An alternative splicing signature defines the basal-like phenotype and predicts worse clinical outcome in pancreatic cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is characterized by extremely poor prognosis. PDAC presents with molecularly distinct subtypes, with the basal-like one being associated with enhanced chemoresistance. Splicing dysregulation contributes to PDAC; however, its involvement in subtype specification remains elusive. Herein, we uncover a subtype-specific splicing signature associated with prognosis in PDAC and the splicing factor Quaking (QKI) as a determinant of the basal-like signature. Single-cell sequencing analyses highlight QKI as a marker of the basal-like phenotype. QKI represses splicing events associated with the classical subtype while promoting basal-like events associated with shorter survival. QKI favors a plastic, quasi-mesenchymal phenotype that supports migration and chemoresistance in PDAC organoids and cell lines, and its expression is elevated in high-grade primary tumors and metastatic lesions. These studies identify a splicing signature that defines PDAC subtypes and indicate that QKI promotes an undifferentiated, plastic phenotype, which renders PDAC cells chemoresistant and adaptable to environmental changes.

Keywords: RNA processing; alternative splicing; chemoresistance.

Graphical abstract

Figure 1

Identification of the PDAC subtype-specific splicing signature (A–C) Pie charts of splicing patterns differentially regulated in classical vs. basal-like PDAC subtypes (A), splicing events up-regulated in classical or basal-like tumors (B), and biological processes of splicing-regulated genes (C). (D) List of the splicing-regulated genes. (E and F) PSI distribution of “classical” exons in the SEC31A, CTNND1, and ADD3 genes and of “basal-like” exons in the LRRFIP2, SEC16A, and SYTL2 genes in TCGA PDAC samples. The regulated EC is shown in orange in the x axis. Statistical analyses of splicing regulation (|Δ median PSI| ≥ 0.1 and false discovery rate [FDR] ≤ 0.01) was performed by the Wilcoxon rank-sum test with Benjamini-Hochberg (FDR) adjustment. (G and H) Kaplan-Meier curves of relapse-free survival (RFS) of patients with PDAC segregated for inclusion (blue line) or skipping (black line) of classical (G) and basal-like (H) exons. Type of event, regulated exon(s), and genomic coordinates of the regulated exon(s) and of the flanking splice sites are indicated. See also Figure S1 and Table S1.

Figure 2

Inclusion of the ADAM15 and SEC16A ECs is associated with prognosis in unselected patients with PDAC (A) Representative splicing assays of the ADAM15 EC in PDAC samples. (B) Pearson’s correlation analysis between the ADAM15 exon 22 PSI and overall survival (OS) in unselected patients with PDAC (n = 22). The blue and pink dots correspond to, respectively, basal-like and classical samples identified by RNA-seq (Figure S2). Black dots correspond to non-sequenced samples. (C) Kaplan-Meier curve displaying the OS of patients with PDAC exhibiting high (red) or low (black) inclusion levels of ADAM15 exon 22 (m.s., median survival). (D) Representative splicing assays of the SEC16A EC in PDAC samples. (E) Pearson’s correlation analysis between the PSI of SEC16A exon 33 and OS in unselected patients with PDAC (n = 22). (F) Kaplan-Meier curve displaying OS of patients with PDAC exhibiting high (red) or low (black) inclusion levels of SEC16A exon 33. See also Figure S1 and Table S2.

Figure 3

Identification of RBPs involved in the establishment of the subtype-specific splicing signature (A) Percentage of the subtype-specific splicing events that correlate to the indicated RBPs. RBPs expressed at higher levels in classical tumors are listed in red; those expressed at higher levels in basal-like tumors are listed in blue. (B) Correlation analysis between the expression of GATA6 and that of RBM47 or RBMS1. (C) Quantitative real-time PCR (real-time qPCR) analysis of RBP expression in PDAC cell lines. (D and E) Representative splicing assay (n = 3) after silencing RBPs in HPAF-II (D) or PANC-1 (E) cells. (C–E) Mean ± standard deviation (SD) of three independent experiments. Student’s t test; ∗∗∗p = 0.001. See also Figures S2–S4 and Table S3.

Figure 4

Transcriptome analysis of PANC-1 cells depleted of QKI or RBFOX2 (A) Western blot analysis of QKI and RBOFOX2 in PANC-1 cells transfected with the corresponding small interfering RNAs (siRNAs) (48 h). (B) QKI or RBFOX2 splicing-regulated genes represented as percentage (orange) of total genes expressed in PANC-1 cells. (C) Percentage of the indicated splicing patterns regulated by depletion of either QKI or RBFOX2. EC, exon cassette; AFE, alternative first exon; ALE, alternative last exon; Comp, complex event; IR, intron retention; A5′, alternative 5′ splice site; A3′, alternative 3′ splice site; MEE, mutually exclusive exon; IED, internal exon deletion. (D) Overlap between QKI- and RBFOX2-regulated events identified in PANC1 cells. (E) Venn diagram showing the direction of the regulated events in QKI- or RBFOX2-depleted cells. (D and E) Statistical analysis was performed by the hypergeometric test. (F) Logo representation of the motif most significantly enriched in proximity of the regulated exons identified in (C). RBPs predicted to bind to the indicated sequences (Tomtom motif comparison tool) are shown below the logos. (G and H) Bar graphs showing the percentage of QKI- or RBFOX2-regulated exons (up or down) containing the identified motifs in comparison with their abundance in reference non-regulated ECs (ref) or constitutive exons (const). (I and J) Distribution of the indicated QKI (I) and RBFOX2 (J) motifs within exons and flanking introns (±250 nt) in the up-regulated (red) and down-regulated (green) exons. (K) RT-PCR analyses for the indicated ECs: QKI-specific (left), RBFOX2-specific (middle), and common (right) targets were selected. Percentage of splicing inclusion ± SD (n = 3) of ECs was evaluated by densitometry. See also Figures S4 and S5 and Table S4.

Figure 5

QKI regulates the basal-like splicing signature in PDAC cells (A) Subtype-specific events expressed in PANC-1 cells and regulated by QKI and RBFOX2. (B) Representative splicing assay of LRRFIP2, SEC16A, and SYTL2 ECs in PANC-1 cells silenced for QKI, RBFOX2, or both. Mean ± SD (n = 3) is reported below the panels. (C) Schematic representation of the QKI and RBFOX2 binding sites in the sequence flanking SEC16A exon 33. (D) Schematic representation of the SEC16A minigene (top), western blot analysis, and splicing assay of the SEC16A minigene performed in PANC1-1 cells depleted of QKI, RBFOX2, or both. (E) Mutations of the QKI and RBFOX2 binding sites performed in the SEC16A minigene. (F) Splicing assay of QKI/RBFOX2 mutants in PANC-1 cells. Mean ± SD (n = 3) is reported below the panels. See also Figure S5 and Table S4.

Figure 6

QKI expression is a marker of the basal-like identity of PDAC cells (A) Pie chart of subtype-specific events regulated by QKI. (B) Splicing assays of ADD3 and CTNND1 ECs in HPAF-II, Capan-1, and PANC-1 and MiaPaCa-2 (MP) cells silenced for QKI, RBFOX2, or both. Mean ± SD of the PSI value (n = 3) is reported below the panels. (C) Overlap between events regulated by QKI and those regulated in PANC-1 vs. HPAF-II. Statistical analysis was performed by the hypergeometric test. (D) Histograms representing the splicing events of the overlap in (C) and showing the type of regulation (up or down) in HPAF-II cells with respect to PANC-1 cells. Events were separated between those up- or down-regulated in QKI-depleted PANC-1 cells. (E) Splicing assays of ESYT2, RAI14, and EXOC1 ECs in HPAF-II, Capan-1, PANC-1, and MP cells. The latter two cell types were also depleted or not for QKI. Mean ± SD of the PSI value (n = 3) is reported below the panels. (F) Violin plot of QKI expression level in classical and basal-like cells identified by single-cell transcriptomic analysis. (G) Violin plot showing of QKI expression in tumors classified as classical, intermediate, and basal-like by single-cell transcriptomic analysis. (F and G) Statistical analyses were performed by Wilcoxon’s test. (H) Histogram showing percentage of classical/intermediate/basal-like patients that were evaluated as positive or negative for QKI (QKI^+/−) or GATA6 (GATA6^+/−) expression (left) or classified as GATA6⁺/QKI⁻ or GATA6⁻/QKI⁺ (right). (I and J) Representative images of PDAC PDO-9 and PDO-12 lines cultured for 9 days in regular or minimal medium (I) and corresponding splicing assay of the ADD3, CTNND1, LRRFIP2, SYTL2, and SEC16A ECs (J). The PSI value is reported below the panels. Scale bar: 50 μm. (K) Immunohistochemistry (IHC) analysis of QKI in PDO-9 and PDO-12. Scale bar: 20 μm. (L) RT-qPCR analysis of QKI expression in PDOs after 9 days in minimal media. (M) Splicing assay of indicated ECs in PDO-9 silenced or not for QKI and cultured in minimal media. PDOs were electroporated with the indicated siRNAs (40 nM) and collected after 4 days. Analysis of QKI expression level is shown. See also Figures S5 and S6 and Tables S2, S4, and S5.

Figure 7

QKI promotes a pro-mesenchymal (MES) phenotype in PDAC (A) Overlap between ECs regulated by QKI and ECs regulated during EMT in breast cancer. Statistical analysis was performed by the hypergeometric test. (B) EC events present in the overlap in (A) showing the type of regulation (up or down) in breast epithelial (EPI) cells with respect to MES cells. Events were separated between those up- or down-regulated in QKI-depleted PANC-1 cells. (C) Splicing assays of the NFYA, CLSTN1, and SPAG9 ECs in HPAF-II and in PANC-1 depleted or not for QKI. Mean ± SD of the PSI value (n = 3) is reported below the panels. (D) Wound-healing assays in PANC-1 cells silenced for QKI (left). Scale bar: 400 μm. Silencing of QKI expression was assessed by western blot analysis (right). The histograms report the quantification of wound area. Data are reported as the mean ± SD (n = 3; two-way ANOVA test; ∗p < 0.05). (E) Kaplan-Meier curve displaying the OS of patients with PDAC exhibiting high (third and fourth upper quartiles, red line; n = 6) or low (first and second quartiles, black line; n = 16) QKI expression, as determined by real-time qPCR in EUS-TA samples. (F) IHC analysis of QKI in patients with PDAC. Dashed lines highlight the area of tissue containing G1/G2 tumors (blue) or G3 tumor (red). Arrows point to representative individual G1/G2 tumor cells (blue) or G3 tumor cells (red). Scale bar: 200 μm. (G) Percentage of G1/G2 (blue) and G3 (red) tumors that were positive (QKI⁺) or negative (QKI⁻) for QKI expression by the IHC analysis. Statistical analysis was performed by the Fisher’s test. (H) Violin plot of QKI expression in resectable (R), borderline/locally advance (B/LA), and metastatic (M) tumors as determined by single-cell transcriptomic analysis. Statistical analysis was performed by Wilcoxon’s test. (I) QKI expression levels in primary lesion and liver metastasis from nine patients with PDAC (Student’s t test; ∗p < 0.05 and ∗∗p < 0.01). (J) Clonogenic assays of PANC-1 cells silenced or not for QKI after 10 days. The histogram reports the percentage of seeded cells that formed colonies (Student’s t test; ∗p < 0.05 and ∗∗p < 0.01). (K–M) Analysis of cell sensitivity performed by crystal violet in PANC-1 cells silenced for QKI (K and L) or for three basal-like ECs (SEC16A, SYTL2, and LRRFIP2) (M). Treatments were carried out for 6 days with increasing concentrations of irinotecan or mFOL (n = 3; mean ± SD; two-way ANOVA test; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p = 0.001, and ∗∗∗∗p < 0.0001). See also Figure S7 and Tables S2 and S6.