A network of DZF proteins controls alternative splicing regulation and fidelity

Abstract

Proteins containing DZF (domain associated with zinc fingers) modules play important roles throughout gene expression, from transcription to translation. Derived from nucleotidyltransferases but lacking catalytic residues, DZF domains serve as heterodimerization surfaces between DZF protein pairs. Three DZF proteins are widely expressed in mammalian tissues, ILF2, ILF3 and ZFR, which form mutually exclusive ILF2-ILF3 and ILF2-ZFR heterodimers. Using eCLIP-Seq, we find that ZFR binds across broad intronic regions to regulate the alternative splicing of cassette and mutually exclusive exons. ZFR preferentially binds dsRNA in vitro and is enriched on introns containing conserved dsRNA elements in cells. Many splicing events are similarly altered upon depletion of any of the three DZF proteins; however, we also identify independent and opposing roles for ZFR and ILF3 in alternative splicing regulation. Along with widespread involvement in cassette exon splicing, the DZF proteins control the fidelity and regulation of over a dozen highly validated mutually exclusive splicing events. Our findings indicate that the DZF proteins form a complex regulatory network that leverages dsRNA binding by ILF3 and ZFR to modulate splicing regulation and fidelity.

Graphical Abstract

DZF proteins form mutually exclusive ZFR-ILF2 and ILF3-ILF2 heterodimers. ZFR and ILF3 use zinc fingers and double-stranded RNA binding domains, respectively, to recognize RNA duplexes and regulate alternative splicing.

Figure 1.

ZFR binds large intronic domains flanking regulated exons. (A) Schematic of ZFR, ILF2, and ILF3 protein features and heterodimerization through DZF domains. (B) Top, crystal structure of the heterodimer interface between ILF2 and ILF3 DZF domains (PDB 4AT7; (14)). Bottom, Alphafold-Multimer prediction of the structure of the heterodimer formed by the ZFR and ILF2 DZF domains (36). (C) Chart indicating fraction of exons identified in rMATS analyses as skipped, included, or not affected by siZFR treatment that are flanked by introns containing ZFR eCLIP peaks. Enrichment of introns containing the indicated number of ZFR eCLIP peaks or greater was assessed by permutation testing with the RegioneR package (** P < 0.01; *** P < 0.001; **** P < 0.0001; (50)). (D) Metagene analysis of ZFR eCLIP peak position and frequency relative to cassette exons skipped or included upon siZFR treatment, generated using RBP-Maps software (47). (E) ZFR eCLIP and ZFR and control knockdown RNA-seq read histograms from MACROH2A1 are shown below gene models corresponding to MXE splicing products. Direction of transcription is left to right. Introns flanking ZFR-regulated exons are indicated by yellow shading. (F) As in E but for a CE splicing event in ESYT2.

Figure 2.

ZFR binds detained introns and regulates the splicing of adjacent exons. (A) Schematic illustrating the relationship between detained introns and ZFR-mediated skipping (top) and inclusion (bottom) of CEs. Empirical P values were determined by permutation testing with the RegioneR package (50). Analysis was limited to CE splicing events in which both exon skipping and inclusion were observed in control samples (min. 5% representation of least abundant isoform). (B) ZFR eCLIP read histograms from the indicated genes containing ultraconserved exons. Gene models corresponding to alternative splice isoforms are shown. (C) Isoform specific RT-qPCR of RNA isolated from ZFR knockout or parental control HEK-293 cells to assess alternative splicing of ultraconserved exons. Relative levels of mRNA isoforms containing the indicated exon junctions were determined by normalization of levels of the included splice isoform to levels of the skipped splice isoform, followed by normalization to control samples (37). Statistical significance was determined by two-tailed Student's t-test, with correction for multiple comparisons by the Holm–Śidák method (n = 2; error bars indicate range; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001).

Figure 3.

ZFR prefers dsRNA or DNA-RNA hybrids over ssRNA. (A) ZFR variants containing (ZFR_long) or lacking zinc-fingers (ZFR_DZF) were co-expressed and purified with ILF2. Binding of the indicated ZFR/ILF2 protein concentrations to fluorescently labeled single-stranded 12-mer oligo-U RNA was assayed by EMSA. (B) As in A, with dsRNA probe formed from 12-mer oligo-U and oligo-A sequences. (C) As in (A), with U-rich (left) and C-rich dsRNA 12-mer probes. (D) As in C, with 24-mer probes. (E) As in D, with U-rich and C-rich 24-mer RNAs duplexed to complementary A-rich and G-rich DNAs.

Figure 4.

DZF proteins ILF2, ILF3 and ZFR form a complex regulatory network. (A) Immunoblot of input (left) and bound (right) fractions from immunoaffinity purification of the indicated stably expressed FLAG-tagged proteins in the presence (+) and absence (–) of MNase. Membranes were probed with antibodies recognizing the endogenous ILF2, ILF3, and ZFR proteins as indicated. All panels are derived from identical exposures of the same membrane. (B) Immunoblot of whole-cell extracts from cells treated with the indicated siRNAs, using antibodies against the endogenous ZFR, ILF2, ILF3 and β-actin. (C) Scatterplot comparing response of CE events meeting rMATS read count cutoffs and regulated in at least one DZF protein knockdown condition (|ΔΨ| > 0.1 and FDR < 0.05) to depletion of ZFR or ILF2. The number of events regulated in both ZFR and ILF2 in each quadrant is indicated in red. Dotted lines indicate ΔΨ ± 0.1. (D) As in (C), comparing the effects of ZFR and ILF3 depletion. (E) ZFR eCLIP, Encode ILF3 eCLIP (45) and DZF knockdown RNA-seq read histograms from the PPP4R1 gene, with gene models indicating mRNA isoforms produced by alternative CE splicing. Direction of transcription is left to right. ΔΨ values from rMATS analyses are indicated. (F) As in (E), for the LRIG2 gene.

Figure 5.

ZFR binds to and functions on introns containing conserved dsRNA elements. (A) Correspondence between CEs skipped or included in siZFR, siILF2, or siILF3 treatment relative to siNS controls (FDR < 0.05 and |ΔΨ > 0.2| in rMATS analysis) and exons adjacent to introns containing the indicated numbers of conserved sequences predicted to form dsRNA (PCCR pairs; (49)). Enrichment of introns containing the indicated number of PCCRs or greater was assessed by permutation testing with the RegioneR package (*P < 0.05; ***P < 0.001; **** P < 0.0001). (B) Correspondence between eCLIP peaks identified in ZFR and PTBP1 eCLIP (red circles; this study) and ENCODE eCLIP studies (black circles except for ILF3 in red; (45)) and predicted conserved intronic dsRNA elements (49). Circle sizes correspond to the number of PCCR sites occupied by the corresponding RBPs. Odds ratios and q values were computed using the LOLA package (53).

Figure 6.

DZF proteins determine regulation and fidelity of MXE splicing. (A) Schematic of possible outcomes of alternative splicing of a gene containing two MXEs. Accurate MXE splicing results in pre mRNAs containing junctions A and B or C and D, while failure of MXE splicing fidelity is reflected by skipping (junction E) or fusion (junction F). (B) Heatmap of validated MXE splicing events in DZF protein knockdowns relative to siNS control. Asterisks indicate events with rMATS |ΔΨ| > 0.1 and FDR < 0.05. Positive values indicate increased usage of the 5′ exon relative to the 3′ exon in the MXE pair under knockdown conditions. (C) Sashimi plot of DNM2 MXE splicing in the indicated knockdown conditions, ZFR eCLIP and Encode ILF3 eCLIP (45) read histograms, and gene model indicating mRNA isoforms resulting from MXE alternative splicing. Mapped reads from three RNA-seq samples per experimental condition were pooled to generate Sashimi plots and junction read counts. Blue lines and read counts correspond to use of the upstream MXE and purple lines and read counts correspond to use of the downstream MXE. Direction of transcription is indicated by chevrons in gene models. Junctions with fewer than 10 reads are not shown. Positions of PCCRs separated by <1000 nt are indicated (49). (D) Sashimi plot of CASK MXE splicing as in (C). Red lines and read counts correspond to isoforms in which all potential MXEs were skipped, and orange lines and read counts correspond to isoforms containing multiple MXEs (fusion events). (E) Heatmap of MXE skipping scores for validated MXE splicing events in DZF protein knockdowns. Statistical significance was determined by two-way ANOVA with Dunnett's correction for multiple comparisons (n = 3; *P < 0.05; see Supplementary Table S5 for fidelity calculations). The predicted susceptibility of the MXE-skipped mRNA to NMD is indicated. (F) As in F but for MXE fusion events.

Figure 7.

ZFR and alternative RNA structures determine ATE1 MXE splicing. (A) Schematic of ATE1 reporter constructs, as designed in (35). ATE1 MXEs 7A and 7B are shown, with positions of duplex-forming elements indicated. R1 and R4 can engage in mutually exclusive base-pairing with R3. Mutated elements are indicated by crosshatching, and relative strength of potential base pairing is indicated by line weight. Right, predicted free energies of the indicated RNA duplexes are shown. (B) Top, quantification of relative isoform usage in semi-quantitative RT-PCR of the indicated mutant minigenes. Statistical significance was determined by two-tailed Student's t-test, with correction for multiple comparisons by the Holm–Śidák method (n = 3; error bars indicate ± SD; *P< 0.05). Bottom, native PAGE of semi-quantitative RT-PCR showing inclusion of exon 7A, 7B or both (F).

Figure 8.

Landscape of DZF protein-mediated alternative splicing regulation. The DZF proteins form a physical and functional network in which ILF2 forms mutually exclusive heterodimers with two dsRNA-binding proteins, ZFR and ILF3. ZFR and ILF3 can have independent, concordant, or opposing roles in regulation of CE and MXE splicing.