Poster M2 - AI-Enhanced Classification of Mechanical States in Cancer-associated Fibroblasts

Introduction: Understanding the mechanical states of cells is crucial for advancing comprehension of biological phenomena and developing therapeutic strategies. This understanding enhances the depth of insights derived from molecular biology research, bridging gaps in the mechanism of biological processes. Such an approach has similarly manifested in cell classification methodologies. Recent studies have elucidated cellular states classified based on morphology and motility as having distinct molecular biological properties, suggesting cellular states examined by mechanical features could fill knowledge gaps in understanding biological processes. Mechanical features, such as morphology, motility, traction force, and actin filament structure, are critical to this analysis. Although traction force plays a crucial role in cellular interaction with the extracellular matrix, it has only been utilized for some types of cells, including cancer cells. CAFs (Cancer-associated fibroblasts) constitute a significant component in the tumor microenvironment and interact with cancer cells, contributing to tumor growth, invasion, and metastasis. CAFs exhibit heterogeneity in subtypes and functionalities, which affect interaction with cancer cells and responses to therapies, leading to a complicated understanding of cancer malignancy. Therefore, classifying and understanding the mechanical states of CAFs is essential for advancing the understanding of cancer progression and treatment strategies. This study developed a sophisticated pipeline to classify and interpret the mechanical states of CAFs by utilizing phase contrast images and traction force maps. We revealed distinct mechanical states within CAFs, highlighting their potential implications in cancer biology.

Materials and

Methods: The breast CAFs (BioIVT) were cultured on fluorescent beads-embedded 2kPa PA(Polyacrylamide)-gel in DMEM (Dulbecco’s modified eagle’s medium) diluted with 10% FBS and 1% Penicillin-Streptomycin a day before imaging. We captured phase contrast snapshot images and traction force microscopy of breast CAFs. We utilized a DCNN (Deep convolutional neural network) model called Cellpose (Stringer C et al., 2021) with fine-tuning to predict cell-body masks. Using cell-body masks to facilitate traction analysis, we made traction-influenced masks that capture traction generations around cells. We extracted 132 mechanical features from phase contrast images, traction magnitude maps, and traction directionality maps by applying 30 Zernike moments, 9 texture features, and 10 traditional morphology features with cell-body and traction-influenced masks. We executed feature selection through factor analysis that drops features that show high correlation (>0.9) or low communality ( < 0.4), then finally retained 26 mechanical features. In order to find the axes most preserving variance of mechanical features, PCA (Principal component analysis) was performed, and mechanical states were classified by K-means clustering in PC space.

Results, Conclusions, and Discussions: Mechanical features such as spatial distribution of traction force and randomness in traction force direction were shown to be crucial factors in distinguishing mechanical states of cancer-associated fibroblasts. We found that three distinct mechanical states were identified. Mechanical state 1 was characterized by an internal distribution of randomly oriented traction force. Mechanical state 2 showed an internal distribution of traction force as shown in state 1 but showed a uniform orientation of traction force at adjacent local sites in the cell, similar to state 3. Mechanical state 3 showed the traction force in a uniform direction, mainly located at the periphery region of cells.
Based on the characteristics of traction forces, mechanical state 1 may be a state that occurs due to the process of cell division or expansion or maybe an active state for deliberately reorganizing the structure of a well-aligned extracellular matrix. In mechanical state 2, well-aligned traction in the peripheral region could be related to helping the nature of cancer cell invasion, which is penetrating the extracellular matrix or tissue. In mechanical state 3, since the aligned traction force is distributed at the cell periphery, mechanical state 1 is supposed to be a cell state specialized in contractibility or mobility. In summary, it is suggested that the distribution of traction in the cell center may be related to invasive behavior, the distribution of traction around the cell periphery may be related to cell migration or contraction, and the random orientation of traction force may be related to specific cell behavior such as reorganization of extracellular matrix.
The pipeline developed in this study provides implications of mechanical states for cellular functions. However, experimental studies will be needed to confirm the relationship between the mechanical state and cellular functions. The mechanobiological mechanism is expected to be revealed based on a combinational study of the mechanical state and functionality analysis. Since this study demonstrated only breast cancer-associated fibroblasts but can be used for other types of cells, it is expected to help suggest strategies for improving health and treating diseases based on understanding the mechanical characteristics of various cells.

Acknowledgements (Optional): This research was supported by National Research Funding granted by the Korean Government (NRF-2021R1A2C3008408).