Fig 2: FLNB isoform switching promotes the mesenchymal cell state.(A) Schematic of FLNB (Filamin B) protein domain structure, which contains an N-terminal actin-binding domain (ABD, shown in yellow), 24 filamin repeats (shown in blue) and two hinge domains (H1 and H2, shown in red). The exon 30 of FLNB encodes the first hinge domain (H1). (B) CD44-high and CD44-low HMLER cells were sorted and FLNB AS was analyzed by RT-PCR with primers flanking FLNB exon 30. The CD44-high HMLER cells were spontaneously generated. (C) The expression of FLNB was suppressed by siRNAs against FLNB 3’UTR region (FLNB-U1, U2 and UTR1,2) in HMLE cells. FLNB-U1,2 is a pool of FLNB-U1 and FLNB-U2 siRNAs. Protein levels of FN1, VIM (mesenchymal markers), CDH1 (epithelial marker) and FLNB after FLNB suppression was quantified by immunoblotting. (D) The expression of FLNB was rescued by FLNB-L (FLNB long isoform) or FLNB-ΔH1 (FLNB lacking H1 domain) after endogenous FLNB depletion by siRNA targeting the UTR of FLNB (FLNB-U1,2). Protein levels of FN1, VIM (mesenchymal markers), CDH1 (epithelial marker) and FLNB were quantified by immunoblotting. (E) Schematic of CRISPR/Cas9-mediated targeting of the FLNB intron 29 - exon 30 junction. (Left) Guide RNAs were designed to target the junction of FLNB intron 29 (in lower-case black letters) and exon 30 (in capital red letters). (Right) Examples of modifications to the junction induced by sgRNAs sgFLNB-SK2 and sgFLNB-SK4 are shown. (F) FLNB exon 30 skipping and total FLNB levels were quantified in cells expressing sgRNAs targeting the junction of FLNB intron 29 and exon 30 by RT-PCR (upper) and immunoblot (lower) (corresponding to Figure 6E). (G–I) Effect of CRISPR/Cas9-mediated modulation of the FLNB-L/FLNB-ΔH1 ratio on the expression of mesenchymal and epithelial markers as examined by qPCR (G), on the expression of CD44 as analyzed by flow cytometry (H), and on the number of mammosphere formed per 1000 cells cultured in low attachment condition (I). n = 4 for (G), n = 6 for (H), n = 3 for (I). *, p<0.05; **, p<0.01, Student’s two-tailed t-test compared to sgGFP controls.

Fig 3: EMT mediated by FLNB isoform switching is dependent on the FOXC1 transcription factor.(A) Analysis of nuclear and cytoplasmic fractions of FLNB. HMLE cells expressing Cas9 and sgRNAs targeting GFP (sgGFP, as control) or targeting the FLNB intron 29 - exon 30 junction to induce exon skipping (sgFLNB-SK2 and SK4, described in Figure 6E), were subjected to subcellular fractionation. The levels of FLNB proteins in the nuclear and cytoplasmic fractions were analyzed by immunoblotting and the skipping of exon 30 was confirmed by RT-PCR. The nuclear levels of FLNB is diminished when exon 30 is skipped. (B) Interaction between FLNB, FOXC1 and PBX1 in the nucleus as determined by immunoprecipitation followed by immunoblotting. HMLE cells expressing Cas9 and sgRNAs targeting GFP (sgGFP, as control) or targeting the FLNB intron 29 - exon 30 junction to induce exon skipping (sgFLNB-SK4, described in Figure 6E), were subjected to immunoprecipitation analysis using nuclear protein extract. The interaction between FLNB and FOXC1 in the nuclear extract was reduced largely due to the decrease in nuclear FLNB levels. (C) The depletion of FOXC1 diminishes the upregulation of mesenchymal markers (ZEB1 and FN1) mediated by FLNB exon 30 skipping (sgFLNB-SK2 and SK4, shown in red bars) as measured by qPCR. n = 6, *, p<0.05; **, p<0.01, Student’s two-tailed t-test. The knockdown efficiency for FOXC1 is shown in the upper panel. (D) The rate of mammosphere formation in HMLE cells that expressed the indicated sgRNAs and transfected with a FOXC1 siRNA pool. The number of mammospheres per 1000 cells plated was counted. Representative images (left) and quantification (right) are shown. n = 3, **, p<0.01, Student’s two-tailed t-test. (E, F) Depletion of FOXC1 diminishes the upregulation of mesenchymal markers (E) and mammosphere formation (F) mediated by QKI and RBFOX1 expression, as measured by qPCR (E) and mammosphere formation assay (F). n = 6 for (E), n = 4 for (F). *, p<0.05; **, p<0.01, Student’s two-tailed t-test. (G) A model of the mechanism underlying the QKI and RBFOX1-induced mesenchymal and stem-like cell state.

Fig 4: QKI and RBFOX1 cooperatively regulate the alternative splicing of common downstream targets, including FLNB.(A) Venn diagram illustrating the overlap in skipped exons regulated by QKI and RBFOX1, as detected by RNA-seq. The significance of overlap was determined by Fisher’s exact test. (B, C) Binding distributions and de novo motifs identified from eCLIP-seq analysis of significant peaks (p≤10−5, fold-enrichment ≥8). The QKI (B) or RBFOX1 (C) binding distributions are shown in pie charts (upper). Below are shown the top two ranked QKI (B) and RBFOX1 (C) binding motifs. (D) Venn diagram illustrating the overlap in binding peaks detected by eCLIP-seq analysis of QKI and RBFOX1. p<0.001, as determined by random shuffling of one set of peaks and repeated testing of the extent of overlap. (E) Percentage of SE splicing events regulated by both QKI and RBFOX1 (n = 183), that contain CLIP-binding peaks located in the exon or flanking introns, for either or both proteins. (F) Heatmap depicting the association between an EMT signature score (top row) and AS events in a panel of breast cancer cell lines. Shown are the top 20 AS events that are positive or negatively associated with the EMT score. (G) Scatter plot of the change in ‘Percentage Spliced In’ (PSI) for SE events shared between QKI and RBFOX1. ΔPSI values are for each ORF relative to EGFP. Colored dots indicate events in CD44 or FLNB. (H) Tracks indicate eCLIP-seq read density for QKI and RBFOX1 at FLNB exon 30, normalized to a million total usable reads. Below each pair of replicate tracks are shown significant binding peaks called by CLIPper software (p≤10−5, fold-enrichment ≥8). Also shown are size-matched input controls for each pair of replicates.10.7554/eLife.37184.018Figure 4—source data 1.QKI and RBFOX1 overlapping events.10.7554/eLife.37184.019Figure 4—source data 2.QKI and RBFOX1 CLIP peaks.

Fig 5: GPX4 is required for DPP7‐mediated protection against disulfidptosis. (A) Western blot analysis of DPP7 and GPX4 protein expression in RKO and HC29 cells transduced with empty vector (EV), DPP7‐expressing lentivirus (DPP7), or DPP7‐expressing lentivirus plus GPX4‐specific siRNA (DPP7 + si‐GPX4). GAPDH was used as a loading control. (B, C) Cell viability of RKO (B) and HC29 (C) cells with indicated genetic modifications under glucose‐replete (Glc+) or glucose‐deprived (Glc‐) conditions for 12 h. (D, E) Flow cytometry analysis of apoptosis in RKO (D) and HC29 (E) cells with indicated genetic modifications under glucose‐replete or glucose‐deprived conditions. (F, G) Non‐reducing (Non‐Re) and reducing (Re) Western blot analysis of cytoskeletal proteins (Drebrin, FLNA and FLNB) in RKO (F) and HC29 (G) cells with indicated genetic modifications under glucose‐replete or glucose‐deprived conditions. Data are presented as mean ± SD of three independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001. SEF, starvation effect factor.