Comparison of Circulating Tumour Cells and Circulating Cell-Free Epstein-Barr Virus DNA in Patients with Nasopharyngeal Carcinoma Undergoing Radiotherapy

Quantification of Epstein-Barr virus (EBV) cell-free DNA (cfDNA) is commonly used in clinical settings as a circulating biomarker in nasopharyngeal carcinoma (NPC), but there has been no comparison with circulating tumour cells (CTCs). Our study aims to compare the performance of CTC enumeration against EBV cfDNA quantitation through digital PCR (dPCR) and quantitative PCR. 74 plasma samples from 46 NPC patients at baseline and one month after radiotherapy with or without concurrent chemotherapy were analysed. CTCs were captured by microsieve technology and enumerated, while three different methods of EBV cfDNA quantification were applied, including an in-house qPCR assay for BamHI-W fragment, a CE-IVD qPCR assay (Sentosa ®) and a dPCR (Clarity™) assay for Epstein-Barr nuclear antigen 1 (EBNA1). EBV cfDNA quantitation by all workflows showed stronger correlation with clinical stage, radiological response and overall survival in comparison with CTC enumeration. The highest detection rate of EBV cfDNA in pre-treatment samples was seen with the BamHI-W qPCR assay (89%), followed by EBNA1-dPCR (85%) and EBNA1-qPCR (67%) assays. Overall, we show that EBV cfDNA outperforms CTC enumeration in correlation with clinical outcomes of NPC patients undergoing treatment. Techniques such as dPCR and target selection of BamHI-W may improve sensitivity for EBV cfDNA detection.

less is known about CTCs in relatively neglected cancers such as NPC. There has been no previous comparison of performance and utility between circulating biomarkers such as CTCs and more conventional EBV cfDNA approaches, with only a comparison between different EBV DNA qPCR quantification assays involving different targets being previously reported 5 .
Hence, we investigate here the utility of various circulating biomarkers in NPC, with a special interest in the performance of CTC enumeration as a novel biomarker against more conventional EBV cfDNA quantitation using qPCR and digital PCR (dPCR) with EBNA1 and BamHI-W as targets.

Results
Comparison of sensitivity and specificity between EBV cfDNA assays. Benchmarking of the EBV cfDNA was conducted using comparison against results from a College of American Pathologists (CAP)accredited laboratory as well as WHO-approved international EBV standards.
The clinical sensitivity and specificity of the three EBV cfDNA assays was benchmarked against an in-house EBV cfDNA assay targeting EBNA1 in a College of American Pathologists (CAP)-accredited clinical-grade laboratory at the Singapore General Hospital (SGH), with known analytical performance reported as a sensitivity of 79% and specificity of 100%. With this assay, 46 NPC patients (Table 1), 31 (69%) were reported EBV-positive, 14 (31%) EBV-negative (1 case was not done due to logistic reasons). Of 31 EBV-positive patients on the clinical-grade assay, both BamHI-W qPCR and EBNA1-dPCR assays showed 100% matching positivity, whereas the EBNA1-qPCR assay showed 80% match. Of the 14 EBV-negative patients, the BamHI-W qPCR, EBNA1-dPCR and EBNA1-qPCR assay reported 9, 7, and 5 positive cases. Overall, all three EBV cfDNA assays demonstrate high clinical sensitivity and specificity, with particularly high sensitivity shown at baseline for the BamHI-W qPCR assay, as expected.
The only available WHO-approved international EBV standard was used to benchmark the sensitivity and specificity of the three EBV cfDNA assays. The BamHI-W qPCR assay demonstrated the highest reproducible sensitivity. The lowest EBV concentration detected in triplicates was 100 IU/mL for BamHI-W qPCR assay and 1,000 IU/mL for both EBNA1 assays ( Table 2). The BamHI-W qPCR assay was also able to detect positive signal in one replicate of the standard containing 1 IU/mL, whereas EBNA1 assays were not able to. In addition, all assays produced no false-positive detection in five EBV-free standards, indicating their high specificity against EBV cfDNA. The IU of NIBSC standards is derived from a mean value of highly variable EBV copy number measured by various qPCR assays of 28 laboratories in the world 8 . These assays employ different DNA extraction methods, and target a wide range of genes, including a single-copy gene, EBNA1, and a multiple-repeat gene, BamHI-W 8 . However, since dPCR was not included in the evaluation, the relationship between EBV copy number as obtained by dPCR and IU is less clear. Moreover, since the number of BamHI-W fragments varies in different EBV isolates, a fixed conversion ratio of BamHI-W copies to IU will not be always accurate in different patients' sample. Therefore, the NIBSC standards were only used in this study for comparison of sensitivity and specificity between EBV cfDNA assays. The subsequent data were to be reported in copy number of respective EBV targets.
Relationship between NPC circulating biomarkers in pre-treatment samples. Among EBV cfDNA quantitation approaches, BamHI-W qPCR assay yielded the highest concentration of EBV cfDNA levels: 2.4 to 37.7-fold higher than EBNA1-qPCR assay and 2.2 to 25.5-fold higher than EBNA1-dPCR assay ( Table 3). All samples detected EBV-positive by both EBNA1 assays were also detected positive for EBV by BamHI-W assay. The detection rates of canonical CTCs and potential CTCs are 76% and 94% in pre-treatment samples respectively. Overall, potential CTC count was higher and weakly correlated to canonical CTC count (r 2 = 0.21, P-value = < 0.01). No correlation was observed between each type of CTC count and EBV cfDNA levels quantified by different assays. However, among the EBV cfDNA assays, strong correlation was observed between BamHI-W qPCR and EBNA1-dPCR assays (r 2 = 0.99, P-value < 0.0001), but not between BamHI-W and EBNA1-qPCR assays (r 2 = 0.03, P-value = 0.29) nor between EBNA1-qPCR and -dPCR assays (r 2 = 0.06, P-value = 0.11). This result corresponded with the similar detection rate of BamHI-W qPCR (89%) and EBNA1-dPCR (85%) assays, with the detection rate of EBNA1-qPCR assay being 67%.

Relationship between NPC circulating biomarkers and clinical stage.
The clinical stages were re-classified to three groups; stage I, stage II-III, and stage IV ( Table 4). The combination of stage-II and -III NPC patients was in the light of long-term 5-year follow-up data from Singapore showing similar survival outcomes using modern treatment approaches 13 . The EBV cfDNA levels in three assays strongly correlated with clinical stages. In contrast, there was no statistically significant relationship between CTCs and clinical stages. These results indicated a strong association between NPC clinical stage and EBV cfDNA, but not CTCs.

Relationship between NPC circulating biomarkers and treatment outcome. Decreased
EBV cfDNA levels were observed in all EBV-positive patients following treatment, strongly correlating with the local radiological response (Table 5). To evaluate the predictive value of NPC circulating biomarkers for short-term radiological response, we determined that EBV cfDNA levels were significantly reduced after treatment (Wilcoxon's signed rank testing p-value < 0.001 for all three techniques BamHI-W qPCR, EBNA1-dPCR and EBNA1-qPCR assay). In contrast, for both canonical and potential CTCs, decrease was not significant (p = 0.07 and 0.54 respectively). The stratified analysis performed on patients undergoing radiotherapy and chemo-radiotherapy showed the magnitude of decrease of canonical CTCs pre-and post-treatment in each group remains insignificant (Supplementary Table 1). Overall, our results show that EBV cfDNA level correlation with short-term radiological response was much stronger than that of potential or canonical CTC counts.
Relationship between NPC circulating biomarkers and overall survival. Survival analysis demonstrated that there was a stronger correlation between EBV cfDNA and overall survival, as compared to that between CTC counts and overall survival. All three EBV cfDNA techniques showed prognostic value on survival analysis: BamHI-W qPCR, EBNA1-dPCR and EBNA1-qPCR assays yielded corresponding p-values of 0.03, 0.02 and 0.0002 by log-rank testing respectively, whereas canonical CTC and potential CTC counts were not associated with overall survival (p = 0.66 and 0.13 respectively). Kaplan-Meier plots are also shown for dichotomized biomarker variables (Supplementary Figure 4).

Discussion
Non-invasive approaches of NPC diagnosis have been available for the past decade via the detection of immunoglobulin A antibody against EBV antigens in patients' serum 14,15 . However, these techniques are inefficient Spike-in Standards (IU/mL) dPCR  Assay   1,000,000  3  3  3  3   1,000  3  3  3  3   100  3  3  1  2   10  3  2  1   in NPC prognosis and relapse prediction 16,17 . There is considerable ongoing research into EBV cfDNA in NPC patients for prediction of post-treatment outcomes 6,18,19 , and its role in selecting patients for additional adjuvant treatment following definitive therapy.  In our study, good correlation between EBV cfDNA and clinicopathologic outcomes was consistently demonstrated regardless of approach undertaken: BamHI-W qPCR, EBNA1-qPCR or EBNA1-dPCR assays. Decreased EBV cfDNA levels are commonly observed in almost all patients undergoing treatment, corresponding generally to the short-term post-treatment radiological response, which is commonly a complete or near-complete response. Overall, our results demonstrated that EBV cfDNA yielded better results in comparison with CTC count as a circulating biomarker for NPC. Regardless of approach, cfDNA showed far stronger correlation with tumor stage, short-term radiological response as well as overall survival, in comparison with CTC counts.

EBNA1-
The detection rate of the in-house BamHI-W qPCR assay was 89%, which was similar to a separate study targeting the same BamHI-W fragment 18 , reporting 96% positive detection in Hong Kong NPC patients. In comparison with clinically validated assays, the in-house BamHI-W qPCR assay demonstrated better performance. The detection rate of the CE-IVD EBNA1-qPCR assay reported in this study was 67%, despite its claimed clinical sensitivity of 100%, based on 80 EBV-positive samples. Moreover, EBV positive cases reported by the BamHI-W qPCR assay were matched with the ones reported by the SGH assay, which had clinical sensitivity of 79%.
Despite being a powerful tool in NPC prognosis, the quantification of EBV cfDNA faces challenges of standardization. The NIBSC standards, which are derived from whole EBV produced by B95-8 cells 8 provide a consensus estimate of EBV IU, but are not ideal for standardization of BamHI-W copy number. In addition, the NIBSC spike-in standards do not truly represent the NPC plasma samples. Naturally occurring cfDNA has a size of less than 181 bp in NPC plasma 20 whereas DNA obtained from NIBSC was genomic DNA with a size of 170 kb 21 . The differences in DNA size influence the choice of DNA extraction kit, which in turn has meaningful impact on DNA recovery, and subsequently DNA quantification. Unlike BamHI-W qPCR and EBNA1-dPCR assays, the EBNA1-qPCR assay was performed using the automatic Sentosa ® system integrated with both nucleic acid extraction and EBV quantification. The QIAamp Circulating Nucleic Acid Kit (Qiagen) used in BamHI-W qPCR and EBNA1-dPCR assays were both designed for extraction of fragmented cfDNA as short as 75 bp whereas Sentosa ® SX Virus Total Nucleic Acid Kit v2.0 (Vela Diagnostics) used in EBNA1-qPCR assay was optimized for total viral DNA extraction. As the comparison of assay performance was conducted on samples undergoing different extraction methods, the performance differences between the two platform technologies, qPCR and dPCR, may also reflect differences in extraction. However, this caveat does not change the conclusion that EBV cfDNA quantification outperforms CTC quantification. The variation in efficiency of DNA extraction kits could explain why EBNA1-qPCR and EBNA1-dPCR assays target the same EBV single-gene EBNA1, and yet differ much in detection rate in NPC plasma samples. Another reason for the higher detection rate of EBNA1-dPCR assay could be the difference in quantification platform in which dPCR technology carries the advantage of being more sensitive. By targeting the multiple-repeat BamHI-W fragments, the in-house BamHI-W qPCR assay yielded the highest detection rate in NPC pre-treatment samples. It also yielded the highest sensitivity in measurement of NIBSC spike-in standards despite the possible DNA losses due to the DNA extraction method potentially not optimized to genomic DNA. On the other hand, regardless of being different in fundamental techniques of quantification and EBV targets, BamHI-W qPCR and EBNA1-dPCR assays were strongly correlated in the measurement of EBV levels in pre-treatment samples. This correlation could possibly be aided by the same extraction process from which the cfDNA used in BamHI-W qPCR and EBNA1-dPCR assays was extracted. Altogether, in our interpretation, the in-house and dPCR assays are more likely to quantify the true values of EBV cfDNA level in pre-treatment samples of NPC patients. Nevertheless, as the absolute values of EBV cfDNA levels in clinical samples are unknown, it cannot be readily concluded which of the three assays performed with better accuracy. Another factor affecting EBV cfDNA quantification was earlier reported to be the PCR master mix 7 . The harmonization study concluded higher consistency of EBV cfDNA quantification in commercially available Roche master mix after being compared with an in-house master mix, which was more prone to batch-to-batch variations. It is certainly possible that master-mix differences could also contribute to such variation in detection.
The evidence of EBV cfDNA existing in the form of short and freely-floating fragments in the plasma had led to a conclusion that they were released from apoptotic NPC cells 20,22,23 . In other words, the NPC cells releasing EBV cfDNA lysed before they had the chance to enter the bloodstream. This phenomenon could explain the non-correlation between NPC CTC counts and EBV cfDNA levels measured by various assays. Overall, our results are the first comparison between EBV cfDNA and CTC count, showing that EBV cfDNA is a better biomarker than CTC enumeration in NPC prognosis and prediction of treatment outcomes, and reveals heterogeneity between NPC circulating biomarkers at the molecular and cellular levels. Our study also demonstrated that by targeting the multiple-repeat BamHI-W, higher detection rate and sensitivity were achieved. Further, we demonstrate that dPCR is useful as a detection method for EBV cfDNA, with potential advantages over qPCR.

Post-Treatment
Participating laboratories and clinic. Institute of Bioengineering and Nanotechnology (IBN) served as the centralised laboratory of the study (Supplementary Figure 1). Blood samples were collected from consenting NPC patients at National Cancer Centre Singapore, and sent to IBN within the same day of their visits within 4 hours. For each sample, whole blood was used for immediate CTC enumeration, and plasma was obtained, assigned blinded IDs and stored at −80 °C until further use. Each plasma assay had its individually optimized volumes. 250 µL of frozen plasma was distributed to Singapore General Hospital (SGH) where cfDNA extraction and quantification was performed using the Sentosa ® SA EBV Quantitative PCR Test (Vela Diagnostics) following manufacturer's requirements. At IBN, 1 mL of thawed plasma was used for cfDNA extraction of which half was quantified by the in-house BamHI-W assay. The other half of the extracted cfDNA was sent to JN Medsys where cfDNA quantification was conducted using the Clarity TM Digital PCR System (JN Medsys). The equivalent plasma volume per reaction was 60 µL. Each reaction mix was incubated at 40 °C for 10 min to allow UNG to degrade carry-over PCR products, followed by 95 °C for 10 min for UNG inactivation. The reaction mix was partitioned into approximately 10,000 individual reactions in the Clarity TM Digital PCR tube-strip (JN Medsys). Thereafter, the tube-strips were stabilised for 2 min, sealed with 230 µL sealing fluid and subjected to thermal cycling using the following parameters: 1 cycle at 95 °C for 5 min, 40 cycles at 95 °C for 50 sec and 58 °C for 1.5 min. Afterward, the tube-strips were transferred to the Clarity TM Reader (JN Medsys), which detected and quantified fluorescence signals from all partitions. Absolute copy number of EBNA1 in each reaction was determined by the Clarity TM Software (JN Medsys) after analysis of the ratio of positive partitions (i.e. those that contained amplified products) over the total number of partitions, using Poisson statistics.

BamHI
Determination of sensitivity and specificity of EBV cfDNA assays. All three EBV cfDNA assays were benchmarked against the EBV qPCR assay routinely performed by the College of American Pathologists (CAP)-certified laboratory in SGH. The clinical sensitivity and clinical specificity of the SGH assay was reported as 79% and 100% respectively, based on 66 untreated nasopharyngeal carcinoma patients and 30 normal volunteers. In addition, sensitivity and specificity of EBV cfDNA assays were benchmarked against the 1 st World Health Organization (WHO) International Standards for EBV, code 09/260; from National Institute for Biological Standards and Control (NIBSC). The NIBSC standards and nuclease-free water were spiked into EBV-free plasma to obtain 18 standards of 6 known EBV concentrations, ranging from 0 to 1,000,000 IU/mL. In addition, two aliquots of EBV-free plasma served as blank standards. The protocol of DNA extraction, sample distribution and EBV cfDNA assays of spike-in standards was identical to the one for clinical plasma samples.
Enumeration of NPC CTCs. CTCs from 1 mL of whole blood were captured using the microsieve technology and enumerated with the aid of biomarker characterization as described previously 24,25 . The microsieve technology is a size-based method capable of isolating both epithelial and mesenchymal CTCs, unlike the affinity system, which only captures EpCAM-expressed CTCs. Cell counting, and image analysis were performed subject to sample availability, using the MetaMorph software (Molecular Devices) and manually verified by trained laboratory technicians. Cytokeratin-positive and CD45-negative nucleated cells were classified as canonical CTCs.
Other nucleated cells that were negative for both cytokeratin and CD45 biomarkers were defined as potential CTCs. All nucleated cells with CD45-positive were classified as white blood cells (Supplementary Figure 2).

Statistical analysis.
Correlation study was carried out to correlate EBV levels amongst the NPC circulating biomarkers assays. Logistic ordinal regression modelling was used to evaluate pre-treatment circulating biomarker quantitation relative to the dependent variable of clinical stage. Wilcoxon's signed-rank test with continuity correction (R.3.0.0) was conducted to compare paired pre and post-treatment levels of NPC circulating biomarkers. Correlation was performed using Microsoft Excel and the logistic ordinal regression model was performed using the "orm {rms}" library package in R. Alpha was set to 0.05 throughout. Survival analysis was performed using R 3.0.0 survival package to study survival distributions of continuous pre-treatment levels of NPC circulating biomarkers and overall survival (Table 3), using log-rank testing to determine significance at a threshold of 0.05. 1 patient (Patient-025) was omitted from survival analysis, as the patient sought follow-up elsewhere.