Fig 2: ILF3 binds closely to the differential editing sites in editing-expression-correlated genes. a Histogram of distances between differential editing sites in editing-correlated genes and the closest ILF3 eCLIP peaks in A549 cells (turquoise), up to 10 kb. Gray curves represent distances for 10 sets of randomly picked A’s in the same genes as differential editing sites. Number of differential editing sites is given by n. p value was calculated by comparing the area under the curve (AUC) of the distance distribution for differential editing sites to a normal distribution fit to the AUC values of 10,000 sets of random gene-matched A’s. b Scatterplot of Pearson correlation coefficient and significance (log10-transformed adjusted p value) of correlation between ILF3 mRNA expression and mRNA expression of editing-correlated genes. Genes passing 10% FDR are labeled as significant (sig, turquoise), others as nonsig. c Cumulative distributions of distances between ILF3 eCLIP peaks and differential editing sites within editing-expression-associated genes (sig) or differential editing sites in genes without editing-expression associations (nonsig), up to 1 kb. Only genes associated with immune and viral related GO terms were included. p value calculated by the Kolmogorov-Smirnov test. d For each cell type in the lung cancer scRNA-seq dataset, ILF3 mRNA expression was correlated with mRNA expression of editing-expression-correlated genes (identified in the TCGA data) by Pearson correlation. Genes associated with any immune or viral-related GO term are shown. The size of each point indicates significance of correlation and color corresponds to values of the correlation coefficient. e Normalized mCherry expression (mean ± SD) for nonedited or edited versions of sites in the 3′ UTR of PKR in A549 cells. Five biological replicates were performed. p value calculated by two-sided t-test (same below), *p < 0.05. f Normalized mRNA expression (mean ± SD) of endogenous PKR in siControl, siADAR1, and siADAR2 A549 cells. Three biological replicates were performed. *p < 0.05. n.s., not significant. g Read coverage of ILF3 eCLIP-seq in A549 cells for two biological replicates (ILF3 IP1 and ILF3 IP2, turquoise) and size-matched input (SMInput, gray). The five validated 3′ UTR editing sites affecting PKR mRNA abundance in A549 cells are labeled in magenta (left). Right: Validation of PKR eCLIP signal overlapping two editing sites. PKR expression (mean ± SD) was measured by qRT-PCR in the IP or SMInput samples and normalized against the expression of 18s rRNA, *p < 0.05. (n = 3)

Fig 3: ILF3 regulates RNA editing through Alu element binding and ADAR1 interaction.(a) Genome-wide distribution of ILF3 binding sites determined by CUT&Tag analysis, showing normalized read density across gene bodies ±3 kb from transcriptional centers. (b) Overlap between ILF3 binding sites identified by CUT&Tag and eCLIP analyses (RPKM>0.5; top 1% of peaks by area under the curve (AUC). (c) Gene Ontology enrichment analysis of ILF3-bound regions identified by eCLIP-seq and CUT&Tag. (d) Distribution of ILF3 eCLIP signal intensity relative to size-matched input controls (log2FC > 3; P < 0.001). (e) Enrichment analysis of repetitive element classes in ILF3 eCLIP peaks relative to a randomized peak distribution. (f) AlphaFold3-predicted structural model of the ILF2/3 complex bound to Alu RNA. (g) Proteomic analysis of ILF3 interactors identified by immunoprecipitation-mass spectrometry (n = 3 biological replicates). (h) Single nucleotide variants (SNVs) detected in ILF3-depleted versus control hPSCs. Base substitution types include counts for the reverse complement variant. (i) Comparison of editing frequencies at A-to-I edited sites in ILF3 knockdown or control knockdown hPSCs. (n = 2 biological replicates; P values determined by Wilcoxon rank-sum test). (j) Aggregate plot showing ILF2/3 eCLIP signal distribution centered around A-to-I-edited sites after ILF3 depletion. Z-scores calculated relative to a randomized peak distribution. (k) Phylogenetic tree – the diameter of each bubble is proportional to the percentage of each species’ respective genome that aligns to Alu elements (top row) or ILF3 eCLIP targets identified in hPSCs (bottom row). (l) Western blot analysis of interactions between FLAG-tagged ADAR1 and wild-type HA-ILF3 or RNA-binding mutant (HA-ILF3ΔRBM). (m) Editing frequencies in ILF3 wild-type versus ΔRBM rescue conditions (n = 2 biological replicates; P values determined by Wilcoxon rank-sum test). (n) Quantification of NANOG-positive cells under self-renewal and exit conditions following rescue with wild-type or ΔRBM mutant ILF3 (n = 3 biological replicates; P values determined by one-way ANOVA with Tukey’s multiple comparisons test).

Fig 4: ILF2 and ILF3 are required for human gastrulation.(a) Results from genome-wide loss-of-function screens in human pluripotent stem cells (hPSCs) during pluripotency exit induced by either TGFβ and bFGF withdrawal or MAPK pathway inhibition1, depicting mean Z-score from three replicates. (b) CRISPRi (top) and representative western blot with signal quantification (bottom) depicting validation of ILF2 and ILF3 knockdown efficiency after 3 days of doxycycline treatment in hPSCs expressing targeted guide RNAs. (c) Experimental design for pluripotency exit assays. (d) Representative flow cytometry quantification of NANOG in hPSCs under self-renewal and exit conditions by MAPK pathway inhibition following ILF2 or ILF3 knockdown. (e) Schematic of chimpanzee (Pan troglodytes) PSC differentiation experiments. (f) Representative flow cytometry quantification of NANOG-positive chimpanzee PSCs under self-renewal or exit conditions following ILF2 or ILF3 knockdown. (g) Schematic of mouse ESC differentiation experiments. (h) Representative flow cytometry quantification of NANOG-positive mouse ESCs under primed pluripotency or exit conditions following Ilf2 or Ilf3 knockdown. (i) Hierarchical clustering of RNA-seq datasets from hPSCs in self-renewal and exit conditions. (j) Gene Set Enrichment Analysis (GSEA) of pluripotency-associated genes in ILF2- or ILF3-depleted cells versus control during exit conditions (ILF2: NES=2.50, p=3.58E-13; ILF3: NES=2.66, p= 1.54E-13). (k) Differential gene expression analysis comparing control and ILF2 or ILF3-knockdown hPSCs after exit from pluripotency (n = 2 biological replicates; fold change > 1.5; P < 0.05). Red and blue indicate up- and down-regulated genes, respectively. (l) Quantification of pluripotency-associated differentially accessible regions in control and ILF2 or ILF3-depleted cells under self-renewal and exit conditions (n = 2 biological replicates, P values calculated using paired Wilcoxon rank-sum test). (m) Experimental design for three-dimensional human peri-gastruloid formation. (n) Representative brightfield images of ILF2- and ILF3-depleted human peri-gastruloids. Scale bar, 100 μm. (o) Analysis of longest axis length ILF2- and ILF3-depleted human peri-gastruloids (n = 7–10 biological replicates; mean ± s.d; P values determined by two-way ANOVA with Šidák’s multiple comparisons test). (p) Immunofluorescence analysis of human peri-gastruloids showing SOX2 (red), SOX17 (green), T (pink), and nuclear DAPI staining (blue) in control and ILF2 or ILF3-depleted cells. Scale bar, 50 μm.

Fig 5: ILF2 and ILF3 suppression disrupts the transcriptional and chromatin profiles of hPSCs.(a) Differential gene expression analysis in ILF2- or ILF3-depleted cells under self-renewal and exit conditions (n = 2 biological replicates per condition; |FC| > 1.5; P < 0.05; red: upregulated, blue: downregulated). (b) Correlation analysis of transcriptional changes between ILF2- and ILF3-depleted cells under self-renewal and exit conditions. (c) ATAC-seq analysis of chromatin accessibility changes in ILF2- or ILF3-depleted cells (n = 2 biological replicates; |FC| > 1.5; P < 0.05; red: increased accessibility, blue: decreased accessibility). (d) Correlation analysis of chromatin accessibility changes between ILF2- and ILF3-depleted cells under self-renewal and exit conditions. (e) Hierarchical clustering of ATAC-seq datasets from hPSCs in self-renewal and exit conditions. (f) RT-qPCR analysis of differentiation markers in markers in control and ILF2/3-depleted human peri-gastruloids (n = 3 biological replicates; mean ± s.d.; P values determined by one-way ANOVA with Tukey’s multiple comparisons test). (g) Experimental schematic for two-dimensional human gastruloid formation. (h) Representative immunofluorescence analysis of germ layer markers in control and ILF2 or ILF3-depleted 2D human gastruloids. Scale bar, 100 μm. (i) Schematic of mouse gastruloid formation experimental design. (j) Representative brightfield images of Ilf2- and Ilf3-depleted mouse gastruloids. Scale bar, 200 μm. (k) Length analysis of Ilf2- and Ilf3-depleted mouse gastruloids (n = 10 biological replicates; mean ± s.d; P values determined by one-way ANOVA with Tukey’s multiple comparisons test). (l) RT-qPCR analysis of Ilf2- and Ilf3-depleted mouse (n = 3 biological replicates; mean ± s.d.; P values determined by one-way ANOVA with Tukey’s multiple comparisons test).