SUMMARY Single-cell RNA-Sequencing (scRNA-Seq) has proved to be a transformative technology for cancer biology, enabling the unbiased transcriptomic profiling of individual tumor cells and revealing a striking amount of transcriptional heterogeneity in malignant cells. Many reports in recent years have identified a range of cancer cell states in diverse cancer types suggesting that these are stable and functional tumor units, with roles in tumor maintenance and progression. However, a major shortcoming of scRNA-Seq analysis is the loss of spatial information which follows from the dissociation of the tumor prior to sequencing. Lacking knowledge of the general location of each cell within the tissue, as well as its local neighborhood, scRNA-Seq cannot alone inform us about the complex set of relationships among cancer cell states, together with their interactions with the elements of the tumor microenvironment. Spatial transcriptomics is a disruptive new technology that for the first time is able to measure whole transcriptomes in a robust fashion throughout a tissue. While spatial transcriptomics maps the expression of all genes simultaneously – enabling systematic and unbiased transcriptome analysis – it is not itself a single-cell technology and thus also cannot alone inform us on the patterning of cancer cell states together with states of the tumor microenvironment. Sensitive and robust algorithms are thus required to harness the full power implicit in an integration of these technologies. Here we propose to develop a new computational method called SNAP (Single-cell Neighborhood Map) which uses matched scRNA-Seq and spatial transcriptomics data from the same tumor to infer the spatial location of each scRNA-Seq-identified cell by reference to the spatial transcriptomics data, and produces a neighborhood transcriptome for each scRNA-Seq cell. To analyze these novel neighborhood transcriptomes we propose an approach to cluster cells with common patterns of neighbors, thereby identifying sets of colocalizing cell states. SNAP promises to exploit the complementary aspects of single-cell and spatial transcriptomics to link co- localizing cancer cell states and states of the tumor microenvironment. The methodology presented here includes several novel algorithms, all of which will be made freely available to the community, where we expect them to be broadly applicable across cancer biology.

NARRATIVE Single-cell RNA-Seq (scRNA-Seq) is a rapidly advancing technology that provides unprecedented opportunities to study the molecular architecture of tumors at the level of individual cells. However, scRNA-Seq data does not provide spatial information thus limiting biological insights from the location of cell states in the tumor. Here we propose a method that infers relationships among cancer cell state and states of the tumor microenvironment through an integration of scRNA-Seq with matched spatial transcriptomics data.