The Long Non-coding RNA Flatr Anticipates Foxp3 Expression in Regulatory T Cells

Mammalian genomes encode a plethora of long non-coding RNA (lncRNA). These transcripts are thought to regulate gene expression, influencing biological processes from development to pathology. Results from the few lncRNA that have been studied in the context of the immune system have highlighted potentially critical functions as network regulators. Here we explored the nature of the lncRNA transcriptome in regulatory T cells (Tregs), a subset of CD4+ T cells required to establish and maintain immunological self-tolerance. The identified Treg lncRNA transcriptome showed distinct differences from that of non-regulatory CD4+ T cells, with evidence of direct shaping of the lncRNA transcriptome by Foxp3, the master transcription factor driving the distinct mRNA profile of Tregs. Treg lncRNA changes were disproportionally reversed in the absence of Foxp3, with an enrichment for colocalisation with Foxp3 DNA binding sites, indicating a direct coordination of transcription by Foxp3 independent of the mRNA coordination function. We further identified a novel lncRNA Flatr, as a member of the core Treg lncRNA transcriptome. Flatr expression anticipates Foxp3 expression during in vitro Treg conversion, and Flatr-deficient mice show a mild delay in in vitro and peripheral Treg induction. These results implicate Flatr as part of the upstream cascade leading to Treg conversion, and may provide clues as to the nature of this process.


INTRODUCTION
Sequencing of the human genome showed that there are only ∼20,000 protein-coding genes, which is comparable to other less complex organisms such as nematodes or the fruit fly. This would suggest that eukaryotic genome must use other ways to generate biological complexity. Indeed, only 1-2% of the genome encodes protein sequences, with the other 98% potentially contributing to complexity through structural modification of accessibility and the generation of non-coding RNA molecules (ncRNA). One particular class of ncRNA, the long-non-coding RNA (lncRNA), have remained particularly enigmatic, long being dismissed as "transcriptional noise". However, recent studies indicate that the ∼35,000 mammalian lncRNA play a significant role in orchestrating and fine-tuning transcriptional programs both in health and disease (1)(2)(3). These lncRNAs can be located within the nucleus or the cytoplasm of a cell and may or may not be polyadenylated. LncRNAs can be categorized into groups according to their localization (intronic, intergenic) and transcriptional direction (sense, antisense, bidirectional) (1).
Very few functional studies of lncRNA have been performed, however the few in-depth studies published have revealed profound roles for these ncRNA. Within the haematopoeitic system, individual lncRNA regulate the survival of the myeloid lineage (4), with several known to be important for the differentiation of eosinophils (5), granulocytes (6), and dendritic cells (7). In addition to the involvement of lncRNA in differentiation, several studies describe their role in innate and adaptive immune response. The long intergenic non-coding RNA (lincRNA)-COX-2 plays a role in the activation or repression of immune-regulatory gene expression in macrophages (8). THRIL (TNF and heterogeneous nuclear ribonucleoprotein L related immunoregulatory lincRNA) is a key player regulating TNFα transcription (9). PACER (p50-associated Cox2 extragenic RNA) is located upstream of the Cox2 transcriptional start site and helps the production of inflammatory mediators (10). NEAT1 (nuclear enriched abundant transcript 1 or nuclear paraspeckle assembly transcript 1) binds SFPQ (a paraspeckle protein) and prevents inhibition of IL-8 (11). Within T cells, ThymoD (thymocyte differentiation factor) aids thymic T cell development by inducing the expression of Bcl11b (12) while NeST enhances IFN-γ production in CD8 + T cells via and chromatin modification of the IFN-γ locus (13). Despite these important functions, it is notable that very few lncRNA have been identified via genetic screens, either due to intrinsic biases toward protein-coding genes in screening methods (14), or because (unlike the examples above) most lncRNA exert relatively subtle impacts on gene regulation, tweaking regulatory networks rather than controlling major phenotypes.
Regulatory T cells (Tregs) are a specialized subpopulation of CD4 + T cells which are critical for the maintenance of tolerance toward self (15). Dysfunction of these cells can result in severe autoimmune diseases, such as type 1 diabetes, multiple sclerosis, rheumatoid arthritis, colitis, and inflammatory bowel disease (16). However, a hyperactive function of these cells is also detrimental, as it can inhibit beneficial anti-pathogen (17) and anti-tumor immunity (18). Forkhead box P3 (Foxp3) is a key transcriptional regulator of Tregs and is used for their identification. Foxp3 is induced during the thymic selection process in a subset of CD4 + T cells with a certain degree of T cell receptor (TCR) self-reactivity, creating thymic Tregs (tTregs) (19). In addition to tTreg generation, Foxp3 induction can occur in T cells while they are present in other tissues, such the gut, colon or placenta, generating peripheral Tregs (pTregs) (20). Foxp3 expression results in radical transcriptional rewriting and functional differentiation, creating T cells with a largely suppressive phenotype (21). In mice and human, FOXP3deficiency results in defective generation of functional Tregs and a fatal breach in immunological tolerance, causing highly aggressive multi-organ autoimmune pathology (22,23).
Few studies have looked at the function of lncRNA in Tregs. In conventional T cells, expression of the lncRNA lnc-EGFR promotes the differentiation of Tregs, licensing tumor growth (24). In Tregs themselves, genomic deletion of the lncRNA Flicr, encoded within the Foxp3 promoter region, results in altered expression of Foxp3 (25). While this latter study could arguably be due to the altered structure of the Foxp3 promoter, owing to the cis-nature of the reported function, it strongly suggests that lncRNA will be of importance in controlling Treg transcriptional profiles. Here we systematically assess the lncRNA profile of Tregs, identifying a novel lncRNA that anticipates Foxp3 expression. This lncRNA, named here Flatr (Foxp3-specific lncRNA anticipatory of Tregs), is highly conserved and enriched in activated Tregs. Generation of Flatrdeficient mice resulted in a minor impairment of in vitro and peripheral Treg induction, indicating a biomarker, rather than major functional, role in the upstream cascade leading to Foxp3 expression.

The Treg LncRNA Transcriptome Is Shaped by Foxp3 Expression
In order to characterize the Treg-specific lncRNA transcriptome, we started with a high throughput sequencing approach. Foxp3 GFP reporter mice (26) were used as a source of naïve CD4 + T cells (CD4 + CD62L + CD44 − GFP − ) and Treg (CD4 + GFP + ). To ensure efficient capture of all lncRNA, and not just those with a polyadenylated tail, we used ribosomal RNA (rRNA) depletion prior to Illumina HiSeq 2000 sequencing. Expression data was mapped onto known lncRNA, of which 1765 were expressed in Treg. Comparative expression analysis found that 13.8% of lncRNA expressed by Tregs were differentially expressed when compared to expression in naïve CD4 + T cells, with 190 lncRNA upregulated in Tregs and 55 lncRNA downregulated in Tregs ( Figure 1A). Using published datasets, these lncRNA core signature changes in Tregs were reproducible and specific, with a tight correlation between thymic and peripheral expression ( Figure 1B) and a similar Treg-specific expression pattern observed across different stages of T cell differentiation ( Figure 1C).
Transcriptome analysis allowed us to define the sets of nondifferential expressed lncRNA (i.e., expressed in both Treg and naïve CD4 + T cells, at equivalent levels), and the core Tregsignature lncRNA, with either upregulation or downregulation in Tregs. Comparison of the non-differential set to the core signature sets revealed three lines of evidence for active control of the lncRNA transcriptome in Tregs. First, core signature lncRNA, both up and down, were disproportionately likely to be intergenic in genomic position (Figure 1D), consistent with transcriptional control mechanisms independent of mRNA changes. Second, core signature lncRNA were twice as likely to possess a Foxp3 chromatin immunoprecipitation (ChIP) peak (27) within the transcript or promoter region (Figure 1E), suggesting a mechanism for direct Foxp3-guided expression modification. Third, when assessing the expression in Foxp3deficient Tregs [using the Foxp3 KIKO transcriptome (28)], an enrichment was observed for the Treg upregulated lncRNA to be reduced in expression and the Treg downregulated lncRNA to be increased in expression ( Figure 1F). Together, these results suggest that the unique lncRNA transcriptome in Tregs is, at least in part, shaped by Foxp3 expression. (D) Non differential, core Treg upregulated, and core Treg downregulated lncRNAs grouped by genomic location relative to protein-coding genes [sense, antisense, and long intergenic non coding (lincRNA)]. (E) Foxp3 Chip-seq peaks (GSE40686) within the promoter region or the gene body of non-differential expressed lncRNAs or core Treg lncRNAs. (F) Differential expression of core lncRNAs in Foxp3 + Treg and Foxp3KIKO Treg (GSE40686).

The LncRNA Flatr Is Specifically Expressed by a Subset of Activated Tregs
Among the unique lncRNA transcriptome in Tregs, we selected Flatr for further study, as one of the most highly expressed Treg-specific lncRNA. Flatr is an intronic lncRNA, located within the gene Cwc27 on chromosome 13 (Figure 2A). Flatr is highly conserved across species, with both exons showing >70% sequence similarity between mouse, human, and chimpanzee ( Figure 2B). In our RNAseq experiment, Flatr expression was high in Tregs while essentially undetectable in naïve CD4 + T cells ( Figure 2C). By contrast, the host gene, Cwc27, was unchanged between Tregs and naïve CD4 + T cells, indicating a specific alteration in the regulation of the lncRNA ( Figure 2D). Analysis of the Th-express compendium (29) also indicated a Treg-specific expression pattern for Flatr, across different CD4 + T cell subtypes ( Figure 2E). As an independent validation, we sorted lymphocyte populations from the thymus and spleen. Flatr expression was restricted to Tregs in both organs, and further showed 3-fold higher expression in Nrp1 + tTregs than Nrp1 − pTregs ( Figure 2F).
In order to determine whether Flatr was expressed by all Tregs, or by only a subset, we adapted RNAflow to the detection of lncRNA, a procedure complicated by the relative shortness of lncRNA limiting RNAflow probe sites. The resulting RNAflow profiles both confirmed the Treg-specificity of Flatr expression, and demonstrated that only a subset of Treg express Flatr at high levels ( Figure 3A). Phenotyping analysis of Flatr low vs. Flatr high Tregs indicated that Flatr high Tregs were more activated, with a shift toward being CD62LlowCD44hi and also elevated expression of ICOS and KLRG1 ( Figure 3B). When analysing the expression of Foxp3 and Flatr during Treg induction in vitro, Foxp3 induction is not efficiently induced until 24 h post-stimulation, and is dependent upon TGFβ ( Figure 3C). By contrast, Flatr expression precedes Foxp3 expression, with substantial upregulation from 4 h post-stimulation, and was only partially TGFβ-dependent ( Figure 3D). Together, these results demonstrate that Flatr is a Treg-specific lncRNA, with expression enriched within the subset of activated Tregs.

Delayed in vitro and Peripheral Treg Induction in Flatr Knockout Mice
To test the function of Flatr in Tregs, we generated new knockout mice deficient in the lncRNA. Two independent strains were generated. In the first, "exon 1 KO mice", CrispR-Cas9-mediated homologous recombination deleted the promoter region and the majority of exon 1 (Figure 4A). In the second, "exon 2 KO mice", donor-free CrispR-Cas9 deletion was used to remove the entirety of exon 2 ( Figure 4B). In both mouse strains, sorted Tregs demonstrated a complete absence of Flatr transcript (Figure 4C), validating the models for analysis of Flatr deficiency.
A potential role for Flatr in Treg development was assessed through analysis of Exon 1 KO and Exon 2 KO mice. We first assessed the role of Flatr in thymic Treg development in vivo. Both strains showed normal T cell development in the thymus (data not shown), with no alteration in the production rate of Tregs in the thymus (Figure 5A). In the periphery, the relative proportions of CD4 and CD8 T cells were unchanged ( Figure 5B). The percentage of Tregs within the CD4 + T cell compartment remained intact ( Figure 5C), with no alterations in the balance of thymic-and peripherally-derived Tregs (Figure 5D), indicating no essential role for Flatr in Treg differentiation or development. Consistent with this, expression  As Flatr expression anticipated Foxp3 expression during in vitro induction (Figure 3D), we next assessed a functional role for Flatr in this process. Using a TGFβ-dependent Treg induction assay, we compared the ability of wildtype and Exon1KO naïve T cells to convert to the Treg lineage. Wildtype naïve T cells demonstrated progressive conversion to the Foxp3 + Treg lineage, a process which demonstrated a TGFβ dose-response ( Figure 6A). Exon1KO naïve T cells exhibited a small but reproducible defect in this Treg conversion process ( Figure 6A). The defect was present across the limiting TGFβ range and across the conversion period (Figure 6A), consistent with a general dampening of Treg conversion. Similar results were observed in culture conditions where IL-2 was limited, although under these conditions wildtype Tregs showed an expansion-contraction kinetics (Supplementary Figure 1). We further investigated peripheral Treg induction in vivo, using oral tolerance and homeostatic conversion assays. OVAreactive OT-II naïve T cells were transferred into a congenic host and gavaged with OVA ( Figure 6B). Using this system, pTreg induction is observed within the mesenteric lymph nodes and the gut (Figure 6C). These pTregs expressed both Foxp3 ( Figure 6D) and Flatr (Figure 6E), demonstrating Flatr upregulation during pTreg induction. To assess a function role during pTreg induction we switched to a homeostatic expansion system. Wildtype and Exon2 KO naïve T cells were injected into a Rag-deficient host ( Figure 6F). This system results in pTreg conversion ( Figure 6G), with upregulation of both Foxp3 ( Figure 6H) and Flatr ( Figure 6I). Using this system, Flatr-deficient naïve T cells demonstrated a lower pTreg induction ( Figure 6G) and reduced expression of Foxp3 in the converted Tregs ( Figure 6H). Together, these results demonstrate a small but reproducible impact of Flatr expression in Treg induction both in vitro and in vivo.
No substantial difference was observed in Treg suppressive function. In vitro suppression of effector T cells was normal when comparing Tregs from KO mice to those from wildtype mice, with no major changes in the proliferation rate of cocultured naïve T cells following stimulation ( Figure 7A). In an in vivo setting, the expression of the major suppressive mediator, CTLA4, was normal ( Figure 7B). A global transcriptional analysis, performed by RNAseq experiments on purified CD4 + CD25 + Tregs from wildtype, Exon 1 KO and Exon 2 KO mice, showed tight clustering of Tregs from each genotype, indicating that Flatr-deficiency does not dramatically alter the Treg transcriptional profile (Supplementary Figure 2). Analysis of differentially expressed genes found few transcriptional changes -only 36 significantly upregulated genes and 8 significantly downregulated genes were consistent across the two knockout strains when compared to wildtype cells (Supplementary Worksheet 1), with no significantly enriched pathways or known suppressive mediators. A normal suppressive capacity was borne out by ex vivo analysis of effector T cells in Flatr-deficient mice, with no spontaneous increase in effector T cells was observed (Figures 7C-F). These results identify Flatr as an anticipatory lncRNA during Treg induction without substantially impacting Treg suppressive capacity under homeostatic conditions.

DISCUSSION
The stable suppressive phenotype of Tregs is induced and maintained through the activity of the master transcription factor Foxp3 (31,32). Foxp3 is well-documented to act as a transcriptional regulator of mRNA. Foxp3 binds with multiple protein partners (33,34), which can in turn bind to ∼700 gene targets (21) and either enhance or suppress the expression of the resulting mRNA, depending on the composition of the complex (34). Several of these targets include secondary transcription factors which serve to "lock in" the Foxp3-induced mRNA transcriptome (35), providing redundancy to the transcriptional circuit. Beyond the abundant data that Foxp3 drives the Treg fate through classical transcriptional control, there is growing evidence that Foxp3 also initiates non-classical programs that influence the stability and function of the lineage. At the DNA level, Tregs undergo extensive epigenetic reprogramming that stabilizes cell identity (36). This is particularly relevant to the Foxp3 locus itself, which undergoes specific DNA demethylation and histone modifications in Tregs (37)(38)(39), however extensive modifications occur at a genome-wide level in functionally relevant genes (40). At the RNA level, Foxp3 not only initiates a mRNA transcriptional program, but also a unique microRNA signature (41). Loss of the Treg microRNA transcriptome results in a loss of suppressor phenotype during inflammation, resulting in fatal autoimmunity (42)(43)(44). These effects are partially mediated through miR-146a (45), miR-155 (46), and miR-17 (47), and are heavily intertwined with the transcriptional control initiated by Foxp3. Our study, and recent advances, implicate lncRNA transcription as an additional layer of complex cellular regulation initiated by Foxp3 and relevant to the Treg fate.
The impact of lncRNA on cellular biology can occur at the epigenetic, transcriptional and post-transcriptional level. There is growing evidence that lncRNA are involved in Treg biology. MEG3, a lncRNA expressed in CD4 + T cells, contributes to the imbalance of Tregs/Th17 ratio in patients with immune thrombocytopenic purpura (48). Expression of lncRNA DQ786243 in Jurkat cells modulates Foxp3 expression, indicating a potential in vivo function (49). The lnc-epidermal growth factor receptor (lnc-EGFR) is highly expressed in Tregs and activates the expression of EGFR and its downstream AP-1/NF-AT1 axis, which increases Treg immunosuppressive function (24). The lncRNA Flicr, negatively regulates Foxp3 expression in Tregs through modification of chromatin accessibility, and alters the course of autoimmune diabetes (25). Here we have defined the scope and nature of the lncRNA transcriptome in Tregs, and identified Flatr as a biologically relevant lncRNA that anticipates Foxp3 expression during Treg differentiation and plays a functional role in the in vitro and peripheral conversion process.
The induction of Tregs from thymocytes or naïve T cells is a coordinated process involving TCR stimulation and costimulatory signals, which, depending on the context, can include signaling from IL-2, TGFβ, and CD28, among others. These upstream signals leave a permanent signature on the nascent Treg, indeed, part of the "Treg" signature is independent of Foxp3 itself, and instead driven by these upstream coordinators (50). Our results indicate that Flatr expression is part of this upstream induction process, as expression of Flatr precedes that of Foxp3 itself. Nevertheless, after initial expression, Flatr becomes dependent on Foxp3 for continued high level expression, as it is reduced in Foxp3 KO "Tregs", a pattern found in several other Treg signature genes, such as CD25 (51). These signals can be differentially integrated by the Foxp3 locus, which contains three conserved non-coding sequences (CNS) that regulate Foxp3 expression (39). Our data on Flatr indicates both a bias toward activated Tregs. Expression of Flatr is largely driven through αCD3/αCD28 stimulation, and accentuated by TGFβ. Consistent with this, in vitro and in vivo models of Treg induction that are dependent on strong TCR signal demonstrated a (minor) function role of Flatr in the process. CSN1 is involved in peripheral Treg induction (39), with TGFβ working via a Smad3 binding site in the region (52). CNS2 aids heritable activation of the Foxp3 locus (53,54), with TGFβ signaling participating indirectly through Stat5 activation (55). A role in either, or both, of these processes is possible for Flatr, based on the induction data. Both of these elements are less potent than CNS3, which has critical importance for thymic induction, and appears to be TGFβindependent (39). By contrast, Flicr is thought to work via CNS3, which potentially accounts for the more prominent, although still minor, phenotype (25). Like most non-coding RNA, the effect of Flatr and Flicr are subtle, likely to modulate these processes rather than acting as a qualitative "on-off " switch. Nonetheless, the conservation of these lncRNA suggests that such modulatory effects, while biologically silent under homeostatic conditions, are likely to manifest with altered pathology under stress conditions.

Mice
Foxp3GFP mice (26) were used for lncRNA profiling. Two independent lines of Flatr KO mice were generated, targeting exon 1 and exon 2 of the Flatr transcript, respectively. C57BL/6 zygotes were injected with CrispR-Cas9 conjugates with gRNA flanking the targeted exon. Resulting offspring were screened for deletions, crossed to C57BL/6 mice and intercrossed to generate experimental mice. Flatr KO mice were backcrossed to Foxp3 Thy1.1 mice for cell sorting (56). Antigen presenting cells were collected from Rag1 KO mice on the B6 background. Transfer experiment included Rag1 KO on the B6 background. OT-II OVA-specific TCR-transgenic mice on a C57BL6 background were obtained from The Jackson Laboratory (Bar Harbor, ME). This study was carried out in accordance with the recommendations of the European Union (EU) concerning the welfare of laboratory animals as declared in Directive 2010/63/EU and University of Leuven ethics committee. The protocol was approved by University of Leuven ethics committee.

LncRNA Sequencing and Analysis
Tregs (CD4 + GFP + ) and naïve T cells (CD4 + CD62L + CD44 − GFP − ) from Foxp3 GFP reporter mice (26) were sorted by FACS. RNA was isolated using NucleoSpin RNA isolation kit (Macherey-Nagel). RNA concentration and purity were determined spectrophotometrically using the Nanodrop ND-1000 (Nanodrop Technologies) and RNA integrity was assessed using a Bioanalyser 2100 (Agilent). Per sample, 400 ng of total RNA was used as input. Using the Illumina TruSeq R Stranded Total RNA Sample Prep Kit with Ribo-Zero Gold (protocol version "April 2013") rRNA was depleted from the total RNA samples using Ribo-Zero ribosomal RNA reduction chemistry. Subsequently, RNA was purified and fragmented and converted into first strand cDNA in a reverse transcription reaction using random primers. Next, double-stranded cDNA was generated in a second strand cDNA synthesis reaction using DNA Polymerase I and RNAse H. The cDNA fragments were extended with a single "A" base to the 3 ′ ends of the blunt-ended cDNA fragments after which multiple indexing adapters were ligated introducing different barcodes for each sample. Finally an enrichment PCR was carried out to enrich those DNA fragments that have adapter molecules on both ends and to amplify the amount of DNA in the library. Sequence-libraries of each sample were equimolarly pooled and sequenced on 1 lane of a HiSeq2000 flow-cell at 2 × 100 bp. RNA-sequencing data was first cleaned (i.e., removal of adapters and low quality parts) with the fastq-mcf software after which a quality control was performed with FastQC. The reads were then mapped to the Mus musculus (mm10) genome with Tophat2. To identify the gene expression HTSeq-count was used to count the number of reads per gene. These read count numbers were then normalized to the sample size. Differential gene expression analysis was performed with the R-package DESeq2. RNA-seq data from mouse thymic T cell subsets (GSE48138) were used for the analysis of lncRNAs expression in thymocytes (57). The GSE40686 dataset was used to analyze the expression of lncRNAs in wildtype and Foxp3 KIKO Treg, as well as for Foxp3 Chip-seq dataset (27).

Locus Mapping and Analysis
The Flatr 3 ′ terminus was mapped using RNA sequencing of the identical dataset as Foxp3 GFP FACS-sorted Treg RNA sequencing expression analysis and visualized using IGV browser. To map the 5 ′ terminus, RNA was prepared from FACS-sorted Treg from Foxp3 Thy1.1 mice (56) and cells were lysed and stored in TRIzol and RNA was extracted followed by phenol-chloroform extraction. 5 ′ RACE was performed using the RLM-RACE kit (ThermoFisher) as per manufacter's instructions, followed by subcloning using the CloneJET PCR cloning kit (ThermoFisher) and sanger sequencing. The 5 ′ prime terminus mapping was confirmed by analysis of FANTOM(5) CAGE data (58). Analysis of evolutionary conserved regions within the Flatr locus was assessed and visualized using ECR browser (59).

Expression Analysis
T cell subsets were FACS sorted from Foxp3 Thy1.1 mice (56) and were found to be over 98% pure at the postsort purity check. Cells were lysed and stored in TRIzol and RNA was extracted followed by phenol-chloroform extraction. cDNA was prepared using the GoScript Reverse transcription system (Promega) as per manufacturer's protocol. RNAamount were normalized to cell number prior to cDNA generation. Flatr cDNA was amplified using forward primer (5 ′ -ACTGGGACCATGAAAGTGCT-3 ′ ) and reverse primer (5 ′ -TCCTGGCTCAGCAGTGATCT-3 ′ ) using the SYBR Green Real-Time PCR Master mix (ThermoFisher). Expression data was analyzed using the 2 − CT method and normalized to expression levels of Rpl and Ppia.
In Rag KO transfer experiment, naïve T cells (CD4 + CD62L + CD44 − ) from spleen and LN from spleen and lymph nodes of CD45.1 Foxp3 Thy1.1 mice and CD45.2 Foxp3 Thy1.1 Exon2 KO mice were purified using MagniSort-CD4 + T-cell enrichment kit (eBioscience) according to the manufacturer's instructions. Purified cells were FACS sorted and were found to be over 98% pure at the post-sort purity check. Naïve T cells from CD45.1 WT and CD45.2 Exon2 KO mice were mixed at a 1:1 ratio and injected (2 × 10 6 ) i.v. into Rag KO mice. After 8 days mice were sacrificed and cells isolated from spleen and lymph nodes were sorted and analyzed for Treg induction by qPCR and flow cytometry.

Statistics
Statistical analyses were performed with Prism software (GraphPad Software) using the Student t-test. Error bars represent SEM as indicated. p < 0.05 was considered statistically significant.

AUTHOR CONTRIBUTIONS
AB, DF, OB and AC performed the experiments. DF and SB performed bioinformatics analysis. JD, JC, SS, and AL designed and supervised the study. DF, AB, SS, and AL wrote the manuscript. All authors read and approved the manuscript.