Colorectal Cancer Organoid-Stroma Biobank Allows Subtype-Specific Assessment of Individualized Therapy Responses

Abstract

In colorectal cancers, the tumor microenvironment plays a key role in prognosis and therapy efficacy. Patient-derived tumor organoids (PDTO) show enormous potential for preclinical testing; however, cultured tumor cells lose important characteristics, including the consensus molecular subtypes (CMS). To better reflect the cellular heterogeneity, we established the colorectal cancer organoid-stroma biobank of matched PDTOs and cancer-associated fibroblasts (CAF) from 30 patients. Context-specific phenotyping showed that xenotransplantation or coculture with CAFs improves the transcriptomic fidelity and instructs subtype-specific stromal gene expression. Furthermore, functional profiling in coculture exposed CMS4-specific therapeutic resistance to gefitinib and SN-38 and prognostic expression signatures. Chemogenomic library screening identified patient- and therapy-dependent mechanisms of stromal resistance including MET as a common target. Our results demonstrate that colorectal cancer phenotypes are encrypted in the cancer epithelium in a plastic fashion that strongly depends on the context. Consequently, CAFs are essential for a faithful representation of molecular subtypes and therapy responses ex vivo.

Significance: Systematic characterization of the organoid-stroma biobank provides a resource for context dependency in colorectal cancer. We demonstrate a colorectal cancer subtype memory of PDTOs that is independent of specific driver mutations. Our data underscore the importance of functional profiling in cocultures for improved preclinical testing and identification of stromal resistance mechanisms. This article is featured in Selected Articles from This Issue, p. 2109.

Figure 1.

Clinical and genetic features of the colorectal cancer organoid–stroma biobank. A, Collected biomaterials from primary tumors (T), matched organoids (O), and fibroblasts (F). CRC, colorectal cancer; NA, data not available; TCGA, The Cancer Genome Atlas; TIS, tumor in situ; TMA, tissue microarray. B, Summary of clinical parameters in the experimental cohort (also see Supplementary Table S1) and public data (TCGA; ref. 28). C, Total detected alterations (SNPs and indels; log scale) in T and O. D, Mutational concordance. Note that private mutations are more frequent in organoids reflecting increased detection sensitivity in the absence of stroma. E, Recurrent mutations in cancer driver genes (based on OncoKB). Variant allele frequencies (VAF) are color coded. Average mutation frequencies (right) in this cohort reflect public data for colorectal cancer (28). Also see Supplementary Table S4. F, Mean copy-number changes in T, O (n = 30 each), and TCGA data (n = 319); also see Supplementary Fig. S3.

Figure 2.

Microenvironmental context is essential for the manifestation of transcriptomic subtypes. A, Schematic strategy for RNA sequencing in matched samples from primary tumors and organoids (n = 30 each), and following subcutaneous PDTO transplantation in NSG mice. An arbitrary subset of xenotransplants (n = 13) was analyzed by RNA sequencing, and human reads (h) were studied. For information on xenotransplant growth, see Supplementary Table S2. B, PCA shows transcriptomic differences among tumors and organoids and partial normalization in xenografts. C, Single-sample GSVA. Unsupervised clustering of tumors and organoids; the most differentially regulated HALLMARK signatures are shown. Note the restoration of tumor-specific signatures in xenografted organoids. EMT, epithelial-to-mesenchymal transition; OX., oxygen; Xeno, xenotransplant. D, CMS classification in clinical samples (Guinney et al.; ref. 3) and experimental models (see Supplementary Table S5). E, Sankey plots show weak overlap between tumors and organoids and increased concordance upon organoid xenotransplantation. See Supplementary Figs. S4 and S5.

Figure 3.

Organoid xenotransplantation elicits a subtype-specific stromal response. A, Differential analysis of stromal gene expression after xenotransplantation of CMS2-derived (n = 4) and CMS4-derived (n = 5) organoids in NSG mice. Volcano plot of mouse transcripts (base mean expression ≥10). Significantly changed genes (P < 0.05) are colored. B and C, GSEA of differential expression in CMS2 vs. CMS4 tumors and corresponding PDTO xenografts (mouse reads). MSigDB signatures for fibroblasts and inflammation (B) and the CMS4 signature (C) were studied. Normalized enrichment score (NES) and P values are shown. D, Multifluorescent analysis of the TME of matched TMA. Quantification of diverse cell populations. CAFs were defined by VIM expression and grouped into α-SMA–positive and α-SMA–negative cells. Significant changes between CMS2 (n = 9) and CMS4 tumors (n = 7) are labeled (*, P < 0.05; **, P < 0.01; Mann–Whitney U test). The tumor–stroma ratio was determined after image-based tissue segmentation (mean ± SD). E, Histologic analysis of TMA stained for pan-cytokeratin and Vimentin (top) and tumors obtained after xenotransplantation of matched PDTOs (bottom) stained for E-Cadherin and Vimentin. Staining was replicated in ≥2 animals each. All scale bars are 200 μm. See Supplementary Fig. S6.

Figure 4.

Coculture restores the subtype-specific colorectal cancer phenotype that is determined by the tumor cell compartment. A, Direct coculture: luciferase-transgenic PDTOs were single-cell dispersed and co-embedded with autologous or heterologous CAFs in Matrigel followed by culture for 6 days in the reduced medium. B–D, Swapping experiment shows the comparable effect of autologous or heterologous CAFs on organoid number (B), area (C; both in n = 4 wells each), and viability (D; luciferase assay, n = 12 wells each). The cells were derived from CMS2 (blue) or CMS4 (green) tumors. Significant changes compared with organoids alone were determined by two-sided t tests (unequal variance, unpaired, FDR 1%). Experiments show mean ± SD and were repeated twice independently. ***, P < 0.001; **, P < 0.01; *, P < 0.05; n.s., P > 0.05. a.u., arbitrary units; Rel., relative. E, EMT gene expression in tumor organoids after Transwell coculture. qPCR analysis in n = 3 wells each. Shown is relative expression (log₂-fold) compared with monoculture. The response depends on the tumor organoid rather than the source of fibroblasts. The experiment was repeated twice independently. F, EMT gene expression in tumor organoids (O11) after transfer of conditioned media (CM) from CAFs. Experiments in 3 wells each, and Transwell coculture was performed as control. Pooled data from two independent experiments are shown. G and H, RNA sequencing (RNA-seq) of organoids (G) and CAFs (H) after Transwell coculture. PCA plots show that tumor organoids and CAFs are influenced in a similar fashion by autologous and heterologous coculture. Note that the magnitude of PC1 (variation between individuals) exceeds PC2 (separating culture conditions). I and J, Differential expression analysis in CAFs after Transwell coculture (F11 and F17 ± autologous PDTOs). I, Volcano plot shows significantly changed genes (P < 0.05, colored). J, GSEA of HALLMARK signatures. Normalized enrichment score (NES) and P values are shown. K, qPCR analysis of gene expression in CAFs and matched normal fibroblasts (NAF). Relative expression (compared with monocultures) after transwell coculture with autologous PDTOs (left) or transfer of PDTO CM (right). Pooled data from two independent experiments are shown. L, Organoid/fibroblast coculture (O+F) affects organoid number, area, and viability (luciferase assay). Measured in n = 29 models (color indicates original CMS). Significance was analyzed by Wilcoxon matched-pairs signed rank test. ****, P < 0.0001. See also Supplementary Table S7. M, Spearman correlation of phenotypes (from L) to original tumor subtype. Organoids with >1,000 alterations were defined as mismatch repair deficient (dMMR). For cocultures, relative values compared with monoculture were studied. Significance was analyzed by Mann–Whitney U test; n.s., P > 0.05. Luc, luciferase; Rel., relative. N, CAFs induce subtype-specific transcriptomic features. Single-sample GSVA in tumors and in vitro contexts. Unsupervised clustering of differentially regulated HALLMARK signatures between CMS2- and CMS4-derived tumors (n = 9 each). Corresponding organoids (shown in the same order) lose subtype-specific signatures in full medium. Differences are recovered upon culture in reduced (red.) medium and Transwell coculture with fibroblasts. DN, down; HOMEOST., homeostasis; OX, oxidative; RESP., response. O, Pearson correlation of MSigDB signatures (GSEA) compared with tumors. Normalized enrichment scores between CMS2 (n = 9) and dMMR (n = 3), CMS3 (n = 6), or CMS4 (n = 9) models. CAF coculture selectively induces similarity in CMS4-derived PDTOs. Conditions as in N. P, Concordance of CMS and CRIS of PDTOs in different contexts and matched tumors. See Supplementary Fig. S8.

Figure 5.

CAFs’ influence on drug responses is dependent on the tumor cell compartment. A, Experimental setup for drug testing in cocultures using luciferase/GFP transduced organoids. B and C, Sensitivity to 5-FU, oxaliplatin, SN-38, and gefitinib in monoculture and coculture with autologous or heterologous CAFs. Tumor cell viability was measured by luciferase activity (mean ± SD in triplicate wells). Representative dose–response curves (B) and heat map of drug sensitivity [relative (rel.) AUC compared with monoculture; C]. Experiments were repeated twice. D, Measurement of organoid- and CAF-specific toxicity by dual luciferase assay. Coculture was performed using organoids expressing Firefly luciferase and matched CAFs expressing Renilla luciferase. Heat map shows normalized (norm.) AUC. CAFs show reduced sensitivity to all treatments compared with organoids. See Supplementary Figs. S9 and S10 and Supplementary Table S8.

Figure 6.

Coculture exposes subtype-specific therapy resistance and individualized drug vulnerabilities. A, Pharmacotyping of four clinical drugs in biobank (n = 29, color indicates original CMS). Tumor cell viability was assessed in monocultures (O) and cocultures (O+F) by luciferase measurement. Normalized AUC was calculated by dividing the maximum AUC value for each drug. ****, P < 0.0001; n.s.,P > 0.05 (Wilcoxon matched-pairs signed rank test). B and C, Spearman correlation between drug responses and growth characteristics (B; data from Fig. 4) or original tumor subtypes (C). Organoids with >1,000 somatic alterations were defined as dMMR. Significant changes are labeled: *, P < 0.05; **, P < 0.01 (Mann–Whitney U test). Gef, gefitinib; Oxa, oxaliplatin; Rel luc, relative luciferase. D and E, Pearson correlation between different drug treatments. AUC in monoculture (D) and the relative change of drug sensitivity in the presence of CAFs (E; AUC_co/AUC_mono) are shown. F, Heat map of differential drug sensitivity in the presence of CAFs (AUC_co/AUC_mono). Data are sorted according to the CAF drug influence, representing the average relative change of all four treatments. CMS of original tumors are labeled. G and H, Subtype comparison between CAF drug influence, (G) and the average (Avg.) AUC in monoculture (H). Medians are marked. Coculture induces significantly higher resistance in CMS4- compared with CMS2-derived models. Mann–Whitney U test (*, P = 0.029; n.s., P > 0.05). I–K, Pharmacologic screens show patient- and treatment-specific resistance mechanisms. A chemogenomic library containing 186 drugs was tested in O14 and O23 in coculture with F14. Stroma-induced resistance was analyzed by comparison of the library alone or in combination with a sublethal concentration of gefitinib (I/J) or SN-38 (K). Tumor cell viability was assessed by luciferase measurement in transgenic organoids. Mean data from two experimental replicates are shown. Top hits comprise MET inhibitor (METi; BAY-474) and other RTK/RAS pathway–associated proteins (red), BD protein inhibitors (blue), and apoptosis regulators (violet). L, MET inhibitor treatment to overcome gefitinib resistance. Heat map shows differential gefitinib response in resistant cocultures (AUC_co/AUC_mono). Addition of 1 μmol/L BAY-474 restores sensitivity in eight of nine tested cocultures. **, P < 0.01 (Wilcoxon matched-pairs signed rank test). M, Dual luciferase assay in cocultures (O14 and O23); 1 μmol/L BAY-474 induces vulnerability of tumor cells (Firefly, green) but not of CAFs (Renilla, red). Mean viability (+SD in triplicate wells) relative to DMSO alone. Experiments were repeated twice independently. See Supplementary Fig. S11 and Supplementary Tables S9 and S10.

Figure 7.

Drug response in the organoid–stroma biobank is linked to distinct colorectal cancer patient outcomes. A, Strategy for the identification of pharmacotranscriptomic signatures (see Methods). B, Violin plots show the transcriptome-wide distribution of Spearman correlation coefficients (rho). Sensitivity (red) and resistance (blue) genes were filtered (rho < −0.5, rho >0.35, and P < 0.05, respectively). Correlations to the absolute response in monocultures and in cocultures (average AUC of 5-FU, oxaliplatin, SN-38, and gefitinib) and the relative response (AUC_co/AUC_mono) defined as CAF drug influence. C and D, Prognostic value of signatures linked with drug sensitivity. A public cohort (GSE39582) was divided into high- and low-expression groups. C, Forest plot shows HR with 95% confidence interval (CI) for RFS in the high-expression group. Signatures were derived from AUC in monoculture, coculture, and relative change (AUC_co/AUC_mono) from each drug alone and from the average. For each signature, the number of genes and log-rank statistics (P value) are listed. NA, no correlated genes were present in GSE39582. D, Kaplan–Meier plot for high (red) and low (gray) expression of the sensitivity signature (CAF drug influence). E and F, Prognostic value of genes linked to drug resistance. Data are shown as above. F, Kaplan–Meier analysis with resistance signature derived from CAF drug influence. P values for C–F were calculated using a univariate log-rank test. See Supplementary Fig. S12 and Supplementary Table S11.