Functional Data Analysis of Spatial Trajectories measured by spatial protein imaging with application to the ovarian cancer tumor microenvironment
Presenter Status
Post-Doctorial Research
Abstract Type
Translational Research
Primary Mentor or Principal Investigator
Brooke Fridley
Presentation Type
Poster
Start Date
21-5-2026 11:00 AM
End Date
21-5-2026 12:00 PM
Abstract Text
Background: Advances in spatial technologies allow researchers to investigate both the composition and spatial organization of cell types in the tumor microenvironment (TME). To quantify spatial clustering, analyses often rely on summary functions such as Ripley’s K and the nearest‑neighbor G. When computed over a series of radii, these functions provide insight into clustering patterns across different spatial scales within the tissue.
Objectives/Goal: To address this, we propose a functional principal component analysis (FPCA) framework to capture spatial clustering patterns of T‑cell populations in the TME and assess how these patterns relate to survival in high‑grade serous ovarian cancer (HGSOC).
Methods/Design: We applied FPCA to evaluate how spatial clustering of CD3⁺ and CD3⁺CD8⁺ T‑cell populations within the ovarian TME relates to survival. The analysis included five ovarian cancer studies: the Nurses’ Health Study (N=239), Nurses’ Health Study II (N=68), the New England Case–Control Study of Ovarian Cancer (N=175), the African American Cancer Epidemiology Study (N=155), and the North Carolina Ovarian Cancer Study (N=136). For each sample, spatial trajectories were derived using the G statistic, restricted to cases containing at least eight positive cells of the relevant phenotype. FPCA was then applied to the resulting spatial curves, and the leading functional principal component (FPC1) was tested for association with survival, adjusting for cancer stage, age at diagnosis, and overall cell abundance (classified as high vs. low using a 1% threshold). A secondary model evaluated the interaction between cell abundance and spatial clustering. Analyses were conducted separately within each study, and effect estimates were subsequently combined using a random‑effects meta‑analysis.
Results: In models that did not incorporate spatial clustering, higher abundance of CD3⁺ cells (hazard ratio [HR] = 0.81; 95% CI: 0.66–0.98) and CD3⁺CD8⁺ cells (HR = 0.64; 95% CI: 0.52–0.79) was associated with improved survival. When spatial information was included alongside abundance, we observed a significant effect of CD3⁺CD8⁺ spatial clustering (FPC1 HR = 1.17; 95% CI: 1.04–1.33) and a borderline effect for CD3⁺ clustering (FPC1 HR = 1.06; 95% CI: 0.99–1.14). Incorporating an interaction term between abundance and spatial clustering revealed a significant interaction for CD3⁺ cells (HR = 1.23; 95% CI: 1.07–1.42) and a borderline interaction for CD3⁺CD8⁺ cells (HR = 1.19; 95% CI: 0.98–1.43). Motivated by this interaction, we estimated hazard ratios across four tumor profiles defined by high/low abundance and high/low spatial clustering. Patients with high abundance but low spatial clustering of CD3⁺ or CD3⁺CD8⁺ cells exhibited the best survival outcomes, with HRs of 0.74 and 0.41, respectively.
Conclusions: In studying the HGSOC tumor microenvironment using spatial proteomics and FPCA, we found that both the abundance and the spatial clustering of T‑cell populations were associated with survival, with the most favorable outcomes observed in tumors exhibiting diffuse T‑cell infiltration. These findings demonstrate that incorporating spatial architecture alongside cell abundance provides a more nuanced understanding of immune influences on HGSOC prognosis.
Functional Data Analysis of Spatial Trajectories measured by spatial protein imaging with application to the ovarian cancer tumor microenvironment
Background: Advances in spatial technologies allow researchers to investigate both the composition and spatial organization of cell types in the tumor microenvironment (TME). To quantify spatial clustering, analyses often rely on summary functions such as Ripley’s K and the nearest‑neighbor G. When computed over a series of radii, these functions provide insight into clustering patterns across different spatial scales within the tissue.
Objectives/Goal: To address this, we propose a functional principal component analysis (FPCA) framework to capture spatial clustering patterns of T‑cell populations in the TME and assess how these patterns relate to survival in high‑grade serous ovarian cancer (HGSOC).
Methods/Design: We applied FPCA to evaluate how spatial clustering of CD3⁺ and CD3⁺CD8⁺ T‑cell populations within the ovarian TME relates to survival. The analysis included five ovarian cancer studies: the Nurses’ Health Study (N=239), Nurses’ Health Study II (N=68), the New England Case–Control Study of Ovarian Cancer (N=175), the African American Cancer Epidemiology Study (N=155), and the North Carolina Ovarian Cancer Study (N=136). For each sample, spatial trajectories were derived using the G statistic, restricted to cases containing at least eight positive cells of the relevant phenotype. FPCA was then applied to the resulting spatial curves, and the leading functional principal component (FPC1) was tested for association with survival, adjusting for cancer stage, age at diagnosis, and overall cell abundance (classified as high vs. low using a 1% threshold). A secondary model evaluated the interaction between cell abundance and spatial clustering. Analyses were conducted separately within each study, and effect estimates were subsequently combined using a random‑effects meta‑analysis.
Results: In models that did not incorporate spatial clustering, higher abundance of CD3⁺ cells (hazard ratio [HR] = 0.81; 95% CI: 0.66–0.98) and CD3⁺CD8⁺ cells (HR = 0.64; 95% CI: 0.52–0.79) was associated with improved survival. When spatial information was included alongside abundance, we observed a significant effect of CD3⁺CD8⁺ spatial clustering (FPC1 HR = 1.17; 95% CI: 1.04–1.33) and a borderline effect for CD3⁺ clustering (FPC1 HR = 1.06; 95% CI: 0.99–1.14). Incorporating an interaction term between abundance and spatial clustering revealed a significant interaction for CD3⁺ cells (HR = 1.23; 95% CI: 1.07–1.42) and a borderline interaction for CD3⁺CD8⁺ cells (HR = 1.19; 95% CI: 0.98–1.43). Motivated by this interaction, we estimated hazard ratios across four tumor profiles defined by high/low abundance and high/low spatial clustering. Patients with high abundance but low spatial clustering of CD3⁺ or CD3⁺CD8⁺ cells exhibited the best survival outcomes, with HRs of 0.74 and 0.41, respectively.
Conclusions: In studying the HGSOC tumor microenvironment using spatial proteomics and FPCA, we found that both the abundance and the spatial clustering of T‑cell populations were associated with survival, with the most favorable outcomes observed in tumors exhibiting diffuse T‑cell infiltration. These findings demonstrate that incorporating spatial architecture alongside cell abundance provides a more nuanced understanding of immune influences on HGSOC prognosis.
Comments
Restricted to Title/Author List/Abstract only as requested by primary author
Poster Board Number: 33