Mammary epithelial intravital imaging data and MaSCOT-AI Cellpose model for analysis of in vivo cell shape dynamics

The data and deep learning segmentation model deposited here are derived from 3D multicoloured intravital microscopy of mammary epithelial cells during development. We aimed to study in vivo cell shape dynamics in real-time in an unbiased way. This robust and deep analysis revealed that hormone-responsive breast cells are unexpectedly elongated and motile at a high frequency during duct growth. The data is associated with our publication Dawson, Milevskiy et al, Cell Reports 2024, Hormone-responsive progenitors have a unique identity and exhibit high motility during mammary morphogenesis. https://doi.org/10.1016/j.celrep.2024.115073 Deposited data- Single channel intravital movie maximum projections (File:MaSCOT-AI Max projections). These are up to 5 hours long, with timepoints every 10 minutes.- Extracted 5th time points from each movie that we used for model training (File:MaSCOT-AI t5 training)- Segmentation files generated by Cellpose 2.2.2 (File: MaSCOT-AI t5 segmentation files) Analysis scripts:The Trackmate-Cellpose python script, R data processing scripts and example excel data sheet are on github at https://github.com/cadaws/MaSCOT-AI Example analysis and data export:A small set of example data and resulting trackmate-Cellpose output will be uploaded at a later date. Methods27 4D movies were acquired every 10 minutes by multiphoton microscopy of anaesthetised cell-type-specific confetti mice at different stages of development. 350 single channel, single-cell thick layers (10-30 µm sections) were isolated by 3D cropping, then flattened by max projection. The 5th time point from all movies was taken for model training in Cellpose 2.2.2, which was achieved after manual correction of segmentation for 150 images (MaSCOT-AI model). The MaSCOT-AI model was used in a high throughput Trackmate-Cellpose script in ImageJ to track mammary cell shape over time. Software versions:Cellpose 2.2.2 GUI with GPU was installed according to https://pypi.org/project/cellpose/ (March 2024).Trackmate v7.11.1 File name structureDate_mouse-model_developmental-stage_fluorescent-protein_z-span Mouse models:K5: K5-rtTA/tetoCre/ConfettiElf5: Elf5-rtTA/tetoCre/ConfettiPr: PR-Cre/Confetti Developmental stage:no label = Terminal end bud at 5 weeksduct/notpreg = duct at 6 or 9 weeks6dPreg/6dplug = 6 days pregnancy6d MPA = 6 days MPA treatmentMPAveh = 6 days MPA vehicle treatment本仓库存储的数据与深度学习分割模型，源自发育过程中乳腺上皮细胞的三维多色活体显微成像（intravital microscopy）。本研究旨在以无偏倚的方式实时探究体内细胞形态动力学。通过本次严谨深入的分析，我们发现激素响应性乳腺细胞在导管生长过程中，会以极高频率呈现出意外的伸长状态与运动活性。本数据集关联我们已发表的论文：Dawson、Milevskiy等，《Cell Reports》2024年，《激素响应性祖细胞具有独特身份并在乳腺形态发生过程中展现高运动活性》，论文链接：https://doi.org/10.1016/j.celrep.2024.115073 存储数据： 1. 单通道活体显微成像视频的最大强度投影文件（文件名：MaSCOT-AI Max projections）：此类视频最长可达5小时，时间间隔为每10分钟一个时间点。 2. 提取自每段视频的第5个时间点图像，用于模型训练（文件名：MaSCOT-AI t5 training）。 3. 由Cellpose 2.2.2生成的分割文件（文件名：MaSCOT-AI t5 segmentation files）。 分析脚本： Trackmate-Cellpose Python脚本、R语言数据处理脚本与示例Excel数据表已上传至GitHub仓库：https://github.com/cadaws/MaSCOT-AI 示例分析与数据导出： 少量示例数据及对应的Trackmate-Cellpose处理结果将于后续上传。 实验方法： 我们对处于不同发育阶段的麻醉状态下的细胞类型特异性Confetti小鼠进行多光子显微镜成像，每10分钟采集一段4D视频，共获得27段。通过三维裁剪分离出350个单通道单细胞层（切片厚度10~30 µm），再通过最大强度投影将其平整化。选取所有视频的第5个时间点图像，用于Cellpose 2.2.2的模型训练；该模型基于150张经人工校正分割结果的图像训练得到，命名为MaSCOT-AI模型。 我们将MaSCOT-AI模型集成至ImageJ中的高通量Trackmate-Cellpose脚本，用于实时追踪乳腺细胞的形态随时间的变化。 软件版本： 基于GPU加速的Cellpose 2.2.2图形用户界面（GUI）按照https://pypi.org/project/cellpose/ 于2024年3月完成安装；Trackmate版本为v7.11.1。 文件命名规则： 日期_小鼠模型_发育阶段_荧光蛋白_z轴跨度 小鼠模型： K5：K5-rtTA/tetoCre/Confetti Elf5：Elf5-rtTA/tetoCre/Confetti Pr：PR-Cre/Confetti 发育阶段说明： 无标注 = 5周龄小鼠的乳腺末端芽基（Terminal end bud） duct/notpreg = 6或9周龄小鼠的导管组织 6dPreg/6dplug = 妊娠6天样本 6d MPA = 经6天醋酸甲羟孕酮（MPA）处理的样本 MPAveh = 经6天醋酸甲羟孕酮溶剂对照处理的样本