Digital polymerase chain reaction for detecting c-MYC copy number gain in tissue and cell-free plasma samples of colorectal cancer patients

We focused on the utility of the droplet digital polymerase chain reaction (ddPCR) for detecting c-MYC gene copy number (GCN) gain in cell-free plasma and tumor tissue of colorectal cancer (CRC) patients. c-MYC GCN status was determined using dual-color silver in situ hybridization (SISH) and ddPCR in retrospective cohort 1 (192 CRC patients) and prospective cohort 2 (64 CRC patients). In cohort 1, c-MYC GCN gain was observed in 34 (17.5%) patients by SISH, and in 7 (3.6%) patients by ddPCR. c-MYC GCN by SISH significantly correlated with ddPCR results (ρ = 0.532, P < 0.001). Although 40 cases (20.7%) showed intratumoral genetic heterogeneity, it did not cause discordance in results obtained by the two methods. c-MYC GCN gain, by both SISH and ddPCR was independently correlated with worst prognosis (P = 0.002). In cohort 2, c-MYC GCN estimation in tissue by ddPCR was also significantly associated with results obtained by SISH (ρ = 0.349, P = 0.005), but correlated with plasma ddPCR with borderline significance (ρ = 0.246, P = 0.050). Additionally, detecting c-MYC GCN gain in plasma with ddPCR might have relatively low sensitivity but high specificity. Our study suggests that ddPCR can be a useful tool for detecting c-MYC GCN gain as a potential prognostic biomarker in CRC tissue samples; however, this will need further verification in plasma samples.

However, analysis of ctDNA requires a method of high sensitivity, as tumor DNA is present at a very low concentration in plasma; thus, ddPCR is expected to overcome this limitation 11 . This method has better ability to precisely quantify the concentration of DNA in a sample as compared to that of traditional quantitative PCR. The ddPCR has been reported to detect cancer mutational status with high concordance [12][13][14] . Interestingly, previous studies have indicated that ddPCR has the ability to accurately screen for GCN status as well as mutations in plasma DNA 15 . Analysis of ctDNA with ddPCR has the potential to detect HER2 amplification in breast and stomach cancer 16,17 . Moreover, it has been shown that ddPCR can determine the MET GCN status in ctDNA with high accuracy 18 . Therefore, ddPCR seems to be a suitable and highly sensitive technique for GCN detection in ctDNA.
In this study, we aimed to analyze whether ddPCR could be adapted to detect small increases of c-MYC GCN in plasma and compared with the c-MYC GCN detected in the primary CRC tissue, using SISH and ddPCR.

Clinicopathological features and frequency of c-MYC GCN gain in cohort 1. We investigated
c-MYC GCN in 192 CRC tissues of cohort 1 by two different methods: SISH and ddPCR. c-MYC GCN gain, defined as mean c-MYC copies/nucleus ≥ 4.0 in SISH analysis, was observed in 34 (17.5%) cases, while by ddPCR method, was observed in 7 (3.6%) cases. Despite the discordance in frequency between the two methods, results by these two methods were significantly associated by Pearson's correlation (ρ = 0.532, P < 0.001).
We hypothesized that the genetic heterogeneity of c-MYC GCN in each tumor cell might be the cause of discrepancy between the SISH and ddPCR results. Intratumoral genetic heterogeneity of c-MYC GCN gain, which was arbitrarily defined as the tumor cells with c-MYC GCN ≥ 4.0, consisted 5 to 50%. When the cells with c-MYC GCN ≥ 4.0 were less than 5% or more than 50%, the tumor was considered genetically homogenous in terms of c-MYC GCN. Forty cases (20.8%) showed intratumoral genetic heterogeneity for c-MYC GCN gain. However, intratumoral genetic heterogeneity of c-MYC GCN gain was not causal for the discordance in results between SISH and ddPCR methods (Table 1; P = 0.492). Table 2 summarizes the correlations detected between clinicopathological features and c-MYC GCN gain by ddPCR, in cohort 1. Since we have previously demonstrated the correlation between c-MYC GCN gain by SISH, and clinicopathological features of CRC 5 , here we present only the results of the ddPCR analysis. However, no statistically significant correlation was observed between the clinicopathological factors and c-MYC GCN gain by ddPCR.

Overall survival of cohort 1 patients with c-MYC GCN gain by sIsH and ddpCR methods.
Kaplan-Meier survival curves illustrated the prognostic effect of c-MYC GCN gain by different detection methods. The mean overall survival of patients with CRC of cohort 1 was 55 months (range 1-73 months). Regardless of the detection method, c-MYC GCN gain was associated with unfavorable overall survival in primary CRC tissues (Fig. 1A,B; P = 0.028 and P = 0.010, respectively). In Fig. 1C, c-MYC GCN gain by neither SISH nor ddPCR (SISH-/ddPCR-), by only SISH (SISH+/ddPCR-), by only ddPCR (SISH-/ddPCR+), and by both SISH and ddPCR (SISH+/ddPCR+), was found in 156 (81.3%), 29 (15.1%), 2 (1.0%) and 5 (2.6%) cases respectively. We attempted to perform SISH on whole tissue sections from two SISH-/ddPCR+ cases to determine the reason for the discordant results. However, we did not observe c-MYC GCN gain (SISH ≥ 4) in whole sections of these tumors. c-MYC GCN gain by both SISH and ddPCR (SISH+/ddPCR+) was most significantly correlated with unfavorable prognosis ( Fig. 1D; P = 0.001). As there were only two SISH-/ddPCR+ cases, this was not a sufficient number for performing survival analysis.
In addition, multivariate Cox proportional hazards analysis indicated that c-MYC GCN gain with both SISH and ddPCR (SISH+/ddPCR+) independently predicted unfavorable prognosis in cohort 1 (Table 3). Table 4 demonstrates the relationships between c-MYC status (tissue ddPCR) and the clinicopathological parameters of the patients of cohort 2. For analyzing ctDNA, it is necessary to focus on clinicopathological factors that presumably account for high concentrations of ctDNA in plasma. At the time of diagnosis, although only two cases showed distant metastasis, 30 cases had lymph node involvement and 11 cases were T4 stage. Lymphatic and venous invasion was observed in 44 and 16 cases, respectively.

Clinicopathological characteristics of cohort 2 patients.
c-MYC GCN gain (tissue ddPCR) may be correlated with tumor a location of the ascending to descending colon (P = 0.023) and lymph node metastasis (P = 0.028). Histologically, low-grade CRCs seem to lack the c-MYC GCN gain (P = 0.019). However, only four cases showed c-MYC GCN gain (tissue ddPCR), which was not sufficient for a statistically significant result from the χ 2 or Fisher's exact test. Comparative analysis of c-MYC GCN status of tumor tissue and plasma ctDNA in cohort 2. In cohort 2, investigation of c-MYC GCN status was conducted on plasma sample as well as on surgical specimens of 64 CRC patients. c-MYC GCN gain by SISH, was observed in 10 (15.6%) patients, by tissue ddPCR in four (6.3%) patients and by plasma ddPCR in one (1.6%) patient. A patient who was detected with c-MYC GCN gain by plasma ddPCR, also showed c-MYC GCN gain by tissue ddPCR and SISH. All four c-MYC GCN gain cases by tissue ddPCR, also showed c-MYC GCN gain by SISH (Fig. 2a). Intratumoral genetic and regional (central and peripheral) heterogeneity of c-MYC status was found in 29 (45.3%) and 8 (12.7%) cases, respectively. However, these heterogeneities were not causal for the discrepancy in c-MYC status results obtained, via different methods of detection (P > 0.05, Supplementary Table 1). Table 5 demonstrates the correlation coefficient of c-MYC statuses obtained by different methods of determination. c-MYC GCN by SISH was significantly associated with tissue ddPCR results (ρ = 0.349, P = 0.005), but not with plasma ddPCR results (P = 0.620). c-MYC status in plasma was positively associated with tissue ddPCR results but showed only borderline statistical significance (ρ = 0.246, P = 0.050). Moreover, c-MYC GCN gain (≥4.0) by SISH was significantly associated with tissue ddPCR results (P = 0.004), but not with plasma ddPCR results (P = 0.482) in Wilcoxon rank-sum test (Fig. 2b,c).

Discussion
The availability of noninvasive assays to detect and monitor tumor status is a major challenge in oncology. Although, ctDNA has emerged as a potential surrogate for precision medicine, the low levels of ctDNA pose a big challenge for successful detection. Recent studies have suggested that the discordance in detection rate between tumor tissue and plasma can be resolved by high-sensitivity ddPCR [19][20][21][22] . Representatively, the ddPCR assay for EGFR mutation in lung cancer was reported to be a highly sensitive and specific biomarker for clinical blood testing 23,24 . In CRC, the clinical utility of ddPCR to detect KRAS mutation in ctDNA is thought to be promising 14,25 .
Furthermore, approaches for detection of GCN alteration from ctDNA are also under the spotlight 15,26-28 . Bhuvan et al. suggested that point mutations in ctDNA might be difficult to detect due to the low ctDNA concentration derived from early stage cancer 27 . On the contrary, GCN gain can contribute a much larger number of ctDNA fragments to the overall plasma. Hence, detection of GCN gain might hypothetically be easier than that of point mutations in ctDNA-based cancer screening. Studies on HER2 have been the most active in the field of   respectively, in gastric cancer patients 16 . Heidrun et al. indicated that ctDNA is a potential screening tool for HER2 amplification in metastatic breast cancer with a positive and negative predictive value of 70% and 92%, respectively 17 . Although ddPCR for detecting HER2 GCN status in ctDNA may be relatively less sensitive, its specificity seems to be high. In our study, we observed borderline correlation between plasma and tissue c-MYC GCN status by ddPCR. However, there was no significant correlation between c-MYC GCN status in plasma by ddPCR, and in tissue by SISH (Table 5). Interestingly, only one case was detected with c-MYC GCN gain (c-MYC ≥ 4) in plasma. This case also showed c-MYC GCN gain in tissue by both ddPCR and SISH (Fig. 2),  Table 4. The correlation between clinicopathological factor and c-MYC GCN gain with ddPCR in 64 CRC patients of cohort 2. p values are from the χ 2 or Fisher's exact test and were significant at less than 0.05.  16 . Our study consisted of only 2 (3.6%) cases with distant metastasis, 30 (46.9%) cases with lymph node involvement, and 11 (17.2%) cases with T4 stage. Relatively early stage CRC patients were included and hence the quantity of released ctDNA in the plasma might have been insufficient for detection. Indeed, sensitivity might prove to be a limitation in detecting c-MYC GCN status in ctDNA of non-advanced CRC patients.
Focusing on FFPE tissue, previous research demonstrated that ddPCR method was as effective as fluorescence in situ hybridization (FISH) and therefore can become a standard method [29][30][31] . Our study demonstrated that results from SISH positively correlated with the results from ddPCR in FFPE tissues, of both cohort 1 and 2 (ρ = 0.532, P < 0.001 and ρ = 0.349, P = 0.005). Nonetheless, when the GCN gain criteria (c-MYC ≥ 4) are applied, the frequency of GCN gain is observed to be discordant, depending on the detection method used. Several reasons for this discordance can be suggested: first, FFPE tissues require fixation, and this may cause genomic DNA damage and degradation. These conditions can induce false negative results because of the low quality and quantity of DNA. Second, SISH is a microscopy-based method that involves directly and optically counting the target gene in tumor cells. This may be the most accurate method; however, personal observation and manual calculation can be potentially error-prone. We cannot exclude the possibility therefore that the SISH method produces more false positive results than ddPCR. On the other hand, the determination of GCN by ddPCR might be underestimated due to the presence of non-tumor cells, including immune cells and stromal cells in the sample 18,26 . A recent study recommended estimating the tumor content ratio (TCR) of a sample for improving the accuracy of GCN analysis by ddPCR 32 . They suggested that determining the HER2 status using ddPCR, calibrated by the TCR, is advisable in clinical practice because non-tumor cells can influence the GCN status in samples with a relatively small amount of cancer cells. The tumor fraction of our samples might have influenced the c-MYC GCN detection by ddPCR. This may be the main reason that GCN gain (c-MYC ≥ 4) by ddPCR was less frequently observed than that by SISH.
Two cases showed c-MYC GCN gain only by ddPCR (SISH-/ddPCR+), whereas 29 cases showed c-MYC GCN gain only by SISH (SISH+/ddPCR-) in cohort 1 (Fig. 1C). We attempted to discover the reason for discordance between the two methods and hypothesized that this discordance could be induced by intratumoral genetic heterogeneity. If there was intratumoral heterogeneity of c-MYC GCN, an amplified portion would be missed in the SISH test or non-amplified portion would cause the ddPCR result to be negative for c-MYC GCN gain. The intratumoral heterogeneity of c-MYC GCN was arbitrarily defined as GCN gain (c-MYC ≥ 4) in tumor cells between 5% and 50%. Indeed, by SISH microscopy, we detected intratumoral genetic heterogeneity of c-MYC GCN in each tumor cell. However, we were unable to find significant association between discordant results depending on methods and intratumoral heterogeneity of c-MYC GCN.
We previously reported that c-MYC GCN gain with SISH is a poor prognostic marker for CRC patients 5 . In the present study, c-MYC GCN gain was correlated with unfavorable overall survival, not only by SISH but also by ddPCR. Interestingly, c-MYC GCN gain with both SISH and ddPCR (SISH+/ddPCR+) was independently correlated with worst prognosis (Table 3).
In conclusion, we tried to determine the c-MYC GCN status in the ctDNA of preoperative CRC patients, by ddPCR. To the best of our knowledge, we are the first to attempt detecting c-MYC GCN by ddPCR in CRC patients. Although the ddPCR assay showed low sensitivity in detecting c-MYC GCN gain in ctDNA of non-advanced CRC patients, it detected c-MYC GCN gain in ctDNA with high specificity. However, we cannot recommend ddPCR of plasma samples as a first screening tool for c-MYC GCN gain due to the high risk of false negative results. Thus, ddPCR may require further evaluation in plasma samples. There was also discrepancy between c-MYC GCN gain measured by SISH and ddPCR in FFPE tissues; nevertheless, the ddPCR results were significantly correlated with the SISH results. Thus, we can suggest the detection of c-MYC GCN gain by ddPCR as a potential prognostic biomarker in CRC tissue.  Droplet digital pCR. Using the human eukaryotic initiation factor 2C1 (EIF2C1) gene as an internal control to assess the copy number of the MYC gene, MYC-to-EIF2C1 ratios were determined using ddPCR. Each sample was partitioned into 20,000 droplets, with target and background (reference) DNA distributed randomly, but not uniformly, among the droplets. Amplicon lengths for ddPCR reaction of MYC and EIF2C1 were 121 bp and 86 bp, respectively. The following FAM probes were used for ddPCR; MYC: PrimePCR ™ ddPCR ™ Copy Number Assay (Bio-Rad) and HEX probes for EIF2C1: PrimePCR ™ ddPCR ™ Copy Number Assay (Bio-Rad).

Materials and Methods
The reactions were performed in 20 μL reaction volumes that consisted of up to 50 ng of extracted DNA (1 μL), 2x ddPCR supermix for probes (No dUTP) (10 μL), MYC primers/probes (1 μL), EIF2C1 primer/probes (1 μL) and deionized distilled water (7 μL). Emulsified PCRs were run in a 96-well plate on a C1000 Touch ™ Thermal Cycler (Bio-Rad). The plates were incubated at 95 °C for 10 min, followed by 50 cycles of 94 °C for 30 s, 60 °C for 60 s and 10 min incubation at 98 °C. The plates were read on a Bio-Rad QX200 droplet reader using the QuantaSoft v1.7.4 software (Bio-Rad) to assess the number of droplets positive for MYC and/or EIF2C1. MYC gene copy number determined by ddPCR was defined as 2 × MYC/EIF2C1. The cut-off for classifying samples as MYC GCN gain was set as ≥ 4 copies/cell. In addition, positive and negative experimental results were obtained according to the Digital MIQE Guideline and are shown in Supplementary Fig. 1

Availability of Materials and Data
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.