A High Content Imaging and Machine Learning Approach to Identify the Anti-Cancer Effect of Novel Small Molecules on Brain Cancer Patient-Derived Organoids

Microscopy and Microanalysis, vol. 31

Abstract

Brain tumors are some of the most difficult-to-treat cancer types, and according to a recent Saudi cancer registry, more than 300 new brain cancer patients, accounting for 2.8% of cancer cases, were reported in 2016. Intra-tumor heterogeneity plays a major role in the aggressive behavior of brain tumors. Glioblastoma, for example, has a very poor prognosis rate, and only 5% of patients survive past five years of diagnosis [1]. The treatment process for these cancers is lengthy, costly, and with high recurrence rates. The survival of a patient depends heavily on the accurate and timely prognosis and diagnosis of cancer. With the advancements in computational methods, various machine-learning (ML) approaches are already used to develop predictive cancer prognosis tools [2]. Recently, many studies have used ML to accurately identify and detect brain tumors in MRI images [3]. In recent years, significant advances have been made in 3D cell culture technologies using spheroids and organoids. An organoid is a 3D structure grown from stem cells and consists of organ-specific cell types that self-organize through cell sorting and spatially restricted lineage commitment. There is a significant shift towards using organoids instead of monolayer cell cultures. Organoids are more advantageous as they mimic real physiology and have the added advantage of multiple types of cells and physiological complexity [4]. Small molecules (SMs) are a modality of choice to optimize drug potency, selectivity, and accuracy. New classes of drugs are constantly sought out to fuel the ongoing fight against cancer. Deep learning approaches can significantly aid in speeding up this process either through structure-based approaches to discover new compounds [5] or by post-processing data acquired by the cellular treatment of a new regimen of compounds. Numerous cellular assays are available to examine and screen these SMs in biological systems. High-content imaging (HCI) provides an excellent opportunity to generate highly reliable qualitative and quantitative data for screening these molecules [6]. High-content imaging (HCI) uses automated microscopy followed by image analysis, and it is used as a tool in cell-based assays to assess several cellular parameters after drug treatment. HCI generates multivariate data of multiplexed assays and provides detailed cellular and sub-cellular information on the effects of a lead compound. Morphological profiling of cells requires data on shape, size, texture, fluorescence intensity, biomarker intensity, localization, and other parameters. Small molecules are used as probes in phenotypic screening, and HCI is a very powerful technique, particularly for this type of screening. Image-based high-content screening has gained popularity recently due to high-performance computing capabilities to acquire, process, and analyze images [7, 8]. Modern HCI systems can generate stacks of high-resolution images that resolve a multitude of complex biological questions and can ultimately help identify novel classes of compounds in drug discovery. It bridges the gap between live cell imaging and cancer treatment [9]. The biggest challenge is when an often noisy dataset is available to analyze and extract meaningful information. Machine learning approaches using artificial intelligence are instrumental in addressing this question [10]. Developing an image-based assay for screening requires a lengthy process. Generic staining approaches, such as Cell painting, can drastically reduce this time. Cell painting is a recently developed HCI-based technique that provides quantitative data. Using this technique, five channels can be simultaneously imaged, and the data acquired can be used for morphological feature extraction. Here, we explore the anticancer potential of novel in-house made small molecule compounds on patient-derived brain cancer organoids using high-content imaging and analysis (Figure 1). These small molecules are synthetic organic scaffolds with moderate to high activity against several human-derived cancer cell lines. We use cell-based image profile matching to identify biologically relevant similarities or differences and using ML we predict the pathway regulators involved in the response of the 3D cells to the compound treatment (Figure 2). Our results show interesting differences between different drug doses and morphological and cellular profiles. It also showcases the physiological and pathological state of the cells in 3D, which will help predict the response of drugs in the real physiological environment. Schematic representation of drug treatment assay. Using a cell profiler, our data analysis pipeline extracts features from segmented organoid images. These features and the actual image data will be used to train a Convolution Neural Network to identify features that accurately classify responders and non-responders.

Authors 5

1. Rizwan Ali corresponding

   King Saud bin Abdulaziz University for Health Sciences · King Abdullah International Medical Research Center · National Guard Health Affairs

   Affiliation as printed

   King Abdullah International Medical Research Center (KAIMRC), King Saud Bin Abdulaziz University for Health Sciences, Ministry of National Guard-Health Affairs , Riyadh ,

2. Yassinali Tamboli

   King Saud bin Abdulaziz University for Health Sciences · King Abdullah International Medical Research Center · National Guard Health Affairs

   Affiliation as printed

   King Abdullah International Medical Research Center (KAIMRC), King Saud Bin Abdulaziz University for Health Sciences, Ministry of National Guard-Health Affairs , Riyadh ,

3. Saba M. Alsubaie

   King Saud bin Abdulaziz University for Health Sciences · King Abdullah International Medical Research Center · National Guard Health Affairs

   Affiliation as printed

   King Abdullah International Medical Research Center (KAIMRC), King Saud Bin Abdulaziz University for Health Sciences, Ministry of National Guard-Health Affairs , Riyadh ,

4. Sikander Hayat Aachen Department of Medicine 2

   RWTH Aachen University

   Affiliation as printed

   Department of Medicine 2, RWTH Aachen University, Medical Faculty , Aachen, North Rhine-Westphalia ,

5. Mohamed Boudjelal

   King Saud bin Abdulaziz University for Health Sciences · King Abdullah International Medical Research Center · National Guard Health Affairs

   Affiliation as printed

   King Abdullah International Medical Research Center (KAIMRC), King Saud Bin Abdulaziz University for Health Sciences, Ministry of National Guard-Health Affairs , Riyadh ,