Deconvolution of subcellular protrusion heterogeneity and the underlying actin regulator dynamics from live cell imaging

Abstract Cell protrusion is morphodynamically heterogeneous at the subcellular level. However, the mechanistic understanding of protrusion activities is usually based on the ensemble average of actin regulator dynamics at the cellular or population levels. Here, we establish a machine learning-based computational framework called HACKS (deconvolution of Heterogeneous Activity Coordination in cytosKeleton at a Subcellular level) to deconvolve the subcellular heterogeneity of lamellipodial protrusion. HACKS identifies distinct subcellular protrusion phenotypes hidden in highly heterogeneous protrusion activities and reveals their underlying actin regulator dynamics. The association between each protrusion phenotype and the actin regulator dynamics is further corroborated by predicting the protrusion phenotype based on actin regulator dynamics. Using our method, we discovered the hidden rare ‘accelerating’ protrusion phenotype in addition to ‘fluctuating’ and ‘periodic’ protrusions. Intriguingly, the accelerating protrusion was driven predominantly by VASP-mediated actin elongation rather than by Arp2/3-mediated actin nucleation. Our analyses also suggested that VASP controls protrusion velocity more directly than Arp2/3 complex, thereby playing a significant role in the accelerating protrusion phenotype. Taken together, we have demonstrated that HACKS allows us to discover the fine differential coordination of molecular dynamics underlying subcellular protrusion heterogeneity via a machine learning analysis of live cell imaging data.