Computer vision profiling of neurite outgrowth mordphodynamics reveals spatio-temporal modularity of Rho GTPase signaling

Neurite outgrowth is essential to build the neuronal processes that produce axons and dendrites that connect the adult brain. In cultured cells, the neurite outgrowth process is highly dynamic, and consists of a series of repetitive morphogenetic sub-processes (MSPs), such as neurite initiation, elongation, branching, growth cone motility and collapse (da Silva and Dotti 2002). Neurons also actively migrate, which might in part reflect neuronal migration during brain development. Each of the different MSPs inherent to neurite outgrowth and cell migration is likely to be regulated by precise spatio-temporal signaling networks that control cytoskeletal dynamics, trafficking and adhesion events. These MSPs can occur on a range of time and length scales. For example, microtubule bundling in the neurite shaft can be maintained during hours, while growth cone filopodia dynamically explore their surrounding on time scales of seconds and length scales of single microns. This implies that a correct understanding of these processes will require analysis with an adequate spatio-temporal resolution. The Rho family of GTPases are signaling switches that regulate a wide variety of cellular processes, such as actin and adhesion dynamics, gene transcription, and neuronal differentiation (Boguski and McCormick 1993). Rho GTPases are activated by guanine nucleotide exchange factors (GEFs), and are switched off by GTPase activating proteins (GAPs). Upon activation, Rho GTPases can associate with effectors to initiate a downstream response. Current models propose that Rac1 and Cdc42 regulate neurite extension, while RhoA controls growth cone collapse and neurite retraction (da Silva and Dotti 2002). However, until now the effects of Rho GTPases on neurite outgrowth have mostly been assessed using protein mutants in steady-state experiments, most often at late differentiation stages, which do not provide any insight about the different MSPs during neurite outgrowth. However, our proteomic analysis of biochemically-purified neurites from N1E-115 neuronal-like cells (Pertz et al. 2008), has suggested the existence of an unexpectedly complex 220 proteins signaling network consisting of multiple GEFs, GAPs, Rho GTPases, effectors and additional interactors. This is inconsistent with the simplistic view that classical experiments have provided before. In order to gain insight into the complexity of this Rho GTPase signaling network, we performed a siRNA screen that targets each of these 220 proteins individually. We hypothesized that specific spatio-temporal Rho GTPase signaling networks control different MSPs occurring during neurite outgrowth, and therefore designed an integrated approach to capture the whole morphodynamic continuum of this process. Perturbations of candidates that lead to a similar phenotype might be part of a given spatio-temporal signaling network. This approach consisted of: 1) A high content microscopy platform that allowed us to produce 8000 timelapse movies of 660 siRNA perturbations; 2) A custom built, computer vision approach that allowed us to automatically segment and track neurite and soma morphodynamics in the timelapse movies (collaboration with the group of Pascal Fua, EPFL, Lausanne); 3) A sophisticated statistical analysis pipeline that allowed the extraction of morphological and morphodynamic signatures (MDSs) relevant to each siRNA perturbation (collaboration with the group of Francois Fleuret, IDIAP). Analysis of our dataset revealed that each siRNA perturbation led to a quantifiable phenotype, emphasizing the quality of our proteomic dataset. Hierarchical clustering of the MDSs revealed the existence of 24 phenoclusters that provide information about neurite length, branching, number of neurites, soma migration speed, and a panel of additional morphological and morphodynamic features that are more difficult to grasp using visual inspection. This complex phenotypic space can more easily be understood when classified according to the first 4 features. Our screen then suggests the existence of 4 major morphodynamic phenotypes that define distinct stages of the neurite outgrowth process. These consist of phenotypes with short neurites, multiple short neurites, long neurites, and long and branched neurites. Further subdivision using the other features provides more information, with cell migration features being very often affected. This implies a high overlap between the signaling machinery that regulates the neurite outgrowth and cell migration processes. The high phenotypical redundancy (24 clusters for 220 candidate genes) provides only limited information to deduce unambiguous signaling networks regulating distinct MSPs. Further knowledge acquired from other approaches we used to study Rho GTPase signaling (FRET biosensors, and other live cell imaging techniques), made us realize that some morphodynamic phenotypes can only be understood when growth cone dynamics are inspected at a much higher resolution. For this purpose, we decided to further investigate a defined subset of genes using high resolution live cell imaging and a custom built growth cone segmentation and tracking pipeline for accurate quantification (collaboration with the group of Gaudenz Danuser, Harvard Medical School, Boston). These results shed light into how distinct cytoskeletal networks enabling growth cone advance can globally impact the neurite outgrowth process. A clear understanding of spatio-temporal Rho GTPase signaling will therefore require multi-scale approaches. Our results provide the first insight into the complexity of spatio-temporal Rho GTPase signaling during neurite outgrowth. The technologies we devised and our initial results, pave the way for a systems biology understanding of these complex signaling systems.