TET dioxygenases localize at splicing speckles and promote RNA splicing

Abstract

The dynamic regulation of RNA metabolism plays a crucial part in cellular function, with emerging evidence suggesting an important role for RNA modifications in this process. This study explores the relationship between RNA splicing and the TET dioxygenase activity, shedding light on the role of hm5C (RNA 5-hydroxymethylcytosine), and TET proteins in RNA metabolism. Integrating data from mass spectrometry, AlphaFold structural modeling, microscopic analysis, and different functional assays, including in vitro splicing, TET proteins were found to regulate splicing. We show that TET1, TET2, and TET3 interact with the splicing factors U2AF1 and U2AF2. Interestingly, TET dioxygenases localize in splicing speckles in mammalian and Drosophila cells. TET speckles association was found to be RNA-dependent, but also rely on its interaction with splicing factors. Furthermore, cellular splicing assays revealed that all three TET proteins promote splicing efficiency independent of their catalytic activity. Interestingly, though, the oxidation of m5C to hm5C restores splicing efficiency in vitro. The latter highlights the regulatory role of cytosine modifications in RNA metabolism. These findings provide insights into the complex interplay between RNA modifications and splicing, suggesting a multifaceted role for TET proteins in RNA metabolism beyond their canonical function in the oxidation of 5mC in DNA.

Keywords: 5-hydroxymethylcytosine; RNA modifications; TET dioxygenases; U2AF; epitranscriptomics; splicing; splicing speckles.

Graphical abstract

Figure 1.

Mass spectrometry analysis reveals novel interactions between TET proteins and splicing factors. (A) TET1, (B) TET2, and (C) TET3 mass spectrometry analysis of TET-associated factors. GFP-tagged murine TET proteins or GFP as control were expressed in human HEK293T cells. GFP-tagged proteins were pulled down from whole cell lysates with the GFP-binder beads and subjected to LC-MS/MS. Volcano plots: each circle/triangle represents a protein detected by LC-MS/MS; X-axis depicts differences in protein abundance in the respective pulldowns as log2 Fold change (GFP-TET versus GFP control); factors more enriched in TET-IP have positive Log2 Fold change values (label-free quantification by MaxQuant) Y-axis depicts the negative log10 of the p-value of a student’s t-test (triplicate samples) dashed black lines: significance border (FDR = 0.05, S0 = 2), blue circles/triangles denote splicing factors. A higher value corresponds to a lower p-value, indicating stronger statistical significance. Proteins outside the dashed lines (upper right or left corners) are significantly enriched or depleted, respectively. (D) Venn diagram shows splicing-associated factors that were found to interact with different TET proteins. TET3 and TET2 showed slightly less enrichment of splicing-associated factors compared to TET1, as shown in the list below.

Figure 2.

Screening and validation of TET protein-protein-RNA interactions by structural modeling with AlphaFold 3. (A) the diagram shows the process of RNA splicing and the proteins involved, with a pre – mRNA harboring exons and introns (light gray) which are removed during splicing. The splice sites at the start ‘GU’ and the end ‘AG’ of the intron are shown, together with the branch point ‘A.’ The polypyrimidine tract located upstream of the 3’ splice site is represented as ‘YYYYY.’ U1 and U2 snRnps, components of the spliceosome, are responsible for recognizing splicing sites. U2AF1 and U2AF2 proteins assist in binding to the 3’ splice sites and branch points. SR proteins (like SC35) bind to exonic splicing enhancers (ESE) to enhance splicing accuracy. (B) schematic representation of U2AF1, U2AF2, and SC35 harboring different RRM domains (RNA recognition motifs), and TET proteins, with their domains, represented: CXXC (zinc finger domain), the conserved catalytic domain in the C-terminus, including the cysteine-rich domain (CRD) and the double-strand beta-helix domain (DSBH), with an insert in between. In addition to the full protein sequences, different fragments of the aminoacid sequences and domains were used to predict protein-protein-RNA interaction, as indicated in the scheme. (C) heatmap showing the ipTM score (interface predicted template modeling score) obtained for AF structure models in the screening for all interactions tested between TET proteins and the splicing factors U2AF1, U2AF2, and SC35. Labels on the ×and y axes indicate the paired protein fragments for structural modeling, including RNA in the prediction. White tiles indicate pairs that were not subjected to structural modeling. (D) the highest-scored structural models were obtained for TET1-U2AF2-RNA, (E) TET2-U2AF1-RNA, and (F) TET3-U2AF predictions. The full structural models are shown from different perspectives (top part), and magnifications are shown at the bottom. In the magnification, only residues within a maximum distance of 5 angstroms to U2AF are shown. (G) Co-immunoprecipitation analysis of TET proteins interaction with splicing factors. The catalytic domain (CD) of TET1, TET2, and TET3, or GFP as a control, were co-immunoprecipitated using GFP-binder beads. Pull-down fractions were analyzed by western blot using anti-FLAG antibody (left, U2AF1), anti-HA antibody (center, U2AF2), or anti-RFP (right, SC35). Inputs ‘I’ and bound ‘B’ fractions are shown. Full blot images and replicates are shown in Figure S2. Uncropped and unprocessed images can be found in Supplementary data S5.

Figure 3.

TET1 localizes at splicing speckles, colocalizing with splicing factors in mouse and Drosophila tissues. (A) immunostaining of TET1 and SC35 in various differentiated mouse cell lines with quantified colocalization represented by the H-coefficient. J1 ESCs, embryoid bodies (EB) derived from J1 ESCs, mouse embryonic fibroblasts (MEF), and mouse tail fibroblasts (MTF) were stained for TET1 (green) and SC35 (red). DNA was counterstained with DAPI (blue). To distinguish the undifferentiated state, Oct 4 was used as a pluripotency marker. The Oct4 channel was omitted for the merge. Colocalization between TET1 and DAPI (black) as well as TET1 and SC35 (dark gray) was quantified using the H coefficient (n = 10 cells) and is shown as a barplot on the right-hand side. (B) MTF cells were immunostained for tagged U2AF proteins and endogenous TET1 (green). Ectopically expressed U2AF1-FLAG or U2AF2-HA (red) were visualized using anti-HA or anti-FLAG antibodies. DNA was counterstained with DAPI (blue). The barplot illustrates the H-coefficient (n = 10) of TET1 and DNA (black) or U2AF proteins (gray). Scale bar: 5 µm. (C) TET1 localizes to SC35-positive speckles in vivo. Paraffin sections of murine testis, lung, and spleen were immunostained with antibodies against TET1 and SC35. DNA was counterstained with DAPI. Line-profile analysis of selected regions denotes TET1 colocalization with SC35 and anti-colocalization with DNA. (D) colocalization of dTET and nuclear speckles in Drosophila salivary gland cells. Different HA-tagged dTET isoforms (short (row 1), long (row 2), and catalytically dead (cat. dead) short isoform (row 3) were expressed from UAS inserts by elav^C155GAL4, and compared with wild type (raw 4). Third instar salivary glands were stained for speckles with anti-SC35 antibodies and dTET with anti-HA antibodies. The overlay of dTET and SC35 are shown in the merge. The boxplot on the right shows the quantification by image analysis of dTET accumulation in speckles. Values exceeding 1 (dashed black line) denote colocalization with SC35. Scale bar: 15 µm. Error bars represent the standard deviation.

Figure 4.

RNA-seq analysis of alternative splicing in Drosophila wild-type versus TETnull. (A) barplot showing the inclusion level difference of exon skipping genes between wild-type (wt) and TETnull samples. (B) barplot showing the inclusion level difference of intron retention between wild-type (wt) and TETnull samples. (C) barplot showing the inclusion level difference of alternative 5’ splice sites between wild-type (wt) and TETnull samples. (D) barplot showing the inclusion level difference of alternative 3’ splice sites between wild-type (wt) and TETnull samples. For (A), (B), (C), and (D), genes showing significant changes are shown ( > 30% changes and p < 0.01, Supplementary data S4), ordered from lowest to highest p-value from left to right. The inclusion level measures the frequency of a particular event (exon skipping, intron retention, 5’ or 3’ alternative splicing sites) in the final mRNA transcript, represented as a value between 0 (always skipped) and 1 (always included). The inclusion level difference is calculated as the inclusion level in wild-type minus the inclusion level in TETnull condition. Blue bars with positive values indicate an increased inclusion in the wild type, while red bars with negative values indicate an increased inclusion in TETnull. (E) gene ontology (GO) analysis of differentially expressed genes (Supplementary data S3) is shown on the left for biological process and on the right for molecular function, with enrichment indicated in blue color and -log(false discovery rate) by circle size.

Figure 5.

TET1 associates with RNA in splicing speckles and is recruited by an mRNA mimic in vivo. (A) live cell RNase extraction of MTF cells. On the left, the scheme shows the experimental pipeline to study TET1 speckle association upon RNase a treatment. Mouse tail fibroblasts were seeded on gelatine-treated glass coverslips and 24 hours later pre-extracted with 0.01% Triton in PBS, followed by incubation with 1 mg/mL RNase a for seven minutes. Afterward, cells were fixed with paraformaldehyde and immunostained with antibodies against TET1 and SC35. Representative confocal microscopy images are shown in the center. As treatment control, similarly treated cells were stained with propidium iodide (PI), and the respective signal was quantified by high-content microscopy (n > 1000 cells). Representative images of PI staining are shown on the right. Scale bar = 5 µm. (B) TET1-CD is recruited with a mRNA mimic in live-cell rF3H assay (RNA fluorescent three-hybrid assay). Scheme of the rF3H assay. Briefly, genetically modified BHK cells with many copies of the LacO were transfected with an RNA trap, an mRNA mimic, and pPABPCI-mCherry or pmCherry-TET1-CD. The RNA trap is fused to LacI (for LacO targeting) and EGFP (for visualization of the LacO locus). After live-cell imaging of transfected cells, the interaction between the mRNA mimic and the mCherry proteins is quantified by image analysis. (C) boxplot showing the quantification of the relative accumulation of mCherry proteins at the LacO. Two independent replicates were performed. N-number = 26–37. Statistical significance was tested with a paired two-sample wilcoxon test (n.S., not significant, is given for p-values ≥0.05; one star (*) for p-values < 0.05 and ≥ 0.005; two stars (**) is given for values < 0.005 and ≥ 0.0005; three stars (***) is given for values < 0.0005). P-values are shown in Supplementary Table S5. (D) Representative images of the rF3H assay for each condition. The area of the LacO locus is highlighted in the mCherry channel where the accumulation of mCherry proteins in the LacO is visible. Scale bar = 5 µm. (E) heatmaps showing ipTms and pTms scores of AlphaFold protein–protein interaction predictions between TET proteins and splicing factors without RNA. Higher-ranked predictions shown in Fig. 2 were repeated without the short RNA, to show RNA-specific effects on the prediction score. (F) structure showing AlphaFold 3 prediction model for TET1-CD and DNA (LINE1 5’ UTR region) versus TET1 interaction with RNA (intron-exon transition region).

Figure 6.

TET proteins and m5C to hm5C oxidation promote splicing. (A) scheme showing the chimeric vector used for the splicing assay based on the NMD system (non-sense mediated mRNA decay). E7, E8, and E9 correspond to exons 7, 8, and 9 of the HNRNPDL gene (heterogeneous nuclear ribonucleoprotein D-like), represented by black/gray rectangles. Introns between the exons are represented by a black line. The red star indicates premature translation termination codons (PTCs) present in the chimeric ‘pre-mRNA.’ If splicing takes place, the PTC is removed, and translation of the RNA, with GFP at the 5,’ occurs. In the absence of splicing, the PTC leads to the degradation of the ‘pre-mRNA.’ Stop codon ‘TAA’ at the end of the GFP coding sequence. (B) boxplots showing the normalized EGFP intensity in different mEscs (wild type and TET triple knockouts) transfected with the NMD vector and the different TET constructs. The same experiments were performed in Dnmt triple knockout cell lines, and quantification is shown in (C). (D) in vitro splicing assay. On the top is the scheme of the vector used for in vitro splicing. The pre-mRNA encoding adenoviral major late gene sequence exons 1 and 2 interrupted by an intron was created by T7 RNA polymerase run-off transcription. MS2 bacteriophage coat protein binding aptamer (not relevant for this study). For this reaction, the only cytidine sources were either CTP, m5CTP, or hm5CTP. 32p-labeled transcripts were phenol-chloroform purified, and similar RNA amounts were subjected to in vitro splicing reactions with HeLa nuclear extracts (0, 5, 10, 30, 60, 90 min). A diagram representing the different splicing intermediates of this construct is shown on the bottom left. Samples were taken at the indicated time points and analyzed on polyacrylamide gels.