Rational thresholding of circulating tumor DNA concentration for improved surveillance of metastatic breast cancer

Abstract

Background

The use of circulating tumor DNA (ctDNA) concentration for metastatic cancer surveillance is promising, but uncertainty remains about cut-offs with clinical validity.

Materials and methods

This observational study recruited 136 subjects with advanced metastatic breast cancer (irrespective of ERBB2/hormone receptor status) for sequencing of their primary tumor in search for PIK3CA hotspot variants amenable for monitoring by droplet digital PCR (ddPCR). The study analyzed 341 on-treatment samples from 19 patients with PIK3CA variants H1047R or E545K enrolled for long-term (median 85 weeks, range 13-125 weeks), frequent (every 3-5 weeks, median of 14 time points per subject, range 2-29) blood sampling for ctDNA quantification by ddPCR, orthogonally validated by deep sequencing. The diagnostic accuracy of ctDNA versus cancer antigen 15-3 (CA15-3) concentrations to predict disease progression within 12 weeks was investigated using receiver operating characteristic (ROC) analysis. Likelihood ratios were used for rational selection of ctDNA result intervals.

Results

ctDNA [area under the ROC curve (AUC) 0.848, 95% confidence interval (CI) 0.791-0.895] showed superior diagnostic performance than CA15-3 (AUC 0.670, 95% CI 0.601-0.735, P < 0.001) to predict clinical progression within 12 weeks. ctDNA levels below 10 mutant allele copies/ml had high negative predictive value (88%), while levels above 100 copies/ml detected 64% of progressions 10 weeks earlier versus standard of care. Logistic regression analysis indicated complementary value of ctDNA and the presence of two consecutive CA15-3 rises, resulting in a model with 86% (95% CI 74% to 93%) positive predictive value and a clinically meaningful result in 89% of blood draws.

Conclusions

Intensive ctDNA quantification improves metastatic breast cancer surveillance and enables individualized risk-based scheduling of clinical care.

Highlights

• •CA15-3 is widely used to monitor breast cancer treatment and recurrence despite its limited specificity.
• •ctDNA levels show diagnostic value for cancer progression but decision thresholds for individual patients were lacking.
• •We present a novel statistical approach for retrospective analysis of real-world data on ctDNA levels in cancer surveillance.
• •In metastatic breast cancer, ctDNA levels below (above) 10 copies/mL (100 copies/mL) have high NPV (PPV) for progression.
• •Frequent long-term monitoring of ctDNA levels holds value for risk-informed scheduling of standard care in metastatic cancers.

Introduction

Metastatic breast cancer is the leading cause of cancer-related death in women worldwide.¹ Survival can be improved through sequential administration of systemic therapies upon disease progression. Clinical practice for monitoring relapse varies and is mostly based on symptoms, but imaging [computed tomography (CT) scans and bone scans] is recommended every 2-4 months by guidelines.² The use of serum markers such as cancer antigen 15-3 (CA15-3) for surveillance is widely used in Europe,³ but it has limited sensitivity and specificity and is not recommended by the American Society of Clinical Oncology (ASCO) and European Society for Medical Oncology (ESMO). Other organizations such as the European Group on Tumor Markers⁴ as well as the National Academy of Clinical Biochemistry⁵ support using blood biomarkers to monitor therapy efficacy. Decades ago, studies showed no benefit of intensive surveillance on survival or quality of life for metastatic breast cancer patients^6,7 but new therapies and advances in blood biomarkers warrant a reassessment.

The potential of circulating tumor DNA (ctDNA) for cancer diagnosis and monitoring, also known as liquid biopsy, is being explored. ctDNA reflects real-time tumor burden and ESMO recognizes its utility for identifying actionable mutations.⁸ ASCO calls for more interoperability, improved sensitivity and overall more data on clinical validity of ctDNA before inclusion in guidelines.⁹ Several studies have shown that ctDNA can detect minimal residual disease (MRD), predict response to neoadjuvant therapies, provide a lead time for earlier relapse recognition and show strong prognostic power for overall (OS) and progression-free survival (PFS).9, 10, 11, 12, 13, 14 Based on trials such as EMERALD and PADA-1,15, 16, 17 liquid biopsies were also established as test of choice to identify secondary ESR1 mutations as an actionable mechanism of acquired resistance to endocrine therapies in hormone receptor-positive/human epidermal growth factor receptor 2-negative (HR+ HER2−) advanced breast cancers. However, the clinical validity and utility of ctDNA for surveillance of advanced cancers are still being studied by ASCO and ESMO.^8,18,19 Only a few studies have investigated ctDNA in metastatic breast cancer. A feasibility study on sequencing-based liquid biopsy for identification of actionable mutations confirmed the prognostic value of rising ctDNA levels for relapse detection.²⁰ Hrebien et al. reported that early ctDNA dynamics after initiation of a new line of therapy can be used as a surrogate marker for therapeutic response and that intensive 4-weekly ctDNA quantification generates a lead time of 3 months over clinical progression in 74% of patients.²¹ A seminal proof-of-concept analysis by Dawson et al. showed that ctDNA concentrations had superior diagnostic accuracy versus both CA15-3 and circulating tumor cells for relapse prediction and found a correlation at cohort level between ctDNA level expressed in copies/ml and OS.²²

The common limitation of studies carried out thus far is that data analysis remained limited to proportional hazard modeling and survival analysis at cohort level without proposing useful result intervals for clinical use. More data are needed to provide clear guidance on how to interpret ctDNA concentrations for clinical use. In this PrecisionTrack study, we aimed to define clinically relevant ctDNA thresholds and provide proof of concept for their validity.

Materials and methods

Patients and samples

This prospective, single-center, observational diagnostic study (PrecisionTrack) conducted at AZ Delta General Hospital, Roeselare, Belgium, assessed patients with advanced, metastatic breast cancer receiving systemic therapy, irrespective of HR and ERBB2 status, with a minimal life expectancy of 6 months. All patients had initiated systemic treatment before study entry and were receiving it in our day hospital. We selected these patients because it facilitated the collection of blood samples every 3-5 weeks. Since the treatment was ongoing at the time of study entry, all patients had either disease control or an ongoing response at that moment. In a first phase, from 1 July 2019 to 1 December 2021, the study recruited 136 patients for next-generation sequencing (NGS) on breast cancer biopsies (Supplementary Material S1 and Figure S1, available at https://doi.org/10.1016/j.esmoop.2024.102235) with a custom amplicon-based panel to identify pathogenic variants in PIK3CA/TP53, GATA3, ERBB2, ERBB3, ESR1, AKT1, BRCA1, BRCA2, KRAS and PTEN as part of standard of care. In a second phase, subjects with PIK3CA variants were enrolled for long-term, intensive (every 3-5 weeks) plasma sampling (2 Streck tubes/time point). The timing of sampling was mandatory according to the protocol, within a 2-week window. This allowed the samples to be collected on the scheduled date of treatment at the day hospital, which is more convenient for the patients. Depending on the type of treatment, some, for instance, are administered every 3 weeks, while others are given every 4 weeks. There are also various other schedules for different treatments. Thirty-six percent of patients showed PIK3CA variants, of which 58% comprised hotspot variants PIK3CA c.3140A>G:p.(His1047Arg) (henceforth referred to as H1047R) and c.1633G>A:p.(Glu545Lys) (henceforth referred to as E545K). This interim analysis, with a data closure on 30 November 2022, reports on a total of 19 subjects with H1047R (n = 14) or E545K (n = 5) variants monitored over a total of 341 on-therapy blood samples (281 with ctDNA results, 262 with CA15-3 results and 202 with both markers measured). The baseline characteristics are summarized in Table 1 and provided for all included subjects in Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2024.102235. All subjects provided written informed consent. This study was approved by the AZ Delta Ethical Committee under clinical trial number B117201939334. The ctDNA data were not used to alter clinical care or therapeutic decisions.

Table 1.

Characteristics of study cohort at baseline

	Total n = 19 (%)
Median age (range), years	67 (51-81)
Breast cancer phenotype, number of subjects (%)
HR+ HER2−	11 (57.9)
Triple negative	1 (5.3)
HER2+	7 (36.8)
Site of metastases, number of subjects (%)
Bone only	5 (26.3)
Liver	3 (15.8)
Visceral, not liver	11 (57.9)
Number of previous therapeutic lines, number of subjects (%)
0	5 (26.3)
1	3 (15.8)
2	1 (5.3)
3	5 (26.3)
>3	5 (26.3)
Systemic treatment at enrollment, number of subjects (%)
Hormonal + CDK4/6 inhibitor or PI3K inhibitor	8 (42.1)
Hormonal	0 (0)
Anti-HER2 treatment	7 (36.8)
Chemotherapy	4 (21.1)

The table lists distributions of the subjects [number (percentage)] at the moment of start of enrollment for ctDNA monitoring according to breast cancer phenotype (HR+: hormone receptor positive; triple negative, HER2+: ERBB2 amplified), predominant site of metastases, number of prior systemic therapeutic lines and systemic therapy at enrollment. These data for all individual subjects are presented in Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2024.102235.

CDK4/6, cyclin-dependent kinase 4/6; ctDNA, circulating tumor DNA; HER2, human epidermal growth factor receptor 2; PI3K, phosphoinositide 3-kinase.

Assays are described in Supplementary Material S1, available at https://doi.org/10.1016/j.esmoop.2024.102235. In brief, cell-free DNA was extracted using the QIAamp Circulating Nucleic Acid Kit and ctDNA was measured by droplet digital PCR (ddPCR) (Biorad, Hercules, CA) using assays dHSAMDV2010075 for E545K and dHsaMDV2010077 for H1047R. Reflex testing was done by unique molecular identifier-enabled deep sequencing using the Avenio ctDNA Expanded Kit (Roche, Switzerland). Molecular counting of mutant PIK3CA by ddPCR and NGS showed excellent correlation [Supplementary Figure S2, available at https://doi.org/10.1016/j.esmoop.2024.102235, Spearman’s rank correlation ρ = 0.943, 95% confidence interval (CI) 0.893-0.970]. CA15-3 was measured by Elecsys CA 15-3 II immunoassay (Roche, Basel, Switzerland).

Data analysis and dichotomization for outcome

The study’s primary outcome was the presence or absence of clinical progression, defined as a time point with clear radiological progression on imaging (chest/abdomen CT, bone scan with magnetic resonance imaging or positron emission tomography–CT scan in selected cases), reviewed in a blinded manner according to RECIST 1.1 and/or a change of therapeutic regimen. Radiological evaluation was typically done every 12 weeks (range 7-19 weeks). The radiological assessment schedule was determined based on clinical discretion and was independent of the liquid biopsy results. No patients had their treatment changed due to toxicity without experiencing progression. Next, each ctDNA time point [with its associated ctDNA concentration measured as mutant allele copies/ml and variant allele frequency (VAF)] was dichotomized with a value of 1 if clinical progression was recorded within 12 weeks from its sampling date and 0 if not. The study included 281 (41% positive), 262 (41% positive) and 202 (37% positive) informative time points for, respectively, PIK3CA ddPCR, CA15-3 level and simultaneous measurement of both PIK3CA ddPCR and CA15-3 level. Receiver operating characteristic (ROC) analysis and multivariate logistic regression were carried out using MedCalc 12.2.1 (Mariakerke, Belgium). Data were expressed as medians (25th-75th percentiles) and statistical differences were determined using the Mann–Whitney U test. The sample size was not predetermined due to the exploratory design of the study.

Results

Study cohort for liquid PIK3CA variant monitoring

NGS on the primary tumor in 136 metastatic breast cancer patients indicated a pathogenic PIK3CA variant in 36% of patients (Figure 1), of which 58% had H1074R (35%) or E545K (23%) (Supplementary Figure S1, available at https://doi.org/10.1016/j.esmoop.2024.102235) variants. A total of 19 subjects (18 females, 1 male, Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2024.102235) with PIK3CA H1047R or E545K variants were recruited for ctDNA monitoring by ddPCR. At interim analysis, subjects were monitored for a median (range) of 85 (13-125) weeks, with a median (range) of 14 (2-29) ctDNA time points per subject, resulting in a total of 281 on-treatment time points for analysis of ctDNA versus progression status, 262 time points for CA15-3 and 202 time points with both ctDNA and CA15-3. At the time of data closure, the OS was 65%. Out of 19 subjects, 5 (26%) exhibited an ongoing response to treatment or stable metastatic disease at the moment of data closure without evidence of clinical progression. These subjects, further referred to as ‘non-progressors’, had a 100% OS rate at the time of data closure. These non-progressors were studied in a total of 91 ctDNA data points, with a median (range) of 16 (10-21) data points per subject, corresponding to a median (range) of 85 (55-125) weeks’ monitoring. Radiological evaluation was typically done every 12 weeks (range 7-19 weeks). The radiological assessment schedule was determined based on clinical discretion and was independent of the liquid biopsy results. Patients with clinical or radiological disease progressions during follow-up (14 of 19, 74%) showed a median (range) of 2-3 (1-6) progression events. These progressors were studied in a total of 190 ctDNA data points, with a median (range) of 13 (2-29) data points per subject, corresponding to a median (range) of liquid monitoring of 79 (13-121) weeks (Figure 1).

Figure 1.

Study design. The study employed a tumor-informed approach to monitor ctDNA in patients with advanced metastatic breast cancer. Firstly, the primary tumor was sequenced to identify those with PIK3CA hotspot variants. These patients were then enrolled in long-term blood sampling for quantitative ctDNA analysis by ddPCR, along with regular physical exams and imaging as standard care. Progression-free survival was the primary outcome. The analysis focused on the diagnostic value of ctDNA and CA15-3 in predicting clinical progression within 12 weeks, with the aim of identifying clinically useful result intervals. CA15-3, cancer antigen 15-3; CT, computed tomography; ctDNA, circulating DNA; ddPCR, droplet digital PCR; MRI, magnetic resonance imaging; PET, positron emission tomography; ROC, receiver operating characteristic; UMI, unique molecular identifier.

Kinetics of ctDNA levels as a function of disease progression

Time curves were generated to plot the levels of ctDNA and CA15-3 as a function of radiological progression and therapy switch (see Figure 2 and Supplementary Material S2, available at https://doi.org/10.1016/j.esmoop.2024.102235, for a graphic summary of all subjects). Among non-progressors (Figure 2A and B), the median level of mutant PIK3CA copies/ml was 0 (range 0-8). However, shedding of ctDNA was observed above the limit of blank in three out of nine non-progressors, with 23% (21/91) of data points showing ctDNA levels above the limit of blank, up to 390 copies/ml. This suggests that shedding of ctDNA can occur in stable metastatic breast cancer, and is not necessarily related to the introduction of a new therapy since it was observed 20-56 weeks after initiation of a new line of treatment (PT016, Figure 2B).

Figure 2.

Representative cases and data structuring. Each graph displays the PIK3CA mutant copy number/ml, measured by ddPCR on the left x-axis using black dots and lines, and CA15-3 (U/ml) levels, measured on the right y-axis using gray dots and lines, at specific weeks of follow-up after enrollment (x-axis). A bar located above the graph shows whether a CT scan of the abdomen and thorax was conducted (colored square) or not (empty square) and evaluated as stable (green), progressive (purple) or uncertain (gray) compared to the previous assessment. Purple triangles indicate time points that were designated as ‘1’ for the presence of progression for ROC analysis of CA15-3 and ctDNA. Therapy switches are depicted using color fields with randomly assigned colors. Optimal lower/upper thresholds for PIK3CA mutant allele levels are illustrated using black dotted lines with a gray zone between 10 and 100 copies/ml (left y-axis). (A and B) Two patients without disease progression are demonstrated. In PT016 mutant PIK3CA was detected on multiple time points and confirmed by deep sequencing (Supplementary Material, available at https://doi.org/10.1016/j.esmoop.2024.102235). (C-F) Variability of mutant PIK3CA and CA15-3 kinetics in progressive patients is shown. Patient PT017 had oligo-progression restricted to one metastatic lesion in the urinary bladder. A graphical summary for all 19 patients, with reflex testing by NGS, is available in the Supplementary Material, available at https://doi.org/10.1016/j.esmoop.2024.102235. CA15-3, cancer antigen 15-3; CT, computed tomography; ctDNA, circulating DNA; ddPCR, droplet digital PCR; ROC, receiver operating characteristic.

Superior diagnostic accuracy of ctDNA versus CA15-3 for prediction of progression

To investigate the diagnostic accuracy of ctDNA versus CA15-3 for predicting progression in metastatic breast cancer, a binary classification was assigned to each individual ctDNA data point for presence (1) or absence (0) of clinical progression within 12 weeks from the sampling date of that time point (Supplementary Material S3, available at https://doi.org/10.1016/j.esmoop.2024.102235). Figure 3A and B shows the distributions of CA15-3 (U/ml) and ctDNA (PIKC3A mutant allele copies/ml) levels in the non-progressors (n = 5) and in samples of the progressors as a function of absence or presence of progression within 12 weeks (Supplementary Table S2, available at https://doi.org/10.1016/j.esmoop.2024.102235). CA15-3 levels were higher in progressors versus non-progressors (P = 0.0006), but not significantly different within the progressors in samples followed by a progression event or not (P = 0.086) (Figure 3A). PIK3CA mutant copy numbers showed clearly less overlapping distribution than CA15-3: in progressors, time points with progression within 12 weeks showed a median [interquartile range (IQR)] of 200 (0-52 900) copies/ml, as compared to 7 (0-6450) copies/ml (P < 0.0001) in samples not followed by progression. Three of 14 progressive patients showed oligo-progression in a single bone (PT018), brain (PT011) and bladder (PT007) lesion. Oligo-progressors showed slightly lower CA15-3 (median 35 versus 48 U/ml, P = 0.002) but higher ctDNA levels (median 5 versus 0 copies/ml, P = 0.030) than non-progressors (Figure 3A and B).

Figure 3.

Diagnostic accuracy of CA15-3 versus PIK3CA ctDNA level to predict disease progression within 12 weeks. (A and B) The concentrations of CA15-3 (U/ml) and PIK3CA mutant allele (copies/ml) in all data points from subjects who did not experience clinical progression (non-progressors), as well as the aggregated data points from all subjects based on the absence or presence of progression within 12 weeks (Px ≤12w) from sampling (P values determined using the Mann–Whitney U test, with green dots in B indicating outliers by Tukey’s test) are displayed. (C) An ROC analysis of CA15-3 level (U/ml, yellow), percent change in CA15-3 compared to the previous measurement (blue) and the presence of two consecutive CA15-3 rises with at least 25% (green) is presented. (D) An ROC analysis of ctDNA concentration expressed as copies/ml (blue) and VAF (green) is displayed. The inset shows that increasing concentrations of PIK3CA mutant alleles are associated with higher positive LRs for progression within 12 weeks. CA15-3, cancer antigen 15-3; ctDNA, circulating DNA; LR, likelihood ratio; ROC, receiver operating characteristic; VAF, variant allele frequency.

ROC analysis indicated that both the absolute CA15-3 levels and the percent change relative to the previous time point had equivalent, albeit moderate, diagnostic value for predicting progression with, respectively, areas under the ROC curve (AUCs) of 0.680 (95% CI 0.617-0.738) and 0.665 (95% CI 0.602-0.725) (Figure 3C). Also the presence of two consecutive CA15-3 rises of 25% or higher had diagnostic value (AUC 0.570, 95% CI 0.505-0.634) but less (P < 0.01) than absolute CA15-3 levels (Figure 3C). The mutant allele concentration (copies/ml) and VAF (%) of ctDNA showed strong diagnostic accuracy, with equivalent AUC values of 0.846 (95% CI 0.791-0.884) and 0.830 (95% CI 0.778-0.873), respectively (Figure 3D). Analysis of data points with simultaneous measurement of ctDNA (AUC 0.848, 95% CI 0.791-0.895) and CA15-3 (AUC 0.670, 95% CI 0.601-0.735) revealed that ctDNA outperformed CA15-3 (P < 0.0001) in predicting progression within 12 weeks, in line with previous studies.²²

Rational thresholding of ctDNA concentration for clinical utility: the 0/10/100 copy model

Increasing ctDNA concentrations were associated with increasing likelihood ratio (LR) for progression (inset Figure 3D). This allowed the empirical selection of result intervals to either rule out or rule in progression (Table 2). Bootstrapping indicated optimal performance of a 0/10/100 copies model. A result of 0-10 copies/ml was highly suggestive for remission (LR = 0.231, 95% CI 0.136-0.394) with a negative predictive value (NPV) of 88% (95% CI 81% to 92%). A result of 100 copies or higher was strongly suggestive for impending progression (LR = 5.077, 95% CI 3.112-8.265) with a positive predictive value (PPV) of 75% (95% CI 65% to 83%). A result of 10-100 copies/ml, encountered in 19% of blood draws, could be considered as gray zone without diagnostic value. Excluding all data points from the three oligo-progressors further decreased the LR of a result in 0-10 copies/ml to 0.181 (95% CI 0.0932-0.351). The associated NPV of 90% (95% CI 82% to 95%) indicated that ctDNA monitoring has utility to rule out systemic progression. Using >100 copies/ml as threshold, 64% of progression events could be detected earlier than standard of care with a median lead time of 10 (IQR 8-20) weeks. Result interval modeling for VAF resulted in similar performance by a 0%/0.25%/2.5% VAF thresholding (Table 1). Of note, given the observed median wild-type PIK3CA allele concentration of 3785 (IQR 2780-6550) copies/ml, the 2.5% VAF corresponds to 96 copies/ml in our sample set, indicating robustness of the modeling.

Table 2.

Optimal result intervals for ctDNA concentration (copies/ml or VAF%) to rule out or rule in progression within 12 weeks

	All subjects (n = 19)				Excluding localized progressors (n = 16)
	Progression within 12 weeks				Progression within 12 weeks
	Presence	Absence	LR	95% CI	Presence	Absence	LR	95% CI
PIK3CA mutant allele concentration (copies/ml)
0-10	12	86	0.231	0.136-0.394	8	71	0.181	0.0932-0.351
10-100	15	24	1.036	0.581-1.849	9	19	0.761	0.366-1.580
≥100	49	16	5.077	3.1119-8.265	49	16	4.919	3.064-7.896
PIK3CA variant allele fraction (VAF%)
0%-0.25%	13	88	0.245	0.147-0.407	7	72	0.156	0.0766-0.318
0.25%-2.5%	20	23	1.442	0.851-2.443	16	19	1.352	0.750-2.439
≥2.5%a	43	15	4.753	2.842-7.948	43	15	4.604	2.790-7.598
Combined (CA15-3, PIK3CA copies/ml) progression score
0.00-0.26	18	98	0.304	0.203-0.457	11	81	0.218	0.217-0.375
0.26-0.36	10	11	1.506	0.675-3.364	10	11	1.461	0.660-3.233
0.36-1.00	42	7	9.943	4.729-20.905	40	6	10.71	4.830-23.749

The upper and middle panels of the table show the likelihood ratio (LR, bold font) and its 95% CI for progression within 12 weeks based on the selected intervals of PIK3CA mutant allele concentration (copies/ml) or VAF%. The LR is presented for all 19 subjects (left) and for 16 subjects after exclusion of data points from 3 subjects who showed localized progression (right). The optimal rule-out and rule-in intervals were obtained by a 0/10/100 copies/ml model and 0%/0.25%/2.5% VAF model. The lower panel of the table displays clinically relevant result intervals for the calculated probability score based on the presence/absence of two consecutive rises of CA15-3 with at least 25% and the PIK3CA mutant allele concentration (copies/ml).

CA15-3, cancer antigen 15-3; CI, confidence interval; ctDNA, circulating tumor DNA; IQR, interquartile range; LR, likelihood ratio.

^aBased on observed median (IQR) wild-type PIK3CA concentration of 3875 (2780-6550) copies/ml plasma, 2.5% VAF corresponds to 96 copies/ml of mutant PIK3CA.

Improved PPV by combined CA15-3 and ctDNA modeling

Logistic regression analysis revealed that the PIK3CA mutant allele copies/ml [odds ratio (OR) = 1.0034, 95% CI 1.0018-1.0051, P = 0.001] and the presence of two consecutive rises of CA15-3 (OR = 7.3277, 95% CI 1.3447-39.9296, P = 0.0213) were both independent contributors to progression risk. However, neither the absolute CA15-3 level nor the CA15-3 relative change was found to be an independent contributor. Based on this, a new combined PIK3CA/CA15-3 probability score was calculated (lower panel of Table 1), which had a similar AUC of 0.846 (95% CI 0.786-0.894). Progression score result intervals were identified that rule out progression (LR = 0.303), strongly predict progression (LR = 10.114) and do not alter progression risk (LR = 1.506, gray zone). Overall, this generated a clinically meaningful interpretation in 89% of samplings. The combined CA15-3 and ctDNA risk scoring did not have a significant impact on NPV, but it did improve the PPV. Specifically, a progression score above 0.36 had 86% PPV for progression, compared to 75% for the model with only PIK3CA copies/ml above 100 copies/ml.

Discussion

Current research on ctDNA in breast cancer is divided into two application domains: detection of MRD in early cancers and surveillance of relapse in advanced cancers.

In MRD analysis, ctDNA is usually detected at a single or few predetermined time points after curative-intent therapy for prognostic reasons or to determine adjuvant systemic therapy. Detection of ctDNA post-treatment is linked with a worse prognosis in many early cancers.23, 24, 25 In early breast cancer9, 10, 11, 12, 13, 14 detectable ctDNA was associated with worse PFS and OS, and predicted poor responses to neoadjuvant therapy. The ChemoNEAR and Plasma DNA Study by Garcia-Murillas et al., which is the largest surveillance study on molecular relapse in early cancers,⁹ analyzed 695 data points from 101 patients over a monitoring period of 35 months. The study revealed an 11-month lead time for detecting relapse. However, like many landmark studies on MRD, the analysis of ctDNA in this study was qualitative and did not evaluate the diagnostic value of ctDNA concentration. The analysis was also limited to proportional hazards modeling at the cohort level. In advanced breast cancer, only three studies9, 10, 11, 12, 13, 14 have investigated the diagnostic value of ctDNA concentration for surveillance. Increasing ctDNA levels were associated with worse OS,²⁰ and early ctDNA dynamics after initiating a new therapy can be used as a surrogate marker for therapeutic response,²¹ with a reported lead time of around 3 months for relapse detection in 74% of patients, which is comparable to the 10 weeks’ lead time in 64% of patients in our dataset. Finally, Dawson et al.’s seminal study²² on an aggregated 141 blood samples obtained from 30 metastatic patients was the first to demonstrate ctDNA’s superiority over CA15-3 for predicting relapse at the cohort level.

To our knowledge, our report is the first independent confirmation of the Dawson study, using a similarly powered dataset of 341 samples from 19 patients. The report proposes a simple scheme for physicians to use ctDNA quantitatively in patient care (summarized in Figure 4) by selecting clinically useful result intervals for individual patient samples. In this scheme, a ctDNA concentration below 10 copies/ml has around 90% NPV for progression, providing a reassuring result in 49% of blood draws, which is much higher than the NPV around 50% reported in early breast cancer surveillance. ctDNA levels above 100 copies/ml showed 77% PPV for progression. The limited PPV reflects the complexity in the metastatic setting where many patients have high tumor burden and persistently high ctDNA levels despite no radiologically recognizable disease progression.

Figure 4.

Graphical abstract summarizes the proposed patient-centered ctDNA thresholding in mBC surveillance for risk-informed scheduling of clinical care. Blood draws with <10 copies/ml (0.25% VAF) indicate high NPV (safe zone, purple) for progression, and could be used to postpone radiology and reassure patients. Concentrations above 100 copies/ml (2.5% VAF) (danger zone, green) indicate high positive predictive value for progression within 12 weeks, and could trigger escalation to comprehensive genomic testing, multidisciplinary oncology consults and advance radiology. ctDNA levels between 10 and 100 copies/ml (0.25%-2.5% VAF), encountered in ∼10% of blood draws in mBC, represent a gray zone with no diagnostic value. CA15-3, cancer antigen 15-3; ctDNA, circulating DNA; mBC, metastatic breast cancer; NPV, negative predictive value; VAF, variant allele frequency.

We also selected result intervals for ctDNA levels expressed as VAF and found that the optimal intervals of 0%/0.25%/2.5% VAF fit very well with the 0/10/100 copies model. However, VAF can be biased due to fluctuations in wild-type allele concentration. As calibrated or orthogonally validated assays (digital PCR, molecular barcode-enabled sequencing) become more available, we support the recommendation from the 2018 joint review of ASCO and the College of American Pathologists²⁶ to abandon VAF and use ctDNA as a purely quantitative biomarker expressed as mutant allele copies/ml.

Our analysis found that using both ctDNA level and the presence of two consecutive rises of CA15-3 provided complementary diagnostic value, which can increase the PPV for impending progression to 86% and provide a meaningful result (rule in/rule out) in 89% of blood draws. Though such combined modeling is less straightforward for broad implementation, it can be integrated in most laboratory information systems.

Recent guidelines by ASCO,¹⁹ ESMO⁸ and the 2018 joint review by ASCO and the College of American Pathologists²⁶ discourage the routine use of ctDNA to monitor recurrence of early cancers or progression of advanced cancers due to lack of clinical validation. Our study provides evidence for the use of liquid biopsies in clinical decision making, which can improve patient comfort and generate health economics by allowing clinicians to schedule radiology and clinic visits based on ctDNA thresholding. Additionally, our findings suggest that ctDNA levels can be used for optimal timing of reflex testing by comprehensive genomic profiling in search for actionable resistance mechanisms.²⁰ Our results support the Medicare and Medicaid local coverage determination guidelines, which state that ctDNA can identify recurrence earlier and with higher accuracy than other forms of surveillance.²³

Our study has limitations. Firstly, independent validation of the proposed ctDNA thresholds in metastatic breast cancer and other cancer types is needed. Absolute ctDNA quantification will allow pooling of similar datasets.²² Secondly, we used a tumor genotype-informed approach²³ and restricted the monitoring to hotspot mutations of PIK3CA for which singleplex commercial primers with reported analytical sensitivities were available.²⁰ Personalized targeted PCR may not be widely implementable due to complexities in assay development and associated variations in limits of detection, and a bespoke approach by broad NGS panels carries prohibitive costs and requires deep sequencing to achieve sufficient analytical sensitivity. Thirdly, since the liquid biopsy results were not communicated to the medical oncologists during the study, there was no anticipated change in anticancer treatment in the absence of clinical or radiological progression. The study was observational and not designed to influence the treatment decisions as the ddPCR assays used had not been clinically validated in our hospital. Particularly in HR+ HER2− advanced breast cancers, a widespread adoption of ctDNA analysis is expected as a first-choice test to identify actionable ESR1 mutations based on recently concluded or ongoing trials (PADA-1, EMERALD, SERENA-6).^15,17,27 Since these secondary ESR1 mutations are typically subclonal relative to the driver mutations in genes such as PIK3CA, TP53 or GATA3 (as shown in the patients’ graphic summary in Supplementary Material S2, available at https://doi.org/10.1016/j.esmoop.2024.102235), it might be interesting to opt for multiplex PCR assays or NGS panels for combined screening of the driver mutations for early detection of impending progression and the ESR1 mutations as underlying actionable cause.

In summary, our study confirms that ctDNA level is diagnostically superior to CA15-3²² and shows potential complementarity of both biomarkers for surveillance of metastatic breast cancer. Our patient-centered approach to ctDNA data handling resulted in an easy-to-use flagging system for impending progression and ctDNA risk-informed scheduling of standard care. Validation in other datasets and pooled analysis in breast cancer studies, as well as testing on other cancer types, such as lung and colorectal cancer, are needed.