Statistical methods for studying cell-cell interactions using spatial transcriptomics for Alzheimer's disease

The malfunction of neuron-microglia bidirectional signaling interactions in the brain is one of the most prominent but poorly understood mechanisms for Alzheimer's Disease (AD). Genes and pathways that regulate neuron-microglia interactions are barely identified. The development of spatial transcriptomics (ST) provides an unparalleled opportunity for studying neuron-microglia interactions, but the analytical tools have not been well developed. Depending on whether individual cells can be spatially mapped and profiled, ST data can be categorized into bulk and single-cell resolutions. The main challenge for using bulk ST to study cell-cell interactions is the lack of cellular resolution, and the major challenge for using single-cell ST is the low power and accuracy for identifications. In Aim 1, we propose a novel joint spatial network model to enable cell-cell interaction identification with bulk ST and associate the interactions with Alzheimer's Disease. We will integrate the cellular profiles from single-nucleus RNA sequencing (snRNAseq) with bulk ST to annotate the spots in ST, identify neuron and microglia enriched pairs of spots, jointly model multiple neuron-microglia interaction networks using these spots from mice with different age and AD status, and identify the association of AD- related phenotypes with neuron-microglia interactions (network edges). We will apply the proposed joint spatial network model to a recent bulk ST study that profiled the transcriptomics for ~500 spots per hemisphere for 20 cerebral hemispheres from wild-type and transgenic mice of Alzheimer's Disease, and validate discovered cell- cell interactions by leveraging independent single-cell ST studies and in snRNAseq studies of Alzheimer's Disease. In Aim 2, we propose a quantile-based distance-calibrated spatial network method to improve the study power and accuracy for identifying cell-cell interactions with single-cell ST data. We will model the heterogeneous association between gene expression and the entire distribution of cell-cell distance, and further aggregate transcriptome-wide signals to simultaneously identify (1) whether there exists a cell-cell interaction, (2) the physical distance for two cells to have robust interactions, and (3) genes whose expression levels are changed by interactions. Then, we will leveraged the identified distance and interaction changed genes to build spatial networks for cell-cell interactions. We will apply the model to seqFISH+ data that includes 10,000 gene profiles for 2,963 cells covering an area of approximately 0.5 mm2 in the cortex subventricular zone and olfactory bulb regions, and use independent single-cell ST data and bulk ST data for validation. The computational/statistical tools developed in this study enables the identification of cell-cell interactions with bulk ST and single-cell ST data, and is also helpful for a broader scientific community to model ST data for any tissues. The identified genes from this study are potential targets for therapeutic strategies of the Alzheimer's Disease .