Bioprinted Patient-Derived Organoid Arrays Capture Intrinsic and Extrinsic Tumor Features for Advanced Personalized Medicine

Abstract

Heterogeneity and the absence of a tumor microenvironment (TME) in traditional patient-derived organoid (PDO) cultures limit their effectiveness for clinical use. Here, Embedded Bioprinting-enabled Arrayed PDOs (Eba-PDOs) featuring uniformly arrayed colorectal cancer (CRC) PDOs within a recreated TME is presented. This model faithfully reproduces critical TME attributes, including elevated matrix stiffness (≈7.5 kPa) and hypoxic conditions found in CRC. Transcriptomic and immunofluorescence microscopy analysis reveal that Eba-PDOs more accurately represent actual tissues compared to traditional PDOs. Furthermore, Eba-PDO effectively capture the variability of CEACAM5 expression-a critical CRC marker-across different patients, correlating with patient classification and differential responses to 5-fluorouracil treatment. This method achieves an uniform size and shape within PDOs from the same patient while preserving distinct morphological features among those from different individuals. These features of Eba-PDO enable the efficient development of a label-free, morphology-based predictive model using supervised learning, enhancing its suitability for clinical applications. Thus, this approach to PDO bioprinting is a promising tool for generating personalized tumor models and advancing precision medicine.

Keywords: colorectal cancer; embedded bioprinting; inter‐patient variability; patient‐derived tumor organoid; supervised learning; tumor matrix stiffness.

Figure 1

Creating tumor microenvironment (TME)‐inspired embedded bioprinting‐enabled arrayed patient‐derived organoids (Eba‐PDOs). A) A schematic illustrating embedded 3D bioprinting process of PDO‐ink based on a Geltrex™ hydrogel within an alginate bath, designed to mimic the natural colorectal cancer (CRC) tissue surrounded by a rigid matrix, highlighting its importance in cancer progression. B) Photograph showing the dispensation of PDO‐laden bio‐ink into an alginate bath (Scale bar = 1 mm). C) Haematoxylin‐eosin (H&E) staining images displaying Eba‐PDO formation within the alginate bath: self‐assembly of suspended PDO cells on day 4, fusion of assembled PDO cells on day 7, and unified Eba‐PDO formation on day 10 (Scale bar = 50 µm). D) Formation rate of a unified Eba‐PDO over time. E) A comparative illustration (left) and bright‐field microscopy images (right) display the differences between standard PDOs (Std‐PDO) within a Geltrex™ dome and Eba‐PDOs on day 14. Scale bar = 500 µm. F) A comparison of the area and variance of Std‐PDOs and Eba‐PDOs throughout the growth period. Data were obtained using PDOs derived from a single CRC patient (CEA_lo‐11, CEA_lo: Low‐CEACAM5). The results are presented as mean ± SEM.

Figure 2

Characterization of cancer hallmarks including morphological features and pathophysiological environments in Eba‐PDOs. A) H&E staining images displaying the morphological features of native CRC tissue, Std‐PDOs, and Eba‐PDOs (Scale bar = 100 µm (CRC Tissue left), 25 µm (CRC Tissue right, Std‐PDO, and Eba‐PDO)). B,C) Analysis of the proportion of luminal area (B) (n = 6 for Std‐PDOs, Eba‐PDOs and 7 for Tissues) and nuclear area (C) of native CRC tissue, Std‐PDOs, and Eba‐PDOs (n = 47 for Std‐PDOs, 52 for Eba‐PDOs, and 59 for Tissues). D–F) The relative mRNA expression of YAP1 (D), MMP2 (E), and CDH2 and CDH1 mRNA ratio (CDH2/CDH1) indicating the epithelial‐mesenchymal transition (F) throughout the PDO culture period. The quantitative real time polymerase chain reaction results are presented as mean ± SEM. For statistical analysis, Student's t‐test was performed between each culture day (n = 3 for Std‐PDOs and n = 6 for Eba‐PDOs). G–I) Immunofluorescence micrographic images representing the cancer pathophysiological environment including YAP/TAZ (indicative of mechanical stress) (G), HIF1α (indicative of hypoxia) (H), and Collagen I (indicative of ECM remodelling) (I). J–L) Mean fluorescent intensity (MFI) of immunofluorescence micropraphic images labeled with YAP/TAZ (J), HIF1α (K), and COL1A1 (L) in Std‐PDOs and Eba‐PDOs on days 7 and 14 (Scale bar = 50 µm). The absence of nonspecific staining from antibody binding was confirmed in Figure S8 (Supporting Information). Results are presented as mean ± SEM. For statistical analysis, a one‐way analysis of variance and Tukey's test were performed (^* p < 0.05, ^** p < 0.01, ^*** p < 0.001, ^**** p < 0.0001, ns: non‐significant for all statistical analysis). Data were obtained using PDOs and tissues derived from a single CRC patient (CEA_lo‐11).

Figure 3

Transcriptomic comparison of Eba‐PDOs compared to Std‐PDOs and the native CRC tissue (Tissue). A) Principal component analysis (PCA) with normalized counts from RNA‐seq data of all samples. B) Pearson”s correlation analysis with most frequently mutated 1004 genes in CRC from National Cancer Institute (https://portal.gdc.cancer.gov/). C) Comparison of Pearson correlation coefficient (r) values between Std‐PDO and Eba‐PDO platforms compared to the CRC tissue. The results are presented as mean ± SEM. For statistical analysis, Student”s t‐test was performed between each group (n = 6 for Std‐PDOs and Eba‐PDOs, ^**** p < 0.0001). Results are D) Volcano plot showing fold changes for genes differentially expressed between Std‐ and Eba‐PDOs. Genes upregulated and downregulated in Eba‐PDOs compared to Std‐PDOs are highlighted in magenta, and dark grey, respectively. p < 0.01. E) Kyoto Encyclopaedia of Genes and Genomes (KEGG) functional classification, and F) Gene Ontology (GO) terms of biological process showing genes upregulated and downregulated in Eba‐PDOs compared to Std‐PDOs. The size of each dot represents the number of differential genes in the enrichment pathway. D–F) Comparison of Differentially‐Expressed Genes (DEGs) between Eba‐PDOs and Std‐PDOs focuses on genes that display a similar expression pattern between the Eba‐PDOs and the original tissue samples. Data were obtained using PDOs and tissues derived from a single CRC patient (CEA_hi‐02).

Figure 4

Patterns of CEACAM5 expression in Eba‐PDOs influenced by matrix stiffness. A) A schematic representation of Std‐ and Eba‐PDOs in matrices of differing stiffness. B) The compressive modulus of alginate bath‐ink across different alginate concentrations; the red line shows the median compressive modulus of human CRC tissues and the green line indicates that of normal colorectal tissue (n = 3; independent experiments). C) Quantification of mean fluorescence intensity (MFI) within specified regions of interest (ROIs) in confocal immunofluorescent images of sectioned PDOs and normal (N) and tumor tissue (T) sections from the same patients marked with CEACAM5 (n = 6 for Tissue (N), 10 for Tissue (T), 8 for Std‐PDOs, and 13 for Eba‐PDOs, A.U.: Arbitrary Unit). Eba‐PDOs were prepared using 1.5% alginate bath‐ink. D) Quantification of MFI within ROI in confocal immunofluorescent images of sectioned Std‐PDOs and Eba‐PDOs marked with CEACAM5, across different alginate bath conditions (n = 8 for Std‐PDOs, 7 for Eba‐PDOs (0.5%), 9 for Eba‐PDOs (1%), 13 for Eba‐PDOs (1.5%), 7 for Eba‐PDOs (2%). E) Confocal immunofluorescence microscopic analysis of sectioned PDOs and tissue sections from the same patient, labeled with DAPI (blue) and CEACAM5 (green). Yellow arrows highlight CEACAM5 expression on the abluminal side of normal colon and CRC tissues, and PDOs, with their abluminal side defined as the side oriented toward the matrix. Yellow asterisks represent the lumen of tissues or PDOs. Scale bar = 250 µm. F) The MFI of sectioned PDOs and tissues was measured for ROIs designated as luminal and abluminal regions (n = 7 for Tissue (N), 6 for Tissue (T), 6 for Std‐PDOs, and 7 for Eba‐PDOs). All data are presented as mean ± SEM. Statistical significance was determined using a one‐way analysis of variance and Tukey's test (^* p < 0.05; ^** p < 0.01; ^*** p < 0.001; ^**** p < 0.0001, ns: non‐significant). Data was obtained using PDOs and tissues derived from a single CRC patient (CEA_lo‐11).

Figure 5

Interpatient variation in CEACAM5 levels and response to 5‐FU in PDOs. A) H&E stained sections of CRC tissue alongside bright field images of both Std‐ and Eba‐PDOs at 14 days. Scale bar = 100 µm (Tissue) and 200 µm (Std‐ and Eba‐PDOs). B) Preoperative serum CEACAM5 levels for each patient (CEA_hi: High‐CEACAM5, CEA_lo: Low‐CEACAM5). C) The ratio of CEACAM5 mRNA expression in patient CRC tissues, Std‐PDOs, and Eba‐PDOs, compared to the CEA_hi‐01 patient sample (n = 4 for all patients). D) Linear regression analysis between CEACAM5 expression in PDOs and corresponding patient tissues (left: Std‐PDOs; right: Eba‐PDOs). E) Cell viability in Std‐ and Eba‐PDOs on day 6 post 5‐FU treatment at varying concentrations, assessed using the CellTiter‐Glo cell viability assay. Magenta indicates Eba‐PDOs and grey indicates Std‐PDOs (n = 3 for both PDO platforms). F) Individual area under curve (AUC) for Std‐ and Eba‐PDOs following 5‐FU treatment. The average for each group is shown with a dotted line (purple for High‐CEA; yellow for Low‐CEA). The results are presented as mean ± SEM. For statistical analysis, a one‐way analysis of variance and Tukey's test were performed (^*** p < 0.001, ^**** p < 0.0001, ns: non‐significant).

Figure 6

Eba‐PDO platform demonstrating the potential for label‐ and test‐free image‐based patient‐level prediction via supervised learning. A) Schematic illustrating the development of a High‐CEA/Low‐CEA prediction model using bright‐field images of PDO platforms via supervised learning. Morphological features such as area, perimeter, and circularity are extracted from the images of 177 Std‐PDOs and Eba‐PDOs from patients with CRC categorized as High‐CEA and Low‐CEA. These features were used to train Random Forest models. Predictions at the individual PDO level were then compiled using a majority voting algorithm (MVA) to establish patient‐level CEA predictions. The model's validation involved using a trained set and a different batch from the validation set of PDOs from the same patients. Evaluation was conducted using a test set of PDOs from different patients with CRC. B) The extracted morphological features are plotted on the 3D scale to show their distributions (x‐axis: perimeter (µm), y‐axis: area (mm²), z‐axis: circularity). Purple and yellow dots indicate High‐CEA and Low‐CEA, respectively. C) Confusion matrix for the Random Forest classifier displaying the prediction accuracy for High‐CEA/Low‐CEA in individual PDOs from both Std‐ and Eba‐PDO groups. D) Schematic illustration of the MVA to improve High‐CEA/Low‐CEA prediction accuracy via aggregating prediction results of individual PDOs. E) Simulation of patient‐level High‐CEA/Low‐CEA prediction accuracy using MVA, based on the number of PDOs utilized. F) Validation and evaluation of the MVA‐based prediction model using an untrained dataset from the same patients as those used in training, and an external validation set of PDOs from different patients with CRC, respectively. High: High‐CEA, Low: Low‐CEA. S: Success, F: Failure.