Drug screening in 3D microtumors reveals DDR1/2-MAPK12-GLI1 as a vulnerability in cancer-associated fibroblasts

Abstract

Interactions between cancer cells and surrounding stromal cells are critical for tumor biology and treatment response. We compare drug screening results from conventional 2D cancer cell lines with 3D tumor tissues and find that, on average, three times more drugs are effective in 3D microtumors. We confirm the effectiveness of doramapimod, a compound that reduces microtumor viability and suppresses tumor growth in mouse models but has no effect on cancer cell growth in monolayers. Mechanistically, doramapimod targets DDR1/2 and MAPK12 kinases in cancer-associated fibroblasts (CAFs), decreasing extracellular matrix (ECM) production and enhancing interferon signaling. These kinases regulate ECM through GLI1 activity in CAFs, independently of canonical hedgehog signaling. Inhibiting the DDR1/2-MAPK12-GLI axis enhances the effectiveness of chemotherapy and immunotherapy in patient tumor slices and preclinical models. These findings highlight the importance of DDR1/2-MAPK12-GLI axis in CAF function and demonstrate the utility of 3D tissue models in identifying microenvironment-specific therapeutic targets.

Keywords: CAF; functional screening; hedgehog pathway; kinase signaling; microtumor; precision oncology; tumor microenvironment.

Graphical abstract

Figure 1

Comparative drug screening in 2D culture and 3D tissues reveals tumor-specific vulnerabilities (A) Schematic overview of the experimental workflow. Fresh tumor tissues were processed into microtumors. Drug screening was conducted in both 2D cultured cells and microtumors systems, followed by machine learning-based analysis to identify effective compounds. (B) An unexpectedly high number of drugs are effective in 3D tissues compared to 2D cultures. A bar plot showing the number of effective compounds identified in microtumors versus 2D monolayer cultures across three tumor models: E0771 (breast), Py8119 (breast), and KPC (pancreas). A significantly higher number of effective drugs were identified in the microtumors. (C) Overlap of effective drugs for breast and pancreatic cancers in 3D tissue slices. The Venn diagram displays the overlap of drugs predicted to be effective against breast and pancreatic cancers in microtumors. A core set of 12 compounds was found to be commonly effective. (D) The top 3D tumor-specific effective drugs. Heatmaps comparing the responses in microtumors versus 2D cultured cells for the top tumor-specific effective drugs in E0771 (left) and KPC (right) models. (E) Top kinases targeted by kinase inhibitors specifically in microtumors. A plot showing the fraction of tumor-specific effective drugs that inhibit the indicated kinases, based on measured kinase inhibition profiles at 500 nM concentration. (F) Drug response variability across cancer types in 2D and 3D microtumors. A boxplot showing the response of various cancer types to a set of 12 drugs in 2D culture and microtumor models.

Figure 2

Doramapimod treatment suppresses tumor growth in multiple cancer models (A) Efficacy of doramapimod in desmoplastic tumors. Bar graph showing the viability (% of control) of microtumors from various tumor types following treatment with doramapimod (500 nM) or DMSO control. Data represent the mean ± SEM from two-three biological replicates. (B) Minimal effect of doramapimod in 2D cultures. The relative viability of 492 cancer cell lines from diverse lineages treated with doramapimod at 2.5 μM. (C) In vivo efficacy of doramapimod in the E0771 TNBC model. Left: tumor growth curves in mice bearing E0771 tumors treated with systemic doramapimod (30 mg/kg daily) or vehicle. Right: final tumor weights at endpoint. n = 9/group. Unpaired Student’s t test. ∗p < 0.05). (D) Doramapimod treatment in an orthotopic pancreatic cancer mouse model. Left: a schematic of experimental design showing KPC cells implanted into the pancreas of 6-week-old C57BL/6 mice to establish an orthotopic pancreatic cancer model. Mice were treated daily with doramapimod (40 mg/kg) for two weeks. Right: tumor weight and representative images of doramapimod or vehicle-treated KPC tumors. n = 4 (control); n = 5 (treatment). Student’s t test. ∗∗p < 0.01. (E) Tumor growth of 4T1 tumors treated with doramapimod. Left: growth curves of 4T1 tumors in BALB/cJ female mice with subcutaneous 4T1 tumor, treated twice a week with either doramapimod (1 mg/kg) or vehicle control via intratumoral injections. Black arrows indicate the timing of drug administrations. Tumor weight at the endpoint. n = 7 (control); n = 6 (treatment). Right: tumor growth of breast PDX model (HCI001) in mice treated with doramapimod. Doramapimod (1 mg/kg) or vehicle control was intratumorally injected twice a week into HCI-001 PDX tumor-bearing mice. n = 5 (control); n = 5 (treatment). For time course tumor volume measurement, statistical significance was calculated by two-tailed, unpaired multiple Student’s t tests with Holm-Sidak correction. For tumor weight, Student’s t test. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001. Error bars indicate SEM.

Figure 3

Doramapimod inhibits the expression of ECM-associated genes in pancreatic and breast tumors (A) Transcriptional changes in tumor tissues following doramapimod treatment. Volcano plots show differentially expressed genes (DEGs) in KPC (left) and E0771 (right) tumor tissues. Genes significantly upregulated or downregulated (>2-fold, p < 0.05) in doramapimod-treated tumors versus control are highlighted in red and blue, respectively. n = 3 tumors per group. (B) Pathway enrichment of downregulated genes. Functional enrichment of genes downregulated by doramapimod across Gene Ontology, WikiPathways, Reactome, and KEGG databases. ECM-related categories are highlighted in red, indicating suppression of matrix organization and remodeling. (C) Proteomic analysis of ECM components. Heatmaps show mass spectrometry-based quantification of ECM-related proteins in KPC (left) and E0771 (right) tumors treated with doramapimod versus vehicle. n = 3 tumors per group. (D) Reduction of CAF populations in vivo. xCell analysis of bulk RNA-seq data from 4T1 and KPC tumors indicates a reduction in CAF signatures following doramapimod treatment. (E) Gene expression changes in breast CAFs. Volcano plot shows significantly upregulated and downregulated genes (>2-fold, p < 0.05) in primary CAF cultures treated with doramapimod (n = 3). (F) Pathway enrichment in CAFs. Dot plot showing enriched pathways among doramapimod-downregulated genes in breast CAFs. ECM-related pathways are labeled in red. (G) Functional protein network analysis. Protein interaction network of downregulated genes in doramapimod-treated CAFs, generated using Enrichr-KG. Clusters are labeled based on key functional pathways including ECM-receptor interaction, TGF-β signaling, and axon guidance.

Figure 4

Doramapimod unexpectedly targets DDR1/2 and MAPK12, regulating extracellular matrix gene expression in cancer-associated fibroblasts (A) Kinome profiling of doramapimod. Left: residual kinase activity for 370 kinases treated with 500 nM doramapimod using a radioactive ATP assay. Kinases with <20% residual activity are indicated in red. Right: bar chart highlighting top inhibited kinases. (B) CAF gene expression after kinase knockdown. Heatmap showing changes in ACTA2 and CXCL12 expression in breast cancer-derived CAFs following siRNA knockdown of doramapimod target kinases. (C) Plot showing qPCR analysis of CXCL12 expression in breast CAFs following siRNA-mediated knockdown of DDR1, DDR2, and MAPK12, with or without doramapimod treatment. Data represent mean ± SEM; n = 3 per group. Student’s t test, compared with DMSO control. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. (D) Transcriptional changes in CAFs upon DDR1, DDR2, or MAPK12 depletion. Volcano plots showing DEGs (>2-fold, p < 0.05) after siRNA knockdown of DDR1 (left), DDR2 (middle), or MAPK12 (right) in primary breast CAFs. Downregulated genes are shown in blue and upregulated in red. n = 3 per group. (E) Pathway enrichment of downregulated genes following combined DDR1/2 and MAPK12 knockdown. Bar graph showing enrichment across Gene Ontology, Reactome, and KEGG pathways, with ECM-related processes highlighted in red. (F) Neutralization of CAFs’ growth stimulatory effect through depletion of DDR1/2 and MAPK12 kinase expression. Top: schematic of the experimental design showing breast CAFs treated with siRNAs against the indicated kinases, then co-cultured with 4T1 cancer cells labeled with nuclear GFP. Bottom: growth curve of 4T1 cells co-cultured with kinase depleted CAFs, displayed as Mean ± SEM. n = >3 in each group. (G) DDR1/2 enhances p38 phosphorylation. Western blot analysis showing elevated levels of phospho-p38 in CAFs overexpressing DDR1 or DDR2 compared to GFP control. Total p38 and β-actin are shown as loading controls. (H) Proposed model of doramapimod action. Doramapimod inhibits the DDR1/2–MAPK12 signaling axis, which drives ECM production in CAFs.

Figure 5

DDR1/2 and MAPK12 converge at GLI to regulate ECM production and support cancer cell growth (A) Gene overlap and expression analysis. Left: a Venn diagram showing the overlap of genes downregulated in response to knockdown of DDR1/2 and MAPK12 in CAFs. Right: a heatmap displaying changes in the expression of 7 genes commonly downregulated by depletion of DDR1/2 and MAPK12. The transcriptional factors, ECM proteins, and immune regulatory roles of these genes are highlighted. (B) Doramapimod reduces GLI1 nuclear localization in CAFs. Left: representative immunofluorescence images showing reduced nuclear GLI1 signal in CAFs following doramapimod treatment (1 μM for 48 h). Right: quantification of nuclear GLI1 intensity from at least 20 cells per group, indicating significantly reduced nuclear localization. Scale bars, 20 μm; ∗∗p < 0.01; Student’s t test. (C) GLI1 transcriptional activity is inhibited by doramapimod, DDR1/2, and MAPK12 knockdown. Left: luciferase reporter assay in pancreatic CAFs shows decreased GLI1 transcriptional activity upon treatment with doramapimod (1 μM) or GANT61 (1 μM). Right: similar reduction in GLI1 activity observed upon knockdown of DDR1, DDR2, or MAPK12. Data represent mean ± SEM from two-three biological replicates; ∗p < 0.05 and ∗∗p < 0.01; unpaired Student’s t test. (D) GLI1 inhibition downregulates ECM-associated pathways. Pathway enrichment plots based on RNA-seq of breast and pancreatic CAFs treated with GANT61. Genes involved in ECM regulation, including integrin signaling, collagen formation, and matrix remodeling, are significantly downregulated (adjusted p values indicated by color scale). (E) DDR1/2 promotes nuclear localization of GLI. Representative images showing that overexpression of DDR1/2 in normal human pancreatic fibroblasts promotes nuclear localization of phosphorylated p38 MAPK and GLI1. Scale bars, 20 μm. (F) Neutralization of CAFs' growth stimulatory effect by GLI depletion. Left: plot showing growth of GFP-labeled 4T1 cancer cells on CAFs with intact or depleted levels of GLI1, displayed as mean ± SEM. N = >3 in each group. Student’s t tests with Holm-Sidak correction. ∗∗p < 0.01. Right: Gant61 treatment does not affect 4T1 cancer cell growth directly in serum-supported monoculture, but significantly reduces 4T1 cell growth when co-cultured with CAFs in serum-free condition, displayed as mean ± SEM. n indicates at least 3 replicates in each group. (G) Model of the non-canonical hedgehog pathway in CAFs. Schematic illustrating DDR1/2-mediated activation of p38/MAPK12 and GLI drives ECM production in CAFs and supports cancer cell growth.

Figure 6

Doramapimod enhances the efficacy of chemotherapy and immunotherapy in human pancreatic tumor slices and preclinical models (A) Doramapimod promotes apoptosis in pancreatic tumor slices. Left: quantification of the percentage of cleaved caspase-3+ cells in three slices from a PDA-47 patient treated with DMSO, 3 μM doramapimod (low), and 5 μM doramapimod (high) over 6 days. Data represent mean ± SEM from three biological replicates. Middle: quantification of cleaved caspase-3+ cells in slices from a PDA-49 patient treated with DMSO, 3 μM doramapimod, 1 μM gemcitabine, and a combination of 3 μM doramapimod and 1 μM gemcitabine over the same period. Data represent mean ± SEM from three biological replicates. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001 by one-way ANOVA with multiple comparisons. Right: representative IHC images of cleaved caspase-3 staining from PDA-49 slices under indicated treatments. Scale bars, 50 μm. (B) Doramapimod reduces collagen deposition in ex vivo pancreatic tumors. Quantification of collagen-positive staining in PDA-49 tumor slices treated with DMSO, doramapimod (5 μM), gemcitabine (1 μM), or combination for 6 days. Data represent mean ± SEM from three biological replicates. (C) In vivo combination treatment with doramapimod and gemcitabine enhances tumor suppression in orthotopic pancreatic cancer model. Mice were implanted orthotopically with KPC cells in the pancreas and treated with either vehicle, doramapimod, gemcitabine, or the combination. Each dot represents an individual mouse (n = 7–11 per group); ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 by one-way ANOVA with multiple comparisons. Error bars indicate SEM. (D) Doramapimod activates immune-related pathways. Pathway enrichment analysis of upregulated genes in E0771 (left) and KPC (right) tumors following doramapimod treatment, showing activation of interferon, cytokine signaling, and antigen processing pathways (highlighted in red). Pathways were curated across Reactome, Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Ontology (GO) databases. (E) Combination of doramapimod and anti–PD-L1 results in enhanced tumor suppression in the E0771 TNBC model. Growth curves of E0771 tumors in mice treated with vehicle, doramapimod, anti-PD-L1 or the combination. n = 5–7 mice per group. Error bars indicate SEM. (F) Left: final tumor weights from E0771 tumor-bearing mice treated with vehicle, doramapimod, anti-PD-L1, or the combination. Doramapimod alone led to partial tumor growth inhibition (partial responses [PR]) in 5 of 7 mice, with no response (NR) in 2 mice. Anti-PD-L1 alone resulted in PR in all 5 mice. The combination treatment produced complete tumor regression (CR) in 3 mice and partial suppression in the remaining 3. Right: tumor weight inhibition (TWI) is plotted as a percentage relative to the average tumor weight of the vehicle control. Each dot or bar represents an individual mouse (n = 5–7 per group); ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 by one-way ANOVA with multiple comparisons. Error bars indicate SEM.