A novel liquid biopsy assay for detection of ERBB2 (HER2) amplification in circulating tumor cells (CTCs)

Introduction

ERBB2 genomic amplification is a cancer driver occurring in approximately 20% of breast cancer patients (1). Several pharmacological strategies are designed to be efficacious against cancers with HER2 tumor-specific overexpression (1). Clinically established routine tissue biopsies detect HER2 overexpressed or ERBB2 “amplified” HER2 by detecting the number of copies of ERBB2 by in situ hybridization (ISH) or by protein overexpression by immunohistochemistry (IHC) (2-4).

During the initial diagnosis of metastatic cancer, the standard of care is to verify concordance of current HER2 status with the primary tumor by performing a tissue biopsy on the first or most available metastatic site. Previous studies of metastatic breast cancer (MBC) patients have demonstrated a degree of tumor tissue heterogeneity (5-12) and discordance in the HER2 status between metastatic sites and the primary tumor (13-15). Lack of knowledge about potential change in HER2 status may increase the chance that the therapy for patients thought to continue to have HER2-positive MBC will be ineffective at the expense of toxicity. At the same time, patients with a change to HER2-positive status may miss out on an effective treatment. The occurrence of receptor conversion makes longitudinal monitoring important, yet it is challenging to implement due to the invasive nature of tissue biopsies. While assessment of HER2 status through tissue biopsy IHC or ISH is the standard of care, a significant population of MBC patients do not receive an assessment of HER2 status and even fewer receive serial longitudinal tissue biopsies (16-18). As a result, tumor evolution and biomarker receptor conversion, which can contribute to treatment resistance, are often missed. Studies are ongoing to assess whether HER2 receptor conversion impacts treatment efficacy and survival.

When tissue biopsy is contraindicated, refused, or otherwise not available, liquid biopsy can be an alternative. Typical liquid biopsy assays are limited to analysis of only the cell-free deoxyribonucleic acid (cfDNA) component of blood. In contrast, Epic Sciences’ liquid biopsy assay identifies candidate circulating tumor cells (CTCs), confirms they are genomically unstable, and assesses ERBB2 amplification within those selected cells. CTCs are very rare (on the order of one in two million cells), making both their detection and analysis challenging. Historically, the detection of CTCs relied on enrichment techniques based on preconceived knowledge about their biological phenotype, which may not account for evolving and emerging CTCs. Enrichment limits the ability to perform standardized clinical pathology for CTC biomarker assessment (morphology, protein, and molecular identification) on individual isolated cells (19-23). To overcome these limitations, Epic Sciences developed a cell-based assay to analyze the entire population of blood cells from a tube of blood providing a broad and unbiased identification of CTCs. Epic’s assay then performs a thorough and high-definition phenotypic assessment, which includes evaluation by a clinical pathologist of cell morphology and immunofluorescence (IF) protein expression. A subset of CTCs is also selected to obtain genomic data matching cell morphology and IF data.

Using the cell-based assay (24), Epic Sciences previously developed an assay for detecting AR-V7-positive CTCs in patients with metastatic castration-resistant prostate cancer and informing the selection of treatment associated with superior survival on taxane therapy over androgen receptor signaling inhibitor-directed therapy (25-33). Epic’s AR-V7 test was approved by MolDX (Molecular Diagnostic Services Program) for Medicare reimbursement. Epic’s CTC ERBB2 assay expands Epic’s core technology to detect ERBB2 amplification in CTCs from patients with MBC in a multistep workflow. This report describes both the analytical validation and clinical concordance studies that Epic Sciences completed to assess the performance of the CTC ERBB2 Assay.

Methods

CTC ERBB2 assay workflow

Epic’s CTC ERBB2 assay workflow is shown in Figure 1 (further described in the Supplementary methods section). The assay uses a non-enrichment approach where all nucleated blood cells from a patient are placed on microscope slides (24). The slides are stained by IF and scanned by a high-definition imaging system. The image data are analyzed by an algorithm to detect CTC candidates from among the approximately three million white blood cells (WBCs) on each slide (Fig. 1). To minimize the manual cell classification effort, Epic developed an image analysis algorithm, BRIA (Breast Cancer Imaging Algorithm), to improve standardization and scalability of the assay. BRIA excludes the majority of the non-CTCs and identifies CTC candidates. The identified CTC candidates and their coordinates are tracked through the entire workflow from IF cell analysis to genomics. Using an in-house-developed Clinical Viewer for classification, interpretation, and reporting, CTC candidates are manually classified and confirmed by trained, California-licensed clinical laboratory scientists (Figs. 1 and 2).

Following CTC classification, the laboratory director (a medical doctor and board-certified clinical pathologist) reviews the CTC candidates and, by prioritizing CTCs with high HER2 IF staining intensities, selects CTCs for further genomic characterization via single-cell isolation and low-pass whole-genome sequencing (24,34) for detection of ERBB2 amplification and chromosomal instability. DNA extraction from the CTCs is done via cell lysis, Proteinase K, and Tris buffer. Sequencing library preparation comes after whole-genome amplification (WGA). Both WGA and library preparation are done with the Sigma-Aldrich SEQPLEX-I WGA kit. Patient-level ERBB2 status is determined based on assessment of ERBB2 amplification levels across the chromosomally unstable CTCs using a clinical decision tree scoring system (Fig. 3). The board-certified pathologist(s) at Epic use Epic’s Clinical Viewer to review QC data, interpret the results, and generate the clinical report.

CTC immunofluorescence

The CTC HER2 test utilizes a 4-channel IF assay workflow as outlined in Figure 1. Centrifugation facilitates plasma separation from red blood cells (RBCs). Following removal of plasma, RBCs are lysed, and the remaining nucleated cells are deposited on glass slides and adhered to microscope slides at a density of approximately 3 × 10⁶ cells per slide and stored at −80°C prior to testing. To begin, the IF assay slides are thawed, proteins fixed with paraformaldehyde, cell membranes permeabilized via methanol treatment, washed and placed in automated slide staining instrument. Each slide is subjected to nuclear staining with 4′,6-diamidino-2-phenylindole (DAPI), as well as staining with fluorescently labeled antibodies specific to CD45 (WBC marker), CD31 (endothelial cell marker), cytokeratins (CKs; CTC marker), and HER2 through a sequential immunoassay workflow. CD31 (marking endothelial cells) and CD45 (marking WBCs) are exclusionary labels that allow us to filter out the endothelial cells and WBCs. DAPI channel is used for nuclear staining. The anti-pan CK antibody detects multi-CK types within the 555 nm channel as a marker for tumor cells. The Tyramide Signal Amplification (TSA) reagent system is used to generate the signal detected at 488 nm for the HER2 biomarker. Following fluorescent slide scanning image analysis is performed through BRIA.

Breast cancer imaging algorithm

BRIA consists of four main components: cell detection, segmentation, feature extraction, and CTC candidate identification. BRIA’s cell detection and segmentation components were leveraged from its predecessor (24) used for AR-V7, with the addition of 42 newly developed image intensity features. BRIA’s input is a high-resolution image from the ZEISS™ Axio automated scanning platform. BRIA first detects cells with DAPI signal. A multiscale feature enhancement algorithm is used to detect edges and blobs that are defined using the DAPI signal to enable detection of cell centers. The detected cells are further computationally characterized to extract relevant features from the biomarker intensity data in each of the channels (DAPI, CKs, CD45, and CD31). Specifically, there are 9 cell morphology, 42 biomarker signal intensity, and 24 image texture features. The feature values are used as input to the machine-learning generated algorithm that identifies CTC candidates for manual review.

Copy number variation pipeline for ERBB2 and large-scale state transition detection

A copy-number-analysis pipeline was developed for analysis of the CTC sequencing data. The pipeline was previously described and is similar to typical whole-genome sequencing workflows (35). Briefly, it aligns the Illumina sequencing reads to the human reference genome (hg38) and tallies the read coverage of 1-Mb bins across the genome. Using the alignment data, it computes key bioinformatic QC metrics for identifying samples with 1) insufficient sequencing data, 2) significant fractions of low-quality alignments, or 3) excessive coverage noise. Samples not passing bioinformatic QC are excluded from analysis (counts shown in Supplementary Tab. 2). Consistent with typical coverage scaling approaches, the bin coverage is scaled relative to an autosomal baseline. This normalizes for the average gene count in autosomes, and the resulting ERBB2/autosomal-average ratio compensates for the presence of multiple copies of chromosomes. In contrast, ISH-based ERBB2 assays typically use CEP17 as the reference point.

Analytical validation

Breast cancer imaging algorithm

To validate BRIA’s ability to identify CTC candidates, its cell-level performance was evaluated on an analytical validation dataset that consisted of manually confirmed CTC candidates, non-CTCs, and visual artifacts from patients’ samples. The cell-level performance of the HER2 IF protein analysis (Supplementary Fig. 1) was assessed using a high-marker-expressing cell line (MDA-MB-453) and a low-marker-expressing cell line (MDA-MB-231). Also, several breast cancer cell lines spiked into healthy donors’ blood were tested for their HER2 median cellular fluorescence intensity (MFI) values on Epic’s assay (Supplementary Fig. 2).

Copy number variation pipeline for ERBB2 and large-scale state transition detection

The accuracy for ERBB2 amplification detection was assessed using MDA-MB-453 cells as positives and WBCs as negatives; sensitivity and specificity were computed. The precision for ERBB2 amplification detection was assessed across two manufacturing lots of the sequencing-reagent kits, two operators, and two sequencing runs. To assess how the results were impacted by the additional variables, positive percent agreement (PPA) and negative percent agreement (NPA) were computed by comparing the mode across all replicates to each replicate result (Tab. 1).

Clinical concordance

Patients were included in the concordance study with distant metastases and with HER2 status as determined in a matched metastatic tissue biopsy using the standard, IHC, and/or ISH performed at each cancer center site and were tested by Epic’s CTC ER assay. The concordance of ERBB2 amplification was detected by single-cell CTC genomics. Contemporaneous is defined as patients with blood collection performed within about 30 days before tissue biopsy. By computing typical concordance metrics (PPA, NPA, positive predictive value [PPV], and negative predictive value [NPV]), we inferred the probability that the assay will provide a correct diagnosis of HER2 status as determined in a matched metastatic tissue biopsy. The two-sided 95% confidence interval for PPA and NPA was calculated using the Wilson score method and the Bayesian Rule formula (36,37).

Biomarker	LOD (n = 138)	AV cutoff	Accuracy (n = 162)			Precision (n = 162)
			Sensitivity	Specificity		PPA	NPA
ERBB2 amplification	2-fold amplification	>1.77 (ERBB2 Z-score)	0.85	0.94		0.85	0.94

The sensitivity and specificity of ERBB2 amplification detection were computed using WBCs and MDA-MB-453 cells as expected negatives and positives, respectively. Based on the expected statuses, true positives and true negatives, and false positives and false negatives were counted for all samples that passed QC.
AV = analytical validation; LOD = limit of detection; NPA = negative percentage agreement; PPA = positive percentage agreement; WBC = white blood cell.

Results

Analytical validation

BRIA and immunofluorescence

BRIA was built to reliably identify CTC candidates from patient slides with minimal manual review. It is based on earlier versions of Epic’s proprietary digital pathology used in its CTC assays (AR-V7 and others) (24-33,35,38). BRIA classifies cell candidates captured from slide images into two classes, “CTC” and “non-CTC,” using a predictive model. In the analytical validation analysis, the evaluation of the BRIA predictive model relies on “true positive,” which refers to a manually confirmed CTC that BRIA classified as “CTC,” and “true negative” refers to a non-CTC that BRIA classified as “non-CTC.” These labels are used in the calculation of BRIA’s cell-level performance metrics. At the cell level, BRIA had 99% sensitivity, 96% specificity, and 97% overall percentage agreement (OPA) for the validation dataset (Supplementary Tab. 1). The analytical performance at the cell level of the IF HER2 protein analysis was evaluated between the high-marker-expressing cell line (MDA-MB-453) and the low-marker-expressing cell line (MDA-MB-231) (Supplementary Fig. 1), and revealed a sensitivity, specificity, and accuracy of 94%, 97%, and 95%, respectively (Supplementary Fig. 1). To demonstrate the assay’s ability to cover the dynamic range of intensity levels of HER2, Epic spiked cells from several breast cancer cell lines into tubes of healthy donor blood and tested the spiked blood on Epic’s CTC ERBB2 assay. The resulting single-cell data showed that the overall HER2 intensity levels were consistent with expectations for well-characterized cell lines based on reported IHC values (39,40) (Supplementary Fig. 2).

Single-cell genomic assay

To validate the single-cell genomic pipeline’s ability to identify ERBB2 copy number amplification and chromosomal instability, we analyzed WBCs from healthy donors and well-characterized breast cancer cell lines (SK-BR-3, MDA-MB-453, and MCF-7) with known chromosomal instability and known ERBB2 amplification status as negative and positive controls, respectively. The experiments, cell types, and the number of cells used for analytical validation are summarized in Supplementary Tab. 2. To evaluate the limit of detection (LOD) for the detection of chromosomal instability, a copy number variation simulator was used to introduce different amounts of large copy number changes (which would result in large-scale state transitions [LSTs]) to the WBC base genomes (Supplementary Fig. 3). Modified genomes with at least 10 breakpoints were consistently called as chromosomally unstable (sensitivity of 91%), which was considered the LOD for chromosomal instability detection (Fig. 3). ERBB2 amplification was not detected in WBCs and MCF-7 cells but was detected in MDA-MB-453 and SK-BR-3 cells (Tab. 1 and Fig. 3). Given that MDA-MB-453 cells had an expected 2-fold ERBB2 amplification, that 2-fold level was considered the LOD for detection of ERBB2 amplification. The sensitivity and specificity for detecting the 2-fold ERBB2 amplification were 85% and 94%, respectively (Tab. 1). Precision was assessed across two sequencing runs, two lots, and two operators and the overall PPA and NPA were 85% and 94%, respectively (Tab. 1).

Discussion

This report shows the analytical and clinical performance of Epic’s CTC ERBB2 (HER2) cell-based liquid biopsy assay’s ability to detect ERBB2 amplifications in MBC patients. The assay is a test intended for a population with very limited options available to get individualized sequential information about the treatment that is most likely to be efficacious.

A substantial improvement in the CTC ERBB2 (HER2) assay’s characterization of CTCs over conventional liquid biopsy assays is that, as a criteria of CTC classification and enumeration, it takes advantage of a characteristic genomic feature of tumor cells, which is the presence of chromosomal instability, a biological mechanism of tumor evolution and adaptation to environmental pressures that enables tumor progression and metastatic capabilities (42-44). Peer-reviewed data showed that genomic sequencing applied to CTCs from Epic Sciences’ assay allowed identification of high chromosomal instability within CTCs (35,42,45) that was independently associated with worse survival in men with high-risk metastatic castration-resistant prostate cancer treated with abiraterone/enzalutamide that may benefit alternative treatments (42). In this report, concordance of ERBB2 status was analyzed only in the subgroup of patients with CTCs with high LSTs by the CTC ERBB2 (HER2) assay, which indicates the detection of chromosomal instability. Only 50% of patients were found to have CTCs with high numbers of chromosomal instability. Clinical studies will be performed to test whether patients with detection of chromosomally unstable CTCs reported here will associate with patient’s outcome.

Typically, reliable tumor identification in histopathological tissue biopsy analysis of HER2 status does not require genomic assessment but instead requires a pathologist to visually examine patterns of cell shapes and regions of tissue structure to identify cancer cells by morphological assessment and to perform proper semiquantitative scoring of HER2 expression on those cancer cells. However, liquid biopsy assays are limited to the analysis of those few rare CTCs (or ctDNA) in the sample that are in suspension, without their tissue contexture, and sometimes may be morphologically indistinguishable from circulating nontumor cells by the trained pathologist (or from DNA shed from endogenous germ-line cells, called cell-free DNA [cfDNA]). To avoid this problem, detection of chromosomal instability in CTCs, a common feature of metastatic cancer cells (42,43), increases the probability that the CTCs on which ERBB2 is evaluated represent truly neoplastic components of the tumor, thus making the ERBB2 CTC assay diagnostic assessments tumor specific. Therefore, as compared to analysis based on cfDNA, Epic’s assay can detect ERBB2 amplifications with high sensitivity because it is evaluating a pool of DNA that comes entirely from a tumor cell with chromosomal instability, as detected by high numbers of LSTs. Development efforts are ongoing to test whether artificial intelligence algorithms trained on thousands of chromosomally unstable CTCs will be capable of differentiating subsets of cancer cells by using only morphological patterns.

Potential confounding factors impacting our concordance analysis are the following: First, there are no available orthogonal assays to demonstrate concordance in the same blood sample. As an alternative, we measured concordance comparing the blood assay results to metastatic tumor tissue biopsies rather than comparing the same blood sample analyzed with different CTC-based assays. Second, bone tissues often represent the location of first distant metastasis in patients (46) but bone biopsies utilizing older decalcification techniques could be falsely negative for HER2 overexpression or ERBB2 amplification (17,41). In fact, bone biopsies from bone-only metastatic patients are often excluded from clinical trials of HER2 expression as exemplified by the recently published DAISY trial (47). Additional factors contribute to assay variability, such as lack of central laboratory verification for pathological tissue biopsy results. These may reflect variations in pathologists’ assessment of HER2 status in tissue biopsies that do not control and are not accounted for within reported analytical variation. Validation studies using a contemporaneous dataset of tissue biopsy samples analyzed in central pathology labs will be needed to confirm these correlative findings in a larger dataset. Clinical utility studies have been designed to test whether the CTC ERBB2 assay will inform improved treatment decision-making to HER2-targeted therapies over the current standard of care for MBC patients.

Acknowledgments

Research support for this study was funded by Epic Sciences. The data are available within the article and the Supplementary Information provided with this article. For additional data that may be under restricted access in compliance with patient consent and ethical principles for data sharing request, please contact the corresponding author.

Disclosures

Conflict of interest: E.T.L., E.W., D.Z., D.B., M.B., N.D., M.S., and R.W. are or have been employed by Epic Sciences. L.S. has received honoraria for consulting from Epic Sciences.

Funding: The research and the work presented were funded by Epic Sciences.

Author contributions: G.D.C. designed the clinical study, performed the statistical analysis, and drafted the manuscript. E.T.L., E.W., and D.Z. designed and developed the genomic platform and performed statistical analysis. D.B. and M.B. designed and developed the immunofluorescent platform. N.D. is a board-certified clinical pathologist who reviewed the CTC candidates and interpreted the clinical results. M.S. coordinated clinical study samples’ accrual, clinical data collection, and data entry. L.S. and R.W. provided scientific guidance, revised the manuscript draft with all authors contributing and providing feedback and advice.