Promoter hypermethylation of SFRP1 is an allele fraction-dependent prognostic biomarker in metastatic pancreatic ductal adenocarcinoma

Abstract

Introduction: Metastatic pancreatic ductal adenocarcinoma (PDAC) is highly lethal. Promoter hypermethylation of SFRP1 (phSFRP1) in cell-free DNA is an established prognostic biomarker in PDAC. We used digital droplet PCR (ddPCR) to examine whether the prognostic impact of phSFRP1 was allele fraction (AF) dependent.

Methods: Prospectively collected plasma samples were analyzed blinded. Dual-strand methylation ddPCR assays were designed for SFRP1, with single-strand assay for the reference gene EPHA3. Patients were stratified into unmethylated SFRP1 (umSFRP1), low phSFRP1 AF (phSFRP1^low), and high phSFRP1 AF (phSFRP1^high). Survival was assessed with Kaplan-Meier curves. The 3-, 6-, and 12-month absolute risk difference (ARD) was calculated, and performance assessed with ROC analyses.

Results: Overall, 354 patients were included. Patients with umSFRP1 (n=137) had a mOS of 9.1 months compared to 7.2 months in phSFRP1^low (n=78) and 3.4 months in phSFRP1^high (n=143, P<0.01). phSFRP1^high was associated with increased mortality at 3 (ARD 26%, 95%CI: 15, 37), 6 (ARD 37%, 95%CI: 26, 48), and 12 months (ARD 23%, 95%CI: 14, 33). phSFRP1^low was associated with increased mortality at 12 months (ARD 13%, 95%CI: 2, 25) but not at 3 (ARD -3%, 95%CI: -13, 8) or 6 months (ARD 3%, 95%CI: -10, 17). phSFRP1 significantly improved performance in predicting mortality compared to only clinical variables (AUC: 0.70-0.71 vs. 0.54-0.57).

Discussion: Patients with phSFRP1^high had significantly shorter survival than phSFRP1^low or umSFRP1, indicating AF-dependent prognostic effects. phSFRP1^low had a worse prognosis than umSFRP1 at only 12 months, indicating dynamic changes. This could help personalize the treatment of PDAC.

Keywords: Sfrp1; biomarker; epigenetics; liquid biopsy; methylation; pancreatic ductal adenocarcinoma.

Figure 1

Kaplan–Meier survival distributions of patients with stage IV PDAC. Patients were stratified according to SFRP1 promoter methylation status and initial treatment with either best supportive care or chemotherapy. The risk table shows the number of patients at risk in 3-month intervals. (A) Dichotomized analysis of methylation status. umSFRP1, no detectable phSFRP1; phSFRP1, any detectable phSFRP1 AF. (B) SFRP1 methylation status grouped by phSFRP1 AF. umSFRP1, no detectable phSFRP1; phSFRP1^low, phSFRP1 AF < 0.53%; phSFRP1^high, phSFRP1 AF > 0.53%.

Figure 2

Univariable and multivariable ARD for patients with stage IV PDAC. A positive number indicates an increased risk of death, and a negative number indicates a lower risk of death. Analyses were performed for 3, 6, and 12 months according to SFRP1 promoter methylation status and covariates. Cases with missing data were excluded from the analysis. phSFRP1^low, phSFRP1 AF below 0.53%; phSFRP1^high, phSFRP1 AF above 0.53%; Gem + cap, gemcitabine and capecitabine; Gem + nab, gemcitabine and nab-paclitaxel; Gem + nab + toci, gemcitabine, nab-paclitaxel, and tocilizumab. The references were umSFRP1, ECOG PS ≤ 1, age ≤ 65, male sex, CA19-9 ≤ 860, and treatment with gemcitabine monotherapy.

Figure 3

The predictive performance of the models for 3-, 6-, and 12-month mortality, validated by 10-fold cross-validation. The 4 models included the following variables: 1) Only the clinical variables age > 65, sex, and ECOG PS > 1. 2) Clinical variables as well as CA19-9 > 860. 3) Clinical variables as well as AF-dependent SFRP1 methylation status (umSFRP1, phSFRP1^low or phSFRP1^high). 4) Clinical variables as well as CA19-9 > 860 and AF-dependent SFRP1 methylation status. Cases with missing data were excluded from the analysis. (A) Performance of models at 3 months. (B) Performance of models at 6 months. (C) Performance of models at 12 months.

Figure 4

Comparative predictive performance of dichotomous versus allele fraction-dependent SFRP1 methylation analysis. The predictive models included the variables age > 65, sex, ECOG PS > 1, CA19-9 > 860 and either dichotomous SFRP1 methylation status (umSFRP1 or phSFRP1) or allele fraction-dependent SFRP1 methylation analysis (umSFRP1, phSFRP1^low, or phSFRP1^high). Performance was assessed at 3-, 6-, and 12 months, and models were computed using 10-fold cross-validation. (A) Performance of models at 3 months. (B) Performance of models at 6 months. (C) Performance of models at 12 months.