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 redirection of T cell activity using bispecific antibodies is one of the most promising cancer immunotherapy approaches currently in development, but it is limited by cytokine storm-related toxicities, as well as the pharmacokinetics and tumor-penetrating capabilities of current bispecific antibody formats. Here, we have engineered the ATTACK (Asymmetric Tandem Trimerbody for T cell Activation and Cancer Killing), a novel T cell-recruiting bispecific antibody which combines three EGFR-binding single-domain antibodies (VHH; clone EgA1) with a single CD3-binding single-chain variable fragment (scFv; clone OKT3) in an intermediate molecular weight package. The two specificities are oriented in opposite directions in order to simultaneously engage cancer cells and T cell effectors, and thereby promote immunological synapse formation. EgA1 ATTACK was expressed as a homogenous, non-aggregating, soluble protein by mammalian cells and demonstrated an enhanced binding to EGFR, but not CD3, when compared to the previously characterized tandem bispecific antibody which has one EgA1 VHH and one OKT3 scFv per molecule. EgA1 ATTACK induced synapse formation and early signaling pathways downstream of TCR engagement at lower concentrations than the tandem VHH-scFv bispecific antibody. Furthermore, it demonstrated extremely potent, dose-dependent cytotoxicity when retargeting human T cells towards EGFR-expressing cells, with an efficacy over 15-fold higher than that of the tandem VHH-scFv bispecific antibody. These results suggest that the ATTACK is an ideal format for the development of the next-generation of T cell-redirecting bispecific antibodies.
The resources and expertise needed to use Deep Learning (DL) in bioimaging remain significant barriers for most laboratories. We present https://github.com/HenriquesLab/ZeroCostDL4Mic/wiki , a platform simplifying access to DL by exploiting the free, cloud-based computational resources of Google Colab. https://github.com/HenriquesLab/ZeroCostDL4Mic/wiki allows researchers to train, evaluate, and apply key DL networks to perform tasks including segmentation, detection, denoising, restoration, resolution enhancement and image-to-image translation. We demonstrate the application of the platform to study multiple biological processes.
In order to mount high-affinity antibody responses, B cells internalise specific antigens and process them into peptides loaded onto MHCII for presentation to TH cells. While the biochemical principles of antigen processing and MHCII loading have been well dissected, how the endosomal vesicle system is wired to enable these specific functions remains much less studied. Here, we performed a systematic microscopy-based analysis of antigen trafficking in B cells to reveal its route to the MHCII peptide-loading compartment (MIIC). Surprisingly, we detected fast targeting of internalised antigen into peripheral acidic compartments that possessed the hallmarks of MIIC and also showed degradative capacity. In these vesicles, internalised antigen converged rapidly with membrane-derived MHCII and partially overlapped with Cathepsin-S and H2-M, both required for peptide loading. These early compartments appeared heterogenous and atypical as they contained a mixture of both early and late markers, indicating specialized endosomal route. Together, our data suggests that, in addition to previously-reported perinuclear late endosomal MIICs, antigen processing and peptide loading could start already in these specialized early peripheral acidic vesicles (eMIIC) to support fast peptide-MHCII presentation.
Cells are highly dynamic and adopt variable shapes and sizes. These variations are biologically important but challenging to investigate in a spatiotemporally controlled manner. Micropatterning, confining cells on microfabricated substrates with defined geometries and molecular compositions, is a powerful tool for controlling cell shape and interactions. However, conventional binary micropatterns are static and fail to address dynamic changes in cell polarity, spreading, and migration. Here, a method for dynamic micropatterning is reported, where the non‐adhesive surface surrounding adhesive micropatterns is rapidly converted to support specific cell‐matrix interactions while allowing simultaneous imaging of the cells. The technique is based on ultraviolet photopatterning of biotinylated polyethylene glycol‐grafted poly‐L‐lysine, and it is simple, inexpensive, and compatible with a wide range of streptavidin‐conjugated ligands. Experiments using biotinylation‐based dynamic micropatterns reveal that distinct extracellular matrix ligands and bivalent integrin‐clustering antibodies support different degrees of front‐rear polarity in human glioblastoma cells, which correlates to altered directionality and persistence upon release and migration on fibronectin. Unexpectedly, however, neither an asymmetric cell shape nor centrosome orientation can fully predict the future direction of migration. Taken together, biotinylation‐based dynamic micropatterns allow easily accessible and highly customizable control over cell morphology and motility.
Efficient generation of antibodies by B cells is one of the prerequisites of protective immunity. B cell activation by cognate antigens via B cell receptors (BCRs), or pathogen-associated molecules through pattern-recognition receptors, such as Toll-like receptors (TLRs), leads to transcriptional and metabolic changes that ultimately transform B cells into antibody-producing plasma cells or memory cells. BCR signaling and a number of steps downstream of it rely on coordinated action of cellular membranes and the actin cytoskeleton, tightly controlled by concerted action of multiple regulatory proteins, some of them exclusive to B cells. Here, we dissect the role of Missing-In-Metastasis (MIM), or Metastasis suppressor 1 (MTSS1), a cancer-associated membrane and actin cytoskeleton regulating protein, in B cell-mediated immunity by taking advantage of MIM knockout mouse strain. We show undisturbed B cell development and largely normal composition of B cell compartments in the periphery. Interestingly, we found that MIM−/− B cells are defected in BCR signaling in response to surface-bound antigens but, on the other hand, show increased metabolic activity after stimulation with LPS or CpG. In vivo, MIM knockout animals exhibit impaired IgM antibody responses to immunization with T cell-independent antigen. This study provides the first comprehensive characterization of MIM in B cells, demonstrates its regulatory role for B cell-mediated immunity, as well as proposes new functions for MIM in tuning receptor signaling and cellular metabolism, processes, which may also contribute to the poorly understood functions of MIM in cancer.