Prospective derivation of a living organoid biobank of colorectal cancer patients

Abstract

In Rspondin-based 3D cultures, Lgr5 stem cells from multiple organs form ever-expanding epithelial organoids that retain their tissue identity. We report the establishment of tumor organoid cultures from 20 consecutive colorectal carcinoma (CRC) patients. For most, organoids were also generated from adjacent normal tissue. Organoids closely recapitulate several properties of the original tumor. The spectrum of genetic changes within the "living biobank" agrees well with previous large-scale mutational analyses of CRC. Gene expression analysis indicates that the major CRC molecular subtypes are represented. Tumor organoids are amenable to high-throughput drug screens allowing detection of gene-drug associations. As an example, a single organoid culture was exquisitely sensitive to Wnt secretion (porcupine) inhibitors and carried a mutation in the negative Wnt feedback regulator RNF43, rather than in APC. Organoid technology may fill the gap between cancer genetics and patient trials, complement cell-line- and xenograft-based drug studies, and allow personalized therapy design. PAPERCLIP.

Copyright © 2015 Elsevier Inc. All rights reserved.

Figures

Figure 1. Derivation of organoids from primary tissue

A) Overview of the procedure. A total of 22 tumor organoids and 19 normal control organoids were derived and analyzed by exome-sequencing, RNA expression analysis and high-throughput drug screening. To determine the concordance between tumor organoids and primary tumor, DNA from the primary tumor was also isolated. B) Organoids architecture resembles primary tumor epithelium. H&E staining of primary tumor and the tumor organoids derived of these. A feature of most organoids is the presence of one or more lumens, resembling the tubular structures of the primary tumor (e.g. P8 and P19b). Tumors devoid of lumen give rise to compact organoids without lumen (P19a). Scale bar = 100 μM. See also Supplemental Data File S1

Figure 2. CRC subtypes are present in organoid cultures

A) Whole exome sequencing of the tumor and corresponding biopsy, when available, revealed the presence of hypermutated (>10mutations/Mb) and non-hypermutated subtypes within the organoids. Comparable rates of mutations were observed in the tumor organoid (O) and tumor biopsy (B). Organoids without corresponding biopsy are indicated in with red (O). B) Comparison of somatic copy number alterations found in the biopsies and corresponding organoids (Biop/Org) and TCGA CRC in hypermutated and non-hypermutated samples.

Figure 3. Genomic alterations found in CRC are represented in organoid cultures

A) Concordance of somatic mutations detected in organoid and corresponding biopsies. Bar graph represents the proportion of coding alterations that are concordant between the biopsy and the corresponding organoid culture, and those that are found only in organoid or biopsy specimen. N/A indicates cases in which exome-sequencing was not performed on the corresponding biopsy. B) Overview of the mutations found in the tumor organoids. The hash-mark in each box represents each allele and whether it was subject to deletion, mutation, frame-shift alteration, nonsense mutation or splice site mutation. Those alterations present in greater than 10% of cases are compared to the percentage of cases reported by the TCGA CRC. * indicates discordant mutations targeting the same gene between the two sites in P19 and P24. See also Tables S1I-J. C) Somatic copy number alterations in organoids amongst commonly amplified genes identified in TCGA CRC.

Figure 4. RNA expression analysis

A) Correlation heat map of normal organoids versus tumor organoids based on 2186 genes (the top 10% of genes in terms of standard deviation). The normal organoids are very highly correlated with each other, whereas the tumor samples exhibit more heterogeneity. The colors represent pairwise Pearson correlations after the expression values have been logged and mean-centered for every gene. The hierarchical clustering is based on one minus correlation distance. The affix N = normal, T = tumor. B) MA plot of logged normal versus tumor gene expression. P-values are computed with the R package limma, by comparing normal versus tumor gene expression. Cancer associated genes, e.g. APCDD1, PROX1 and PTCH1 are shown in the top half. C) CRC molecular subtypes are represented by the organoid panel. Genes by samples heat map of normalized gene expression of 22 organoid samples and 431 TCGA RNA-seq tumor tissue samples, organized by subtype. Within each subtype, samples are sorted by their mean gene expression for the signature genes associated with that specific subtype.

Figure 5. Development of a high-throughput drug screening assay utilizing organoid models

A) Autocrine/paracrine WNT signaling in P19b. A small panel of tumor organoids was incubated with increasing amounts of the Porcupine inhibitor IWP2. Growth of the RNF43 mutant P19b was inhibited, indicative of dependency on autocrine/paracrine WNT signaling. See also Figure S5 B) Scatterplot of (1-AUC) values for all technical replicates of drug screening data. Plots show the correlation between the 3 different technical replicates and each data point represents the (1-AUC) value for an individual organoid. C) Scatterplots of the correlation in (1-AUC) values for 3 compounds (GDC0941, obatoclax mesylate and trametinib) screened twice during every screening run. Values are the mean of 3 technical replicates.

Figure 6. Heatmap of IC50’s of all 85 compounds against 19 colorectal cancer organoids.

A) Organoids have been clustered based on their IC50 values across the drug panel. The drug names and their nominal target(s) are provided in the bottom panel. B) Drugs with the same nominal targets have similar activity profiles across the organoid panel. (1-AUC) values are plotted for inhibitor of PI3K (GDC0941 and BYL719), IGF1R (OSI-906 and BMS-536924), EGFR (cetuximab and gefitinib) and BRAF (PLX4720 and dabrafenib).

Figure 7. Gene-drug associations and differential drug sensitivity profiles of interest.

(A) Association of TP53 mutational status with nutlin-3a response. Viability response curves of the altered (blue) and wild-type organoids (grey) as well as scatter plots of cell line IC50 (μM) values. IC50 values are on a log scale comparing TP53 mutant and wild type (WT) cell lines. Each circle represents the IC50 of one cell line and the red bar is the geometric mean. (B) Immunohistochemical staining showing stabilization of TP53 in organoid P18 (Scale bar = 100 μM). (C) Association of KRAS status and cetuximab response. (D) Dose-response curves after 6 days treatment with MK2206, AZD8931 and gemcitabine. Error bars are the standard deviation of triplicate measurements. (E) Reproducibility of drug response profiles for 11 drugs. The Pearson correlation score of (1-AUC) values from the primary screen compared to (1-AUC) values from validation screens are used for comparison. The validation screen was performed twice (run 1 and 2) with >1 month elapsed between each screen. NA - data unavailable for this drug. (F) The correlation of 1-AUC values from the primary and validation screens for AZD8931, gemcitabine and nutlin-3a.