Calibration of cell-free DNA measurements by next-generation sequencing

Abstract

Objectives

Accurate monitoring of disease burden depends on accurate disease marker quantification. Although next-generation sequencing (NGS) is a promising technology for noninvasive monitoring, plasma cell-free DNA levels are often reported in misleading units that are confounded by non–disease-related factors. We proposed a novel strategy for calibrating NGS assays using spiked normalizers to improve precision and to promote standardization and harmonization of analyte concentrations.

Methods

In this study, we refined our NGS protocol to calculate absolute analyte concentrations to (1) adjust for assay efficiency, as judged by recovery of spiked synthetic normalizer DNAs, and (2) calibrate NGS values against droplet digital polymerase chain reaction (ddPCR). As a model target, we chose the Epstein-Barr virus (EBV) genome. In patient (n = 12) and mock (n = 12) plasmas, NGS and 2 EBV ddPCR assays were used to report EBV load in copies per mL of plasma.

Results

Next-generation sequencing was equally sensitive to ddPCR, with improved linearity when NGS values were normalized for spiked DNA read counts (R² = 0.95 for normalized vs 0.91 for raw read concentrations). Linearity permitted NGS calibration to each ddPCR assay, achieving equivalent concentrations (copies/mL).

Conclusions

Our novel strategy for calibrating NGS assays suggests potential for a universal reference material to overcome biological and preanalytical variables hindering traditional NGS strategies for quantifying disease burden.

KEY POINTS.

• Although next-generation sequencing (NGS) of plasma DNA has great promise for noninvasive disease monitoring, measurements are influenced by preanalytical factors.

• Assays such as droplet digital polymerase chain reaction (ddPCR) are highly sensitive and truly quantitative but do not provide the same breadth or depth of data sequencing assays provide.

• By normalizing NGS measurements to spiked-in oligonucleotide recovery and calibrating to an orthogonal ddPCR assay, NGS results can be reported quantitatively (copies/mL) and with less preanalytical bias.

INTRODUCTION

Next-generation sequencing (NGS) has revolutionized molecular diagnostics by broadening the spectrum of analytes that can be characterized in a single assay. Circulating cell-free DNA (cfDNA) panels hold great promise for noninvasive assessment of disease status and for serial monitoring of disease burden. Current methods of cfDNA quantification, however, are reported in units of measure that are biased by unpredictable levels of non–disease-related factors (eg, tumor markers reported as “allele fraction” or pathogen loads reported as “per million human reads”). Irrelevant DNA in the denominator of such fractions is biased by natural and in vitro preanalytical factors unrelated to disease burden. Cell-free DNA is naturally present in all healthy individuals, and imprecision in biomarker measurement is exacerbated when infection or inflammation artificially depresses fractional calculations by increasing the DNA released from healthy cells, particularly cells with a short half-life, such as leukocytes.^1,2 Other sources of interference include physical exertion and treatment-induced healthy cell turnover, as with radiation therapy or chemotherapy.³ In vitro contributors include leukocyte lysis during specimen processing and shear stress during blood collection not amenable to amelioration by Streck cfDNA blood collection tubes or similar “preservative” strategies ex vivo. In summary, a wide range of non–disease-related factors influence fractional units for reporting cfDNA biomarker levels.

Absolute quantification of analytes in copies per mL using digital polymerase chain reaction (PCR) may help overcome this problem. Balaji et al⁴ used droplet digital PCR (ddPCR) to demonstrate that tumor marker levels reported in copies per mL rather than allele fraction were more predictive of survival. Digital PCR is widely considered the gold standard for quantifying DNA per mL of solute,^5,6 but it is limited in scalability given that only a few targets are measurable per assay. In contrast, a single NGS assay can quantify multiple targets, but NGS measurements are not reproducible because read counts are affected by run-specific factors such as depth of coverage and extraction efficiency.^7-9 Allegretti et al⁷ reported that plasma concentration of extracted cfDNA (ng/mL) varied more than 500-fold across their cohort of patients with cancer, despite blood processing within 1 hour of collection, suggesting that natural and ex vivo factors can vary dramatically. In an individual patient whose biomarker levels are monitored serially, NGS read concentrations per mL do not always trend with fractional values.^8,10 Because total patient cfDNA concentrations vary over time, serial measurements reported as allele fractions may be misleading if the patient acquires a transient acute infection (or another cfDNA-inciting event), causing a temporary decrease in serial fractional biomarker levels, even when the disease of interest remains stable or worsens.¹¹ In this regard, fractional concentration per total DNA can be misleading, but absolute concentration per mL better reflects disease burden.

The implication is that patients and their health care professionals can better manage disease when the kinetics of serial measurements reflect a change in disease burden while also revealing the trajectory of new, actionable variants, such as emerging drug-resistance factors. In contrast, stable serial levels of a biomarker suggest that the underlying condition does not immediately threaten health, such as clonal hematopoiesis or the silent virome.

Prior NGS studies attempted to estimate analyte concentrations in ng per mL by multiplying the extracted cfDNA concentration (ng/mL) by the allele fraction.^11,12 This strategy may be more accurate if all cfDNA were human in origin and if all cells in every person had the same molecular weight (eg, diploid with fixed telomeres). Nevertheless, this simple ng per mL calculation has merit. In fact, Liu et al¹² found that per mL alteration levels were more prognostic than allele fractions in a cohort of patients with pancreatic cancer. Likewise, Chabon et al¹³ reported a stronger correlation between marker levels and metabolic tumor volume when reported per mL compared with allele fraction (P = .002 vs P = .003). In mathematical models for estimating tumor volume, Avanzini¹⁴ favored per mL over allele fraction as the informative measurement. Another factor in favor of measuring disease markers per mL rather than per ng is that cfDNA originates from both human and nonhuman sources. In particular, severe illness and other stressors can elevate plasma virome levels (eg, herpes-, polyoma-, adeno-, papilloma-, anello-, circo-, and parvoviruses) in excess of a million viral genomes/mL.

To address concerns with current practices, we developed a novel strategy to report NGS values in copies per mL. We first designed synthetic oligonucleotides with no significant homology to the human genome or to known human pathogen genomes.⁹ By spiking them into plasma, and then normalizing NGS read counts for the proportional recovery of synthetic DNA reads, we could account and adjust for variable efficiencies from cfDNA extraction through library preparation, sequencing, and informatics. In replicate tests, volumetric concentrations of human and viral targets were more precise when adjusted for the efficiency of the total test system.⁹ Unique molecular identifiers helped ensure that input cfDNA samples were counted as unique molecules after collapsing read families comprised of amplicons generated during PCR.¹⁵

In the current study, we applied this new normalization strategy alongside an assay calibration strategy to report absolute concentrations. We selected the Epstein-Barr virus (EBV) genome as a model system because plasma EBV viral loads are among the most longstanding cfDNA markers routinely used in patient care.^16,17 High levels of EBV in plasma are correlated with primary viral infection, reactivated infection, and EBV-related tumor burden. Serial changes in EBV load help track incipient tumor, disease progression, resolution, and treatment response for a range of diseases.^18-21 Although quantitative PCR is typically used to measure EBV viral load, it is now feasible to quantify EBV and many other infection- or cancer-related cfDNA samples by NGS.^2,22,23 We demonstrate proof of principle that concentrations of plasma EBV genomes are measurable in copies per mL by normalizing NGS read counts to those of spiked synthetic oligonucleotides and by calibrating NGS to the gold standard, ddPCR. We show how this method can overcome the shortcomings of traditional quantification strategies.

METHODS

Figure 1 shows an outline of the experiments to quantify the EBV genome in each plasma specimen by ddPCR and, in parallel, by an NGS panel targeting the EBV genome, among other biomarkers. Next, EBV concentrations by ddPCR were plotted against those by NGS to determine the degree of linearity and thus the suitability for calibrating NGS values to those of ddPCR. The protocol is detailed in the sections that follow.

FIGURE 1.

Workflow for quantifying Epstein-Barr virus (EBV) viral load by droplet digital polymerase chain reaction (ddPCR) and next-generation sequencing (NGS). EndoGenus spikes were added to plasma before extraction so that variations in the efficiency of the NGS total test system could be quantified and used to normalize EBV read counts.

Patient and Mock Plasma DNA Selection

Twenty-four plasma specimens (12 patient plasmas and 12 mock plasmas) were selected from a prior study⁹ in which NGS was performed to quantify EBV DNA alongside 23 cancer gene variants and additional oncogenic pathogen genomes. These 24 plasmas included EBV-positive plasmas (n = 21) across a wide range of EBV viral loads and EBV-negative control plasmas (n = 3) Table 1. The research protocol was approved by the University of North Carolina biomedical institutional review board.

TABLE 1.

Epstein-Barr Viral Load in Patient Plasmas by 2 Units of Measurement

Sample	Clinical diagnosis	EBV load by NGS, copies/million human reads	EBV load by NGS, precalibration normalized copies/mL
P1	Gastroesophageal adenocarcinoma	8	1
P2	Gastroesophageal adenocarcinoma	22	16
P3	Gastroesophageal adenocarcinoma	20	2
P4	Colonic adenocarcinoma	47	4
P5	Hodgkin lymphoma	36	11
P6	Nasopharyngeal carcinoma	18,315	724
P7	Nasopharyngeal carcinoma	18,941	1,804
P8	Acute liver failure	994	701
P9	Posttransplant viremia	343	288
NC1	Gastroesophageal adenocarcinoma	0	0
NC2	Pancreatic neuroendocrine tumor	0	0
NC3	Pancreatic neuroendocrine tumor	0	0

EBV, Epstein-Barr virus; NGS, next-generation sequencing.

ddPCR Quantification of EBV Genomes

Two ddPCR assays were designed to quantify the EBV genome in mock and patient plasma. Primers and probes were designed and purchased (Bio-Rad) to amplify 2 separate segments of the EBV genome within the highly conserved BMRF1 and BALF4 genes, with amplicon lengths of 67 and 75 base pairs (bp), respectively. Following duplex 20-µL ddPCR assays (QX200 instrument and reagents, QuantaSoft analysis software [Bio-Rad]), EBV load was calculated by dividing the copies per PCR by the volume in mL of plasma represented in the PCR assay, reported as copies per mL.

DNA Extraction and NGS

Plasma cfDNA was sequenced using the EndoGenus Plasma Mutation Panel, as previously described,⁹ and EBV loads were reported in 2 ways: (1) EBV genome count per million human reads and (2) EBV genome copies per mL after applying the normalizer formula to adjust for NGS assay efficiency (described below). Briefly, 12 EDTA anticoagulated blood specimens were double-centrifuged (2,000g for 10 minutes each) to prepare patient plasma. Additionally, 2-mL aliquots of 4 mock plasma specimens were tested in duplicate (Seraseq ctDNA Reference Material v2 at human gene variant allele fractions 0.125%, 0.25%, 1% [also tested at 1- and 4-mL inputs], and 2%; SeraCare) containing DNA prepared in part from lymphoblastic cell line DNA²⁴ harboring EBV DNA. Manufacturer documentation denotes the mock plasma DNA length as approximately 170 bp, closely matching the fragment length of natural tumor-derived DNA.²⁴ Both patient and mock plasma samples were spiked with synthetic oligonucleotides (EndoGenus spikes), and DNA was extracted from 1- to 4-mL aliquots. Mock plasmas were extracted in duplicate, whereas natural patient plasmas were extracted only once because of limited sample volumes.

To study the impact of plasma input volumes, duplicate analysis of the 1% mock plasma specimen was performed at each of 3 input volumes (1, 2, or 4 mL), for a total of 6 replicates. DNA from both natural and mock plasmas was extracted into 60 µL of eluate using the Maxwell RSC and LV extraction kit (Promega). Purified DNA was quantified by fluorimetry (Qubit HS DNA kit, Thermo Fisher Scientific). Library preparation was performed per manufacturer instructions (LiquidPlex, ArcherDX), modified to add 26 primer sets targeting the EBV genome, primers for EndoGenus spikes, 28 human cancer genes, and additional pathogens.^9,24,25 Pools of 16 libraries at 1.5 pmol/L (quantified by real-time PCR; KAPA Library Quantification kit, Kapa Biosystems) were sequenced on an Illumina NextSeq instrument using v2 mid-reagents, and FASTQ files were uploaded to Archer Analysis software, version 6.0 (ArcherDX), for alignment and unique read counts.

In each plasma specimen, unique read counts were generated for 26 segments of the EBV genome and for the 5 EndoGenus spikes targeted by the NGS panel. The EBV viral load was calculated in 2 ways: in copies per million human reads and in relative copies per mL of plasma after applying the following normalizer formula to adjust for assay efficiency. This formula normalizes EBV levels for the proportional recovery of synthetic DNA spike reads⁹:

$${\displaystyle \begin{array}{lll}& \mathrm{R}\mathrm{e}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{E}\mathrm{B}\mathrm{V}\mathrm{l}\mathrm{o}\mathrm{a}\mathrm{d}\left(\mathrm{c}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{e}\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{m}\mathrm{L}\right)=& \mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{E}\mathrm{B}\mathrm{V}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{d}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}\mathrm{x}\frac{75\mathrm{c}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{e}\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{m}\mathrm{L}}{\mathrm{O}\mathrm{b}\mathrm{s}\mathrm{e}\mathrm{r}\mathrm{v}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{n}\mathrm{d}\mathrm{o}\mathrm{G}\mathrm{e}\mathrm{n}\mathrm{u}\mathrm{s}\mathrm{s}\mathrm{p}\mathrm{i}\mathrm{k}\mathrm{e}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{d}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}}\end{array}}$$

The average EndoGenus spike read count is 75 copies per mL, and any increase or decrease in the observed read count proportionally influences the relative EBV load (copies per mL).

NGS Calibration Method

The NGS normalizer formula above yields relative values for EBV load that were previously shown to be precise and reproducible.⁹ These values are not absolute concentrations of EBV per mL of plasma, however, because assay calibration is required to achieve accuracy. To convert relative to absolute concentrations by NGS, we first determined the relationship between NGS values and ddPCR values generated after parallel tests on aliquots of the same cfDNA samples. Pearson correlation using linear regression was used to assess linearity between different test protocols. Relative NGS and ddPCR values with a linear relationship were then converted to absolute values using a slope/intercept strategy and the standard equation for a line (y = mx + b).

RESULTS

ddPCR Quantification

Our ddPCR generated distinct, reproducible double-positive, double-negative, and single-positive cloud patterns when droplets were evaluated for the BMRF1 and BALF4 segments of the EBV genome. Precise counts for each target molecule in copies per PCR were determined based on Poisson statistics of partitioning among the droplets. The known input volume was used to calculate EBV load in units of copies per mL of plasma.

In both mock and human plasma-derived DNAs, EBV load by BMRF1 ddPCR was consistently higher than by BALF4 ddPCR Table 2. This difference was unexpected given that (1) the 2 ddPCR assays targeted conserved segments of the same viral genome; (2) ddPCR is widely considered to be an accurate method of quantifying DNA; and (3) these ddPCR assays had similar amplicon lengths, although the slightly longer BALF4 amplicon (75 bp vs 67 bp) could be a factor.²⁶ Nevertheless, the 2 ddPCR assays yielded EBV genome values that strongly correlated with each other across a wide range of EBV loads, with a high linear correlation coefficient (R² = 0.99) Figure 2.

TABLE 2.

The BMRF1 ddPCR Assay Yields Higher EBV Loads Than the BALF4 ddPCR Assay

	Mean EBV BMRF1:BALF4 ratio, copies per mL	Range,a (CV)
EBV-positive patient plasma (n = 7)	2.6	1.4-3.9 (0.27)
EBV-positive mock plasma (n = 11)	1.4	1.4-1.5 (0.02)

CV, coefficient of variation; ddPCR, droplet digital polymerase chain reaction; EBV, Epstein-Barr virus.

^a Sampling error may contribute to more precise measurements in mock rather than natural patient plasmas because of lower viral loads.

FIGURE 2.

Epstein-Barr virus (EBV) droplet digital polymerase chain reaction assays are linear but not equivalent. The EBV loads by BMRF1 amplification tend to generate higher values than BALF4. Linearity is excellent, suggesting the potential to use either of these assays to track changes in viral load over time in serial patient specimens (R² = 0.9935).

NGS Analyses and Linear Relationship to ddPCR Values

By NGS analyses, the number of EBV genomes was estimated based on the median read count across 26 segments of the viral genome in the targeted NGS panel. Viral genome concentrations by NGS were reported in 2 ways: first by calculating EBV genome load in copies per million human reads and second by normalizing EBV genome counts for the efficiency of the NGS assay using the normalizer formula to report EBV load in relative copies per mL Table 3.

TABLE 3.

EBV Viral Loads by ddPCR Assays and NGS in 24 Plasma Samples

			ddPCR					NGS
	Specimen	Plasma extraction volume, mL	DNA input, ng	EBV BMRF1 load, copies/ddPCR	EBV BMRF1 load, copies/mL	EBV BALF4 load, copies/ddPCR	EBV BALF4 load, copies/mL	DNA input, ng	EBV load, copies/million human reads	EBV load, copies/mL
EBV-positive plasma samples	P1a	1	1.3	0	0	0	0	9	8	1
	P2	1	10	20	313	8.8	138	72	22	16
	P3a	1	1.3	1.4	15.0	0	0	9	20	2
	P4	2	2.6	4.2	23	1.4	7.5	18	47	4
	P5	2	10	46	246	11.8	63	73	36	11
	P6	4	2.3	3,140	8,411	1,154	3,091	16	18,315	724
	P7	3	3.7	4,200	15,000	1,660	5,929	26	18,941	1,804
	P8	2	18.5	2,340	12,536	1,036	5,550	132	994	701
	P9	2	20	756	4,050	536	2,871	146	343	288
EBV-negative plasma samples	NC1	1	1.8	0	0	0	0	13	0	0
	NC2	2	23	0	0	0	0	165	0	0
	NC3	2	30	0	0	0	0	211	0	0
EBV-positive mock plasma samples	M1a	2	2.9	6,820	36,535	4,840	25,928	21	42,251	8,630
	M1b	2	4.1	9,240	49,500	6,440	34,500	29	41,820	9,647
	M2a	2	3.4	7,560	40,500	5,200	27,857	24	39,490	8,690
	M2ba	2	3.6	0	0	0	0	26	39048	7,615
	M3a	1	2.7	4,840	51,857	3,340	35,786	19	42,725	10,662
	M3b	1	2.5	4,820	51,643	3,340	35,786	18	41,721	12,222
	M3c	2	3.7	7,960	51,466	5,860	37,888	26	41,524	10,619
	M3d	2	3.8	9,920	53,143	6,960	37,286	27	41,826	10,502
	M3e	4	7.6	18,940	59,188	13,200	41,250	54	39,514	9,332
	M3f	4	8.1	20,600	55,179	15,180	40,661	58	40,333	10,305
	M4a	2	2.9	6,880	36,857	4,880	26,143	20	40,223	10,063
	M4b	2	4.2	7,440	39,857	5,280	28,285	30	39,296	7,123

ddPCR, droplet digital polymerase chain reaction; EBV, Epstein-Barr virus; NGS, next-generation sequencing.

^aThree of 24 plasma samples were excluded from data analyses because of undetectable EBV by 1 of 2 ddPCR assays or by both ddPCR assays compared with NGS detection, suggesting sampling or technical error.

The NGS loads were consistently higher than the ddPCR loads, yet there was a linear relationship between NGS results and each ddPCR assay Figure 3. This correlation was weaker when NGS values were reported in copies per million human reads rather than relative copies per mL, after normalizing to spike recovery (R² = 0.91 vs 0.95). This observation suggests that normalizing read counts for assay efficiency adds value over raw read concentrations.

FIGURE 3.

Calibration of Epstein-Barr virus (EBV) loads by next-generation sequencing (NGS) to those of the droplet digital polymerase chain reaction (ddPCR) assays. There is a linear relationship between EBV levels by NGS and EBV levels by either ddPCR assay. Linearity is weaker when NGS values are expressed fractionally as copies/million human reads (A) (BMRF1: y = 1.1362x + 443.54, R² = 0.9216; BALF4: y = 0.8128x – 383.8, R² = 0.9092) vs as copies/mL (B) (BMRF1: y = 4.6516x + 2104.1, R² = 0.9515; BALF4: y = 3.3614x + 641.08, R² = 0.9578). A linear relationship between NGS and each ddPCR assay permits NGS assay calibration whereby EBV read counts by NGS can be extrapolated and reported as EBV loads in absolute copies/mL.

NGS Calibration to ddPCR Values

The strong linear relationship between NGS and ddPCR values permitted calibration of NGS to each ddPCR assay to extrapolate EBV load in copies/mL plasma. By this strategy, NGS values are converted to ddPCR-based concentrations using the equation for a line: y = mx + b. These linear equations are shown in Figure 3b for EBV BALF4 and BMRF1 ddPCR assays.

Although these 2 linear equations reflect the relationship between NGS and each ddPCR assay, it remains unclear whether 1 of the ddPCR assays is more accurate. If we assume that true EBV load is an average of the 2 measured ddPCR concentrations, the average slope of the 2 linear equations in Figure 3B is 4.1. Applying this slope as a 4.1-fold multiplier, with an intercept of 0, we can generate NGS concentrations more comparable to those of ddPCR by using the following refined NGS normalizer formula:

$${\displaystyle \begin{array}{lll}& \mathrm{E}\mathrm{B}\mathrm{V}\mathrm{l}\mathrm{o}\mathrm{a}\mathrm{d}\left(\mathrm{c}\mathrm{o}\mathrm{p}\mathrm{i}\mathrm{e}\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{m}\mathrm{L}\right)=& \mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{E}\mathrm{B}\mathrm{V}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{d}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}\mathrm{x}\frac{308\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{m}\mathrm{L}}{\mathrm{O}\mathrm{b}\mathrm{s}\mathrm{e}\mathrm{r}\mathrm{v}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{n}\mathrm{d}\mathrm{o}\mathrm{G}\mathrm{e}\mathrm{n}\mathrm{u}\mathrm{s}\mathrm{s}\mathrm{p}\mathrm{i}\mathrm{k}\mathrm{e}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{d}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}}\end{array}}$$

Using this refined equation, our NGS EBV measurements are now reportable in copies per mL to achieve substantially equivalent values to those of the gold-standard ddPCR, which is a hallmark of assay calibration.

DISCUSSION

In this study, we show proof of principle that plasma analytes detected by NGS can be reported quantitatively in copies per mL. We also demonstrate that NGS sensitivity and linearity are comparable to ddPCR across a dynamic range up to a calibrated level of approximately 50,000 copies per mL. To our knowledge, ours is the first study to show how spiked normalizers can be applied to generate measurements that are substantially equivalent to those of a reference procedure.

We chose to calibrate NGS to ddPCR because ddPCR is widely considered a gold-standard method for absolute quantification of nucleic acid. Simplicity (relatively few wet-lab steps compared with NGS) combined with a Poisson statistics–based strategy for achieving accuracy without the need for calibration are features that render ddPCR an effective measurement tool. Interestingly, the 2 EBV ddPCR assays that we developed behaved differently from each other, as illustrated by divergent slopes in the linear equations shown in Figure 3B, despite both assays targeting conserved regions of the same viral genome. Possible explanations for these differences include nonrandom cfDNA fragmentation patterns, active viral replication, cross-reactivity or polymorphisms affecting primer or probe binding, and segmental deletion or duplication within the viral genome.^27,28 Noting emerging evidence of the EBV genome’s propensity to recombine,^29-31 our results in Tables 2 and 3 are consistent with gene copy number variation within the viral genomes of patient plasmas (deletion of BALF4 or duplication of BMRF1), which should be explored in a larger cohort as a potential mechanism of oncogenesis. Regardless of the underlying reason, the fact that both the BMRF1 and BALF4 ddPCR assays demonstrate strong linearity relative to each other and relative to NGS values demonstrates the potential for any of these assays to monitor serial changes in EBV concentrations in copies per mL. Assays that have good analytic performance (eg, sensitive, specific, linear, precise) are candidates for tracking rising or falling viral loads in serial samples from a given patient, using the kinetics of change to inform medical decision-making.³²

As for absolute quantification, our findings call into question whether any single analytic procedure could be relied on for accuracy. A reminder of laboratory medicine principles seems relevant to this discussion. A major purpose of calibration is to achieve accuracy (trueness) that is traceable to a higher-order reference material or measurement procedure.³³ Accuracy is biologically desirable; more realistically achievable is intra- and interlaboratory equivalence of values (harmonization), ideally extending across measurement systems.

Standard reference material has been developed for a handful of infectious and cancer-related DNA measurements,^34-40 though rarely are both fractional and copies per mL values assigned to these materials, and these materials are not intended to mimic the cfDNA fragmentation patterns of natural biomarkers. The paltry availability of reference materials, considering the millions of disease markers (eg, pathogen genomes, somatic variants) potentially measurable by NGS, suggests that a methods-based strategy for NGS calibration is a practical solution. In this regard, spiked oligonucleotides could serve a larger role as reference material for calibrating measurements by normalizing values generated across multiple measurable biomarkers and test systems. The commercial availability of such a synthetic DNA reference material having an assigned concentration could help standardize NGS measurements when accompanied by a bed file to align spike reads as well as a detailed description of the material to guide design of targeted enrichment and to help testing laboratories create a robust informatics pipeline. The same reference material could potentially be tested for its ability to normalize values that other types of test systems generate, such as quantitative PCR assays. Success of this strategy depends on how closely the reference material mimics the natural analytes with respect to the efficiency of each step in the protocol. Measurements at intermediate steps within a protocol may assist in troubleshooting problematic assays.

We acknowledge that although additional experiments are required to evaluate the accuracy of EBV and other analyte levels that this NGS measurement system generates,^25,41-44 the principle is now established that NGS concentrations can be calibrated and reported in copies per mL. We recommend a methods-based approach to help standardize such concentrations across many analytes and test systems.

It should be noted that our innovative NGS strategy does not preclude reporting in traditional units of measurement that remain useful—for example, by filtering out noise at low allele fractions or by helping to distinguish disease markers from confounders (eg, germline variants, coexisting benign clones or biomarkers).

Trends in biomarker levels in serial plasma cfDNA samples are increasingly used to monitor disease burden in cancer and infectious disease settings.^22,45,46 Next-generation sequencing is a powerful method to quantify multiple markers at once, improving our ability to detect minimal residual disease, quantify the impact of medical intervention, and help guide selection of a different therapy when there are early signs of disease progression or drug resistance.

Our novel NGS normalization strategy suggests that we can generate not only fractional values but also concentrations in copies per mL that must be evaluated in clinical trials for their ability to improve short-term decisions and long-term medical outcomes. Further investigation is warranted into how spiked normalizers could help standardize measurements across various NGS panels and in different testing laboratories.

Conflict of interest disclosure

The authors have nothing to disclose.

Funding

This study was supported by the National Center for Advancing Translational Sciences (NCATS) UL1TR002489, the National Cancer Institute (Innovative Molecular Analysis Technologies) NCI R21 CA22903, the University of North Carolina (UNC) Translational Science Center NIH UL1TR002489, the UNC Department of Pathology and Laboratory Medicine, and the Lineberger Comprehensive Cancer Center.