Clinicopathologic characteristics of the cohort
A total of 198 patients with EC and with at least 6 months of post-surgery follow up have been prospectively included in the study. Clinical characteristics are summarised in Supplementary Table 1. The cohort included patients with endometrioid (76%) and non-endometrioid carcinomas (24%), low (G1-2) and high grade (G3) (58% and 42%, respectively), FIGO stage I-IV (65%, 15%, 15%, 4.7%, respectively) tumours from all TCGA groups (POLE 7,7%, MSI 39%, NSMP 27%, HCN 26%) classified accordingly with the updated ESMO/ESGO consensus [1]. A total of 37 (19%) patients had a relapse with a median DFS of 13.9 months [2.6–49.2]. Twenty-four patients (12%) died of disease, showing the global cohort a median DSS of 19.1 months [8.6–45.7].
Targeted sequencing of UAs for personalized ctDNA detection
The UAs from all the patients were subjected to NGS using a targeted panel previously used to characterize EC patients [10, 26]. With this strategy we identified pathogenic mutation in 189 of the patients (95.46%) The 10 most frequently mutated genes for SNPs were PTEN (54.46%), PIK3CA (48.51%), TP53 (30.69%), ARID1A (28.71%), KRAS (20.79%), CTNNB1 (19.31%), PIK3R1 (17.82%), FBXW7 (14.36%), PPP2R1A (13.37%) and FGFR2 (9.41%). Focusing on genomic alterations, 19 patients (9.59%) had CNAs. The top 10 altered genes for CNAs are CCNE1 (4.46%), ERBB2 (2.48%), CDK2 (1.98%), AKT2 (0.99%), MDM2 (0.99%), MYC (0.99%), PIK3CA (0.99%), AR (0.50%), AXL (0.50%) and CCND3 (0.50%). These data are consistent with the most common alterations of EC described in tissue and UAs samples [26, 28].
Based on the genetic alterations found in the UAs, specific ddPCRs were designed to monitor those with the highest allelic frequency. In MSI tumours, a panel of 5 microsatellite markers was also assessed by ddPCR. Overall, we could design effective assays to monitor ctDNA in 177 patients. In addition to the pre-surgical time point, we obtained follow-up blood samples from 130 patients (65.66%) to study the value of longitudinal cfDNA/ctDNA monitoring (Fig. 1).
Schematic representation of the cfDNA and ctDNA analyses in the cohort of endometrial cancer patients. Consort plot showing the number of patients excluded at each time point of the study
Pre-surgery CfDNA levels have independent prognostic value in endometrial cancer patients
The total cfDNA concentration was evaluated to determine its association with the pathological findings and risk of recurrence. Total cfDNA was isolated from plasma samples (3-5mL) obtained at the time of tumour resection and quantified using Qubit fluorometry. The total cfDNA concentration at surgery ranged between 3.62 and 366.80 ng/mL with a mean value at 21.94 ng/mL and a median of 15.12 ng/mL. Higher levels of cfDNA correlated with traditional high risk of recurrence markers, although statistically significance was only found for myometrial and lympho vascular infiltration (Mann Whitney test p-value > 0.05) (Fig. 2A-H). Accordingly, significantly higher pre-surgery cfDNA levels were found in patients who showed disease recurrence or died because of disease (Mann Whitney test p-value < 0.01) (Fig. 2I-J). An optimal cut-off at 25ng/mL was determined using on the RegParallel package to group patients into high or low cfDNA levels and to explore the utility of cfDNA as a predictor of clinical outcome. Following this strategy 20.70% (41/198) of patients showed high pre-surgery cfDNA levels (Fig. 2J). These patients had a significantly shorter DFS (Log-rank test p-value < 0.0001; HR = 3.91; 95% CI [2.04–7.51]) and DSS (Log-rank test p-value < 0.0001; HR = 6.54; 95% CI [2.83–15.10]) than those with low levels of pre-surgery cfDNA (Fig. 2L-M, respectively). In addition, multivariant analyses showed that cfDNA levels had independent prognostic value to predict DFS (Log-rank test p-value = 0.008; HR = 2.98; 95% CI [1.35–6.61]) and DSS (Log-rank test p-value < 0.001; HR = 9.13; 95% CI [2.82–29.50]) (Supplementary Table 3). Moreover, the correlation between cfDNA levels and standard blood biomarkers used clinical for follow-up, such as CA-125 or CEA, was analysed in subset of the cohort. Importantly, cfDNA levels did not correlate with CA-125 and CEA levels (Spearman R < 0.1) (Supplementary Fig. 1A-C).
The value of pre-surgery cfDNA in identifying patients with poor clinical outcome. A-J. Violin plots of the pre-surgery cfDNA levels (Log10 ng/mL) according to the clinicopathologic variables of the tumours. Statistical significance was evaluated by based on Mann–Whitney U test **p < 0.01. K. Classification of the patients as low or high pre-surgery cfDNA based on the optimal cut point (25 ng/mL). L-M. Kaplan Meier curves showing DFS (L) and DSS (M) based on pre-surgery cfDNA levels. Univariate Cox proportional-hazards model was used to estimate HR and log-rank test to report p-value
No correlation between cfDNA levels and leucocytes count was found, although these blood cells has been reported as the main responsible for the plasma cfDNA content in other tumour types [29]. Furthermore, to ensure that the value of cfDNA and ctDNA were not biased towards tumour size or volume we performed a comparison of cfDNA levels and ctDNA positivity with tumour length (Supplementary Fig. 1D-F, respectively) and with tumour volume (Supplementary Fig. 1E-G, respectively), finding no differences between these variables.
ctDNA as a minimally invasive prognostic tool in endometrial cancer patients
The levels of ctDNA were determined in a total of 177 patients using personalized ddPCR assays based on the mutational profile found in the UA and 52 (29.38%) of them showed detectable levels of ctDNA (Fig. 3K) with a variant allelic frequency (VAF) in a range from 0.01 to 39.10%, an average of 4.08% and a median of 0.44%. Pre-surgery ctDNA positivity was significantly associated with higher levels of cfDNA (Mann Whitney, p-value < 0.01) (Supplementary Fig. 2A) being this association partially explained by a lower cfDNA input used for the ddPCR in patients with low cfDNA (Supplementary Fig. 2B). However, pre-surgery cfDNA levels and ctDNA VAF were not correlated in those patients with detectable ctDNA (Spearman R ≤ 0.2) (Supplementary Fig. 2C-D). Higher detection rates and VAFs were observed in tumours with clinico pathological features of high risk, more specifically in patients with high grade, FIGO III-IV, over 50% myometrial infiltration or LVSI (Mann Whitney, p-value < 0.01) (Fig. 3A-H) (Supplementary Table 2), confirming that higher risk EC tumours shed into circulation higher ctDNA contents.
The value of ctDNA analyses in endometrial cancer. A-J. Box plots showing the highest variant allelic frequency (VAF %) of the alterations found in the ctDNA accordingly the patient clinical variables. Statistical significance was assessed based on Mann–Whitney U test **p < 0.01, **p < 0.001, ****p < 0.0001. K. Percentage of patients with positive and negative ctDNA levels. L-M. Kaplan Meier curves showing DFS (L) and DSS (M) in patients with positive vs. negative levels of ctDNA. Univariate Cox proportional-hazard model was used to estimate HR and log-rank test to report p-value
With the aim to understand if pre-surgery ctDNA can provide additional information to predict the disease prognosis we grouped patients in positive and negative for ctDNA presence and performed survival analyses. Of note, patients with detectable levels of ctDNA showed significant shorter DFS (Log-rank test p-value < 0.001; HR = 3.63; 95% CI [1.80–7.30]) and DSS (Log-rank test p-value < 0.01; HR = 3.91; 95% CI [1.57–9.74]) when compared to patients with undetectable ctDNA (Fig. 3L-M, respectively, Supplementary Table 4).
Combinatory analysis of cfDNA and ctDNA identifies patients with worst clinical outcomes
Since cfDNA and ctDNA levels independently provide prognostic information, we aimed to explore whether combining both we could improve the identification of patients at higher risk of EC recurrence. To this end, patients were stratified according to cfDNA levels and the presence/absence of ctDNA at surgery. Using this approach 4 groups were set up: ‘Group 1’ cfDNA-low/ctDNA-negative (58.76%); ‘Group 2’ cfDNA-low/ctDNA-positive (20.34%); ‘Group 3’ cfDNA-high/ctDNA-negative (11.86%) and ‘Group 4’ cfDNA-high/ctDNA-positive (9.04%). The clinical characteristics of the patients included in each group are summarised on Supplementary Table 5.
Patients in group 4 (cfDNA-High/ctDNA-positive) had the worst results in terms of DFS (Log-rank test p-value < 0.0001; HR = 9.25; 95% CI [4.49–19.10]) and DSS (Log-rank test p-value < 0.0001; HR = 11.20; 95% CI [4.73–26.60]) when compared to the remaining groups (Fig. 4A-B, respectively). Patients in group 3 (cfDNA-high/ctDNA-negative) showed poorer survival rates than patients in group 1 (cfDNA-low/ctDNA-negative) but similar with the group 2 (cfDNA-low/ctDNA positive) (Supplementary Fig. 3A-B). These results could be associated with the presence of very low ctDNA levels in patients included in group 3 that were not detected due to the limitations of the ddPCR approach, although cfDNA high levels indicate an aggressive disease.
Combined analyses of cfDNA and ctDNA identify the patients with the worst clinical outcome. A-B. Kaplan Meier curves showing DFS (A) and DSS (B) in patients according to the pre-surgery high levels of cfDNA and detectable levels of ctDNA. C. Bar plot with the early recurrence status according to the combinatorial approach and the ESGO risk classification D. Graphical representation of the univariate (blue) and multivariate (red) Cox proportional-hazard models. P-value > 0.05 is represented with the *symbol. E-F. Kaplan-Meier curves showing DFS in patients according to the pre-surgery high levels of cfDNA and detectable levels of ctDNA in patients with low or intermediate (E) and high-intermediate or high (F) risk of recurrence based on the ESGO-ESTRO-ESP risk stratification
Notably, 75% of the patients included in group 4 (characterized by high cfDNA and ctDNA positivity) had a disease relapse, and 69% of them died during the follow-up as a result of the disease. Of note, 73% these patients had a relapse within the first year after the surgery while only the 30% of tumours classified as high-intermediate and high risk of recurrence, according to the latest ESGO-ESTRO-ESP risk classification, showed an early relapse (Fig. 3C). Importantly, this combinatory approach showed independence over the traditional risk factors and molecular subtype (Fig. 3D). Besides, this combinatory approach remains clinically significant when stratifying patients based on or according to histology, grade or FIGO stage (Supplementary Fig. 3C-F). Importantly, the presence of high levels of pre-surgery cfDNA and ctDNA positivity in patients classified as low or intermediate risk based on ESGO-ESTRO-ESP criteria was associated with a quick relapse in three cases, although most of the patients had a good prognosis (Fig. 4E). And notably, patients classified as high-intermediate/high risk of recurrence based on ESGO-ESTRO-ESP criteria and with high levels of pre-surgery cfDNA and the ctDNA positivity showed a very aggressive disease (Fig. 4F). Therefore, the analysis of liquid biopsy clearly complements the current tools to anticipate disease relapse (Supplementary Fig. 3G-H).
Combination of risk classification and cfDNA/ctDNA to predict the patients’ outcome
We combined the current risk stratification tools with the risk groups derived from the liquid biopsy analyses. With this strategy we considered a patient in the group of poor prognoses if she has a high-intermediate or high-risk tumour (ESGO-ESTRO-ESP criteria) or high cfDNA/ctDNA positivity at surgery. This approach identified 54.23% (96/177) of the cohort as poor prognoses. With this approach the HRs of the poor prognosis group associated with the DFS (Log-rank test p-value < 0.0001; HR = 7.91; 95% CI [2.39–26.20]) and DSS (Log-rank test p-value < 0.0001; HR = 16.10; 95% CI [2.14–120]) were even more prominent than when analyse this classification strategies independently (Supplementary Fig. 4A-B, respectively). Moreover, 90% and 95% of the patients who underwent disease recurrence and died because of the disease respectively, were classified as high-risk patients thanks to the inclusion of the liquid biopsy in the analysis.
CfDNA and ctDNA as a monitoring tool for EC
To explore the value of cfDNA and ctDNA monitoring as a surrogate of the EC burden, a total of 372 longitudinal blood samples were analysed in a subset of 130 patients. From these patients, 22 showed disease progression. Significant reduction on the cfDNA levels were found after surgical resection at 1, 6 and 24 months (Wilcoxon signed-rank test, p-value < 0.05, Supplementary Fig. 5A-F), although the dynamics of cfDNA levels in the longitudinal samples did not show value to anticipate the disease reappearance in our cohort of patients.
Notably, the analyses of the specific tumour fraction through longitudinal samples allowed for the identification of the disease recurrence months before (4.68 ± 2.98) the clinical confirmation of relapse, mainly in patients with detectable pre-surgery ctDNA (80%, 8/10, Fig. 5A). However, in 4 (18.18%) of the 22 patients with recurrent disease, ctDNA was not detectable with the ddPCR approach (Supplementary Fig. 5G). Most of these patients were also ctDNA negative at surgery, showing the need to improve the sensitivity of the ctDNA detection to monitor low-shedding tumours. In addition, only 3 of the 37 samples analyzed one month after surgery were positive. Accordingly in these patients the surgery was not radical, and the presence of residual disease was known by the gynaecologists.
Longitudinal analyses of cfDNA/ctDNA allow for early detection and accurately reflect the disease kinetics. A. Swimmer plot of the 18 patients that underwent tumour progression divided based on the combinatorial approach with longitudinal samples collected at least 6 months prior to the relapse (cfDNA and ctDNA). B-D. Example figures of the cfDNA (blue dotted line) and ctDNA kinetics (yellow line) in patients with advanced disease. (1) Radiotherapy, (2) Carboplatin-Paclitaxel, (3) Dostarlimab, (4) Exemestane, (5) Doxorrubicin and Avastin, (6) Lenvatinib and Pembrolizumab, (7) Topotecan and Bevacizumab
Importantly, longitudinal analyses of ctDNA proved to be a powerful tool to identify patients undergoing an early relapse, as reflected the patient #1 described in Fig. 5B. This patient was diagnosed with a FIGO stage IB endometrioid tumour that was positive for a pathogenic mutation in FGFR2 in pre-surgical cfDNA (0.06%), which was found increased 5 months after surgery (4.52%) and clinically y confirmed two months later as an abdominal recurrence (18%). The patient started to receive Dostarlimab (a PD-1 blocker) and the ctDNA strongly decreased (0.38%) in line with the partial response defined based on CT-Scan. The patient is currently in response (0%) and is being monitored by means of the ctDNA along with the imaging.
Furthermore, longitudinal analyses also allow for the dynamic characterization of the disease in response to therapy pressure (Fig. 5C-D). For example, in Fig. 5C shows the case of a patient that was diagnosed with high-grade serous histology of EC and FIGO IIIA. NGS analyses of the UA showed alterations within the PPP2R1A and TP53 genes and they were followed through the course of the disease. The patient had high levels of cfDNA at surgery as well as detectable levels of ctDNA. According to clinical guidelines the patient was treated with carboplatin-paclitaxel combination therapy and radiotherapy. Afterwards, the ctDNA levels were measured again and still detectable levels were found, indicating persistence of the disease (0.29%). Shortly thereafter, the patient was confirmed to have a liver recurrence by CT-scan, which showed a spike in ctDNA levels (16.90%). Following the relapse, the patient was treated with a second-line chemotherapy. Although there was an initial reduction on ctDNA levels, they started to increase again and the patient showed progressive disease with peritoneal affectation and entered PS-ECOG 4 and could no longer be treated.
Another example is shown in Fig. 5D, a patient diagnosed with a mixed histology phenotype (initially diagnosed as endometrioid), grade 2, FIGO IB tumour that showed high levels of cfDNA and ctDNA at surgery. After a year, patient showed symptoms compatible with a disease relapse at the lungs. At this moment ctDNA was positive confirming the recurrence of the disease. Due to the nature of the relapse the patient was closely monitored throughout the course of the disease, and the cfDNA/ctDNA kinetics reflect the evolution of the disease and the response to therapy, allowing the detection of disease recurrence prior to clinical evidence. Thanks to this approach, clinicians have been able to adjust the treatment and anticipate CT-scans according to the tumour kinetics. It is important to note that traditionally clinical variables identified this patient as being of intermediate risk, but our combinatory approach of cfDNA and ctDNA analyses classified it as of being of high-risk of recurrence, reinforcing the additional value of liquid biopsy to anticipate the disease relapse.
Add Comment