EMT-ciliary signaling in quasi-mesenchymal-stem-like cells drives therapeutic resistance and is a druggable vulnerability in triple-negative breast cancer

Abstract

Cancer therapeutic resistance is mediated, in part, by phenotypic heterogeneity and the plasticity of tumor cells, the latter being enabled by epithelial-mesenchymal transition (EMT). However, EMT in human cancer therapeutic response remains poorly understood. We developed patient-derived organoids (PDOs) from human triple-negative breast cancer (TNBC) and investigated their response to chemotherapy. We found that chemotherapy treatment kills the bulk of tumor cells in PDOs, but there is selective survival of malignant cells that had activated an EMT program, entered a quasi-mesenchymal, stem cell-like state and display primary cilia. We developed a family of small-molecule inhibitors of ciliogenesis and show that treatment with these inhibitors, or genetic ablation of primary cilia, is sufficient to suppress this chemoresistance via NFκB-induced cell death. We conclude that an EMT-ciliary signaling axis induces chemoresistance in quasi-mesenchymal ciliated stem-like cells to help tumors evade chemotherapy and represents a druggable vulnerability in human TNBC.

Keywords: EMT; Primary cilia; Therapeutic Resistance; Triple-Negative Breast Cancer.

Figure 1. Quasi-mesenchymal cancer cells assemble primary cilia in human TNBCs.

(A) Patient-derived tumor biopsies (PDB) were stained for the indicated proteins (n = 13, representative result for patient#13 is shown). Scale bar: 2 mm (insets: 3×). (B) The percentage of Ecad + /Vim+ cells was quantified in three distinct Ecad+ tumor region of interest (ROI) for each PDB (n = 13, mean ± s.e.m.). (C) The percentage of Ecad+ ciliated cells was quantified in the same ROIs (n = 13 mean ± s.e.m.). (D) The percentage of Ecad + /Vim+ vs Ecad + /Vim- ciliated cells was determined (n = 13, mean ± s.e.m.; Student’s t test: **P = 0.0090). (E) Patient-derived tumor organoids (PDOs) from distinct patient samples were stained for the indicated proteins (n = 7, representative result for patient #5). Scale bar: 100 μm (insets: 3×). (F, G) The percentage of Ecad + /Vim+ cells and the percentage of ciliated cells were quantified in optical sections per PDO in each patient sample (n ≥5 PDOs/patient sample, mean ± s.e.m.). (H) The percentage of Ecad + /Vim+ vs Ecad + /Vim− ciliated cells was determined. (n = 68 PDOs, six distinct patient samples, mean ± s.e.m.; Student’s t test: ****P < 0.0001). (I) Whole-mount immunofluorescence staining was conducted on PDOs for the indicated markers. 3-dimensional imaging of PDOs was performed. The 3D representation (3D view) as well as internal optical sections (int. sec.) are shown for representative PDOs for patient sample#3. Scale bar: 50 μm (inset: 2×). (J) Signal intensity for Ecad and Vim staining was measured across cells in individual PDOs for the three distinct patient samples, in cells residing within PDOs (inner) in comparison to cells residing at the periphery of PDOs (outer, n ≥5 PDOs/patient sample, mean ± s.e.m.; Student’s t test: n.s. P = 0.63; ****P < 0.0001). (K) Distribution of the Ecad and Vim signal across one representative PDO is shown. (L) Ciliation of cells residing within PDOs in comparison to cells on the outer part of PDOs was quantified (n = 26 PDOs from five distinct patient samples, mean ± s.e.m.; Student’s t test: **P = 0.0102). (M) Gene expression in cancer cells of PDOs was analyzed by scRNAseq (n = 4888 cells). UMAP illustrating transcriptional heterogeneity between cancer cells of PDOs. Each point represents a cell colored according to its cell cluster (0–4), with clustering performed at a resolution of 0.2. (N) Heatmap highlighting marker genes of each cluster. (O–R) UMAP and violin plots illustrating the expression of EMT and ciliogenesis signatures in the distinct cell clusters. Kruskal–Wallis test: ****P < 2.2e-16. Box plots show the median (center line), the 25th and 75th percentiles (lower and upper bounds of the box), and whiskers extending up to 1.5 times the interquartile range from the box limits. Data points beyond this range are considered outliers and are shown individually. Cluster 0 n = 2201, cluster 1 n = 1558, cluster 2, n = 747, cluster 3 n = 242, cluster 4 n = 140. Source data are available online for this figure.

Figure 2. Quasi-mesenchymal ciliated cells mediate chemoresistance.

(A) PDOs from three distinct patient (pat.) samples were stained for the indicated proteins. Scale bars: 100 μm (insets: 3×). The percentage of ciliated cells was quantified in optical sections in PDOs for each patient sample (n ≥33 PDOs/treatment condition from three distinct patient samples, mean ± s.e.m.; Student’s t test: ***P = 0.0008, ****P < 0.0001). (B) Gene expression in cancer cells of PDOs was analyzed by scRNAseq (n = 820 cells). UMAP plot integrating data from three samples (CTL = DMSO-treated, Doxo. = doxorubicin, Taxol). Each point represents a cell colored by sample of origin. (C) UMAP illustrating the different cell clusters identified by transcriptional heterogeneity. Each point represents a cell colored according to its cell cluster (0–4), with clustering performed at a resolution of 0.9. (D) Heatmap highlighting marker genes of each cluster. (E) Percentage of cells in each cluster per sample. (F, G) UMAP and violin plots illustrating the expression of EMT and ciliogenesis signatures in the distinct cell clusters. Kruskal–Wallis test: ****P < 2.2e-16. Box plots show the median (center line), the 25th and 75th percentiles (lower and upper bounds of the box), and whiskers extending up to 1.5 times the interquartile range from the box limits. Data points beyond this range are considered outliers and are shown individually. Cluster 0 n = 251, cluster 1 n = 234, cluster 2 n = 213, cluster 3 n = 66, cluster 4 n = 56. (H) Distribution of cells according to their expression of EMT and ciliogenesis programs. The expression thresholds (dashed line) used to define an EMT^high/CilioUp^high population were calculated based on one standard deviation above the mean of expression of all cells. (I) The percentage of EMT^high/CilioUp^high cells was determined for each sample and normalized to the control. Source data are available online for this figure.

Figure 3. Naonedin-3 represses primary ciliogenesis and chemoresistance of quasi-mesenchymal ciliated stem-like cancer cells.

(A) Schematic representation of the screening strategy for the identification of novel small-molecule inhibitors of primary ciliogenesis. (B) PDOs from three distinct patient (pat.) samples were treated with DMSO (CTL) or Naonedin-3 (Nao-3) and stained for Arl13b. Scale bar: 50 µm. (C) The percentage of ciliated cells was quantified. (n = 6 PDOs/patient sample/treatment condition; Student’s t test: ****P < 0.0001). (D) PDOs arising from patient (pat.) sample #6 were treated at the indicated concentration of Nao-3 and stained for the indicated markers. Representative images of PDOs are shown. Scale bar: 50 μm. (E) The percentage of ciliated cells was quantified in optical sections in PDOs and the IC₅₀ is shown. (n ≥8 PDOs/concentration, mean ± s.e.m.). (F) PDOs were stained for the indicated proteins after the indicated treatments (Tax. = Taxol, Nao-3. = Naonedin-3). (G) The percentage of cleaved-caspase 3 (CC3)+ cells was measured in individual PDO for the distinct patient samples (n = 6 PDOs/patient sample/treatment condition; Student’s t test: CTL vs. Nao-3 ***P = 0.0006, CTL vs. Taxol ***P = 0.0002, CTL vs. Taxol + Nao-3 ****P < 0.0001, Taxol vs. Taxol + Nao-3 **P = 0.0045). Scale bar: 50 µm. (H, I) The percentage of Vim^high cells in PDOs after each treatment was determined by FACS using dissociated samples (n = 3, mean ± s.e.m.; Student’s t test: CTL vs. Nao-3 *P = 0.033, CTL vs. Nao-3 + Taxol ****P < 0.0001). Source data are available online for this figure.

Figure 4. EMT-driven cell plasticity enables formation of primary cilia in quasi-mesenchymal cells to mediate chemoresistance.

(A) Morphology and western blot analysis of EMT markers in E-like (shCTL) and M-like (shECAD) HMLER cells. Scale bar: 100 µm. (B) Cells were stained for the indicated proteins to determine the percentage of ciliated cells (n = 3, mean ± s.e.m.; Student’s t test: ***P = 0.0008). Scale bar: 15 µm. (C) The sensitivity of shCTL and shEcad HMLER cells to Taxol was determined by quantifying incorporation of propidium iodide (PI) in cancer cells in response to Taxol (1 µM) over time (30 h) (n = 3, mean ± s.e.m.; Student’s t test: shCTL vs. shEcad 30 h *P = 0.0193). Results are normalized to the HMLER shCTL. Scale bar: 100 µm. (D) KIF3A and IFT20 knockouts were validated by western blot of extracts for the indicated HMLER variants. (E, F) The impact on ciliogenesis and on Taxol sensitivity was assessed as described in (B, C). Scale bars: 15 µm (E) and 100 µm (F). n = 3, mean ± s.e.m.; Student’s t test (E): ***P = 0.0002. (F) sgCTL vs. sgKIF3A **P = 0.0054, sgCTL vs. sgIFT20 ****P < 0.0001). Source data are available online for this figure.

Figure 5. Primary cilia mediate chemoresistance by repressing NFκB-mediated cell death in quasi-mesenchymal ciliated cancer cells.

(A) PDOs were stained for the indicated proteins after treatment with DMSO (CTL) or Naonedin-3 (Nao-3). Scale bar: 100 µm. (B) Gene expression in cancer cells of PDOs was analyzed by scRNAseq (n = 902 cells). UMAP plot integrating data from the distinct samples (CTL = DMSO-treated, Nao-3. = Naonedin-3). Each point represents a cell colored by the sample of origin. (C) UMAP illustrating the different cell clusters identified by transcriptional heterogeneity. Each point represents a cell colored according to its cell cluster (0–2), with clustering performed at a resolution of 0.1. (D) Heatmap highlighting marker genes of each cluster. (E) Percentage of cells in each cluster per sample. (F–I) UMAP and violin plots illustrating the expression of apoptosis and NFκB signatures in the distinct cell clusters. Kruskal–Wallis tests: ****P < 2.2e-16. Box plots show the median (center line), the 25th and 75th percentiles (lower and upper bounds of the box), and whiskers extending up to 1.5 times the interquartile range from the box limits. Data points beyond this range are considered outliers and are shown individually. Cluster 0 n = 499, cluster 1 n = 281, cluster 2 n = 122. (J) Relative levels of the indicated gene transcripts in HMLER shECAD sgCTL or sgIFT20 variants treated with Taxol were determined by real-time qPCR analysis. n = 3, mean ± s.e.m. (K, L) The sensitivity of HMLER shEcad sgCTL or sgIFT20 variants to Taxol alone or in combination with Bay11-7085 was determined by quantifying incorporation of propidium iodide (PI) in cancer cells over time (48 h). Scale bar: 100 µm. n = 3, mean ± s.e.m., results are normalized to the HMLER shEcad sgCTL after Taxol treatment. Student’s t test: sgCTL vs. sgIFT20 *P = 0.0213, sgIFT20 vs. sgIFT20 + Bay11-7085 *P = 0.0299. sgCTL vs. sgCTL + Bay11-7085 n.s. P > 0.45. Source data are available online for this figure.

Figure EV1. Cancerous lesions in patient-derived biopsies and additional features of patient-derived cancer organoids.

(A) Paraffin sections from patient-derived biopsies (PDBs) were stained with H&E (n = 13, a representative result for patient#13 is shown), Scale bars: 2 mm (low magnification) and 100 μm (high magnification). (B) The morphology of PDOs from distinct patient samples (pat. #1-7) was examined by brightfield microscopy. Scale bar: 100 μm. (C) Paraffin sections of tumors were stained for the indicated proteins (a representative image is shown). Scale bar: 15 μm, inset: 3x. (D) Patient-Derived Organoids (PDOs) were stained for the indicated proteins (a representative image is shown). Scale bar: 15 μm, inset: 3x. (E) Heatmap showing inferred CNVs in cells of PDOs from patient sample#3 with unsupervised hierarchical clustering. (F–I) UMAP and violin plots illustrating the expression of mammary stem cell and pluripotency signatures in the distinct cell clusters composing PDOs from patient sample#3. Kruskal–Wallis test: ****P < 2.2e-16. Box plots show the median (center line), the 25th and 75th percentiles (lower and upper bounds of the box), and whiskers extending up to 1.5 times the interquartile range from the box limits. Data points beyond this range are considered outliers and are shown individually. Cluster 0 n = 2201, cluster 1 n = 1558, cluster 2, n = 747, cluster 3 n = 242, cluster 4 n = 140.

Figure EV2. Impact of chemotherapy on PDOs and additional features of quasi-mesenchymal ciliated stem-like cells.

(A) The dose-dependent impact of chemotherapy on cell viability in PDOs was tested for two distinct patient (pat.) samples (#3 and #7). Representative images of PDOs treated with DMSO (CTL) or with doxorubicin (doxo., 100 nM) and Taxol (10 nM) after 48 h of treatment are shown. The cell viability was measured at different concentrations and the IC50 is shown for each drug (mean ± s.e.m., representative analysis of 3 independent experiments). Scale bar: 100 μm. (B) The impact of drugs on the ability of chemoresistant cancer cells to reconstitute PDOs after drug removal at two intermediate drug concentrations is shown and measured at the indicated concentrations. Scale bar: 100 μm (n ≥213 PDOs/treatment condition, mean ± s.e.m.). (C–F) UMAP and violin plots illustrating the expression of MYLK, TPM2, mammary stem cell and pluripotency signatures in the distinct cell clusters from PDOs. Box plots show the median (center line), the 25th and 75th percentiles (lower and upper bounds of the box), and whiskers extending up to 1.5 times the interquartile range from the box limits. Data points beyond this range are considered outliers and are shown individually. Cluster 0 n = 251, cluster 1 n = 234, cluster 2 n = 213, cluster 3 n = 66, cluster 4 n = 56. (G, H) Expression levels of MYLK and TPM2 was examined in cells from control and post-chemotherapy PDOs. The proportion of cells expressing high levels of MYLK and TPM2 ( > 2) was determined for each sample and normalized to the control. MYLK > 2 CTL n = 12, Doxo. n = 42, Taxol n = 131; TPM2 > 2 CTL n = 18, Doxo. n = 37, Taxol n = 128.

Figure EV3. Distribution of cells along the pseudotime trajectory and progressive expression of EMT and ciliogenesis programs.

(A) UMAP combining data from 3 samples (PDOs from patient sample #3, CTL/Doxo./Taxol) colored by pseudotime value. Cells are ordered along an inferred trajectory; a black line connects all clusters. (B, C) Distribution of cells from distinct samples and clusters along the pseudotime trajectory and correlative analysis of the expression of EMT and CilioUp signatures. Wilcoxon test: Hallmark EMT score CTL R = 0.75 P < 2.2e-16, Doxo. R = 0.85 P < 2.2e-16, Taxol R = 0.86 P < 2.2e-16; CilioUp score CTL R = 0.42 P = 1.4e-07, Doxo. R = 0.56 P < 2.2e-16, Taxol R = 0.42 P < 2.2e-16. (D) Violin plots illustrating the expression of EMT and ciliogenesis transcriptional programs in individual cells of cluster 2 per treatment condition. CTL n = 21, Doxo. n = 48, Taxol n = 144.

Figure EV4. Identification of small-molecule inhibitors of primary ciliogenesis.

(A) PDOs from distinct patient (pat.) samples were treated with DMSO (CTL) or Ciliobrevin A (CilA, 100 µM, 24 h) and stained for Arl13b. Scale bar: 50 µm. The percentage of ciliated cells was quantified in distinct PDOs for each patient sample (n ≥4 PDOs/patient sample/treatment condition, means ± s.e.m.; Student’s t test: CTL vs. CilA n.s. P = 0.98). (B) A secondary screen with small molecules identified as candidate ciliogenesis inhibitors in a primary screen (Fig. 3A) at two distinct concentrations (1 and 10 µM) was conducted. The ciliogenesis index was determined 24 h after each treatment. Results are relative to CTL, DMSO-treated cells (n = 3, mean ± s.e.m.). (C) The impact of the most potent inhibitors on ciliogenesis is shown (10 μM, Nao-1 = 398_C6; Nao-2 = 398-E5; Nao-3 = 408-A10; Nao-4 = 408_C11; Nao-5 = 408_H10). Scale bar: 15 µm. (D) Chemical structures of Naonedins.

Figure EV5. Mechanisms of EMT-cilium-dependent regulation of chemoresistance.

(A) The correlation in the expression levels of EMT and apoptosis gene expression signatures in cancer cells from PDOs treated with Nao-3 was determined using a Spearman test, r = 0.45, ***P < 0.001. (B) Violin plots illustrating the expression levels of the indicated gene transcripts in cell clusters from DMSO-treated and Nao-3-treated PDOs. Kruskal–Wallis test: ns. P = 0.18; ***P = 0.00085; ****P < 0.0001. Box plots show the median (center line), the 25th and 75th percentiles (lower and upper bounds of the box), and whiskers extending up to 1.5 times the interquartile range from the box limits. Data points beyond this range are considered outliers and are shown individually. Cluster 0 n = 499, cluster 1 n = 281, cluster 2 n = 122. (C) Heatmap illustrating the expression levels of the indicated gene transcripts in HMLER shEcad cells treated with a vehicle control (CTL, DMSO) or Ciliobrevin A (CilA). GSE160549 dataset.