Physics approaches to the spatial distribution of immune cells in tumors

Clare C. Yu, Juliana C. Wortman, Ting-Fang He, Shawn T. Solomon, Bob Zhang, Anthony Rosario, Ruo‐Qian Wang, Travis Y. Tu, Daniel Schmolze, Yuan Yuan, Susan E. Yost, Xuefei Li, Herbert Levine, Gurinder S. Atwal, Peter P. LeeReports 2020 · DOI

TL;DR

This paper introduces novel physics-inspired methods, particularly a maximum entropy approach, to quantify the spatial distribution of immune cells within tumors. They found that the way immune cells (B and T cells) are spread out, not just their density, is a significant predictor of whether triple-negative breast cancer will recur. This suggests that spatial arrangement plays a crucial role in immune surveillance and patient prognosis.

What Problem Does This Paper Address?

The effectiveness of cancer immunotherapy often hinges on the immune system's ability to infiltrate and attack tumor cells. While higher densities of immune cells within tumors are generally associated with better outcomes, this paper questions whether the spatial distribution of these cells, beyond mere density, is equally, if not more, important for predicting cancer recurrence. Traditional methods often average cell densities across entire tissues, potentially missing crucial spatial heterogeneity. The authors aim to develop quantitative methods to capture this spatial information and link it to clinical outcomes.

Methods

The authors developed and applied several novel quantitative techniques to analyze the spatial distribution of immune cells (specifically CD20+ B cells and CD8+ T cells) in tumor tissue images:

  • Occupancy: This method involves overlaying a grid on tissue images and asking a binary question for each square (e.g., "Is there at least one CD20+ B cell in this square?"). The "occupancy" is the fraction of squares answering "yes." This is analyzed across various square sizes (length scales) to understand spatial distribution.
  • Fractal Dimension: They used a variation of the box-counting method to quantify the irregularity of cell arrangements. A higher fractal dimension generally indicates a more complex or irregular shape. They specifically looked at the difference in fractal dimension across different length scales (Δs) to distinguish between clustered and spread-out cells.
  • Hotspot Analysis: This technique identifies regions within the image where cell density is significantly above average. They developed a local method using Gaussian distributions to represent cell density and calculated the fraction of the area with "hotspots."
  • Maximum Entropy Approach: This is a core novel contribution. They divide images into 3x3 matrices of squares and analyze the probability of different patterns (states) of cell presence (1s) and absence (0s). By maximizing entropy subject to constraints (like observed occupancy), they can infer underlying probability distributions. They analyze various "connected information moments" (S₁, S₂, S₃, S<0xE2><0x82><0x99>) and Kullback-Leibler divergences (D₀₁, D₁₂, D₂₃, D₀<0xE2><0x82><0x99>) derived from these probabilities to quantify spatial correlations and differences between distributions.

Datasets

The study utilized 2D images of archived formalin-fixed paraffin-embedded (FFPE) primary tumor tissue from 37 patients with triple-negative breast cancer (TNBC). These patients were categorized into two groups:

  • Good Clinical Outcome (n=24): No recurrence within 5 years after surgery.
  • Poor Clinical Outcome (n=13): Recurrence within 3 years after surgery.

The images were processed using multispectral staining to identify and label different cell types (T cells and B cells) with different color chromophores. Necrotic and fibrotic areas were excluded.

Key Insights and Ideas

  • Spatial Distribution Matters: The central idea is that how immune cells are arranged in a tumor, not just how many there are, is a critical factor in predicting cancer recurrence and patient outcome.
  • Physics-Inspired Metrics: The paper demonstrates the power of applying concepts from statistical physics and information theory (entropy, fractal dimensions, occupancy) to biological imaging data.
  • Multi-Scale Analysis: The developed techniques allow for the examination of spatial patterns across a range of length scales, providing a more comprehensive view than single-scale analyses.
  • Maximum Entropy as a Powerful Tool: The maximum entropy framework offers a principled way to infer underlying spatial distributions and quantify complex relationships between cell patterns and clinical outcomes.
  • Immune Surveillance Hypothesis: The findings suggest that a more spread-out distribution of immune cells might be more effective for immune surveillance, allowing for better interaction with cancer cells and presentation of antigens.

Key Results

  • B Cells: For CD20+ B cells, the area under the curve (AUC) for occupancy and entropy measures (S₁, S₂, S₃, S<0xE2><0x82><0x99>) was significantly larger for patients with good clinical outcome compared to poor outcome. This indicates B cells are more spread out in good outcome cases. The fractal dimension difference (Δs) was also greater for good outcome, supporting this finding.
  • CD8+ T Cells: While CD8+ T cell densities differed significantly between good and poor outcomes, the entropy curves showed less clear distinctions, often crossing. However, the fractal dimension difference (Δs) was greater for good outcome, suggesting T cells are also more spread out in these cases.
  • Predictive Power: The spatial distribution metrics, particularly S₁, I₁, and D₀₁, were sensitive to differences between good and poor outcomes, suggesting their potential as biomarkers for predicting recurrence. The authors note that these methods can predict recurrence with 60-80% accuracy.

Limitations and Future Work

  • Mechanism Unknown: The paper acknowledges that the exact reason why spatially dispersed immune cells lead to better outcomes is not fully understood. They speculate on immune surveillance and chemokine responsiveness but state it's an open question.
  • Physical Topography: While they investigated collagen I distribution, they did not find differences in physical topography that could explain the spatial distribution differences, suggesting other factors are at play.
  • Generalizability: The study focused on triple-negative breast cancer. Future work could explore these methods in other cancer types and for other immune cell populations.
  • Broader Applications: The authors suggest their flexible approaches could be applied to analyze spatial distributions in metastatic tumors, other cell types, and even in non-biological fields like astrophysics (galaxy distribution).
  • Co-localization: They mention straightforward extensions to quantify co-localization of different cell types or cells with other structures (e.g., blood vessels), which would be a valuable next step.

Summary updated

Abstract

The goal of immunotherapy is to mobilize the immune system to kill cancer cells. Immunotherapy is more effective and, in general, the prognosis is better, when more immune cells infiltrate the tumor. We explore the question of whether the spatial distribution rather than just the density of immune cells in the tumor is important in forecasting whether cancer recurs. After reviewing previous work on this issue, we introduce a novel application of maximum entropy to quantify the spatial distribution of discrete point-like objects. We apply our approach to B and T cells in images of tumor tissue taken from triple negative breast cancer patients. We find that the immune cells are more spatially dispersed in good clinical outcome (no recurrence of cancer within at least 5 years of diagnosis) compared to poor clinical outcome (recurrence within 3 years of diagnosis). Our results highlight the importance of spatial distribution of immune cells within tumors with regard to clinical outcome, and raise new questions on their role in cancer recurrence.
BibTeX
@inproceedings{yu2020physics,
  author = {Clare C. Yu and Juliana C. Wortman and Ting-Fang He and Shawn T. Solomon and Bob Zhang and Anthony Rosario and Ruo‐Qian Wang and Travis Y. Tu and Daniel Schmolze and Yuan Yuan and Susan E. Yost and Xuefei Li and Herbert Levine and Gurinder S. Atwal and Peter P. Lee},
  title = {{Physics approaches to the spatial distribution of immune cells in tumors}},
  booktitle = {Reports},
  year = {2020},
  doi = {10.1088/1361-6633/abcd7b},
}
Figure