SMAD4 induces opposite effects on metastatic growth from pancreatic tumors depending on the organ of residence

Abstract

The role of driver gene mutations in sustaining tumor growth at metastatic sites is poorly understood. SMAD4 inactivation is a paradigm of such mutations and a hallmark of pancreatic ductal adenocarcinoma (PDAC). To determine whether metastatic tumors are dependent on SMAD4 inactivation, we developed a mouse model of PDAC that enables spatiotemporal control of Smad4 expression. While Smad4 inactivation in the premalignant pancreas facilitated the formation of primary tumors, Smad4 reactivation in metastatic disease suppressed liver metastases but promoted lung metastases. These divergent effects were underpinned by organ-biased differences in the tumor cells' chromatin state that emerged in the premalignant pancreas and were distinguished by the dominance of KLF4 versus RUNX1 transcription factors. Our results show how epigenetic states favored by the organ of residence can influence the output of driver mutations in metastatic tumors, which has implications for interpreting tumor genetics and therapeutically targeting metastatic disease.

Fig. 1. A Smad4-restorable GEMM of PDAC.

a, Schematic of GEMM alleles. rtTA3, third-generation reverse tetracycline transactivator; TRE, tetracycline response element. b, Overall survival of KC-shRen mice and KC-shSmad4 mice (expressing one of two independent Smad4 shRNAs: 591 or 1599) after Dox administration (n = 7 KC-shRen mice; n = 11 KC-shSmad4 mice). Statistical analysis was conducted using a log-rank test. c, H&E staining of primary tumors and liver and lung metastases from KC-shSmad4 mice. Data are representative of ten, six and three independent mice, respectively. d, sWGS analysis of the Cdkn2a/b locus in KC-shSmad4 tumor-derived cell lines (n = 10 independent mice). e, Western blot analysis of Dox response in vitro. Data are representative of six independent KC-shSmad4 cell lines. Source Data

Fig. 2. Smad4 restoration has organ-specific effects on tumor growth.

a, Schematic of orthotopic, intrasplenic and tail-vein injection experiments. This image was created using BioRender.com. b, Analysis of primary tumor growth after orthotopic transplantation of KC-shSmad4 cells and subsequent Smad4 restoration. Left, fold-change quantifications of tumor volume on day 30 versus day 0 of Dox withdrawal (mean ± s.e.m.; n = 8 independent mice per group). Different color shading indicates independent cell lines. Right, representative ultrasound images of tumors (demarcated by dashed yellow lines). c, Analysis of liver metastasis burden after intrasplenic injections of KC-shSmad4 cells and subsequent Smad4 restoration. Left, percentage area quantifications on day 30 after Dox withdrawal (mean ± s.e.m.; n = 6 independent mice per group). Different color shading indicates independent cell lines. Right, representative macroscopic images of tumor-bearing livers. d, Analysis of lung metastasis burden after tail-vein injections of KC-shSmad4 cells and subsequent Smad4 restoration. Left, fold-change quantifications of bioluminescence signal on day 30 versus day 0 of Dox withdrawal (mean ± s.e.m.; n = 8 and 7 independent mice per group, respectively). Different color shading indicates independent cell lines. Right, representative bioluminescence images of tumor-bearing mice. Statistical analyses were conducted using an unpaired two-sided t-test (b) or two-sided Mann–Whitney U-test (c,d). Source Data

Fig. 3. Smad4 induces different transcriptional programs in liver versus lung metastases.

a, GO analysis of upregulated genes in liver or lung metastases 7 or 14 days after Dox withdrawal. Combined scores and P values for the top KEGG (left) or Hallmark (right) pathways are shown for Smad4 on versus off for each organ and time point. Complete GO lists are provided in Supplementary Table 2. b, Heat map of representative genes from Smad4’s cytostatic or apoptotic (tumor-suppressive) and fibrogenic (tumor-promoting) transcriptional programs. The average RNA-seq log₂ fold change (FC) and P values are shown for Smad4 on versus off for each organ and time point. c, Representative IF staining for Ki67 in KC-shSmad4 liver and lung metastases ± Smad4 restoration. mKate2 was used to label tumor cells. Right, quantifications (n = 12, 15, 16 and 14 independent tumors from three, four, three and three mice for the groups shown, from left to right). Analysis was performed 7 days after Dox withdrawal. d, Representative IF staining for α-SMA in KC-shSmad4 liver and lung metastases ± Smad4 restoration. mKate2 was used to label tumor cells. Right, quantifications (n = 12, 10, 12 and 11 independent tumors from three, three, three and four mice for the groups shown, from left to right). Analysis was performed at the experimental endpoint (30 days for liver; 45 days for lungs). Statistical analyses were conducted using Fisher’s exact test with Benjamini–Hochberg adjustment (a), a Wald test with negative binomial modeling and Benjamini–Hochberg adjustment (b), an unpaired two-sided t-test (c) or a two-sided Mann–Whitney U-test (d). Source Data

Fig. 4. Smad4’s tumor-suppressive function depends on p57.

a, Representative macroscopic images of tumor-bearing livers ± Smad4 restoration on the background of the indicated stable shRNAs. Arrowheads highlight metastatic tumors. Data are representative of the number of independent mice per group specified in b. b, Quantifications of metastasis burden (fold change of percentage tumor area) on day 40 after Dox withdrawal (mean ± SEM; n = 9, 8, 10 and 7 independent mice per groups shown, from left to right). c,d, Representative IF staining (c) and quantifications (d) for Ki67 in KC-shSmad4 liver metastases ± Smad4 restoration on the background of the indicated stable shRNAs (n = 14, 18, 10 and 12 independent tumors from three, four, three and four mice for groups shown, from left to right). Analysis was performed 7 days after Dox withdrawal. Statistical analysis was conducted using an unpaired two-sided t-test (b,d). Source Data

Fig. 5. Liver and lung metastases harbor distinct chromatin states.

a, Heat map of differentially accessible chromatin regions in tumor (mKate2⁺) cells isolated from the pancreas, liver or lungs (in vivo ATAC-seq). Smad4 on corresponds to 14 days of Dox withdrawal. Each column represents an independent mouse. A complete list of differentially accessible ATAC-seq peaks is provided in Supplementary Table 3. b, Top-scoring TF motifs identified by HOMER de novo motif analysis of in vivo ATAC-seq peaks enriched in liver versus lung metastases. Enrichment P values are shown in parentheses. c, Analysis of liver metastasis burden after intrasplenic injections of lung metastasis-derived KC-shSmad4 cell lines with or without subsequent Dox withdrawal. Left, percentage area quantifications at endpoint (mean ± s.e.m.; n = 5 and 6 independent mice per group, respectively; different color shading indicates independent cell lines). Right, representative macroscopic images of tumor-bearing livers. Shown are matching brightfield and mKate2 fluorescence images. Arrowheads highlight metastatic tumors. d, Analysis of lung metastasis burden after tail-vein injections of liver metastasis-derived KC-shSmad4 cell lines with or without subsequent Dox withdrawal. Left, percentage area quantifications at endpoint (mean ± s.e.m.; n = 10 and 7 independent mice per group, respectively; different color shading indicates independent cell lines). Right, representative macroscopic images of tumor-bearing lungs. Shown are matching brightfield and mKate2 fluorescence images. Arrowheads highlight metastatic tumors. e,f, Quantifications of the percentage of Ki67⁺ tumor cells in KC-shSmad4 lung-to-liver (e) and liver-to-lung (f) metastases ± Smad4 restoration. Right, representative IF images (n = 8 and 6 independent tumors from two mice per respective group for liver; n = 19 and 15 independent tumors from five mice per respective group for lung). mKate2 was used to label tumor cells. Analysis was performed 7 days after Dox withdrawal. g,h, Quantifications of the percentage of p57⁺ cells in KC-shSmad4 lung-to-liver (g) and liver-to-lung (h) metastases ± Smad4 restoration. Right, representative IHC images (n = 6 and 5 independent tumors from two mice per respective group for liver; n = 6 independent tumors from five mice per group for lung). mKate2 was used to label tumor cells. Analysis was performed 7 days after Dox withdrawal. Statistical analyses were conducted using a hypergeometric test (b) or unpaired two-sided t-test (c–h). Source Data

Fig. 6. Liver and lung metastasis-like chromatin states emerge early in tumorigenesis.

a, UMAP visualization of scATAC-seq profiles of KC-shSmad4 (+Dox) primary tumor cells (mKate2⁺). Signature scores based on liver-specific or lung-specific ATAC-open peaks from bulk ATAC-seq data are displayed in color per individual cell. Data represent three independent mice. b, Mutual exclusivity of liver-specific and lung-specific open chromatin signatures in primary tumors. The plot shows the distribution of cells from KC-shSmad4 (+Dox) primary tumors according to their enrichment of liver-specific (LiverOPEN) and lung-specific (LungOPEN) open chromatin signatures derived from bulk ATAC-seq (Methods). Each dot corresponds to an independent cell. The proportions of cells with liver-only (blue) and lung-only (red) signatures were compared to those with both liver and lung signatures (green) and those with neither signature (purple). c, GO analysis of upregulated genes in liver metastasis-like versus lung metastasis-like cell subpopulations of KC-shSmad4 primary tumors. Combined scores and P values for the top KEGG pathways are shown. Complete GO lists are provided in Supplementary Table 4. d, UMAP visualization of scRNA-seq profiles of KC-shRen and KC-shSmad4 (+Dox) premalignant pancreatic tissue (mKate2⁺). Cell type classification based on established gene signatures (Methods) are displayed in color per cell. Data represent four independent mice. e, Assessment of cell states enriched in shRen (red) versus shSmad4 (blue) conditions using Milo analysis (Methods). f, Distribution of premalignant pancreas cell neighborhoods in shRen or shSmad4 mice across transcriptional states. The x axis indicates the log₂(fold change) in differential abundance of shRen (<0) and shSmad4 (>0). Each neighborhood was associated with a cell type if more than 80% of the cell states in the neighborhood belonged to the respective state; otherwise, it was annotated as ‘mixed’. g, Fractions of premalignant pancreas cells corresponding to different transcriptional states under the indicated conditions (n = 1 mouse per condition; Methods). PBS was used as the control. Cer, cerulein. Statistical analyses were conducted using a chi-squared test (b) or Fisher’s exact test with Benjamini–Hochberg adjustment (c). Source Data

Fig. 7. KLF4 and RUNX1 are associated with liver-biased and lung-biased chromatin states.

a, ATAC-seq and RNA-seq combined scores for the indicated TF families in liver versus lung metastases. This metric infers the probability that a given TF family with a significantly enriched motif in the ATAC-open regions impacts SMAD4-induced gene expression changes on the basis of a consistent RNA-seq change in the Smad4 on versus off comparison (Methods). Top TF families scored using the HOMER de novo motif analysis in the liver versus lungs are shown. b, Combined scores for the indicated TF families in liver and lung metastases from PDAC patients. This metric infers the activity of a given TF family on the basis of enrichment or depletion of its predicted target genes in the respective metastases versus primary tumors but not in the corresponding normal tissues (Methods). Top TFs scored using JASPAR or TRANSFAC position weight matrix data are shown. c, Representative IHC staining for KLF4 and RUNX1 in KC-shSmad4 (+Dox) liver or lung metastases. Right, higher-magnification views of the dashed areas to highlight tumor-specific nuclear signal. Data are representative of 20 metastases from four independent mice. Source Data

Fig. 8. Klf4 and Runx1 depletions antagonize Smad4 in an organ-specific manner.

a, Representative macroscopic images of tumor-bearing livers ± Smad4 restoration on the background of the indicated stable shRNAs. Arrowheads highlight metastatic tumors. Insets, shSmad4-linked GFP reporter. Data are representative of the number of independent mice per group specified in b. For shRen, the same cohort as shown in Fig. 4 was used. b, Quantifications of metastasis burden (fold change of percentage tumor area) on day 30 after Dox withdrawal (mean ± s.e.m.; n = 9, 8, 7, 8, 9 and 9 independent mice per group, respectively). For shRen, the same cohort as shown in Fig. 4 was used. c, Representative bioluminescence images of tumor-bearing lungs ± Smad4 restoration on the background of the indicated stable shRNAs. Data are representative of the number of independent mice per group specified in d. d, Quantifications of metastasis burden (fold change of bioluminescence signal) on day 21 after Dox withdrawal (mean ± s.e.m.; n = 7, 8, 8, 8, 8 and 7 independent mice per group, respectively). Statistical analysis was conducted using an unpaired two-sided t-test (b,d). Source Data

Extended Data Fig. 1. Characterization of the KC-shSmad4 mouse model.

(a) Overall (pie chart) and organ-specific (stacked bar graph) frequency of metastasis in KC-shSmad4 mice. (b) Macroscopic images of tumor-bearing pancreas, liver, and lungs from KC-shSmad4 mice ( + Dox) at endpoint. Fluorescent images match the corresponding brightfield images. Data are representative of 10 (pancreas), 6 (liver), and 3 (lungs) independent mice. (c) Immunohistochemistry (IHC) staining for SMAD4 in primary and metastatic KC-shSmad4 tumors ( + Dox). Dashed lines demarcate metastases. Data are representative of 3 independent mice. (d) sWGS analysis of genome-wide copy number alterations in KC-shSmad4 primary tumor-derived cell lines (n = 10 independent mice). Frequency plot is shown on the top and individual sample tracks are provided on the bottom. (e) sWGS analysis of genome-wide copy number alterations in KC-shSmad4 tumor-derived cell lines from matched primary tumors, liver and lung metastases. The Kras and Cdkn2a/b loci are highlighted (arrows). Data are representative of 3 independent mice. (f) Frequency of the indicated homozygous deletions in SMAD4-altered (mutated or homozygously deleted, n = 959) or wild-type (WT, n = 3188) PDAC tumors in the MSK-IMPACT cohort (CDKN2A/B, p < 10⁻¹⁰; q < 10⁻¹⁰). CDKN2A/B and their adjacent gene MTAP are highlighted. Source Data

Extended Data Fig. 2. Smad4 restoration in vitro and in murine liver and lung metastases, and SMAD4 expression in human metastatic recurrence.

(a) Proliferation assays of KC-shRen and KC-shSmad4 cells +/-Dox +/-TGFβ in vitro (mean + SEM; n = 3 and 4 independent cell lines per genotype, respectively). (b) Brightfield images of KC-shRen and KC-shSmad4 cells after TGFβ treatment +/-Dox for 6 days. Data represent 3 independent cell lines per genotype. (c) H&E staining of liver and lung metastases generated by intrasplenic or tail vein injections of KC-shSmad4 cell lines. Data represent 4 independent mice per organ. (d) Percent-area quantification of liver metastasis burden at day 0 of Dox withdrawal after intrasplenic injections of KC-shSmad4 cell (n = 8 independent mice). Different color shading indicates independent cell lines. (e) H&E staining of liver and lung metastases from KC-shSmad4 cells without (Smad4 OFF) or with Smad4 restoration (Smad4 ON). Data are representative of 6 independent mice per organ. (f) Schematic of orthotopic experiments with KC-shRen and KC-shSmad4 cells for analysis of metastasis burden. Created in BioRender. Tsanov, K. (2025) https://BioRender.com/yihgg0q. (g) Analysis of liver metastasis burden after orthotopic injections of KC-shRen or KC-shSmad4 cells with or without subsequent Dox withdrawal. (Left) Percent-area quantifications at endpoint (mean + SEM; n = 5, 3, 8 and 8 independent mice for groups shown, left to right). (Right) Representative macroscopic images of tumor-bearing livers. (h) Analysis of lung metastasis burden after orthotopic injections of KC-shRen or KC-shSmad4 cells with or without subsequent Dox withdrawal. (Left) Percent-area quantifications at endpoint (mean + SEM; n = 4, 5, 8 and 8 independent mice for groups shown, left to right). (Right) Representative macroscopic images of tumor-bearing lungs. Different color shading in (g-h) indicates independent cell lines, arrowheads highlight metastatic tumors, and insets show mKate2 reporter. (i) (Left) Schematic of metastasis recurrence studies in human PDAC^,. Created in BioRender. Tsanov, K. (2025) https://BioRender.com/dt9i4jd. (Right) Fraction of patients whose tumors stained SMAD4-positive or -negative by IHC, according to site of recurrence. Exact numbers are indicated on each bar. Statistical analyses were two-way ANOVA (a), unpaired two-sided t-test (g, h), and Fisher’s exact test (i). Source Data

Extended Data Fig. 3. Effects of Smad4 restoration on gene expression and signaling.

(a) Overlap between upregulated genes in tumor cells from the pancreas, liver, or lungs at 7 and 14 days after Dox withdrawal. Numbers reflect genes in each category. Complete gene lists are provided in Supplementary Table 1. DEGs = differentially expressed genes. (b) Normalized Smad4 mRNA levels (RNA-seq data; tpm = transcripts per million) in KC-shSmad4 tumor cells from the pancreas, liver, or lungs +/-Dox for 7 or 14 days (n = 5, 5, 6, 5, 6 and 8 independent mice for groups shown, left to right). Box plots represent median +/- upper/lower quartile (box) +/-1.5 x IQR between the first and third quartile. (c) Representative IHC staining for pSMAD2 in KC-shSmad4 tumors ( + Dox) from the pancreas, liver, or lungs. Quantifications (%pSMAD2 cells) are shown on the right (n = 6 independent primary tumor regions or metastatic tumors from 2 mice per organ). (d) Heatmap of differentially expressed genes from the indicated KEGG pathways. Average log2 fold-change and p-values are shown for each organ and timepoint. (e) Heatmap of SMAD4-dependent SMAD2/3 ChIP target genes that are differentially upregulated in liver vs. lung metastases at 7 or 14 days after Dox withdrawal. Average RNA-seq log₂ fold-change values are shown for Smad4 ON vs. OFF in each organ and timepoint. The type of genomic binding region is color-annotated on the left. Complete gene lists are provided in Supplementary Table 1. (f) Fraction of upregulated genes in liver and/or lung metastases at 7 or 14 days of Dox withdrawal that are SMAD4-dependent SMAD2/3 ChIP-seq targets. Absolute number of genes per group is shown in parentheses. Statistical analyses were one-way ANOVA (b, c), and Wald test with negative binomial modeling and Benjamini-Hochberg adjustment (d). Source Data

Extended Data Fig. 4. Impact of Smad4 restoration on pathways in vitro vs. in vivo, on primary tumor phenotypes, and in GEMMs.

(a) Comparison of gene ontology (GO) analyses of upregulated genes after Dox withdrawal in vitro (x axis) and in vivo in the pancreas, liver or lungs (y axis). See Methods for experimental details. Combined scores, correlation coefficients (ρ) and correlation p-values for the enriched KEGG pathways are shown for each organ. Complete GO lists and combined scores are provided in Supplementary Table 2. (b) Representative immunofluorescence staining for Ki67 in KC-shSmad4 primary tumors +/-Smad4 restoration. mKate2 labels tumor cells. Quantifications are shown on the right (n = 8 independent tumor regions from two mice per group, respectively). Analysis was performed 7 days after Dox withdrawal. (c) Representative immunofluorescence staining for a-SMA in KC-shSmad4 primary tumors +/-Smad4 restoration. mKate2 labels tumor cells. Quantifications are shown on the right (n = 9 and 6 independent tumor regions from two mice per group, respectively). Analysis was performed at experimental endpoint (45 days after Dox withdrawal). (d) Representative immunofluorescence staining for Ki67 in KC-shSmad4 GEMM primary tumors and liver and lung metastases +/-Smad4 restoration. mKate2 labels tumor cells. Quantifications are shown on the right (n = 8, 8, 7, 5, 8 and 8 independent primary tumor regions or metastatic tumors from 3, 4, 4, 3, 3 and 3 mice for groups shown, left to right). Analysis was performed 7 days after Dox withdrawal. (e) Western blot analysis of shRNA efficiency. Smad4-restored cells were pre-treated with TGFb to induce p57 expression. Actin was used as a loading control. (f) Quantifications of baseline metastasis burden (% tumor area) after intrasplenic injections of KC-shSmad4 cells under continuous Dox administration (mean + SEM; n = 9 and 10 independent mice per group, respectively). Livers were harvested 68 days after injection. Statistical analyses were Pearson correlation (a), and unpaired two-sided t-test (b, c, d, f). Source Data

Extended Data Fig. 5. Analysis of in vivo ATAC-seq data.

(a) Principal component analysis (PCA) of ATAC-seq data from tumor (mKate2⁺) cells isolated from the pancreas, liver, or lungs (based on peak normalization). Each sample corresponds to an independent mouse. Circled areas highlight separation based on organ. (b) Number of statistically significant differentially accessible in vivo ATAC-seq peaks. Exact numbers of gained/lost peaks are shown for each comparison (absolute fold change ≥1.5; FDR ≤0.1). (c) Representative in vivo ATAC-seq tracks at loci with liver-specific, lung-specific, or liver/lung-shared chromatin opening in KC-shSmad4 tumor cells (mKate2⁺GFP⁺). ‘Housekeeping’ genes are shown as a reference. Data are representative of 3 independent tumors. y-axis scale range is indicated per lane as normalized read counts. Source Data

Extended Data Fig. 6. Analysis of in vitro ATAC-seq data and tumor-derived cell lines.

(a) Heatmap of differentially accessible chromatin regions in Smad4 OFF tumor-derived cell lines ( + Dox) from the pancreas, liver, or lungs. Each column represents an independent cell line. A complete list of differentially accessible ATAC-seq peaks is provided in Supplementary Table 3. (b) Principal component analysis (PCA) of in vitro ATAC-seq data (based on peak normalization). Each sample corresponds to an independent cell line. Circled areas highlight separation based on organ site. (c) Top-scoring TF motifs identified by HOMER de novo motif analysis of in vitro ATAC-seq peaks enriched in liver vs. lung metastasis-derived cell lines. The numbers in parentheses indicate enrichment p-values. (d) Proliferation assays of KC-shSmad4 tumor-derived cell lines from the pancreas, liver, or lungs +/-Dox +TGFβ in vitro (n = 3 independent experiments per group). Statistical analyses were hypergeometric test (c), and one-way ANOVA (d). Source Data

Extended Data Fig. 7. Analysis of liver and lung metastasis-like chromatin and transcriptional states in murine premalignant pancreas and human PDAC.

(a) UMAP visualization of scATAC-seq profiles of Kras-mutant pancreatic epithelial cells (mKate2⁺) +/- cerulein injury (n = 1 mouse for each) from ref. (see Methods for details). Signature scores based on liver- or lung-specific ATAC-open peaks from bulk ATAC-seq data are displayed in color per individual cell. (b) Schematic of scRNA-seq experiment in premalignant pancreas. Created in BioRender. Tsanov, K. (2025) https://BioRender.com/52xbslm. (c) (Left) UMAP visualization of scRNA-seq profiles of KC-shRen and KC-shSmad4 ( + Dox) pre-malignant pancreatic tissue (mKate2⁺) (identical to Fig. 6d). (Middle & Right) Signature scores based on liver- or lung-specific genes corresponding to organ-enriched ATAC-open peaks are displayed in color per individual cell. Data represent 4 independent mice. (d) UMAP visualization of scRNA-seq profiles of human primary PDAC tumor cells from ref. (see Methods for details). Signature scores based on scRNA-seq profiles of mouse liver- or lung-specific ATAC-open cell populations (from the matching scATAC-seq multiomics data) are displayed in color per individual cell. (e) Gastric- and progenitor-like signature scores corresponding to panel (d). (f) Correlation analysis of liver/lung metastasis-like vs. gastric/progenitor-like transcriptional signatures. The size and color of each circle reflects correlation strength. Exact correlation scores are shown for each comparison. Statistical analysis by Pearson correlation. Source Data

Extended Data Fig. 8. Analysis of KLF and RUNX TF expression in liver and lung metastases from murine transplant models, GEMMs, and PDAC patients.

(a) Representative IHC staining for the indicated TFs in KC-shSmad4 ( + Dox) liver or lung metastases. Data are representative of 20 metastases from 4 independent mice. (b) Normalized Klf4 and Runx1 mRNA levels (RNA-seq data; tpm = transcripts per million) in KC-shSmad4 tumor cells from the pancreas, liver, or lungs ( + Dox for 7 or 14 days) (n = 5, 6, 6, 5, 6 and 6 independent mice for groups shown, left to right). Box plots represent median +/- upper/lower quartile (box) +/-1.5 x IQR between the first and third quartile. (c) Representative IHC staining for KLF4 and RUNX1 in KC-shSmad4 ( + Dox) GEMM liver or lung metastases. Higher magnifications of dashed areas are shown on the right to highlight tumor-specific nuclear signal. Quantification of the fraction of KLF4+ or RUNX1+ metastases is shown on the right (n = 10 and 7 independent tumors from 5 and 4 mice for liver and lung, respectively). (d) Representative IHC staining for KLF4 and RUNX1 in liver or lung metastases from PDAC patients. Higher magnifications of dashed areas are shown on the right to highlight tumor-specific nuclear signal. Quantification of the KLF4 or RUNX1 IHC signal is shown on the right (n = 8 and 9 independent tumors from 7 and 8 patients for liver and lung, respectively). Statistical analyses were one-way ANOVA (b), chi-squared test with Yates correction (c), and two-sided Mann-Whitney U test (d). Source Data

Extended Data Fig. 9. Analysis of KLF4 and RUNX1 expression in murine and human primary PDAC tumors.

(a) Representative IHC staining for KLF4 and RUNX1 in serial sections of a primary tumor from the KC-shSmad4 ( + Dox) transplant model (left), autochthonous model (middle), and human PDAC patient (right). Data are representative of 4 (transplant model) and 6 (autochthonous model) independent mice and 8 patients. (b) Net change in ATAC-RNA combined scores for the KLF and RUNX TF families in Smad4 ON vs. OFF primary tumors at day 14. This metric infers the probability that a given TF family is impacted by Smad4 restoration, based on a combined change in ATAC-seq motif accessibility and RNA-seq levels of its predicted target genes (see Methods for details). (c) Representative IHC staining for KLF4 or RUNX1 in KC-shSmad4 primary tumors +/-Smad4 restoration. Quantifications are shown on the right (n = 6, 6, 5 and 5 independent tumors per group, respectively). Analysis was performed 14 days after Dox withdrawal. Statistical analysis was unpaired two-sided t-test. Source Data

Extended Data Fig. 10. Analysis of Klf4 and Runx1 depletions in vitro and in vivo.

(a) Western blot analysis of shRNA efficiency. Actin was used as a loading control. (b) Proliferation assays of KC-shSmad4 cell lines harboring the indicated stable shRNAs +/-Dox and +/-TGFβ in vitro (mean + SEM; n = 3 independent experiments per group). (c) Schematic of experiments to assess early metastatic colonization of KC-shSmad4 cells with TF knockdowns. Created in BioRender. Tsanov, K. (2025) https://BioRender.com/ufvotxg. (d-e) Bioluminescent analysis of livers (d) and lungs (e) harvested 7 days after intracardiac injection of cells expressing the indicated shRNAs. Quantification is shown on the left and representative bioluminescent images on the right (mean + SEM; n = 9, 10 and 10 independent mice per organ for groups shown, left to right). (f-g) Schematic of intrasplenic (f) and tail vein (g) experiments with KC-shSmad4 cells with TF knockdowns. Created in BioRender. Tsanov, K. (2025) https://BioRender.com/ur1sduv and https://BioRender.com/hxar0yr. (h) Quantifications of baseline liver metastasis burden (% tumor area) after intrasplenic injections of KC-shSmad4 cells under continuous Dox administration (Smad4-OFF) (mean + SEM; n = 9, 7 and 9 independent mice per group, respectively). Livers were harvested 64-66 days after injection. (i) Quantifications of baseline lung metastasis burden (bioluminescent signal) after tail vein injections of KC-shSmad4 cells under continuous Dox administration (Smad4-OFF) (mean + SEM; n = 7, 8 and 8 independent mice per group, respectively). Lungs were harvested 57 days after injection. (j) Model of the organ-specific interplay between SMAD4 and KLF4/RUNX1-associated chromatin states. In the liver, active KLF4 allows SMAD4 to act at genes that promote cytostasis, thus necessitating the inactivation of Smad4. In the lungs, more speculatively, active RUNX1 allows SMAD4 to act at pro-fibrotic genes unopposed by its cytostatic program. Partially created in BioRender. Tsanov, K. (2025) https://BioRender.com/74y5yim. Source Data