The tumor microenvironment is undeniably complex. Specialized immune cells interact directly and indirectly with tumor cells in a constantly evolving relationship that influences the biology of the tumor. Spatial context is essential to uncovering the mysteries of cell behavior and interactions. Just as individuals may behave differently in cities, suburbs, and rural areas, individual cells adopt different roles depending on their surroundings, or neighborhoods.
A pioneering study published in Cell by Dr. Garry Nolan’s lab at Stanford University introduced a novel analysis framework which can be used to study tissue biology at two levels – the distinct regions of the tissue and the cell types present in these regions.
Using the CODEX® system, Schürch et al. generated high-dimensional spatial maps of colorectal cancer (CRC) tissue and identified distinct cell types and “cellular neighborhoods” to understand how interactions between cells and neighborhoods can influence CRC patient outcomes. Keep reading to learn more about their findings and this novel framework for spatial analysis.
Spatial phenotyping reveals cell type distribution in colorectal cancer
The authors selected samples from two different CRC patient groups for their study. The first group exhibited de novo formation of numerous tertiary lymphoid structures (TLSs), known as a “Crohn’s-like reaction” (CLR). The second group lacked these structures and displayed diffuse inflammatory infiltration (DII). Previous studies have established that patients in the CLR group have significantly higher survival rates than those from the DII group, suggesting that TLSs play some role in this discrepancy.
The TLS is a type of cellular neighborhood, defined by the study authors as regions of tissue with characteristic cell type enrichment. Seeking to identify additional such neighborhoods that may play a role in CRC patient outcomes, Dr. Nolan’s team turned to multiplex imaging-based spatial phenotyping using CODEX. They first validated a novel 56-marker panel on a variety of common normal and cancer tissues and then stained and imaged a series of CRC tissue microarrays (TMAs) containing the tumor invasive front of 17 CLR and 18 DII patient samples.
The difference between the two groups is readily apparent. As shown in the image below, the CLR patient sample has an organized structure, with an easily recognizable TLS located in its center, while the DII sample lacks any such structure.
Representative TMA cores for CLR and DII patients depicted as 7-color images (top). Voronoi diagrams of clustered cell types (CTs), merged to reduce complexity (bottom). Source: Schürch et al. / CC BY
Through subsequent digital analysis of CODEX imaging data, the authors were able to assign a single color to a cell type or functional state as defined by multiple co-expressed markers, producing Voronoi plots which map the location of cell phenotypes within tissue.
They then used unsupervised x-shift clustering to find distinct phenotypes and annotate them based on biological function, revealing 28 unique cell types, divided into structural cell types and cancer and immune cells.
Tissues are organized into cellular neighborhoods
The team developed an algorithm which revealed cellular neighborhoods by examining individual cells at the tumor invasive front and determining which neighboring cells they interact with directly. Using this approach, the authors identified nine distinct cellular neighborhoods representative of recurring patterns of cellular organization.
The heat map below describes the cell type enrichment of each of the nine neighborhoods. Each neighborhood is characterized by the presence of different functional and structural cell types. All the identified cellular neighborhoods – except neighborhood five – were present in both CLR and DII patients. Neighborhood five, which corresponds to the presence of tertiary lymphoid structures, was absent in the DII group.
Identification of 9 distinct cellular neighborhoods based on the 28 original CTs and their respective frequencies (enrichment score) within each neighborhood (pooled data from both patient groups). Source: Schürch et al. / CC BY
There were differences across other neighborhoods, as well. In general, CLR samples had larger, more organized cellular neighborhoods compared to the more fragmented neighborhoods present in DII samples.
Going a step further, the authors applied tensor decomposition to identify differences in cell type and cellular neighborhood coupling between the two patient groups. They used these differences to identify modules based on cell type and cellular neighborhoods. In turn, the differences in these modules were used to identify distinct tissue modules.
Left: Decomposition results for CLR patients. Tissue modules correspond to an “immune compartment” (top) and a “tumor compartment” (bottom). Right: Decomposition results for DII patients. Tissue modules correspond to an “immune and tumor compartment” (top) and a “granulocyte compartment” (bottom). Source: Schürch et al. / CC BY
In CLR patients, there were two distinct tissue modules which corresponded to the tissue and immune compartments. The DII patient group, on the other hand, displayed a combined tumor and immune compartment and a separate granulocyte compartment. This suggests that granulocyte-enriched neighborhoods may play a significant role in DII-type colorectal cancer.
Cell type frequencies influence patient outcomes
The authors examined the frequency of functional markers – which were not included during cell phenotyping – across the neighborhoods they identified. The frequency of CD4+ T cells, as well as the ratio of CD4+ to CD8+ T cells in the tumor boundary (neighborhood six), was highly predictive for survival in DII patients. The same frequencies in the bulk tumor did not indicate strong prognostic value.
They also studied the expression of immune checkpoint markers and markers of activation and proliferation in the T cell-enriched, macrophage-enriched, bulk tumor, and tumor boundary neighborhoods. In each of these neighborhoods, functional activation of CD8+ T cells, CD4+ T cells, and regulatory T cells was different. Proliferating CD4+ T cells, for example, were present in higher frequency in the bulk tumor compared to other neighborhoods. The inducible T cell costimulator, ICOS, on the other hand, was more frequent in the immune compartments than in the bulk tumor.
Cellular neighborhoods communicate and interact with one another
As we’ve seen, different cellular neighborhoods recruit similar combinations of cell types. As the authors note, this can be interpreted as a form of communication between the cellular neighborhoods. In examining the communications networks of the CLR and DII groups, the authors observed clear differences.
In CLR patients, the follicle neighborhoods (containing the TLS) interacted with T cell- and macrophage-enriched neighborhoods and the vascularized smooth muscle. However, this interaction was absent in DII samples. Instead, in DII samples, the granulocyte-enriched neighborhood appeared to interact with the tumor boundary and bulk tumor – an interaction absent in the CLR group.
These results confirm that the tertiary lymphoid structure plays a major role in antigen presentation and the recruitment of T cells and macrophages. T cells appear to actively exchange information with the tumor boundary, which may account for an effective antitumoral response resulting in improved survival.
In DII patients, on the other hand, the macrophage-enriched neighborhood four appeared to suppress immune activation in the T cell-enriched neighborhood, and the T cell neighborhood did not exchange information with the tumor boundary, which may impair the antitumoral response.
Conceptual framework for describing CRC immune tumor microenvironment spatial behavior. Source: Schürch et al. / CC BY
The framework for cellular neighborhood analysis presented in this study is useful for interpreting spatial biology in a dynamic tissue, such as the tumor microenvironment. This framework can be applied to any CODEX dataset, and as the authors note in their paper, “Applying these tools to large, well-annotated patient cohorts could yield clinical biomarkers, therapeutic strategies, and insight into how spatial tissue behaviors facilitate antitumoral immunity.”