10X Visium spatial RNA-seq from adult mouse brain sections paired to single-nucleus RNA-seq

We developed cell2location, a principled and versatile Bayesian model that is designed to resolve fine-grained cell types in spatial transcriptomic data and create comprehensive cellular maps of diverse tissues. To validate cell2location in real tissue, we applied the model to data from the mouse brain, which features diverse neural cell types organised in a well characterised spatial architecture across brain areas, thus presenting a canonical use case to test spatial genomics. We generated matched single nucleus (sn, this submission) and Visium spatial RNA-seq (10X Genomics) profiles of adjacent mouse brain sections that contain multiple regions from the telencephalon and diencephalon. To assess the biological and intra-organ technical variation in spatial mapping, we assayed two mouse brains and serial tissue sections from each brain (total of 3 and 2 matched sections from two animals, respectively, and an extra section for snRNA-seq), creating a rich multi-modal and replicated transcriptomic dataset. Tissue processing. Brains of wild-type adult C57BL/6 mice (postnatal day 56, 1 female and 1 male) were dissected, snap frozen, embedded in optimal cutting temperature compound (Tissue-Tek) and stored at -80oC. Brain hemispheres were cryosectioned at -20oC using a cryostat (Leica, CM3050S). To assess tissue quality, RNA was extracted from test tissue sections using the RNeasy Pico Kit (Qiagen) and yielded high RIN values (9.6 and 9.7) on an Agilent Bioanalyser, indicating high RNA quality. For matched single nuclei and Visium RNA-seq experiments, brain hemispheres were cryosectioned to adjacent thick (200 µm) and thin (10 µm) coronal sections, respectively, and processed the same day. In total, four consecutive sets of thick and thin tissue sections were collected from each brain. Five sets of tissue sections yielded both good quality single nuclei and Visium data (three adjacent sections from mouse 1 and two sections from mouse 2) while one additional section from mouse 2 yielded good single nuclei; these were considered for analysis in this study. Visium spatial transcriptomics. Thin (10 µm) mouse brain sections were cryosectioned and mounted directly onto separate capture areas on 10X Visium Spatial Gene Expression slides (beta product version). Processing was done per manufacturer's protocols. Briefly, sections were methanol-fixed, hematoxylin and eosin (H&E)-stained, and imaged on a NanoZoomer 2.0 slide scanner (Hamamatsu). Sections were then permeabilized and further processed to obtain cDNA libraries that were quality controlled using the Agilent Bioanalyser. The cDNA libraries were sequenced on the Illumina HiSeq 4000 system, aiming at 300 million raw reads per section with read lengths 28cy R1, 8cy i7 index, 0cy i5 index, 91cy read 2. 10X Visium spatial sequencing data was aligned to mouse pre-mRNA genome reference version mm10 using 10X SpaceRanger and mRNA count matrices were generated by adding intronic and exonic reads for each gene in each location. The paired histology H&E images were processed using 10X SpaceRanger to select locations covered by tissue by aligning pre-recorded spot locations with fiducial border spots in the histology image. This allows evaluating the correspondence between cell maps produced using our method and the known brain anatomy. This also allows identifying the number of nuclei in each spot using nuclear segmentation as described in Suppl. Methods and reported in Fig S8A-D. The histology image was used to manually annotate cortical layers in the primary somatosensory cortex (SSp) region using the lasso tool in the 10X Loupe browser.本研究开发了cell2location——一款兼具严谨性与通用性的贝叶斯模型（Bayesian model），旨在解析空间转录组数据中的精细细胞类型，并构建多种组织的完整细胞图谱。为在真实组织中验证cell2location的性能，我们将该模型应用于小鼠脑数据集。小鼠脑拥有多样化的神经细胞类型，且在脑区中呈现出已被充分表征的空间架构，因此是检验空间基因组学的典型应用场景。 我们采集了相邻小鼠脑切片的匹配单细胞核（single nucleus，本提交中简称sn）转录组与Visium空间RNA测序（10X Genomics）谱数据，这些切片涵盖端脑与间脑的多个区域。为评估空间映射中的生物学差异与器官内技术变异，我们对2只小鼠的脑组织及其连续组织切片进行了检测（两只小鼠分别对应3对与2对匹配切片，且额外为单细胞核RNA测序预留1片切片），由此构建了内容丰富的多模态重复转录组数据集。 ### 组织处理流程 选取2只健康成年C57BL/6野生型小鼠（出生后第56天，1雌1雄）的脑组织，解剖后迅速冷冻，以OCT包埋剂（optimal cutting temperature compound，Tissue-Tek品牌）包埋，并于-80℃保存。使用冷冻切片机（Leica CM3050S）在-20℃下对大脑半球进行冰冻切片。为评估组织RNA质量，我们从测试切片中提取RNA，采用RNeasy Pico试剂盒（Qiagen），经Agilent生物分析仪检测后获得了9.6与9.7的高RNA完整性数值（RIN），表明RNA质量优异。 针对匹配的单细胞核与Visium RNA测序实验，我们将大脑半球分别切为相邻的厚层（200μm）与薄层（10μm）冠状切片，并于同日完成处理。每只小鼠的脑组织共计采集4组连续的厚层与薄层切片对。其中5组切片可同时获得高质量单细胞核数据与Visium测序数据（小鼠1对应3对相邻切片，小鼠2对应2对），此外小鼠2还额外获得1片高质量单细胞核切片，本研究均纳入分析。 ### Visium空间转录组实验流程 将10μm厚的小鼠脑冰冻切片直接贴于10X Visium空间基因表达玻片（测试版）的独立捕获区域，实验流程严格遵循厂商官方指南。简要流程如下：切片经甲醇固定、苏木精-伊红（H&E）染色后，使用NanoZoomer 2.0玻片扫描仪（Hamamatsu）完成成像。随后对切片进行通透处理，进一步构建cDNA文库，并通过Agilent生物分析仪完成文库质量控制。将cDNA文库在Illumina HiSeq 4000测序平台上进行测序，每片切片预期产出3亿条原始读段，测序读长设置为：R1为28个循环，i7索引为8个循环，i5索引为0个循环，read 2为91个循环。 使用10X SpaceRanger工具将10X Visium空间测序数据比对至小鼠pre-mRNA基因组参考版本mm10，并通过统计每个捕获位点上每个基因的内含子与外显子读段数，生成mRNA计数矩阵。使用10X SpaceRanger处理配对的H&E组织学图像：通过将预先记录的捕获位点坐标与组织学图像中的基准标记边界点对齐，筛选出被组织覆盖的捕获位点。该步骤可用于评估本研究方法生成的细胞图谱与已知小鼠脑解剖结构的匹配度。同时，该流程还可通过核分割（详见补充方法，结果见图S8A-D）统计每个捕获位点的细胞核数量。研究人员还借助10X Loupe浏览器的套索工具，在组织学图像上手动注释了初级躯体感觉皮层（SSp）区域的皮层分层。