In developing vertebrate neurons, a neurite is formed by more than a hundred microtubules. While individual microtubules are dynamic, the microtubule array has been regarded as stationary. Using live-cell imaging of neurons in culture or in brain slices, combined with photoconversion techniques and pharmacological manipulations, we uncovered that the microtubule array flows retrogradely within neurites to the soma. This flow drives cycles of microtubule density, a hallmark of the fluctuating state before axon formation, thereby inhibiting neurite growth. The motor protein dynein fuels this process. Shortly after axon formation, microtubule retrograde flow slows down in the axon, reducing microtubule density cycles and enabling axon extension. Thus, keeping neurites short is an active process. Microtubule retrograde flow is a novel type of cytoskeletal dynamics, which changes the hitherto axon-centric view of neuronal polarization.

All data was obtained from live-cell imaging experiments that were performed on an Andor spinning disk Nikon Eclipse Ti microscope with Perfect Focus system (Nikon) and a Yokogawa CSU-X1 Spinning Disk Unit (CSUX1-A1N-E/FB2 5000rpm Control, FW, DMB 95L100016). Detailed methods for each experiment are available in the method section of the linked publication (upon acceptance; currently in revision).

All scripts were written using Python 3.7 or 3.8, with environments build with conda (Anaconda Inc.; 2020. Available from: https://docs.anaconda.com/) and using the following packages: SciPy 1.6.2 (for FigureFlow) or 1.5.2 (https://pypi.org/project/scipy/) (67), sci-kit image 0.17.2 or 0.18.1 (https://pypi.org/project/scikit-image/;) (68), numpy 1.20.3 (https://pypi.org/project/numpy/) (69), seaborn 0.11.1 (for FigureFlow) or 0.11.0  (https://pypi.org/project/seaborn/) (70), scikit-posthocs 0.6.7 (https://pypi.org/project/scikit-posthocs/) (71), matplotlib 3.34 or 3.4.2 (https://pypi.org/project/matplotlib/) (72), pandas 1.3.0 (for FigureFlow) or 1.1.3 (https://pandas.pydata.org/) (73), and python-pptx 0.6.19 (https://pypi.org/project/python-pptx/). For viewing data and for some manual analysis detailed in the methods section of the publication (upon acceptance; currently in revision) ImageJ (NIH, RRID: SCR_002074; https://imagej.nih.gov/ij/) was used, including several ImageJ plugins detailed in the methods section of the publication.