CUBIC-Cloud: An Integrative Computational Framework Towards Community-driven Whole-Mouse-Brain Mapping

Abstract

Recent advancements in tissue clearing technologies have offered unparalleled opportunities for researchers to explore the whole mouse brain at cellular resolution. With the expansion of this experimental technique, however, a scalable and easy-to-use computational tool is in demand to effectively analyze and integrate whole-brain mapping datasets. To that end, here we present CUBIC-Cloud, a cloud-based framework to quantify, visualize and integrate whole mouse brain data. CUBIC-Cloud is a fully automated system where users can upload their whole-brain data, run analysis and publish the results. We demonstrate the generality of CUBIC-Cloud by a variety of applications. First, we investigated brain-wide distribution of PV, Sst, ChAT, Th and Iba1 expressing cells. Second, Aβ plaque deposition in AD model mouse brains were quantified. Third, we reconstructed neuronal activity profile under LPS-induced inflammation by c-Fos immunostaining. Last, we show brain-wide connectivity mapping by pseudo-typed Rabies virus. Together, CUBIC-Cloud provides an integrative platform to advance scalable and collaborative whole-brain mapping.

Mesoscale microscopy and image analysis tools for understanding the brain

“…Although there exists a very large number of laboratories focused on the vital function of glia in the maintenance of brain homeostasis, researchers interested in the detection and mapping of neuronal somata have been the main driver for establishing high-throughput imaging pipelines for brain segmentation, cell identification and counting. In addition to mapping the location of neuronal cell types ( Mano et al., 2020 ), such methods are also used for mapping brain activity ( Renier et al., 2016 ) and understanding cell-to-cell connectivity ( Vélez-Fort et al., 2014 ). Until recently, neuronal cell detection has been performed manually in whole-brain images ( Ogawa et al., 2014 ; Vélez-Fort et al., 2014 ; Watabe-Uchida et al., 2012 ), but this does not scale for routine use, when many thousands of cells can be labelled in each brain.…”

“…The second class are machine learning approaches. Many studies have used random forest classifiers, implemented using Ilastik ( Berg et al., 2019 ) which has been used in CUBIC-Cloud ( Mano et al., 2020 ) and also in ClearMap. More recently, deep learning ( Lecun et al., 2015 ), and in particular convolutional neural networks (CNNs) have been applied for high-performance cell detection ( Iqbal et al., 2019b ).…”

“…The 14TB image can be represented by point clouds, taking up less than 3 GB, however data must be prepared in a particular way to use atlases of this form (an image must exist with all cells labelled). It remains to be seen whether this approach will become as widespread as image-based atlases, but tools are being developed to take advantage of this approach ( Mano et al., 2020 ).…”

DeepSlice: rapid fully automatic registration of mouse brain imaging to a volumetric atlas

“…Although long-recognized as a critical step for standardization and integration of complex multimodal datasets 2,3 ,registration of data to coordinate systems such as the CCF remains laborious and reliant on human skill 4,5 . Existing tools typically assume programming proficiency, take weeks or even months to register typical datasets, and offer low levels of automaticity [6][7][8][9][10] . In contrast, Convolutional Neural Networks (CNNs) have shown great promise in the automated analysis of other types of imaging data, including cellular histology 11 and pose estimation 12 , but to date no equivalent tool for the registration of neuroimaging data to volumetric atlases such as the CCF has been described.…”

Tools for accurate post hoc determination of marker location within whole-brain microscopy images

“…This is problematic, as it does not allow researchers to use more recently developed atlases (Chon et al, 2019;Hoops et al, 2021;Kenney et al, 2021;Perens et al, 2020;Young et al, 2021), nor adapt software developed for one model organism, to another. Tools also exist for the detection and analysis of structures in whole-brain images such as neuronal somata (Furth et al, 2018;Goubran et al, 2019;Iqbal et al, 2019;Mano et al, 2020;Renier et al, 2016;Song et al, 2020;Tyson et al, 2020;Young et al, 2020), axons (Friedmann et al, 2020;Goubran et al, 2019) and vasculature (Kirst et al, 2020;Todorov et al, 2020). While mapping implanted devices within the brain has been performed using non-invasive magnetic resonance imaging (MRI) or computed tomography (CT) (Borg et al, 2015, Rangarajan et al, 2016, Király et al, 2020Kollo et al, 2020), there has been only one study using 3D whole brain microscopy (Liu et al, 2020).…”