Association of Epigenetic Clock with Consensus Molecular Subtypes and Overall Survival of Colorectal Cancer Free

Colorectal cancer is the third leading cause of cancer-related death in the United States. Colorectal cancer is highly heterogeneous and the five-year survival rate varies drastically with tumor stage at diagnosis (1–2). Age is a strong predictor of risk and outcome of colorectal cancer and other common complex diseases, including heart disease, stroke, and Alzheimer's disease (3). Several DNA methylation-based clocks developed recently have shown promise as biomarkers for biological aging. In particular, the Horvath clock, a multiple tissue age predictor, has been shown to be strongly and robustly correlated with chronologic age (4). Epigenetic age acceleration, that is, the difference between DNA methylation age and chronologic age, has been associated with increased risk of many diseases (5–8), including colorectal cancer and other cancers (9–12), as well as with all-cause mortality and known colorectal cancer risk factors such as obesity and smoking (13–15).

Studies of the association between epigenetic age acceleration and cancer outcomes, however, have generated inconsistent results. A large cohort study of 3,086 patients found no significant association of blood-based epigenetic age acceleration with survival in seven cancer types after adjusting for sociodemographic and lifestyle variables [HR = 1.02; 95% confidence interval (CI): 0.98–1.06; P = 0.38; ref. 16]. A study using The Cancer Genome Atlas (TCGA) data reported that the association of tissue-based (for solid tumors) and blood-based (for acute myeloid leukemia) epigenetic age acceleration with survival varies with cancer type (17). A negative relationship was observed in thyroid carcinoma and renal clear cell carcinoma while a positive relationship was observed in esophageal carcinoma. Another TCGA study of 1,076 patients with breast cancer reported that decreased epigenetic age (assessed in tumor tissue samples) is associated with poor prognosis after adjusting for major clinical variables, including tumor stage and estrogen receptor status (18). We speculate that the observed inconsistencies are due, in part, to molecular heterogeneity of cancers that were not fully taken into account in these studies.

The Colorectal Cancer Subtyping Consortium (CRCSC) has recently developed the Consensus Molecular Subtype (CMS), a robust gene expression-based molecular classification of colorectal cancer, which unifies six different classification systems (19). CMS groups colorectal cancer into four molecular subtypes: CMS1 (MSI-immune), CMS2 (Canonical), CMS3 (Metabolic), and CMS4 (Mesenchymal). CMS classifications have been shown to correlate with clinical characteristics of colorectal cancer and can be recapitulated both in vitro and in vivo (20–22). It has also been shown that CMS predicts chemotherapeutic efficacy in metastatic colorectal cancer (23). To date, the relationship of epigenetic clock and CMS of colorectal cancer remains unexplored. In this study, we examined the associations of epigenetic age acceleration with molecular characteristics (i.e., CMS), tumor clinical characteristics as well as with outcomes in 345 patients with colorectal cancer from TCGA.

Colonic tissue DNA methylation data (Illumina 450K platform) and available clinic information of 389 colorectal cancer samples and 38 matched normal samples were downloaded from TCGA. CMS assignments of these cancers were obtained from a large-scale study conducted by The CRC Subtyping Consortium (19). The CMS classification system based on 18 gene expression datasets including the TCGA classifies each colorectal cancer into one of the four molecular subtypes: CMS1 (MSI-immune), CMS2 (Canonical), CMS3 (Metabolic), CMS4 (Mesenchymal), or Unknown (19). Of the 389 colorectal cancer cases in our study, 44 patients that do not have CMS information were excluded, leaving 345 patients available for analysis, 32 of whom also have matched normal colon tissues.

We used Horvath model to calculate DNA methylation age (4). Briefly, the Horvath model uses beta values of 353 CpG loci to calculate DNA methylation age as following:

where F is a function for transformation of age and |{b_0},\ {b_1} \ldots \ {b_{353}}\ $| are coefficients obtained from the elastic net regression model. Epigenetic age acceleration is then estimated as the residual of regression of DNA methylation age on chronologic age (14, 16, 18, 24).

All statistical analyses were performed using R (Version: 3.4.0). Multivariable linear regression models were used to assess the associations of epigenetic age acceleration with clinical features of colorectal cancer, including molecular subtypes, tumor position, and stage. Kaplan–Meier curves were generated for survival rates of patients with different molecular subtypes and tumor stages and log-rank test was used to test the significance of difference. Cox proportional hazards regression was used to assess the association of epigenetic age acceleration with survival in both stratified and unstratified analyses.

The relationship between DNA methylation age and chronologic age is shown in Supplementary Fig. S1. Among the 32 patients with available adjacent normal colonic tissues, chronologic age is highly correlated with DNA methylation age in the colonic tissue (r = 0.898; Supplementary Fig. S1A). However, this correlation is largely lost in cancer tissues (r = 0.276; Supplementary Fig. S1B). These results indicate that the pattern of DNA methylation observed in normal colonic tissues is disrupted in colorectal cancer tissues.

Among the 345 patients with colorectal cancer with both methylation and CMS information, 77.4% were white, 46.1% were females, and 53.9% males. In univariate analysis, we observed significant association of CMS with epigenetic age acceleration (P = 3.47E-11; Table 1). CMS1 shows the biggest increase in DNA methylation age with an average of 14.76 years and CMS2 shows the biggest decrease with an average of 4.78 years (Table 1). Using CMS2 as the reference, epigenetic age acceleration is statistically significantly associated with CMS classifications after adjusting for age, gender, and tumor stage. Compared with CMS2, the other three CMS subtypes all show significant epigenetic age acceleration: 23.90 years for CMS1 (P = 5.55E-11; 95% CI: 17.10–30.69), 9.16 years for CMS3 (P = 5.84E-03; 95% CI: 2.68–15.65), and 6.05 years for CMS4 (P = 2.69E-02; 95% CI: 0.70–11.41; Table 2). All pairwise comparisons, except for the comparison between CMS3 and CMS4, show statistically significant differences (P < 0.05) in epigenetic age acceleration.

We next assessed the relationship of epigenetic age acceleration with patient survival. Epigenetic age acceleration showed no significant association with mortality in univariate analysis (Table 3), while age and tumor stage were found to be significant predictors of survival. Older patients had a HR of 1.73 (95% CI: 1.02–2.95) when compared with the younger group. Compared with patients with stage I disease, patients with stage IV colorectal cancer had the worst survival with a HR of 5.98 (95% CI: 2.00–19.92), and patients with stage III CRC had a HR of 2.82 (95% CI: 0.98–8.12). The Kaplan–Meier estimates were suggestive of an association between CMS classification and survival (P = 0.087; Supplementary Fig. S2).

We then stratified our analysis by age, tumor stage, and molecular subtype, and investigated whether epigenetic age acceleration is associated with survival in each of these patient strata. The stratified analysis showed no significant association for epigenetic age acceleration (Supplementary Fig. S3), except for patients with stage II colorectal cancer (HR = 1.36; 95% CI: 0.99–1.85; P = 0.056; Supplementary Fig. S3B) and CMS4 (HR = 1.42; 95% CI: 1.00–2.02; P = 0.052; Supplementary Fig. S3C).

We further examined the association of epigenetic age acceleration with survival for each CMS subtype, adjusted for age and tumor stage. We observed worsening survival with increased epigenetic age acceleration in CMS4 colorectal cancer (HR = 1.47; 95% CI: 1.02–2.11; P = 0.041; Supplementary Fig. S3D). In contrast, our analysis was suggestive of a beneficial but nonsignificant effect of epigenetic age acceleration on survival in CMS1 colorectal cancer (HR = 0.52; 95%CI: 0.24–1.14; P = 0.102). No significant association with epigenetic age acceleration was observed for CMS2 and CMS3 colorectal cancer.

Multivariate survival analysis adjusting for age, tumor stages, and molecular subtypes revealed no overall association for epigenetic age acceleration, while age and tumor stage remain significantly associated with patient survival (Table 4).

Our above analyses indicated a complex relationship of epigenetic age acceleration with overall survival in colorectal cancer, likely due partially to the molecular heterogeneity of colorectal cancer and potential nonlinear relationship between epigenetic age acceleration and overall survival. In subsequent analyses, we classified patients into two distinct groups based on their epigenetic age acceleration estimates: those showing decelerated aging, that is, with a negative epigenetic age acceleration estimate; and those with accelerated aging, that is, with a positive epigenetic age acceleration estimate. Interestingly, the Kaplan–Meier curves showed statistically significant difference between these two groups (P = 0.026) in survival (Fig. 1). Cox proportional regression analyses demonstrated that the accelerated aging group has worse survival with a HR of 1.69 (95% CI: 1.06–2.68; P = 0.028) in an unadjusted model. After adjusting for tumor stage and molecular subtypes, this association become even more pronounced (HR = 1.97; 95% CI: 1.14–3.39; P = 0.014; Table 5).

Colorectal cancer is a heterogeneous disease. Molecular characteristics-based CMS profile of colorectal cancer has been shown to be correlated with clinical features and outcomes in patients with cancer (19–21). Accumulating evidence suggests epigenetic age acceleration as a novel biomarker for cancer risk (9–12). In this study of 354 patients with colorectal cancer in TCGA, we systematically examined the association of epigenetic age acceleration with colorectal cancer tumor molecular characteristics (CMS subtypes), clinical characteristics, and patient outcomes. We found epigenetic age acceleration varies with CMS subtypes. Specifically, we found that CMS1 colorectal cancer exhibits the largest epigenetic age acceleration while CMS2 shows the least. We further showed that patients with accelerated aging as indicated by a positive epigenetic age acceleration estimate had worse survival than patients with decelerated aging. A previous large study of 2,129 patients with colorectal cancer reported that CMS1 colorectal cancer had worse overall survival compared with CMS2 colorectal cancer (19). Taken together, our data support epigenetic age acceleration as a promising prognostic factor for colorectal cancer.

Tissue source appears to have a noticeable effect on the reported roles of epigenetic age on cancers. In the study by Dugué and colleagues, blood-based epigenetic ages showed no association with clinical outcomes of several types of cancers (16), while the study by Lin and colleagues of 25 different types of cancers found varying correlations of tissue-based epigenetic age with cancer outcomes (17). Neither study revealed correlation of epigenetic age with colorectal cancer clinical outcomes, regardless of the tissue sources (blood or tumor tissue) used for epigenetic age assessment (16–17). In this study, epigenetic ages were derived from DNA methylation of colorectal cancer tissues. Consistent with the study by Lin and colleagues using tumor tissue–based epigenetic age (17), our univariate or stratified analysis, whether by age, by tumor stage, or by CMS, found no association of epigenetic age acceleration with patient outcomes. However, when we further examined the association of epigenetic age acceleration with survival in each molecular subtype adjusting for age and tumor stage, we observed a tendency for positive relationship of age acceleration with survival in CMS1 while negative relationship in CMS4. When we classified patients into two groups based on the positivity or negativity of their epigenetic age acceleration estimates, a strong association with patient overall survival emerged, and the association became more significant after adjusting for tumor stage and CMS subtype. Taking together, our results suggest that epigenetic age acceleration is a promising biomarker for colorectal cancer prognosis. When combined with tumor stage and molecular subtype, epigenetic age acceleration may provide better prediction for patient outcome.

One inherent limitation of our study is that we could not control all variables that may affect colorectal cancer survival due to the lack of information in TCGA. For example, treatment information and smoking status, which are important factors for patient prognosis and have been shown to be associated with epigenetic age acceleration (15), are not completely available in TCGA. Validation of our results in large prospective cohort studies is warranted.

No potential conflicts of interest were disclosed.

Conception and design: C. Zheng, L. Li, R. Xu

Development of methodology: C. Zheng, L. Li, R. Xu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Zheng

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Zheng, L. Li, R. Xu

Writing, review, and/or revision of the manuscript: C. Zheng, L. Li, R. Xu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Xu

Study supervision: L. Li, R. Xu

This work was supported by NIH grants (DP2HD084068, R01 AG057557-01, R01 AG061388-01, and R56 AG062272-01, to R. Xu; U01CA181770 and P20CA233216, to L. Li) and an American Cancer Society grant (RSG-16-049-01-MPC, to R. Xu).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.