Coordinated Cellular Neighborhoods Orchestrate Antitumoral Immunity at the Colorectal Cancer Invasive Front

Summary Antitumoral immunity requires organized, spatially nuanced interactions between components of the immune tumor microenvironment (iTME). Understanding this coordinated behavior in effective versus ineffective tumor control will advance immunotherapies. We re-engineered co-detection by indexing (CODEX) for paraffin-embedded tissue microarrays, enabling simultaneous profiling of 140 tissue regions from 35 advanced-stage colorectal cancer (CRC) patients with 56 protein markers. We identified nine conserved, distinct cellular neighborhoods (CNs)—a collection of components characteristic of the CRC iTME. Enrichment of PD-1+CD4+ T cells only within a granulocyte CN positively correlated with survival in a high-risk patient subset. Coupling of tumor and immune CNs, fragmentation of T cell and macrophage CNs, and disruption of inter-CN communication was associated with inferior outcomes. This study provides a framework for interrogating how complex biological processes, such as antitumoral immunity, occur through concerted actions of cells and spatial domains.


Data S4. Supervised Annotation of CRC Clusters Based on Marker Expression, Tissue Localization and
Morphology, Related to Figure 3 and STAR Methods. After X-shift clustering, single cells from the 143 resulting initial clusters were overlaid on the raw data fluorescent images and on H&E stains of TMAs based on X/Y positions and visually verified based on marker expression profiles, morphology, and localization within the tissue. Similar clusters were manually merged, resulting in 28 final clusters. Here, 18 representative clusters are shown as yellow crosses based on X/Y coordinates of the cells contained in that cluster, overlaid on stitched montages of different TMA cores. For each of these 18 clusters, three examples of markers important for cluster identification (2 positive and 1 negative) and DRAQ5 nuclear stain (right panels) are shown as well as a global overview of cellular distribution of that cluster within a single TMA spot (yellow crosses on black background, left panel (1) Cleanup gating: Nucleated cells were selected for by gating on cells positive for Hoechst (cycle 1) and DRAQ5 (cycle 23), and out-of-focus events were removed by gating on the focused Z planes.

Group 1 (CLR)
Group 2 (diffuse) We motivate our use of tensor methods for describing differences in the variation across patients' joint CN-CT compositions by discussing the limitations of traditional PCA for this purpose. One possibility for describing the differences, between patient groups, in variation across patients' joint CN-CT compositions, would have been to first perform PCA (by flattening each patient's 2D matrix to a 1D vector), and subsequently describe how the identified axes were different. However, this would have eliminated the information that CNs and CTs form two distinct but coupled views of the iTME. This coupling corresponds exactly to the fact that the underlying biological programs drive multiple distinct CTs to be found together in multiple distinct CNs. For example, multiple CTs might share combinations of cytokine receptors, and cytokine gradients might promote combinations of CNs.
An example which illustrates how underlying biology could give rise to the tensor decomposition output is depicted as a schematic in the Figure  In the left region, the blue and red CNs share a recruitment factor (heart-shaped indentation), so share a common cell type (green) with a cognate localization factor (heart). In the right region, the orange and the gray cells share a localization factor (circle), so are found in multiple CNs. The green CN uses multiple recruitment factors, one shared with the yellow CN. Distinct interacting pairs of recruitment and localization factors co-occur across patients (red and blue found together, and yellow and green found together), each co-occurring collection of interacting pairs corresponding to a tissue module. These recruitment and localization factors are inferred from the tensor decomposition output, visualized as tissue modules comprised of CN modules and cell type (CT) modules, with interactions between them represented as edges (Panel 3). Note that there is a common collection of CT modules and CN modules that are present to different extents in each tissue module. The contribution of each CN module and CT module to each tissue module is represented by its shading (Panel 3). In tissue module 1 (top box), the CN module in the first row is interpreted as the recruitment factor with a circular indentation. This is because it contains yellow and green CNs, and there is a strong edge with the CT module containing the orange and grey cell types, and a weak edge with the CT module containing the blue cell type. The CN module with just the green CN (row 2) is interpreted as the recruitment factor with the square indentation. This is because that CN module does not contain any other CNs and has only one edge with one CT module containing the blue cell type. Since the red and green CNs are not found in the same patients, the CN module with the red and blue CNs and its cognate CT module with just the green cell type are faint in tissue module 1 and form tissue module 2. Note that the CN modules and the cell type modules are identified by their mutual dependence.

Data S7. Tensor Decomposition, Related to Figure 5.
Schematic illustrating the interpretation of the tensor decomposition output.
(1) Legend of components: A CN module corresponds to a cell recruitment program utilized by the CNs comprising that module, and a CT module corresponds to a cell type localization program utilized by the cell types comprising that module. Different pairs of recruitment programs and localization programs interact to different strengths.
(2) Different pairs of interacting recruitment programs and localization programs co-occur to form the tissue through balanced interactions between recruitment and localization factors. These combinations yield similar combinations of CNs and cell types within them across patients.
(3) Graphical representation of tissue modules corresponding to combinations of interacting pairs, indicated by edges, of CN modules (left column) and CT modules (right column). CN modules and CT modules are common across both tissue modules. In each tissue module, the transparency of each CN module and CT module corresponds to the weight of the maximum edge of which it is part, i.e. indicating its contribution to that tissue module.