Deep Learning (DL) methods are powerful analytical tools for microscopy and can outperform conventional image processing pipelines. Despite the enthusiasm and innovations fuelled by DL technology, the need to access powerful and compatible resources to train DL networks leads to an accessibility barrier that novice users often find difficult to overcome. Here, we present ZeroCostDL4Mic, an entry-level platform simplifying DL access by leveraging the free, cloud-based computational resources of Google Colab. ZeroCostDL4Mic allows researchers with no coding expertise to train and apply key DL networks to perform tasks including segmentation (using U-Net and StarDist), object detection (using YOLOv2), denoising (using CARE and Noise2Void), super-resolution microscopy (using Deep-STORM), and image-to-image translation (using Label-free prediction - fnet, pix2pix and CycleGAN). Importantly, we provide suitable quantitative tools for each network to evaluate model performance, allowing model optimisation. We demonstrate the application of the platform to study multiple biological processes.
The actin cytoskeleton plays a fundamental role in various motile and morphogenetic processes involving membrane dynamics. We show that actin-binding proteins MIM (missing-in-metastasis) and IRSp53 directly bind PI(4,5)P2-rich membranes and deform them into tubular structures. This activity resides in the N-terminal IRSp53/MIM domain (IMD) of these proteins, which is structurally related to membrane-tubulating BAR (Bin/amphiphysin/Rvs) domains. We found that because of a difference in the geometry of the PI(4,5)P2-binding site, IMDs induce a membrane curvature opposite that of BAR domains and deform membranes by binding to the interior of the tubule. This explains why IMD proteins induce plasma membrane protrusions rather than invaginations. We also provide evidence that the membrane-deforming activity of IMDs, instead of the previously proposed F-actin–bundling or GTPase-binding activities, is critical for the induction of the filopodia/microspikes in cultured mammalian cells. Together, these data reveal that interplay between actin dynamics and a novel membrane-deformation activity promotes cell motility and morphogenesis.
Actin-depolymerizing factor (ADF)/cofilins are essential regulators of actin filament turnover. Several ADF/cofilin isoforms are found in multicellular organisms, but their biological differences have remained unclear. Herein, we show that three ADF/cofilins exist in mouse and most likely in all other mammalian species. Northern blot and in situ hybridization analyses demonstrate that cofilin-1 is expressed in most cell types of embryos and adult mice. Cofilin-2 is expressed in muscle cells and ADF is restricted to epithelia and endothelia. Although the three mouse ADF/cofilins do not show actin isoform specificity, they all depolymerize platelet actin filaments more efficiently than muscle actin. Furthermore, these ADF/cofilins are biochemically different. The epithelial-specific ADF is the most efficient in turning over actin filaments and promotes a stronger pH-dependent actin filament disassembly than the two other isoforms. The muscle-specific cofilin-2 has a weaker actin filament depolymerization activity and displays a 5–10-fold higher affinity for ATP-actin monomers than cofilin-1 and ADF. In steady-state assays, cofilin-2 also promotes filament assembly rather than disassembly. Taken together, these data suggest that the three biochemically distinct mammalian ADF/cofilin isoforms evolved to fulfill specific requirements for actin filament dynamics in different cell types.