Optimization of Sources of Circulating Cell-Free DNA Variability for Downstream Molecular Analysis

Abstract

Circulating cell-free DNA (ccfDNA) is used increasingly as a cancer biomarker for prognostication, as a correlate for tumor volume, or as input for downstream molecular analysis. Determining optimal blood processing and ccfDNA quantification are crucial for ccfDNA to serve as an accurate biomarker as it moves into the clinical realm. Whole blood was collected from 50 subjects, processed to plasma, and used immediately or frozen at −80°C. Plasma ccfDNA was extracted and concentration was assessed by real-time quantitative PCR (qPCR), fluorimetry, and droplet digital PCR (ddPCR). For the 24 plasma samples from metastatic pancreatic cancer patients, the variant allele fractions (VAF) of KRAS G12/13 pathogenic variants in circulating tumor DNA (ctDNA) were measured by ddPCR. Using a high-speed (16,000 × g) or slower-speed (4100 × g) second centrifugation step showed no difference in ccfDNA yield or ctDNA VAF. A two- versus three-spin centrifugation protocol also showed no difference in ccfDNA yield or ctDNA VAF. A higher yield was observed from fresh versus frozen plasma by qPCR and fluorimetry, whereas a higher yield was observed for frozen versus fresh plasma by ddPCR, however, no difference was observed in ctDNA VAF. Overall, our findings suggest factors to consider when implementing a ccfDNA extraction and quantification workflow in a research or clinical setting.

Cell-free DNA refers to DNA present in various body fluids, with circulating cell-free DNA (ccfDNA) referring specifically to cell-free DNA present in the bloodstream. Increased ccfDNA levels have been associated with a wide range of physiologic and pathophysiologic conditions including physical exercise, acute and chronic inflammatory disease, trauma, cardiovascular disease, and allograft rejection.¹ In cancer research, ccfDNA has been found to correlate with tumor burden, and tumor-derived ccfDNA, or circulating tumor DNA, has been recognized increasingly as a promising noninvasive biomarker applicable to cancer diagnosis, prognosis, and monitoring.^2,3

Many potential sources of variation exist between blood collection and final ccfDNA analysis; blood collection tube (BCT) and ccfDNA extraction have been reviewed extensively.4, 5, 6 Here, the focus was on the variability of ccfDNA quantification, plasma processing, and sample storage. Minimizing variability is essential for accurate quantification and molecular characterization of ccfDNA. Preservative BCTs allow for storage of whole blood for up to 7 days without significant white blood cell (WBC) lysis and shedding of genomic DNA, which might otherwise confound ccfDNA quantification. Conversely, nonpreservative BCTs (eg, EDTA) generally show significant WBC lysis after 1 day of storage, leading to higher amounts of contaminating WBC DNA compared with blood processed immediately.⁷ Once blood is collected in a BCT, it is processed to plasma in a series of centrifugation steps designed, with each successive step, to further reduce the contamination of plasma by cells. Although some studies have shown no difference between one- and two-step centrifugation protocols for plasma preparation,^8,9 most studies have shown a decrease in yield with a second centrifugation step as a result of decreased WBC contamination.10, 11, 12 Some laboratories use a high-speed (14,000 to 16,000 × g) second spin,¹³ which may not be feasible in all clinical laboratory settings. The high-speed centrifuges required for these spins (which accommodate common plasma volumes) may not be available in many clinical laboratories. A suboptimal alternative is the use of microcentrifuges that can reach these speeds but require plasma processing to be performed in multiple small batches. To our knowledge, no study has rigorously investigated the effect of second spin speed or whether the addition of a third centrifugation step would decrease WBC contamination further. Once plasma is obtained, most research laboratories bank samples at −80°C to allow for batched ccfDNA extraction; however, this practice may be infeasible in a clinical workflow requiring rapid assay turnaround times. Studies comparing the ccfDNA yield from fresh versus frozen plasma also have been inconsistent.14, 15, 16, 17

Common methods for quantification of ccfDNA concentration in plasma include fluorescence-based (ie, Qubit, Thermo Fisher Scientific, Waltham, MA) and PCR-based, including quantitative real-time PCR (qPCR) using a standard curve or absolute quantification by digital PCR, usually droplet digital PCR (ddPCR). Quantification method correlation was assessed in earlier studies with mixed results. Good correlation has been shown between qPCR and ddPCR,^18,19 Qubit and qPCR,²⁰ and Qubit and ddPCR,¹⁰ whereas other studies observed poor correlation between Qubit and qPCR.^21,22 Notably, each of these studies only compared two methods and used qPCR assays that targeted single-genome copy elements, unlike the higher-sensitivity²³ repeat-element qPCR techniques commonly used today. In addition, some studies used ddPCR detection of circulating tumor DNA as a sensitive read out of WBC contamination, noting a decreased variant allele fraction (VAF) with delayed processing of nonpreservative BCTs.24, 25, 26

Here, three ccfDNA concentration quantification methods and ddPCR quantification of VAF were used to address three important questions for the translation of research studies into clinical laboratory tests, as follows: i) the necessity of a high-speed (16,000 × g) versus lower-speed (4100 × g) second spin, ii) the potential benefit of a third spin added to a two-spin plasma processing protocol, and iii) the effect of frozen versus fresh plasma. Finally, the intrasample variability and correlation of the three ccfDNA quantification methods were compared. These factors were chosen while leaving discussions of extraction kit and BCT choice to the already expansive literature on these topics.

Materials and Methods

Specimen Collection

Whole blood was collected from patients at the Hospital of the University of Pennsylvania with written informed consent. Blood was obtained in Streck cell-free DNA BCTs (Omaha, NE) from 26 healthy subjects and 24 subjects with metastatic pancreatic ductal adenocarcinoma (mPDAC). Different subsets of specimens were used in each experiment depending on timing and specimen availability (Supplemental Table S1).

Whole Blood Processing and Plasma Banking

A workflow schematic shown in Figure 1 summarizes the blood processing steps. The initial centrifugation step (first-spin) for all specimens was 1600 × g for 10 minutes. For the second spin speed comparison, plasma underwent a second centrifugation step at either 16,000 × g or 4100 × g for 15 minutes and was banked at −80°C. For the experiment investigating the addition of a third centrifugation step, the plasma from the first (1600 × g) spin underwent an additional one or two centrifugation steps (two- or three-spin protocol, respectively) at 4100 × g for 15 minutes and was banked at −80°C. For the experiment investigating fresh versus frozen plasma, the plasma from the first spin underwent a second centrifugation at 4100 × g and was banked at −80°C or used immediately. Specimens were transported, stored, and processed at room temperature and all centrifugation steps were performed with brake on. Frozen plasma was thawed at room temperature for 30 to 60 minutes before ccfDNA extraction.

Figure 1.

Schematic of circulating cell-free DNA (ccfDNA) workflow. Displayed are the main elements of a workflow for processing of whole blood into plasma, followed by ccfDNA extraction, quantification, and variant detection. The double-headed arrows indicate the conditions compared for each element and figure. In brief, ccfDNA yields and variant detection were compared for the following: lower-speed (4100 × g) versus high-speed (16,000 × g) second spins (Figure 2), two-spin versus three-spin protocols (Figure 3), and fresh versus frozen plasma (Figure 4). Variation in real-time quantitative PCR (qPCR)-, Qubit-, and droplet digital PCR (ddPCR)-based quantification methods are compared in Figure 5.

ccfDNA Extraction from Plasma

Extractions were performed using the QIAGEN (Hilden, Germany) QIAsymphony DSP Circulating DNA Kit according to the manufacturer's instructions on a QIAsymphony SP. Extractions were performed using 2 to 4 mL plasma (depending on specimen availability), with a final elution volume of 60 μL.

ccfDNA Quantification by qPCR

The concentration of extracted ccfDNA was quantified by qPCR for a 115-bp amplicon of the Alu repetitive element.²³ Briefly, qPCR was performed on extracted ccfDNA, in quadruplicate reactions, using Power SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA) according to the manufacturer's instructions on a ViiA 7 Real-Time PCR System (Applied Biosystems). Results were normalized to a standard curve of reference DNA (Promega, Madison, WI) using QuantStudio Real-Time PCR Software version 1.3 (Applied Biosystems) and reported as nanograms of ccfDNA per milliliter of plasma.

ccfDNA Quantification by ddPCR

The concentration of extracted ccfDNA was quantified by ddPCR for an amplicon in the RRP30 locus, which is used frequently as a reference locus for copy number variation assays. Briefly, ddPCR was performed on extracted ccfDNA, without pre-amplification, using the ddPCR copy number assay according to the manufacturer's instructions on the QX200 AutoDG Droplet Digital PCR System (Bio-Rad, Hercules, CA). Results were analyzed using QuantaSoft Version 1.7.4.0917 (Bio-Rad). Results are reported as copies per milliliter of plasma.

ccfDNA Quantification by Qubit Fluorometry

Briefly, the Qubit dsDNA HS Assay Kit (Invitrogen, Carlsbad, CA) was used to quantify the ccfDNA concentration of the extracted samples according to the manufacturer's instructions on a Qubit 4 Fluorimeter (Invitrogen). Results are reported as nanograms of ccfDNA per milliliter of plasma.

KRAS G12/13 Variant Detection and VAF Quantification by Pre-Amplification ddPCR

The KRAS G12/13 locus was interrogated for VAF as a sensitive measure of WBC DNA contamination. These patients and this locus were chosen because approximately 85% of mPDAC patients' tumor tissues harbor pathogenic variants detected by this assay at this locus.²⁷ First, the KRAS G12/13 locus was pre-amplified as previously described,²⁸ and then analyzed for variants G12A/C/D/R/S/V and G13D using the ddPCR KRAS G12/G13 Screening Assay Kit (Bio-Rad) on either the QX200 Droplet Digital PCR System (Bio-Rad) or the QX200 AutoDG Droplet Digital PCR System (Bio-Rad) according to the manufacturer's instructions. Results were analyzed using QuantaSoft Version 1.7.4.0917 (Bio-Rad). Variant copies detected at greater than three SDs above the average false-positive rate (data not shown, determined based on the results from healthy subjects 1 to 24) (Supplemental Table S1) and also a VAF higher than a 0.04% threshold (the VAF of three SDs above the average false-positive rate of the average total copies for the same 24 healthy controls) were considered positive. VAF was calculated as variant copies minus the average false-positive rate divided by the total number of positive copies (variant and wild-type) minus the average false-positive rate.

Statistical Analysis

To examine the differences in the ccfDNA concentration or VAF between the different processing methods, the ratio of the yields or VAFs were computed, respectively, for each pair of values and tested against the null value of 1 (ie, no difference) using the Wilcoxon matched-pairs signed rank test, a nonparametric equivalent of the paired t-test. In addition, agreement of yield values from different processing methods was assessed using Lin concordance correlation coefficient (CCC) and the associated P value.²⁹ All correlations were measured with the nonparametric Spearman rank correlation coefficient, rho. Coefficients of variation were computed for three replicate quantifications by each method for each specimen, and the Friedman test (nonparametric equivalent of analysis of variance) was used to determine differences. The correlation of these values was assessed using the nonparametric Spearman rank correlation coefficient, rho, because differences in scale between ddPCR and qPCR/Qubit did not allow for assessment of agreement using Lin CCC for all comparisons. Analyses were performed in Prism version 9.0.2 (GraphPad, San Diego, CA) and Stata/IC 16.1 (StataCorp LLC, College Station, TX).

Results

Comparison of High-Speed with a Lower-Speed Second Spin

A high-speed second centrifugation step (16,000 × g) was compared with a lower-speed (4100 × g) second centrifugation step within a standard two-spin plasma processing protocol for 24 plasma samples, including healthy control specimens (n = 12) and disease specimens (mPDAC patients, n = 12). The ccfDNA concentration was not affected significantly by second-spin speed because ccfDNA concentration ratios for the high- and lower-speed second spins (Figure 2A) did not differ significantly from 1.0 by any of the three quantification methods (qPCR P = 0.208, Qubit P = 0.468, and ddPCR P = 0.509, Wilcoxon signed rank test). Furthermore, high-speed and lower-speed ccfDNA concentrations agreed (Supplemental Figure S1A) (CCC = 0.981, P < 0.001 for qPCR; CCC = 0.982, P < 0.001 for Qubit; and CCC = 0.876, P < 0.001 for ddPCR). Finally, investigation of the KRAS G12/13 locus for variants by pre-amplification ddPCR in the 12 mPDAC specimens showed that the VAF ratio did not differ significantly from 1.0 when a variant was detected in both samples (n = 6, P = 0.438) (Figure 2B) and the values agreed (CCC = 0.973, P < 0.001) (Supplemental Figure S1B). Two samples had very low VAFs (<0.1%) detected in the lower-speed specimen, but not in the high-speed specimen. In addition, four specimens had no KRAS G12/13 variants detected.

Figure 2.

A comparison of circulating cell-free DNA (ccfDNA) yields and variant detection for high-speed (16,000 × g) and lower speed (4112 × g) second spins. Whole-blood samples from 24 individuals [healthy subjects 1 to 12 and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients 1 to 12] were subjected to a first spin at 1600 × g and the plasma supernatant was isolated and divided into aliquots for further processing. Plasma then was subjected to either a high-speed (16,000 × g) or lower-speed (4100 × g) second spin. To assess the effect of the second spin speed on ccfDNA yield, ccfDNA was extracted, divided into aliquots, and then quantified by real-time quantitative PCR (qPCR), Qubit, and droplet digital PCR (ddPCR). A: High-speed and lower-speed second spin yield ratios were not significantly different from 1.0 (qPCR P = 0.208, Qubit P = 0.468, ddPCR P = 0.509). Bars show geometric means and geometric SDs of the yield ratios, dotted line indicates ratio of 1.0 (open circles, healthy controls; filled circles, PDAC patients). To assess the effect of the second spin speed on KRAS G12/13 variant allele fraction (VAF) in ccfDNA, ddPCR was performed. B: High-speed and lower-speed second spin KRAS G12/13 VAF ratios were not significantly different from 1.0, when a variant was detected in both samples (P = 0.438). Bars show geometric means and geometric SDs of the VAF ratios, dotted line indicates ratio of 1.0. Two samples had the KRAS G12/13 variants detected at very low VAFs (<0.1%) in the lower-speed specimen but not in the high-speed specimen. Four samples did not have a KRAS G12/13 variant detected in either specimen. n = 12 healthy controls (A); n = 12 PDAC patients (A); n = 6 PDAC patients (B).

Comparison of a Two-Spin Plasma Preparation Protocol with a Protocol with an Additional Third Spin

Next, the addition of a third (4100 × g) centrifugation step (three-spin) to a two-spin (1600 × g first spin and 4100 × g second spin) plasma processing protocol was assessed to determine whether it reduced ccfDNA concentration by removing contaminating cellular components. These spin steps were performed with a standard blood processing centrifuge that, unlike high-speed centrifuges, is available in most laboratories. Different cohorts of healthy subject specimens (n = 12) and disease specimens (mPDAC patients, n = 12) were tested than in the previous analysis (Supplemental Figure S1). The ccfDNA concentration was not affected significantly by spin protocol because the concentration ratios after the two-spin or three-spin plasma preparation (Figure 3A) did not differ significantly from 1.0 (qPCR P = 0.317, Qubit P = 0.439, and ddPCR P = 0.160). Furthermore, concentrations after the two-spin or three-spin plasma preparation agreed (Supplemental Figure S2A) (CCC = 0.959, P < 0.001 for qPCR; CCC = 0.986, P < 0.001 for Qubit; and CCC = 0.996, P < 0.001 for ddPCR). Finally, the VAF ratio for the KRAS G12/13 locus for the 12 mPDAC specimens did not differ from 1.0 when a variant was detected in both samples (n = 9, P = 0.496) (Figure 3B) and the values agreed (CCC = 0.998, P < 0.001) (Supplemental Figure S2B). Two samples with very low VAF (<0.1%) had a variant detected in one specimen but not in the other (one in the three-spin specimen but not in the two-spin specimen, and one in the two-spin specimen but not in the three-spin specimen). One sample had no variants detected in either specimen.

Figure 3.

A comparison of circulating cell-free DNA (ccfDNA) yields and variant detection for a two- versus three-spin plasma preparation protocol. Whole-blood samples from 24 individuals [healthy subjects 13 to 24 and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients 13 to 24] were subjected to a first spin at 1600 × g and a lower speed (4100 × g) second spin (two-spin) or two successive lower-speed spins (three-spin). To assess the effect of the third spin on ccfDNA yield, ccfDNA was extracted, divided into aliquots, and then quantified by real-time quantitative PCR (qPCR), Qubit, and droplet digital PCR (ddPCR). A: Three-spin and two-spin yield ratios were not significantly different from 1.0 (qPCR P = 0.317, Qubit P = 0.439, and ddPCR P = 0.160). Bars show geometric means and geometric SDs of the yield ratios, dotted line indicates ratio of 1.0 (open circles, healthy controls; filled circles, PDAC patients). To assess the effect of the second spin speed on KRAS G12/13 variant allele fractions (VAFs) in ccfDNA, ddPCR was performed. B: Three-spin and two-spin KRAS G12/13 VAF ratios were not significantly different from 1.0, when a variant was detected in both samples (P = 0.496). Bars show geometric means and geometric SDs of the VAF ratios, dotted line indicates ratio of 1.0. Two samples with very low VAF (<0.1%) were detected in one specimen but not the other (one in the three-spin specimen but not the two-spin specimen, and one in the two-spin specimen but not the three-spin specimen). One specimen had no KRAS G12/13 variants detected. n = 12 healthy controls (A); n =12 PDAC patients (A); n = 9 PDAC patients (B).

Comparison of Yield from Freshly Prepared Plasma versus Frozen Plasma

Here, ccfDNA yield from freshly prepared plasma was compared with frozen plasma derived from the same subject's blood sample. Concentrations of ccfDNA from fresh plasma were significantly higher than frozen plasma by qPCR and Qubit as shown by the geometric mean of the ccfDNA concentration ratios of 1.134 (95% CI, 1.047 to 1.229) for qPCR (P = 0.001) and 1.206 (95% CI, 1.140 to 1.276) for Qubit (P < 0.001) (Figure 4A). In contrast, the concentration of ccfDNA from frozen plasma was significantly higher than from fresh plasma by ddPCR with a geometric mean of the ccfDNA concentration ratios of 0.933 (95% CI, 0.889 to 0.979, P = 0.010) (Figure 4A). Despite these differences, the ccfDNA concentration for fresh and frozen plasma agreed per Lin concordance correlation coefficient at 0.987, 0.931, and 0.990 for qPCR, Qubit, and ddPCR, respectively (all P < 0.001) (Supplemental Figure S3A). In addition, KRAS G12/13 VAF ratios did not differ from 1.0 when a variant was detected in both samples (n = 7, P = 0.219) (Figure 4B) and the values agreed (CCC = 0.997, P < 0.001) (Supplemental Figure S3B). Four samples with very low VAF (<0.1%) had a variant that was detected in one specimen but not in the other (three in the fresh specimen but not the frozen specimen, and one in the frozen specimen but not in the fresh specimen). One specimen had no variants detected in either specimen.

Figure 4.

A comparison of circulating cell-free DNA (ccfDNA) yields and variant detection from fresh versus frozen plasma. Whole-blood samples from 24 individuals [healthy subjects 13 to 24 and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients 13 to 24] were subjected to a first spin at 1600 × g and a lower-speed (4100 × g) second spin (two-spin). To assess the effect of the plasma banking conditions on ccfDNA yield, ccfDNA was extracted from matched fresh or frozen plasma, divided into aliquots, and then quantified by real-time quantitative PCR (qPCR), Qubit, and droplet digital PCR (ddPCR). A: By qPCR and Qubit, there was a significant increase in yield from fresh plasma compared with frozen because the concentration ratios were statistically significantly different from 1.0 (qPCR geometric mean of the ratios, 1.134; P = 0.001; Qubit geometric mean of the ratios, 1.206; P = < 0.001), whereas by ddPCR the fresh plasma yield was lower compared with frozen plasma (geometric mean of the ratios, 0.933; P = 0.010). Bars show geometric means and geometric SDs of the yield ratios, dotted line indicates ratio of 1.0 (open circles, healthy controls; filled circles, PDAC patients). To assess the effect of plasma banking on KRAS G12/13 variant allele fractions (VAFs) in ccfDNA, ddPCR was performed. B: Fresh versus frozen plasma KRAS G12/13 VAF ratios were not significantly different from 1.0, when a variant was detected in both samples (P = 0.219). Bars show geometric means and geometric SDs of the VAF ratios, dotted line indicates ratio of 1.0 . Four samples with very low VAF (<0.1%) were detected in one specimen but not the other (three in the fresh specimen but not the frozen specimen, and one in the frozen specimen but not the fresh specimen). One specimen had no KRAS G12/13 variants detected. n = 12 healthy controls (A); n =12 PDAC patients (A); n = 7 PDAC patients (B).

Variation in ccfDNA Quantification Methods

Finally, the variability and correlation of ccfDNA concentration measurements were evaluated. Three independent replicate measurements of ccfDNA concentration were performed on 24 QIAsymphony ccfDNA extracts (12 healthy controls and 12 mPDAC patients) for each of the three quantification methods. Quantification coefficients of variation did not differ significantly by method at 15.15% ± 5.54% (means ± SD) by qPCR, 13.79% ± 5.31% by Qubit, and 16.49% ± 8.00% by ddPCR (Figure 5A) (Friedman test P = 0.582). Furthermore, the methods were highly correlated (Figure 5, B–D) (qPCR versus Qubit, Spearman rho = 0.959; Qubit versus ddPCR, Spearman rho = 0.960; and ddPCR versus qPCR, Spearman rho = 0.957).

Figure 5.

A comparison of circulating cell-free DNA (ccfDNA) quantification variability and correlation. Whole-blood samples from 24 individuals [healthy subjects 3 to 12 and 25 to 26 and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients 1 to 12] were subjected to a first spin at 1600 × g and the plasma supernatant was isolated and then subjected to a lower-speed (4100 × g) second spin. To assess the variability in ccfDNA quantification measurement, ccfDNA was extracted, divided into aliquots, and then quantified by real-time quantitative PCR (qPCR), Qubit, and droplet digital PCR (ddPCR), three times for each measurement approach. A: No significant difference was observed in quantification variability by the three different methods, the coefficients of variation (CV) for the three replicate quantifications were not significantly different (Friedman test P = 0.582). Bars show means and SD (open circles, healthy controls; filled circles, PDAC patients). To assess the correlation of ccfDNA quantification measurements by the three methods, one of the replicate quantifications for each was compared with one of each of the others. B–D: All methods were highly correlated: qPCR versus Qubit, Spearman r = 0.959 (B); Qubit versus ddPCR, Spearman r = 0.960 (C); and ddPCR versus qPCR, Spearman r = 0.957 (D). Dotted line in B isrepresents the line of identity. n = 12 healthy controls (A); n = 12 PDAC patients (A).

Discussion

This study examined the impact of different blood processing and nucleic acid assessment platforms on the yield and accessibility of ccfDNA for downstream analyses in the molecular diagnostic laboratory. There was no reduction in plasma ccfDNA concentration with a high-speed (16,000 × g) centrifugation step or the addition of a third centrifugation step when compared with a two-step centrifugation protocol. These data suggest neither a high-speed step nor a third centrifugation further reduces WBC contamination and need not be included in blood processing protocols. There was also no significant difference in the VAF as determined by ddPCR-based measurement of known KRAS pathogenic variants in the plasma of a subset of patients with mPDAC. When considering fresh versus frozen plasma, different results were observed by quantification method, with increased yield from fresh plasma by qPCR and Qubit, but decreased yield by ddPCR. These data may explain the conflicting results observed in prior studies that have investigated the effect of plasma freeze/thaw.14, 15, 16, 17 This is an important consideration when maximizing yield is important or when comparing batched research yields with clinical yields. Finally, examination of ccfDNA quantification methodologies by their relative variability and correlation did not identify significant differences. The data show a similar degree of variability in ccfDNA quantification across methods (average coefficients of variation, approximately 15%), and all methods correlated well.

Many challenges to evaluating ccfDNA methodologies exist; variable concentrations between samples necessitate paired analyses, and, coupled with limited blood draw volumes, this makes testing multiple conditions difficult. Furthermore, low sample concentrations test the limits of quantification assays. Many studies are limited in their applicability because they do not use ratio paired statistical analyses and therefore do not account for sample-to-sample differences in ccfDNA concentration.^10,12 Here, replicate extractions and quantifications across multiple paired samples with ratio paired statistical testing were used to overcome these limitations. Other studies have used pooled or contrived samples to overcome some of these difficulties.⁹ Pooled samples were not used here because they fix analyses at one concentration rather than sampling across a dynamic range. Contrived samples were not used in this study because no available material exists to replicate endogenous ccfDNA accurately that is fragmented based on histone patterning and largely protein bound. To test these methods across a wider dynamic range both healthy donor and mPDAC samples were used. Furthermore, the use of mPDAC samples facilitated the analysis of circulating tumor DNA VAF within the ccfDNA.

This study evaluated three sources of potential variability that are highly relevant to implementation of ccfDNA testing in both a research and a clinical laboratory setting. First, time and complexity may be removed from blood processing protocols if a higher speed (16,000 × g) second spin can be replaced by a lower speed (4100 × g) second spin. This allows for plasma processing in a standard centrifuge, no longer necessitating the use of high-speed centrifuges/rotors or subaliquoting for processing in microcentrifuges. Second, omitting an unnecessary third centrifugation step saves time and effort. Third, most ccfDNA research studies batch their extractions and quantifications by banking plasma samples at −80°C, a step that likely would not occur in a clinical laboratory setting because samples typically are processed and extracted in short order. However, our data here are mixed; a single freeze-thaw cycle reduced the resultant ccfDNA concentration as measured by qPCR and Qubit, but increased it by ddPCR.

The ccfDNA yield differences for fresh versus frozen plasma may be a peculiarity of the ddPCR-based RRP30 locus copy number methodology used because it is a narrow and specific surrogate measure of genome copies compared with Qubit and qPCR-based measure of Alu elements. Further experiments would be necessary to explain this phenomenon because it may be the result of differences in detected fragment sizes between methodologies or robustness of methodologies to contaminating products of freeze/thaw (ie, denatured proteins). However, these data do mirror contradictory data in the literature. One study observed that the ccfDNA concentration increased with the freeze–thaw cycle number when the membrane-based QIAGEN QIAamp Circulating Nucleic Acid kit was used, but did not increase when the bead-based kits were used.¹⁴ Other studies showed no change in ccfDNA concentration,^15,16 and yet another showed a decreased ccfDNA concentration¹⁷ with increased freeze–thaw cycles. Many of these studies, however, were limited by small sample sizes, use of spiked DNA rather than clinically obtained blood samples, and unpaired analyses. Notably, the average differences observed all were relatively small (≤20%), on par with the error of the quantification methods themselves (approximately 15%), and agreement and correlation still were high. Therefore, this finding is unlikely to noticeably affect the quantification of any one sample, although it may be important in considering samples in aggregate because laboratories seek to implement ccfDNA-based tests and validate concentration cut-off points established in the literature (where freeze/thaw is common). Finally, the data showing similar variability and high correlation between ccfDNA quantification methods suggest that the increased time and effort associated with qPCR and ddPCR-based quantification methods may not be necessary if Qubit data are equivalent for a given application.

Future studies are needed to expand these results and may include investigations into the many and varied centrifugation speeds and conditions (such as centrifuge temperature and use of centrifuge brake) that exist in the literature and in real-world research and clinical laboratories. Furthermore, as ccfDNA quantification and downstream analyses are used in disease areas other than cancer, future studies should explore these factors in those specific patient cohorts. Overall, it is critical to investigate specific methodologic choices in terms of variability, especially in comparison with other established methods, when implementing ccfDNA analysis in the research or clinical setting.

Supplemental Data

Supplemental Figure S1.

Agreement between circulating cell-free DNA (ccfDNA) yields for high-speed versus lower-speed second-spin plasma preparation protocols. A: The plots show the agreement of the 4100 × g versus 16,000 × g second-spin yields for healthy controls (open circles) and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients (filled circles) by real-time quantitative PCR (qPCR) (left panel: CCC = 0.981, P < 0.001), by Qubit (middle panel: CCC = 0.982, P < 0.001), and droplet digital PCR (ddPCR) (right panel: CCC = 0.875, P < 0.001). B: The plot shows the agreement of the 4100 × g versus 16,000 × g second spin KRAS G12/13 variant allele fractions (VAFs) by pre-amplification ddPCR when a variant was detected in both samples for mPDAC patients (CCC = 0.973, P < 0.001). Dotted lines represent the line of identity. n = 12 healthy controls (A); n = 12 mPDAC patients (A); n = 6 mPDAC patients (B).

Supplemental Figure S2.

Agreement between circulating cell-free DNA (ccfDNA) yields for two-spin versus three-spin plasma preparation protocols. A: The plots show the agreement of the two-spin versus three-spin protocol yields for healthy controls (open circles) and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients (filled circles) by real-time quantitative PCR (qPCR) (left panel: CCC = 0.959, P < 0.001), by Qubit (middle panel: CCC = 0.986, P < 0.001), and droplet digital PCR (ddPCR) (right panel: CCC = 0.996, P < 0.001). B: The plot shows the agreement of the two-spin versus three-spin protocol KRAS G12/13 variant allele fractions (VAFs) by pre-amplification ddPCR when a variant was detected in both samples for mPDAC patients (CCC = 0.496, P < 0.001). Dotted lines represent the line of identity. n = 12 healthy controls and n = 12 mPDAC patients (A); n = 9 mPDAC patients (B).

Supplemental Figure S3.

Agreement between circulating cell-free DNA (ccfDNA) yields for fresh versus frozen plasma. A: The plots show agreement of the fresh versus the frozen plasma extraction yields for healthy controls (open circles) and metastatic pancreatic ductal adenocarcinoma (mPDAC) patients (filled circles) by real-time quantitative PCR (qPCR) (left panel: CCC = 0.987, P < 0.001), by Qubit (middle panel: CCC = 0.931, P < 0.001), and droplet digital PCR (ddPCR) (right panel: CCC = 0.990, P < 0.001). B: The plot shows the agreement of the fresh versus frozen plasma KRAS G12/13 variant allele fractions (VAFs) by pre-amplification ddPCR when a variant was detected in both samples for mPDAC patients (CCC = 0.997, P < 0.001). Dotted lines represent the line of identity. n = 12 healthy controls and n = 12 mPDAC patients (A); n = 7 mPDAC patients (B).

Supplemental Table

S1