Bram Herpers1,5, Rob Roovers2,5, Berina Eppink2,5, Marc van de Wetering4,5, Kuan Yan1,5, Lucia Salinaro1,5, Wim de Lau4,5, Hans Clevers4,5, Robert Vries3,4,5, Mark Throsby2,5 and Leo Price1,5
1OcellO B.V. Leiden, The Netherlands
2Merus Utrecht, The Netherlands
3HUB, Utrecht, The Netherlands
4The Hubrecht Institute, Utrecht, The Netherlands
5suppresSTEM consortium, European Community FP7, grant agreement number 601876
Background. The high failure rate in cancer drug research has been linked to the poor predictive capacity of in vitro models using 2D cultures of cancer cell lines. Compared to 2D monolayer cultures, 3D cultured tissues show gene expression patterns, differentiation- and functional characteristics which more closely reflect the situation in vivo. Furthermore, patient-derived organoid cultures retain the gene expression profiles and histological characteristics of the original tumor tissues which are often lost during long term selection on tissue culture plastics. Organoid cultures therefore increase the scope for predicting drug responses in patients, discriminating different drug responses and flagging toxicity. We used a high content screening platform and a panel of 40 colorectal cancer organoids to characterize the responses to signaling pathway inhibitors and a panel of 545 bispecific antibodies. These antibodies comprised a HER3 or EGFR targeting arm combined with a LGR4, LGR5, ZNRF3 or RNF43 targeting arm to target stem cells. Active antibodies were rescreened and candidate leads were selected.
Results. The broad mutation spectrum of the organoids was reflected in a broad heterogeneity of organoid phenotypes. Some CRC organoids formed well-differentiated spheroids with a single lumen that resembled the phenotype of normal wild type organoids, whereas others had multiple lumens or were poorly differentiated without a luminal cavity. A rich set of morphological features was extracted from 3D image data, including organoid size and shape, planar cell polarity, lumen formation as well as cell number and nucleus shape. Some features, such as those that described lumen formation, were more sensitive in detecting drug treatment than features associated with cell proliferation, improving the sensitivity of the assay to detect active molecules. A set of 10 features was selected to create a drug response profile. We observed that the absence of activating mutations did not always correlate with sensitivity to corresponding pathway inhibitors, underscoring the need for empirical testing of drugs to predict patient sensitivity. The bispecific antibody screen was performed in two stages: a primary screen in three different tumoroid models (18,908 wells) and a validation screen in 30 different tumoroid models (27,840 wells). These screens simultaneously measured morphological alterations associated with growth, differentiation and survival (e.g. apoptosis) and identified a panel of bispecific antibodies that potently inhibited a significant majority of colorectal cancer tumoroid models tested.
Conclusions. These results demonstrate that high content screening of CRC organoids is an effective strategy to identify novel inhibitors of CRC tumor outgrowth and enable identification of bispecific antibodies that target colorectal cancer stem cells with different mutational backgrounds.