Bayesian Classification of Mrna and Kinetochore Transport Dynamics

Abstract

Single-particle tracking is commonly used to quantify molecular dynamics in live cells from high-resolution fluorescence imaging data. Examples include long-range transport of mRNAs in neurons and oscillatory dynamics of sister kinetochores during cell division. Inferring the mechanism of transport from single-particle trajectories that can include hundreds to thousands of trajectories requires unbiased model-based inference to classify single-particle motion. Previously, we introduced a Bayesian inference procedure for this purpose that is based on a Hidden Markov Model (HMM) that accounts for both diffusive and directed transport models of motion. While HMMs generally model stochastic switching between discrete models based on a sequence of observations, formulating the HMM in the context of Bayesian inference is essential in biological data analysis in order to render this annotation objective and automated. Here, we apply this procedure to neuronal mRNAs labeled using the MS2 viral capsid reporter system and sister kinetochores labeled in HeLa cells expressing GFP-tagged CENP-A and GFP-tagged CSAP to label spindle poles. Neuronal mRNA trajectories exhibit a diverse range of behaviors along tracks including cytosolic diffusion and retrograde/anterograde transport. From our analysis, we obtain detailed distributions of transport rates and diffusivities along individual mRNA trajectories, as well as the lifetimes of each state of motion. We find that the rate of anterograde transport is greater than that of retrograde transport, indicating a possible role of forwards-backwards transport of the mRNA particles in distributing these molecules throughout the cell. To illustrate the broad applicability of our approach, we also use it to classify the heterogeneous oscillatory motion of sister kinetochores during HeLa cell division. While our results are consistent with sliding window averages commonly employed to analyze kinetochore dynamics, the Bayesian HMM infers local switching and pausing dynamics that are uniquely resolved by single-particle-based analysis.