SPP1 is required for maintaining mesenchymal cell fate in pancreatic cancer

Abstract

Elucidating the complex network of communication between tumour cells is central to understanding cell fate decisions and progression of pancreatic ductal adenocarcinoma (PDAC)^1,2. We previously showed that constant suppression of BMP activity by the BMP antagonist GREM1 secreted by mesenchymal PDAC cells is essential for maintaining the fate of epithelial PDAC cells³. Here we identify SPP1 (also known as osteopontin)⁴ as a key regulator of mesenchymal cell fate in pancreatic cancer. Proteomic analysis of plasma from patients with PDAC showed that SPP1 is substantially upregulated in late-stage disease. Inactivation of Spp1 led to a delay in tumorigenesis in mouse PDAC models and abolished metastasis formation. Spp1 was expressed in epithelial PDAC cells, and Spp1 inactivation resulted in a conversion of mesenchymal to epithelial PDAC cells. Mechanistically, SPP1 bound the CD61 receptor on mesenchymal PDAC cells to induce Bmp2 and Grem1 expression, and GREM1 inhibition of BMP signalling was required for Spp1 expression in epithelial cells, thereby forming an intercellular regulatory loop. Concomitant inactivation of Grem1 reverted the epithelial phenotype of Spp1 knockout to fully mesenchymal PDAC. Conversely, Grem1 heterozygosity combined with Spp1 knockout resulted in wild-type PDAC histology, a result that confirmed the direct antagonistic functions of these factors. Hence, mesenchymal and epithelial PDAC cell fates are determined by the reciprocal paracrine regulation of the soluble factors GREM1 and SPP1.

Fig. 1. EPCs secrete SPP1 to maintain MPC identity.

a, Schematic showing a comparison of the proteomic profiles of plasma proteins of patients with stage I or II PDAC and patients with stage III or IV PDAC. LC–MS, liquid chromatography–mass spectrometry. b, Volcano plot showing differential protein expression between plasma samples from patients with early-stage pancreatic cancer (stage I or II) or late-stage pancreatic cancer (stage III or IV). Blue dots indicate proteins with increased expression in early-stage pancreatic cancer (stage I and II), and red dots indicate proteins with increased expression in late-stage pancreatic cancer (stage III and IV). Statistical analysis was performed using two-sided t-tests. c, ELISA of plasma SPP1 levels in KPCY and WT mice (n = 3 mice). d, Immunofluorescence analysis of VIM (red), YFP (grey) and SPP1 (green) in KPCY tumours. Scale bar, 50 μm. e, Bright-field and fluorescence images of VIM–GFP in Spp1^WT/WT KPFV and Spp1^Δ/Δ KPFV organoids. Arrowheads indicate VIM–GFP⁺ cells. Scale bar, 50 μm. f, RT–qPCR analysis of expression of the epithelial markers Krt19, Cdh1 and Epcam and the mesenchymal markers Fn1, S100a4 and Vim between Spp1^WT/WT KPFV and Spp1^Δ/Δ KPFV organoids (n = 3 biological replicates), normalized using Spp1^WT/WT KPFV values. g, Images of subcutaneous tumours formed from Spp1^WT/WT KPF and Spp1^Δ/Δ KPF organoids. Scale bar, 0.5 cm. h, Sizes of subcutaneous tumours formed from Spp1^WT/WT KPF and Spp1^Δ/Δ KPF organoids (n = 7 mice). Data are the mean ± s.e.m. (c,f,h). P values were calculated using two-sided t-tests. The diagram in a was created in BioRender. Ruiz, J. (2025) (https://biorender.com/32og8hu). Source Data

Fig. 2. CD61 is required for MPC identity.

a, Immunofluorescence analysis of CD61 (red), YFP (grey) and SPP1 (green) expression in KPCY orthotopic tumours (n = 3 mice). Scale bar, 50 μm. b, Comparative RT–qPCR analysis of the epithelial markers Krt19, Cdh1 and Epcam and the mesenchymal markers Fn1, S100a4 and Vim in Itgb3^WT/WT KPFV and Itgb3^Δ/Δ KPFV organoids (n = 3 biological replicates). Values are normalized to Itgb3^WT/WT KPFV values. c, Comparative analysis of VIM–GFP expression between Itgb3^WT/WT KPFV and Itgb3^Δ/Δ KPFV organoids. Arrowheads indicate VIM–GFP⁺ cells. Scale bar, 50 μm. d, Images of subcutaneous tumours formed from Itgb3^WT/WT KPF and Itgb3^Δ/Δ KPF organoids. Scale bar, 0.5 cm. e, Sizes of subcutaneous tumours formed from Itgb3^WT/WT KPF and Itgb3^Δ/Δ KPF organoids (n = 4 mice). f, Immunofluorescence staining for GFP (grey), KRT19 (green) and VIM (red) of Itgb3^WT/WT KPF and Itgb3^Δ/Δ KPF subcutaneous tumours. Arrowheads indicate VIM⁺ cancer cells. Scale bar, 50 μm. g, Comparison of the percentages of different cancer cell subpopulations between subcutaneous tumours formed from Itgb3^WT/WT KPF and Itgb3^Δ/Δ KPF organoids (n = 3 mice). KRT19⁺VIM^– marks epithelial cancer cells, whereas KRT19⁺VIM⁺ marks EMT hybrid cancer cells, and KRT19^–VIM⁺ indicates mesenchymal cancer cells. Data are the mean ± s.e.m. (b,e,g). P values were calculated using two-sided t-tests. Source Data

Fig. 3. Inactivation of Spp1 inhibits PDAC development and metastasis formation.

a, Schematic showing how the Spp1^fl/fl KPFCT mouse model was generated. b, Haematoxylin and eosin (H&E) staining of Spp1^WT/WT KPFCT and Spp1^fl/fl KPFCT tumours (n = 3 mice). Scale bar, 1.0 mm. c, Immunofluorescence staining for GFP (grey), KRT19 (green) and VIM (red) in Spp1^WT/WT KPFCT and Spp1^fl/fl KPFCT tumours. Scale bar, 200 μm. d, Comparison of the proportions of different cancer cell subpopulations between Spp1^WT/WT KPFCT and Spp1^fl/fl KPFCT tumours (n = 5 mice). e, Kaplan–Meir survival curves of Spp1^WT/WT KPFCT (n = 15) and Spp1^fl/fl KPFCT (n = 16) mice. The P value was calculated using the log-rank test. f, Kaplan–Meir survival curves of Spp1^WT/WT KPCY (n = 14) and Spp1^fl/fl KPCY (n = 14) mice. The P value was calculated using the log-rank test. g, Images of liver and lung metastases in Spp1^WT/WT KP^hetFCT and Spp1^fl/fl KP^hetFCT mice. Arrowheads indicate metastases. h, Percentage of mice with liver metastases (left) and quantification of the number of liver metastases (right) in Spp1^WT/WT KP^hetFCT (n = 15) and Spp1^fl/fl KP^hetFCT (n = 20) mice. i, Percentage of mice with lung metastases (left) and quantification of the number of lung metastases (right) in Spp1^WT/WT KP^hetFCT (n = 15) and Spp1^fl/fl KP^hetFCT (n = 20) mice. Data are the mean ± s.e.m. (d,h (right), i (right)). P values were calculated using two-sided Fisher-exact tests (h (left) and i (left)) or two-sided t-tests (d,h (right), i (right)). Source Data

Fig. 4. SPP1 induces BMP2 through CD61 to maintain MPC fate.

a, Volcano plot showing gene expression profiles between Spp1^WT/WT KPFV and Spp1^Δ/Δ KPFV organoids. Blue dots indicate genes with higher expression in Spp1^Δ/Δ KPFV organoids, and red dots indicate genes with higher expression in Spp1^WT/WT KPFV organoids. P values were calculated using edgeR (two-sided). b, RT–qPCR analysis of Grem1 and Bmp2 expression in Spp1^WT/WT KPFV and Spp1^Δ/Δ KPFV organoids (n = 3 biological replicates). Values are normalized to Spp1^WT/WT KPFV values. c, RT–qPCR analysis of BMP2 target genes Id1, Id2, Id3 and Grem1 in Itgb3^WT/WT KPFV and Itgb3^Δ/Δ KPFV organoids (n = 3 biological replicates), normalized to Itgb3^WT/WT KPFV values. d, RT–qPCR analysis of Spp1, Id1, Id2, Id3, Bmp2 and Grem1 levels in KPF organoids treated with recombinant BMP2 protein or vehicle (n = 3 biological replicates), normalized to vehicle values. e, Immunofluorescence staining of GFP (grey), KRT19 (green) and VIM (red) in WT control (Spp1^WT/WTGrem1^WT/WT KPFCT), Spp1 homozygous knockout (Spp1^fl/flGrem1^WT/WT KPFCT), Grem1 homozygous knockout (Spp1^WT/WTGrem1^fl/fl KPFCT), Grem1 heterozygous knockout (Spp1^WT/WTGrem1^fl/WT KPFCT), Grem1 heterozygous knockout combined with Spp1 homozygous knockout (Spp1^fl/flGrem1^fl/WT KPFCT) and Grem1 and Spp1 double-homozygous knockout mice (Spp1^fl/flGrem1^fl/fl KPFCT) (n = 5 mice). Scale bar, 100 μm. f, Quantification of the proportion of KRT19⁺VIM^– and KRT19^–VIM⁺ cancer cells in e. Data are the mean ± s.e.m. (b–d,f). P values were calculated using two-sided t-tests. Source Data

Extended Data Fig. 1. The role of SPP1 in human and mouse PDAC.

a, UMAP plot of single cells from a scRNA-seq dataset for human PDACs (24 PDAC and 11 normal pancreas samples). Ductal cell type 1 and 2 represent the two main PDAC cancer cell populations. b, Cancer cells were categorized into the high SPP1 expression population (SPP1^hi, n = 6980 cells) and the low SPP1 expression population (SPP1^lo, n = 6981 cells) groups based on the expression of SPP1. c, EMT signature score plot for SPP1 high (n = 6980 cells) and SPP1 low (n = 6981 cells) cancer cell populations. d, Normalised RNA read counts (RSEM) for SPP1 from the PDAC dataset of GSE71729, ICGCarray, ICGCseq, TCGA, GSE62452 and GSE79668. The samples were categorised into SPP1^hi (n = 321) and SPP1^lo groups (n = 323) based on its expression, higher or lower than the median value. e, GSEA plot for enrichment of the hallmark EMT gene set in the transcriptome of SPP1^hi PDACs. NES, normalized enrichment score; FDR, false discovery rate. f, Kaplan-Meir test for the survival analysis of SPP1 high expression patients (n = 321) and SPP1 low expression patients (n = 323). P value was calculated with the logrank test. g, KPFV pancreatic cancer mouse model schematic diagram. h, Validation of CRISPR/Cas9 Spp1 knockout by Western blot. n = 3 independent experiments. Gel source data are provided in Supplementary Fig. 1. i, H&E and immunohistochemistry for Krt19 and Vimentin in KPFV; Spp1^wt/wt (n = 3 mice) and KPFV; Spp1^Δ/Δ allografts (n = 2 mice). Scale bar: 2.5 mm. j, Immunofluorescence staining for DAPI (Blue), GFP (gray), Krt19 (green), and Vimentin (red) of KPF; Spp1^wt/wt (n = 3 mice) and KPF; Spp1^Δ/Δ subcutaneous tumours (n = 2 mice). Scale bar is 50um. For b and d data, the lines in violin plots indicate the median, the upper quartile and the lower quartile respectively. For b, c, d data, P value is from two-sided Mann-Whitney test.

Extended Data Fig. 2. Spp1 overexpression induces mesenchymal transition of PDAC.

a, Immunofluorescent analysis of Vimentin (yellow), DAPI(blue), and Krt19 (red) in KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression organoids. Scale bar length is 50μm. b, RT-qPCR analysis of the expression of epithelial markers Krt19, Cdh1, and Epcam, and mesenchymal markers Fn1, S100a4, and Vim between KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression organoids (n = 3 biological replicates), normalised using KP^R172H/+C; Plv-con values. c, Immunofluorescent staining for GFP (gray), Krt19 (green), and Vimentin (red) in KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression allografts. n = 3 mice. Scale bar: 50um. d, H&E and Immunohistochemistry for Krt19 in KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression allografts. n = 3 mice. Scale bar: 1.0 mm. For b data are mean ± s.e.m. P values were calculated using two-sided t-tests. Source Data

Extended Data Fig. 3. The role of SPP1 in human PDAC development.

a, RT-qPCR analysis of the expression of epithelial markers KRT19, CDH1, and EPCAM, and mesenchymal markers FN1, S100A4, and VIM between Human PDAC; SPP1^wt/wt and Human PDAC; SPP1^Δ/Δ organoids (n = 3 biological replicates), normalised using SPP1^wt/wt organoids values. b, Validation of CRISPR/Cas9 SPP1 knockout by Western blot. n = 3 independent experiments. Gel source data are provided in Supplementary Fig. 1. c, Image of the subcutaneous tumours formed by Human PDAC; SPP1^wt/wt and Human PDAC; SPP1^Δ/Δ organoids. Scale bar length is 0.5 cm. d, The sizes of subcutaneous tumours of (c) (n = 6 mice). e, Immunofluorescence staining for KRT19 (green), and VIMENTIN (red) of subcutaneous tumours from (c). Scale bar is 50um. f, Comparison of the percentages of different cancer cell subpopulations of (e) (n = 5 mice). g, h, H&E and immunohistochemistry for KRT19 of subcutaneous tumours from (c). n = 3 mice. Scale bar: 2 mm. i, RT-qPCR analysis of the expression of epithelial markers KRT19, CDH1, and EPCAM, and mesenchymal markers FN1, S100A4, and VIM between Human PDAC; Plv-con and Human PDAC; SPP1 overexpression organoids (n = 3 biological replicates), normalised using Plv-con organoids values. j, Image of the subcutaneous tumours formed by Human PDAC; Plv-con and Human PDAC; SPP1 overexpression organoids. Scale bar length is 0.5 cm. k, The sizes of subcutaneous tumours of (j) (n = 6 mice). l, Immunofluorescence staining for KRT19 (green), and VIMENTIN (red) of subcutaneous tumours from (j). Scale bar is 50um. m, Comparison of the percentages of different cancer cell subpopulations of (l) (n = 5 mice). n, o, H&E and immunohistochemistry for KRT19 of subcutaneous tumours from (j). n = 3 mice. Scale bar: 2 mm. For a, d, f, i, k, m data are mean ± s.e.m. P values were calculated using two-sided t-tests. Source Data

Extended Data Fig. 4. Spp1 to maintain identity of MPCs by Cd61 and the effect of tumor stromal composition and immune cell infiltration upon Spp1 deletion.

a, Flow cytometry for Cd61 in Cd61^wt/wt and Cd61^Δ/Δ PDAC organoids. b, GSEA of the hallmark EMT gene set between of KPFV; Cd61^wt/wt and KPFV; Cd61^Δ/Δ transcriptomes. c, Flow cytometry for Cd45⁻/Cd31⁻/GFP⁻ in tumors from KPFCT; Spp1^wt/wt and KPFCT; Spp1^fl/fl mice. d, Comparison of the percentages of Cd45⁻/Cd31⁻/GFP⁻ in tumors from KPFCT; Spp1^wt/wt and KPFCT; Spp1^fl/fl mice (n = 5 mice). e, Immunofluorescence staining for aSMA (red), GFP (green), and Vimentin (Pink) of KPFCT; Spp1^wt/wt and KPFCT; Spp1^fl/fl tumours (n = 5 mice). Scale bar is 50um. f, Comparison of the percentages of aSMA cell subpopulations with Cancer cells between tumours formed by KPCTF; Spp1^wt/wt and KPFCT; Spp1 ^fl/fl. g-m, Comparison of the percentages of Cd4 T cells (g), Cd8 T cells (h), B220 B cells (i), NK cells(j), CD11b monocytes cells (k), F4/80 macrophages cells (l), dendritic cells (m), neutrophils cells (n) in tumors from KPFCT; Spp1^wt/wt and KPFCT; Spp1 ^fl/fl mice (n = 5 mice). For d, f, g, h, i, j, k, l, m and n data are mean ± s.e.m. P values were calculated using two-sided t-tests. Source Data

Extended Data Fig. 5. Loss of Spp1 inhibited PDAC development and metastasis formation.

a, H&E and immunohistochemistry of Krt19 in tumors from KP^hetFCT; Spp1^wt/wt and KP^hetFCT; Spp1^fl/fl mice. n = 3 mice. Scale bar: 2.5 mm. b, Immunofluorescence staining for GFP (gray), Krt19 (green), and Vimentin (red) of KP^hetFCT; Spp1^wt/wt and KP^hetFCT; Spp1^fl/fl tumours. n = 3 mice. Scale bar: 50 μm. c, H&E and immunohistochemistry for Vimentin in livers from KP^hetFCT; Spp1^wt/wt (n = 15 mice) and KP^hetFCT; Spp1^fl/fl mice (n = 20 mice). Scale bar: 10 mm. d, H&E and immunohistochemistry for Vimentin in lungs from KP^hetFCT; Spp1^wt/wt (n = 15 mice) and KP^hetFCT; Spp1^fl/fl mice (n = 20 mice). Scale bar: 5 mm.

Extended Data Fig. 6. Spp1 overexpression promotes PDAC metastasis formation.

a, Ultrasound imaging of pancreatic tumours in Pancreas Orthotopic transplantation mice. The dashed red circles indicate the tumours in the pancreas. b, H&E immunohistochemistry in tumour from KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression Pancreas Orthotopic transplantation mice. n = 3 mice. Scale bar: 10 mm. c, Percentage of mice with of liver metastases (left) and quantification of number of liver metastases (right) in KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression Pancreas Orthotopic transplantation mice (n = 4 mice). d, Percentage of mice with of lung metastases (left) and quantification of number of lung metastases (right) in KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression Pancreas Orthotopic transplantation mice (n = 4 mice). e, Immunohistochemistry for Vimentin in livers from KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression Pancreas Orthotopic transplantation mice. n = 4 mice. Scale bar: 10 mm. f, Immunohistochemistry for Krt19 in lung from KP^R172H/+C; Plv-con and KP^R172H/+C; Spp1 overexpression Pancreas Orthotopic transplantation mice. n = 4 mice. Scale bar: 5 mm. For c(right) and d(right) data are mean ± s.e.m. c (left) and d (left) P values were calculated using two-sided Fisher exact test. c (right) and d (right) P values were calculated using two-sided t-tests. Source Data

Extended Data Fig. 7. Spp1 antibody treatment induces epithelial differentiation by blocking MPCs in pancreatic cancer.

a, Scheme depicting the experimental approach for Spp1 neutralization experiments in KP^R172H/+CY Model. b, Kaplan-Meir test for the survival analysis of KP^R172H/+CY mice treat with Spp1 antibody or IgG control. P value was calculated with the logrank test. c, Percentage of mice with of liver metastases (left) and quantification of number of liver metastases (right) in KP^R172H/+CY mice treat with Spp1 antibody or IgG control (n = 12 mice). d, Percentage of mice with of lung metastases (left) and quantification of number of lung metastases (right) in KP^R172H/+CY mice treat with Spp1 antibody or IgG control (n = 12 mice). e, Immunofluorescent staining for YFP (gray), Krt19 (green), and Vimentin (red) in KP^R172H/+CY mice treat with Spp1 antibody or IgG control. Scale bar: 50um. f, Comparison of the proportions of different cancer cell subpopulations between KP^R172H/+CY mice treat with Spp1 antibody or IgG control tumours (n = 5 mice). g, H&E immunohistochemistry in tumors from KP^R172H/+CY mice treat with Spp1 antibody or IgG control. n = 3 mice. Scale bar: 2 mm. h, Immunohistochemistry of Krt19 in tumors from KP^R172H/+CY mice treat with Spp1 antibody or IgG control. n = 3 mice. Scale bar: 2 mm. i, Immunohistochemistry for Krt19 in livers from KP^R172H/+CY mice treat with Spp1 antibody or IgG control mice. n = 12 mice. Scale bar: 5 mm. j, Immunohistochemistry for Krt19 in lung from KP^R172H/+CY mice treat with Spp1 antibody or IgG control. n = 12 mice. Scale bar: 2.5 mm. For c(right), d(right) and f data are mean ± s.e.m. c (left) and d (left) P values were calculated using two-sided Fisher exact test. c (right), d (right) and f, P values were calculated using two-sided t-tests. The diagram in a was created in BioRender. Ruiz, J. (2025) (https://biorender.com/adqe56v). Source Data

Extended Data Fig. 8. Molecular analysis of Bmp2 regulation by Spp1 and Bmp2 indirectly modulates Spp1 expression by regulating mesenchymal cell fate.

a, RT-qPCR analysis of the expression of Id1, Id2, Id3 and Grem1 in control and Cd61 overexpressing KPFV organoids. Values are normalised to control. b, GSEA of the TNFα NF-κB hallmark gene set between KPFV; Cd61^wt/wt and KPFV; Cd61^Δ/Δ organoids. c, RT-qPCR analysis of the NF-κB signaling pathway target genes Ccl2, Clap2, and Tnf in KPFV; Cd61^wt/wt and KPFV; Cd61^Δ/Δ organoids. Values are normalised to the KPFV; Cd61^wt/wt values. d, RT-qPCR analysis of Ccl2, Clap2, and Tnf expression in control and Cd61 overexpressing KPFV organoids. Values are normalised to the control. e, Potential binding sites of RelA in the Bmp2 promoter were predicted using JASPAR. f, Schematic diagram of the binding sites of RelA to the Bmp2 promoter. g, ChIP analysis for RelA binding to the Bmp2 promoter and Tnf promoter in KPFV organoids. ChIP was conducted using IgG control or RelA antibody, with or without treatment with TAK1 inhibitor (5Z-7), followed by qPCR analysis. Values are normalised to the input. h, RT-qPCR analysis of the expression of Bmp2 in Vehicle and TAK1 inhibitor (5Z-7) treated organoids. Values are normalised to vehicle treatment. i, RT-qPCR analysis of the expression of Spp1, Id1, Id2, Id3, Bmp2 and Grem1 in KPF organoids treated with the Bmp2 inhibitor LDN193189 and vehicle. Values are normalised to the vehicle. 5Z-7: 5Z-7-oxozeaenol. j, Comparative RT-qPCR analysis of Id1, Id2, Id3, Spp1, Krt19 and Vim, in KPF organoids treated with Bmp2 for 0, 3, 6, 9 and 24 h. Values are normalised to Bmp2 0 h treatment. k, Comparative RT-qPCR analysis of Id1, Id2, Id3, Spp1, Krt19 and Vim, in KPF organoids treated with LDN193189 for 0, 3, 6, 9 and 24 h. Values are normalised to the LDN193189 0 h treatment. For a, c, d, g, h, i, j, k data are mean ± s.e.m and n = 3 biological replicates. P values were calculated using two-sided t-tests. Source Data

Extended Data Fig. 9. Histological analysis of Spp1 and Grem1 compound mutant mice.

a, Schematic diagram showing the mouse model KPFCT; Grem1^fl/fl. b, Schematic diagram showing the KPFCT; Spp1^fl/fl; Grem1^fl/fl mouse model. c, Immunohistochemical staining of Krt19 in control (KPFCT; Spp1^wt/wt; Grem1^wt/wt), homozygous Spp1 knock out (KPFCT; Spp1^fl/fl; Grem1^wt/wt), homozygous Grem1 knock out (KPFCT; Spp1^wt/wt; Grem1^fl/fl), heterozygous Grem1 knock out (KPFCT; Spp1^wt/wt; Grem1^fl/wt), heterozygous Grem1 knock out combined with homozygous Spp1 knock out (KPFCT; Spp1^fl/fl; Grem1^fl/wt) and homozygous double knock out mice (KPFCT; Spp1^fl/fl; Grem1^fl/fl). n = 3 mice. The scale bar: 500 μm.

Extended Data Fig. 10. Grem1 deletion reverses the epithelial cell state induced by Spp1 knock out.

a, Scheme depicting the experimental approach for establishing tumours from Spp1 CRISPR/Cas9 knockout and Tamoxifen-inducible Grem1 knockout organoids. b, RT-qPCR analysis of the expression of epithelial markers Krt19, Cdh1, and Epcam, and mesenchymal markers Fn1, S100a4, and Vim in organoids from either KPFC; Spp1 ^wt/wt; Grem1^wt/wt, KPFC; Spp1^Δ/Δ; Grem1^fl/fl organoids and subsequently treated with either vehicle or with 4-OH Tamoxifen as noted in the figure (n = 3 biological replicates); all data was normalised against the KPFC; Spp1 ^wt/wt; Grem1^wt/wt group. c, Image of the subcutaneous tumours generated by the implantation of KPFC; Spp1^Δ/Δ; Grem1^fl/fl organoids into nude mice and subsequently treated with either vehicle or with Tamoxifen as noted in the figure. Scale bar length is 1 cm. d, Quantification of the subcutaneous tumours detailed in (c) (n = 5 mice). e, Immunofluorescence staining for GFP (gray), Krt19 (green), and Vimentin (red) of KPFC; Spp1^Δ/Δ; Grem1^fl/fl & GFP (treated with vehicle) and KPFC; Spp1^Δ/Δ; Grem1^fl/fl & GFP (treated with Tamoxifen) subcutaneous tumours. Scale bar is 1 mm. f, Comparison of the proportions of epithelial cancer cell subpopulations between the tumours detailed in (e) (n = 5 mice). g, Comparison of the proportions of mesenchymal cancer cell subpopulations the tumours detailed in (e). h, Immunohistochemistry for H&E from subcutaneous tumours formed by KPFC; Spp1^Δ/Δ; Grem1^fl/fl &GFP treated with vehicle or Tamoxifen. n = 3 mice. Scale bar: 1 mm. For b, d, f and g data are mean ± s.e.m. P values were calculated using two-sided t-tests. The diagram in a was created in BioRender. Ruiz, J. (2025) (https://biorender.com/2uaw5xd). Source Data

Extended Data Fig. 11. Grem1 overexpression reverses the mesenchymal cell state established by initial Spp1 overexpression.

a, Scheme depicting the experimental approach for establishing tumours from Spp1 overexpression and doxycycline-inducible Grem1 overexpression organoids. b, RT-qPCR analysis of the expression of epithelial markers Krt19, Cdh1, and Epcam, and mesenchymal markers Fn1, S100a4, and Vim in organoids from either KPF; Plv-GFP&pINDUCER-con or KPF; Plv-Spp1-GFP&pINDUCER-Grem1 mice treated with either vehicle or doxycycline as noted in the figure (n = 3 biological replicates). c, Image of the subcutaneous tumours formed by KPF; Plv-Spp1-GFP&pINDUCER-Grem1 organoids treated with either vehicle or Doxycycline as indicated. Scale bar length is 1 cm. d, Quantification of the subcutaneous tumours detailed in (c). e, Immunofluorescence staining for GFP (gray), Krt19 (green), and Vimentin (red) of the subcutaneous tumours formed by KPF; Plv-Spp1-GFP&pINDUCER-Grem1 organoids treated with either vehicle or Doxycycline as indicated (n = 5 mice). Scale bar is 1 mm. f, Comparison of the proportions of epithelial cancer cell subpopulations between of the subcutaneous tumours detailed in (e) (n = 5 mice). g, Comparison of the proportions of mesenchymal cancer cell subpopulations of the subcutaneous tumours detailed in (e) (n = 5 mice). h, Immunohistochemistry for H&E from subcutaneous tumours formed by KPF; Plv-Spp1-GFP&pINDUCER-Grem1 treated with vehicle or Doxycycline. n = 3 mice. Scale bar: 1 mm. For b, d, f and g data are mean ± s.e.m. P values were calculated using two-sided t-tests. The diagram in a was created in BioRender. Ruiz, J. (2025) (https://biorender.com/2uaw5xd). Source Data

Extended Data Fig. 12. Model for the mechanism of Spp1, Bmp2 and Grem1 in maintaining cellular heterogeneity in pancreatic cancer.

Schematic cartoon depicting the mechanism for the paracrine interaction between epithelial and mesenchymal cancer cells. This diagram was created in BioRender. Ruiz, J. (2025) (https://biorender.com/e3c38ka).