QKI ensures splicing fidelity during cardiogenesis by engaging the U6 tri-snRNP to activate splicing at weak 5' splice sites

Abstract

During organogenesis, precise pre-mRNA splicing is essential to assemble tissue architecture. Many developmentally essential exons bear weak 5' splice sites (5'SS) yet are spliced with high precision, implying unknown yet active splicing fidelity mechanisms. By combining transcriptome and alternative splicing profiling with temporal eCLIP mapping of RNA interactions across development, we identify the RNA-binding protein QKI as an essential direct regulator of splicing fidelity in key cardiac transcripts. Although QKI is dispensable for cardiac specification, its loss disrupts sarcomere assembly despite intact expression of sarcomere mRNAs through exon skipping and nuclear retention of mis-spliced RNAs. QKI-dependent exons in essential cardiac genes have weak 5'SS and frequently show poor complementarity with U6 snRNA. We show that QKI directly interacts with U6 snRNA using an overlapping interface to its traditional intronic binding activity, securing U4/U6·U5 tri-snRNP to ensure splicing fidelity. Thus, QKI exemplifies how context-aware RBPs enforce splicing fidelity at structurally vulnerable splice sites during organogenesis.

Extended Data Fig. 1:. QKI is dynamically regulated and predominantly nuclear localized during cardiac differentiation of hESCs

a, RNA-seq quantification of individual QKI isoforms (QKI-202, −203, −205, −206) across cardiac differentiation stages (FPKM values from Frank et al., 2019), showing dynamic isoform expression patterns. b, qRT–PCR analysis of total QKI transcript levels at seven key time points during directed cardiac differentiation of hESCs, showing stage-resolved dynamic regulation. Data represent mean ± SEM from n = 3 biological replicates. c, Immunoblot of total QKI protein during cardiac differentiation from D0 to D14, confirming stage-specific regulation. α-Tubulin serves as a loading control. d, Immunofluorescence staining for QKI and stage-specific lineage markers at matched differentiation stages: OCT4 (pluripotent hESCs), T (primitive streak/mesoderm), HAND2 (cardiac mesoderm), ISL1 (cardiac progenitors), and cTnT (cardiomyocytes). QKI expression is observed throughout all stages and localizes near-exclusively to the nucleus. e, Western blot analysis of nuclear and cytosolic fractions in D8 cardiomyocytes, showing enrichment of QKI in the nuclear compartment. DNMT3B and α-Tubulin serve as nuclear and cytosolic markers, respectively. Error bars represent ±SEM; p-values are calculated using Student’s t-test; biological replicates n = 3

Extended Data Fig. 2:. Generation and validation of QKI knockout hESCs

a, Schematic of the genomic QKI locus in humans along with the CRISPR/Cas9 strategy targeting exon 3 of QKI. b, Illustration of QKI knockout alleles in the indicated hESC clones, each harboring unique homozygous frameshift mutations disrupting the open reading frame of QKI. c, Immunoblot confirming loss of QKI protein in three independent knockout clones; α-Tubulin serves as loading control. d, qPCR validation of pluripotency markers (OCT4, NANOG, SOX2, KLF4, MYC, DNMT3B) showing no significant differences between WT and QKI KO hESCs clones (normalized to RPL37A). e, Immunoblot confirming the expression of pluripotency markers OCT4 and NANOG in QKI KO clones, comparable to the isogenic WT counterpart. f,g,h, Immunofluorescence analysis of pluripotency markers in WT and QKI KO hESCs. Representative images showing OCT4, NANOG, and SOX2 expression in wild-type and QKI knockout hESCs. Note the reduced SOX2 protein levels in QKI knockout cells compared to wild type, despite intact expression of OCT4 and NANOG. Error bars represent ±SEM; p-values are calculated using Student’s t-test; biological replicates n = 3

Extended Data Fig. 3:. QKI knockout does not impair trilineage differentiation

a, Schematic of the directed mesoderm differentiation protocol. Below, qRT-PCR analysis of mesoderm markers (TBXT, MESP1, CDX2, FOXF1) in WT and three independent QKI KO hESC clones. All KO clones exhibit robust induction of mesoderm markers, indicating that QKI loss does not impair mesoderm commitment. b, Schematic of the directed endoderm induction protocol. qRT-PCR analysis of definitive endoderm markers (SOX17, FOXA2, GSC, CDH2) following directed endoderm differentiation of WT and QKI KO hESCs. All three KO clones show comparable induction of endodermal genes, demonstrating intact endodermal potential in the absence of QKI. c, Schematic of the directed ectoderm induction protocol. qRT-PCR quantification of ectoderm lineage markers (PAX6, OTX2) during directed ectoderm differentiation. All QKI KO clones activate canonical ectoderm genes to levels similar to WT, confirming that QKI is dispensable for ectoderm specification. Error bars represent ±SEM; p-values are calculated using Student’s t-test; biological replicates n = 3

Extended Data Fig. 4:. Validation and specificity of QKI eCLIP libraries

a, Immunoblot confirming successful immunoprecipitation (IP) of endogenous QKI from lysates under eCLIP conditions from hESCs (D0), cardiac progenitor (D4), and cardiomyocyte (D8) stages. α-Tubulin serves as loading control. b, Bars indicate fraction of significant peaks (p ≤ 10⁻³, fold-enrichment ≥ 8) from individual replicate eCLIP experiments, which overlap indicated RNA region annotations. c, Clustering analysis comparing QKI eCLIP replicates from this study across differentiation stages with publicly available QKI eCLIP datasets (ENCODE) and unrelated RBPs (IGF2BP1, IGF2BP3, and RBFOX2). QKI replicates cluster tightly together and separates clearly from unrelated RBPs, confirming both the quality and target specificity of QKI eCLIP across stages. Additionally, stage-resolved QKI datasets exhibit progressive differences, indicating developmental regulation of QKI binding. d, Histogram of peak fold-enrichment in D0 QKI eCLIP for peaks identified in D8 QKI eCLIP (as shown in Fig. 2i). Genes were separated as ‘cardiomyocyte-specific’ (five-fold or higher expression in D8 than D0) or ‘other genes’.

Extended Data Fig. 5:. QKI loss leads to nuclear sequestration of mis-spliced cardiac transcripts

a-d, Representative read density tracks showing RNA-seq and QKI eCLIP signal for differentially spliced exons in ACTN2, PICALM, RBFOX2, and CAPZB across cardiac differentiation. e, Representative RT-PCR assays to validate splicing defects for indicated QKI-dependent targets using (left) minigene reporters and corresponding endogenous transcripts (NPHP3, NIN, CLASP2, and RYR2) in HEK293T cells following QKI knockdown, or (right) C2C12 cells.

Extended Data Fig. 6:. QKI knockdown leads to mis-splicing and nuclear retention of RNA for essential cardiac factors.

a-b, Whole-gene read density plots for (a) ACTN2 and (b) NEBL, including QKI eCLIP (and paired input), D0, D4, and D8 polyA RNA-seq, and D8 nuclear fraction rRNA-depleted RNA-seq. c, Scatter plot indicates lack of correlation between RNA and protein expression changes in D8 QKI knockout versus wild-type cardiomyocytes. Sarcomere genes (Gene Ontology ID: 0030017) are indicated in red. d-e, Distribution of indicated transcript categories in (d) nuclear and (e) whole-cell RNA-seq between WT and QKI KO cardiomyocytes. Transcripts containing QKI KO–excluded exons show modest but significant nuclear enrichment compared to all cassette exons (P = 7.5 × 10⁻⁷, two-sided Wilcoxon rank-sum test), consistent with the nuclear retention of mis-spliced RNAs. f, Maximum intensity projections of TTN and RYR2 smRNA FISH combined with immunofluorescence for QKI in human cardiomyocytes. TTN transcripts form discrete nuclear foci that wrap around QKI-enriched puncta, suggestive of localized accumulation at TTN transcription sites. In contrast, RYR2 transcripts form distinct foci that do not colocalize with either TTN or QKI, indicating that QKI–TTN condensates do not serve as general hubs for all QKI-dependent transcripts. This spatial segregation underscores the transcript-specific nature of the nuclear interactions of QKI in cardiomyocytes.

Extended Data Fig. 7:. U6 complementarity determines QKI dependency at weak 5ʹSS

a, MaxEnt 5ʹSS strength scores of all cassette exons, QKI-regulated cassette exons (ΔPSI ≥ 0.1, FDR ≤ 0.01), and QKI-regulated exons bound by QKI with strong inclusion (ΔPSI ≥ 0.5), revealing a significant shift toward weaker 5ʹSS in QKI-dependent targets. b, Schematic of 5ʹSS with indicated base-pairing potential to U1, U5, and U6 snRNAs for exons in cardiac-related genes that lose exon inclusion upon QKI knockout. c, For indicated exon classes, bars show the fraction of exons with their 5ʹSS containing the indicated number of complementary nucleotides to U5 snRNA. d, Schematic of the RYR2 MUT3 minigene, which harbors a +2C mutation in the context of the inherent good complementarity to U6 snRNA in the RYR2 5ʹSS. e, Representative RT-PCR image of the RYR2 MUT3 minigene splicing profile relative to WT, MUT1, and MUT2 minigenes that have been described previously. RYR2 MUT3 results in consistent inclusion of exon 75, similar to MUT1, indicating a context-specific significance for specific nucleotides at +2 in QKI-dependent exon inclusion. f, Schematic of NIN minigenes harboring a +2T>G (MUT2) and a +2T>A (MUT3) 5ʹSS mutations to test the significance of nucleotide identity +2T in the NIN 5ʹSS context. g, RT-PCR image showing splicing profiles of NIN MUT2 and MUT3 relative to WT and MUT1 minigenes. In this context, altering the nucleotide identity at +2 leads to a complete loss of inclusion in a QKI-independent manner. h, As in f but with splice sites harboring mutations that improve complementarity to U6 snRNA. i, RT-PCR showing comparative splicing profiles of NIN MUT5 and MUT6 to the NIN WT minigene. Improving complementarity to U6 snRNA, in this context, does not enhance inclusion of NIN exon 18, indicating a context-specific requirement for +2T in the NIN 5ʹSS. j, As in Fig. 5j, predicted U5 complementarity was scored for QKI knockout- or knockdown-excluded exons from our profiling of D8 and D0 QKI knockout cells, our profiling of QKI siRNA knockdown in 293T cells, and published QKI knockdown in BEAS2B and neuronal stem cell (NSC) cells^, , with no enrichment seen for weak U5 complementarity.

Extended Data Fig. 8:. Molecular basis of QKI–U6 snRNA interaction

a, Representative 1H‒15N HSQC NMR titrations of recombinant 15N-labeled QKI KH‒QUA2 domain with increasing concentrations of U6 RNA. Chemical shift perturbations confirm specific interaction with the 5ʹ region of U6, consistent with structured RNA recognition. Amino acid residues interacting with the RNA molecules (disappearing peaks and peaks with chemical shift perturbation > 0.05) are highlighted in the structural models (pdb: 4jvh). b, Molecular docking model showing the 3D arrangement of U6 snRNA interacting with the QKI dimer. U6 wraps around QKI, positioning its 5ʹ end along the positively charged RNA-binding groove, consistent with the binding footprint revealed by NMR. c, Electrostatic surface potential of QKI dimer highlighting a positively charged groove that accommodates U6 snRNA. The U6 backbone follows a trajectory consistent with the modeled RNA-binding interface.

FIGURE 1 |. QKI uncouples cardiac fate specification from sarcomere formation and contractile function

a, Schematic summarizing the cardiac differentiation efficiency along the WNT-BMP morphogen gradient (“cardiac corridor”) for wild-type (WT) and QKI knockout (KO) hESCs. QKI KO cells efficiently generated cardiomyocytes, albeit with shifted optimal WNT requirements. (n = 3 independent biological replicates). b-e, QKI KO cardiomyocytes show robust expression of core cardiac transcription factors but severely disrupted sarcomere organization. Immunofluorescence staining for cardiomyocyte markers (b) ACTN2 and cardiac troponin T (cTnT) and indicated cardiac transcription factors (c) NKX2.5, (d) GATA6, and (e) MEF2C in WT and QKI KO cardiomyocytes. Scale bars, 20 μm. f, Second harmonic generation (SHG) imaging of sarcomere structure in WT and QKI KO cardiomyocytes reveals lack of organized contractile structures in QKI KO. Scale bar, 10 μm. g, Transmission electron microscopy (TEM) showing sarcomeric architecture in WT and QKI KO cardiomyocytes at day 8. QKI KO cardiomyocytes lack defined Z-lines and regular sarcomere units. Scale bar, 500 nm. h, Schematic of cardioid generation method. i, Representative brightfield images of WT and QKI KO cardioids on day 8 showing comparable size and morphology. j, Immunofluorescence staining for TTN and ACTN2 in WT and QKI KO cardioids at day 10 shows disorganized sarcomere structure in QKI KO organoids. Scale bar, 20 μm. k, Illustration of the strategy used to re-express QKI in QKI-KO cardiomyocytes using PiggyBac-based strategy driving QKI-mVENUS using doxycycline (DOX) inducible promoter. Right: immunoblot confirming QKI re-expression at WT levels. l. Schematic of the generation and metabolic maturation of cardiomyocytes to deplete any carry over of proliferating cells post differentiation. m. Rescue of sarcomere assembly following DOX-induced QKI re-expression in QKI KO cardiomyocytes. Immunostaining shows progressive restoration of sarcomere structure upon QKI re-expression in QKI KO cardiomyocytes. Cells were differentiated without DOX, metabolically selected, replated and induced with DOX for 96 hours. Sarcomeres reappear 24–48 hours post-induction. Scale bar, 20 μm. n, Heatmap showing transcriptome-wide expression profiles in wild-type and QKI KO cells (average FPM across 3 replicates) at key stages of directed cardiac differentiation—pluripotency (D0), cardiac progenitor (D4), and cardiomyocyte (D8). Despite loss of QKI, global transcriptional programs proceed largely unchanged between WT and QKI KOs in the indicated stages. o, QKI loss does not impair functional gene activation during cardiac differentiation. Gene Ontology (GO) enrichment analysis of differentially expressed genes in wild-type and QKI KO cells at key stages of differentiation reveals near-identical, stage-specific activation of gene programs associated with cardiomyocyte specification, sarcomere assembly, and cardiac contractility. p. Heatmap showing that genes specifically induced during cardiomyocyte differentiation exhibit near-identical expression dynamics in wild-type and QKI knockout cells at day 8, indicating intact transcriptional activation of cardiac identity programs despite QKI loss. (n = 3 in all experiments unless otherwise stated).

FIGURE 2 |. Stage-resolved eCLIP-seq maps transcriptome-wide QKI binding landscapes during human cardiac specification

a, Schematic of stage-resolved eCLIP-seq across human cardiac differentiation from pluripotency (D0), cardiac progenitors (D4), to cardiomyocytes (D8). b, Total number of reproducible, significantly enriched QKI binding eCLIP peaks per stage normalized against the paired SMInput (n=3 biological replicates at each stage of differentiation, fold change ≥ 2; q-value ≤ 0.05). Pie chart shows the genomic distribution of QKI peaks across introns, 3ʹUTRs, coding regions, and other regions across all stages versus total nucleotides covered in Gencode (v19) transcripts. c, Metagene plot shows intronic QKI eCLIP peak distribution for significant peaks (p ≤ 10⁻³, fold-enrichment ≥ 8) from example individual replicate eCLIP, showing enrichment near 5ʹSS in all indicated time points of cardiac differentiation. d, Metagene plot for cardiomyocyte QKI eCLIP peaks along a meta-mRNA shows enrichment in the 3’UTR. e, Top enriched QKI-binding motif (ACTAAY) and bipartite consensus (ACTAA[C/T]N₁–₂₀TAA[C/T]) derived from HOMER analysis of IDR eCLIP peaks, with fold-enrichment above background for the indicated time points f, Percent of peaks at each stage containing the canonical bipartite QKI motif. g, Reverse transcription stop-site (RT-stop) analysis across QKI-bound motifs reveals increased termination at −1, 0 and −7/−8 positions relative to QKI motif locations. h, Peak read density in QKI eCLIP (normalized across both timepoints) for QKI eCLIP peaks identified in D8, separated by their location in (bottom) D8-specific genes (D8 expression more than 5-fold increased from D0) or (top) other genes. i, GO enrichment analysis of QKI-bound mRNAs at each stage, showing a shift from early neuronal and signal transduction components (D0, D4) to sarcomere and contraction-related categories at D8. j, Heatmap of representative QKI-bound transcripts grouped by function. At D8, QKI selectively binds mRNAs encoding cardiac-relevant RBPs such as RBM20, RBM24, RBFOX2, MBNL1/2, as well as sarcomeric and calcium-handling proteins, including TTN, RYR2, ACTN2, TPM1, NEBL, and CAMK2D.

FIGURE 3 |. QKI loss leads to altered RNA splicing, RNA nuclear retention, and protein abundance of sarcomere genes

a, Bar plot showing the number of significant splicing events (FDR ≤ 0.01, ΔPSI ≥ 10%) identified in QKI KOs across D0, D4, and D8, categorized into skipped exon (SE), mutually exclusive exon (MXE), retained intron (RI), alternative 3ʹ (A3SS), and 5ʹ splice sites (A5SS). b, Dot plots highlighting differentially included (red, ΔΨ > 0) and excluded (blue, ΔΨ < 0) cassette exons in QKI KO versus WT across differentiation stages. c, Bar plots showing the number of significant alternative splicing events detected per timepoint in QKI knockout. d, Gene Ontology (GO) term enrichment analysis of cassette exon (left) exclusion and (right) inclusion events in QKI KO events across D0, D4, and D8 stages indicates D8 QKI-dependent exon inclusion at D8 enriched in contractile and sarcomeric programs while exclusion events are enriched in neuronal and migration-related genes. e,f, Read density of QKI eCLIP and RNA-seq at alternative exons in (e) RYR2 and (f) NIN with significant exclusion in QKI KO versus WT indicates QKI binding proximal to the alternative exon. Grey PSI (Percent Spliced In) indicates quantitation from less than 10 junction-spanning reads. g, Volcano plot of quantitative whole cell LC-MS/MS proteomics (WT vs. QKI KO cardiomyocytes, n = 3 biologically independent replicates) indicates depletion of key sarcomeric proteins among QKI targets. P values are calculated using Student’s t test; biological replicates n = 3 h, Read density of QKI eCLIP and nuclear RNA-seq at TTN in WT and QKI KO cardiomyocytes indicates QKI binding near (top) 5ʹSS of exons with QKI-dependent inclusion and (bottom) nuclear accumulation of mis-spliced TTN transcripts in QKI KO cardiomyocytes. i, Shown are all sarcomere genes with differential (log₂(FC) ≤ −0.5) protein abundance in QKI KO versus WT, along with fold-change observed in proteomic, whole-cell polyA RNA-seq, and nuclear total RNA-seq, along with indication of the presence of eCLIP peaks and QKI KO-dependent splicing changes. eCLIP indicates the maximum fold-enrichment across 3 biological replicates, with reproducible peaks across all replicates indicated with black outline. j, Histogram indicates the increase in nuclear versus total RNA abundance in QKI KO versus WT for sarcomere genes. Significance was determined by Kolmogorov-Smirnov test. k, Representative immunofluorescence images showing near-complete loss of TTN and ACTN2 protein in QKI KO cardiomyocytes, despite transcriptional induction. Scale bar, 10 μm.

FIGURE 4 |. QKI controls a conserved splicing regulatory program through binding to the 5ʹSS- and 3ʹSS-adjacent intronic regions

a, ‘Splicing maps’ for QKI generated by integrating stage-matched eCLIP and splicing data reveal positional logic for QKI-mediated regulation. QKI binding near the 5ʹSS promotes exon inclusion, whereas binding near the 3ʹSS promotes exon exclusion. b, Read density of QKI eCLIP and RNA-seq at alternative exons show examples of 5ʹSS-proximal binding at exon 4 of AKAP9 leading to decreased (QKI-dependent) inclusion upon QKI loss. c, Reactome pathway analysis and Gene Ontology (GO) enrichment analysis of QKI-dependent splicing events in cardiomyocytes, stratified by QKI binding position and splicing outcome. Exons bound by QKI near the 5ʹSS and excluded upon QKI loss (QKI-dependent inclusion) are highly enriched for genes involved in sarcomere organization and cardiac contractility. In contrast, exons bound near 3ʹSS and aberrantly included in QKI KO cells (QKI-dependent exclusion) are enriched for neuronal and migration-associated genes. d,e, Schematic of classification logic used to assign QKI-bound, dependent (d) exon inclusion or (e) exon exclusion events into “consistent,” “discordant,” or “expression-driven” categories by comparing splicing and binding at D0 and D8. (center) Splicing maps generated for the indicated categories comparing D0 and D8 indicate that consistent events show consistent QKI binding in D0 and D8, whereas discordant events appear to be indirectly regulated by QKI. D8 exclusive exon inclusion events (green) show enriched QKI binding at D8 but not D0, indicating QKI control of splicing but not underlying gene expression at these events.

FIGURE 5 |. QKI enhances exon inclusion at suboptimal 5ʹ splice sites with poor U6 complementarity

a, Schematic indicating key steps of snRNA recognition of the 5ʹSS, with insets indicating complementarity between canonical 5ʹSS and (top) U1 or (bottom) U6 snRNA. b, Illustration of base pairing interactions of U1, U5, and U6 snRNAs at a canonical 5ʹSS and five example exons with QKI-dependent inclusion and QKI eCLIP peaks near the 5ʹSS. Indicated splice site scores are from Maximum Entropy (MaxEnt) modeling. c, For indicated exon classes, bars show the fraction of exons with their 5ʹSS containing the indicated number of complementary nucleotides to U6 snRNA. Significance of QKI-regulated exon enrichment for weak U6 base pairing was calculated by Fisher’s Exact Test comparing high (3–5 nt) versus low (0–2 nt) complementarity. d, Schematic of wild-type and mutant NIN 5ʹSS minigenes, illustrating enhanced U6 base-pairing in MUT1 to test the functional relevance of U6 complementarity in QKI-dependent exon inclusion. Reporters spanning endogenous exons and flanking introns were transfected into HEK293XT cells with QKI siRNA knockdown or non-targeting control. e, Representative image of minigene RT-PCRs using wild-type (WT) and U6-optimized (MUT1) NIN reporters with or without adjacent QKI-binding motifs, either in the presence or absence of QKI. Bottom, immunoblot confirming QKI knockdown; GAPDH serves as a loading control. f, Bar graph representing densitometric quantification of minigene RT-PCRs in (f) for NIN, n=3. Significance was determined by two-way ANOVA; p < 0.01(**), p ≥ 0.05 (ns) g, Schematic of wild-type (WT), U6-enhanced (MUT1), and U6-weakened (MUT2) RYR2 5ʹ splice site minigene constructs, designed to test the functional relevance of U6 base-pairing strength in QKI-dependent exon inclusion. MUT1 improves, and MUT2 reduces, U6 complementarity relative to WT. h, Representative image of minigene RT-PCRs using wild-type (WT) and U6-modified (MUT1, MUT2) RYR2 reporters with enhanced or reduced U6 complementarity, respectively, either in the presence or absence of QKI. Bottom, immunoblot confirming QKI knockdown; Actin serves as a loading control. i, Bar graph representing the densitometric quantification of the minigene RT-PCRs in (f) for RYR2, n=3. Significance was determined by two-way ANOVA; p < 0.001(***), p ≥ 0.05 (ns). j, As in (c), predicted U6 complementarity was scored for QKI knockout- or knockdown-excluded exons from our profiling of D8 and D0 QKI knockout cells, our profiling of QKI siRNA knockdown in 293T cells, and published QKI knockdown in BEAS2B and neuronal stem cell (NSC) cells^, . Across diverse human cell types and datasets, exons excluded upon QKI loss consistently harbor weak 5ʹSS with reduced U6 base-pairing potential, indicating a conserved requirement for QKI in stabilizing U6 engagement at structurally suboptimal 5ʹSS. Significance was determined by Fisher’s Exact Test.

FIGURE 6 |. QKI directly engages U6 snRNA independent of Its intronic binding activity

a, QKI eCLIP-seq reveals robust and selective binding to U6 snRNA across all stages of human cardiac differentiation (D0, D4, D8). Significance was determined by the Mann-Whitney U test. b, QKI shows negligible enrichment for other spliceosomal snRNAs (U1, U2, U4, U5, U6ATAC, U11, U12, U4atac) compared to U6 snRNA across all stages of human cardiac differentiation. Significance was determined by the Mann-Whitney U test. c, Comparative enrichment analysis of QKI on U6 snRNA (log₂ fold change over size-matched input) relative to 155 ENCODE-profiled RBPs, showing selective U6 engagement (comparable to canonical U6-binding factor SMNDC1) but not to U1 d, QKI shows selective binding to U6 snRNA, but not to the minor spliceosomal U6ATAC snRNA, in contrast to SMNDC1, which binds to both snRNAs. e, QKI binds specifically to the 5ʹ region of U6 snRNA across all stages and replicates, distinct from canonical U6-binding splicing factors (e.g., PRPF8, RBM22), which predominantly bind the 3ʹ region. f, Domain architecture of QKI and schematic of the KH domain double mutant (K120A/R124A) used to assess functional contributions of canonical RNA binding to U6 engagement. g, Normalized read density in eCLIP of transgenic MYC-tagged wild-type and KH domain mutant QKI in HEK293T cells, shown for eCLIP peaks identified in independent experiments with anti-QKI antibody in HEK293T. h, Quantification of (g) shows a significant loss of eCLIP signal with the KH domain mutant. i, Enrichment of QKI binding at indicated snRNAs in cells expressing WT or KH mutant QKI indicates that U6 binding is preserved upon mutation of the canonical RNA recognition domain. j, Base-resolution eCLIP mapping of QKI binding to U6 snRNA reveals that the selective enrichment at the 5ʹ stem-loop and adjacent single-stranded region critical for 5ʹSS engagement is preserved in the QKI KH domain mutant. k, RT-PCR using a NIN exon 18 minigene shows that the KH-domain mutant QKI fails to restore splicing despite retaining U6 binding. l, Schematic showing secondary structure of human U6 snRNA with key regions including the 5ʹ stem-loop, 5ʹSS interacting domain, internal stem-loop, and telestem. Colors indicate regions used for NMR analysis. m-o, NMR titration of recombinant QKI KH–QUA2 protein with in vitro transcribed U6 RNA fragments. Residues affected based on chemical shift perturbations overlap with the canonical RNA-binding surfaces. m, The U6 fragment containing both the 13-mer sequence with a partial QRE motif and the 5ʹ stem-loop induces the strongest chemical shift changes, consistent with an extended RNA-binding interface and a higher affinity interaction. n, The 13-mer alone also triggers pronounced shifts, yet to a lesser extent compared to (m). o, In contrast, the 5ʹ stem-loop alone causes only minor perturbations, consistent with a weak interaction. Error bars represent ±SEM; p-values are calculated using Student’s t-test; biological replicates n = 3

FIGURE 7 |. QKI integrates into native spliceosomal complexes and enforces U6 engagement at weak 5ʹ splice sites

a, Schematic of the RNA-aware nuclear complexome and IP-proteome workflows. UV-crosslinked nuclei from WT and QKI KO cardiomyocytes were subjected to BN-PAGE and LC-MS/MS with or without RNase treatment (complexome), or immunoprecipitation of endogenous QKI followed by RNase-treated LC-MS/MS (IP proteome). b, RNA-aware complexome profiling without RNase reveals extensive co-migration of QKI with spliceosomal assemblies, including PRPF38A, TXNL4A, USP39, SMU1, RBM42, and SNRNP27, suggesting RNA-mediated interactions with multiple splicing stages. c, RNase-treated complexome narrows the QKI interactome, retaining defined associations with early spliceosomal and tri-snRNP proteins, indicating direct protein-protein interactions independent of RNA. d, RNase-treated immunoprecipitation proteome confirms a coherent subset of QKI interactors—TXNL4A, PRPF4, SMU1, RBM42, USP39, and SART1, overlapping with complexome results and highlighting QKI association with early spliceosome complexes. e-g, Western blot validation of QKI co-immunoprecipitation with (e)TXNL4A, (f) SNRNP27, and (g) RBM42 confirms direct interaction with QKI.