNKT cells, which may be exposed to chronic endogenous activation in situ.
Since we had identified a distinct population of NKT10 cells in adipose tissue after GalCer, we wondered whether we could also identify an ⍺ ( Figure 6H), and another population (Cluster C) expressed KLRG1 + associated markers, including Gzmb, Cd244 and Klrk1 ( Figure 6H).
Overall, this data indicates that cMAF + and KLRG1 + cell populations are induced by antigen experience in the spleen, and are constitutively present in adipose tissue.

Identification of a conserved cMAF-associated and NKT FH -like transcriptional state in NKT10 cells
Having identified transcriptional signatures of regulatory iNKT cells in adipose tissue and after serial antigen activation, we next sought to describe shared transcriptional features of these different NKT10 cell populations. Gene expression analysis identified 110 genes enriched among splenic NKT10 cells (Table S7), including Ctla4, Pdcd1, Lag3, Il21, Maf, Hif1a and Ccr5 ( Figure 7A), all of which were already identified in adipose NKT10 cells. We identified 39 genes conserved across NKT10 cells from both tissues ( Figure 7B), including Tr1 cell markers, Tgfb1, and the tolerogenic factors Slfn2 and Vsir 52,53 ( Figure 7B). We also found that splenic NKT10 cells expressed the adipose iNKT cell marker Nfil3 ( Figure   7A), which we previously linked to IL-10 production by regulatory adipose  Table S8). Module scoring of our resting scRNA-Seq data demonstrated that NKT10/cMAF + cells but not KLRG1 + cells are enriched for NKT FH cell gene signatures ( Figure 6G). Overall, this suggests that NKT10/cMAF + cells are transcriptionally similar to NKT FH cells, and these two memory-like iNKT cell populations may phenotypically and functionally overlap. We have shown that E4BP4 (Nfil3) rather than FOXP3 regulates production of IL-10 by adipose iNKT cells 17

Data Availability
scRNA-seq data from this manuscript have been deposited in the Gene Expression Omnibus under accession code GSE190201.

Declaration of Interests
The authors declare no competing interests.