Fragment size and level of cell-free DNA provide prognostic information in patients with advanced pancreatic cancer

Between September 2012 and July 2017, we included 61 patients with locally advanced (n = 6) or metastatic (n = 55) pancreatic cancer that were admitted to Stavanger University Hospital (SUH) and Haukeland University Hospital (HUH). Patients received first-line treatment with gemcitabine (SUH only), nab-paclitaxel plus gemcitabine (SUH only), or FOLFIRINOX (both centers). Peripheral venous blood samples (9 ml EDTA tubes) were drawn before initiation of chemotherapy (n = 61; denoted B1) and after the first cycle of chemotherapy (n = 39; denoted B2). Treatment response was defined with a standard disease evaluation of radiological images, based on the RECIST 1.1 criteria [21]. We also included a control group (median age = 54), which comprised 28 healthy individuals with no prior or current cancer diagnosis. Blood samples collected at SUH were processed within 2 h; samples collected at HUH were processed at SUH within 24 h, after overnight transport. All patients and healthy controls provided written informed consent to participate in the study. The project was approved by the Regional Committee for Medical and Health Research Ethics (REK-Vest 2011/475, REK-Vest 2013/1743).

Blood samples were separated with density centrifugation using Lymphoprep™ (Axis Shield) density gradient media, according to the manufacturer’s instructions. We used 4 mL (1–2 mL for the first 8 patients) of plasma (diluted 1:1 with 0.9% NaCl for the density gradient separation protocol) for isolating total cfDNA with the QIAamp Circulating Nucleic Acid kit (Qiagen), as described by the manufacturer. cfDNA was eluted in 50 µL of Buffer AVE (Qiagen) and stored at − 80 °C until further analysis.

The cfDNA fragment size and cfDNA level were determined for each sample with an Agilent 2100 Bioanalyzer and the Agilent High Sensitivity DNA chip, according to the manufacturer’s instructions. The fragment size of cfDNA was determined, with the Agilent 2100 Bioanalyzer software, and defined as the mode of the main peak (corresponding to one nucleosome plus linker, derived from apoptotic cells) in the electropherogram (Additional file 1: Figure S1). cfdna level were also determined with the software, calculated as the area under the main peak in the electropherogram (Additional file 1: Figure S1). Determination of cfDNA fragment size and cfDNA levels were performed by individuals blinded to clinical data.

All statistical analyses were performed with IBM SPSS version 24.0 (http://​www.​spss.​com/​). All tests were two-sided, and p-values less than 0.05 were considered statistically significant. Missing data were automatically excluded from the analyses. Clinicopathological patient data were compared with the χ² or Fisher’s exact test, for categorical data, and with the independent-samples t-test or Mann–Whitney U test for continuous data. The Mann–Whitney U test was also used to compare cfDNA fragment size and cfDNA levels between control and patient samples, and between centers. The Wilcoxon signed rank test or the sign test were used to compare samples taken before and after initiation of chemotherapy. Correlation tests were performed between cfDNA fragment size and cfDNA levels with the non-parametric Spearman’s rank correlation coefficient for the continuous variables, and Cohen’s Kappa coefficient for the categorical variables. Data from the two study centers were compared statistically (Additional file 1: Table S1 and Figure S2). We found significant differences between centers in first-line treatment and cfDNA fragment size.

Survival analyses were performed with Kaplan–Meier estimates, the log-rank test, and Cox proportional hazards regression. The primary endpoints were progression-free survival (PFS) and overall survival (OS). PFS was defined as the time between inclusion (or sampling date for the B2 sample) and progression or death due to any cause, when the patient died before evidence of progression was obtained. OS was defined as the elapsed time between inclusion (or sampling date for the B2 sample) and death due to any cause.

Univariable Cox regression analyses were used to investigate the effects of single variables on survival. We tested variables measured before initiation of chemotherapy, including the B1 cfDNA fragment size (both continuous and dichotomized), B1 cfDNA levels (both continuous and dichotomized), the combination of B1 cfDNA fragment size and B1cfDNA levels, carbohydrate antigen 19-9 (CA 19-9) status, study center, age, sex, tumor size, tumor location, T-stage, clinical stage, metastatic location, N-stage, Eastern Cooperative Oncology Group (ECOG) performance status, first line treatment, second line treatment, and prior anti-cancer surgery. We also tested selected variables 1–2 months after the start of chemotherapy: the B2 cfDNA fragment size (both continuous and dichotomized) and B2 cfDNA levels (both continuous and dichotomized).

Multivariable Cox regression modeling was also performed for selected variables, including the B1 cfDNA fragment size (dichotomized), B1 cfDNA levels (dichotomized), CA 19-9 status, study center, age, sex, tumor size, tumor location, clinical stage, ECOG performance status, first line treatment, and prior anti-cancer surgery. Several variables were excluded from the multiple regression model, due to associations with other markers, including the B1 cfDNA fragment size (continuous), B1 cfDNA level (continuous), the combination of B1 cfDNA fragment size and B1 cfDNA levels. Other variables were excluded due to missing data (T-stage, metastatic location, N-stage), or because they were not acquired at baseline (B2 cfDNA fragment size, B2 cfDNA levels, and second-line treatment). The multivariable analyses were performed with both forward and backward stepwise selection of covariates.

The 61 patients included in this study had either locally advanced (n = 6) or metastatic tumors (n = 55). The cohort included a slight preponderance of men (n = 35, 57.4%), and the mean age was 64 years at diagnosis. Primary tumors were predominately located in the head of the pancreas (55.7%), and the median size was 37 mm. The first six (9.8%) patients included in the study received Gemcitabine monotherapy; the remaining patients received either FOLFIRINOX (n = 30, 49.2%) or nab-paclitaxel plus Gemcitabine (n = 25, 41%). All baseline patient characteristics and clinicopathological data are summarized in Table 1.

cfDNA fragment size and cfDNA levels were measured in plasma samples from the patients and from partially age-matched healthy controls (n = 28). The cfDNA fragment size in healthy control samples were longer (median 176.5 bp, range 168–185 bp) than the fragment size in patient samples (locally advanced: median 170 bp, range 167–173 bp, p = 0.001; metastatic: median 167 bp, range 148–180 bp, p < 0.001; Figs. 1 and 2). Moreover, the levels of cfDNA in healthy control samples (median 0.33 ng/mL plasma, range 0.03–1.61 ng/mL plasma) were significantly lower than those in patient samples (locally advanced: median 3.26 ng/mL plasma, range 1.16–7.98 ng/mL plasma, p < 0.001; metastatic: median 6.58 ng/mL plasma, range 0.53–1911.63 ng/mL plasma, p < 0.001; Figs. 1 and 2). When samples from patients with locally advanced tumors were compared to samples from patients with metastatic tumors, we found significant differences in cfDNA fragment size (p = 0.044; Fig. 2a), but not in cfDNA levels (p = 0.257; Fig. 2b).

cfDNA fragment size and cfDNA levels were determined in plasma samples obtained before the initiation of chemotherapy (n = 61 patients), and in plasma samples obtained after the first cycle of chemotherapy (n = 39 patients). Neither the median cfDNA fragment (p = 0.956) nor the median cfDNA level (p = 0.110) changed significantly during treatment (Additional file 1: Figure S3).

Patients were followed for a median of 7.7 months (range 0.3–25.8 months). During follow-up, progression was observed in 54 (88.5%) patients, and 51 (83.6%) patients died. We dichotomized cfDNA fragment size and cfDNA levels in patient samples based on the median values of all patient samples; thus a short fragment size was defined as ≤ 167 bp, and a high cfDNA level was defined as > 4.66 ng/mL plasma. We found that a short pre-treatment cfDNA fragment size (≤ 167 bp) was highly associated with both shorter PFS (4.0 vs 7.7 months; log-rank p = 0.003; Fig. 3a) and shorter OS (4.6 vs 10.5 months; log-rank p = 0.001; Fig. 3b). Analyses of in-treatment cfDNA fragment size demonstrated that short fragment size were again associated with shorter PFS (3.1 vs 6.8 months; log-rank p = 0.049; Additional file 1: Figure S4A), but not with shorter OS (7.0 vs 7.0 months; log-rank p = 0.634; Additional file 1: Figure S4B). Similarly, patients with high pre-treatment cfDNA levels had both shorter PFS (3.3 vs 7.7 months; log-rank p < 0.001; Fig. 3c) and shorter OS (5.4 vs 8.8 months; log-rank p = 0.002; Fig. 3d) compared to patients with low cfDNA levels. In samples obtained after initiation of chemotherapy, high cfDNA levels were neither associated with shorter PFS (4.5 vs 6.9 months; log-rank p = 0.185; Additional file 1: Figure S4C), nor shorter OS (7.0 vs 7.4 months; log-rank p = 0.368; Additional file 1: Figure S4D).

We found a moderately negative correlation between pre-treatment cfDNA fragment size and cfDNA levels (spearman correlation − 0.608, p < 0.001). Short fragment size (≤ 167 bp) and high cfDNA levels (> 4.66 ng/mL plasma) were also concordant in 45/61 (74%; Kappa = 0.475; 95% CI 0.253–0.696) patients. To see if it provided additional prognostic value, we performed survival analysis on the combination of the pre-treatment cfDNA fragment size and cfDNA levels. The patients were grouped according to whether they had neither short cfDNA fragment size nor high cfDNA levels (negative for both markers), either short cfDNA fragment size or high cfDNA levels (positive for 1 marker), or both (positive for two markers). The survival analyses demonstrated that both PFS (8.1 vs 4.6 vs 2.6 months; log-rank p < 0.001; Fig. 3e) and OS (10.6 vs 7.9 vs 3.8 months; log-rank p = 0.001; Fig. 3f) were significantly prolonged in patients negative for both markers, compared to patients that were positive for one or two markers.

Both univariable and multivariable Cox proportional hazards regression analyses were performed to estimate the prognostic impact of cfDNA fragment size and cfDNA levels on survival, relative to other clinicopatological parameters. The univariable regression analyses (Table 2) confirmed the prognostic impact of cfDNA fragment size and cfDNA levels, as demonstrated by the Log-Rank test. In addition, the univariable analysis showed that radiologically determined tumor size, ECOG performance status, first-line treatment, second-line treatment and study center also had prognostic value. The multivariable Cox regression analyses (Table 3) demonstrated that the dichotomized cfDNA level was an independent prognostic factor for both PFS (HR = 3.049, p = 0.005) and OS (HR = 2.236, p = 0.028). In addition, CA 19-9 was an independent prognostic factor for OS. Finally, the ECOG performance status and first-line treatment were independent prognostic factors for both PFS and OS. No other covariates were identified as significant predictors of survival in the multivariable model.