Motility efficiency and spatiotemporal synchronization in non-metastatic vs. metastatic breast cancer cells

Interpretable Fine-Grained Phenotypes of Subcellular Dynamics via Unsupervised Deep Learning

Uncovering fine-grained phenotypes of live cell dynamics is pivotal for a comprehensive understanding of the heterogeneity in healthy and diseased biological processes. However, this endeavor poses significant technical challenges for unsupervised machine learning, requiring the extraction of features that not only faithfully preserve this heterogeneity but also effectively discriminate between established biological states, all while remaining interpretable. To tackle these challenges, a self-training deep learning framework designed for fine-grained and interpretable phenotyping is presented. This framework incorporates an unsupervised teacher model with interpretable features to facilitate feature learning in a student deep neural network (DNN). Significantly, an autoencoder-based regularizer is designed to encourage the student DNN to maximize the heterogeneity associated with molecular perturbations. This method enables the acquisition of features with enhanced discriminatory power, while simultaneously preserving the heterogeneity associated with molecular perturbations. This study successfully delineated fine-grained phenotypes within the heterogeneous protrusion dynamics of migrating epithelial cells, revealing specific responses to pharmacological perturbations. Remarkably, this framework adeptly captured a concise set of highly interpretable features uniquely linked to these fine-grained phenotypes, each corresponding to specific temporal intervals crucial for their manifestation. This unique capability establishes it as a valuable tool for investigating diverse cellular dynamics and their heterogeneity.

Learning dynamical models of single and collective cell migration: a review

Single and collective cell migration are fundamental processes critical for physiological phenomena ranging from embryonic development and immune response to wound healing and cancer metastasis. To understand cell migration from a physical perspective, a broad variety of models for the underlying physical mechanisms that govern cell motility have been developed. A key challenge in the development of such models is how to connect them to experimental observations, which often exhibit complex stochastic behaviours. In this review, we discuss recent advances in data-driven theoretical approaches that directly connect with experimental data to infer dynamical models of stochastic cell migration. Leveraging advances in nanofabrication, image analysis, and tracking technology, experimental studies now provide unprecedented large datasets on cellular dynamics. In parallel, theoretical efforts have been directed towards integrating such datasets into physical models from the single cell to the tissue scale with the aim of conceptualising the emergent behaviour of cells. We first review how this inference problem has been addressed in both freely migrating and confined cells. Next, we discuss why these dynamics typically take the form of underdamped stochastic equations of motion, and how such equations can be inferred from data. We then review applications of data-driven inference and machine learning approaches to heterogeneity in cell behaviour, subcellular degrees of freedom, and to the collective dynamics of multicellular systems. Across these applications, we emphasise how data-driven methods can be integrated with physical active matter models of migrating cells, and help reveal how underlying molecular mechanisms control cell behaviour. Together, these data-driven approaches are a promising avenue for building physical models of cell migration directly from experimental data, and for providing conceptual links between different length-scales of description.

A deep learning-based segmentation pipeline for profiling cellular morphodynamics using multiple types of live cell microscopy

MOTIVATION

Quantitative studies of cellular morphodynamics rely on extracting leading-edge velocity time series based on accurate cell segmentation from live cell imaging. However, live cell imaging has numerous challenging issues regarding accurate edge localization. Fluorescence live cell imaging produces noisy and low-contrast images due to phototoxicity and photobleaching. While phase contrast microscopy is gentle to live cells, it suffers from the halo and shade-off artifacts that cannot be handled by conventional segmentation algorithms. Here, we present a deep learning-based pipeline, termed MARS-Net (Multiple-microscopy-type-based Accurate and Robust Segmentation Network), that utilizes transfer learning and data from multiple types of microscopy to localize cell edges with high accuracy, allowing quantitative profiling of cellular morphodynamics.

SUMMARY

To accurately segment cell edges and quantify cellular morphodynamics from live-cell imaging data, we developed a deep learning-based pipeline termed MARS-Net (multiple-microscopy-type-based accurate and robust segmentation network). MARS-Net utilizes transfer learning and data from multiple types of microscopy to localize cell edges with high accuracy. For effective training on distinct types of live-cell microscopy, MARS-Net comprises a pretrained VGG19 encoder with U-Net decoder and dropout layers. We trained MARS-Net on movies from phase-contrast, spinning-disk confocal, and total internal reflection fluorescence microscopes. MARS-Net produced more accurate edge localization than the neural network models trained with single-microscopy-type datasets. We expect that MARS-Net can accelerate the studies of cellular morphodynamics by providing accurate pixel-level segmentation of complex live-cell datasets.

Tumor-Associated Protrusion Fluctuations as a Signature of Cancer Invasiveness

The generation of invasive fluctuating protrusions is a distinctive feature of tumor dissemination. During the invasion, individual cancer cells modulate the morphodynamics of protrusions to optimize their migration efficiency. However, it remains unclear how protrusion fluctuations govern the invasion of more complex multi-cellular structures, such as tumors, and their correlation with the tumor metastatic potential. Herein, a reductionist approach based on 3D tumor cell micro-spheroids with different invasion capabilities is used as a model to decipher the role of tumor-associated fluctuating protrusions in cancer progression. To quantify fluctuations, a set of key biophysical parameters that precisely correlate with the invasive potential of tumors is defined. It is shown that different pharmacological drugs and cytokines are capable of modulating protrusion activity, significantly altering protrusion fluctuations, and tumor invasiveness. This correlation is used to define a novel quantitative invasion index encoding the key biophysical parameters of fluctuations and the relative levels of cell-cell/matrix interactions, which is capable of assessing the tumor's metastatic capability solely based on its magnitude. Overall, this study provides new insights into how protrusion fluctuations regulate tumor cell invasion, suggesting that they may be employed as a novel early indicator, or biophysical signature, of the metastatic potential of tumors.

Emerging machine learning approaches to phenotyping cellular motility and morphodynamics

Cells respond heterogeneously to molecular and environmental perturbations. Phenotypic heterogeneity, wherein multiple phenotypes coexist in the same conditions, presents challenges when interpreting the observed heterogeneity. Advances in live cell microscopy allow researchers to acquire an unprecedented amount of live cell image data at high spatiotemporal resolutions. Phenotyping cellular dynamics, however, is a nontrivial task and requires machine learning (ML) approaches to discern phenotypic heterogeneity from live cell images. In recent years, ML has proven instrumental in biomedical research, allowing scientists to implement sophisticated computation in which computers learn and effectively perform specific analyses with minimal human instruction or intervention. In this review, we discuss how ML has been recently employed in the study of cell motility and morphodynamics to identify phenotypes from computer vision analysis. We focus on new approaches to extract and learn meaningful spatiotemporal features from complex live cell images for cellular and subcellular phenotyping.

Persistent and polarized global actin flow is essential for directionality during cell migration

Cell migration is hypothesized to involve a cycle of behaviours beginning with leading edge extension. However, recent evidence suggests that the leading edge may be dispensable for migration, raising the question of what actually controls cell directionality. Here, we exploit the embryonic migration of Drosophila macrophages to bridge the different temporal scales of the behaviours controlling motility. This approach reveals that edge fluctuations during random motility are not persistent and are weakly correlated with motion. In contrast, flow of the actin network behind the leading edge is highly persistent. Quantification of actin flow structure during migration reveals a stable organization and asymmetry in the cell-wide flowfield that strongly correlates with cell directionality. This organization is regulated by a gradient of actin network compression and destruction, which is controlled by myosin contraction and cofilin-mediated disassembly. It is this stable actin-flow polarity, which integrates rapid fluctuations of the leading edge, that controls inherent cellular persistence.

QuimP: analyzing transmembrane signalling in highly deformable cells

Summary

Transmembrane signalling plays important physiological roles, with G protein-coupled cell surface receptors being particularly important therapeutic targets. Fluorescent proteins are widely used to study signalling, but analyses of image time series can be challenging, in particular when cells change shape. QuimP software semi-automatically tracks spatio-temporal patterns of fluorescence at the cell membrane at high spatial resolution. This makes it a unique tool for studying transmembrane signalling, particularly during cell migration in immune or cancer cells for example.

Availability and implementation

QuimP (http://warwick.ac.uk/quimp) is a set of Java plugins for Fiji/ImageJ (http://fiji.sc) installable through the Fiji Updater (http://warwick.ac.uk/quimp/wiki-pages/installation). It is compatible with Mac, Windows and Unix operating systems, requiring version >1.45 of ImageJ and Java 8. QuimP is released as open source (https://github.com/CellDynamics/QuimP) under an academic licence.

Supplementary information

Supplementary data are available at Bioinformatics online.

Lévy-like movement patterns of metastatic cancer cells revealed in microfabricated systems and implicated in vivo

Metastatic cancer cells differ from their non-metastatic counterparts not only in terms of molecular composition and genetics, but also by the very strategy they employ for locomotion. Here, we analyzed large-scale statistics for cells migrating on linear microtracks to show that metastatic cancer cells follow a qualitatively different movement strategy than their non-invasive counterparts. The trajectories of metastatic cells display clusters of small steps that are interspersed with long "flights". Such movements are characterized by heavy-tailed, truncated power law distributions of persistence times and are consistent with the Lévy walks that are also often employed by animal predators searching for scarce prey or food sources. In contrast, non-metastatic cancerous cells perform simple diffusive movements. These findings are supported by preliminary experiments with cancer cells migrating away from primary tumors in vivo. The use of chemical inhibitors targeting actin-binding proteins allows for "reprogramming" the Lévy walks into either diffusive or ballistic movements.

Profiling cellular morphodynamics by spatiotemporal spectrum decomposition

Cellular morphology and associated morphodynamics are widely used for qualitative and quantitative assessments of cell state. Here we implement a framework to profile cellular morphodynamics based on an adaptive decomposition of local cell boundary motion into instantaneous frequency spectra defined by the Hilbert-Huang transform (HHT). Our approach revealed that spontaneously migrating cells with approximately homogeneous molecular makeup show remarkably consistent instantaneous frequency distributions, though they have markedly heterogeneous mobility. Distinctions in cell edge motion between these cells are captured predominantly by differences in the magnitude of the frequencies. We found that acute photo-inhibition of Vav2 guanine exchange factor, an activator of the Rho family of signaling proteins coordinating cell motility, produces significant shifts in the frequency distribution, but does not affect frequency magnitude. We therefore concluded that the frequency spectrum encodes the wiring of the molecular circuitry that regulates cell boundary movements, whereas the magnitude captures the activation level of the circuitry. We also used HHT spectra as multi-scale spatiotemporal features in statistical region merging to identify subcellular regions of distinct motion behavior. In line with our conclusion that different HHT spectra relate to different signaling regimes, we found that subcellular regions with different morphodynamics indeed exhibit distinct Rac1 activities. This algorithm thus can serve as an accurate and sensitive classifier of cellular morphodynamics to pinpoint spatial and temporal boundaries between signaling regimes.

Many studies in cell biology employ global shape descriptors to probe mechanisms of cell morphogenesis. Here, we implement a framework in this paper to profile cellular morphodynamics very locally. We employ the Hilbert-Huang transform (HHT) to extract along the entire cell edge spectra of instantaneous edge motion frequency and magnitude and use them to classify overall cell behavior as well as subcellular edge sectors of distinct dynamics. We find in fibroblast-like COS7 cells that the marked heterogeneity in mobility of an unstimulated population is fully captured by differences in the magnitude spectra, while the frequency spectra are conserved between cells. Using optogenetics to acutely inhibit morphogenetic signaling pathways we find that these molecular shifts are reflected by changes in the frequency spectra but not in the magnitude spectra. After clustering cell edge sectors with distinct morphodynamics we observe in cells expressing a Rac1 activity biosensor that the sectors with different frequency spectra associate with different signaling intensity and dynamics. Together, these observations let us conclude that the frequency spectrum encodes the wiring of the molecular circuitry that regulates edge movements, whereas the magnitude captures the activation level of the circuitry.

Application of a co‑expression network for the analysis of aggressive and non‑aggressive breast cancer cell lines to predict the clinical outcome of patients

Breast cancer metastasis is a demanding problem in clinical treatment of patients with breast cancer. It is necessary to examine the mechanisms of metastasis for developing therapies. Classification of the aggressiveness of breast cancer is an important issue in biological study and for clinical decisions. Although aggressive and non‑aggressive breast cancer cells can be easily distinguished among different cell lines, it is very difficult to distinguish in clinical practice. The aim of the current study was to use the gene expression analysis from breast cancer cell lines to predict clinical outcomes of patients with breast cancer. Weighted gene co‑expression network analysis (WGCNA) is a powerful method to account for correlations between genes and extract co‑expressed modules of genes from large expression datasets. Therefore, WGCNA was applied to explore the differences in sub‑networks between aggressive and non‑aggressive breast cancer cell lines. The greatest difference topological overlap networks in both groups include potential information to understand the mechanisms of aggressiveness. The results show that the blue and red modules were significantly associated with the biological processes of aggressiveness. The sub‑network, which consisted of TMEM47, GJC1, ANXA3, TWIST1 and C19orf33 in the blue module, was associated with an aggressive phenotype. The sub‑network of LOC100653217, CXCL12, SULF1, DOK5 and DKK3 in the red module was associated with a non‑aggressive phenotype. In order to validate the hazard ratio of these genes, the prognostic index was constructed to integrate them and examined using data from the Cancer Genomic Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. Patients with breast cancer from TCGA in the high‑risk group had a significantly shorter overall survival time compared with patients in the low‑risk group (hazard ratio=1.231, 95% confidence interval=1.058‑1.433, P=0.0071, by the Wald test). A similar result was produced from the GEO database. The findings may provide a novel strategy for measuring cancer aggressiveness in patients with breast cancer.