A Patient-Derived Organ-on-Chip Platform for Modeling the Tumor Microenvironment and Drug Responses in Pancreatic Cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is a fatal malignancy. Current conventional chemotherapeutics are inadequate in controlling the disease; hence, there is an urgent need for precision medicine. Ex vivo models that replicate the tumor and its microenvironment can advance precision medicine in PDAC. Patient-derived organoids (PDOs) offer a promising solution by retaining the functional features of the tumor, allowing for individualized study of cancer biology and drug response. However, PDOs fall short in replicating the tumor microenvironment (TME), which includes various stromal and immune cells influencing tumor growth and chemoresistance. We hypothesize that combining PDO technology with organ-on-a-chip (OoC) systems can enhance ex vivo cancer modeling. Here, we develop a patient-derived platform by incorporating PDOs with key components of the TME (fibroblasts, endothelial cells, and immune cells) within a microfluidic system. This OoC model represents the crosstalk between cancer and stroma observed in PDAC in vivo. Targeting the stroma improves the effectiveness of standard chemotherapy in this OoC. Further, using this platform, we are able to model and assess the efficacy of immune checkpoint blockade for T cell cytotoxicity in PDAC. The OoC provides a pathophysiologically applicable system to support future investigations aimed at utilizing precision medicine and testing therapeutics in PDAC.

Keywords: microfluidic; organ on chip; pancreatic cancer; patient derived organoids; tumor microenvironment.

FIGURE 1

Establishment of pancreatic organoids and immune cell isolation workflow. (A) Schematic diagram of the workflow for developing and maintaining human Pancreatic Ductal Organoids (PDOs) from patient tissue. Organoids are cultured in 3D Matrigel, passaged weekly, and maintained in complete organoid media. Created in BioRender. Adnan, D. (2025) https://BioRender.com/y93dwq2 (B) Representative brightfield images of (i) a normal PDO and (ii) a cancer PDO derived from human pancreas samples. (C) Variability of nuclear‐to‐cytoplasmic (N/C) ratio: cancer vs normal. Each dot = single cell (normal n = 16 cells, cancer n = 17 cells). Variance compared by Fligner–Killeen test (two‐sided): χ ²(1) = 14.19, *p < 0.001. Bars show variance on log10 scale ±SE. (D) PDO growth in culture from single cells: mean diameter over days for normal vs cancer lines. Analysis by linear mixed‐effects model (lme4) with fixed effects for Group, Day, and Group×Day and random intercept for PDO line; day effect β = 0.188, p = 0.004; cancer > normal β = 0.551 (log‐scale), p < 0.001; Group×Day interaction p = 0.027, bars show mean ± SE. (E) Immunofluorescence staining of PDOs from normal and cancer donors showing expression of CA19‐9, a pancreatic cancer marker. (F) IF staining of Cytokeratin‐19 (CK19) in normal (i) vs cancer (ii) PDOs (representative fields). (G) Quantified CK19 signal (log10 scale), n = 6 per group. Comparison by Wilcoxon rank‐sum test (two‐sided), showing CK19 expression in both groups, with higher levels observed in tumor‐derived PDOs consistent with known PDAC features: W = 36, p = 0.002. (H) Representative immunoblots for p‐ERK, total ERK, and β‐actin in PDO lysates. (I) p‐ERK/ERK ratio. Each dot = one PDO sample (cancer n = 3, normal n = 2). Welch's t‐test (two‐sided): t = 3.90, p = 0.037. Bars show mean ± SD. (J) Brightfield/IF images of isolated human monocytes. (K) IF confirming polarization: CD68⁺ M1 and CD163⁺ M2 macrophages (representative). (L–P) Cytokine profiling of polarized macrophages (M1 vs M2) at 1, 6, and 24 h. Each point n = 2 per group per time. Two‐sided Welch t‐tests per timepoint with Benjamini–Hochberg (BH) correction within each cytokine. Significant examples: IFN‐γ (M1>M2 at 1 h p = 0.0217, 6 h p = 0.00571, 24 h p = 0.0128), IL‐6 (all times p < 0.0415), TNF‐α (24 h p = 0.0149), IL‐4 (M2>M1 at 1 h and 24 h p = 0.0382; M1 6 h NA), IL‐10 (ns at all times). Lines show mean ± SD.

FIGURE 2

Building Patient‐Derived ex vivo model of PDAC. (A) Schematic diagram of the microfluidic chip platform (The Chip). The device consists of two chambers separated by a 3 µm porous membrane. The upper chamber mimics a vascular channel seeded with human umbilical vein endothelial cells (HUVECs), while the lower chamber represents the tumor compartment and contains a co‐culture of patient‐derived pancreatic ductal organoids (PDOs), primary pancreatic stellate cells (PSCs), and monocyte‐derived macrophages. Right panels show representative immunofluorescence staining confirming cellular components: CD31 (HUVECs), EpCAM (epithelial/tumor cells), α‐SMA (PSCs), and CD68 (macrophages). Created in BioRender. Adnan, D. (2025) https://BioRender.com/8x9rr7b (B) Timeline schematic detailing the sequential seeding of each cell type and initiation of flow via microfluidic pump. Created in BioRender. Adnan, D. (2025) https://BioRender.com/x5kxoqg (C) Representative brightfield images showing tumor growth in the Chip at days 0, 4, and 9. (D) Scatter plot quantifying PDO diameter over time. PDOs in the full multicellular model exhibited significantly greater growth than PDOs cultured alone Linear mixed‐effects model (random intercept for Chip): Group×Day interaction p = 0.0016. Follow‐up per‐day Welch tests with BH correction: Day 0 ns (p = 0.896), Day 4 p = 0.0478, Day 9 p < 0.001 (full>PDO only). n = 3 Chips per group. Points show Chip means; lines show mean ± SE. (E) Scatter plot illustrating percent change in surface area (S.A.%) of PDOs from day 0 to day 9. Linear mixed‐effects model: Group×Day p = 0.173; per‐day Welch tests BH‐adjusted: Day 4 ns (p = 0.183), Day 9 ns (p = 0.183); values consistently higher in full model. n = 3 Chips per group. (F‐G) Representative BF (F) and biopsy H&E (G) illustrating desmoplastic morphology (no stats). (H) Immunofluorescence staining of PDOs from normal and cancer donors showing differential expression of phosphorylated ERK (p‐ERK), indicating activation of MAPK signaling (representative). (I) Quantified p‐ERK intensity: n = 4 per group. Wilcoxon rank‐sum (two‐sided): W = 16, p = 0.02. Bars show mean ± SD.

FIGURE 3

Tumor‐derived signaling modulates stromal activation and fibroblast gene expression. (A) Immunofluorescence staining of tumor models containing PDOs, pancreatic stellate cells (PSCs), and monocytes, showing increased collagen deposition (red) and EpCAM+ PDOs (green), indicating stromal remodeling. (B) Immunofluorescence image of tumor model containing only PDOs and PSCs (without monocytes), showing reduced collagen content and less elongated PSC morphology compared to the full model in (A). (C) PSC activation quantified by spread area after co‐culture with cancer PDOs vs normal PDOs. Each dot = single PSC (cancer n = 13 cells, normal n = 12 cells). Wilcoxon rank‐sum: W = 24, p = 0.002. Bars show mean ± SE, dots = cells. (D) Schematic of the transwell co‐culture experiment in which primary pancreatic fibroblasts were seeded in the lower chamber and co‐cultured with either cancerous PDOs or Matrigel‐only controls in the upper insert (3 µm pore size), allowing for exchange of soluble factors. Created in BioRender. Adnan, D. (2025) https://BioRender.com/zc49f8x (E) PCA plot of RNA‐seq data from co‐cultured fibroblasts (n = 1 per group, fibroblasts monoculture vs fibroblasts + cancer PDOs), demonstrating clustering near cancer‐associated fibroblasts (CAFs) identified in a previously published single‐cell RNA‐seq dataset of PDAC. (F) Volcano plot showing significantly upregulated and downregulated genes in fibroblasts co‐cultured with tumor organoids versus controls. EdgeR with TMM normalization and BH‐FDR; genes significant at FDR<0.05. n = 1 per group (G) Heatmap displaying all 219 differentially expressed genes (FDR < 0.05) from the transwell co‐culture experiment. (H) Focused heatmap of selected genes involved in stellate cell activation and extracellular matrix remodeling. (I) Gene set enrichment analysis showing statistically significant hallmark pathways enriched in fibroblasts co cultured with cancerous PDOs, indicating activation of pro‐tumorigenic pathways. (J–O) Monocyte cytokines in Transwell Monocyte alone vs Monocyte co‐cultured with PDOs. Conditioned media collected at 1, 6, 24 h, n = 2 per group per timepoint. Two‐sided Welch t‐tests with BH within cytokine across times. IL‐6 higher at 1, 6, 24 h (p = 0.0108, 0.0272, 0.0233), TNF‐α higher at 24 h (p = 0.0022), IL‐1β higher at 6 and 24 h (p = 0.0090, 0.0090); IFN‐γ ns; IL‐10 and IL‐4 ns. Lines show mean ± SD. (P) MDS positioning shows a shift of Monocytes when co‐cultured with PDOs toward mixed polarization relative to M1/M2 references.

FIGURE 4

Evaluation of the tumor Chip model for drug response and stromal modulation. (A) Left: Representative brightfield image of PDOs cultured in an Elplasia 96‐well round‐bottom microplate prior to chemotherapy testing. Right: Dose‐response curve of gemcitabine treatment showing non‐linear regression used to calculate the IC₅₀ value based on five serial concentrations (100, 20, 4, 0.8, 0.16 µm). (B) Immunofluorescence images of tumor (control) or tumor + all‐trans retinoic acid (ATRA), demonstrating that ATRA reduces stromal density and PSC elongation, as evidenced by decreased collagen staining. (C) PSC extension length with or without ATRA in the full model. Each dot = PSC process length (Tumor n = 9 cells, Tumor+ATRA n = 15 cells). Wilcoxon rank‐sum: W = 110.5, p = 0.011. (D) Representative immunofluorescence images showing cleaved caspase‐3 (c.c.3⁺) and EpCAM⁺ cells in tumor Chips treated with DMSO (control), gemcitabine alone, or gemcitabine in combination with ATRA and Clodrosome (a macrophage‐depleting agent). (E) Quantification of apoptotic tumor cells (c.c.3⁺/EpCAM⁺ double‐positive cells) revealed significantly increased apoptosis in the combination treatment group (gemcitabine + ATRA + Clodrosome) compared to gemcitabine alone or DMSO control (DMSO n = 7, gemcitabine n = 6, gemcitabine+ATRA+Clodrosome n = 8). One‐way ANOVA: p < 0.001; Tukey post‐hoc: Combo > DMSO p < 0.001, Combo > gem p = 0.011, gem vs DMSO p = 0.084 (ns). Box/whiskers show median, IQR, 1.5 × IQR.

FIGURE 5

Incorporation of patient‐derived T cells into the tumor Chip reveals stromal effects on T cell infiltration and function. (A) Brightfield and immunofluorescence images showing T cells successfully migrating from the upper vascular chamber into the tumor compartment of the Chip containing PDO only (1K vs. 10K PDOs) and further in PDO + Stroma. (B) Bar plots showing quantification of T cell infiltration into the lower chamber demonstrating increased migration with higher tumor burden (10K vs. 1K PDOs) and further enhancement when stroma (PSC + Monocytes) were present, indicating that the tumor microenvironment facilitates T cell recruitment. 1K (n = 6), 10K (n = 4), 10K+Stroma (n = 5). One‐way ANOVA with Tukey: 10K>1K p = 0.009, 10K+Stroma>1K p < 0.001, 10K+Stroma vs 10K p = 0.15 (ns). Bars show mean ± SD. (C) Bar plot indicating a significantly higher T cell count at the lower chamber in the full model Chips (PDOs + PSCs + Monocytes) compared to the PDOs only Chips. PDO only (n = 3) vs PDO+Stroma (n = 6). Welch's t‐test: p = 0.028. (D) Despite increased T cell infiltration, quantification of cancer cell apoptosis (e.g., cleaved caspase‐3⁺/EpCAM⁺ cells) revealed reduced tumor cell killing in Chips with stromal components compared to PDO‐only Chips, suggesting that stromal cells contribute to T cell dysfunction in this ex vivo model of pancreatic cancer. PDO only (n = 3) vs PDO+Stroma (n = 6). Welch's t‐test: p = 0.028. Bars show mean ± SD. (E) Schematic of the transwell co‐culture experiment, where tumor models (PDOs + PSCs + monocytes) or Matrigel controls were placed in the upper insert, while activated T cells were seeded in the lower well, allowing for exchange of signaling without direct contact. Created in BioRender. Adnan, D. (2025) https://BioRender.com/zc49f8x (F) PCA plot shows that blood‐derived bulk RNA‐seq data from T cells co‐cultured with tumor models cluster closely with tissue‐derived tumor‐infiltrating T cells from a previously published single‐cell RNA‐seq dataset. (G) Heatmap displaying all significantly up‐ and downregulated genes in T cells co‐cultured with the tumor model versus control, indicating substantial transcriptional changes, (edgeR; BH‐FDR<0.05, n = 1 sample per group). (H) Volcano plot highlighting differentially expressed genes from the transwell co‐culture, (edgeR; BH‐FDR<0.05, n = 1 sample per group). (I) Heatmap of selected exhaustion‐associated genes (e.g., CD38 and NOTCH2NL) in T cells co‐cultured with the tumor model, demonstrating the induction of an exhausted T cell phenotype, (edgeR; BH‐FDR<0.05, n = 1 sample per group).

FIGURE 6

(A–C) Representative flow plots showing memory subsets (CD45RA/CCR7) over time and PD‐1 expression after co‐culture. (D) Dose‐dependent response to PD‐1 blockades with pembrolizumab in the PDAC tumor Chip model. Quantification of T cell infiltration into the tumor chamber following treatment with low‐dose (10 µg/mL) or high‐dose (100 µg/mL) pembrolizumab. Control (n = 6), pembrolizumab 10 µg/mL (n = 7), pembrolizumab 100 µg/mL (n = 6). One‐way ANOVA: F(2,16) = 4.64, p = 0.026; Tukey: 100 µg/mL > 10 µg/mL p = 0.028; 100 µg/mL vs control p = 0.110 (ns); 10 µg/mL vs control p = 0.800 (ns). Bars show mean ± SD. (E) Representative immunofluorescence images of tumor Chips showing DAPI (nuclei, blue), cleaved caspase‐3 (C.C.3, red), and EpCAM (tumor marker, green). Merged images illustrate enhanced tumor cell apoptosis in Chips treated with high‐dose pembrolizumab. (F) Quantification of tumor cell apoptosis (percent cleaved caspase‐3⁺ among EpCAM⁺ cells). High‐dose pembrolizumab (100 µg/mL) significantly increased tumor cell killing compared to both the low‐dose group (10 µg/mL) and control. Control (n = 6), 10 µg/mL (n = 7), 100 µg/mL (n = 6). One‐way ANOVA: F(2,16) = 7.03, p = 0.006 (η ² = 0.47). Tukey: 100 µg/mL > control p = 0.007; 100 µg/mL > 10 µg/mL p = 0.028; 10 µg/mL vs control p = 0.709 (ns).