The Allen Institute for Cell Science has spent the past decade developing open tools and resources for the quantitative analysis of image-based cellular data. The projects below highlight key applications of these tools to study dynamic stem cell states. These findings lay the foundation for the next phase of our research—explore our vision for the future, called CellScapes.
Nuclear Morphogenesis
|
The nucleus is one of the most critical parts of our cells, housing our genetic material (DNA) and regulating cell division. While the importance of the nucleus is well established, the factors that determine how it grows and behaves are still poorly understood. For example, in a colony of cells that have identical DNA, their nuclei still show differences in growth and behavior. This project seeks to understand why that might be.
This project studies how the nuclei of human induced pluripotent stem cells (hiPSCs) with identical DNA change across space and time. Using a fluorescently labeled nuclear membrane, we captured 3D movies of typical multi-cell colony growth over multiple days, and developed computational tools to analyze nuclear height, volume, and growth dynamics.
Follow these links to read the article or explore the related datasets now in our web-based tool Timelapse Feature explorer.
|
|
Key findings
- Nuclear growth is influenced by combinations of local interactions between cells, cell-intrinsic behaviors, and inherited traits.
- The cellular environment impacts different aspects of nuclear growth dynamics on distinct timescales (for example, growth rate is impacted on a short timescale of a few hours and nuclear height can be impacted over days).
- The amount of time that a cell’s nucleus grows is dependent on the starting volume of the nucleus. For example, smaller nuclei grow for a longer time period before dividing, while larger nuclei grow for a shorter time period. This provides a compensation mechanism for nuclear growth control.
- Nuclear starting volume and growth duration are both inherited across generations, while other analyzed aspects of nuclear growth are not inherited but instead determined by their context in space and time.
Data
The publicly available datasets, Tracked hiPSC FOV-nuclei timelapse datasets, contains:
- Three core movies following three colonies with different starting sizes.
- Segmented nuclei with their trajectories and other quantitative features.
- Matched pairs of 20X and 100X 3D images of nuclei used to train a Vision Transformer (ViT) deep-learning-based segmentation model.
Analysis tools
- CytoDL (GitHub) – Vision Transformer-Based Segmentation: novel segmentation and validation workflow predicting high-resolution segmentations essential for our quantitative analysis of nuclear growth from low-resolution images that preserve cell health
- Visualize & analyze segmented time-series microscopy data in Timelapse Feature Explorer (web), an open-source interactive viewer
- Analysis code (GitHub)
Epithelial-to-Mesenchymal Transition (EMT)
|
The epithelial-to-mesenchymal transition (EMT) is a dynamic and complex change in cell state that plays a crucial role in both healthy processes, such as development and wound healing, and disease contexts, including fibrosis and cancer metastasis. Despite its significance, EMT remains poorly defined due to the diverse ways in which it has been characterized. Changes in gene expression, cell polarity, basement membrane interactions, and cell migration behavior have each been used independently to describe EMT, yet how these different observations relate remains unclear. By inducing EMT in hiPSCs, this project aims to bring these diverse observables into a single holistic picture, providing a more comprehensive definition of EMT and advancing our broader understanding of cell state transitions.
|
|
Key findings
- EMT can be induced in hiPSCs grown in different cell culture geometries.
- hiPS cell-based EMT models provide an experimentally tractable system for the integration of multiple types of data across different observables.
- The timing of cell migration and changes in expression of EMT-related protein markers are influenced by the starting cell culture geometry.
Data and analysis tools
- Query & explore the dataset with BioFile Finder (web) – an open-source interactive file explorer
- Open & visualize 3D EMT time-series movies from BioFile Finder via Vol-E (web) directly in your web browser
- Image analysis code (GitHub)
- Data analysis code (GitHub)
Cell Representations & Learning
|
Imagine that you collect a movie of a cell growing and you want to measure aspects of this complex process to better understand cell biology. Some changes of the cell, like height or volume, are obvious to the casual observer. However, many important subtle changes that occur may not be so clear. To address this challenge, our team turned to complex computer models. This project applies Artificial Intelligence to automatically extract meaningful features from 3D cell structures, directing researchers towards areas of interest and helping them uncover new insights such as differences in organelles after drug treatments.
|
|
Key findings
- The model identifies key features that describe complex organelles, such as DNA replication foci, centrioles, nuclear speckles, and nucleoli.
- Selecting the right computational model for each structure and purpose is critical for accurate analysis.
- The framework detects drug-induced differences in organelle organization, demonstrating its potential for screening studies.
Data and tools
- Download dataset via Quilt
- Machine learning models & training code using CytoDL Framework (GitHub)
Variations in Intracellular Organization of hiPSCs
|
Every single cell in your body is packed with organelles—specialized structures which support essential functions like growth, energy production, and waste removal. While these functions are well understood, how organelles are spatially arranged within cells to achieve those functions remains a complex question. To address this challenge, we labeled 25 key organelles with fluorescent markers and used 3D microscopy to see their organization in hiPSCs. We applied computational tools to measure the spatial distribution of organelles at high resolution. With this information, we were able to create a 3D map of an “average” cell and explore the variations in intracellular organization depending on the cell’s shape, size, location within the colony, or stage of cell division.
|
|
Key findings
- Cells in interphase maintain a common intracellular organization despite having very different shapes and sizes.
- Cells at the edges of colonies show a more polarized localization of their major organelles but the interactions between them stay the same.
- Cells in different stages of cell division display different organization of their structures and organelles, both the location and the degree of variation.
Data
- This dataset, WTC-11 hiPSC Single-Cell Image Dataset v1, includes:
- A collection of over 200,000 3D images of live human stem cells spanning 25 organelles.
- A set of measurements to analyze cellular structure locations and their variations within a population of cells and compare these between two different populations.
- Explore the dataset with:
- Cell Feature Explorer (web) – an open-source interactive multidimensional image viewer
- BioFile Finder (web) – an open-source interactive file explorer
Analysis tools
- Generalizable analysis frameworks (GitHub) to create:
- Cell shape coordinate system – Quantifying modes of 3D cell shape variation via spherical harmonics and PCA analysis.
- Internal locations coordinate system within cells to permit integration of multiple structures' location into the same cell shapes.