Functional Precision Oncology
The concept of precision oncology (PM) is being used worldwide to identify new therapy strategies for individual patients, mainly for high-risk patients who suffer from uncurable, refractory or relapsed aggressive diseases. State-of-the art next-generation sequencing techniques are used to identify the drug that efficiently works for the respective patient by targeting the tumor-driving (epi-)genetic aberration(s) (e.g., fusion, mutation, or amplification) present in the individual tumor cells (Letai et al., 2017). However, it has become apparent that sequencing of the biopsy sample alone is not sufficient to assign a tailor-made drug for each patient and most patients do not currently benefit from therapy selected based on the molecular target identification (Kornauth et al., 2022).
To address these challenges of current genomic PM programs and to close the gap between the molecular knowledge of individual cancers and what can be directly applied in the clinic, comprehensive functional information is needed to identify the vulnerability of the respective tumor.
Our group functional precision oncology is addressing this need by performing comprehensive image-based ex vivo drug efficacy testing on patient-derived tumor cells to provide functional information at the individual tumor cell level.
Using this approach our groups is aiming 1) to better understand cancer- and drug-induced phenotypes to discover new drug efficacies and drug repositioning opportunities; 2) to investigate possible biomarkers and underlying mechanisms for the drug effects seen as well as for the resistances that emerge, and 3) to ultimately translate the results directly into the clinic.
To these aims we are carrying out the following projects:
Establishment of predictive relevant ex vivo disease models:
We receive fresh and fresh-frozen patient-derived tumor material, which is used to establish predictive disease models. Current main cancer entities cover adult and pediatric leukemia (CLL, AML) as well as selected solid tumor entities. Here, we focus on establishing ex vivo tumor models that closely reflect the situation in the human body and at the same time are suitable for high-content drug testing in multi-well formats. We investigating different culture protocols that particularly mimic the tumor microenvironment and the influence of cellular matrices, including 2D (co-)cultures as well as 3D spheroid cultures with emphasize on short-term models that allow a relevant clinical turn-around time of our drug testing results.
Short-term and long-term tumor models are further used to discover underlying mechanisms for the drug efficacies and resistance seen. In reverse translation studies, co-culture systems including stromal cells, immune cells and/or cytokine signaling are established (Oppermann et al., Blood, 2016) to particularly investigate possible factors leading to in vivo drug resistance.
Functional drug response profiling:
We are carrying out functional ex vivo drug sensitivity and resistance profiling (‘drug response profiling’, DRP) using primary human material derived from cancer patients and healthy donors. Key methodology is the use of high-content fluorescent spinning- disc confocal microscopy (HCM, bioimaging) allowing to test large drug libraries for their effect on human-derived cancer cells in a medium- to high-throughput format.
We test different compound libraries, including donated chemical probe library obtained through the Structural Genomics Consortium (SGC) as well as customized drug libraries consisting mainly oncology drugs, that are FDA/EMA approved including conventional chemotherapeutic agents according to guideline and standard-of-care treatments for the respective diagnoses (see above), as well as a variety of targeted new drugs (e.g., (multi) kinase inhibitors) and compounds currently tested in clinical late phase (Phase II/III) for the respective diagnoses. For mechanistic studies and training controls, different cell death modulators (e.g., for apoptosis, necroptosis, ferroptosis and ER-stress-induced cell death, differential/epigenetic and metabolic modifiers) are included. Drugs are tested in a minimum of 5 concentrations covering the clinical achievable plasma and steady state concentration and replicates, mainly in a 384 multi-well format.
As automated analysis tools are crucial for handling and management of big data and to reach a quick turnaround time for clinical translation of results, we use and further develop bioinformatical drug analysis pipelines allowing the quantification of mono- and combination therapy responses for individual patients and disease cohorts and mapping of drug hits with drug target genes and multi-omics data (ElHarouni et al., Pharmacol Res., 2022).
Together with the molecular profiling of the respective tumor samples, the functional information will further used for establishment of hypotheses on cancer subtype-selective drug combinations and their predictive biomarkers. This synergistic approach ultimately also supports the optimal design of new clinical trials (e.g., new basket trials). Finally, functional measurements in combination with molecular cancer profiles provide valuable information for subsequent mechanistic studies aimed at validating potentially novel/drug effective targets
High-content phenotypical image analysis using artificial intelligence:
We apply high-content 2D and 3D image analysis to phenotypical drug response profiling of the patient-derived tumor cell cultures. Here we use the multiplex/cell painting approach, combining different fluorescent dyes to extract phenotypical information of single cells and applied drugs. Using this phenotypical profiling approach, we investigate how the different drugs affect the different subpopulations of tumor and non-tumor cells and further classify induced cell death mechanisms as well as entity and patient specific phenotypes and drug response. We employ both classical image analysis approaches (MIP (2D), volumetric 3D analysis, 3D and 2D nuclei and single cell segmentation, advanced feature extraction (i.e morphology, shape, intensity, texture), supervised and unsupervised phenotype clustering) as well as end-to-end deep learning with the aim to gain more complex and detailed information of drug response and cellular behavior.
For quantification of image-based drug responses, we have previously developed a patient-by-patient deep transfer learning method using a convolutional neural network which transfers image information into image-based cell viabilities (Berker Y. et al., IEEE-Trans Med Imaging (TMI), 2022). Our group is now building up on the established pipelines and is developing new and advanced AI-based image analysis models. Here, we closely collaborate with the local and international bioinformatics, AI and image-analysis groups.
The group is co-affiliated with the group of Clinical Pharmacy, FB 14 at the Institute of Pharmacology and Clinical Pharmacy, Goethe-University, Frankfurt am Main.