Search

Histopathologic assessment is indispensable for diagnosing colorectal cancer (CRC). However, manual evaluation of the diseased tissues under the microscope cannot reliably inform patient prognosis or genomic variations crucial for treatment selections. To address these challenges, we develop the Multi-omics Multi-cohort Assessment (MOMA) platform, an explainable machine learning approach, to systematically identify and interpret the relationship between patients’ histologic patterns, multi-omics, and clinical profiles in three large patient cohorts (n = 1888). MOMA successfully predicts the overall survival, disease-free survival (log-rank test P-value<0.05), and copy number alterations of CRC patients. In addition, our approaches identify interpretable pathology patterns predictive of gene expression profiles, microsatellite instability status, and clinically actionable genetic alterations. We show that MOMA models are generalizable to multiple patient populations with different demographic compositions and pathology images collected from distinctive digitization methods. Our machine learning approaches provide clinically actionable predictions that could inform treatments for colorectal cancer patients.

Routine tumor-node-metastasis (TNM) staging of colorectal cancer is imperfect in predicting survival due to tumor pathobiological heterogeneity and imprecise assessment of tumor spread. We leveraged Bayesian additive regression trees (BART), a statistical learning technique, to comprehensively analyze patient-specific tumor characteristics for the improvement of prognostic prediction. Of 75 clinicopathologic, immune, microbial, and genomic variables in 815 stage II–III patients within two U.S.-wide prospective cohort studies, the BART risk model identified seven stable survival predictors. Risk stratifications (low risk, intermediate risk, and high risk) based on model-predicted survival were statistically significant (hazard ratios 0.19–0.45, vs. higher risk; P < 0.0001) and could be externally validated using The Cancer Genome Atlas (TCGA) data (P = 0.0004). BART demonstrated model flexibility, interpretability, and comparable or superior performance to other machine-learning models. Integrated bioinformatic analyses using BART with tumor-specific factors can robustly stratify colorectal cancer patients into prognostic groups and be readily applied to clinical oncology practice.

Although tumor-infiltrating T cells hold a beneficial prognostic role in colorectal cancer, other lymphocytic populations are less characterized. We developed a multiplexed immunofluorescence assay coupled with digital image analysis and machine learning to identify natural killer (NK) cells (NCAM1+CD3⁻), natural killer T-like (NKT-like) cells (NCAM1⁺CD3⁺), and T cells (NCAM1⁻CD3⁺) within the PTPRC⁺ (CD45⁺) cell population and to measure their granzyme B (GZMB; cytotoxicity marker) and FCGR3A (CD16a; NK-cell maturity marker) expression. We evaluated immune cell densities and spatial configuration in 907 incident colorectal carcinoma cases within two prospective cohort studies. We found that T cells were approximately 100 times more abundant than NK and NKT-like cells. Overall, NK cells showed high GZMB expression and were located closer to tumor cells than T and NKT-like cells. In T and NKT-like cells, GZMB expression was enriched in cells in closer proximity to tumor cells. Higher densities of both T and NKT-like cells associated with longer cancer-specific survival, independent of potential confounders (Ptrend < 0.0007). Higher stromal GZMB+ and FCGR3A+ NK-cell densities associated with longer cancer-specific survival (Ptrend < 0.003). For T and NKT-like cells, greater proximity to tumor cells associated with longer cancer-specific survival (Ptrend < 0.0001). These findings indicate that cytotoxic NCAM1⁺CD3⁻GZMB⁺ NK cells and NCAM1vCD3v NKT-like cells are relatively rare lymphocytic populations within the colorectal cancer microenvironment and show distinct spatial configuration and associations with patient outcome. The results highlight the utility of a quantitative multimarker assay for in situ, single-cell immune biomarker evaluation and underscore the importance of spatial context for tumor microenvironment characterization.

Background: The relationships between tumor stromal features (such as desmoplastic reaction, myxoid stroma, and keloid-like collagen bundles) and immune cells in the colorectal carcinoma microenvironment have not yet been fully characterized. Methods: In 908 tumors with available tissue among 4,465 incident colorectal adenocarcinoma cases in two prospective cohort studies, we examined desmoplastic reaction, myxoid stroma, and keloid-like collagen bundles. We conducted multiplex immunofluorescence for T cells [CD3, CD4, CD8, CD45RO (PTPRC), and FOXP3] and for macrophages [CD68, CD86, IRF5, MAF, and MRC1 (CD206)]. We used the inverse probability weighting method and the 4,465 incident cancer cases to adjust for selection bias. Results: Immature desmoplastic reaction was associated with lower densities of intraepithelial CD3⁺CD8⁺CD45RO⁺ cells [multivariable odds ratio (OR) for the highest (vs. lowest) density category, 0.43; 95% confidence interval (CI), 0.29–0.62; Ptrend <0.0001] and stromal M1-like macrophages [the corresponding OR, 0.44; 95% CI, 0.28–0.70; Ptrend = 0.0011]. Similar relations were observed for myxoid stroma [intraepithelial CD3⁺CD8⁺CD45RO⁺ cells (Ptrend <0.0001) and stromal M1-like macrophages (Ptrend = 0.0007)] and for keloid-like collagen bundles (Ptrend <0.0001 for intraepithelial CD3⁺CD8⁺CD45RO⁺ cells). In colorectal cancer-specific survival analyses, multivariable-adjusted hazard ratios (with 95% confidence intervals) were 0.32 (0.23–0.44; Ptrend <0.0001) for mature (vs. immature) desmoplastic reaction, 0.25 (0.16–0.39; Ptrend <0.0001) for absent (vs. marked) myxoid stroma, and 0.12 (0.05–0.28; Ptrend <0.0001) for absent (vs. marked) keloid-like collagen bundles. Conclusions: Immature desmoplastic reaction and myxoid stroma were associated with lower densities of tumor intraepithelial memory cytotoxic T cells and stromal M1-like macrophages, likely reflecting interactions between tumor, immune, and stromal cells in the

Evidence indicates the pathogenic role of epigenetic alterations in early-onset colorectal cancers diagnosed before age 50. However, features of colorectal cancers diagnosed at age 50–54 (hereafter referred to as “intermediate-onset”) remain less known. We hypothesized that tumor long interspersed nucleotide element-1 (LINE-1) hypomethylation might be increasingly more common with decreasing age of colorectal cancer diagnosis. In 1356 colorectal cancers, including 28 early-onset and 66 intermediate-onset cases, the tumor LINE-1 methylation level measured by bisulfite-PCR-pyrosequencing (scaled 0 to 100) showed a mean of 63.6 (standard deviation (SD) 10.1). The mean tumor LINE-1 methylation level decreased with decreasing age (mean 64.7 (SD 10.4) in age ≥70, 62.8 (SD 9.4) in age 55–69, 61.0 (SD 10.2) in age 50–54, and 58.9 (SD 12.0) in age <50; p < 0.0001). In linear regression analysis, the multivariable-adjusted β coefficient (95% confidence interval (CI)) (vs. age ≥70) was −1.38 (−2.47 to −0.30) for age 55–69, −2.82 (−5.29 to −0.34) for age 50–54, and −4.54 (−8.24 to −0.85) for age <50 (Ptrend = 0.0003). Multivariable-adjusted hazard ratios (95% CI) for LINE-1 methylation levels of ≤45, 45–55, and 55–65 (vs. >65) were 2.33 (1.49–3.64), 1.39 (1.05–1.85), and 1.29 (1.02–1.63), respectively (Ptrend = 0.0005). In conclusion, tumor LINE-1 hypomethylation is increasingly more common with decreasing age of colorectal cancer diagnosis, suggesting a role of global DNA hypomethylation in colorectal cancer arising in younger adults.

Macrophages are among the most common cells in the colorectal cancer microenvironment, but their prognostic significance is incompletely understood. Using multiplexed immunofluorescence for CD68, CD86, IRF5, MAF, MRC1 (CD206), and KRT (cytokeratins) combined with digital image analysis and machine learning, we assessed the polarization spectrum of tumor-associated macrophages in 931 colorectal carcinomas. We then applied Cox proportional hazards regression to assess prognostic survival associations of intraepithelial and stromal densities of M1-like and M2-like macrophages while controlling for potential confounders, including stage and microsatellite instability status. We found that high tumor stromal density of M2-like macrophages was associated with worse cancer-specific survival, whereas tumor stromal density of M1-like macrophages was not significantly associated with better cancer-specific survival. High M1:M2 density ratio in tumor stroma was associated with better cancer-specific survival. Overall macrophage densities in tumor intraepithelial or stromal regions were not prognostic. These findings suggested that macrophage polarization state, rather than their overall density, was associated with cancer-specific survival, with M1- and M2-like macrophage phenotypes exhibiting distinct prognostic roles. These results highlight the utility of a multimarker strategy to assess the macrophage polarization at single-cell resolution within the tumor microenvironment.

Immunotherapy targeting the CD274 (PD-L1)/PDCD1 (PD-1) immune checkpoint axis has emerged as a promising treatment strategy for various cancers. Experimental evidence suggests that phosphatidylinositol-4,5-bisphosphonate 3-kinase (PI3K) signaling may upregulate CD274 expression. Thus, we hypothesized that PIK3CA mutation, PTEN loss, or their combined status might be associated with CD274 overexpression in colorectal carcinoma. We assessed tumor CD274 and PTEN expression by immunohistochemistry and assessed PIK3CA mutation by pyrosequencing in 753 patients among 4,465 incident rectal and colon cancer cases that had occurred in two U.S.-wide prospective cohort studies. To adjust for potential confounders and selection bias due to tissue availability, inverse probability weighted multivariable ordinal logistic regression analyses used the 4,465 cases and tumoral data including microsatellite instability, CpG island methylator phenotype, KRAS and BRAF mutations. PIK3CA mutation and loss of PTEN expression were detected in 111 of 753 cases (15%) and 342 of 585 cases (58%), respectively. Tumor CD274 expression was negative in 306 (41%), low in 195 (26%), and high in 252 (33%) of 753 cases. PTEN loss was associated with CD274 overexpression [multivariable odds ratio (OR) 1.83; 95% confidence interval (CI), 1.22–2.75; P = .004]. PIK3CA mutation was statistically-insignificantly (P = .036 with the stringent alpha level of 0.005) associated with CD274 overexpression (multivariable OR, 1.54; 95% CI, 1.03–2.31). PIK3CA-mutated PTEN-lost tumors (n = 33) showed higher prevalence of CD274-positivity (82%) than PIK3CA-wild-type PTEN-lost tumors (n = 204; 70% CD274-positivity) and PTEN-expressed tumors (n = 147; 50% CD274-positivity) (P = .003). Our findings support the role of PI3K signaling in the CD274/PDCD1 pathway.

Background: Myeloid cells represent an abundant yet heterogeneous cell population in the colorectal cancer microenvironment, and their roles remain poorly understood. Methods: We used multiplexed immunofluorescence combined with digital image analysis to identify CD14⁺ monocytic and CD15⁺ granulocytic cells and to evaluate their maturity (HLA-DR and CD33), immunosuppressive potential (ARG1) and proximity to cytokeratin (KRT)-positive tumor cells in 913 colorectal carcinomas. Using covariate data of 4465 incident colorectal cancers in two prospective cohort studies, the inverse probability weighting method was used with multivariable-adjusted Cox proportional hazards models to assess cancer-specific mortality according to ordinal quartiles (Q1–Q4) of myeloid cell densities. Immune cell–tumor cell proximity was measured with the nearest neighbor method and the G-cross function, which determines the likelihood of any tumor cell having at least one immune cell of the specified type within a certain radius. Results: Higher intraepithelial (Ptrend=0.0002; HR for Q4 (vs Q1), 0.48, 95% CI 0.31 to 0.76) and stromal (Ptrend <0.0001; HR for Q4 (vs Q1), 0.42, 95% CI 0.29 to 0.63) densities of CD14⁺HLA-DR⁺ cells were associated with lower colorectal cancer-specific mortality while, conversely, higher intraepithelial densities of CD14⁺HLA-DR⁻ cells were associated with higher colorectal cancer-specific mortality (Ptrend=0.0003; HR for Q4 (vs Q1), 1.78, 95% CI 1.25 to 2.55). Spatial analyses indicated that CD15⁺ cells were located closer to tumor cells than CD14⁺ cells, and CD14⁺HLA-DR⁺ cells were closer to tumor than CD14⁺HLA-DR⁻ cells (p<0.0001). The G-cross proximity measurement, evaluating the difference in the likelihood of any tumor cell being colocated with at least one CD14⁺HLA-DR⁺ cell versus CD14⁺HLA-DR⁻ cell within a 20 µm radius, was associated with lower colorectal cancer-specific mortality (Ptrend <0.0001; HR for Q4 (vs Q1), 0.37, 95% CI 0.24 to 0.5

Purpose: Although high T-cell density is a well-established favorable prognostic factor in colorectal cancer, the prognostic significance of tumor-associated plasma cells, neutrophils, and eosinophils is less well-defined. Experimental Design: We computationally processed digital images of hematoxylin and eosin (H&E)–stained sections to identify lymphocytes, plasma cells, neutrophils, and eosinophils in tumor intraepithelial and stromal areas of 934 colorectal cancers in two prospective cohort studies. Multivariable Cox proportional hazards regression was used to compute mortality HR according to cell density quartiles. The spatial patterns of immune cell infiltration were studied using the GTumor:Immune cell function, which estimates the likelihood of any tumor cell in a sample having at least one neighboring immune cell of the specified type within a certain radius. Validation studies were performed on an independent cohort of 570 colorectal cancers. Results: Immune cell densities measured by the automated classifier demonstrated high correlation with densities both from manual counts and those obtained from an independently trained automated classifier (Spearman’s ρ 0.71–0.96). High densities of stromal lymphocytes and eosinophils were associated with better cancer-specific survival [Ptrend < 0.001; multivariable HR (4th vs 1st quartile of eosinophils), 0.49; 95% confidence interval, 0.34–0.71]. High GTumor:Lymphocyte area under the curve (AUC0,20μm; Ptrend = 0.002) and high GTumor:Eosinophil AUC0,20μm (Ptrend < 0.001) also showed associations with better cancer-specific survival. High stromal eosinophil density was also associated with better cancer-specific survival in the validation cohort (Ptrend < 0.001). Conclusions: These findings highlight the potential for machine learning assessment of H&E-stained sections to provide robust, quantitative tumor-immune biomarkers for precision medicine.

Accumulating evidence suggests that metabolic reprogramming has a critical role in carcinogenesis and tumor progression. The usefulness of formalin-fixed paraffin-embedded (FFPE) tissue material for metabolomics analysis as compared with fresh frozen tissue material remains unclear. LC/MS-MS–based metabolomics analysis was performed on 11 pairs of matched tumor and normal tissues in both FFPE and fresh frozen tissue materials from patients with colorectal carcinoma. Permutation t test was applied to identify metabolites with differential abundance between tumor and normal tissues. A total of 200 metabolites were detected in the FFPE samples and 536 in the fresh frozen samples. The preservation of metabolites in FFPE samples was diverse according to classes and chemical characteristics, ranging from 78% (energy) to 0% (peptides). Compared with the normal tissues, 34 (17%) and 174 (32%) metabolites were either accumulated or depleted in the tumor tissues derived from FFPE and fresh frozen samples, respectively. Among them, 15 metabolites were common in both FFPE and fresh frozen samples. Notably, branched chain amino acids were highly accumulated in tumor tissues. Using KEGG pathway analyses, glyoxylate and dicarboxylate metabolism, arginine and proline, glycerophospholipid, and glycine, serine, and threonine metabolism pathways distinguishing tumor from normal tissues were found in both FFPE and fresh frozen samples. This study demonstrates that informative data of metabolic profiles can be retrieved from FFPE tissue materials.

Background: Smoking has been associated with worse colorectal cancer patient survival and may potentially suppress the immune response in the tumor microenvironment. We hypothesized that the prognostic association of smoking behavior at colorectal cancer diagnosis might differ by lymphocytic reaction patterns in cancer tissue. Methods: Using 1474 colon and rectal cancer patients within 2 large prospective cohort studies (Nurses’ Health Study and Health Professionals Follow-up Study), we characterized 4 patterns of histopathologic lymphocytic reaction, including tumor-infiltrating lymphocytes (TILs), intratumoral periglandular reaction, peritumoral lymphocytic reaction, and Crohn’s-like lymphoid reaction. Using covariate data of 4420 incident colorectal cancer patients in total, an inverse probability weighted multivariable Cox proportional hazards regression model was conducted to adjust for selection bias due to tissue availability and potential confounders, including tumor differentiation, disease stage, microsatellite instability status, CpG island methylator phenotype, long interspersed nucleotide element-1 methylation, and KRAS, BRAF, and PIK3CA mutations. Results: The prognostic association of smoking status at diagnosis differed by TIL status. Compared with never smokers, the multivariable-adjusted colorectal cancer–specific mortality hazard ratio for current smokers was 1.50 (95% confidence interval = 1.10 to 2.06) in tumors with negative or low TIL and 0.43 (95% confidence interval = 0.16 to 1.12) in tumors with intermediate or high TIL (2-sided Pinteraction = 0.009). No statistically significant interactions were observed in the other patterns of lymphocytic reaction. Conclusions: The association of smoking status at diagnosis with colorectal cancer mortality may be stronger for carcinomas with negative or low TIL, suggesting a potential interplay of smoking and lymphocytic reaction in the colorectal cancer microenvironment.

Background/Objectives: Tumour budding and poorly differentiated clusters (PDC) represent forms of tumour invasion. We hypothesised that T-cell densities (reflecting adaptive anti-tumour immunity) might be inversely associated with tumour budding and PDC in colorectal carcinoma. Methods: Utilising 915 colon and rectal carcinomas in two U.S.-wide prospective cohort studies, and multiplex immunofluorescence combined with machine learning algorithms, we assessed CD3, CD4, CD8, CD45RO (PTPRC), and FOXP3 co-expression patterns in lymphocytes. Tumour budding and PDC at invasive fronts were quantified by digital pathology and image analysis using the International tumour Budding Consensus Conference criteria. Using covariate data of 4,420 incident colorectal cancer cases, inverse probability weighting (IPW) was integrated with multivariable logistic regression analysis that assessed the association of T-cell subset densities with tumour budding and PDC while adjusting for selection bias due to tissue availability and potential confounders, including microsatellite instability status. Findings: Tumour budding counts were inversely associated with density of CD3+CD8+ [lowest vs. highest: multivariable odds ratio (OR), 0.50; 95% confidence interval (CI), 0.35–0.70; Ptrend < 0.001] and CD3+CD8+CD45RO+ cells (lowest vs. highest: multivariable OR, 0.44; 95% CI, 0.31–0.63; Ptrend < 0.001) in tumour epithelial region. Tumour budding levels were associated with higher colorectal cancer-specific mortality (multivariable hazard ratio, 2.13; 95% CI, 1.57–2.89; Ptrend < 0.001) in Cox regression analysis. There were no significant associations of PDC with T-cell subsets. Interpretation: Tumour epithelial naïve and memory cytotoxic T cell densities are inversely associated with tumour budding at invasive fronts, suggesting that cytotoxic anti-tumour immunity suppresses tumour microinvasion.