Detection of circulating tumor DNA using a tissue-free epigenomic assay is a highly prognostic biomarker in early-stage triple negative breast cancer

Abstract

Purpose:

Clinical tools to monitor treatment response and metastatic risk could improve early-stage triple-negative breast cancer (TNBC) care. While molecular residual disease (MRD) assays show promise, their use in the neoadjuvant setting requires rapid turnaround times. Tissue-informed approaches may be challenging for patients with limited biopsy samples. The objectives were to determine the surveillance sensitivity for detecting metastatic recurrence and evaluate the ctDNA response to neoadjuvant therapy using a tissue-free epigenomic assay.

Patients and Methods:

Patients with stage II or III TNBC undergoing neoadjuvant docetaxel and carboplatin chemotherapy on a clinical trial (NCT02124902) followed by surgery with or without adjuvant therapy were included in this study. Blood samples were prospectively collected prior to, during, and after completion of neoadjuvant therapy (NAT), and after surgery at pre-specified surveillance time points. Plasma samples were analyzed by Guardant Reveal.

Results:

A total of 119 TNBC patients were included in the analysis. ctDNA was detected in the post-surgical setting in 8.9% (7/79) of patients, with an 83% (5/6) patient-level surveillance sensitivity for metastatic recurrence and 99.5% (197/198) sample-level specificity. Post-surgical ctDNA detection was prognostic for shorter recurrence-free interval (RFI; HR 37.7, p<0.0001). ctDNA detection at the post-NAT pre-surgical time point was also associated with shorter RFI in patients with residual disease at surgery (HR 28.2, p<0.0001).

Conclusions:

In early-stage TNBC patients, a tissue-free, epigenomic assay and demonstrated high specificity and sensitivity for metastatic recurrence. ctDNA detection in the neoadjuvant setting indicated poor prognosis, highlighting its potential role across breast cancer care.

Introduction

Breast cancer is a leading cause of cancer-related deaths in women, with an estimated 300,000 new breast cancer diagnoses expected annually in US (1). Despite effective systemic therapies, up to 30% of patients with breast cancer will exhibit intrinsic or acquired resistance to therapy and develop metastatic disease (2–4). Early triple-negative breast cancer (TNBC) is associated with aggressive tumor biology, characterized by a high metastatic propensity and a relative lack of actionable tumor biomarkers, posing a significant challenge for achieving a curative outcome for many patients (5,6). Therefore, validated tools to identify molecular residual or early metastatic disease is an unmet clinical need in the management of TNBC.

Technologies developed for detecting circulating tumor DNA (ctDNA) as a molecular residual disease (MRD) analyte have demonstrated their capabilities to improve post-treatment recurrence risk stratification across diverse cancer types since ctDNA can often be detected before clinical evidence of recurrence (7–11). However, in recent studies investigating the efficacy of therapeutic intervention upon molecular relapse in TNBC, metastatic disease on imaging was identified upon the first detection of ctDNA in up to 70% of patients allocated to treatment intervention (12,13), underscoring the need to develop more sensitive technologies for patients with TNBC that can detect MRD during a window in which a patient may be treated with curative intent and have the opportunity to participate in clinical trials.

Current breast cancer clinical guidelines do not recommend routine screening for metastatic disease in patients without symptoms suggestive of recurrent disease (14,15). This is primarily due to studies performed in the 1980’s which demonstrated that although intensive image-based screening for recurrence resulted in earlier detection of metastasis, this did not result in improved survival outcomes of patients in these studies (16,17). Evolving treatment paradigms over the last several decades and the availability of sensitive screening technologies warrants re-investigation into whether early metastatic detection and intervention improves patient survival.

Patients with stage II-III TNBC are often treated with neoadjuvant systemic therapy to achieve tumor downstaging and identify those who are at low risk of recurrence. Achieving a pathological complete response (pCR) significantly reduces the risk of recurrence, establishing pCR a valuable prognostic indicator in breast cancer (18,19). However, while patients with non-pCR are at a higher risk of recurrence, it is not clear which patient with non-pCR will have favorable clinical outcomes and may be spared from additional adjuvant chemotherapy. Recently, a study demonstrated that the evaluation of ctDNA at the time of surgery in patients with non-pCR improved the patient prognostication, resulting in clinically meaningful stratification of patient outcomes based on ctDNA status (20). However, neoadjuvant studies in breast cancer to date have primarily utilized tumor-informed ctDNA tests, which may present unique challenges for prospective use in neoadjuvant decision-making given tissue quantity may be limited and turnaround times may be unacceptably long due to the need for prior tissue sequencing. Tissue-free ctDNA tests with clinically meaningful performance have greater potential to address these limitations effectively. Additional studies are needed to validate the prognostic value of ctDNA at this time point and whether ctDNA detection status might be a clinically useful biomarker to guide patient care.

Herein, we demonstrate the clinical performance of a tissue-free epigenomic assay for detecting MRD in patients with early-stage TNBC. Additionally, we demonstrate the prognostic significance of ctDNA at pre-intervention, post-neoadjuvant therapy (NAT) and post-surgery time points which may inform future interception studies aimed at improving clinical outcomes for patients with early breast cancer.

Methods

Patients and samples

Patients who were enrolled between July 2014 and June 2021 with clinical stages II or III TNBC undergoing neoadjuvant docetaxel and carboplatin chemotherapy on a clinical trial (NCT02124902) were included in this study. The primary objective of the clinical study was to determine if neoadjuvant docetaxel and carboplatin would increase the pCR rate in TNBC compared to historical expectations. Overall, the pCR rate was 45.7%, and genomic and immune profiling differentiated between pCR and non-pCR patients as previously described (21). The study was conducted in accordance with the Declaration of Helsinki, approved by the Institutional Review Board of Washington University in St. Louis, and all patients provided written informed consent to participate. As part of a translational study described here, blood samples were prospectively collected for the purposes of ctDNA analysis prior to, during, and after completion of neoadjuvant chemotherapy, and at pre-specified surveillance time points after surgery (every 6 months for 5 years; Supplementary Fig. S1).

Whole blood was collected in K2 EDTA or Streck tubes and immediately processed for plasma isolation. Tubes were centrifuged at 1000g for ten minutes at 4°C. The plasma was aspirated in 1 mL aliquots and transferred to cryovials to be frozen and stored in liquid nitrogen vapor or at −70°C.

ctDNA analysis

Isolated plasma was analyzed for the presence of ctDNA using the commercially available Guardant Reveal assay (Guardant Health, Redwood City, CA). Detailed methods for the Guardant Reveal workflow and computation of methylation-based tumor fraction have been previously described (22). This assay was analytically validated according to College of American Pathologists (CAP) and New York State Department of Health (NYSDOH) requirements in a CLIA-certified laboratory. Briefly, Guardant Reveal is a tissue-free, epigenomic assay that assesses cfDNA for the presence of tumor-derived methylation signatures for MRD detection. The assay’s primary intended use is for post-surgical detection of MRD to predict recurrence, however we also explored its ability to predict response to neoadjuvant therapy in this study. The bioinformatic algorithm for breast cancer interrogates ~3,000 differentially methylated regions (DMRs) of DNA most strongly associated with breast cancer to classify each sample as ctDNA detected or not detected and estimate quantitative tumor fraction. Samples were analyzed and resulted blinded to the patients’ clinical data.

Analysis plan and statistical methods

The primary objective of this biomarker study was to determine the post-operative surveillance sensitivity and specificity for detecting distant metastatic recurrence in patients with TNBC. Surveillance sensitivity was defined as the percentage of patients with distant recurrence who had at least one ctDNA+ sample in the post-surgical setting and specificity was defined as the percentage of ctDNA- post-treatment samples from patients with no evidence of disease recurrence at the time of data cut-off. Patients were eligible to be included in the surveillance sensitivity analysis if they had at least one evaluable sample in the post-surgical setting that was collected within one year of distant recurrence to estimate sensitivity for recurrence in a breast cancer population undergoing routine ctDNA surveillance monitoring (e.g., every 3–6 months). For specificity analysis, patients must have had at least one post-treatment sample, defined as a sample collected at least 21 days post-surgery if no adjuvant therapy was administered or at least 21 days post-adjuvant therapy if adjuvant therapy was administered, and patients must have had at least one year of follow-up from their last surveillance sample for the sample to be included. We also evaluated lead interval was defined as the time from the first post-surgical ctDNA+ test to the detection of diagnosis of distant metastatic recurrence. The association between distant recurrence, TN stage, and likelihood of pCR was also evaluated based on the ctDNA detection status from samples collected at baseline, on-NAT, and post-NAT were included as an exploratory analysis. To evaluate dynamics between two time points, patients must have had samples collected at baseline and pre-surgery that passed quality control (QC).

Data were summarized as medians and proportions, as specified. Comparisons between categorical data including ctDNA detection rates among patient characteristics were analyzed using nonparametric tests. The Mann-Whitney test was conducted for comparisons between quantitative data including methylation score (tumor fraction). Recurrence-free interval (RFI) was calculated using the Kaplan-Meier method. P-values less than 0.05 were considered significant. Statistical analyses were performed using GraphPad Prism 10 (GraphPad Prism; RRID:SCR_002798).

Data availability

Data supporting the generation of all manuscript figures, including ctDNA detection results and patient-level clinical data, are provided in Supplemental Table S1. Raw sequencing data were generated at Guardant Health and are not publicly available to comply with patient consent forms. However, additional details regarding clinical and test data are available from the corresponding author upon reasonable request.

Results

Patient and sample overview

A total of 122 patients with clinical stage II-III TNBC who were enrolled in clinical trial NCT02124902 underwent prospective whole blood collection intended for cfDNA analysis. Of 578 total samples, 452 (78%) samples from 119 patients passed QC and were included in the analysis. A total of 29 samples were collected less than 21 days after surgery and were excluded in the analysis to align with the recommended sample specifications for the Guardant Reveal product configuration (Fig. 1). The median plasma input volume was 3 mL (range, 1–5 mL) with a median cfDNA input of 8.2 ng (range, 0.1–30 ng). The majority (88.2%) of patients had clinical stage II disease, had lymph node-negative disease at diagnosis (58.0%), and grade 3 tumors (89.9%). Most (64.8%) did not receive adjuvant therapy. Most participants were Caucasian (74.8%) but notably, there was a relatively high representation of African American participants in this study (24.4%). The median duration of follow-up was 50 months (range, 7–101 months). The clinical characteristics of the evaluable cohort are summarized in Table 1 and details of patient-level evaluable samples are available in Supplemental Table S1.

Fig. 1.

Summary of evaluable patients and samples.

Table 1.

Cohort characteristics

	N (%)
Total N	119 (100%)
Age (years)
<65	102 (85.7%)
≥65	17 (14.3%)
Race
Caucasian	89 (74.8%)
African American	29 (24.4%)
Other	1 (0.8%)
Stage at diagnosis
IIA	67 (56.3%)
IIB	38 (31.9%)
IIIA	12 (10.1%)
IIIC	2 (1.7%)
Lymph nodes at diagnosis
Node-negative	69 (58.0%)
Node-positive	50 (42.0%)
Grade at diagnosis
G2	9 (7.6%)
G3	107 (89.9%)
Unknown	3 (2.5%)
pCR
Yes	52 (43.7%)
No	67 (56.3%)
Adjuvant therapy
Chemotherapy-containing regimen	35 (29.4%)
Other	7 (5.9%)
None	77 (64.7%)
Recurrence
Distant	13 (10.9%)
Local	5 (4.2%)
None	101 (84.9%)

ctDNA detection at the baseline time point

A total of 72 patients had evaluable samples collected at baseline (pre-NAT) with a detection rate of 78% (56/72). The ctDNA detection rate was higher in patients with clinical lymph node-positive compared to lymph node-negative cancers (100% vs. 63%; p < 0.0001), but was not associated with T stage (p = 0.194) or likelihood of pCR (p = 0.784; Fig. 2A). The median tumor fraction (TF) of ctDNA for all baseline samples was 0.22% (range, 0–45.1%), with ctDNA being detected as low as 0.005%. The baseline TF was significantly higher in patients with lymph node-positive cancers compared to those with lymph node-negative cancers (median 0.84% vs. 0.03%, p < 0.0001). Additionally, the baseline TF was higher in patients with clinical T3 tumors relative to those with T1/T2 tumors (median TF 2.0% vs. 0.12%, p = 0.0047; Fig. 2B), confirming a positive correlation between higher TF and disease burden, as expected. There was no difference in the baseline TF by pCR status (p = 0.899).

Fig. 2. ctDNA detection and prognostic significance at baseline and cycle 1 day 3 (C1D3) of neoadjuvant chemotherapy.

A) Proportion of ctDNA+ samples by lymph node (N) status, tumor stage (T), and pathological complete response (pCR) at baseline. B) Tumor fraction by defined clinical variables at baseline. C) Proportion of ctDNA+ samples for defined clinical variables at C1D3 of NAT. D) Tumor fraction for defined clinical variables at C1D3 of NAT. E) RFI by ctDNA detection at baseline and C1D3.

ctDNA detection and prognostic significance during neoadjuvant chemotherapy

We explored the significance of ctDNA detection at an early on-treatment time point, Cycle 1 Day 3 (C1D3) of NAT, hypothesizing that early changes in ctDNA could serve as an indicator of chemotherapy sensitivity and correlate with clinical outcomes. Overall, ctDNA was detected in 78% (65/83) of patients at C1D3 of NAT. ctDNA detection was not correlated with clinical lymph node status (p = 0.118), clinical T stage (p = 0.062), and was not predictive of pCR (p = 0.108, Fig. 2C). Analysis of C1D3 TF revealed that clinical T stage was the only feature associated with C1D3 TF; patients with T3 tumors had a higher C1D3 TF compared to those with T1/T2 tumors (p = 0.0012, Fig. 2D).

ctDNA detection at the baseline or C1D3 time point was not prognostic for recurrence-free interval (RFI; hazard ratio [HR] 2.5, p = 0.380; HR 0.68, p = 0.516, respectively; Fig. 2E).

ctDNA detection at the post-NAT time point and ctDNA dynamics on NAT

To determine if the detection of ctDNA after completion of NAT was associated with patients’ clinical outcomes, blood samples were collected after completion of all NAT and prior to surgery. A total of 55 patients had evaluable samples at this time point, which were collected a median of 34 days (range, 8–73 days) after completion of NAT. The ctDNA detection rate at this time point was 3.6% (2/55). Both ctDNA+ patients had non-pCR at the time of surgery (pathological stage IIA and IIIC, Fig. 3A) and developed distant metastatic recurrence at 8.7- and 8.2-months post-surgery, respectively. Detection of ctDNA post-NAT was associated with distant metastatic recurrence, despite a small number of events (p = 0.014, Fig. 3B). Next, we evaluated the prognostic significance of ctDNA at the post-NAT time point stratified by pCR status. Patients with non-pCR at the time of surgery who were ctDNA+ had a shorter RFI compared to patients with non-pCR and were ctDNA- (HR 28.2, p < 0.0001). Patients in the non-pCR/ctDNA- group had a similar RFI compared to patients in the pCR/ctDNA- group (HR 0.25, p = 0.17). All patients who exhibited disease recurrence and were ctDNA- recurred greater than one year from surgery, suggesting that the presence of ctDNA at this time point may effectively capture early distant recurrence events.

Fig. 3. Prognostic significance of ctDNA detection and clearance at the pre-surgery time point.

A) Proportion of patients who had detectable ctDNA at the pre-surgery time point based on pCR status and B) recurrence status. C) RFI by ctDNA detection at the pre-surgery time point stratified by pCR status. D) proportion of patients who had detectable ctDNA at both baseline and pre-surgery time points (+/+), cleared ctDNA (+/−), or were negative at both time points (−/−) based on pCR status and E) recurrence status. F) RFI grouped by ctDNA detection at the baseline and pre-surgery time point. R=recurrence.

To determine if ctDNA clearance on NAT was a positive prognostic indicator, the ctDNA dynamics between the baseline and post-NAT time points were assessed. Neither of the two patients (0%, 0/2) who were ctDNA+ at both time points achieved a pCR with NAT, whereas 59% (13/22) of patients who cleared ctDNA achieved a pCR, and 33% (2/6) of patients who were ctDNA- at both time points achieved a pCR. The difference in pCR rate between patients who were persistently ctDNA+ vs. those who cleared ctDNA was not significant; however, this may be due to the few patients in the persistent ctDNA+ group (p = 0.199, Fig. 3D). When evaluating the same categories with disease recurrence rate, 100% (2/2) of patients who were ctDNA+ had a distant recurrence compared to 5% (1/22) who cleared ctDNA (p = 0.011, Fig. 3E). Patients who were persistently ctDNA+ had a shorter RFI compared to those who cleared ctDNA (HR 45.3, p < 0.0001; Fig. 3F).

Post-surgical surveillance and metastatic recurrence prediction

The primary objective of this biomarker study was to determine the post-operative surveillance sensitivity and specificity for detecting distant metastatic recurrence in patients with TNBC. The post-surgical surveillance analysis was performed on 242 unique samples obtained from 85 patients (13 with recurrence). Six patients with recurrence had at least one evaluable post-surgical sample collected within one year of recurrence. The ctDNA surveillance sensitivity for distant recurrence was 83% (5/6, 95% confidence interval [CI] 35.9–99.6%). Of these five patients, ctDNA was detected in two with lead intervals to metastatic recurrence of 14.1 and 5.0 months, and the remaining three had ctDNA detected at the time of recurrence. In these three patients, two had evaluable post-surgical samples only at the time of recurrence precluding ability to demonstrate lead interval. In one of these patients, the time from surgery to recurrence was less than one month. In the ctDNA+ post-surgical samples from the patients included in the surveillance analysis, the median TF was 2.8% (range, 0.05%-61.8%). We evaluated the quantitative ctDNA dynamics in one patient with multiple post-surgical samples prior to recurrence. This patient’s first post-surgical sample was detected at 0.05% TF, with subsequent samples demonstrating an increase in TF over time up leading up to recurrence, aligning with the expected clinical course for this patient (Supplementary Fig. S2). The remaining seven patients with distant recurrence either did not have any post-surgical samples available (n=6) or had their last surveillance sample collected greater than one year from recurrence (Fig. 4A).

Fig. 4. Detection and prognostic significance of post-surgical ctDNA.

A) Swimmer plot of patients with distant recurrence summarizing patient’s treatment and ctDNA testing. Surveillance sensitivity is defined as the rate of ctDNA detection any time post-surgery in patients with clinical recurrence and samples collected within one year of recurrence. B) RFI based on ctDNA detection at the first post-surgery time point analyzed, and C) RFI based on ctDNA detection at any point post-surgery.

We explored the ctDNA detection rates in the post-surgical period by the site of metastasis. ctDNA was detected in 3 of 4 patients with multiple sites of metastatic recurrence, 1 of 1 patient with bone recurrence, and 1 of 1 patient with a lymph node recurrence. One recurred patient who did not have detectable ctDNA during the post-surgical period had a lung-only recurrence. Additionally, there was one patient with multiple sites of metastasis who was ctDNA- at all three post-surgical time points; this patient was not included in the surveillance sensitivity analysis since their last sample was collected 13.5 months prior to recurrence. As an exploratory objective, we evaluated the surveillance sensitivity for local recurrence. In the post-surgical setting, ctDNA was detected in 1 of 3 patients who had a local recurrence (Supplementary Fig. S3).

In the 198 samples collected among 69 patients who did not have disease recurrence, one sample was ctDNA+, resulting in a patient specificity of 98.6% (68/69, 95% CI 92.2–100%) and sample specificity of 99.5% (197/198, 95% CI 97.2–100%; Supplementary Fig. S4).

The prognostic significance of ctDNA detection during the post-surgical follow-up period was evaluated. Patients who had detectable ctDNA at the first time point post-surgery had a shorter median RFI compared to patients who were ctDNA- at the first post-surgery time point (median RFI 10.8 months vs. not reached [NR], HR 38.8, p < 0.0001; Fig. 4B). Patients with detectable ctDNA at any point during the post-surgical period had a shorter median RFI compared to those who never had detectable ctDNA during the post-surgical period (median RFI 8.2 months vs. NR, HR 37.7, p < 0.0001; Fig. 4C). The 3-year RFI rate was 14.3% for patients who were ctDNA+ at any point during follow-up, compared to 94.3% for those who remained ctDNA-. RFI was further assessed by stratifying patients based on their primary surgical outcomes (pCR vs. non-pCR). Among patients with non-pCR, those who were ctDNA+ in the post-surgical period exhibited shorter RFI compared to patients who were ctDNA-, identifying a subgroup of non-pCR patients who were at a lower risk of recurrence (Supplementary Fig. S5).

Discussion

Longitudinal MRD surveillance testing in patients with breast cancer who have completed curative-intent therapy may provide an opportunity for clinicians to tailor treatment or surveillance strategies for individual patients. Our study evaluated the clinical performance of the Guardant Reveal MRD test in patients with early-stage TNBC, demonstrating a high specificity and sensitivity for distant metastatic relapse. Additionally, we explored ctDNA detection and dynamics during NAT and observed that ctDNA clearance compared to ctDNA persistence was a positive prognostic indicator for disease recurrence.

In this study, the surveillance sensitivity for distant recurrence among patients with samples collected within one year of recurrence was 83% (5/6). We assert that sensitivity in a surveillance population should be estimated using clinically meaningful time-bound analyses to emulate a MRD surveillance program of testing at least every three to six months during follow-up. Indeed, several studies have consistently shown that more frequent ctDNA sampling may enhance surveillance sensitivity and improve the likelihood of detecting ctDNA prior to clinical detection of recurrence. For MRD tests in breast cancer including all subtypes, reported sensitivities for recurrence range from 80% to 89% with greater than 95% specificity (7,23–27). One limitation in the present study was the infrequent blood collections during the post-surgical period. Blood collections were planned every six months during follow-up as per protocol; however, the COVID-19 pandemic and patient attrition contributed to some missed blood collections. While MRD was detected prior to metastatic disease in a subset of patients, three patients only had available samples at the time of recurrence, which were ctDNA+. We hypothesize that earlier and more frequent ctDNA sampling during the surveillance period in our study could have improved metastatic recurrence prediction and provided a more precise lead interval.

Our study also explored the surveillance ctDNA sensitivity by site of recurrence. We observed that the patient with false negative results during the post-surgical period had a lung-only metastatic recurrence. The lower sensitivity or shorter lead times for ctDNA detection in patients with lung-only metastasis has previously been observed in breast as well as other cancers (11,25,28). We additionally explored the ability to detect ctDNA in patients with local recurrences and observed numerically lower ctDNA detection rates in these patients, consistent with other studies (24,25). Since MRD studies to date lack the statistical power to estimate sensitivity for local recurrences, these sensitivity estimates are currently observational. However, considering the existing clinical tools for monitoring local recurrences and new breast primary tumors (such as mammography and breast MRI) (29), along with the favorable outcomes associated with local versus distant recurrences (30), ctDNA monitoring has greater potential to address the clinical need for surveilling patients for distant metastasis. In this context, ctDNA analysis could enhance patient care by complementing local imaging tools with valuable insights into distant metastasis and local disease recurrence.

TNBC is a particularly challenging disease for surveillance monitoring, as others have shown relatively short lead intervals from ctDNA detection to recurrence presumably driven by the more aggressive disease biology and rapid recurrence after surgery compared to other breast cancer subtypes (12,24). In a prospective study involving early stage TNBC patients, metastatic relapse on imaging at the time of ctDNA positivity was observed in approximately 70% of cases (12). Similar findings of high metastatic relapse rates following the first ctDNA+ test were recently reported in the ZEST clinical trial (13). This trial evaluated whether niraparib could improve disease-free survival in patients with TNBC or HER2-negative, BRCA-mutated breast cancer who had completed treatment and exhibited ctDNA positivity without radiographic evidence of recurrence. Additionally, in a separate study using an MRD assay in the real-world setting, among TNBC patients who underwent restaging scans within one year of a positive ctDNA result, up to 40% of TNBC patients who underwent restaging scans within one year of a positive ctDNA result were found to have detectable metastatic disease (31). These findings are primarily driven by the earlier detection of asymptomatic metastatic disease through imaging prompted at the time of ctDNA detection rather than at the time that clinical symptoms develop, as clinical guidelines do not recommend image-based surveillance outside of clinical signs and symptoms suggestive of metastasis. Therefore, due to the retrospective nature of our study, the reported lead intervals may differ from those determined through prospective MRD testing in clinical practice. When designing adjuvant and surveillance studies, clinical lead intervals from ctDNA detection to metastatic recurrence, and the expected rate of image-detectable metastatic disease following a positive ctDNA result should be considered.

Current clinical guidelines recommend NAT for patients with stage II-III TNBC (14). However, it remains unclear whether a subset of patients can safely have systemic therapy de-escalated or if escalated therapy is necessary to minimize recurrence risk for another subset. The detection of ctDNA at a pre-NAT time point offers both prognostic value by identifying patients with a relatively higher burden of disease and potential utility in monitoring ctDNA dynamics during neoadjuvant therapy. In this study, ctDNA was detected in 78% of patients at the pre-NAT time point, including 100% of those with lymph node-positive disease at diagnosis. Published studies on patients with TNBC undergoing neoadjuvant therapy report pre-NAT ctDNA detection rates of 82% to 93% when using tumor-informed assays (20,26,32). These studies, however, included patients with varying levels of disease burden which may introduce selection bias toward those at a very high risk of recurrence. For example, in the I-SPY2 trial, ctDNA was detected at the pre-NAT time point in 91% of TNBC patients (20). Notably, this analysis focused on patients whose tumors were MammaPrint high (where 88% of the TNBC patients were MammaPrint ultra-high 2), a group characterized by tumors with a high proliferative advantage, likely contributing to higher tumor DNA shedding into the blood. Consequently, ctDNA detection rates at diagnosis should be interpreted with caution, as they are influenced by the underlying tumor biology and selection criteria of the study cohort.

Integrating ctDNA status with clinicopathologic features, such as pCR status, may better risk stratify patients to aid in adjuvant therapy decisions or surveillance approaches. In this study, ctDNA was detected in 4% of patients at the post-NAT/pre-surgery time point and effectively predicted metastatic recurrence, whereas ctDNA- patients at this time point had a low rate of distant metastatic recurrence. All ctDNA+ samples in patients with distant recurrence were from those who did not achieve a pCR with NAT, including pathological stages IA, IIA, IIIA, and IIIC (Supplemental Table S1). Post-surgical ctDNA was detected in only one patient who achieved a pCR, consistent with other studies showing 0–5% ctDNA detection rates in this subgroup (20,26). This finding is clinically significant because under current guidelines, non-pCR patients are recommended to receive additional adjuvant therapy. Our findings on the prognostic significance of pre-surgery ctDNA detection align with the I-SPY 2 trial, which reported a 22% detection rate in TNBC patients and a positive association with disease recurrence. The higher detection rate in that study reflects the greater recurrence risk, as 74% of TNBC patients who were ctDNA+ at the pre-surgery time point experienced recurrence. These data suggest that ctDNA status following NAT can identify a subset of very high-risk patients who may require additional systemic therapy. This approach may offer an opportunity to further risk stratify and guide adjuvant therapy decisions in a patient population at high risk of recurrence shortly after surgery. Particularly, our data indicate that stage II/III TNBC patients with residual disease at surgery may benefit from adjuvant and surveillance ctDNA monitoring, with testing performed earlier and more frequently, as positive ctDNA detection both at the time of surgery and throughout follow-up predicted poor outcomes in those without a pCR. By contrast, patients achieving a pCR consistently demonstrated ctDNA negativity both at the time of surgery and during follow-up. Aligned with findings from historical clinical trials, our study demonstrated a very low recurrence rate in patients achieving a pCR, supporting its potential as an additional prognostic factor to incorporate into ctDNA surveillance strategies.

Recent studies have demonstrated the prognostic significance of ctDNA dynamics during NAT using tissue-informed assays; patients who cleared ctDNA at various time points with NAT had improved clinical outcomes compared to patients who did not clear ctDNA with NAT (20,33). Indeed, early clearance of ctDNA after three weeks of neoadjuvant therapy in the I-SPY 2 trial was associated with an improved prediction of pCR and better identification of patients at lower risk of disease recurrence, compared to patients who remained persistently ctDNA+ during NAT. Consistent with these findings, our study demonstrated that ctDNA clearance between the pre-NAT and post-NAT/pre-surgery time points was a positive prognostic indicator. We explored the prognostic significance of ctDNA collected at an early on-treatment time point at three days after starting neoadjuvant chemotherapy (C1D3). The detection of ctDNA at C1D3 of NAT was not predictive of distant recurrence, and we hypothesize that ctDNA detection this time point may be indicative of persistent ctDNA detected at baseline in addition to tumor DNA shed in response to the acute effects of chemotherapy. Further research is needed to understand the optimal on-treatment time points and to determine if the magnitude of change in ctDNA TF can provide valuable prognostic and predictive information to inform adaptive neoadjuvant therapy approaches.

In the clinical setting, it is essential for a ctDNA assay to deliver rapid turnaround times and maintain a high QC pass rate to support adaptive neoadjuvant strategies and enable effective surveillance monitoring. However, obtaining sufficient tissue for a tumor-informed MRD test can be challenging, particularly in breast cancer cases where diagnostic biopsies yield limited specimens or in patients who achieve complete responses to neoadjuvant therapy. Studies in TNBC evaluating standard neoadjuvant therapy have reported pCR rates as high as 65% (4), potentially excluding a significant proportion of patients from accessing tumor-informed monitoring assays if there is insufficient tissue availability from diagnostic biopsies. Additionally, a retrospective analysis of ctDNA in the I-SPY2 trial demonstrated that time points as early as three weeks after initiating neoadjuvant therapy provided clinically meaningful insights, correlating ctDNA dynamics with patient outcomes (20). However, conducting prospective interception studies using an adaptive neoadjuvant therapy design requires rapid turnaround times for clinical decision-making at early time points, which may pose a limitation for tumor-informed tests. Thus, a tissue-free assay is ideally suited for this use case.

The utilization of TF in early-stage cancer is currently being investigated. In this study, we assessed the quantity of ctDNA at various time points by analyzing methylation-based TF. Among the 56 ctDNA+ samples at baseline, TF showed a positive correlation with disease burden, as expected. Specifically, baseline TF was higher in patients with lymph node-positive cancer compared to those with lymph node-negative cancer, and in patients with T3 tumors compared to those T1/T2 tumors. These results suggest that methylation-based TF in early breast cancer can enhance the precision of evaluating ctDNA detection and dynamics. Future studies should explore the validity and clinical value of quantitative ctDNA dynamics in the breast cancer neoadjuvant setting by utilizing clinically relevant on-treatment time points to assess the relationship between TF and treatment response. In the post-surgical setting, we observed a relatively higher median TF in ctDNA+ samples compared to the baseline time point. This is also expected, as post-surgical ctDNA positivity, particularly in samples collected at or near the time of recurrence, is indicative of early or a higher burden of metastatic disease. To support this, in the one patient who had multiple ctDNA+ post-surgical samples, we observed an increase in TF over time leading up to distant recurrence (Supplementary Fig. S2).

A limitation in our study was the small number of recurrence events, which restricted the analysis of ctDNA dynamics. Despite this, most participants in this study had both baseline and post-NAT samples available, allowing us to demonstrate a strong association between ctDNA clearance and favorable clinical outcomes. Studies with a larger proportion of patients who exhibit disease recurrence will be needed to further validate these findings. Furthermore, although we were able to analyze ctDNA status at the post-NAT/pre-surgical time point in relation to pathological staging, a limitation was our inability to associate ctDNA status with residual cancer burden (RCB). Incorporating RCB data could have provided additional insights into current prognostic measurements. An additional limitation of this study was the unexpectedly high number of QC failures. The primary cause of sample failure was low cfDNA extraction yield, which was partially influenced by the subset of samples with low plasma input (1–2 mL). This finding is not typical in clinical settings, where blood collections often yield consistently higher amounts of cfDNA. In a sampling of approximately 10,000 clinical blood collections from patients diagnosed with early-stage breast cancer, the median yield of cfDNA was 13.7 ng (interquartile range, 9.3–20.4 ng) with a QC pass rate of 98.2% (Guardant Health, data on file). Thus, this highlights a potential limitation of analyzing biobanked specimens. Another strength of our study was its inclusion of a relatively high proportion of African American participants (24%), ensuring a robust representation of a diverse population.

In summary, we evaluated ctDNA in patients with stage II/III TNBC using a tissue-free MRD assay and demonstrated high specificity and sensitivity for distant recurrence detection. Detection and dynamics of ctDNA in the neoadjuvant setting correlated with clinical features and outcomes, indicating the need for further validation of ctDNA-based adaptive NAT approaches. Our data support the validation of this tissue-free epigenomic assay to identify TNBC patients with a high risk of distant metastatic recurrence during the post-surgical surveillance period.

Supplementary Material

1

Supplementary Fig. S1. Study schema for the ctDNA biomarker analysis in the study. NAT=Neoadjuvant therapy

Supplementary Fig. S2. Methylation based tumor fraction (TF) in patient NTN096 who had a distant recurrence with multiple post-surgical samples. Blue shading=adjuvant therapy window; red dotted line=time of metastatic recurrence; filled circles=ctDNA+; open circles=ctDNA-.

Supplementary Fig. S3. Swimmer plot of patients with local recurrence summarizing patient’s treatment and ctDNA testing.

Supplementary Fig. S4. Swimmer plot of patients with no known recurrence summarizing patient’s treatment and ctDNA testing.

Supplementary Fig. S5. RFI based on ctDNA detection at any point post-surgery stratified by pCR status.

Statement of translational relevance.

Effective surveillance tools for patients with early-stage triple negative breast cancer (TNBC) following completion of curative-intent therapy are lacking. Further, there is room for improved methods of determining prognosis beyond standard cancer staging and pathologic response to neoadjuvant systemic treatment. In this study, we determined the feasibility of a tissue-free, epigenomic molecular residual disease (MRD) assay for detecting circulating tumor DNA (ctDNA) in TNBC patients undergoing neoadjuvant chemotherapy followed by definitive surgery. Our findings demonstrate a surveillance sensitivity of 83% for detecting distant metastatic recurrence with a sample specificity of 99.5%. Furthermore, we demonstrate ctDNA detection during neoadjuvant therapy has prognostic significance in predicting patient outcomes. These results underscore the potential of a MRD test to complement current monitoring tools during neoadjuvant therapy and post-surgical surveillance and help inform prognosis. Integrating this test into clinical practice as an additional tool could allow clinicians to personalize treatment decisions for patients.