Comprehensive Analysis of DNA Methylation and Prediction of Response to NeoadjuvantTherapy in Locally Advanced Rectal Cancer

Abstract

The treatment for locally advanced rectal carcinomas (LARC) is based on neoadjuvant chemoradiotherapy (nCRT) and surgery, which results in pathological complete response (pCR) in up to 30% of patients. Since epigenetic changes may influence response to therapy, we aimed to identify DNA methylation markers predictive of pCR in LARC patients treated with nCRT. We used high-throughput DNA methylation analysis of 32 treatment-naïve LARC biopsies and five normal rectal tissues to explore the predictive value of differentially methylated (DM) CpGs. External validation was carried out with The Cancer Genome Atlas-Rectal Adenocarcinoma (TCGA-READ 99 cases). A classifier based on three-CpGs DM (linked to OBSL1, GPR1, and INSIG1 genes) was able to discriminate pCR from incomplete responders with high sensitivity and specificity. The methylation levels of the selected CpGs confirmed the predictive value of our classifier in 77 LARCs evaluated by bisulfite pyrosequencing. Evaluation of external datasets (TCGA-READ, GSE81006, GSE75546, and GSE39958) reproduced our results. As the three CpGs were mapped near to regulatory elements, we performed an integrative analysis in regions associated with predicted cis-regulatory elements. A positive and inverse correlation between DNA methylation and gene expression was found in two CpGs. We propose a novel predictive tool based on three CpGs potentially useful for pretreatment screening of LARC patients and guide the selection of treatment modality.

Keywords: 5-fluorouracil; high-throughput DNA methylation analysis; predictive biomarker; translational research.

Figure 1

Methylation profile of locally advanced rectal carcinomas (LARC). (a) The unsupervised hierarchical clustering analysis of the methylation data using the most variable probes (6284 probes presenting β values with Standard Deviation >0.2) in the tumor samples revealed three major groups. Rows indicate the CpG sites, while columns represent samples. Clinical features of each case are represented below the heatmap along with targeted next-generation sequencing data for specific genes. Metastases were identified during the follow-up and after the treatment. (b) Distribution of the differentially methylated CpG probes in the comparison of pathological complete (pCR) or incomplete (pIR) response and normal tissue (NT) cases. The proportion of CpGs in relation to its location relative to the promoter, gene body, and intergenic regions; and to the CpG islands context was based on the Illumina EPIC annotation.

Figure 2

Sequential steps used to develop the DNA methylation classifier to predict neoadjuvant chemoradiotherapy (nCRT) response in rectal cancer (ReCa). (a) The methylation status of 99 ReCa and 7 adjacent normal tissue (ANT) samples from TCGA (Infinium Human Methylation 450K BeadChip platform, Illumina) was assessed for comparison. The identified differentially methylated (DM) probes were contrasted to those detected in our study (32 ReCa vs. 5 Normal tissues evaluated using the Infinium Human Methylation EPIC BeadChip platform–850K, Illumina), confirming 81.2% of DM probes in both studies (limma, False Discovery Rate < 5%; |∆β| > 0.15). (b) The ReCa cases were divided according to the response (11 pCR and 21 pIR) to nCRT to identify DM probes with potential predictive value. After filtering, a predictive classifier was trained by Diagonal Linear Discriminant Analysis (DLDA) using three probes (cg13770628, cg01072658, cg03085846). This classifier was able to distinguish pCR from pIR cases with 100% sensitivity and 90% specificity using the Leave-One-Out Cross-Validation (LOOCV) model. (c) The methylation status of these three probes was confirmed by bisulfite pyrosequencing in a set of 68 samples (32 array-dependent and 36 array-independent). * Only high-quality pyrosequencing results were included in the validation set. Sensitivity and specificity were calculated using LOOCV. FF: fresh frozen; FFPE: Formalin-fixed paraffin-embedded; DM: differentially methylated; ReCa: Rectal Cancer; NT: Normal Tissue; ANT: Adjacent Normal Tissue; pCR pathological Complete Response; pIR: pathological Incomplete Response.

Figure 3

Development and performance of the classifier associated with response to treatment. (a) Venn diagram showing the number of differentially methylated probes overlapping among the group comparisons (pCR vs. NT, pIR vs. NT, and pCR vs. pIR) that were used for developing a predictive model based on Diagonal Linear Discriminant Analysis (DLDA). (b) Tridimensional distribution of samples according to the methylation values of the three probes included in the classifier and their performance in discriminating patients with pCR or pIR tested using Leave-One-Out Cross-Validation (LOOCV). Normal tissue (NT) samples were also tested using the classifier. (c) Methylation status of three CpGs (CpG-A: cg01072658, CpG-B: cg03085846, CpG-C: cg13770628) identified as potential biomarkers of response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer evaluated by bisulfite pyrosequencing. BS-pyroseq: bisulfite pyrosequencing, pCR: pathological complete response, pIR: pathological incomplete response.

Figure 4

Correlation between the methylation levels of the three CpGs of LARC samples determined by microarray (Infinium MethylationEPIC BeadChip array, Illumina) and bisulfite (BS) pyrosequencing analysis. The scatterplots show a high positive correlation for three CpGs identified as a predictive classifier of response to neoadjuvant chemoradiotherapy. (r) Pearson’s correlation coefficient, (p) p-value from Pearson’s correlation test.

Figure 5

DNA methylation levels of 45 CpGs mapped in the region of the OBSL1 gene in (a) normal rectal tissue and LARC samples classified according to nCRT response (b) pCR—complete response and (c): pIR—incomplete response). The symbol * indicates significant differences between the groups; the CpG-A (cg01072658) is highlighted in pink. (d) Correlation between the DNA methylation levels of the CpG-A and the OBSL1 gene expression levels in 27 LARC from our cohort and (e) in a set of LARC from TCGA-READ data collection. (f) The expression levels of the OBSL1 gene were lower in the pIR group (p < 0.05).