Tumor Panel sequencing and establishment of ddPCR assays
To identify the disease-associated mutations, we first conducted NGS analysis of tumor samples. To this end, tumor samples and corresponding blood samples of all nine recruited patients were collected and tumor DNAs were analyzed using the Tumor Panel. Somatic mutations were identified in all of the cases. Somatic PTEN mutations were identified in 7 of 9 patients, which are of high frequency as previously reported [25]: 2 were missense mutations and 5 were multi-hit mutations (more than one type). TP53 mutations were identified in all of the 6 serous carcinoma patients (EM001, 002, 003, 007, 008, 009), most of which were solely missense mutations, but these were absent in cases of non-serous carcinoma (EM004, 005, 006). In our analysis, the most frequently mutated genes were PTEN, FAT4, ARID1A, TP53, ZFHX3, ATM, and FBXW7 (Fig. 2a). The results were partially in accordance with the mutation pattern of the subgroup of serous-like/copy number high of EC, which was characterized by mutations of TP53, PIK3CA, FBXW7, PPP2R1A, PIK3R1, CHD4, PTEN, and CSMD3 (PPP2R1A, CHD4, and CSMD3 were not included in Tumor Panel) [25]. We focused on TP53, PTEN, PIK3CA, PIK3R1, and FBXW7, and pathogenic or likely pathogenic mutations in these genes that were recorded in CLISING and cBioportal database (Additional file 3: Table S3) to determine whether they were related to recurrence or FIGO stage. However, there were no specific mutation patterns between recurrent/non-recurrent cases or advanced/early stage cases. Regarding the counts of mutated genes, PTEN was also the top one, and we could see multi-locus mutations in most of the genes (Fig. 2b). In EM001, 003, 006, the count of variants was much higher than that in other cases; however, it was not correlated with prognosis or any other clinical characteristics (Fig. 2c).
In order to track somatic mutations in plasma DNA and identify the presence of ctDNA, we chose one-to-three mutations of high frequency to design personalized ddPCR assays for each case (Additional file 4: Table S4), and germline mutations were ruled out. These mutations are usually reported frameshift, stopgain or nSNV variants on exons with high agreement on detection rate between TPS and ddPCR with certain ddPCR assay.
Personalized tumor-specific ddPCR analysis was effective and reliable
After somatic mutations were identified for each case and personalized ddPCR assays were established, we investigated the potential utility of ctDNA analysis in high-risk EC in a prospectively cohort of 9 women presenting with high-risk EC (Fig. 2).
The data showed TPS and ddPCR analysis had a high level of agreement in the assessment of the mutant allele fractions in baseline tumor tissue DNA (Fig. 3a), demonstrating the robust ability to develop ddPCR assays for diverse mutations. DdPCR accurately quantified mutant DNA at single-molecule sensitivity, even in the presence of vast amounts of wild-type DNA (60 mutant copies in 64,060 cfDNA copies, 0.09%) (Fig. 3b). Tumors of EM004 and EM005 patients harbored the same somatic mutation (PTEN-c.389G > A), and ctDNA was not detected in the disease-free case while it was detected in the relapse case (Fig. 3b), demonstrating the effectivity and reliability of this method. In patients with more than one mutation identified in the primary tumor, we tracked all mutations in the plasma with stable agreement for present/absent mutation in the same plasma (Fig. 3c), emphasizing the reproducible and robust nature of the assays developed.
Post-operative ctDNA status was correlated with disease-free survival
We determined whether ctDNA status analyzed by ddPCR was associated with tumor relapse. The personalized ddPCR assays were used to track mutations in serial plasma samples collected as previously mentioned. We assessed the correlation between relapse and ctDNA status at different time points. Consistent with previous observations [16], ctDNA was detected in 67% (6 of 9) of the baseline plasma samples. Baseline level including median cfDNA level, mutant copies, mutant allele fraction, and ctDNA detection rate were associated with the advanced status of the disease such as FIGO stage and node status, but not correlated to other clinicopathological characteristics including age, pathology subtypes, tumor size, myometrial invasion, LVSI, and involvement of lower uterine segment (Table 1).
CtDNA detection at baseline, i.e. before any treatment, was not predictive of DFS (Fig. 4a). CfDNA and ctDNA levels at baseline were higher in patients who relapsed than in those who did not relapse (cfDNA copies: median of 344,200 versus 52,760/mL; mutant copies: median of 1840 versus 30/mL; mutant allele fraction: median of 3.25% versus 0.045%, for relapse and DFS, respectively), although not at a statistically significant level (Fig. 4c).
CtDNA tracking in serial post-operative plasma samples predicted tumor relapse. We then assessed the potential of ctDNA tracking to report tumor relapse. CtDNA was detected in post-operative blood tests in 44% (4 of 9) of the patients (Fig. 4b), with highly variable mutational loads (median of 900 copies/mL; range, 290 to 81,300 copies/mL, mutant allele fraction from 0.15% to 79.47%). In these samples, ctDNA detection was predictive of tumor relapse (DFS: median of 9 months [ctDNA detected] versus median undefined [ctDNA not detected]; hazard ratio [HR], 17.43 [95% CI, 1.614 to 188.3]) (Fig. 4b). In ctDNA tracking, in accordance with baseline values, ctDNA detection was correlated with FIGO stage and node metastasis (Additional file 5: Table S5), which along with tumor size and myometrial invasion, were predictors of tumor relapse (Additional file 6: Table S6). These results suggest that post-operative ctDNA detection is closely related to tumor relapse, although due to small sample size, we cannot statistically prove that ctDNA is an independent prognostic factor in a multivariate cox regression analysis (P = 0.336, HR, 0.007, 95% CI 0.000–164.011).
CtDNA was superior to CA125 or HE4 in detection of tumor relapse
Commonly used epithelial tumor markers CA125 and HE4 were concurrently monitored whenever ctDNA was analyzed. Of the patients who did not relapse, 5/6 did not show ctDNA in any post-operative plasma sample (P = 0.048). One patient (EM003) had ctDNA detected 8 months after surgery but stayed disease free until 26 months after surgery in the follow-up period (Fig. 5a). Of the patients who relapsed in the follow-up period, all cases (3/3) showed ctDNA in post-operative plasma samples (Fig. 5b).
The performance of common tumor markers was less reliable than ctDNA detection. CA125 was negative in all relapsed cases with an extremely high false-negative rate, and half of the disease-free cases had CA125 positive (3/6). HE4 had both sensitivity and specificity of 66.7% in estimating tumor relapse, with a kappa index of 0.308, which only suggested a fair concordance of this marker and tumor relapse estimation (Fig. 5c). Imaging had a perfect consistency with relapse, with 100% sensitivity and specificity. In general, the sensitivity of post-operative ctDNA detection to estimate tumor relapse was 100%, and the specificity was 83.3%, with a kappa index of 0.769, which indicated substantial concordance of post-operative ctDNA detection and tumor relapse (Fig. 5d).