Development of an Automated Liquid Biopsy Assay for Methylated Markers in Advanced Breast Cancer

Current molecular liquid biopsy assays to detect recurrence or monitor response to treatment require sophisticated technology, highly trained personnel, and a turnaround time of weeks. We describe the development and technical validation of an automated Liquid Biopsy for Breast Cancer Methylation (LBx-BCM) prototype, a DNA methylation detection cartridge assay that is simple to perform and quantitatively detects nine methylated markers within 4.5 hours. LBx-BCM demonstrated high interassay reproducibility when analyzing exogenous methylated DNA (75–300 DNA copies) spiked into plasma (coefficient of variation, CV = 7.1%–10.9%) and serum (CV = 19.1%–36.1%). It also demonstrated high interuser reproducibility (Spearman r = 0.887, P < 0.0001) when samples of metastatic breast cancer (MBC, N = 11) and normal control (N = 4) were evaluated independently by two users. Analyses of interplatform reproducibility indicated very high concordance between LBx-BCM and the reference assay, cMethDNA, among 66 paired plasma samples [MBC N = 40, controls N = 26; Spearman r = 0.891; 95% confidence interval (CI) = 0.825–0.933, P < 0.0001]. LBx-BCM achieved a ROC AUC = 0.909 (95% CI = 0.836–0.982), 83% sensitivity and 92% specificity; cMethDNA achieved a ROC AUC = 0.896 (95% CI = 0.817–0.974), 83% sensitivity and 92% specificity in test set samples. The automated LBx-BCM cartridge prototype is fast, with performance levels equivalent to the highly sensitive, manual cMethDNA method. Future prospective clinical studies will evaluate LBx-BCM detection sensitivity and its ability to monitor therapeutic response during treatment for advanced breast cancer. Significance: We technically validated an automated, cartridge-based, liquid biopsy prototype assay, to quantitatively measure breast cancer methylation in serum or plasma of patients with MBC, that demonstrated high sensitivity and specificity.


Introduction
Breast cancer is now the most common type of cancer worldwide (1). In newly updated data, Globocan 2020 estimates that there were nearly 2.3 million new cumulative methylation (CM) levels of a panel of markers (7,8). In patients with MBC, a 10-gene panel consisting of AKRB, COLA, HOXB, RAS-GRF, RASSF, HISTHC, GPX, ARHGEF, TMEFF, and TMSF detected ctDNA in 300 μL sera with high sensitivity (91%) and specificity (96%; ref. 7). In the Translational Breast Cancer Research Consortium (TBCRC005) cohort of patients undergoing chemotherapy for MBC, the index of CM (CMI) of a minimal 6-gene subset (AKRB, HOXB, RASGRF, RASSF, HISTHC, and TMSF) was a strong predictor of survival outcomes in MBC (8). Yet, because cMethDNA involves a minimum of 1 week to complete and requires high technical competency, translation into a widely available diagnostic laboratory assay would be very challenging.
In fact, to date there is no commercially available in vitro diagnostic (IVD) circulating cell-free DNA (cfDNA) methylation assay developed for breast cancer (9)(10)(11). Therefore, our long-term goal is to use the principles of cMethDNA to develop an assay for routine use as a clinical management tool by making it faster and easier to perform through automation. As a first step, in collaboration with the diagnostics company Cepheid, and using their GeneXpert® platform (refs. 12-15; https://www.cepheid.com/) we have developed a prototype assay. The cartridge-based, Liquid Biopsy for Breast Cancer Methylation (LBx-BCM) assay is intended to be used at point of care as a rapid ancillary assay to support current clinical approaches to evaluate breast cancer. LBx-BCM is a prototype in development and is not for use in clinical diagnostic procedures and not reviewed by any regulatory body. Here, we report technical validation of LBx-BCM, demonstrating that it is possible to automate processing of plasma and serum samples and quantitatively assess DNA methylation for nine target genes within 4.5 hours, with less than 15 minutes of hands-on time.

Study Design and Sample Collections
We used prospective blood collections from studies that followed women with

Ethics Approval and Consent to Participate
We obtained written informed consent from the patients following approval of each study from the JH Institutional Review Board. The studies were conducted in accordance with recognized ethical guidelines (Belmont Report).

DNA Marker Selection
To ensure good breast cancer coverage for the LBx-BCM prototype assay, we chose nine CpG DNA markers from among a larger panel of cMethDNA genes that, together, recognized all four histologic subtypes of breast cancer (7,8,14,16).

The Prototype LBx-BCM
The GeneXpert® system (Cepheid) is a closed, automated PCR-based molecular diagnostic testing platform using self-contained cartridges to perform nucleic acid extraction and PCR. The LBx-BCM prototype was developed to meet the increased technical sensitivity required for detection of picograms of free ctDNA in blood. One cartridge is used for bisulfite conversion of unmethylated cytosine residues to uracil, which changes the DNA sequence specifically for unmethylated DNA, but not for methylated DNA (the conversion cartridge). Two additional cartridges are used for the performance of methylation-specific PCR (the methylation detection cartridges); these two detection cartridges contain reagents, in each cartridge, for nested multiplex real-time quantitative PCR of 4-5 target genes and ACTB as the internal reference, using six different fluorophores. Primer and probe sequences are presented in Supplementary  Table S1. The entire assay is completed within 4.5 hours and requires approximately 15 minutes of hands-on time. LBx-BCM is a research use only prototype in development, not for use in diagnostic procedures, and has not been reviewed by any regulatory body.

LBx-BCM Algorithm for Methylation
The method of calculating CM is described in Supplementary Table S2. Step 1: GeneXpert® software assigns the C t at the end of the run; the user assigns C t = 45 if no signals were detectable during the run; C t (C t gene − C t ACTB) is calculated to normalize all results to the ACTB reference DNA. If some samples have negative C t (C t gene − C t ACTB) for a gene, all samples are transformed by adding a constant value to give positive integers for that gene.
Step 2: If C t (C t gene − C t ACTB) is higher than the historical replicate median of 300 copies + 13 C t units, the user adjusts to C t (C t gene − C t ACTB) = 0, thereby removing signals from the analysis that are too low to quantitate (less than 0.04 copies of target; Supplementary Fig. S1; Supplementary Table S2).
Step 3: Gene methylation (M) = [1/ C t (C t gene − C t ACTB)] * 1,200. This is a robust transformation intended to raise the methylation values from baseline and increase the assay dynamic range.
Step 4: Calculate CM as follows, where CM = sum of M in the 9-gene panel.

Sample Processing
Plasma from STRECK Cell-free DNA BCT tubes (STRECK, Omaha, NE, #218962) was collected, harvested, and frozen at −80°C within 5 days. Two sequential centrifugation steps ensured that the plasma was free of cells prior to freezing. Serum was harvested from serum separation tubes (BD, #367988), and frozen at −80°C within 4 hours. Plasma and serum were stored frozen at −80°C in aliquots. Before using, the samples were thawed at room temperature, inverted 10 times, then microcentrifuged at 14,000 rpm for 15 minutes at room temperature.

The LBx-BCM Assay
For the LBx-BCM assay, plasma or serum (1.0 mL) was mixed with 50 μL proteinase K (600 units/mL; PK; Roche Diagnostics Corp.), 2.0 mL Lysis Buffer (Cepheid) and incubated for 10 minutes at room temperature. After incubation, absolute ethanol (1.5 mL) was added and the sample was loaded into the bisulfite conversion cartridge for processing (2.5 hours). The bisulfite-converted DNA sample was divided equally into two LBx-BCM methylation detection cartridges and methylation specific-PCR was performed (1 hour and 45 minutes; AKRB, TMSF, ZNF, TMEFF target genes and ACTB reference gene in Cartridge A; COLA, HISTHC, RASGRF, HOXB, RASSF target genes and ACTB in Cartridge B). The following reactions were run in each methylation detection cartridge: (i) a methylation-independent, nested multiplexed PCR that preamplified the 9-gene panel for 20 cycles, and (ii) a methylationspecific, nested quantitative 6-plex real-time PCR that uses internal primers and first PCR. The assay is completed within 4.5 hours including approximately 15 minutes of preparation time.

Preparation of Analytic Replicates
We spiked 600, 300, 150, or 0 copies of a laboratory stock of methylated human control DNA (#N2131, Promega Corp.) that was previously quantified by digital droplet PCR into 1.0 mL of commercial normal plasma or serum (female human pooled plasma, K 2 EDTA anticoagulated or pooled serum, BioIVT). After adding PK (50 μL), lysis buffer (2 mL) and absolute ethanol (1.5 mL), each sample was transferred to a bisulfite cartridge. Within this cartridge, DNA was extracted, then converted with sodium bisulfite (D5030-1, Lightning Conversion Reagent, Zymo Research) and afterward transferred in equal amounts to each of two detection cartridges for quantitative nested methylation-specific real-time PCR.

Interuser Reproducibility
J0888 repository samples obtained from patients with MBC (N = 11) and normal controls (N = 4) were aliquoted into duplicate sample sets. One set was tested by User A and the other set was tested by User B on separate days using the same reagents. Users were blinded to the origin of the samples. LBx-BCM was performed as described above. Interuser concordance was evaluated using the Spearman correlation coefficient.

Statistical Analysis
Analyses of independent groups were performed and data were visualized using box whisker plots. Differences between groups were evaluated using the nonparametric Mann-Whitney test. The performance of the 9-gene panel was characterized by estimating the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and likelihood ratio along with the 95% confidence intervals (CI). Classification accuracy = TP + TN/TP + TN + FP + FN using ROC-derived laboratory methylation cutoffs (38.5 CM units for LBx-BCM; 1.5 CM units for cMethDNA). All statistical tests were two sided and considered statistically significant at P < 0.05. Spearman correlation was performed to compare the CM of the reference laboratory assay, cMethDNA, with CM obtained in the LBx-BCM system in the test set samples (40 cancer and 26 control noncancer samples). GraphPad Prism version 9.0 (GraphPad Software) was used for all analyses.

Data Availability
Patient datasets generated and/or analyzed during this study are not publicly available due to the sensitivity of the data, but are available from the corresponding author upon reasonable request.

The LBx-BCM Prototype Assay
The LBx-BCM's quantitative PCR workflow is depicted in Fig. 1A. Steps 1-4 involve pre-processing of the sample for DNA extraction. In step 5, the mixture is placed in the bisulfite conversion cartridge. In step 6, the bisulfite-treated DNA is divided equally into two LBx-BCM methylation detection cartridges to amplify and detect nine methylated genes (up to five methylated genes plus ACTB per cartridge). At the end of the assay (4.5 hours), the cycle threshold (C t ) for each gene is provided.
For LBx-BCM marker development, we selected a 9-marker panel. We selected primer/probe combinations that performed optimally in the presence of the other markers and fluorophores in the 6-plex reaction in each cartridge (7,8,14). The final 9-gene panel consisted of HOXB, RASGRF, AKRB, TMSF, COLA, HISTHC, TMEFF, RASSF, and ZNF. Primer and probe sequences for the gene panel are shown in Supplementary Table S1.

Interassay Reproducibility
The challenge in development of an automated assay for ctDNA is the ability to detect only a few hundred picograms or less of target DNA in a vast abundance of normal cfDNA. We developed the cartridge-based LBx-BCM assay (Fig. 1A), including the method for calculating CM (Supplementary Fig. S1; Supplementary Table S2). The analytic performance of the assay was evaluated by spiking replicates of 300, 150, 75, and 0 copies (1 ng-250 pg) of fully methylated target DNA into 0.5 mL of either commercial pooled normal plasma (  Table S3). On the basis of the C t (C t Gene − C t ACTB), scatter diagrams showed that nearly all replicates of 75-300 copies of target DNA were detected. In the absence of target DNA (0 spiked copies), either the input DNA was too low to quantitate, or no PCR amplification was observed (  Table S3). The C t increased with decreasing number of copies of target for each gene. CM of the 9-gene panel was significantly different for 0 versus 75 copies (P < 0.0001), 75 versus 150 copies (P = 0.0003), and 150 versus 300 copies (P < 0.0001; Mann-Whitney Analysis;  Table S2), we used the replicate control median C t of 300 copies, as shown in Supplementary Table S3.

Interuser Reproducibility
We evaluated LBx-BCM reproducibility between users to determine whether the method gave similar results independent of the operator. A total of 15 samples, including patients with MBC (N = 11), and healthy controls (N = 4), were divided into duplicate sets and assayed on different days using cartridges from the same batch. The Spearman r = 0.887 indicated a high level of interuser reproducibility (Fig. 1D).

LBx-BCM-Based Detection of MBC in the Training Set
The LBx-BCM ctDNA method was initially evaluated in JH repository J0888 samples (patient characteristics and sample sets presented in Tables 1 and 2) to verify that LBx-BCM could distinguish between MBC versus normal serum using circulating cfDNA. For many of these patients, blood was collected while they were undergoing chemotherapy. We examined CM of the 9-marker panel in serum samples (MBC, N = 20; control normal, N = 20), and observed significantly higher methylation in the cancer sera compared with normal controls as shown in the histogram (Supplementary Fig. S3A) and in box whiskers plot (Supplementary Fig. S3B; Mann-Whitney test P = 0.002). The ROC-derived threshold that provided the highest combined sensitivity and specificity was 38.5 CM units ( Supplementary Fig. S3C). At this threshold the ROC AUC = 0.766 (95% CI, 0.616-0.916; P = 0.004), with 75% sensitivity (95% CI, 53. 1-88.8) and 65% specificity (95% CI, 43.3-81.9).

Accuracy of LBx-BCM to Detect MBC in the Test Set
We then locked existing assay parameters and tested an independent, well an-

Interplatform Concordance between LBx-BCM and cMethDNA
LBx-BCM and cMethDNA assays utilize nearly identical primer/probe sequences and similar nested quantitative multiplex methylation-specific PCR strategies. However, cMethDNA normalizes methylation to a gene-specific recombinant standard of 50 methylated copies spiked into 300 μL of plasma or serum prior to purification of DNA, while LBx-BCM normalizes methylation to the endogenous actin reference in the DNA present in 500 μL plasma or serum. As a technical verification step to determine whether LBx-BCM achieved a similar level of performance as cMethDNA, we performed cMethDNA on the    entire IMAGE II/J0888 test set (Fig. 3). Consistent with the LBx-BCM results, cMethDNA detected significantly more methylation in plasma samples from breast cancer than from normal or benign individuals as shown in histogram and box whiskers plots (Fig. 3A and B; Mann-Whitney test P < 0.0001). At the CMI threshold of 1.5 units, for cMethDNA the ROC AUC = 0.896 (95% CI = 0.817-0.974; P < 0.001), with a sensitivity of 83% (95% CI = 68.1-91.3) and a specificity of 92% (95% CI = 75.9 -98.6; Fig. 3C). LBx-BCM and cMethDNA methylation results were highly concordant (Spearman r = 0.891, P < 0.0001; N = 66 paired samples; Fig. 3D). Descriptive statistics for interplatform reproducibility between LBx-BCM and cMethDNA are provided in Table 3.

Changes in LBx-BCM Methylation During Treatment of MBC
We had previously reported results of longitudinal studies in serial blood collections for cMethDNA (7,8  concordance with this assay, we predicted that LBx-BCM methylation levels would also change during the course of chemotherapy. We analyzed CM by LBx-BCM in serum samples obtained from MBC patients in two prospective clinical studies conducted at JH-J0214 and J0425. Serum was collected prior to the initiation of treatment (baseline), 18-49 days (median 21 days) after starting a new line of treatment, and upon completion of additional cycles. Patients received either 28-day cycles of docetaxel or 21-day cycles of capecitabine. Representative plots of LBx-BCM methylation are shown in Fig. 4 and Supplementary Fig. S5. In these heavily pretreated patients with stage IV breast cancer, changes in CM occurred frequently during the course of treatment. For many patients, there was an initial reduction in methylation after the initiation of therapy. Increased methylation was observed among patients who progressed on treatment and among some patients with stable disease (Fig. 4; Supplementary Fig. S5).

Discussion
Widespread diagnostic implementation of assays that detect ctDNA has not occurred. This is largely due to the technical complexities of such assays and the extensive time required to conduct them (9,(17)(18)(19). We sought to overcome these obstacles by developing, to the best of our knowledge, the first automated ctDNA methylation assay capable of simultaneously quantitating methylation levels in a panel of markers. Our goal was to develop an assay that would be sophisticated yet simple to perform in underserved regions worldwide, and could be used at point of care to provide same-day feedback to clinicians and patients (7,8,14,20). We report the development of a nested, quantitative, multiplexed methylation-specific PCR assay, called LBx-BCM run on the GeneExpert® system. It can be performed in approximately 4.5 hours sample to answer and it uses many of the same markers and principles as the highly sensitive manual cMethDNA method that served as its foundation (7,8,14). LBx-BCM is a prototype for research use only.
In the current study, we report technical development and validation of LBx-BCM. We chose a ctDNA marker panel consisting of HOXB, RASGRF, AKRB, TMSF, COLA, HISTHC, TMEFF, RASSF, and ZNF (7,8,14) and then developed LBx-BCM using the training set of J0888 repository samples. LBx-BCM was validated by performing interassay, interuser and interplatform reproducibility studies comparing LBx-BCM with the reference method cMethDNA (7). Interassay reproducibility studies demonstrated that although the sensitivity to predict tumor origin was only 29.6%. Shen and colleagues (24) used 1-10 ng cfDNA to perform methyl-DNA immunoprecipitation followed by high-throughout sequencing to profile methylation patterns typical of tumor cfDNA in several tumor types. However, validation of the method based on differentially methylated regions was not performed in breast cancer (24).
our study cohort was primarily from patients (29/39) with ER + /PR + /HER2 − breast cancer. The LBx-BCM assay needs to be evaluated in large, prospectively designed studies which include a balanced representation of all histologic subtypes of breast cancer and a greater ethnic diversity. Such a cohort has already been identified in the prospective TBCRC 005 trial of patients with stage IV breast cancer for whom serial blood sampling was performed at baseline, 3-4 weeks after initiation of a new chemotherapy treatment and 8-12 weeks later (8). It would also be important to evaluate the performance of this automated system in detecting disease in patients with earlier stages of breast cancer.
In conclusion, we have developed and technically validated a quantitative multiplexed and automated assay for methylated markers in the GeneXpert® system for assaying cfDNA from a liquid biopsy which can be implemented at the point of care.