Automated cell cluster analysis provides insight into multi-cell-type interactions between immune cells and their targets

Abstract

Understanding interactions between immune cells and their targets is an important step on the path to fully characterizing the immune system, and in doing so, learning how it combats disease. Many studies of these interactions have a narrow focus, often looking only at a binary result of whether or not a specific treatment was successful or only focusing on the interactions between two individual cells. Therefore, in an effort to more comprehensively study multicellular interactions among immune cells and their targets, we used in vitro longitudinal time-lapse imaging and developed an automated cell cluster analysis tool, or macro, to investigate the formation of cell clusters. In particular, we investigated the behavior of cancer-specific CD8⁺ and CD4⁺ T cells on how they interact around their targets: cancer cells and antigen-presenting cells. The macro that we established allowed us to examine these large-scale clustering behaviors taking place between those four cell types. Thus, we were able to distinguish directed immune cell clustering from random cell movement. Furthermore, this macro can be generalized to be applicable to systems consisting of any number of differently labeled species and can be used to track clustering behaviors and compare them to randomized simulations.

Keywords: Cancer; Cell cluster analysis; Immune cell interactions; Macro.

Fig. 1.. Example of the distribution of the four cell types as seen through the microscope.

(A – D) Examples of images of the four cell types are shown: (A) 6132A cerulean positive cancer cells (blue), (B) DiD stained CD11b⁺ cells (magenta) isolated from spleen of C3H/HeN mice together, (C) 6132A-specific, TCR-transduced CD4⁺ T cells (GFP, green) and (D) 6132A-specific, TCR-transduced CD8⁺ T cells (mCherry, red). (E) shows an example of the even distribution of the four cell types throughout the entire area of an 8 × 8 montage. This 8 × 8 image is used to ensure that the four 2 × 2 time-lapsed positions are representative of the broader area of the 8 × 8 montage. See Materials and Methods for details.

Fig. 2.. Four cell type cluster formation is not efficiently detectable by eye, even in a 2 × 2 montage.

Representative 2 × 2 montage using time-lapse Olympus VivaView microscopy. The image shows 6132A cerulean-positive cancer cells (blue) with DiD-stained CD11b⁺ cells (magenta) together with 6132A-specific, TCR-transduced CD8⁺ T cells (mCherry, red) and 6132A-specific, TCR-transduced CD4⁺ T cells (GFP, green). See Materials and Methods for details.

Fig. 3.. Computational steps involved in the cell cluster-finding macro.

In the first row, the macro begins with background-subtraction, thresholding the image, and creating a binary mask, followed by combining mask overlays of different channels and selecting pairwise overlaps with cancer cells. After this, the overlap of cancer cells with the GFP channels is marked as “hasG” and repeated for the other channels accordingly to form the “mask output.” In the next row, the various “mask outputs” from the GFP, RFP, and Cy5 channels are combined via Boolean “AND” operations, which distinguish each cancer cell by which other cells are present at overlapping spatial coordinates. This is used to form double and triple overlaps, which correspond to different cluster types. The clusters are combined with the original cancer cell image to create a final image output with labels and outlines. These images are representations for easier understanding, and do not correspond directly to actual images produced by the software.

Fig. 4.. Image-processing using the established cell cluster-finding macro.

An example with imaged data files of Step 1 from Fig. 3 is shown, along with before and after images of Step 7. The other diagrammed steps from Fig. 3 occur in the background and are not represented on the computer screen.

Fig. 5.. Examples of cell clusters detected by the macro.

(A) The C+11 clusters are automatically counted by the macro but not labeled or outlined. They have therefore been manually encircled (dashed line) and labeled. Note that any other cluster that contains cancer and CD11b⁺ is also a C+11 cluster. (B – E) Examples of clusters that are automatically counted, outlined, and labeled by the macro, which only outlines the target cell of the cluster. The immune cells detected by the macro as part of the cluster are manually encircled to highlight the entire cluster. The labels have also been enhanced for legibility. (B) A “C+4+8” cluster. (C) A “C+8+11” cluster. (D) A “C+4+11” cluster. (E) A “Four-Cell” cluster. See Materials and Methods for details.

Fig. 6.. Comparison of cell cluster formation between real and simulated immune cells in three cell type culture.

(A) The upper panel shows cell cluster analyses of the first 2 × 2 position from 0 – 6 h and the second 2 × 2 position from 6 – 12 h. The lower panel shows the simulation controls, using the real cancer cells with simulated immune cells. Statistical analysis shows significant differences between both measured cell clusters in the real and simulated experiments (Table 3). (B) Same data as A, but with watershed segmentation applied to automatically split touching cancer cells. The significant differences remain. See Materials and Methods for details.

Fig. 7.. Comparison of cell cluster formation comparison between real and simulated immune cells in four cell type culture.

Analogous to Fig. 6, but with four cell types. (A – B) There are significant differences between “Four-Cell”, “C+4+11”, and “C+11” clusters between the experimental and simulated data, and insignificant differences among the other clusters (Table 3). The “Four-Cell” clusters line was enhanced for viewability. See Materials and Methods for details.