HIGH-THROUGHPUT SPATIAL TRANSCRIPTOMICS ON 50 UM-THICK VIBRATOME SECTIONS TO ENABLE INTEGRATION WITH TWO-PHOTON CALCIUM IMAGING

In the mammalian brain, distinct neuronal types interact to generate a broad range of behaviors, yet the relationship between the activity patterns of neural ensembles and their molecular identity remains poorly understood. Traditional approaches have been limited by low scalability and insufficient transcriptional resolution. Spatial transcriptomics can identify fine cell types by localizing hundreds of genes at subcellular resolution using methods such as CoppaFISH (Bugeon et al., Nature 2022). Nevertheless, the thin cryostat sections employed by previous methods require careful handling and represent a throughput bottleneck. CoppaFISH 3D (Prankerd et al., SfN 2023; Zhou et al., SfN 2024) adapts CoppaFISH for 50 µm-thick vibratome sections from PFA-fixed mouse brains, enabling easier tissue handling and more efficient registration to calcium recordings than thin fresh-frozen sections. Here, we optimize a more sensitive spatial transcriptomic method, Direct Hybridization-based In Situ Sequencing (Lee et al., Scientific Reports 2022), for 50 µm-thick vibratome sections from PFA-fixed mouse brains. We target a gene panel designed for discrimination of cortical cells into supertypes as defined in the Allen Brain Cell Atlas (Yao et al., Nature 2023), using a six-round combinatorial barcoding scheme. We present a cell-typing workflow employing a quasi-3D heterogeneous graph transformer that transfers labels from a single-cell reference onto spatial data while incorporating local cellular neighborhood information. This pipeline integrates with standard systems neuroscience workflows and should facilitate registration between spatial transcriptomics data and in vivo two-photon recordings, providing a practical framework for investigating how transcriptomically defined cell types contribute to circuit function and behavior.