A Deep Learning Method to Identify and Characterize Fluorescently Labelled Proteins from Live-cell Imaging

Anastasia Razdaibiedina

We have developed an unsupervised deep learning method to group proteins based on features extracted from micrographs using a convolutional neural network (CNN). Since functionally related proteins often have similar localization and morphological patterns, we can leverage this information to group proteins with similar visual profiles into similarity clusters. Unsupervised features obtained from this method reveal the hierarchical organization of a cell at different levels.

A High-Content Screen to Survey Effects of Single-Gene Perturbations on Subcellular Compartments

Alex Daiejavad

Lots of low-throughput experiments have been done to study the impacts of small numbers of mutations in the cell at a time, but can we scale-up and automate this process? The Marker Project in our lab leverages SGA technology and high content screening to visualize different subcellular compartments in millions of S. cerevisiae cells, each carrying a perturbation in one of ~4000 genes. We utilize bioinformatics, data analysis, and domain knowledge to identify the range of defects for each subcellular compartment and the genes associated with each defect. This will improve our understanding of how each gene contributes to cell morphology and fitness, the level of pleiotropy, and the characteristics of essential and nonessential genes.

Understanding How Cells Die

Dara Lo and Athanasios Litsios

Understanding how cell die is important both for keeping them alive and for promoting cell death in situations where that is important, but this topic has not been well explored. To study the pattern of subcellular events that occur as cells die, we are using 350 temperature-sensitive mutants of essential genes, each expressing one of 20 different GFP-tagged proteins. By imaging cells over a time course at nonpermissive temperature, we are uncovering the order of events that occur as cells die. We find that similar terminal phenotype trajectories tend to occur in mutants with defects in related bioprocesses. Comparison with wild-type cells dying of old age, in collaboration with the Nystrom lab (University of Gothenburg), gives us further insight into replicative aging.

Spatiotemporal dynamics of abnormal proteins

Kyle Wang

Despite cataloging millions of missense variants across human genomes, how these mutations affect phenotypes at the protein level is not well understood. Abnormal protein phenotypes such as subcellular mislocalization, changes in abundance and aggregation have been hallmarks of many diseases, rare and common, that affect human health today. To gain an understanding of this genotype to phenotype relationship, we need to examine a broad spectrum of abnormal proteins and catalogue their phenotypic landscape. To do this, we look to the budding yeast, Saccharomyces cerevisiae, for its genetic tractability, scalability, and availability of high throughput reagents. We’ve constructed a library of over 1000 essential gene mutants, where each mutant allele is fused to a fluorescent protein. We imaged every mutant with its wild type counterpart in 12 hr time course experiments, capturing dynamic protein localization and abundance data in millions of live cells. We trained a convolutional neural network to classify subcellular localization and extracted protein abundance data at the single cell level. Using this dynamic dataset, we aim to systematically explore the phenotypic landscape of abnormal proteins in the cell and decipher the rules that govern how mutations at the DNA level results in subcellular phenotypes.

Defining the Spatiotemporal Proteome of the Eukaryotic Cell Division Cycle

Athanasios Litsios

Cell cycle progression in eukaryotic cells relies on coordinated changes in the composition and subcellular localization of the proteome. By combining two distinct convolutional neural networks - one for predicting cell cycle stage and one for predicting protein localization - on images of millions of live yeast cells, we resolved proteome level dynamics in both concentration and localization during the cell cycle, with resolution of ~20 subcellular localization classes. We show that a quarter of the proteome displays cell cycle periodicity, with proteins tending to be controlled either at the level of localization or concentration, but not both. Distinct levels of protein regulation are preferentially utilized for different aspects of the cell cycle, with changes in protein concentration being mostly involved in cell cycle control, while changes in protein localization in the biophysical implementation of the cell cycle program. Multiomics integration of proteome concentration, gene expression, and translational efficiency measurements, reveals the importance of post-translational regulation in the shaping of cell cycle proteome dynamics across eukaryotes. Our pipeline illustrates how spatiotemporal phenomena can be systematically explored from static images of live cells. Collectively, we have generated a high resolution, quantitative map of proteome dynamics during the cell cycle.

Systematically Quantifying Proteostasis in Live Single Cells

Harsha Garadi Suresh

We are interested in unraveling the principles and mechanisms of action of proteostasis machineries that govern protein quality control by performing a variety of unbiased and genome-wide genetic and cell biological screens in yeast and human cells. Here we present one such case study which involved discovery of a novel functional interplay between redundant DUBs Ubp14/Ubp2 and HECT-family E3 Ubiquitin (Ub) ligases Ufd4/Hul5 in recycling and synthesis, respectively of the unconventional and unanchored K29-linked polyUb chains. In the absence of regulation of their abundance, these polyUb chains aberrantly associate with maturing ribosomes to disrupt their assembly and target unassembled orphan ribosomal proteins to Intranuclear Quality control compartment (INQ) for their degradation. These studies provide unique perspectives into understanding the basis of a group of diseases that are characterized by mutations in genes encoding ribosomal proteins that are collectively termed as "Ribosomopathies". Thus, as we focus on understanding the fundamentals of proteostasis, our work often provides vital insights into the clinically pressing problems of cell biology.

Identifying Proteins that Migrate to the Periphery of the Vacuole

Marine Hemmerle

The fungal vacuole is a central hub for cellular homeostasis and participates in various essential regulatory roles. In mammalian cells, the lysosome is the equivalent of the fungal vacuole and mutations in lysosomal proteins are associated with Lysosomal Storage Disorders and neurodegenerative diseases. Studies have shown that cytosolic proteins can move to the vacuole/vacuole periphery. However, the pathways that bring these proteins to the vacuole periphery and the functions of the proteins at this localization remain unknown. We are developing a fluorescence-microscopy-based platform for the systematic identification of proteins localized in the vacuole periphery in live yeast cells using a Split-Fluorescence-activating and Absorption Shifting Tag (split-FAST) system. We will generate a library of 6000 yeast strains using automated genetics and perform high throughput imaging under various stress conditions to study the kinetics and regulation of the vacuolar localization. By comparing the proteins identified in the split-FAST screens with what we already know, we expect to find new pathways regulating the vacuole/lysosome and downstream effectors, perturbation of which could lead to disease states in humans.

Global Mapping of Genetic Interactions in a Haploid Human Cell Line, HAP1

We are using the CRISPR-Cas9 system with the lentiviral Toronto knockout library (KOv3) to systematically introduce mutations in a ~haploid human cell line using single-guide RNAs (sgRNAs) in a pool, allowing the collection of large-scale GI data in human cells.

Mapping Genetic Interactions of Human Essential Genes

Zi Yang Wang

In yeast, we found that hypomorphic alleles of essential genes have substantially more GIs than deletions of non-essential genes. We are using several approaches to systematically create hypomorphic alleles of human essential genes and cataloguing a collection of stable cell lines. These are then used as queries to screen for genetic interactions via a genome-wide CRISPR-Cas9 knockout method.

Generating A Mitochondrial Network in Health and Disease

Sanna Masud

Barth syndrome (BTHS) is a multi-systemic, X-linked rare congenital heart disease and is caused by mutation of the TAFAZZIN (TAZ) gene. TAZ encodes a highly conserved transacylase that remodels immature forms of cardiolipin into mature cardiolipin, the hallmark phospholipid of the inner mitochondrial membrane. BTHS patients have complex clinical presentation, exhibiting a diverse spectrum of phenotypes and disease severity, making it difficult to both diagnose and manage, with no disease-specific therapeutics available. This phenotypic variance observed in BTHS patients suggests that there are important relationships between mutant TAZ and other genes involved in mitochondrial/cellular function, also known as genetic interactions or genetic modifiers, that mediate expressed phenotypes. My project leverages genome-scale pooled CRISPR loss of function screens in human cell models to interrogate the genetic dependencies of TAZ and identify suppressors as a therapeutic opportunity for BTHS, and other conditions that are caused by aberrant cardiolipin. To further understand TAZ and the pleiotropic nature of BTHS/ other mitochondrial diseases, we have also performed ~100 mitochondrial query genetic interaction screens in the HAP1 model to generate a comprehensive mitochondrial profile similarity network, Mitonet.

Exploring the Biological Role of a Deglycosylase Implicated in a Rare Disease

Sylvia Almeida

NGLY1 Deficiency is a rare multi-systemic autosomal recessive disease caused by mutations in N-glycanase 1 (NGLY1) and has no known cure. Patients with the disease present with a broad range of symptoms including developmental delays, hypotonia, intellectual and motor impairments, liver dysfunction, and seizures. NGLY1 is a deglycosylase which has been implicated in protein quality control and regulation of the proteasome. However, much is still unknown about its biological role in the cell and the mechanisms of disease pathogenesis are poorly understood.
Our goal is to elucidate the function of NGLY1 in human cells. One known role of NGLY1 is in activating the response to proteotoxic stress and NGLY1 mutant cells are hypersensitive to proteasome inhibition. Therefore, we aim to investigate the mechanisms by which cells can become resistant to proteotoxic stress to identify pathways which can suppress the proteostatic defects caused by loss of NGLY1. A second aim of this project is to survey NGLY1 function on a global scale. To achieve this, we are using genome-wide CRISPR-Cas9 screening of the nearly haploid human cell line HAP1 as an unbiased strategy to profile NGLY1 and potentially identify novel roles for this gene in the cell.

Functional Characterization of Chemical-Genetic Interactions

Urvi Bhojoo

We are developing a high-throughput, human cell-based platform for chemical genetic profiling of drugs and novel bioactive compounds using CRIPSPR technology. In collaboration with interdisciplinary research groups, we have generated a unique dataset of ~150 compounds that reveal how drugs work at the genetic level in human cells. These efforts have led to the identification of 22 previously uncharacterized small molecules with promising therapeutic potential from the RIKEN natural product library. With further experimentation and analysis, we uncovered the critical molecular pathway for a novel compound cytotoxicity and devised a sensitization strategy that can be exploited in cancer treatment. This work is not only expanding our understanding of the interactions between drugs and our genome, but also hold promise for the development of targeted therapies for various diseases.

Complex Haploinsufficiency of the Yeast Essential Genome

Thuy Nguyen

A fitness defect from hemizygosity is known as simple haploinsufficiency, whereas complex haploinsufficiency (CHI) occurs when hemizygosity at two or more loci in a diploid cell confers a fitness defect. This type of interaction is likely to explain some disease phenotypes in humans, who are obligate diploids. Using automated yeast genetics, we are screening for CHI interactions in the Saccharomyces cerevisiae essential genome.

Identifying Bypass Suppressors of Essential Genes and Synthetic Lethal Gene Pairs

Radha Subramaniam and Clarence Yeung

Genetic modifiers, such as suppressor or enhancer mutations, significantly influence observed characteristics and disease phenotypes. This influence extends to the expression of disease phenotypes in both Mendelian and complex genetic disorders. The manifestation of genetic resilience, as exemplified by individuals harboring mutations associated with severe Mendelian diseases yet remaining clinically unaffected, underscores the impact of background-specific genetic interactions. We use Saccharomyces cerevisiae, employing an automated, high- throughput yeast genetics platform to systematically analyze bypass suppressor networks across both essential genes and synthetic lethal gene pairs. This will contribute towards building a comprehensive, genome-wide network of suppressor genetic interactions to further our understanding of suppressor genetic networks on a broader scale while addressing questions related to genetic resilience, functional genetic relationships, evolutionary conservation of genetic pathways, and development of therapeutics.

Robustness of Genetic Interactions within a Species

Thiago Fernandes

An individual’s genome, termed genetic background, is unique due to the distinct collection of variants across all chromosomes. The biological consequence of this variation on our traits is still unknown. Previously, our lab has shown that gene essentiality and signaling network structure can be altered in two closely related S. cerevisiae genetic backgrounds (S288c vs. Σ1278b). To understand the spectrum of change within a single species, we have constructed seven genome-wide deletion collections in diverse yeast backgrounds. With this, we are surveying gene essentiality change within a species. Additionally, these collections have been integrated into the Synthetic Genetic Array (SGA) pipeline to study the robustness of the genetic interaction network to differences in genetic background. This project will further our understanding of phenotype and network dynamics due to natural variation.

The YETI Collection (Yeast Estradiol strains with Titratable Induction)

The ability to switch a gene from off to on and monitor dynamic changes provides a powerful approach for probing gene function and elucidating causal regulatory relationships. We have constructed the YETI (Yeast Estradiol strains with Titratable Induction) collection, in which >5,600 yeast genes are engineered for transcriptional inducibility with single-gene precision at their native loci and without plasmids. Each strain contains SGA screening markers and a unique barcode, enabling high-throughput genetics. We have characterized the YETI collection using growth phenotyping as well as SGA and BAR-seq screens.

Scoring Gene Essentiality across Natural Yeast Isolates

Guihong Tan

Different wild yeast strains are specialized for different niches, which is reflected by their different genetics. To gain comparative insights into background-dependent fitness phenotypes and associated modifier interactions, and to dissect the core essential genes of yeast requires the generation of systematic deletion mutant collections in genetically diverse strains. Traditional PCR-based is costly and can be a major bottleneck at this scale. We have developed a collection of CRISPR-Cas9 based gene deletion plasmids consisting of ~5800 individual plasmids (> 98% of the genome), each of which targets and deletes a unique gene in the yeast genome through homology-based repair. These plasmids are designed to take into account the sequence divergence among different wild strain backgrounds, such that the guide RNA can target every strain in our collection despite their genome diversity, and the repair fragments include large homology regions of ~200 bp ensuring efficient repair in any background. We can score gene essentiality across natural yeast isolates individually in an array format. In addition, deletion mutants generated with these plasmids carry unique 20 bp DNA barcodes, which allows for their identification using a pooled barcode sequencing strategy. With these resources, we are focusing on ~10 diverse genetic backgrounds to generate systematic gene deletion collections.

Sequence Analysis of our Temperature-sensitive Mutant Collection and Construction of a GFP-tagged Version of the TS Collection

Zhijian Li and Kyle Wang

We have previously constructed temperature-sensitive (TS) mutants for all ~1000 essential genes in yeast. Most of these TS alleles are hypomorphic, even at permissive temperature. Now, to understand what amino acid changes confer these mutant phenotypes, we have sequenced the TS collection and are using structure prediction and variant effect mapping tools to explore the data. In addition, to identify the consequences of each mutation on the localization and abundance of the protein in vivo, we have constructed GFP fusions of all the TS alleles.