Clinical Testing for Tumor Cell-Free DNA

Abstract

Context.—

Clinical testing for tumor cell-free DNA (cfDNA) has evolved rapidly, but no practice guidelines exist.

Objective.—

To summarize cfDNA laboratory practices based on self-reporting and assess preanalytical, analytical, and postanalytical trends that may influence the quality, accuracy, and consistency of cfDNA testing.

Design.—

Data were derived from the College of American Pathologists cfDNA proficiency testing program submitted by 101 participating laboratories from 2018 to 2019.

Results.—

Most laboratories performing clinical circulating tumor DNA testing are commercial/nonhospital (71.2%; 72 of 101) and international (77.2%; 78 of 101) laboratories. Commercial laboratories had higher monthly test volumes than hospital-based laboratories (median, 36 versus 7–8) and tended to have larger gene panels (median, 50 versus 11 genes) when panel-based testing was offered. The main clinical indications include therapy selection and treatment/disease monitoring. Plasma is the most commonly accepted specimen, which is predominantly collected in cell-stabilizing tubes. Equal proportions of laboratories use next-generation sequencing (NGS) and non-NGS methods to assess key genes, including EGFR, BRAF, KRAS, NRAS, and IDH1. Most laboratories reported a lower limit of detection (LLOD) of 0.5%, variant allele frequency or less, which did not differ by method, NGS or non-NGS, except for EGFR. Sixty-five percent (17 of 26) of laboratories using the FDA-approved non-NGS EGFR assay report analytical sensitivities higher than 0.5%, as compared to 15% (16 of 104) of laboratories using an alternative NGS or non-NGS method. There is also a wider range in LLODs obtained for the FDA-approved EGFR assay than nonapproved assays.

Conclusions.—

These results highlight emerging practice trends and serve as a foundation to initiate future practice recommendations.

Cell-free DNA (cfDNA) was first reported in whole blood in 1948, which consists of short fragments of circulating DNA that are approximately 150 to 200 base pairs in length.^1,2 cfDNA may be released from any cell in the body secondary to cell lysis, necrosis, apoptosis, and secretion.^2–6 Several studies have shown that the main components of cfDNA present in whole blood represent circulating leukocyte DNA.^7,8 Since initial discovery, cfDNA has been identified in multiple bodily fluids, such as plasma or serum, cerebral spinal fluid, saliva, urine, and peritoneal and pleural fluid, and has shown potential as a robust biomarker in a variety of clinical settings, including pregnancy, transplant, end-stage renal disease, traumatic brain injury, cerebral infarction, myocardial infarction, and oncology.^3,4,9–19 In cancer patients, cfDNA may contain tumor-derived DNA, which is denoted as circulating tumor DNA (ctDNA), and the term liquid biopsy has been applied to the assessment of ctDNA.

The clinical utility and validity of cfDNA as a biomarker has been clearly demonstrated in the field of noninvasive prenatal screening (NIPS), whereby circulating fetal DNA in expectant mothers can be screened for genetic aneuploidy.^9,10,20 Given the widespread adoption of NIPS, the American College of Medical Genetics and Genomics published a position statement in 2016,²¹ and the Biochemical Molecular Genetics Committee of the College of American Pathologists (CAP) showed that through proficiency testing, clinical NIPS can be successfully performed with high accuracy.²² The clinical utility and validity of ctDNA as a cancer biomarker was first established in non–small cell lung cancer (NSCLC) for the detection of targetable EGFR alterations, which led to the US Food and Drug Administration (FDA) approval of a single-gene companion diagnostic assay in 2016.^23–26 Since then, high-sensitivity sequencing technologies have led to the FDA approval of multigene next-generation sequencing (NGS) assays for ctDNA applicable to many cancer types.^27–29 Although tissue-based tumor mutation profiling remains the standard for initial diagnosis and treatment selection, multiple large-scale studies have demonstrated high concordance between tissue and ctDNA molecular alterations, suggesting that ctDNA could potentially be used as a surrogate for, or adjunct to, tissue-based analysis.^30–34 Nevertheless, technical variation and discordance among ctDNA methods exist, particularly for mutations below 0.5% to 1% variant allele fraction (VAF),^35,36 highlighting a role for rigorous analytical validation, ongoing quality management and proficiency assessment, and regulatory oversight of these clinical assays.

By virtue of ctDNA being present at low levels, often 1% or less of total cfDNA, ultrasensitive methods are required.³ Non-NGS methods, such as quantitative polymerase chain reaction (qPCR) and digital-based PCR, have emerged as fast, cost-effective, and highly sensitive (0.01–0.5% VAF) techniques for ctDNA detection. However, these methods are limited by the number of targets that can be reasonably assessed.^37,38 Panel size is typically not limiting for NGS methods, but dedicated chemistries and bioinformatics pipelines that can perform appropriate error-correction are needed to achieve the required level of sensitivity (at least 0.1% VAF).^19,37–39 Yet, the inherent cost, complexity, processing time, and computational requirements for such NGS assays precludes broad implementation. As such, large NGS-based panel testing is currently limited to large commercial laboratories and major academic cancer centers.

As ctDNA continues to be accepted as a cancer biomarker, integration of ctDNA testing into current diagnostic and oncologic practices is rapidly evolving. At this time, there are no standards or guidelines for clinical ctDNA testing. In 2018, the CAP and the American Society of Clinical Oncology published a joint review on ctDNA as an emerging biomarker in oncology,⁴⁰ which was followed by the development of proficiency testing for tumor cfDNA by the CAP. Concurrently, the CAP Molecular Oncology Committee initiated a laboratory practice survey to assess practice patterns in clinical ctDNA testing as related to preanalytical, analytical, and postanalytical parameters. Herein, we summarize the first 2 years of the CAP CFDNA Proficiency Testing (PT) Program on laboratory practices, which demonstrate clear trends in specimen processing, methodology, targets assessed, and clinical reporting. These results can serve as the basis for future practice recommendations and guidelines and assist laboratories with the development, validation, and quality management of ctDNA testing.

MATERIALS AND METHODS

Data Acquisition

Data were derived from a total of 4 CAP CFDNA PT mailings distributed in 2018 and 2019 (2018-A, 2018-B, 2019-A, 2019-B), which assessed both accuracy of ctDNA performance and self-reported laboratory practices. These PT programs were designed and administered by the CAP Molecular Oncology Committee. The accuracy of cfDNA testing will not be summarized herein. Laboratory practices were assessed by an 18-part questionnaire (Supplemental Table 1; see supplemental digital content containing 5 tables and 1 figure) that accompanied each mailing, and responses were solicited on a voluntary basis. These questions were designed to capture practice characteristics, including test volume, gene and variant content, specimen types tested, as well as other preanalytical, analytical, and postanalytical methods, and future practices.

Laboratory practice setting was extracted from the CAP demographics database and missing classifications were assigned through online review of the laboratory’s Web site. Practice settings were separated by hospital based and non–hospital based, and further separated as academic (university or teaching hospital) or nonacademic, and independent/commercial reference laboratory or other nonhospital laboratory.

Statistical Analysis

After data aggregation, statistical tests were performed for select results. A repeated measures logistic regression model was used to test for participant-specific monthly ctDNA test volume changes over time, and a Kruskal-Wallis test was used to analyze the distribution of monthly ctDNA volumes between practice settings, based on the laboratory’s last reported monthly ctDNA volume. Repeated measures logistic regression models were also used to test for practice changes for cell-stabilizing and K₂EDTA tube usage. Wilcoxon rank sum tests were used to evaluate the number of genes included in a panel between the hospital and nonhospital laboratories. The lower limit of detection (LLOD) distributions were evaluated with 2 tests. A logistic regression model was used to test for differences by gene, and a Fisher exact test was used to test the EGFR-specific LLOD group distribution differences between methods (NGS, non-NGS, and Roche EGFR). The comment response rate between practice settings was analyzed with a Pearson χ². A significance level of .05 was used for these analyses, and all summaries and testing were compiled with SAS 9.4 (SAS Institute, Cary, North Carolina).

RESULTS

Laboratory Practice Characteristics and Test Volumes

A total of 101 individual laboratories responded to the analysis questions. During the 2-year period, the number of laboratories participating in the program increased from 51 in 2018-A to 70 participants in 2018-B, 83 in 2019-A, and 86 in 2019-B. With respect to laboratory location, most participants (77.2%; 78) were international, particularly China based (25.7%; 26), and the remaining (22.8%; 23) were United States based (Table 1). The practice setting for the 101 laboratories were mostly independent/commercial laboratories (60.4%; 61), which was followed by academic hospital-based laboratories (14.9%; 15), nonacademic hospital-based laboratories (13.9%; 14), and other nonhospital laboratories (10.9%; 11). There were statistically significant differences in self-reported cfDNA test volume amongst the different practice settings (Table 2). Independent/commercial reference laboratories reported a median of 36 tests (5th–95th percentile, 1–672) performed each month, while academic hospital-based laboratories, nonacademic hospital-based laboratories, and nonhospital laboratories reported a median of 8, 7, and 10 tests each month, respectively. During the 2-year period, the self-reported test volumes showed no significant change in hospital and nonhospital practice settings (P = .07) (Figure 1), or for the entire cohort (P = .17) (Supplemental Figure 1; Supplemental Table 2).

Table 1.

Participant Demographics

Practice Demographic	No. (%)
Practice location	101
Other, international	52 (51.5)
China	26 (25.7)
United States	23 (22.8)
Practice setting	101
Independent/commercial reference laboratory	61 (60.4)
Hospital/medical center laboratory – academic	15 (14.9)
Hospital/medical center – nonacademic	14 (13.9)
Nonhospital laboratory	11 (10.9)

Table 2.

Monthly Circulating Tumor DNA Test Volume^a

Group	n	5th Percentile	25th Percentile	Median	75th Percentile	95th Percentile
Overall	101	1	5	15	80	650
Independent/commercial reference laboratory	61	1	10	36	200	672
Hospital/medical center laboratory – academic	15	0	2	8	16	200
Hospital/medical center – nonacademic	14	0	3	7	17	50
Nonhospital laboratory	11	2	10	10	50	100

Abbreviation: ctDNA, circulating tumor DNA.

^aData were derived from the last reported monthly ctDNA volume. P =.002.

Figure 1.

Monthly circulating tumor DNA (ctDNA) test volume separated by laboratory practice setting over time. The outlier counts are listed above the modified axis. The median value is embedded within the plot. There was no statistically significant monthly volume change between these mailings (P=.07).

Gene and Variants Tested

The spectrum of genes and variants tested were similar between domestic and international laboratories. Nearly all surveyed laboratories tested for alterations in EGFR (Figure 2, A and B) followed by BRAF, KRAS, NRAS, ERBB2, ALK, MET, ROS1, IDH1, and other genes (Supplemental Table 3). Of the types of alterations tested, most were single nucleotide variants; however, BRAF multinucleotide variants and ALK and ROS1 fusion events were also tested. The 101 respondents indicated that cfDNA testing was performed either by a gene panel (43.6%; 44), individual genes (42.6%; 43), or a combination of both (13.9%; 14). Although the most commonly tested genes did not differ among practice settings, the number of genes tested varied by setting. The smallest panels were encountered in hospital settings (median of 11 genes; n = 9), and the largest panels were encountered in nonhospital settings (median of 50 genes; n = 49) (Table 3), but this difference did not reach statistical significance (P = .07) based on the Wilcoxon rank sum test.

Figure 2.

The most commonly assessed genes and variants tested by participating laboratories. A, Most commonly assessed genes with ALK and ROS1 represented fusion detection, whereas the remainder represented DNA mutation detection. B, Most commonly assessed variants. Note that the number of variant-level responses varied by program.

Table 3.

Number of Genes in the Tested Panels^a

Group	n	Min	25th Percentile	Median	75th Percentile	Max
All laboratories	58	3	11	33	70	694
Nonhospital	49	3	12	50	77	694
Hospital	9	5	11	11	19	70

Abbreviations: Min, minimum; Max, maximum.

^aP =.07.

Preanalytical Parameters and Specimen Handling

Given that cfDNA can be derived from different body fluids, a range of specimen types may be tested. The program revealed that almost all of the 101 laboratories accepted plasma (99.0%; 100) for cfDNA testing. Other accepted specimens included pleural effusion fluid/bronchial washings (22.8%; 23), cerebrospinal fluid (21.8%; 22), serum (7.9%; 8), urine (4.0%; 4), and saliva (1.0%; 1) (Table 4). Five laboratories (5.0%) indicated that they used a non-plasma/serum variation of peripheral blood, such as whole blood. However, it was unclear whether this specimen was used for cfDNA testing, or as a source of comparative germline DNA.

Table 4.

Preanalytical Practice Characteristics

	No. (%)
Specimens accepted for ctDNA testinga	101
Plasma	100 (99.0)
Pleural effusion fluids/bronchial washings	23 (22.8)
Cerebrospinal fluid	22 (21.8)
Serum	8 (7.9)
Urine	4 (4.0)
Saliva	1 (1.0)
Other	5 (5.0)
Blood collection tubesa	101
Cell-stabilizing tube	85 (84.2)
Streck cell-free DNA BCT	74 (73.3)
Roche Cell-Free DNA Collection Tube	13 (12.9)
PaxGene Blood ccfDNA tube	8 (7.9)
Other cell-stabilizing tube	7 (6.9)
K2EDTA	38 (37.6)
Acid citrate dextrose	2 (2.0)
Other	1 (1.0)
Maximal amount of time between collection and processing with non-cell-stabilizing tubes	36
0–2 h	15 (41.7)
3–4 h	14 (38.9)
5–6 h	5 (13.9)
7–12 h	1 (2.8)
13–24 h	1 (2.8)
Reagents used for DNA extraction	101
QIAamp Circulating Nucleic Acid Kit (Qiagen)	44 (43.6)
Roche cobas cfDNA Sample Preparation Kit	22 (21.8)
MagMAX Cell-Free DNA Isolation Kit	14 (13.9)
QIAamp MinElute ccfDNA Mini Kit	3 (3.0)
Other	18 (17.8)
Minimum DNA required for testing	99
Yes	60 (60.6)
No	39 (39.4)

Abbreviation: ctDNA, circulating tumor DNA.

^aMultiple responses allowed.

Preanalytical variables, such as collection tube and time-to-processing, can greatly affect the quality and quantity of recovered cfDNA. The program responses showed that cell-stabilizing tubes were used by most laboratories (84.2%; 85 of 101) (Table 4). Of the laboratories that use cell-stabilizing tubes, the Streck cell-free DNA BCT tube is the most commonly used (74), followed by the Roche Cell-Free DNA Collection Tube (13), PaxGene Blood ccfDNA tube (8), and other cell-stabilizing tubes (7). As for non–cell-stabilizing tubes, K₂EDTA tubes continue to be used by 37.6% (38 of 101) of laboratories. There was no significant change in the usage of cell-stabilizing tubes (ranging from 81.5% to 86.2% from 2018-A through 2019-B; P = .21) or K₂EDTA tubes (ranging from 33.3% to 34.5% from 2018-A through 2019-B; P = .45) during the 2-year period when analyzed by a repeated measures logistic regression model. Thirty-six laboratories that only used non–cell-stabilizing tubes reported on time-to-processing, which was defined as the maximum amount of time between collection and processing. The allowable interval was 6 hours or less for 94.4% (34 of 36) for the majority of these laboratories. While most of these laboratories were hospital based (69%; 25 of 36), a third represented nonhospital (31%; 11 of 36) laboratories.

With regard to cfDNA extraction (Table 4), most laboratories reported using the QIAamp Circulating Nucleic Acid Kit (Qiagen) (43.6%; 44 of 101) followed by the Roche cobas cfDNA Sample Preparation Kit (21.8%; 22 of 101), MagMAX Cell-Free DNA Isolation Kit (ThermoFisher) (13.9%; 14 of 101), and QIAamp MinElute ccfDNA Mini Kit (3.0%; 3 of 101). Eighteen percent of the laboratories used an alternative method of cfDNA extraction designated “other.” No significant change in the extraction methods was observed over time (Supplemental Table 4). A minimum amount of DNA was required by 60.6% (60 of 99) of laboratories for clinical testing.

Analytical parameters and assay sensitivity ctDNA testing methodologies varied across laboratories and included both non-NGS and NGS techniques. Of the 101 laboratories, 19.8% (20) reported an enrichment step to enhance for mutant alleles as a part of testing. Approximately equal proportions of laboratories used non-NGS (46.0%; 46 of 100) and NGS (42.0%; 42 of 100) techniques, whereas a few laboratories (12.0%; 12 of 100) used a combination of the two (Table 5). Non-NGS cfDNA testing methods included real-time PCR, digital PCR, and allele-specific PCR; NGS cfDNA testing methods included amplicon based, hybrid capture based, and other sequencing based. On a gene level (Table 6), NGS was the predominant method used by at least 70.0% of laboratories assessing BRAF, KRAS, NRAS, and IDH1 variants, whereas only half of EGFR assays were performed by NGS (54.2%; 52 of 96).

Table 5.

Analytical Practice Characteristics

	No. (%)
Laboratory performs enrichment for mutant alleles	101
Yes	20 (19.8)
No	81 (80.2)
Methoda	100
Non-NGS	46 (46.0)
NGS	42 (42.0)
Both NGS and non-NGS	12 (12.0)
Technologies	74
PCR-based method	33 (44.6)
Real-time PCR	19 (25.7)
Digital PCR	11 (14.9)
Allele-specific PCR	3 (4.1)
Sequencing-based methods	32 (43.2)
Amplicon-based targeted NGS	18 (24.3)
Hybrid capture-based NGS	13 (17.6)
Sequencing based, other	1 (1.4)
Both methods	9 (12.2)
Digital PCR and amplicon-based targeted NGS	4 (5.4)
Digital PCR and hybrid-capture-based NGS	3 (4.1)
Allele-specific PCR and amplicon-based targeted NGS	1 (1.4)
Other PCR and sequencing based	1 (1.4)

Abbreviations: NGS, next-generation sequencing; PCR, polymerase chain reaction.

^aThis factor was determined by selecting a laboratory’s unique method(s) by mailing and then selecting the unique methods for all mailings.

Table 6.

Testing Methods Used by Gene

	Column Number (%)
Methoda	EGFR, n = 96	BRAF, n = 66	KRAS, n = 60	NRAS, n = 56	1DH1, n = 36
Next-generation sequencing	52 (54.2)	49 (74.2)	48 (80.0)	46 (82.1)	31 (86.1)
Roche cobas EGFR mutation	22 (22.9)	N/A	N/A	N/A	N/A
Digital PCR	14 (14.6)	9 (13.6)	4 (6.7)	3 (5.4)	2 (5.6)
Other commercial kit	9 (9.4)	6 (9.1)	5 (8.3)	6 (10.7)	3 (8.3)
PCR allele specific, quantitative	3 (3.1)	3 (4.5)	3 (5.0)	1 (1.8)	0 (0.0)
ARMS	2 (2.1)	1 (1.5)	1 (1.7)	0 (0.0)	0 (0.0)
Single base extension	2 (2.1)	2 (3.0)	2 (3.3)	1 (1.8)	0 (0.0)
LNA/PNA	1 (1.0)	1 (1.5)	1 (1.7)	0 (0.0)	0 (0.0)
PCR allele specific, nonquantitative	1 (1.0)	0 (0.0)	1 (1.7)	1 (1.8)	0 (0.0)
Sanger sequencing	1 (1.0)	1 (1.5)	1 (1.7)	0 (0.0)	2 (5.6)

Abbreviations: ARMS, amplification-refractory mutation system; LNA/PNA, locked nucleic acid and peptide nucleic acid; N/A, not applicable; PCR, polymerase chain reaction.

^aMultiple responses allowed.

The reported analytical sensitivity, or LLOD, for detecting cfDNA variants was similar for NGS and non-NGS assays. Eight-five percent or more of laboratory-developed tests (LDTs), both NGS and non-NGS, for EGFR, BRAF, KRAS, IDH1, and NRAS had LLODs of 0.5% VAF or less (Figure 3, A through E; Table 7). The proportion of stated LLODs was also similar when binned into different ranges below 0.5% VAF. Stated LLODs showed greater variation across laboratories for EGFR, especially the FDA-approved Roche cobas EGFR qPCR assay. Among laboratories using the Roche cobas EGFR assay (20.0%; 26 of 130 responses), LLODs ranged from 0.1% to 10.0% VAF as compared to a LLOD of 0.1% VAF for the p.L858R or p.T790M variants quoted by the manufacturer.²⁵ Given this wide range, fewer laboratories performing the FDA-approved assay reported achieving EGFR LLODs of 0.5% or less (34.6%; 9 of 26 responses), as compared to laboratories performing laboratory-developed assays for EGFR (84.6%; 88 of 104 responses) (Table 8).

Figure 3.

Stated LLOD for commonly assessed genes, based on variant allele frequency (%) and detection method. A, EGFR. B, BRAF. C, IDH1. D, KRAS. E, NRAS. Results are based on laboratory-specific nonduplicate method/LLOD testing combinations. LLODs above 1.0% are not plotted for BRAF, IDH1, KRAS, or NRAS, which include 2 for BRAF, 1 for IDH1, 2 for KRAS, and 2 for NRAS. Multiple responses were allowed. Abbreviations: LLOD, lower limit of detection; NGS, next-generation sequencing.

Table 7.

Lower Limit of Detection (LLOD) (%) Distribution Based on Gene

			LLOD Intervals – Row Number (%)
Gene	No. of Responsesa	Range of LLODs	≤0.10	0.11–0.25	0.26–0.50	>0.50
BRAF	83	0.01–5.00	27 (32.5)	18 (21.7)	27 (32.5)	11 (13.3)
EGFR	130	0.01–10.00	47 (36.2)	21 (16.2)	29 (22.3)	33 (25.4)
IDH1	47	0.05–2.00	13 (27.7)	10 (21.3)	20 (42.6)	4 (8.5)
KRAS	79	0.01–2.00	28 (35.4)	19 (24.1)	24 (30.4)	8 (10.1)
NRAS	70	0.01–2.00	18 (25.7)	16 (22.9)	26 (37.1)	10 (14.3)

^aMultiple responses allowed.

Table 8.

EGFR Lower Limit of Detection (LLOD) (%) Based on Method^a

			Summary of Most Common LLODs - Row Number (%)
EGFR Methodb	No. of Responsesb	Range of LLODs	≤0.10	0.11–0.25	0.26–0.50	0.51–1.00	1.01–3.00	>3.00
NGS	68	0.01–2.00	22 (32.4)	16 (23.5)	21 (30.9)	8 (11.8)	1 (1.5)	0 (0.0)
Non-NGS	36	0.01–5.00	17 (47.2)	5 (13.9)	7 (19.4)	4 (11.1)	2 (5.6)	1 (2.8)
Roche FDA	26	0.10–10.00	8 (30.8)	0 (0.0)	1 (3.8)	9 (34.6)	2 (7.7)	6 (23.1)

Abbreviations: FDA, US Food and Drug Administration; NGS, next-generation sequencing.

^aP <.001.

^bMultiple responses allowed.

Postanalytical Parameters and Reporting Practices

As cfDNA testing becomes more routine in the clinical setting, the utility of such testing is likely expanding. While the majority (95.0%; 96 of 101) of cfDNA testing is performed to identify drug targets (Table 9), other main indications include monitoring of disease activity (54.5%; 55 of 101) and detecting residual disease (32.7%; 33 of 101) and determining prognosis (29.7%; 30 of 101). A few laboratories perform ctDNA testing to resolve differential diagnoses that include malignancy (10.9%; 11 of 101), and fewer still perform these tests to screen unaffected individuals (6.9%; 7 of 101). The latter 7 laboratories were all nonhospital laboratories (6 independent/commercial reference laboratories and 1 “clinic” laboratory) with 5 being international sites. Despite the fact that most laboratories perform cfDNA testing for treatment selection and prognosis, which is heavily dependent on cancer type, 41.6% of laboratories accepted cfDNA specimens without a histopathologic diagnosis. As far as report content, the majority of the 101 laboratories (67.3%; 68) provided an interpretative comment related to therapies, and 38.6% provided interpretative comments related clinical trials. Nonhospital laboratories were significantly more likely to provide interpretation related to clinical trials, particularly international nonhospital laboratories (Supplemental Table 5). Finally, when a negative result is encountered, interpretative comments related to the possibility of a false-negative result is provided by more than half of laboratories (66.3%; 67).

Table 9.

Postanalytical and Clinical Reporting

	No. (%)
ctDNA assay intended usesa	101
Prediction for specific targeted therapy	96 (95.0)
Monitoring disease activity	55 (54.5)
Residual disease detection	33 (32.7
Determination of prognosis	30 (29.7)
Differential diagnosis of malignancy	11 (10.9)
Screening unaffected patients	7 (6.9)
Other	3 (3.0)
ctDNA specimens accepted without a histopathologic diagnosis	101
Yes	59 (58.4)
No	42 (41.6)
Laboratory provides an interpretive comment related to therapy	101
Yes	68 (67.3)
No	33 (32.7)
Laboratory provides an interpretive comment related to therapies in clinical trials	101
Yes	39 (38.6)
No	62 (61.4)
Laboratory provides an interpretive comment related to possible false-negative results if tumor DNA is not present in the specimen tested	101
Yes	67 (66.3)
No	34 (33.7)

Abbreviation: ctDNA, circulating tumor DNA.

^aMultiple responses allowed.

Future Practices

Future cfDNA laboratory practices were also assessed to better understand the direction of clinical liquid biopsy testing. Most respondents (62.0%; 49 of 79) did not plan to expand their ctDNA offerings in the next year (2020). Of the 26 participants (25.7%, 26 of 101) that responded to the types of liquid biopsy assays in development in their laboratory, 12 (46.2%) planned to offer circulating tumor cell assays, 6 (23.1%) planned to offer circulating RNA assays, 5 (19.2%) planned to offer exosome testing, and 3 (11.5%) planned to offer an alternative, unspecified type of liquid biopsy assay, including miRNA (microRNA) and methylation assessment.

DISCUSSION

As clinical ctDNA assays have become more widely available, it is increasingly important to assess testing accuracy and understand current trends in clinical laboratory practices. In 2018, CAP initiated both cfDNA proficiency testing, as well as a laboratory practice survey, which has been summarized herein. To the best of our knowledge, this is the first and largest interlaboratory assessment of clinical liquid biopsy practice trends, which can serve as a framework for initiating future clinical laboratory practice guidelines and standards.

Data from the 2018 and 2019 programs revealed several important demographic insights. First, most laboratories performing cfDNA testing were nonhospital or commercial/reference laboratories. Most of the participating laboratories were international with similar numbers of US-based and China-based nonhospital laboratories. While this finding may represent differences in the clinical adoption of cfDNA testing, it may also reflect a bias in laboratory subscription to CAP PT. Although monthly test volumes remained relatively constant during these 2 years, there was a 1.7× increase in the number of participating laboratories from 2018 to 2019, suggesting a modest expansion of clinical tumor cfDNA testing. Irrespective of practice setting, the major genes tested were similar across the entire cohort, reflecting a selective emphasis on targetable variants, particularly variants corresponding to FDA-approved therapies and those incorporated into National Comprehensive Cancer Network guidelines, such as for NSCLC.⁴¹

Preanalytical conditions can greatly affect the quantity and quality of cfDNA recovery and assay sensitivity, including specimen collection, handling and storage conditions, and time to processing.³⁸ The program responses showed clear trends in specimen collection and handling practices with a few exceptions. Nearly all laboratories preferred plasma collected in cell-stabilizing tubes, which are designed to stabilize, and limit ex vivo lysis of leukocytes for prolonged intervals.^42–44 However, select laboratories continue to accept plasma collected in non–cell-stabilizing tubes or serum, both of which have been shown to reduce analytical sensitivity of cfDNA assays.^45–48 As expected, almost all laboratories accepting non–cell-stabilizing collection tubes reported processing times of 6 hours or less. Although workflow details are unclear from this program, a notable 31% of nonhospital/commercial laboratories use non–cell-stabilizing collection tubes with most reporting processing time of 6 hours or less. Only 2 such participants reported processing times greater than 6 hours. If stated processing times are not achievable, the risk of a false-negative result should be noted and addressed (see below). With regard to serum, similar issues with false-negative studies exist since serum contains higher levels of total DNA, which is mostly gDNA (genomic DNA), secondary to the clotting reaction.^48,49 Laboratories accepting either types of specimens should carefully review analytical sensitivity of their assays and document any deviations from validated processes. In addition to blood, up to 23.0% of participants test an alternative source of cfDNA. Given demonstration of ctDNA in other bodily fluids, such as pleural effusion⁵⁰ and bronchial washings,⁵¹ cerebrospinal fluid,⁵² urine,⁵³ and saliva,⁵⁴ which can show higher amounts of tumor DNA,³⁹ it is anticipated that ctDNA testing from body fluids will continue to increase in the future.

DNA extraction and quantification methods are also important preanalytical variables that can greatly impact ctDNA results. The program results show no significant change in preferred cfDNA extraction methods over time. Of the 4 most common methods, 1 is silica/column based (QIAamp Circulating Nucleic Acid Kit), 1 is glass fiber based (cobas cfDNA Sample Preparation Kit), and 2 are bead based (MagMAX Cell-Free DNA Isolation Kit and QIAamp minElute ccfDNA mini kit). As new ctDNA extraction kits and methodologies continue to emerge, laboratories looking to adopt these newer methods should recognize the differences in yield and purity of cfDNA, and associated analytical sensitivity, for any new cfDNA extraction method.^55,56 Laboratories are also advised to perform appropriate validation studies to account for potential effects of cfDNA extraction method on LLOD.⁵⁷ While most laboratories reported using a minimum amount or concentration of DNA needed for cfDNA testing, which correlates with the ability to meet minimum LLOD thresholds,³⁶ the program did not address how DNA concentration is being measured or defined by laboratories. For instance, a concentration may be defined as a measurement of total DNA or specifically the cfDNA component or a tumor mutant allele–enriched component. In the future, it will be important to develop a consensus definition of cfDNA quantification to enable direct comparison of ctDNA results obtained in different clinical laboratories.

With regard to analytical characteristics, the field is divided approximately equally between NGS and non-NGS assays. To this end, both NGS and non-NGS LDTs reported similarly robust LLODs of 0.5% VAF or less. The only significant difference in the reported LLOD was noted between the FDA-approved Roche cobas EGFR assay and other laboratory-developed EGFR assays.

Referencing the manufacturer’s label,²⁵ the LLOD is calculated to be a 0.1% VAF for both p.L858R or p.T790M mutations, whereas the LLODs reported by participants ranged from 0.1% to 10.0% VAF. Only one-third of laboratories achieved the manufacturer’s intended level of sensitivity 0.1%. Since FDA-approved assays may be performed with deviations, which render them LDTs, such differences may have accounted for the wide-ranging LLODs, although this was not specifically addressed in these programs. Laboratories that do not achieve LLODs of 0.1% for this assay should review their protocols per manufacturer’s recommendations and reassess whether their LLODs are adequate to address the clinical needs of their patient population. To our knowledge, this is the first analysis showing differing LLODs for an FDA-approved in vitro diagnostic assay after clinical implementation. As the field continues to progress with increased demands on assay sensitivity, all laboratories are encouraged to perform periodic cfDNA assay optimization to meet changing clinical indications.

Analytical sensitivity directly relates to potential false-negative studies, and appropriately 66.3% of the laboratories provide an interpretative comment related to the possibility of a false-negative result in their reports. Laboratories that do not ascribe to this practice should consider this addition since reliability of ctDNA testing depends on a number of factors, including features inherent to the tumor (eg, stage, ctDNA shedding, degree of clonality, and subclonality), preanalytical variables (eg, DNA input, DNA quality and fragmentation), as well as analytical conditions (eg, read-depth, bioinformatic pipelines). In fact, there have been reports of up to a 30.0% false-negative rate in ctDNA testing in NSCLC.⁵⁸

At this time, tissue diagnosis and molecular characterization of tumors remains the gold standard. Interpreting ctDNA testing results in the context of prior clinical, pathologic, and/or molecular information is critical. For instance, variants may be detected that are unrelated to the cancer type in question, such as those related to clonal hematopoiesis,^59–62 that may confound the cfDNA result. Since most targeted therapies are disease specific, laboratories (59.0%) that accept cfDNA specimens without a known histopathologic diagnosis should strive to confirm the pathologic diagnosis in order to provide appropriate cancer-type–guided treatment recommendations. Lastly, when there is no known diagnosis, cfDNA may be useful as a screening modality for cancer. Although there have been a number of large-scale screening studies,^63–65 only a small number of laboratories offer cfDNA cancer screening at this time.

The heavy commercial/reference laboratory representation in the program participants may reflect bias in subscription, but it may also indicate barriers of entry for hospital-based laboratories. At present, high-sensitivity sequencing techniques available as manufactured kits are still limited and continue to show cross-assay variability.³⁶ While academic/industry consortia are working toward guidelines for validating ctDNA assays,⁵⁷ cross-platform standardization becomes challenging, especially with a lack of reference or quality control materials, although this is an active area of research and collaboration.⁶⁶ Therefore, substantial financial investment, prospective patient recruitment and sampling, sophisticated molecular diagnostic capacity, and robust bioinformatics support are needed to develop and validate an institutional ultrasensitive tumor cfDNA assay. Consistent testing volumes are also required to justify development of such assays, and most hospital-based laboratories may not generate sufficient demand, which is suggested by the comparably lower monthly volumes reported by hospital-based laboratories. In addition, most hospital-based laboratories depend on adequate reimbursement to remain solvent, and reimbursement for cfDNA tests remain tenuous. At present, cfDNA testing recommendations are beginning to enter clinical guidelines. The current National Comprehensive Cancer Network NSCLC (version 4.2021) guidelines are the first to incorporate cfDNA biomarker testing in limited clinical scenarios,⁴¹ but tissue biopsy is still recommended when cfDNA test results are negative. As ctDNA testing continues to evolve, there will be an increasing need to establish guidelines to harmonize laboratory testing and reporting practices. Much of the trends demonstrated in this study could guide future regulatory requirements and practice recommendations.