CaMI—a computational framework to profile emergent patterns of single-cell motility

Introduction:: The ability of individual cells to navigate and interact within their local microenvironments drives emergent homeostatic and disease processes, such as immune surveillance or cancer metastasis. Quantifying changes in cell motility provides a biomarker for how age or disease states may alter these critical cellular behaviors. Current measurements of cell motility are often limited to individual parameters such as average speeds and average displacements. These approaches limit our ability to assess cellular heterogeneity and extract single-cell motility patterns that may be enriched or reduced in response to perturbations. Here, we describe CaMI (Fig.1A), a computational framework that enables us to identify emergent, single-cell motility patterns in a large dataset (n=74,253 cells spanning 409 biological conditions), extract cell type-specific motility phenotypes, and measure how environmental perturbations can induce fractional redistributions across motility behaviors in various cell populations.¹ This tool can be applied to a variety of motility datasets to assess dynamic behaviors at the single-cell level and quantify how perturbations induce shifts in motility behaviors that inform functional changes at the tissue and organ level.


Materials and Methods:: To identify emergent cell motility patterns across a wide variety of cell types and microenvironments, we curated data from several cell motility datasets acquired using brightfield or phase microscopy. The dataset includes 15 cell types spanning 409 experimental conditions (such as T-cell lymphomas moving within 3D collagen gels, epithelial cells in 2D and 3D microenvironment, etc.). Cell trajectories ranged from 2.5 to 16 hours in duration with frequencies of 2 to 10 minutes. To render each trajectory comparable, we developed a novel pseudo-Monte Carlo approach that applies directional biases to interpolate positions along each cell trajectory. Using this method, we integrated non-unform cell motility datasets to extract 74,253 short-term (2.5-hour) trajectories and 24,753 long-term (8-hour) trajectories at 5-minute intervals. After comparison and optimizations, we identified eight motility parameters that captured critical cell movement information according to the anisotropic persistent random walk model.² These parameters include the mean squared displacement over short and long-durations (MSD10, MSD100), persistence time in primary and non-primary axes (Pp, Pnp), diffusivities in primary and non-primary axes (Dp, Dnp, Dtot), and the anisotropic index (phi). Using these motility parameters, we performed unsupervised k-means clustering to identify emergent short-term and long-term motility patterns. Furthermore, with this new framework, we evaluated multiple test cases to demonstrate that we can develop new insights on cell motility patterns and changes in heterogeneity across relevant biological conditions.


Results, Conclusions, and Discussions:: Unsupervised clustering identified 17 short-term motility patterns across our dataset. Each motility pattern comprises a unique combination of displacement, persistence time, and diffusivity (Fig.1B). To interpret the broader characteristics of these trajectories, we performed hierarchical clustering on the 17 short-term trajectories to extract four behavioral classes (SG1-4). SG1 (patterns 1-7) and SG2 (patterns 8-11) exhibit low displacement and diffusivity, while SG3 (patterns 12-14) and SG4 (patterns 15-17) have high displacement and diffusivity. Simultaneously, SG1 and SG3 are more persistent in the primary axis, while SG2 and SG4 are more persistent in the non-primary axis. Using CaMI, we mapped how perturbations to cells altered emergent cell motility phenotypes. For example, treatment of T-cell lymphoma cells (Ly12) with increasing concentrations of actin depolymerizer, Cucurbitacin E, progressively increases the relative proportion of motility phenotypes 14-17 and decreases overall cellular heterogeneity (Fig.1C). These results suggest that inhibiting actin polymerization within T-cell lymphoma induces longer and more tortuous trajectories, which could have broader impacts on the immune microenvironment of this cancer. Taken together, these results describe how our computational framework can map individual perturbations to emergent changes in dynamic behaviors at the single-cell level. The application of this framework to distinct biological systems could allow for robust identification of motility patterns that exist within the constituent cell types of these systems. We are now in the process of coupling this framework with emerging single-cell omics technologies to understand how molecular changes drive distinct cellular behaviors associated with healthy and disease states.

Acknowledgements (Optional): :

References (Optional): : 1. Profiling emergent behaviors from single-cell motility. Maity, D. et al. Biorxiv, 2022. Profiling emergent behaviors from single-cell motility. Maity, D. et al. Biorxiv, 2022. https://doi.org/10.1101/2022.09.21.508955

2. Statistical analysis of cell migration in 3D using the anisotropic persistent random walk model. Wu, P. et al. Nature protocols, 2015.https://doi.org/10.1038/nprot.2015.030