Purpose:

We previously demonstrated that high levels of circulating methylated DNA are associated with subsequent disease progression in women with metastatic breast cancer (MBC). In this study, we evaluated the clinical utility of a novel liquid biopsy-breast cancer methylation (LBx-BCM) prototype assay using the GeneXpert cartridge system for early assessment of disease progression in MBC.

Experimental Design:

The 9-marker LBx-BCM prototype assay was evaluated in TBCRC 005, a prospective biomarker study, using plasma collected at baseline, week 4, and week 8 from 144 patients with MBC.

Results:

At week 4, patients with MBC with high cumulative methylation (CM) had a significantly shorter median PFS (2.88 months vs. 6.60 months, P = 0.001) and OS (14.52 months vs. 22.44 months, P = 0.005) compared with those with low CM. In a multivariable model, high versus low CM was also associated with shorter PFS (HR, 1.90; 95% CI, 1.20–3.01; P = 0.006). Change in CM from baseline to week 4 (OR, 4.60; 95% CI, 1.77–11.93; P = 0.002) and high levels of CM at week 4 (OR, 2.78; 95% CI, 1.29–5.99; P = 0.009) were associated with progressive disease at the time of first restaging. A robust risk model based on week 4 circulating CM levels was developed to predict disease progression as early as 3 months after initiating a new treatment.

Conclusions:

The automated LBx-BCM prototype assay is a promising clinical tool for detecting disease progression a month after initiating treatment in women with MBC undergoing routine care. The next step is to validate its clinical utility for specific treatments.

Translational Relevance

Predictive clinical biomarkers to identify early disease progression are needed for women with metastatic breast cancer given the heterogeneity of the disease. These biomarkers could also help minimize toxicity and optimize the use of available and effective therapies. We demonstrate the successful translation of a panel of nine DNA methylated markers shed in the blood of patients with breast cancer from a laboratory assay to an automated liquid biopsy Breast Cancer Methylation (LBx-BCM) assay for identifying early disease progression and predicting survival outcomes. In our study, high versus low cumulative methylation based on the median cut point was associated with significantly shorter progression-free and overall survival. In addition, a robust prediction model was developed to detect disease progression as early as 3 months after initiating a new treatment based on circulating cumulative methylation levels measured at week 4.

New treatment options for MBC, including biomarker targeted therapies, have led to meaningful improvements in survival and in some situations, a treatment approach akin to other chronic diseases, where clinicians work through a decision tree of treatment options while carefully monitoring progress (1). Given the heterogeneity of MBC and the cost to conduct randomized clinical trials, it is not feasible to evaluate the efficacy of each treatment sequence. In clinical practice, medical oncologists frequently initiate treatment changes when there is evidence of disease progression on imaging or toxicity. There is currently no biomarker in clinical practice proven to predict early disease progression in women with MBC (2). The hope is that predictive biomarkers can be used for the early identification of patients unlikely to benefit from prescribed therapies. This type of biomarker would also minimize exposure to toxic therapies and accelerate decisions toward initiating new therapies or palliative measures.

Epigenetic alterations are among the most common molecular abnormalities in human cancers (3), and are a vital source of information about underlying disease. Tumors commonly release altered DNA into the bloodstream; these aberrant DNAs could serve as valuable markers of disease (4, 5). Several studies have tested methylation markers in the blood for diagnosis (6–9). However, none have been validated for the prediction of disease progression in women with breast cancer (10–12). Using genes observed to be methylated in tumor tissue of patients with breast cancer and detectable in their blood, but not found in breast tissue or blood of women without cancer (13), we developed and analytically validated a highly sensitive high-throughput quantitative multiplex methylation-specific polymerase chain reaction assay called “cMethDNA” to detect circulating cell-free methylated DNA (13). Next, in TBCRC 005, we prospectively confirmed the clinical utility and prognostic power of circulating DNA methylation for both disease progression and survival in women with MBC (14). We also compared the cumulative methylation index (CMI) based on cMethDNA to circulating tumor cells (CTC). Although both CTCs ≥5 cells/mL of blood and high CMI were associated with worse prognosis, the CMI level at week 4 was more sensitive than was a CTC level ≥5 cells/mL at identifying women with progressive disease (PD).

Given the promising results of the manual cMethDNA research assay, we devoted efforts to developing an assay that could be used in the clinical setting. A collaboration with Cepheid led to the development of the Liquid Biopsy Breast Cancer Methylation (LBx-BCM) prototype assay (currently for Research Use Only, not for use in patient management), based on the principles of the original cMethDNA assay, but running on the widely available GeneXpert cartridge system. In a recent paper, we reported the analytical validity of this automated cartridge assay that can generate reproducible DNA methylation results within 5 hours and potentially be available at the point-of-care (15). In addition, the cartridge-based assay requires fewer laboratory personnel to run and less intensive training. The assay can also be used with both plasma and serum samples (14). Cepheid GeneXpert instrument systems are widely deployed in the United States, Europe, and low-middle income countries. The system uses bar-coded, automated cartridges for each specimen and is scalable, capable of handling from one to 80 individual cartridges simultaneously. The conditions of the LBx-BCM prototype assay were optimized (which included a rigorous analytical validation) to test 9 genes that reflect the three major breast cancer subtypes in two cartridges, each of which analyzes 4 to 5 genes plus actin as the loading control. This included optimization of the assay to detect as low as 75 copies of methylated DNA of each of the 9 genes. Importantly, the new automated clinical assay can interchangeably use serum, EDTA-, or STRECK-preserved plasma with similar signal detection (15). The assay uses a total of 1.0 mL of blood divided into two cartridges (0.5 mL per cartridge). Consistent with most PCR techniques, our results show incrementally lower efficiency below 300 copies of target DNA per assay. The goal of the study was to establish the prognostic and predictive performance of the new LBx-BCM in women with MBC for identifying disease progression early so that ineffective therapies can be stopped early, and toxicities can be minimized.

Study population and design

A total of 144 of 185 study participants enrolled in TBCRC 005 were included in this study (see Supplementary Fig. S1 for study schema). TBCRC 005 was the first prospective study designed to test the clinical utility of circulating DNA methylation biomarkers in blood from individuals with MBC. Patients were enrolled between January 2007 and June 2009. Plasma and serum were collected at baseline immediately before starting a new regimen, at 4 weeks, and at first restaging which was between 8 and 12 weeks. Treatment response was documented at first restaging by the treating physician. Eligible participants for TBCRC 005 were females 18 years of age or older with histologically confirmed MBC, measurable disease, an Eastern Cooperative Oncology Group (ECOG) performance status of 0 to 2, who were starting a new systemic therapy and being treated at one of seven participating US academic medical centers. To evaluate the potential utility of novel assays, we chose to design the study to reflect real-world clinical practice and therefore the choice of therapy was at the discretion of the treating physician.

Measurable/evaluable disease was defined as a lesion 1 cm or greater on CT scan or MRI or a superficial/palpable lesion 1 cm or greater. Patients with a diagnosis of second cancer in the previous 5 years were excluded, except for basal or squamous cell carcinoma of the skin and/or cervical carcinoma in situ. An additional criterion for this study was that all participants had to have 1 mL of available plasma collected at both baseline and 4 weeks. All patients provided written informed consent. The institutional review board (IRB) at each study site approved this study. The study was compliant with Good Clinical Practice Guidelines and the Declaration of Helsinki.

To evaluate within and between batch reproducibility, EDTA plasma from 36 patients with MBC enrolled in Johns Hopkins Breast Cancer Repository (a similar study population to TBCRC 005) was pooled, and 1.0 mL aliquots were made. Two aliquots were included in each of the 22 batches.

Sample preparation

Laboratory personnel was blinded to any sample details or associated clinical data. Samples were processed in random order ensuring that repeated samples from the same individual were analyzed in the same batch. Individual samples were pre-processed for LBx-BCM by mixing 1.0 mL of plasma with proteinase K and lysis buffer. Ethanol was added and the sample was transferred in its entirety to a single cartridge and treated with sodium bisulfite as described previously (15). After a 2-hour step, unmethylated cytosine was converted to uracil (a DNA sequence change that is detectable by PCR), whereas methylated DNA is protected and remains unchanged. The bisulfite-treated DNA sample was then split equally and transferred to two methylation detection cartridges where two separate PCR reactions occurred in series: PCR 1 is methylation-independent, nested multiplexed PCR that serves as a pre-amplification step, and PCR 2 is a methylation-specific qRT-PCR reaction that measures amplicons generated in the first PCR. PCR 2 uses six fluorophores in the multiplex to quantify four to five methylated targets/cartridge and a β-actin (ACTB) reference gene, nine targets in a total of two cartridges. A dilution series is used to provide reference levels for each marker. Briefly, target DNA is diluted to 0, 75, 150, and 300 copies and spiked into commercial, normal pooled plasma to provide a reference dilution curve. Replicates (n = 10–12) are evaluated in the cartridge.

Calculation of DNA methylation

Methylation in the following 9 target genes was analyzed: AKR1B1, TM6SF1, ZNF671, TMEFF2, plus ACTB reference in Cartridge A; COL6A2, HIST1H3C, RASGRF2, HOXB4, and RASSF1 plus ACTB reference in Cartridge B. Normalized methylation levels were calculated in several steps. (1). Calculate Ct values: The GeneXpert software assigns Ct values describing the PCR cycle threshold at which fluorescence signal exceeds background, defaulting to Ct = 45 if no signals were detectable during the run. (2). Normalize to the total amount of DNA: Ct values are normalized to the total abundance of DNA in each sample, represented by β-actin levels (ACTB). The resulting ΔCt = Ctgene – CtACTB is the difference between the Ct of the individual target gene and the Ct of ACTB within the cartridge for each marker. If some samples have negative ΔCt because the estimated abundance of the gene was slightly higher than actin, all samples were similarly transformed by a constant value to give positive integers for that gene (3). Standardize individual markers to a common level. The ΔCt value representing standard 300 copies of DNA was estimated from the dilution experiment described above, and ΔCt values for each marker were scaled to a common level (4). Determine the lower limit of detection. If ΔCt (gene – ACTB) was higher than the replicate median of 300 copies + 13 ΔCt units, then we adjusted to ΔCt (gene – ACTB) = 0, thus removing signals from the analysis that were too low to reliably quantitate (less than 0.03 copies of target) (5). Invert and scale measures. Scaled ΔCt were inverted so that higher values corresponded to a greater abundance of DNA and multiplied by a factor of 1,200 for compatibility with measures made using the laboratory-based cMethDNA technique. Gene methylation (M) = [1 / ΔCt (gene – ACTB)] × 1,200. Finally, cumulative methylation CM (CM) was calculated as the sum of all M for the 9 genes in the panel.

Data analysis

Inter- and intrabatch variation was evaluated using the coefficient of variation (CV) for CM and each of the nine markers (16). Median progression-free survival (PFS) and overall survival (OS) with 95% CIs were estimated using the Kaplan–Meier method. Survival distributions were compared between patients with high and low CM at week 4 using the log-rank test, controlling for age, ethnicity, body mass index (BMI), prior therapy, tumor subtype, and visceral/nonvisceral disease. High and low CM levels were based on the median. Change in CM from baseline to week 4 was also assessed. The change was defined simply as CM week 4 - CM baseline. Response at first restaging was categorized as progressive disease (PD), partial response/complete response (PR/CR), or stable disease (SD) based on the treating physician's assessment. Spearman ρ was used to detect monotonic trends. Similar multivariable models were used to evaluate the independent effect of CM after adjustment for CTC, CA27–29, and CEA. Predictive models of early PD were assessed by area under a ROC curve (AUC) and reported with bootstrap 95% confidence intervals (95% CI). Time dependent ROC (17) was used to estimate the AUC when predicting progression at specific time points (3, 6, and 9 months). Models requiring training were evaluated in leave-one-out cross-validation to achieve unbiased estimates of performance. Several models using a binary threshold, random forest, and linear discriminant were tested to evaluate combinations of individual markers, cumulative methylation for all 9 genes, and both absolute methylation levels at week 4 as well as change in methylation at week 4 versus baseline. All tests were two-sided and considered statistically significant at P < 0.05 and were performed using the R statistical software suite (available at http://www.r-project.org) with both standard packages and custom code.

Data availability statement

Deidentified patient data and assay results are available upon request to the corresponding authors.

Table 1 describes the patient characteristics for 144 females with MBC enrolled in TBCRC 005. In 124 of the 144 patients, a third plasma sample was available at the time of first restaging, which was between 8 and 12 weeks after the start of treatment. The median age of participants was 56 years, 19% were Black, and 87% were postmenopausal. Most patients were diagnosed with an ER-positive/PR-positive/HER2-negative tumor; 23% were HER2-positive and 19% were triple negative. Twenty-six percent of patients had no prior chemotherapy and/or endocrine-based therapy and 50% had elevated CTCs (≥5 cells/mL of blood at baseline Cell Search CTC System). The median CM at baseline was 562.63, at week 4 was 229.69, and at week 8 was 8216.61. All patients were followed until death or the last follow-up. The median follow-up time of the cohort was 71.1 months (about 6 years). The inter- and intrabatch CV for CM were 0.06 and 0.05, respectively. The overall CV for individual genes ranged from 0.07 to 0.55 and the within batch CV for individual genes ranged from 0.05 to 0.33. (See Supplementary Table S1 for further details.)

Table 1.

Baseline patient characteristics of the analytic population.

CharacteristicAnalytic population (N = 144)
Age, years, median (range) 56 (28–84) 
Race 
aWhite 117 (81) 
 Black 27 (19) 
Menopausal status 
 Postmenopausal 122 (87) 
 Perimenopausal/premenopausal 18 (13) 
BMI, kg/m2, median (range) 26 (18–44) 
bTumor phenotype 
 ER positive/PR positive/HER2 negative 84 (58) 
 HER2 positive (any ER) 33 (23) 
 Triple negative 27 (19) 
Disease burden 
 Visceral only (liver, lung, brain) 25 (17) 
 Non visceral only (bone and/or soft tissue) 49 (34) 
 Both 70 (49) 
Prior therapy 
 None 38 (26) 
 Chemotherapy only 32 (22) 
 Endocrine-based therapy only (ET) 34 (24) 
 Chemotherapy/ ET 40 (28) 
Elevated CTC level (≥5) 63 (50) 
Progression-free survival, months, median (95% CI) 4.8 (3.72–6.00) 
Follow Up, months, median (95% CI) 71.1 (66.70–NA) 
Cumulative Methylation levels 
 Baseline, median (range) 562.63 (0.00–2706.77) 
 Week 4, median (range) 229.69 (0.00–2075.94) 
 Week 8, median (range) 216.61 (0.00–3086.98) 
CharacteristicAnalytic population (N = 144)
Age, years, median (range) 56 (28–84) 
Race 
aWhite 117 (81) 
 Black 27 (19) 
Menopausal status 
 Postmenopausal 122 (87) 
 Perimenopausal/premenopausal 18 (13) 
BMI, kg/m2, median (range) 26 (18–44) 
bTumor phenotype 
 ER positive/PR positive/HER2 negative 84 (58) 
 HER2 positive (any ER) 33 (23) 
 Triple negative 27 (19) 
Disease burden 
 Visceral only (liver, lung, brain) 25 (17) 
 Non visceral only (bone and/or soft tissue) 49 (34) 
 Both 70 (49) 
Prior therapy 
 None 38 (26) 
 Chemotherapy only 32 (22) 
 Endocrine-based therapy only (ET) 34 (24) 
 Chemotherapy/ ET 40 (28) 
Elevated CTC level (≥5) 63 (50) 
Progression-free survival, months, median (95% CI) 4.8 (3.72–6.00) 
Follow Up, months, median (95% CI) 71.1 (66.70–NA) 
Cumulative Methylation levels 
 Baseline, median (range) 562.63 (0.00–2706.77) 
 Week 4, median (range) 229.69 (0.00–2075.94) 
 Week 8, median (range) 216.61 (0.00–3086.98) 

Note: Data are presented as No. (%) unless otherwise indicated.

Abbreviations: BMI, body mass index; CTC, circulating tumor cells; ER, estrogen receptor; PR, progesterone receptor.

aIncludes one Asian.

bBreast tumor phenotype.

CM and disease outcomes

Figure 1A and B display the Kaplan–Meier curves for progression-free survival (PFS) and overall survival (OS) by high versus low methylation at week 4 based on a median CM of 229.69. We confirmed that the median PFS was significantly shorter for patients with high CM (2.88 months; 95% CI, 2.52–4.08 months) versus low week 4 CM (6.60 months; 95% CI, 5.76–8.52 months; P = 0.001) based on the new LBx-BCM assay. Median OS was also significantly shorter for patients with high CM (14.52 months; 95% CI, 11.16–19.80) compared with low CM (22.44 months; 95% CI, 20.16–30.60; P = 0.005). In multivariate analyses, patients with high versus low week 4 CM levels had a shorter PFS (HR, 1.90; 95% CI, 1.20–3.01; P = 0.006) after adjustment for age, ethnicity, menopausal status, BMI, tumor phenotype, visceral tumor burden, and prior systemic therapy. A similar result was observed for OS (HR, 1.19; 95% CI, 0.74–1.91; P = 0.482) after adjustment for the same covariates, although the association was not statistically significant. Details of both analyses are described in Supplementary Table S2.

Figure 1.

A and B, Kaplan–Meier curves for PFS and OS among women with metastatic breast cancer stratified by CM level beginning at week 4. High versus low CM was based on the median (CM > median versus CM ≤ median.) The corresponding multivariate Cox regression models are shown in Table 2.

Figure 1.

A and B, Kaplan–Meier curves for PFS and OS among women with metastatic breast cancer stratified by CM level beginning at week 4. High versus low CM was based on the median (CM > median versus CM ≤ median.) The corresponding multivariate Cox regression models are shown in Table 2.

Close modal

CM and disease status

The association between CM and disease status at first restaging was evaluated next. At baseline, 5% (7/144) of patients had no detectable CM. For each individual, absolute level of CM at week 4 and the change in CM from baseline to week 4 is shown in Fig. 2A. In 77% of patients, absolute CM levels decreased in the first 4 weeks and then remained stable through time to the first restaging. In 18% of patients with stable disease and 37% of patients with PD, an increase in CM was observed from baseline to week 4. Further, there was no increase in CM levels among responders. CM levels were stable from week 4 to week 8.

Figure 2.

A, The relationship between CM level at week 4 and the change in CM level between baseline and Week 4. B and C use box plots to display the relationship between disease status at first restaging and week 4 CM level or change in CM level (baseline – week 4).

Figure 2.

A, The relationship between CM level at week 4 and the change in CM level between baseline and Week 4. B and C use box plots to display the relationship between disease status at first restaging and week 4 CM level or change in CM level (baseline – week 4).

Close modal

Figure 2B displays week 4 CM by disease status using box plots. The median CM at week 4 was highest in women with PD (CM = 57.37) compared with women with SD (CM = 41.48) and responsive disease (CM = 12.81), which includes individuals with either PR or CR. A significant trend in CM levels was observed on the basis of the response at first restaging (Spearman ρ = 0.33; P = 6.76e−05). A significant trend was also observed for change in CM from baseline to week 4 by disease status at first restaging as shown in Fig. 2C (Spearman ρ = 0.22; P = 0.01). and for week 8 CM levels but not for a change in CM from week 4 to week 8 (Supplementary Fig. S2).

In the univariate analysis shown in Table 2, high versus low CM at week 4 was associated with PD at first restaging (OR, 3.25; 95% CI, 1.57–7.87; P = 0.001). The OR was attenuated in a multivariable model (OR, 2.78; 95% CI, 1.29–5.99; P = 0.01), also shown in Table 2. A change in CM from baseline to week 4 was also associated with PD at first restaging in univariate analysis (OR, 3.81; 95% CI, 1.64–8.82; P = 0.002). The OR increased to 4.60 (95% CI, 1.77–11.93; P = 0.002) in a multivariable model. Similar patterns were observed for the association between CM levels at week 8 and PD (OR, 5.12; 95% CI, 2.02–12.94; P = 0.001) and change in CM between week 4 and week 8 and PD (OR, 3.273; 95% CI, 1.36–7.85; P = 0.01; see Supplementary Table S3). Supplementary Figure S3 displays baseline CM levels by disease status using box plots. The median CM was highest for PD (CM = 713.43), compared with SD (CM = 533.65), and responsive disease (CM = 425.48), which includes PR and CR. There was no significant trend observed (Spearman ρ = 0.03; P = 0.70).

Table 2.

Association of week 4 CM and changes in CM with disease status at first restaging.

PredictoraCM level at week 4bChange in CM level btw baseline and week 4
OR (95% CI)POR (95% CI)P
Univariate analysis 
 CM 3.25 (1.57–7.87) 0.001 3.81 (1.64–8.82) 0.002 
 Age (continuous) 0.97 (0.94–1.00) 0.086 0.97 (0.94–1.00) 0.070 
Multivariate analysis 
 CM 2.78 (1.29–5.99) 0.009 4.60 (1.77–11.93) 0.002 
 Age (continuous) 0.97 (0.94–1.01) 0.186 0.97 (0.94–1.01) 0.183 
Race 
cWhite (ref) 1.00  1.00  
 Black 1.09 (0.39–3.06) 0.870 1.23 (0.43–3.53) 0.697 
Tumor Phenotype 
 ER +ve/PR +ve/ HER2 −ve (ref) 1.00  1.00  
 HER2 (any ER) +ve 0.80 (0.29–2.23) 0.675 1.30 (0.46–3.72) 0.623 
 Triple -ve 0.67 (0.20–2.27) 0.523 0.75 (0.21–2.62) 0.650 
Prior therapy 
 None (ref) 1.00  1.00  
 Chemotherapy 1.44 (0.48–4.37) 0.519 1.29 (0.42–4.02) 0.656 
 Endocrine-based therapy (ET) 0.67 (0.20–2.24) 0.514 0.78 (0.23–2.67) 0.698 
 Chemo/ ET 0.83 (0.26–2.59) 0.744 1.05 (0.33–3.32) 0.932 
Disease burden 
 Visceral/both (ref) 1.00  1.00  
 Nonvisceral 0.69 (0.30–1.54) 0.362 0.65 (0.28–1.47) 0.301 
BMI, kg/m2 
 <25 (ref) 1.00  1.00  
 25–30 0.44 (0.17–1.18) 0.104 0.38 (0.14–1.00) 0.051 
 >30 1.34 (0.51–3.56) 0.555 0.87 (0.31–2.45) 0.796 
Menopausal status 
 Postmenopausal (ref) 1.00  1.00  
 Premenopausal 0.91 (0.25–3.31) 0.887 0.92 (0.25–3.39) 0.895 
PredictoraCM level at week 4bChange in CM level btw baseline and week 4
OR (95% CI)POR (95% CI)P
Univariate analysis 
 CM 3.25 (1.57–7.87) 0.001 3.81 (1.64–8.82) 0.002 
 Age (continuous) 0.97 (0.94–1.00) 0.086 0.97 (0.94–1.00) 0.070 
Multivariate analysis 
 CM 2.78 (1.29–5.99) 0.009 4.60 (1.77–11.93) 0.002 
 Age (continuous) 0.97 (0.94–1.01) 0.186 0.97 (0.94–1.01) 0.183 
Race 
cWhite (ref) 1.00  1.00  
 Black 1.09 (0.39–3.06) 0.870 1.23 (0.43–3.53) 0.697 
Tumor Phenotype 
 ER +ve/PR +ve/ HER2 −ve (ref) 1.00  1.00  
 HER2 (any ER) +ve 0.80 (0.29–2.23) 0.675 1.30 (0.46–3.72) 0.623 
 Triple -ve 0.67 (0.20–2.27) 0.523 0.75 (0.21–2.62) 0.650 
Prior therapy 
 None (ref) 1.00  1.00  
 Chemotherapy 1.44 (0.48–4.37) 0.519 1.29 (0.42–4.02) 0.656 
 Endocrine-based therapy (ET) 0.67 (0.20–2.24) 0.514 0.78 (0.23–2.67) 0.698 
 Chemo/ ET 0.83 (0.26–2.59) 0.744 1.05 (0.33–3.32) 0.932 
Disease burden 
 Visceral/both (ref) 1.00  1.00  
 Nonvisceral 0.69 (0.30–1.54) 0.362 0.65 (0.28–1.47) 0.301 
BMI, kg/m2 
 <25 (ref) 1.00  1.00  
 25–30 0.44 (0.17–1.18) 0.104 0.38 (0.14–1.00) 0.051 
 >30 1.34 (0.51–3.56) 0.555 0.87 (0.31–2.45) 0.796 
Menopausal status 
 Postmenopausal (ref) 1.00  1.00  
 Premenopausal 0.91 (0.25–3.31) 0.887 0.92 (0.25–3.39) 0.895 

Abbreviations: BMI, body mass index; Chemo, chemotherapy; ER, estrogen receptor; ET, endocrine-based therapy; PR, progesterone receptor; +ve, positive; −ve, negative.

aCM binary measure: CM level > median versus CM level ≤ median.

bΔCM binary measure > 0 vs. ΔCM ≤ 0.

cIncludes one Asian.

Next, we assessed whether high CM levels were independently associated with PD in the presence of other circulating markers that have been studied or used for monitoring disease progression in patients with MBC (see Supplementary Tables S4–S6). The association between Week 4 CM levels at first restaging and PD remained strongly significant in multivariable models that either included CTC (OR, 6.93; P = 0.002), CEA (OR, 5.48; P = 0.003), or CA27–29 (OR, 4.18; P = 0.012). Of note, the association between circulating CTC (OR, 1.04; P = 0.952) and CEA (OR, 1.16; P = 0.747) and PD were not statistically significant. However, circulating CA27-29 remained statistically significant after adjustment for CM (OR, 4.18; P = 0.004).

Development of a new predictive model for early disease progression

Finally, we developed and evaluated robust models for women with MBC to predict disease progression at 3, 6, and 9 months and first restaging based on CM measurements. Figure 3A and B display time updated ROC curves predicting disease progression based on high versus low week 4 CM and change in CM from baseline to week 4, respectively. Week 4 CM outperformed a change in CM based on disease progression at each time point. The combined model that includes both week 4 CM (high vs. low) and change in CM yielded the best performance across the multiple end points (AUC = 0.668–0.733). Alternative models, ranging from random forest models to a simple threshold model in which individual markers were binarized to low versus high methylation yielded AUCs ranging from 0.557 to 0.679 (similar to naive week 4 CM), reflecting the robustness of the overall model (see Supplementary Fig. S4). Time updated ROC analyses combining high versus low week 8 CM and PD as well as change in CM from week 4 to week 8 and PD was slightly weaker (see Supplementary Fig. S5).

Figure 3.

Time updated ROC curves for (A) CM level at week 4, (B) change in CM level (Week 4 – Baseline), and (C) the Combination of week 4 CM level and Change in CM level (Week 4 – Baseline) using linear discriminant analysis. On each graph, the black curve shows PD at first restaging. The red, blue, and green ROC curves on each panel display the ROC curve for disease progression at 3, 6, and 9 months, respectively.

Figure 3.

Time updated ROC curves for (A) CM level at week 4, (B) change in CM level (Week 4 – Baseline), and (C) the Combination of week 4 CM level and Change in CM level (Week 4 – Baseline) using linear discriminant analysis. On each graph, the black curve shows PD at first restaging. The red, blue, and green ROC curves on each panel display the ROC curve for disease progression at 3, 6, and 9 months, respectively.

Close modal

The results of this real-world clinical study support the continued development and validation of this new DNA methylation LBx-BCM prototype assay in women with MBC, as a tool to predict early disease progression within 1 month of initiating therapy. This study addresses an unmet clinical need given the increasing armamentarium of drugs with similar response rates available for use in patients with MBC. The LBx-BCM prototype assay could enable medical oncologists to change less effective therapies early after the initiation of treatment and thereby minimize unnecessary toxicity for patients. Using a novel DNA methylation assay that could be used in the clinical setting, we confirmed our 2017 findings that higher week 4 CM levels are associated with worse disease outcomes in women with MBC (14). We identified that both week 4 CM levels and change in CM from baseline to week 4 are independent determinants of disease progression. CM continued to be an independent predictor of PD after adjustment for CTC, CEA, and CA27–29, which are frequently measured molecular markers of disease status in women with MBC. A new risk model with good discriminatory power was also developed to predict disease progression as early as 4 weeks after initiating a new treatment.

There is a paucity of data on the dynamics of circulating methylation levels over time in patients with cancer. The results of our study suggest that in most patients with MBC, methylation decreases immediately after treatment is initiated which is likely due to the initial response to treatment. Methylation levels then either remain at a new baseline or increase. More frequent measurements within the first 6 weeks could inform whether there is an optimal time to measure CM after initiating treatment. The methylated marker panel used in this study includes 9 of the 10 genes evaluated in the original TBCRC 005 study and a new gene ZNF671 (associated with ER-negative breast cancer; ref. 14). In addition, actin was added as a reference control in each of the two cartridges for a total of 4 to 5 genes of interest plus one reference gene per cartridge. In developing our panel, we sought markers that were robustly and consistently methylated in breast tumors, to detect tumor-derived DNA in as many women as possible, and our findings here suggest that this is true in practice as well. We believe that our markers are largely interchangeable indicators of the presence of a tumor, rather than specialized indicators of the state of the tumor, so that differences in CM reflect differences in overall tumor burden.

To our knowledge, this is the first methylation-based prediction model focused on early disease progression in patients with MBC. The predictive risk model was robust when tested under many different statistical assumptions. Alternative models from binary threshold to random forest did not perform better than using CM values based on week 4 or change in CM. The discriminatory performance of our prediction model was well within the range of breast cancer risk models being used in clinical practice that report AUCs ranging from 0.53 to 0.66 (18, 19). The model prediction was also in line with a recent model developed to predict survival outcomes in women with early-stage triple-negative breast cancer (20) reporting an AUC of 0.59 to 0.61.

The strengths of our study include the multicenter prospective design, the repeated samples, the careful selection of genes and the automated assay, and the ability to adjust for other tumor molecular markers. Limitations of the study include the lack of central adjudication for outcomes at first restaging, the lack of repeated measures before week 4, and a large enough sample size to evaluate significant differences for specific treatments used in various breast cancer phenotypes.

In conclusion, the automated, easy to use LBx-BCM prototype assay, a novel DNA methylation test, was able to successfully identify early disease progression after initiation of therapy in women with MBC. Further development of the LBX-BCM assay will include evaluation of weekly CM level after treatment initiation in women with metastatic disease to identify the optimum time to measure CM, and subsequent validation and refinement of our model in similar patient populations as well as in early-stage disease.

K. Visvanathan reports grants from Cepheid during the conduct of the study, also has a patent for US 10,316,361 issued. L. Cope reports grants from Cepheid during the conduct of the study; also has a patent for US 10,316,361 issued. M. Fackler reports grants and personal fees from Cepheid during the conduct of the study; personal fees from Cepheid outside the submitted work, also has a patent 20200190586 pending, a patent 10450609 issued, a patent 9416404 issued, a patent 20150057188 issued, a patent 8822155 issued, a patent 20140220561 issued, a patent 8062849 pending, a patent 7858317 issued, a patent 8062849 issued, a patent 2013266341 issued, a patent 2852690 issued, a patent for ZL201380039009.7 issued, a patent for 2015-51732 pending, a patent 2874035 issued, and a patent for US-2020-0190586 pending. L.J. Sokoll reports nonfinancial support from Veridex, LLC during the conduct of the study. A. Forero-Torres reports other support from SeaGen outside the submitted work. J.N. Ingle reports grants from The Breast Cancer Research Foundation during the conduct of the study. N.U. Lin reports grants from Genentech, Pfizer, Merck, and Zion Pharmaceuticals; grants and personal fees from SeaGen, Olema Pharmaceuticals, and AstraZeneca; and personal fees from DSI, Prelude Therapeutics, Aleta Biopharma, Voyager Therapeutics, Blueprint Medicines, Janssen, Artera, and Up to Date outside the submitted work. R. Nanda reports personal fees from Astrazeneca, BeyondSpring, Fujifilm GE, Gilead, Infinity, iTeos, Merck, OBI, Oncosec, Sanofi, and Seagen, and grants from Arvinas, AstraZeneca, Celgene, Corcept Therapeutics, Genentech/Roche, Gilead/Immunomedics, Merck, OBI, Pharma, OncoSec, Pfizer, Relay, Seattle Genetics, Sun Pharma, and Taiho outside the submitted work. A.M. Storniolo reports personal fees from Astra Zeneca outside the submitted work. S.A. Campbell reports other support from Cepheid outside the submitted work. M. Bates reports other support from Cepheid and Danaher during the conduct of the study; other support from Cepheid and Danaher outside the submitted work. A.C. Wolff reports grants from Cepheid during the conduct of the study; also has a patent for methylation in breast cancer issued. S. Sukumar reports grants and personal fees from Cepheid during the conduct of the study; personal fees from Cepheid outside the submitted work, also has a patent 20200190586 pending, a patent 10450609 issued, a patent 9416404 issued, a patent 20150094222 pending, a patent 20150057188 issued, a patent 8822155 issued, a patent 20140220561 issued, a patent 8062849 pending, a patent 7858317 issued, a patent 8062849 issued, a patent 2013266341 issued, a patent 2852690 issued, a patent for 201380039009 issued, and a patent 2874035 issued. No disclosures were reported by the other authors.

K. Visvanathan: Conceptualization, data curation, methodology, writing–original draft, project administration, writing–review and editing. L. Cope: Conceptualization, data curation, formal analysis, supervision, methodology, writing–review and editing. M.J. Fackler: Conceptualization, data curation, methodology, project administration, writing–review and editing. M. Considine: Formal analysis, writing-review and editing. L. Sokoll: Data curation, writing–review and editing. L.A. Carey: Resources, data curation, writing–review and editing. A. Forero-Torres: Resources, data curation, writing–review and editing. J.N. Ingle: Resources, data curation, writing–review and editing. N.U. Lin: Resources, data curation, writing–review and editing. R. Nanda: Resources, data curation, writing–review and editing. A.M. Storniolo: Resources, data curation, writing–review and editing. S. Tulac: Resources, data curation, writing–review and editing. N. Venkatesan: Resources, data curation, methodology, writing–review and editing. N.C. Wu: Data curation, writing–review and editing. S. Marla: Resources, data curation, writing–review and editing. S. Campbell: Data curation, writing–review and editing. M. Bates: Resources, data curation, writing–review and editing. C.B. Umbricht: Writing–review and editing. A.C. Wolff: Conceptualization, resources, writing–review and editing. S. Sukumar: Conceptualization, resources, data curation, writing–review and editing.

This work was supported by a DOD grant (W81XWH-18–1–0018) and a Cepheid research agreement (#90066820) to S. Sukumar as well as funding support to the TBCRC from The Breast Cancer Research Foundation and Susan G. Komen. CellSearch reagents are provided by Veridex, LLC. We would also like to thank Jennifer Lehman, Akhila Hadji, and Betty May for their help with various aspects of data checking and sample aliquoting.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

1.
NCCN
.
Breast Cancer Guidelines Version 2
.
2022
; https://www.nccn.org.
2.
Henry
NL
,
Somerfield
MR
,
Dayao
Z
,
Elias
A
,
Kalinsky
K
,
McShane
LM
, et al
.
Biomarkers for systemic therapy in metastatic breast cancer: ASCO guideline update
.
J Clin Oncol
2022
;
40
:
3205
21
.
3.
Pixberg
CF
,
Schulz
WA
,
Stoecklein
NH
,
Neves
RP
.
Characterization of DNA methylation in circulating tumor cells
.
Genes (Basel)
2015
;
6
:
1053
75
.
4.
Angus
L
,
Beije
N
,
Jager
A
,
Martens
JW
,
Sleijfer
S
.
ESR1 mutations: moving towards guiding treatment decision-making in metastatic breast cancer patients
.
Cancer Treat Rev
2017
;
52
:
33
40
.
5.
Forte
VA
,
Barrak
DK
,
Elhodaky
M
,
Tung
L
,
Snow
A
,
Lang
JE
, et al
.
The potential for liquid biopsies in the precision medical treatment of breast cancer
.
Cancer Biol Med
2016
;
13
:
19
40
.
6.
Lennon
AM
,
Buchanan
AH
,
Kinde
I
,
Warren
A
,
Honushefsky
A
,
Cohain
AT
, et al
.
Feasibility of blood testing combined with PET-CT to screen for cancer and guide intervention
.
Science
2020
;
369(6499)
:
eabb9601
.
7.
Panagopoulou
M
,
Karaglani
M
,
Balgkouranidou
I
,
Biziota
E
,
Koukaki
T
,
Karamitrousis
E
, et al
.
Circulating cell-free DNA in breast cancer: size profiling, levels, and methylation patterns lead to prognostic and predictive classifiers
.
Oncogene
2019
;
38
:
3387
401
.
8.
Uehiro
N
,
Sato
F
,
Pu
F
,
Tanaka
S
,
Kawashima
M
,
Kawaguchi
K
, et al
.
Circulating cell-free DNA-based epigenetic assay can detect early breast cancer
.
Breast Cancer Res
2016
;
18
:
129
.
9.
Xu
Z
,
Sandler
DP
,
Taylor
JA
.
Blood DNA methylation and breast cancer: a prospective case-cohort analysis in the sister study
.
J Natl Cancer Inst
2020
;
112
:
87
94
.
10.
Van De Voorde
L
,
Speeckaert
R
,
Van Gestel
D
,
Bracke
M
,
De Neve
W
,
Delanghe
J
, et al
.
DNA methylation-based biomarkers in serum of patients with breast cancer
.
Mutat Res
2012
;
751
:
304
25
.
11.
de Ruijter
TC
,
van der Heide
F
,
Smits
KM
,
Aarts
MJ
,
van Engeland
M
,
Heijnen
VCG
, et al
.
Prognostic DNA methylation markers for hormone receptor breast cancer: a systematic review
.
Breast Cancer Res
2020
;
22
:
13
.
12.
Pan
Y
,
Liu
G
,
Zhou
F
,
Su
B
,
Li
Y
.
DNA methylation profiles in cancer diagnosis and therapeutics
.
Clin Exp Med
2018
;
18
:
1
14
.
13.
Fackler
MJ
,
Lopez Bujanda
Z
,
Umbricht
C
,
Teo
WW
,
Cho
S
,
Zhang
Z
, et al
.
Novel methylated biomarkers and a robust assay to detect circulating tumor DNA in metastatic breast cancer
.
Cancer Res
2014
;
74
:
2160
70
.
14.
Visvanathan
K
,
Fackler
MS
,
Zhang
Z
,
Lopez-Bujanda
ZA
,
Jeter
SC
,
Sokoll
LJ
, et al
.
Monitoring of serum DNA methylation as an early independent marker of response and survival in metastatic breast cancer: TBCRC 005 prospective biomarker study
.
J Clin Oncol
2017
;
35
:
751
8
.
15.
Fackler
MJ
,
Tulac
S
,
Venkatesan
N
,
Aslam
AJ
,
de Guzman
TN
,
Mercado-Rodriguez
C
, et al
.
Development of an automated liquid biopsy assay for methylated markers in advanced breast cancer
.
Cancer Res Commun
2022
;
2
:
391
401
.
16.
Tworoger
SS
,
Hankinson
SE
.
Use of biomarkers in epidemiologic studies: minimizing the influence of measurement error in the study design and analysis
.
Cancer Causes Control
2006
;
17
:
889
99
.
17.
Heagerty
PJ
,
Lumley
T
,
Pepe
MS
.
Time-dependent ROC curves for censored survival data and a diagnostic marker
.
Biometrics
2000
;
56
:
337
44
.
18.
Meads
C
,
Ahmed
I
,
Riley
RD
.
A systematic review of breast cancer incidence risk prediction models with meta-analysis of their performance
.
Breast Cancer Res Treat
2012
;
132
:
365
77
.
19.
Okpechi
IG
,
Schoeman
HS
,
Longo-Mbenza
B
,
Oke
DA
,
Kingue
S
,
Nkoua
JL
, et al
.
Achieving blood preSsure goals sTudy in uncontrolled hypertensive patients treated with a fixed-dose combination of ramipriL/hydrochlorothiazide: the ASTRAL study
.
Cardiovasc J Afr
2011
;
22
:
79
84
.
20.
Polley
M-YC
,
Leon-Ferre
RA
,
Leung
S
,
Cheng
A
,
Gao
D
,
Sinnwell
J
, et al
.
A clinical calculator to predict disease outcomes in women with triple-negative breast cancer
.
Breast Cancer Res Treat
2021
;
185
:
557
66
.

Supplementary data