High-sensitivity mutation analysis of cell-free DNA for disease monitoring in endometrial cancer

Abstract

Purpose:

We sought to determine whether sequencing analysis of circulating cell-free (cf)DNA in prospectively accrued endometrial cancer (EC) patients captures the mutational repertoire of the primary lesion and allows for disease monitoring.

Design:

Peripheral blood was prospectively collected from 44 newly diagnosed EC patients over a 24-month period (i.e., baseline, post-surgery, every six months after). DNA from the primary ECs was subjected to targeted next-generation sequencing (NGS) of 468 cancer-related genes, and cfDNA to a high-depth NGS assay of 129 genes with molecular barcoding. Sequencing data were analyzed using validated bioinformatics methods.

Results:

cfDNA levels correlated with surgical stage in ECs, with higher levels of cfDNA being present in advanced stage disease. Mutations in cfDNA at baseline were detected pre-operatively in 8 of 36 (22%) patients with sequencing data, all of whom were diagnosed with advanced stage disease, high tumor volume and/or aggressive histologic type. Of the 38 somatic mutations identified in the primary tumors also present in the cfDNA assay, 35 (92%) and 38 (100%) were detected at baseline and follow-up, respectively. In six patients with recurrent disease, changes in circulating tumor (ct)DNA fraction/ variant allele fractions in cfDNA during follow-up closely mirrored disease progression and therapy response, with a lead time over clinically-detected recurrence in two cases. The presence of ctDNA at baseline (p<0.001) or post-surgery (p=0.014) was significantly associated with reduced progression-free survival.

Conclusions:

cfDNA sequencing analysis in EC patients at diagnosis has prognostic value and serial post-surgery cfDNA analysis enables disease and treatment response monitoring.

INTRODUCTION

Endometrial cancer (EC) is the fourth most common female malignancy in the United States with more than 65,000 new cases estimated in 2022 (1). Despite advances in therapy, over the course of 2013–2017, uterine cancer death rates have increased more than any other type of cancer among women (2), and when compared to 1975, the overall 5-year relative survival for EC has decreased (3). EC is histologically, clinically and molecularly heterogeneous (4). The majority of low-risk histology, early-stage disease can be treated with primary surgery and carries a favorable prognosis (5), whereas serous carcinomas and carcinosarcomas, representing only about 15% of all ECs, account for the majority of deaths. In these high-risk ECs, the efficacy of standard therapeutic regimens (i.e. paclitaxel/ carboplatin) (6,7), is limited, and novel targeted therapies are emerging (8). Other than radiologic imaging and serum CA125 level assessment in some patients, currently there is no non-invasive form of treatment response or disease progression monitoring available for patients with EC.

Sequencing analysis of cell-free DNA (cfDNA) has emerged as a potential biomarker for early cancer detection, diagnosis, disease monitoring, and prognostication (9–11), and has been well studied in patients with early and advanced disease in a number of cancer types, including breast, colorectal and lung cancer (12–14). Unlike those solid tumors, distant metastatic disease in EC is not common (approximately 4%) (15). To date, the role of cfDNA sequencing analysis in women with EC is less defined. Small gene panels assessed by digital or quantitative PCR have identified circulating cell-free tumor (ct)DNA in subsets of patients with EC, often those with high-risk disease (16–21); these studies were retrospective, however, and only provided limited longitudinal information. Next generation sequencing (NGS) of hotspot amplicons of CTNNB1, KRAS, PTEN and PIK3CA in plasma sampled at hysterectomy from patients with endometrioid EC revealed mutations in 33% of patients, and found to be associated with advanced stage, deep myometrial/vascular invasion, and primary tumor size (22).

In this proof-of-principle study, we sought to define whether i) the presence of cfDNA is associated with the histologic subtype or disease stage in newly diagnosed EC patients, ii) whether sequencing analysis of longitudinal cfDNA samples can be used for disease monitoring in this disease and iii) whether the presence of ctDNA in plasma derived from EC patients is associated with disease progression or recurrence. To address these aims, we performed a retrospective analysis of prospectively collected plasma samples from patients with newly diagnosed EC using a high-sensitivity NGS-based mutation profiling assay.

MATERIAL AND METHODS

Patient selection

In this prospective proof-of-principle study, 44 biopsy-confirmed newly diagnosed EC patients consented to an institutional research protocol between 09/2017 and 07/2018 were included (Supplementary Figure S1, Consort Diagram). Rather than a consecutive series, cases representative of all histologic types and FIGO stages were included to assess the presence of mutations in cfDNA according to histologic type and stage. All surgical specimens were subjected to a central pathology review by a dedicated gynecologic pathologist (L.H.E.) to i) define the histologic type following the World Health Organization of Female Genital Tumors (23), ii) assess lymphovascular invasion, iii) confirm the presence of isolated tumor cells (ITCs) in the lymph nodes, and iv) confirm the FIGO staging. Clinical data, including demographic information, surgical staging, chemotherapy, radiation treatment, imaging, and follow-up were extracted from the electronic medical records from the date of patient consent. This study was approved by Memorial Sloan Kettering Cancer Center’s (MSK’s) institutional review board (IRB), and written informed consent was obtained from all patients. This study was conducted in accordance with the Declaration of Helsinki.

Liquid biopsies

Peripheral blood samples were collected from each patient in two Streck Cell-Free DNA Blood Collection Tubes (Streck, La Vista, NE). Samples were collected prior to surgery, at the post-operative visit (median 22 days post-surgery, range 9–57 days), and at 6, 12, 18, and 24 months. Blood was processed and cfDNA extracted using the QIAsymphony SP system (Qiagen), quantified, and stored following validated standard operating procedures at MSKs cfDNA extraction laboratory in the Department of Pathology and Laboratory Medicine, as previously described (24).

MRI-based FIGO staging and tumor volume analysis

Pretreatment magnetic resonance imaging (MRI) scans were obtained on a 3 Tesla magnet (GE Healthcare, Chicago IL) in accordance with guidance from the European Society of Urogenital Radiology (ESUR) (25). A board-certified radiologist (Y.L.) reviewed all MR images for the presence of tumor and, if present, assigned a preliminary MRI-based FIGO stage following the FIGO classification system, and according to the rules laid out in the ESUR guidelines (25,26). Briefly, tumor was considered present if MR images demonstrated an intermediate T2 signal intensity (SI) lesion with diffusion restriction (high SI on high b-value DWI and low SI on ADC map), and hypoenhancement compared to myometrium on later phase DCE. The MRI-based FIGO stages are detailed in the Supplementary Methods. In addition to the above qualitative evaluation, two radiologists (J.N. and Y.L.) manually segmented every lesion on each tumor-containing slice using AquariusNet (Terarecon, Foster City, USA) to automatically generate total tumor volume for each patient.

Tumor and cfDNA NGS sequencing analysis

Primary ECs underwent clinical tumor-normal targeted DNA NGS of 468 cancer-related genes using MSK-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) (27,28). NGS was performed on the primary tumor specimen in all but one case; for CD23, NGS was performed on a diagnostic biopsy (Supplementary Table S1). In addition, DNA was extracted from microdissected extra-uterine tumor sites from 6 cases who presented with stage III or stage IV disease and subjected separately to MSK-IMPACT sequencing at MSK’s Integrated Genomics Operation (IGO), as described (29). The median depth of coverage for tumor samples was 618x (range 101x-1030x) and for normal samples was 504x (range 325x-729x). MSK-IMPACT sequencing data were analyzed using previously described bioinformatics approaches (27,30). Somatic mutations identified in the primary tumor or extra-uterine site from a given patient were interrogated in the matched respective primary tumor/ extra-uterine site by manual inspection of BAM files using mpileup files (SAMtools mpileup; v1.2 htslib 1.2.1) (SAMTOOLS, RRID:SCR_002105) (31), as previously described (30). Primary ECs were stratified into The Cancer Genome Atlas (TCGA) molecular subtypes (32), as previously described (28).

For each patient, cfDNA obtained at baseline and post-surgery was subjected to MSK-ACCESS (Analysis of Circulating cfDNA to Examine Somatic Status), a high-depth NGS assay with molecular barcoding technology for the detection of very low frequency somatic alterations in 129 genes, a subset of the 468 MSK-IMPACT genes (33). Of the 44 ECs, 36 had a pre- and/or post-operative blood draw and the required plasma cfDNA input of 10–20ng cfDNA for MSK-ACCESS sequencing (Supplementary Figure S1 and Table S1). In addition, longitudinal cfDNA samples from 25 patients with sufficient cfDNA and/or detectable mutations at baseline or post-surgery and/or disease progression were also subjected to MSK-ACCESS sequencing (Supplementary Table S1). cfDNA samples were sequenced to a median raw coverage of 25,611x (range 13,295 – 58,743x); post collapsing, the median duplex coverage was 1,340x (range 212x – 2,839x). MSK-ACCESS raw sequence data were demultiplexed and processed as previously described (33) (Supplementary Methods). Variant calling was performed in a matched tumor-informed manner (“genotyping”) using Waltz (https://github.com/mskcc/Waltz), and required at least 2 duplex consensus reads, comprising both strands of DNA, to call a somatic single nucleotide variant (SNV) or short insertion and deletion (indel) at a site known to be altered in the matched tumor sample from a given patient, as previously described (33). Mutation-specific ctDNA fraction values were computed for all somatic mutations determined to be clonal in the tissue sample using MSK-IMPACT and covered in the MSK-ACCESS liquid biopsy panel (Supplementary Methods).

Statistical analysis

Comparisons between cfDNA concentrations and clinicopathologic parameters were performed using the Wilcoxon rank sum test. Progression-free survival (PFS) was defined from the date of surgery to the time of recurrence/progression or death or last follow-up whichever occurred first. The median PFS and PFS rates were estimated applying the Kaplan-Meier method. The association between somatic mutation in pre- and post-surgery cfDNA with PFS was examined using Log-rank test with permutation (34). To accommodate the time dependent nature of post-surgery cfDNA mutation, a landmark analysis was performed using the landmark time of 51 days (or 1.68 months) which is the longest time interval between surgery date and post-surgery cfDNA measurement date in this cohort. A p-value of p<0.05 was considered statistically significant. All statistical analyses were performed using R version 4.1.1 (https://cran.r-project.org/) (CRAN, RRID:SCR_003005).

DATA AVAILABLITY

Clinicopathologic information and the somatic mutations identified in the primary tumors and cfDNA are available in Supplementary Tables S1–S3. Targeted sequencing data that support the findings of this study is available at cBioPortal [www.cbioportal.org; Endometrial Carcinoma cfDNA (MSK, Clin Cancer Res 2022)].

RESULTS

EC patient population and association with cfDNA

In this study, 77% (34/44) of EC patients had FIGO stage I disease and 23% (10/44) had advanced-stage disease at diagnosis (stages III and IV; Table 1). As expected, ECs of endometrioid histology were the most common (24/44, 55%), followed by serous carcinoma (7/44; 16%) and carcinosarcoma (6/44; 14%; Table 1). All but one EC patient underwent primary surgery; half of the patients (21/43; 49%) received no further treatment, and the remaining patients were treated with either radiation and/or chemotherapy following surgery (Supplementary Table S1). Despite a subset of the 18- and 24-month timepoints occurring during the COVID-19 pandemic, 229 of the intended 258 (89%) longitudinal blood draws were collected. Median length of clinical follow-up was 33 months (range 0.3–44), with 11 patients being diagnosed with recurrent or persistent disease (Supplementary Table S1).

Table 1:

Histologic type, stage, tumor volume and cfDNA concentration of endometrial cancer patients included this study.

	n (%)	Mean tumor volume cm3 (Std Dev)	MSK-IMPACT sequencing primary tumor n (%)	Mean cfDNA concentration pre-op/baseline ng/ml (Std Dev)	MSK-ACCESS sequencing cfDNA pre-op/baseline n (%)
Surgical stage	44		42		33
Stage IA	30* (68%)	-	28* (67%)	2.68 (1.94)	19* (58%)
Stage IB	4* (9%)	-	4* (10%)	4.05 (2.29)	4* (12%)
Stage II	0 (0%)	-	0 (0%)	-	0 (0%)
Stage III	8 (18%)	-	8 (19%)	4.78 (2.56)	8 (24%)
Stage IV	2 (5%)	-	2 (5%)	11.61 (12.74)	2 (6%)
MRI stage	37	105.2 (446)	35		24
No measurable disease	7 (21%)	0 (0)	5 (14%)	3.03 (1.76)	4 (17%)
Stage IA	18 (53%)	14.4 (26)	18 (51%)	2.91 (1.89)	11 (46%)
Stage IB	3 (9%)	42.2 (47)	3 (9%)	6.01 (3.82)	3 (13%)
Stage II	1 (3%)	12.6 (NA)	1 (3%)	3.10 (NA)	1 (4%)
Stage III	3 (9%)	171.7 (128)	3 (9%)	4.06 (2.91)	3 (13%)
Stage IV	2 (6%)	1332.5 (1804)	2 (6%)	11.61 (12.74)	2 (8%)
Histologic type	44		42		33
Endometrioid	24 (55%)	28.7 (73)	23 (55%)	2.90 (1.69)	16 (48%)
Serous	7 (16%)	16.0 (27)	7 (17%)	4.01 (2.45)	6 (18%)
Carcinosarcoma	6 (15%)	671.6 (1291)	6 (14%)	5.52 (7.63)	5 (15%)
High-grade carcinoma	4 (9%)	13.8 (20)	3 (7%)	3.58 (2.21)	3 (9%)
Mixed	1 (2%)	7.7 (NA)	1 (2%)	7.47 (NA)	1 (3%)
De-differentiated	1 (2%)	95.4 (NA)	1 (2%)	8.89 (NA)	1 (3%)
Adenosarcoma	1 (2%)	111.0 (NA)	1 (2%)	2.07 (NA)	1 (3%)

^*One stage IA and two stage IB cases had isolated tumor cells in the lymph nodes on pathology. cfDNA: cell-free DNA; NA, not available; Std Dev, standard deviation.

Median cfDNA concentration in pre-operative cfDNA samples across cases was 2.71 ng/mL (range 0.66–20.62; Supplementary Table S1). We found EC patients with stage III/IV disease at diagnosis to have significantly higher amounts of cfDNA compared to those with stage I disease (Wilcoxon rank sum test, p=0.028; Figure 1A). In addition to surgical stage, MRI stage (25) was assigned based on a pre-operative MRI in 34/44 (77%) patients, which aligned with final surgical stage in 29/34 patients (Table 1, Supplementary Table S1). cfDNA concentrations were found to be numerically higher in histologic types associated with an aggressive clinical behavior, including carcinosarcoma and high-grade carcinomas of no specific type, however this association was not statistically significant likely due to the small sample size in some of the groups (p>0.1; Figure 1B).

Figure 1. Baseline characteristics of newly diagnosed endometrial cancers.

A, cfDNA concentrations in ng/ml plasma at baseline in newly diagnosed endometrial cancers according to surgical stage. Wilcoxon rank sum test. B, cfDNA concentrations in ng/ml plasma at baseline in newly diagnosed endometrial cancers according to histologic type. HG Carc, high-grade carcinoma; Dediff, dedifferentiated; Carcinosarc, carcinosarcoma. Wilcoxon rank sum test was used for comparisons. C, somatic mutations identified in primary newly diagnosed endometrial cancers included in this study. Mutation types are color coded according to the legend. Information on the histologic subtype, tumor grade of endometrioid endometrial cancers, and molecular subtype are provided on top.

The primary ECs of 42 patients were subjected to clinical tumor-normal FDA-authorized MSK-IMPACT sequencing of 468 cancer-related genes; for two cases there was no/ insufficient tumor for molecular profiling (Supplementary Figure S1). A median of 7.5 non-synonymous mutations were detected (range 1 – 92; Supplementary Table S2). Consistent with previous reports (32), the most commonly mutated genes were TP53 (52%), PTEN (50%), PIK3CA (50%), ARID1A (36%), and FBXW7 (29%; Figure 1C, Supplementary Table S2). Of all 42 cases included, ten ECs (24%) were of microsatellite instable (MSI) molecular subtype, 12 (29%) of copy number-low molecular subtype and 20 (48%) of copy number-high molecular subtype (Figure 1C). Of the 23 endometrioid ECs, 11 (48%) were of copy number-low, 9 (39%) of MSI, and 3 (13%) of copy number-high molecular subtype (Figure 1C).

Landscape of somatic mutations in baseline cfDNA

We next sought to define whether the landscape of somatic mutations in baseline/ pre-operative cfDNA would be representative of that identified in the respective primary ECs. A high-depth UMI-based sequencing assay (MSK-ACCESS) revealed the baseline presence of tumor-derived mutations in cfDNA in eight of 36 cases subjected to MSK-ACCESS sequencing (22%), six of which had stage III or stage IV disease (Figure 2, Supplementary Figures S1 and S2, Supplementary Table S1). ctDNA was also detected in a stage 1A high-grade adenosarcoma and a stage 1B high-grade mixed endometrioid/ serous carcinoma with isolated tumor cells (ITCs) and lymphovascular space invasion (Figure 2, Supplementary Figure S2), however none of the low-grade stage I endometrioid EC patients had detectable mutations in their pre-operative cfDNA. Of these 8 ECs with ctDNA being present in plasma at baseline, 6 displayed lymphovascular space invasion (LVSI), which was not significantly different from ECs without detectable ctDNA at baseline (5/25; Fisher’s exact test, p=0.137; Supplementary Table S1). When defining the cancer cell fractions of the mutations identified (i.e., bioinformatically inferred percentage of tumor cells harboring the mutation in a given sample), we observed that all mutations that were clonal in the primary tumor and present in the MSK-ACCESS assay were also detected in baseline/ pre-operative cfDNA (Figure 2, Supplementary Figure S2). The only exception was case CD28, a stage IB mixed EC with ITCs, where a clonal CDKN2A E61Pfs*54 and a TERT promoter C228T mutation present in the primary tumor were not detected in the plasma at baseline, but were detected at later time points (see below), whereas a subclonal TP53 mutation was (Figure 2, Supplementary Figure S2).

Figure 2. Mutations detected in cfDNA at baseline of newly diagnosed endometrial cancers.

In eight patients with newly diagnosed endometrial cancer, mutations present in the primary tumor were detected in plasma at baseline using a high-sensitivity sequencing assay (MSK-ACCESS). For each case, information on the histologic type, surgical stage, MRI-based tumor burden/ volume and circulating tumor (ctDNA) fraction is provided on top. The cancer cell fractions (CCFs) of the somatic mutations identified in the primary tumor (T) and extra-uterine sites are shown on the left, and the mutation types in the primary tumor (T), extra-uterine sites and cfDNA (cf) on the right, color-coded according to the legend. Only genes/ mutations that are present in both, the tumor sequencing assay (MSK-IMPACT) and the cfDNA sequencing assay (MSK-ACCESS) are shown (see Supplementary Figure S2 for all alterations). FT, fallopian tube; ITCs, isolated tumor cells in the lymph nodes; LN, left hypogastric lymph node; LLN¹, left external iliac sentinel lymph node; LLN², right exterior iliac sentinel lymph node; MP, mesenteric implant; SLN¹, left exterior iliac sentinel lymph node; SLN², obturator sentinel lymph node.

In four stage III/IV cases (i.e., CD20, CD22, CD26 and CD36) with detectable mutations in the cfDNA at baseline, in addition to the primary EC, DNA from extra-uterine sites, including (sentinel) lymph nodes, peritoneal implants and fallopian tube, was extracted and subjected to MSK-IMPACT sequencing. Heterogeneity between the primary tumor and extra-uterine sites was observed, with mutations identified in the primary EC not being present in the respective lymph nodes, whereas some lymph nodes/ peritoneal implants harbored additional mutations that were not detected in the respective primary lesions (Supplementary Figure S2). In case CD20, the primary tumor, the left exterior iliac sentinel lymph node (SLN) and obturator SLN shared a set of mutations, including an ARID1A truncating mutation and PIK3CA and KRAS hotspot mutations that were subclonal in the primary tumor but were found to become clonal in the sentinel lymph nodes (Figure 2; Supplementary Figure S2). A clonal PTEN mutation and RNF43, TSC1 and SMARCA4 frameshift mutations were, however, only shared between the primary EC and left exterior iliac SLN but were absent in the obturator SLN. In turn, AGO2 and FOXA1 missense mutations were restricted to the two SLNs. In case CD22, the primary EC and the mesenteric implant and three (sentinel) lymph nodes shared hotspot PIK3CA, TP53 and PPP2R1A mutations. An ARID1A truncating mutation was present in the primary lesion and three of the four extra-uterine sites as well as in cfDNA (Figure 2; Supplementary Figure S2). Whereas the primary EC and the fallopian tube lesion of case CD36 shared all somatic mutations identified, in case CD26, while FBXW7, NTRK3 and PIK3CA mutations were shared between all sites and cfDNA, a different PIK3CA hotspot mutation and a TP53 splice site mutation were restricted to the primary tumor. All mutations, however, were detected in the pre-surgery cfDNA. Sequencing analysis of the liquid biopsy at baseline was generally more representative of the repertoire of somatic mutation in the primary EC than that of the (sentinel) lymph nodes (Figure 2).

The ctDNA fraction in plasma obtained at baseline varied and was between 0.7% and 2.0% in the majority of cases (i.e., CD11, CD20, CD26, CD28, CD36 and CD37; Figure 2, Supplementary Table S3). The ctDNA fraction in pre-operative plasma was elevated in two cases, however. In CD22, a stage IVB endometrial serous carcinoma, the ctDNA fraction was 5.8%, and in CD23, a stage IVB carcinosarcoma with a very high tumor burden (2,608cm³) the ctDNA fraction was 74.6%; both patients died from their disease 31 and 30 months after the baseline sample, respectively (Figure 2; Supplementary Tables S1 and S3).

Mutation tracking in serial cfDNA samples during follow-up

Next, we sought to define whether cfDNA sequencing of serial plasma DNA samples could be employed for disease monitoring including the assessment of treatment response and identification of early disease recurrence. For this, blood samples were collected every 6 months during follow-up; for 25 patients with either recurrent disease and/or detectable post-surgery ctDNA in plasma, the longitudinal cfDNA samples were subjected to MSK-ACCESS sequencing. In six of these 25 EC patients (24%), ctDNA was detected in the serial blood samples (Supplementary Table S1). The change in the mutant allele fraction (MAFs) of the mutations identified as well as of the ctDNA fraction during follow-up closely followed disease progression and therapy response (Figures 3–5, Supplementary Figures S3 and S4).

Figure 3. Longitudinal disease monitoring by sequencing of plasma-derived cfDNA samples from patients CD20 and CD26 with newly diagnosed endometrial cancer.

A, Case CD20, B, Case CD26 with detectable pre-operative circulating tumor (ct) DNA and corresponding graphs of mutation allele frequencies (MAFs) according to the time of blood collection, treatment timeline and a pictographic representation of disease burden. MAFs of mutation in plasma DNA (left y-axis) extracted from peripheral blood samples obtained at baseline, post-surgery and serially every 6 months during follow-up, color-coded according to the legend. The ctDNA fraction (right y-axis) is shown by a dashed line. Below the graph, information about surgery, radiation therapy, chemotherapy, last follow-up, and schematic radiographic representations of disease burden are shown. AWD, alive with disease; DOD, dead of disease; EC, endometrial cancer; F/U, follow-up; MAF, mutant allele fraction.

Figure 5. Association of baseline ctDNA and disease progression.

A, Longitudinal disease monitoring of case CD36. MAFs of mutation in plasma DNA (left y-axis) extracted from peripheral blood samples obtained at baseline, post-surgery and serially every 6 months during follow-up, color-coded according to the legend. The ctDNA fraction (right y-axis) is shown by a dashed line. Below the graph, information about surgery, chemotherapy, last follow-up, and schematic radiographic representations of disease burden are shown. AWD, alive with disease; F/U, follow-up, MAF, mutant allele fraction. B, Progression-free survival according to the detection of ctDNA in the baseline/ pre-surgery plasma. P value determined by log rank test. C, Swimmer plot depicting disease progression in newly diagnosed endometrial cancer patients according to the detection of ctDNA at baseline/ pre-surgery and stage.

In case CD20, a stage IIIC2 dedifferentiated carcinoma, a marked increase in the MAFs was observed following the second course of chemotherapy (i.e., 8–9 months post diagnosis). Imaging at that time revealed increased pelvic, supraclavicular and thoracic lymph nodes (Figure 3A). The overall ctDNA fraction and the allele fractions of the 10 mutations detected in plasma decreased during another course of chemotherapy and a clinical response with decreased size of lymph nodes was observed at 18 months. Mirroring the steep increase in MAFs/ ctDNA fraction between 18 and 24 months, imaging at 24 months revealed increased metastatic retroperitoneal adenopathy throughout (Figure 3A). Similarly, in case CD26, a stage IIIC1 carcinosarcoma, tumor was cleared during radiation and chemotherapy post-surgery with almost undetectable levels of tumor mutations in cfDNA at 6 months. Following chemotherapy, at the 12 months blood draw, a marked increase in the MAFs of all five mutations detected in the cfDNA, including two PIK3CA and a TP53 mutation, was observed, which was consistent with the increased tumor burden observed in this patient, including increase in lymph node sizes at 13 months and multiple spinal metastases at 15 months (Figure 3B).

In patient CD22, a stage IVB serous carcinoma, a sharp reduction in the MAFs of all four somatic mutations (i.e., ARID1A, PIK3CA, TP53 and PPP2R1A mutations) was detected post-surgery and chemotherapy. At 12 months, ctDNA levels increased, consistent with new lung nodules and increased perirectal metastases observed by imaging (Figure 4A). Despite the administration of 3 courses of chemotherapy over the next 12 months, the levels of ctDNA continued to rise; this lack of response aligned with the imaging data, which showed an increase in tumor burden and new splenic nodules (Figure 4A; Supplementary Figure S3). In case CD28, a stage IB grade 3 mixed endometrioid/ serous carcinoma with ITCs, only 3 of the 5 mutations (i.e., two PTEN and a TP53 mutation) present in the primary tumor were detected in ctDNA at baseline (Figure 2). Following the first course of chemotherapy plus radiation, an increase in the ctDNA fraction was observed. At 6 months, also the TERT promoter mutation and CDKN2A p.E61Pfs*54 mutation not detected at baseline were now identified in plasma, whereas there was no evidence of disease by imaging (Figure 4B). A sharp increase in MAFs were observed for six months before the detection of new pulmonary nodules on imaging at 12 months. Also in patient CD28 the MAFs of the mutations kept rising despite additional courses of chemotherapy and radiation (Figure 4B). This lack of response to chemo/ radiation therapy was also found in the imaging studies, which showed increased disease volume and new distant metastases at 12, 18 and 24 months (Figure 4B; Supplementary Figure S3). Whilst all five primary tumor mutations (i.e., FBXW7, PIK3R1, APC, PTEN and ERBB3 mutations) were identified in plasma at baseline in CD36, a patient with stage IIIA grade 1 endometrioid EC, these remained undetectable following surgery and chemotherapy for 18 months, and the patient was rendered disease-free (Figure 5A). We did observe an increase in the MAFs of the mutations at the 30-month blood draw, without any change in tumor burden on imaging until 36 month when the patient was noted to have an increased pelvic mass and sidewall nodule and underwent surgery for the removal of these lesions (Figure 5A). Whilst exploratory and further studies would be required to confirm these findings, these data provide evidence demonstrating that the detection of ctDNA changes can provide a lead time over the radiologically and/or clinically detected recurrence in EC patients.

Figure 4. Longitudinal disease monitoring by sequencing of plasma-derived cfDNA samples from patients CD22 and CD28 with newly diagnosed endometrial cancer and comparison with serum CA125 levels.

A, Case CD22, B, Case CD28 with detectable pre-operative circulating tumor (ct) DNA and corresponding graphs of mutation allele frequencies (MAFs) according to the time of blood collection, treatment timeline and a pictographic representation of disease burden. MAFs of mutation in plasma DNA (left y-axis) extracted from peripheral blood samples obtained at baseline, post-surgery and serially every 6 months during follow-up, color-coded according to the legend. The serum CA125 levels (right y-axis) in U/ml are shown by a dashed line, with <35 U/mL CA125 being the normal range. Below the graph, information about surgery, radiation therapy, chemotherapy, last follow-up, and schematic radiographic representations of disease burden are shown. DOD, dead of disease; EC, endometrial cancer; F/U, follow-up; ITCs, isolated tumor cells in the lymph nodes; MAF, mutant allele fraction; ND, not determined.

In contrast to ovarian cancer, the role of the tumor antigen CA125 for surveillance in EC is less well established. For two of the ECs with detectable mutations in the cfDNA during follow-up, regular CA125 measurements were obtained. In both CD22 and CD28, elevated CA125 levels were detected, however only 5.8 months and 17 months after the increase in MAFs of tumor-derived mutations in cfDNA (Figures 4A and B). This exploratory finding suggests that the detection of ctDNA in plasma may be more sensitive than the CA125 levels in serum of EC patients, warranting further studies.

ctDNA detection and prediction of progression-free survival

We examined whether the detection of ctDNA at different time points would have the potential to predict recurrences in this series of newly diagnosed ECs. Of the eight patients with detectable ctDNA mutations at baseline, all but one (CD37, adenosarcoma) recurred within the time of follow-up (median 33 months, range 0.3–44; Supplementary Table S1). The four EC patients with detectable mutations in cfDNA immediately post-surgery displayed recurrent or persistent disease following surgery (Supplementary Tables S1 and S3). Furthermore, for CD33, a stage IA carcinosarcoma, while no mutations were detected in the baseline/ pre-surgery blood sample, we observed an increase in the MAFs of the mutations at 6 months, the only longitudinal sample, which mirrored the disease recurrence observed by imaging (Supplementary Figure S4 and Table S3). We found that the presence of ctDNA at baseline/ pre-operatively in this series of newly diagnosed ECs was significantly associated with progression-free survival (p<0.001; Hazard ratio (HR) 11.14 (95% confidence interval (CI) 2.72–45.59); Figures 5B and C). Due to the limited number of events, a meaningful multivariate analysis could not be performed. Finally, we performed a landmark analysis showing that the detection of ctDNA immediately post-surgery also to be significantly associated with disease progression (p=0.014; HR 15.56 (95% CI 2.16–112.16), Supplementary Figure S5), however, the number of cases with measurable ctDNA after surgery in this cohort was small.

DISCUSSION

Our findings demonstrate that the amount of cfDNA present in plasma correlates with surgical stage in newly diagnosed EC patients, and that the detection of mutations in cfDNA using a high-sensitivity high-depth liquid biopsy sequencing assay was almost exclusively restricted to EC patients with advanced disease and more aggressive histology types. We have also shown that mutations detected in plasma/ peripheral blood of EC patients are representative of those present in the primary tumor, with 35/38 (92%) of the primary tumor mutations identified in the cfDNA at diagnosis. Whilst there was limited genetic heterogeneity between the primary ECs and extra-uterine sites/ metastases analyzed by MSK-IMPACT sequencing, the mutations detected in cfDNA captured also those present in the extra-uterine sites. Our study provides evidence that ctDNA detection is not only feasible in patients with EC but also that ctDNA, levels closely mirror treatment response and disease progression. Finally, we demonstrate that the detection of mutations in cfDNA, both pre- and post-surgery, correlates with disease progression in women with newly diagnosed EC.

Our study illustrates that even using a high-sensitivity high-depth UMI-based sequencing assay, ctDNA is detected only in a subset of patients with EC and is associated with stage and histologic type (although not significant for histology, probably due to low numbers). Akin to other cancer types (35,36), we found that cfDNA analysis may be useful in the setting of disease monitoring and detection of minimal residual disease (MRD) to predict disease recurrence. Furthermore, although we and others have previously shown that in early breast cancer serial sampling during follow-up was required for accurate MRD detection (14,37–39), in EC, the analysis of even a single blood sample taken before surgery offered information on disease recurrence. In contrast to breast or ovarian cancer, ECs are characterized by a high number of recurrently mutated cancer genes (4,40). Therefore, sequencing panels targeting cancer-related genes rather than personalized tumor-specific assays are well suited for longitudinal cfDNA-based disease monitoring. At the individual mutation level, the MSK-ACCESS assay employed in this study detects through de novo and tumor-informed analysis high confidence mutations at 0.5% and 0.1%, respectively (33). With the ongoing development of assays for MRD detection following curative-intent treatment and during the surveillance period for relapse, which often target larger numbers of mutations and thereby increase sensitivity for ctDNA (11), future studies using ultra-sensitive assays, in particular for EC patients with low-grade low-stage disease, are warranted.

Shedding of tumor DNA from cancer cells has been shown to be highly variable (12,41), and the fraction of ctDNA in plasma in the ECs studied here was frequently undetectable in early-stage disease or low-grade histologies using a deep sequencing method. ECs with high-risk histologic types, however, including carcinosarcoma and serous ECs, had higher ctDNA fractions in plasma. In addition to disease burden, these higher ctDNA fractions may be due to higher levels of processes that play a role in cfDNA release from cells such as proliferation, apoptosis, and necrosis (42).

Case CD37 included in this study was, following review of the resection specimen, an adenosarcoma rather than an endometrial carcinoma. The differential diagnosis of adenosarcoma includes benign and malignant endometrial conditions, including carcinosarcomas, and the clinical presentation may mimic that of EC with abnormal uterine bleeding (43). Uterine adenosarcomas are a genetically heterogeneous group, and recurrent DICER1, FGFR2 and TP53 mutations, MDM2/CDK4/HMGA2 and TERT amplification as well as NCOA2/3 gene fusions, amongst others, have been reported in these tumors (43,44). The primary adenosarcoma CD37 in this study was found to harbor 12 somatic mutations, including pathogenic DICER1, FGFR2, PTEN and ARID1A mutations (Supplementary Figure S2). Notably, the 3 mutations covered by the MSK-ACCESS assay were detected in the cfDNA at baseline (Figure 2), providing evidence to suggest that cfDNA analysis may be used at least for a subset of uterine sarcomas. The diagnostic performance of liquid biopsy in sarcoma in general has been reported to vary and to be dependent on tumor type (45). Larger studies to demonstrate the prognostic value and clinical utility of cfDNA assays in uterine sarcoma are warranted.

Our study has several limitations. Despite the prospective accrual of patients with EC, the number of patients when stratified according to stage or histologic type is small, and PFS analyses are based on limited number of events. Furthermore, due to the COVID-19 pandemic, 11% of samples could not be collected during the follow-up period as initially planned. Nevertheless, our findings provide a proof-of-principle of the role of cfDNA detection in prognostication and disease monitoring in EC.

It is currently still unclear which early-stage EC patients would benefit from adjuvant therapy after surgery with radiation, chemotherapy or both (6,46–48). In addition, recurrent EC can still be cured with salvage therapy if it is caught before wide-spread metastasis occurs (49). Currently, the role of ITCs in the evaluation of risk of recurrence in EC is also unanswered (50). Given these challenges, cfDNA analysis could benefit not only those EC patients with advanced disease for early detection of recurrence but could also prevent women the unnecessary morbidity of adjuvant therapy. Based on our findings, larger, prospective studies are warranted to incorporate cfDNA analysis as a monitoring tool and to guide additional therapies in EC patients at high risk for recurrence.

In conclusion, we have shown here that ctDNA mutation detection in plasma from newly diagnosed EC patients is not only feasible but has prognostic value and enables disease and treatment response monitoring.

TRANSLATIONAL RELEVANCE.

The use of minimally invasive methods for risk stratification, therapy response monitoring, and recurrence detection may impact treatment decision-making and outcomes for patients with endometrial cancer (EC), the leading cause of gynecologic cancer mortality. We demonstrate in EC patients that the detection of circulating tumor (ct)DNA at baseline or immediately post-surgery allowed for the identification of patients with high-risk of disease progression. Further, analyses of longitudinally collected plasma samples revealed that changes in ctDNA fraction mirrored the disease course and therapeutic response and had lead time compared to CT imaging in a subset of EC patients. This proof-of-principle study paves the way for future work assessing the clinical utility of ctDNA-based risk stratification and disease monitoring. Future studies are warranted to define whether EC patients with advanced disease who have changes in ctDNA fraction during surveillance would benefit from immediate therapeutic intervention.