Spatial Analysis of Malignant-Immune Cell Interactions in the Tumor Microenvironment Using Topological Data Analysis

Seol Ah Park, Yongsoo Kim, Davide Gurnari, Paweł Dłotko, Jooyoung HahnbioRxiv 2025 · DOI

TL;DR

This paper introduces a novel framework using Topological Data Analysis (TDA) to analyze the complex spatial interactions between malignant and immune cells within the tumor microenvironment (TME). By combining global shape analysis of tumor cell clusters with local density measurements, the method provides a more robust understanding of these interactions, leading to better predictions of patient survival.

What Problem Does This Paper Address?

Understanding how malignant cells and immune cells interact spatially within the tumor microenvironment (TME) is crucial for predicting cancer progression and treatment response. However, existing methods often struggle with the highly heterogeneous and complex distributions of these cells, relying heavily on local features that can be sensitive to noise and variations in imaging. This paper aims to overcome these limitations by developing a framework that integrates both global and local spatial information for a more comprehensive and robust analysis of these critical interactions.

Methods

The core of this work is a framework built upon Topological Data Analysis (TDA), specifically using the Ball Mapper algorithm.

  • Topological Malignant Clusters (TopMC): For global analysis, the authors introduce TopMC. This method uses persistence diagrams from TDA to capture the overall spatial shape of malignant cell distributions. It quantifies tumor-immune cell infiltration by measuring distances from the boundaries of these TopMCs, making it robust to segmentation errors and cell density variations.
  • Distance-Density (DD) Space: Local interactions are assessed by evaluating the density of malignant cells around immune cells. This local density, combined with the global distance from TopMC boundaries, forms a "distance-density" (DD) space for each immune cell.
  • Feature Matrix: Principal Component Analysis (PCA) is applied to the DD space to summarize key features, creating an 18-dimensional feature vector for each image.
  • Ball Mapper for High-Dimensional Data: The 18D feature vectors are then fed into the Ball Mapper algorithm. This creates a graph where nodes represent groups of images with similar spatial interaction patterns, and edges connect nodes with shared features. This allows for visualization and analysis of complex, high-dimensional data without explicit dimension reduction.
  • Two-Stage Hierarchical Clustering (HC) and Survival Analysis: To link these spatial patterns to patient outcomes, a two-stage HC approach is used. First, nodes in the Ball Mapper graph are clustered. Then, patients are clustered based on the distribution of their images across these node clusters. Finally, Kaplan-Meier survival analysis is performed on these patient clusters to identify survival-linked spatial patterns.

Datasets

The study utilizes multiplex immunofluorescence (mIF) imaging data from Diffuse Large B Cell Lymphoma (DLBCL) patients. The images are categorized into two types:

  • TR images: Malignant-enriched regions of the tumors.
  • BR images: Border regions of the tumors, showing interactions with adjacent tissue.

The data is sourced from previous studies ([Roemer et al., 2023] and [Kim et al., 2023]). A total of 972 mIF images were analyzed out of 1337, including 687 TR images and 285 BR images.

Key Insights and Ideas

  • Global Shape Matters: The study highlights that the global spatial arrangement of malignant cells (captured by TopMC) is as important as local cell densities for understanding tumor-immune interactions and predicting prognosis.
  • Robustness to Heterogeneity: TDA-based methods, particularly TopMC, prove to be robust to variations in cell distribution and segmentation quality, making them suitable for complex biological data.
  • Distance-Density Space as a Rich Feature Space: Combining distance from tumor boundaries with local malignant cell density provides a nuanced view of immune cell behavior and its relationship with tumor cells.
  • Ball Mapper Uncovers Complex Relationships: The Ball Mapper algorithm effectively visualizes and summarizes high-dimensional spatial features, revealing continuous transitions in immune-malignant cell interactions.
  • Tumor Border as a Prognostic Indicator: A higher proportion of border (BR) images in a patient's data correlates with improved survival, suggesting that well-defined tumor borders are a positive prognostic sign.
  • Optimal Immune Cell Positioning: The positioning of specific immune cell types (like active T cells) within areas of strong infiltration appears critical for good survival outcomes, a pattern not seen in poor survival cases.

Key Results

  • The proposed TDA-based framework successfully integrates global and local spatial features to analyze malignant-immune cell interactions.
  • The study demonstrates that the TopMC method is robust to variations in cell distribution, enabling consistent analysis across different image types (TR and BR).
  • Survival analysis revealed significant correlations between spatial patterns of immune phenotypes and patient survival probability.
  • Specifically, a higher proportion of BR images was associated with better survival.
  • In good survival groups, immune cells exhibited lower mean densities of nearby malignant cells, and certain T cell populations were found in more favorable positions within the TME.

Limitations and Future Work

  • Sensitivity of PCA: The authors acknowledge that PCA-based analysis in the DD space can be sensitive to minor variations, especially with sparse data. Future work could focus on developing methods to mitigate this impact.
  • Further Validation: While the study uses DLBCL data, applying the framework to other cancer types with diverse morphologies could further validate its generalizability.
  • Interpreting Complex Phenotypes: The method shows promise in interpreting challenging immune phenotypes like PD1+ and PDL1+ cells, which are often difficult to analyze due to low abundance. Further research could refine this aspect.
  • Clinical Translation: The ultimate goal would be to translate these findings into clinical diagnostics and therapeutic strategies.

Summary updated

Abstract

Abstract The spatial interactions between malignant and immune cells in the tumor microenvironment (TME) play a crucial role in cancer biology and treatment response. Understanding these interactions is critical for predicting prognosis and assessing immunotherapy effectiveness. Conventional methods, which focus on local spatial features, often struggle to achieve robust analysis due to the complex and heterogeneous cellular distributions. We propose a Topological Data Analysis (TDA)-based framework using both global and local spatial features between malignant and immune cells. For the global aspects, we introduce Topological Malignant Clusters (TopMC), a method utilizing persistence diagrams in TDA to capture the global spatial shapes of malignant cell distributions. It quantifies tumor-immune cell infiltration at a global scale by computing distances from the boundaries of the TopMC. Local interactions are evaluated through the density of malignant cells. Using high-resolution multiplex immunofluorescence (mIF) imaging in Diffuse Large B Cell Lymphoma, we integrate distance and density measures into a distance-density space to analyze patterns of malignant-immune cell interactions in TME. This study shows the robustness of the proposed approach to variations in cell distribution, enabling consistent analysis irrespective of whether images are acquired from malignant-enriched or border regions of the tumors. Furthermore, we elucidate the correlation between spatial patterns of immune phenotypes and patient survival probability.
BibTeX
@misc{park2025spatial,
  author = {Seol Ah Park and Yongsoo Kim and Davide Gurnari and Paweł Dłotko and Jooyoung Hahn},
  title = {{Spatial Analysis of Malignant-Immune Cell Interactions in the Tumor Microenvironment Using Topological Data Analysis}},
  year = {2025},
  doi = {10.1101/2025.03.27.645762},
  howpublished = {bioRxiv},
}
Figure