Serum microRNAs as biomarkers for recurrence in melanoma

Abstract

Background: Identification of melanoma patients at high risk for recurrence and monitoring for recurrence are critical for informed management decisions. We hypothesized that serum microRNAs (miRNAs) could provide prognostic information at the time of diagnosis unaccounted for by the current staging system and could be useful in detecting recurrence after resection.

Methods: We screened 355 miRNAs in sera from 80 melanoma patients at primary diagnosis (discovery cohort) using a unique quantitative reverse transcription-PCR (qRT-PCR) panel. Cox proportional hazard models and Kaplan-Meier recurrence-free survival (RFS) curves were used to identify a miRNA signature with prognostic potential adjusting for stage. We then tested the miRNA signature in an independent cohort of 50 primary melanoma patients (validation cohort). Logistic regression analysis was performed to determine if the miRNA signature can determine risk of recurrence in both cohorts. Selected miRNAs were measured longitudinally in subsets of patients pre-/post-operatively and pre-/post-recurrence.

Results: A signature of 5 miRNAs successfully classified melanoma patients into high and low recurrence risk groups with significant separation of RFS in both discovery and validation cohorts (p = 0.0036, p = 0.0093, respectively). Significant separation of RFS was maintained when a logistic model containing the same signature set was used to predict recurrence risk in both discovery and validation cohorts (p < 0.0001, p = 0.033, respectively). Longitudinal expression of 4 miRNAs in a subset of patients was dynamic, suggesting miRNAs can be associated with tumor burden.

Conclusion: Our data demonstrate that serum miRNAs can improve accuracy in identifying primary melanoma patients with high recurrence risk and in monitoring melanoma tumor burden over time.

Figure 1

Kaplan-Meier analysis for RFS by recurrence risks defined by the Cox proportional hazards model. Patients defined as high recurrence risk (dashed line) by the Cox proportional hazards model demonstrated significantly reduced RFS compared to patients defined as low recurrence risk (solid line) in both the (A) discovery (p = 0.0036 by log rank test) and (B) validation (p = 0.009 by log rank test) cohorts

Figure 2

Kaplan-Meier analysis for RFS by recurrence risks defined by logistic regression risk model. Patients defined as high recurrence risk (dashed line) by the logistic regression model demonstrated significantly reduced RFS compared to patients defined as low recurrence risk (solid line) in both the (A) discovery (p < 0.0001 by log rank test) and (B) validation (p = 0.033 by log rank test) cohorts

Figure 3

Proof-of-principle logistic regression subgroup analysis of stage II patients. (A) ROC curve for the miRNA containing logistic risk model defined for stage II patients had good classification performance (AUC = 0.89). (B) Kaplan-Meier analysis for RFS of high and low recurrence risk groups in stage II patients showed significant separation of RFS curves (p < 0.001 by log rank test)

Figure 4

Longitudinal evaluation of miRNA expression in pre- and post-recurrence serum samples. Difference between miRNA expression levels of miR-103 and miR-221 at primary diagnosis and at recurrence was statistically significant (p = 0.012 and p = 0.026, respectively). The horizontal axis represents time of blood draw for 17 patients. The vertical axis represents –(Ct) value