Unraveling mitotic protein networks by 3D multiplexed epitope drug screening

Lorenz Maier, Stefan M. Kallenberger, Katharina Jechow, Marcel Waschow, Roland Eils, Christian Conrad

TL;DR

This paper introduces 3D SPECS, an automated high-content screening platform for analyzing protein localization in 3D cell cultures during mitosis. By combining iterative antibody staining, 3D imaging, and machine learning with a novel spherical coordinate system (SpheriCell), they can map protein distributions and predict drug interactions, offering a more physiologically relevant approach than traditional 2D methods.

What Problem Does This Paper Address?

Understanding how proteins are spatially organized within cells is crucial for deciphering cellular processes. Traditional methods often rely on 2D cell cultures or tissue sections, which lack the physiological relevance of 3D environments. While multiplexed immunofluorescence and 3D microscopy offer richer data, integrating these techniques with automated analysis, especially for complex dynamic events like mitosis, has been challenging. Furthermore, cell-to-cell variability in orientation and structure hinders direct comparisons. This paper tackles these limitations by developing an automated workflow to precisely map protein localization in 3D cell cultures during mitosis and to analyze the impact of drug treatments on these spatial networks.

Methods

The core innovation is 3D SPECS (3D Spatial characterization of Protein Expression Changes by microscopic Screening). This workflow involves:

  • Iterative Antibody Staining: A technique to overcome spectral limitations by sequentially staining for multiple proteins, with fluorescent labels that can be bleached between rounds. This is adapted for 3D spheroids.
  • Automated High-Content 3D Imaging: Confocal microscopy coupled with automated detection of mitotic cells using machine learning.
  • SpheriCell Coordinate System: A novel, segmentation-free method to standardize spatial protein localization. It maps protein distributions into a spherical coordinate system based on the cell's mitotic axis, creating standardized 3D partitions (sectors and shells). This allows for direct comparison across cells, regardless of their orientation.
  • Mathematical Modeling: A non-linear ODE-based model is used to predict protein affinities to mitotic regions and mutual affinities between proteins, helping to explain observed spatial distributions and infer interactions.

Datasets

The study utilized two human epithelial cell lines: MCF10A (non-tumorigenic) and MCF10CA (tumorigenic). They screened 12 mitotic cancer drugs across these cell lines and two mitotic phases (metaphase and segregation). The resulting dataset comprises 6,272 confocal image stacks, leading to 1,217 mitotic events and 284,778 mean intensity values for 3D partitions. The raw microscopy data is available in the BioStudies database under identifier S-BSST176.

Key Insights and Ideas

  • 3D is Crucial for Mitotic Analysis: The study highlights that 3D cell culture provides a more physiologically relevant context for studying mitosis compared to 2D cultures, influencing cell behavior like differentiation and growth arrest.
  • SpheriCell Standardizes Complex 3D Data: The SpheriCell system is a powerful tool for normalizing and comparing protein localization data from 3D spheroids, overcoming the challenges of cell-to-cell variability and orientation.
  • Drug Effects are Spatially Resolved: Beyond just measuring drug efficacy (e.g., IC50), 3D SPECS reveals how drugs alter the spatial distribution of proteins within the cell, offering a deeper understanding of their mechanisms.
  • Predicting Novel Protein Interactions: The mathematical modeling component allows for the prediction of previously unknown protein-protein interactions, guiding future experimental validation.
  • Malignancy Correlates with Stress: Tumorigenic MCF10CA cells showed higher levels of γ-H2AX (a marker for DNA double-strand breaks), suggesting increased chromosomal stress compared to non-tumorigenic MCF10A cells.

Key Results

  • Differential Protein Localization: Distinct differences in protein localization patterns were observed between MCF10A and MCF10CA cells, particularly during metaphase.
  • Drug Impact on Spatial Distribution: Twelve mitotic drugs were tested, revealing that while most drugs primarily affected protein concentrations, a few (Haspin, Aurora B, PLK1) also significantly altered spatial distributions.
  • Inhibitor Sensitivity Varies: MCF10CA cells appeared more sensitive to certain mitotic spindle inhibitors (Aurora A, PLK1, paclitaxel) and CHK1 inhibition compared to MCF10A cells.
  • Model Predicts Interactions: The mathematical model successfully predicted 16 novel mutual affinities between proteins, in addition to known interactions, and these predictions were validated against literature.
  • Robustness of Cellular Architecture: Changes in protein affinities due to drug treatments were more pronounced than changes in individual protein localizations, suggesting a degree of robustness in the cellular architecture.

Limitations and Future Work

  • Mitotic Phase Selection: The study acknowledges that the automated selection of mitotic cells was not randomized across all mitotic phases, which could limit the analysis of inhibitor effects on the duration of these phases.
  • Model Assumptions: The mathematical model relies on simplifying assumptions, such as free protein concentrations being proportional to average cellular concentrations, which are justified but still assumptions.
  • Future Directions: The authors suggest extending the method to analyze drug effects on mitotic phase duration, incorporating phospho-specific antibodies for more detailed analysis, and integrating cell polarity landmarks for non-dividing cells. The SpheriCell approach could also be applied to other complex cellular events. The ability to predict novel interactions opens avenues for further experimental investigation into functional relations and protein complexes.

Summary updated

Abstract

Three-dimensional protein localization intricately determines the functional coordination of cellular processes. The complex spatial context of protein landscape has been assessed by multiplexed immunofluorescent staining or mass spectrometry, applied to 2D cell culture with limited physiological relevance or tissue sections. Here, we present 3D SPECS, an automated technology for 3D Spatial characterization of Protein Expression Changes by microscopic Screening. This workflow comprises iterative antibody staining, high-content 3D imaging, and machine learning for detection of mitoses. This is followed by mapping of spatial protein localization into a spherical, cellular coordinate system, a basis for model-based prediction of spatially resolved affinities of proteins. As a proof-of-concept, we mapped twelve epitopes in 3D-cultured spheroids and investigated the network effects of twelve mitotic cancer drugs. Our approach reveals novel insights into spindle fragility and chromatin stress, and predicts unknown interactions between proteins in specific mitotic pathways. 3D SPECS's ability to map potential drug targets by multiplexed immunofluorescence in 3D cell culture combined with our automated high-content assay will inspire future functional protein expression and drug assays.
BibTeX
@inproceedings{maier2018unraveling,
  author = {Lorenz Maier and Stefan M. Kallenberger and Katharina Jechow and Marcel Waschow and Roland Eils and Christian Conrad},
  title = {{Unraveling mitotic protein networks by 3D multiplexed epitope drug screening}},
  booktitle = {Molecular},
  year = {2018},
  doi = {10.15252/msb.20188238},
}
Figure