A Liquid Biopsy Assay for Noninvasive Identification of Lymph Node Metastases in T1 Colorectal Cancer

Abstract

Background & aims: We recently reported use of tissue-based transcriptomic biomarkers (microRNA [miRNA] or messenger RNA [mRNA]) for identification of lymph node metastasis (LNM) in patients with invasive submucosal colorectal cancers (T1 CRC). In this study, we translated our tissue-based biomarkers into a blood-based liquid biopsy assay for noninvasive detection of LNM in patients with high-risk T1 CRC.

Methods: We analyzed 330 specimens from patients with high-risk T1 CRC, which included 188 serum samples from 2 clinical cohorts-a training cohort (N = 46) and a validation cohort (N = 142)-and matched formalin-fixed paraffin-embedded samples (N = 142). We performed quantitative reverse-transcription polymerase chain reaction, followed by logistic regression analysis, to develop an integrated transcriptomic panel and establish a risk-stratification model combined with clinical risk factors.

Results: We used comprehensive expression profiling of a training cohort of LNM-positive and LMN-negative serum specimens to identify an optimized transcriptomic panel of 4 miRNAs (miR-181b, miR-193b, miR-195, and miR-411) and 5 mRNAs (AMT, forkhead box A1 [FOXA1], polymeric immunoglobulin receptor [PIGR], matrix metalloproteinase 1 [MMP1], and matrix metalloproteinase 9 [MMP9]), which robustly identified patients with LNM (area under the curve [AUC], 0.86; 95% confidence interval [CI], 0.72-0.94). We validated panel performance in an independent validation cohort (AUC, 0.82; 95% CI, 0.74-0.88). Our risk-stratification model was more accurate than the panel and an independent predictor for identification of LNM (AUC, 0.90; univariate: odds ratio [OR], 37.17; 95% CI, 4.48-308.35; P < .001; multivariate: OR, 17.28; 95% CI, 1.82-164.07; P = .013). The model limited potential overtreatment to only 18% of all patients, which is dramatically superior to pathologic features that are currently used (92%).

Conclusions: A novel risk-stratification model for noninvasive identification of T1 CRC has the potential to avoid unnecessary operations for patients classified as high-risk by conventional risk-classification criteria.

Keywords: Detection Biomarker; Noninvasive Assay; Risk-Stratification Model; Transcriptomic Panel.

Figure 1.

Training phase of a transcriptomic panel for the identification of LNM in patients with T1 CRC. A) Overview of the study. B) A ROC curve for a 4-miRNA and 5-mRNA panel in serum from training cohort patients (LNP = 5, LNN = 41; AUC: 4-miRNA panel = 0.78, 5-mRNA panel = 0.77, combination panel = 0.86). C) Risk score distribution plot in training cohort patients. Modified risk scores were obtained from individual risk scores by using Youden’s index values from the risk model. D) Forest plots with ORs for each panel risk score status in univariate logistic regression analysis in training cohort patients (ORs: 4-miRNA panel = 8.62, 5-mRNA panel = 8.44, combination panel = 14.22).

Figure 2.

Validation phase of the transcriptomic panel for the identification of LNM in patients with T1 CRC. A) A ROC curve for the transcriptomic panel in tissue specimens from validation cohort patients (LNP = 12, LNN = 130, AUC = 0.83). B) A ROC curve for the transcriptomic panel in matched serum samples in validation cohort patients (LNP = 12, LNN = 130, AUC = 0.82). C) Risk score distribution plot in serum specimens from validation cohort patients. D) A nomogram illustrating the probability of LNM risk. For clinical purposes, the scores of each covariate are added, and the total score is depicted on the total score point axis.

Figure 3.

Clinical validation of the risk-stratification model in patients with T1 CRC. A) The risk-stratification model, which combines the transcriptomic panel and pathological risk factors, outperformed detection accuracy of the transcriptomic panel or risk factors alone in serum specimens from validation cohort patients (AUC = 0.90). B–C) Forest plot with ORs of clinicopathological variables, transcriptomic panel, and risk-stratification model in univariate (B) and multivariate (C) logistic regression analysis in validation cohort patients. D) Currently used pathological factors led to the overtreatment of 92% patients with T1 CRC (left panel). The patients in validation cohort using our transcriptomic classifier divided into high (Yellow) and low (Light blue) risk by Youden’s index. Pie chart shows LNM status of LNP (Orange) and LNN (Dark blue). The transcriptomic panel would have led to the overtreatment of only 22% patients with T1 CRC (middle panel), and the risk-stratification model would have led to the overtreatment of only 18% patients with T1 CRC (right panel).