ER remodelling screens leveraging high-content image informatics

The endoplasmic reticulum (ER) is required for the synthesis of secretory proteins. Numerous pathologies are associated with dysfunction of the ER, which can lead to cell death and inflammation via “ER stress” (Smith et al, 2017).

Recent work from our laboratory (Smith et al, 2017) has shown the fragmentation and turnover of dysfunctional ER material within lysosomes, which uses an intracellular trafficking highway called “autophagy”, is critical for cellular health in vivo. This process is hereafter referred to as ER “remodelling”.

Usefully, remodelling can be induced by application of chemical stressors of the ER to cultured mammalian cells. In principle, cultured cells could thus enable high-throughput genetic and chemical screens to identify genes and pathways involved in ameliorating or exacerbating ER stress (i.e. via turning up or down ER remodelling).

However, ER remodelling is a complex phenotype, refractory to conventional screening methods. This is due to the subtle associated changes in the distribution of the ER network throughout the cell, which are currently assessed by manual image analysis.
In this project, the student will a generate reporter cell lines by knocking in multiple spectrally resolvable fluorescent tags onto markers for different subdomains of the ER (such as sheets, reticular network, ER-mitochondrial contact sites etc.), using CRISPR technology.
The student will use automated high-content microscopy and image informatics platforms (e.g. Dao et al, 2016, Warchal et al, 2016), in conjunction with the National Phenotypic Screening Centre in Dundee (2nd supervisor), and supervised machine learning approaches (additional guidance from Guido Sanguinetti at School of Informatics in Edinburgh) in order to extract unbiased signatures for ER remodelling events.

Various positive and negative controls will be used to generate large image datasets for signature generation and refinement, including chemical ER stressors in the presence or absence of key known ER remodelling genes. Additionally, the discrimination of true events from non-specific changes in cell morphology will be addressed by using the multiparametric image signatures generated from above to build a machine learning model which classifies ER remodelling events.

Finally, the trained image signatures of ER remodelling and machine learning model will be put to the test in a proof-of-principle chemogenomic screen for amelioration of ER remodelling, utilising chemical libraries available at IGMM, in collaboration with Neil Carragher (Head of the Edinburgh Phenotypic Assay Centre and Edinburgh Cancer Discovery Unit).

This project will provide the student with key skills in molecular and cellular biology, large dataset generation and analysis, image informatics and image-based phenotypic screening approaches.