Article Text

Download PDFPDF

1275 Mapping the heterogeneous colorectal tumor microenvironment with multiplexed imaging and machine learning
Free
  1. Blanca Cabrera Gil1,
  2. Laura Gui Levy2,
  3. Mario Kreutzfeldt2,
  4. Joanna Kowal3,
  5. Giusy Procopio3,
  6. Deniz Eroglu3,
  7. Doron Merkler2,
  8. Andrew Janowczyk1,
  9. Diego Dupouy3,
  10. Aitana Neves1 and
  11. Thomas McKee2
  1. 1Swiss Institute of Bioinformatics, Geneva, Switzerland
  2. 2Hôpitaux universitaires de Genève, Geneva, Switzerland
  3. 3Lunaphore Technologies SA, Tolochenaz, Switzerland

Abstract

Background Colorectal carcinoma (CRC) represents a major worldwide health burden and shows an increasing incidence particularly in younger patients. The majority of metastatic CRC (microsatellite stable, MSS) do not respond to current immunotherapies in contrast to the small subset of microsatellite unstable (MSI) patients. However, responses to immunotherapy were recently observed in subsets of primary MSS tumors indicating their potential vulnerability to such therapeutic approaches. Here, we use multiplex imaging coupled with powerful image analysis tools to highlight important phenotypic differences between MSI and MSS.

Methods 100 human CRC sections (30MSI, 70 MSS) were examined on the COMET™ platform (Lunaphore) with a multiplex sequential immunofluorescence (seqIF) panel of 12 biomarkers (CD3, CD31, CD4, CD8, FoxP3, TCF-1, TOX, EOMES, CK, CD45RO, S100, D2-40).

Images were preprocessed by background subtraction and local contrast enhancement.1 Cell segmentation was done with the Stardist algorithm.2Cell phenotypes (CT) were defined by threshold-based classifiers. The epithelium was detected based on CK expression. The density of CT was assessed per area of interest. Cell neighborhoods (CN) were defined as the composition of the 10 closest CTs within 300um. CNs were clustered into CN-classes using DB-SCAN.3

Results Cell segmentation and phenotyping algorithms resulted in precise cell detection (figure 1). We compared CT infiltration patterns in MSI vs MSS in the tumor-epithelium and epithelium-stroma interfaces. While we observed no significant differences in CT densities at the interface, CD3 cell density was significantly higher in MSI vs MSS patients in the tumor-epithelium (figure 2), with MSS more heterogeneously distributed with CD3-high and CD3-low outliers.

Differences in cell interactions CRC patients were examined using CN analysis. We observed a CN composed of CD3, CD8 cytotoxic T cells, CD3+CD8+ and tumor cells to be differentially enriched between MSI and MSS patients (figure 3).

Conclusions A more precise characterization of the immune response to cancers is essential to leverage the advantages of immune modulations in the treatment of cancers. We confirm here important differences between MSI and MSS CRCs in part systematic due to the higher antigen load in MSI CRCs but also highlight heterogeneity amongst the MSS tumors. Characterization of the phenotype of the T-lymphocyte population and its localization with respect to elements such as epithelial cells and tertiary lymphoid structures may help to define both the prognosis of tumors and the possibility of a response to immune checkpoint therapy.

References

  1. SM Pizer, et al. Computer Vision, Graphics, and Image Processing 39, 1987.

  2. U Schmidt, et al. MICCAI, 2018.

  3. Martin Ester, Hans-Peter Kriegel, Jörg Sander, and Xiaowei Xu. 1996. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD’96). AAAI Press, 226–231.

  4. Massey FJ 1951. The Kolmogorov-Smirnov Test for Goodness of Fit. Journal of the American Statistical Association. 1951;46(253):68–78.

  5. Cohen J 1988. Statistical power analysis for the behavioral sciences (2nd ed.). Hillside, NJ: Lawrence Erlbaum Associates.

Ethics Approval The study has been approved by HUG.

Abstract 1275 Figure 1

Example results from cell segmentation and phenotyping. Left: Original image patch. Blue: DAPI, green: CD4, red: CD8. Right: Patch with nuclei segmentation (all contours), CD4+ cells (cyan) and CD8+ cells (yellow)

Abstract 1275 Figure 2

CT analysis of the tumor epithelium region. Each boxplot represents the cell density distribution of a CT in MSS and MSI patients. CD3 cell density is observed to be significantly higher in MSI vs MSS patients, with MSS more heterogeneously distributed with CD3-high and CD3-low outliers

Abstract 1275 Figure 3

Cell neighborhood-class consisting of CD3, CD8, and CD3_CD8 cell types is significantly more frequent in MSI vs MSS patients. a) CN frequency in MSI vs MSS patients (KS-test [4] p-value and Cohen d [5] effect size). b) Proportion of CTs in this CN-class

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.