Prognostic significance of spatial and density analysis of T lymphocytes in colorectal cancer

Background Although high T cell density is a strong favourable prognostic factor in colorectal cancer, the significance of the spatial distribution of T cells is incompletely understood. We aimed to evaluate the prognostic significance of tumour cell-T cell co-localisation and T cell densities. Methods We analysed CD3 and CD8 immunohistochemistry in a study cohort of 983 colorectal cancer patients and a validation cohort (N = 246). Individual immune and tumour cells were identified to calculate T cell densities (to derive T cell density score) and G-cross function values, estimating the likelihood of tumour cells being co-located with T cells within 20 µm radius (to derive T cell proximity score). Results High T cell proximity score associated with longer cancer-specific survival in both the study cohort [adjusted HR for high (vs. low) 0.33, 95% CI 0.20–0.52, Ptrend < 0.0001] and the validation cohort [adjusted HR for high (vs. low) 0.15, 95% CI 0.05–0.45, Ptrend < 0.0001] and its prognostic value was independent of T cell density score. Conclusions The spatial point pattern analysis of tumour cell-T cell co-localisation could provide detailed information on colorectal cancer prognosis, supporting the value of spatial measurement of T cell infiltrates as a novel, robust tumour-immune biomarker.

T cell proximity and density score analyses for two cores (one from tumor center (CT) and one from invasive margin (IM)) not represented in Figure 1. The panels show tumor cores stained with CD3 and CD8 (A), corresponding phenotyping maps for T cells, tumor cells and other cells (B), G-cross [Gtumor:T cell] function curves representing the likelihood of any tumor cell in the sample having at least one CD3 + /CD8 + T cell within a radius r (C).

Figure S3
Core-to-core correlation of G-cross (Gtumor:T cell) at 20 µm radius in two randomly chosen cores of tumors with two or more cores measured using Spearman's rank correlation coefficients (R). Red line depicts perfect concordance (slope=1). G-cross correlations for CD3 + and CD8 + cells are presented separately in the tumor center and in the invasive margin. N=913 for CD3 in the tumor center, N=725 for CD3 in the invasive margin, N=877 for CD8 in the tumor center, N=738 for CD8 in the invasive margin.

Figure S4
Core-to-core correlation of T cell densities in two randomly chosen cores of tumors with two or more cores measured using Spearman's rank correlation coefficients (R). Red line depicts perfect concordance (slope=1). CD3 + and CD8 + cell density correlations are presented separately in the tumor center and the invasive margin. N=913 for CD3 in the tumor center, N=725 for CD3 in the invasive margin, N=877 for CD8 in the tumor center, N=738 for CD8 in the invasive margin.

Figure S5
Correlation diagrams of manually and automatedly calculated cell densities measured using Spearman's rank correlation coefficients (R). Red line depicts perfect concordance (slope=1). Cell densities were calculated in 50 tumor regions manually and using the optimized, automated method utilizing the QuPath software. The correlations are presented separately for T cells, tumor cells and other cells.

Figure S6
Examples of tissue microarray cores with distinct T cell infiltration patterns. (A-D) CD3 + and CD8 + stained immunohistochemistry images, corresponding phenotyping maps for T cells, tumor cells and other cells, and G-cross (Gtumor:T cell) as a function of radius (r).

Figure S7
Kaplan-Meier cancer-specific survival curves for combined T cell proximity and density score variable. Statistical significance was determined with Log-rank test.

Figure S8
Kaplan-Meier cancer-specific survival curves for G-cross (Gtumor:T cell) proximity function values at 20 µm radius and for T cell densities. Analyses were done separately for CD3 + and CD8 + cells in the tumor center and invasive margin by using ordinal quartiles Q1-Q4 (from low to high). Statistical significance was determined with Log-rank test.

Table S11
Univariable and multivariable Cox regression models for cancer-specific survival and overall survival according to T cell proximity score and T cell density score in the validation cohort. Table S12. Comparison of prognostic power of T cell proximity score and T cell density score using Cox regression models for cancer-specific survival in the validation cohort. Model 2: Cox proportional hazards regression model including T cell proximity score and T cell density score. Model 3: Cox proportional hazards regression model based on Model 2 that was additionally adjusted for sex, age (<65, 65-75, >75), year of operation (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014), tumor location (proximal colon, distal colon, rectum), disease stage (I-II, III, IV), tumor grade (well/moderately differentiated, poorly differentiated), lymphovascular invasion (negative, positive), mismatch repair (MMR) status (proficient, deficient), BRAF status (wild-type, mutant). Ptrend values were calculated by using the three ordinal categories of T cell proximity score and T cell density score as continuous variables in univariable and multivariable Cox proportional hazard regression models.