Preanalytical Variables for the Genomic Assessment of the Cellular and Acellular Fractions of the Liquid Biopsy in a Cohort of Breast Cancer Patients

Abstract

Liquid biopsy allows assessment of multiple analytes, providing temporal information with potential for improving understanding of cancer evolution and clinical management of patients. Although liquid biopsies are intensely investigated for prediction and response monitoring, preanalytic variables are of primary concern for clinical implementation, including categories of collection method and sample storage. Herein, an integrated high-density single-cell assay workflow for morphometric and genomic analysis of the liquid biopsy is used to characterize the effects of preanalytical variation and reproducibility of data from a breast cancer cohort. Following prior work quantifying performance of commonly used blood collection tubes, this study completes the analysis of four time points to assay (24, 48, 72, and 96 hours), demonstrating precision up to 48 hours after collection for assay sensitivity, highly reproducible rare cell enumeration, morphometric characterization, and high efficiency and capacity for single-cell genomic analysis. For the cell-free analysis, both freezing and use of fresh plasma produced similar quality and quantity of cell-free DNA for sequencing. The genomic analysis (copy number variation and single-nucleotide variation) described herein is broadly applicable to liquid biopsy platforms capable of isolating cell-free and cell-based DNA. Morphometric parameters and genomic signatures of individual circulating tumor cells were evaluated in relation to patient clinical response, providing preliminary evidence of clinical validity as a potential biomarker aiding clinical diagnostics or monitoring progression.

Cancer evolves from the point of initiation to lethal metastatic disease through the course of natural disease progression and in response to treatment pressures. To achieve distant metastasis, cancer cells must travel through the bloodstream to distant sites and adapt to growth in a foreign environment, ultimately proliferating to the point of disrupting the health of the patient. Diagnostic evaluation of this process, which often takes years, is thus a problem of both time and space. In such a complex pathology, the liquid biopsy, based on a simple blood draw, affords a minimally invasive route to assess molecular biomarkers in solid tumor cancers at multiple time points over a course of treatment, allowing for a greater understanding of the time-related changes of cancer progression and treatment response.

The high-density single-cell assay (HD-SCA) employed in this study is an integrated workflow that detects circulating tumor cells (CTCs) and combines molecular assays of DNA extracted from CTCs (single-cell DNA) and cell-free DNA (cfDNA) from a single blood collection tube. The HD-SCA is the only workflow that integrates the analysis of circulating cells and cfDNA, thus providing multiplexed molecular analysis of biomarkers with the increased potential for guiding treatment and improving clinical management of the disease.

To employ CTCs as a biomarker of disease and show clinical utility as a diagnostic, prognostic, or predictive tool in the clinic, a detection platform must be both precise and accurate. The features of cellular and acellular components are likely to differ among cancer types and will certainly vary over the evolution of disease, which has to be recognized as a potential challenge in the development and deployment of a detection system. Most likely specific analytes, such as CTCs and cfDNA, will require separate validation for clinical utility in different disease settings. Recently, a joint review from the American Society of Clinical Oncology and the College of American Pathologists summarized the current information about clinical cfDNA assays to initiate a framework to guide future research.¹ Although general exploration is necessary for biological discovery, in clinical medicine, a procedure must have demonstrated reproducibility with characterization of known deviations or fluctuations. Robust liquid biopsy detection technology must be capable of detecting extremely rare analytes, such as circulating tumor DNA and CTCs, in a background of billions of normal blood cells and plasma. Development of a rare cell detection assay for a specific context of use requires full analytical validation that the assay is fit for the purpose indicated.2, 3, 4 Clinical validation can then be done to generate the evidence that associates the assay output with clinical outcomes.⁵ This is followed by clinical utility, which clearly demonstrates that the use of the assay output in clinical management improves patient outcomes relative to nonuse.2, 3, 4

Quality laboratory practices are important to the health and welfare of the population. Laboratory-developed tests are not subject to US Food and Drug Administration regulation, but certification under Clinical Laboratory Improvement Amendments/College of American Pathologists is required to ensure quality and validity for the purpose of diagnosis, prevention, treatment of disease, or assessment of health (Clinical Laboratory Improvement Amendments, https://www.cdc.gov/clia/law-regulations.html). By characterizing and quantifying all aspects of laboratory testing, including preanalytics, significant improvement can be made in the quality, accuracy, and relevance of laboratory practices (US Food and Drug Administration, https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm). Preanalytical variation in the detection of CTCs includes such categories as collection method, shipping conditions, and sample storage. For moderate- to high-risk laboratory-developed tests, the US Food and Drug Administration's proposed regulation requires US Food and Drug Administration authorization before marketing in the United States through either the premarket approval or 510(k) premarket notification pathways (US Food and Drug Administration, https://www.fda.gov/media/89841/download). The 510(k) is a premarket submission that demonstrates the laboratory-developed test to be marketed (generally class II; moderate risk) is at least as safe and effective as a legally marketed device that is not subject to premarket approval. The enumeration of CTCs of epithelial origin in patients with metastatic breast, colon, or prostate cancer using CellSearch (Menarini Silicon Biosystems, Huntington Valley, PA) currently has US Food and Drug Administration 510(k) approval, and general guidelines include the CellSave Preservative blood collection tube (BCT; Menarini Silicon Biosystems) with ambient temperature storage and a time to assay (TTA) of <96 hours.

The HD-SCA workflow offers significant promise in the multianalyte analysis of the liquid biopsy, while gaining biological insights relevant to disease progression. Using an analogous method, Epic Sciences (San Diego, CA) recently released a predictive and prognostic CTC test for metastatic castration-resistant prostate cancer certified under US Clinical Laboratory Improvement Amendments and accredited by the College of American Pathologists.6, 7, 8 This exemplifies the potential for clinical utility of the HD-SCA workflow for the identification of a biomarker in breast cancer patients.

The HD-SCA workflow is inherently a nonenrichment high-content direct imaging method capable of providing visualization of both typical and atypical tumor-related cells in circulation and molecular parameters at both the single-cell level and cfDNA level. This allows for both the identification of genetic mutations important for targeted therapies and for a greater understanding of the genomic drift in the tumor cell population, which may lead to treatment resistance. The value of single-cell genomic analysis conducted on this platform has been previously reported, showing compatibility with clinical practice.^9,10 The HD-SCA workflow has the potential to transform cancer research and clinical practice by applying a genomic approach to specific cell populations that have only been initially characterized by cytomorphology in the liquid biopsy. In addition, genomic analysis of the circulating cfDNA from plasma isolated from these cancer patients may be measured and evaluated with the same principles, providing complementary genomic information.

Single-cell and next-generation sequencing (NGS) technologies are rapidly evolving, making it imperative to control for technical and biological variations, ensuring reproducible and accurate findings. Sequencing analysis is affected by preanalytical variables that may introduce errors at each step, affecting the quality of the data for analysis and their interpretation. Quality controls for each step in the HD-SCA workflow have been developed and established to detect, prevent, and mitigate errors, while accounting for technical variation and analysis biases of overall CTC enumeration or CTC subclasses, which may be relevant to guide specific clinical decisions. Few studies have rigorously examined the technical reproducibility of single-cell sequencing after CTC enumeration and isolation. Most methods have shown similar overall performances and sensitivity, but they often lack important necessary evidence of assay performance, including accuracy and specificity. This presents a serious barrier to the ultimate goal of identifying signatures that are clinically meaningful.

This study systematically analyzed the influence of preanalytical variables in the context of blood biospecimen collection, handling, and processing, with respect to the analysis of the complete liquid biopsy for scalable clinical utility in routinely obtained peripheral blood samples from breast cancer patients. The variables under investigation were TTA (time from blood draw to cryostorage) for CTC evaluation and fresh versus frozen (FVF) plasma preparations for cfDNA isolation for the application of single-cell and cfDNA genomic analysis. This study completes a set of four TTAs (24, 48, 72, and 96 hours) in the assessment of CTC detection for optimization of the HD-SCA workflow. The results recommend using the Streck cell-free DNA BCT (Streck, La Vista, NE) up to a 48-hour TTA, which ensures cellular retention and the highest efficacy of rare cell identification by the HD-SCA workflow, while providing high-quality single-cell genomic sequencing data. In addition, the quality of isolated cfDNA and resulting sequencing data was independent of freezing. This study does the following: i) validates the potential of the HD-SCA workflow for CTC discovery, ii) demonstrates the feasibility of single-cell and cfDNA copy number variation (CNV) and single-nucleotide variation (SNV) profiling, and iii) outlines the preanalytical variables integral for the scalability and reproducibility of a complete analysis of the liquid biopsy.

Materials and Methods

Study Design

The main goals of this study were to complete a comprehensive analysis of the preanalytical variables for five BCTs, four TTAs, and FVF for the genomic assessment of the cellular and acellular fractions of the liquid biopsy in a large cohort of primary breast cancer patients. To determine the blood collection, handling, and processing variations across a wide range of disease states to represent the clinical scenario, the patient cohort consisted of nonmetastatic, treatment-naive patients and metastatic patients. Patients with breast cancer participated in the ongoing prospective Physical Sciences in Oncology study (PSOC-0068) entitled OPTImization of blood COLLection (Figure 1). A total of 163 patients were enrolled between April 2013 and January 17, 2017, at multiple clinical sites in the United States [namely, Billings Clinic (Billings, MT), Duke University (Durham, NC), City of Hope Comprehensive Cancer Center (Duarte, CA), MD Anderson Cancer Center (Houston, TX), University of Southern California Norris Comprehensive Cancer Center (Los Angeles, CA), University of Southern California Norris Oncology/Hematology (Newport Beach, CA), and Los Angeles County and University of Southern California Medical Center (Los Angeles, CA)]. Patient recruitment took place according to an institutional review board–approved protocol.

Figure 1.

OPTImization of blood COLLection (OPTICOLL) study design. A: Complete analysis of the preanalytical variables for five BCTs, four TTAs, and fresh versus frozen (FVF) for the genomic assessment of the cellular and acellular fractions of the liquid biopsy. B: A wide range of patients with primary breast cancer participated in the ongoing prospective Physical Sciences in Oncology study (PSOC-0068) entitled OPTICOLL.

Recruitment and study schedules were established and unified across clinical sites for patients with nonmetastatic, treatment-naïve disease and metastatic disease.¹¹ Nonmetastatic, treatment-naïve patients had two study-related blood draws: before any initial treatment and 3 to 8 weeks after surgical resection of the primary lesion with or without adjuvant drug treatment as primary therapy. Patients with metastatic disease were eligible for participation at the start of a new line of therapy, either as a first line of therapy or after experiencing progression while on therapy for treatment of metastatic disease. These patients had a potential for up to 14 study-related blood draws at intervals of 8 to 12 weeks, coordinated with routine clinic visits. This collection scheme was designed with a primary goal of screening and reproducible identification of circulating rare cells, and only as a secondary goal for clinical correlations. An institutional review board at each site approved all procedures, and all study participants provided written informed consent.

To complete the prior analysis of the most commonly used BCTs (cfDNA, EDTA, acid-citrate-dextrose, and heparin), the CellSave Preservative BCT (catalog number 7900005; Veridex LLC, Raritan, NJ) was employed to evaluate any differences in rare cell detection compared with the previously superior cfDNA Streck BCT (catalog number 218962; Streck Laboratories, Omaha, NE) in peripheral blood samples collected from breast cancer patients.¹¹ A total of 23 patients provided 33 pairs of BCTs with 8 mL of blood collected in each and processed at 24 hours for rare cell detection with the goal of determining the optimal BCT for the HD-SCA workflow.

For complete analysis of up to 96 hours TTA, study participants underwent phlebotomy of up to 40 mL of blood into a predetermined set of BCTs. Tube brand and lot number were tracked between clinical sites and throughout the study. A total of four Streck BCTs with 8 mL of blood were collected at each draw: 24, 48, 72, and 96 hours TTA. The acceptable period defining each TTA was ±8 hours; therefore, 24-hour TTA samples were processed 16 to 32 hours after sample collection. Plasma from each patient draw collected in the Streck BCT and processed at 24-hour TTA was divided into two equal aliquots to generate fresh and frozen plasma preparations.

This study tested components potentially leading to large degrees of preanalytic variability, made constant several others by strict prescription of methods, and tracked many additional variables that may add variability to the final results, but may not necessarily be controllable. The formally tested variables were TTA for rare cell detection and FVF plasma samples for cfDNA genomic analysis. Those variables held constant by conformity to strict protocols, as well as those inherently less controllable parameters, were tracked, but not controlled for with a standardized questionnaire.¹¹ Clinical variables, as available, were additionally recorded.

Blood Collection

Standard operating procedure for blood collection was provided to each clinical site, including descriptive details on collection into the five specialized and proprietary tube types, in no particular order, through no particular needle type, with room temperature storage. A questionnaire was developed to assess the preanalytic variables related to collection and handling of blood samples with the purpose of tracking what cannot be prescribed. Variables, such as type of draw, anatomic location, collection device, BCT order, and needle gauge, were tracked. The blood draw questionnaire data for each draw were collected from each clinical site. The blood collection questionnaire used in this study can be located at http://kuhn.usc.edu/OPTICOLL/HDSCA_Blood_Draw_Questionnaire.pdf (last accessed August 6, 2019).

Shipping

All BCTs were packaged and shipped to the processing laboratory using a validated temperature-stable shipper, International Organization for Standardization–certified GreenBox Thermal Management System (ThermoSafe Brands, Arlington Heights, IL) and Standard71 shippers (Paradigm Design Solutions, Los Angeles, CA). Shipping events were tracked and monitored throughout the study. Each shipper received from blood collection sites was visually inspected for container integrity on unpacking at the laboratory. When components showed damage, the dysfunctional parts were replaced, or the entire box was discarded. Temperature maintenance tests of 10% of the shippers were completed with the use of XpressPDF temperature labels (PakSense; Emerson Cargo Solutions, Boise, ID).

Parallel Enumeration of Rare Cells by CellSearch and the HD-SCA Workflow

A total set of 40 peripheral blood samples were collected from 25 patients with breast cancer in a multicenter study.¹¹ All patients were diagnosed with organ-confined or metastatic breast cancer. For each patient, up to 10 mL of peripheral blood was collected into CellSave blood collection tubes (BCTs; Veridex LLC) for the CellSearch test and 7.5 mL of blood was collected in Cell-Free DNA BCTs (Streck) for the HD-SCA workflow. The first 2 mL of blood drawn was discarded to remove any potential contamination. Blood was drawn at various time points during treatment. This study was approved by the institutional review board at Billings Clinical Hospital and Duke University Comprehensive Cancer Center. Informed consent was obtained from all participating patients. CTC enumeration was conducted using the CellSearch system (Veridex LLC), according to the manufacturer's protocol by a third-party laboratory. Enumeration of HD-CTCs by HD-SCA was conducted as described below.

Identification and Characterization of Rare Cells Using the HD-SCA Workflow

All blood samples were processed as previously described.^11,12 Each sample was treated independently regardless of TTA. A maximum of 12 replicate slides for rare cell identification and characterization were prepared and stored at −80°C until further analysis. One test for detection of candidate cells consists of two slides. An immunofluorescence staining protocol based on the published HD-SCA workflow¹¹ was used, which included an antibody cocktail of pan-cytokeratin (CK; catalog number C2562; Sigma-Aldrich, St. Louis, MO), anti-CK19 (catalog number M088801; Agilent Dako, Santa Clara, CA), anti-CD45 (catalog number MCA87A-647; Bio-Rad, Hercules, CA), anti–estrogen receptor (SP1; catalog number RM-9101-S; Thermo Fisher Scientific, Waltham, MA), and DAPI.¹¹ Secondary antibodies were catalog number A-21127 and A-11034 from Life Technologies (Carlsbad, CA). The number of total retained cells was estimated using the count of the DAPI-stained nuclei. Cells that were CK⁺, CD45⁻, with intact nucleus, and generally larger and morphologically distinct from surrounding white blood cells (WBCs; HD-CTCs), as well as cells that only partially met these criteria (marginal CTC populations), were recorded.¹³ Marginal populations included the following: i) CTC small: CK⁺, CD45⁻, cells with intact nuclei that were the same size or smaller than neighboring WBCs; ii) CTC low CK: cells with CK levels lower than HD-CTCs or absent, CD45⁻, and large morphologically distinct nuclei; and iii) CTC cfDNA producing: CK⁺ CD45⁻ cells with a DAPI pattern of nuclear condensation and fragmentation and plasma membrane blebs that are common features of apoptotic cells.^14,15 In addition to CTC enumeration, the high-content data consisted of six additional parameters: total number of nucleated cells per slide, total number of candidate cells per slide, relative nuclear area per CTC, relative cytokeratin and estrogen receptor staining intensities per CTC (represented as the SD over the mean signal intensity of the cell of interest to the nearest 50 cells), and estrogen receptor localization per CTC.

Rare cell enumeration was conducted for all samples received. Slides from the Streck BCT at 24-, 48-, 72-, and 96-hour TTA were analyzed to determine the best TTA for rare cell detection using the HD-SCA workflow. CTC-positive samples were defined by detection of ≥1 HD-CTC across a test, consisting of two slides. For each CTC-positive 24- to 48-hour TTA-matched patient sample, defined by the presence of CTC candidate events at both TTAs, the resulting cells were used to determine if a difference in single-cell genomic analysis was detectable between the 24- and 48-hour TTAs (single-cell NGS). Negative pairs were 24- to 48-hour TTA-matched samples that were evaluated and for which CTC candidate events were not found at the 48-hour TTA. Single CTCs harvested from CTC-positive patients were analyzed using NGS-based whole genome CNV and targeted sequencing analysis methods.

Assessment of the Variability between Autostainer Runs

A previously established quality assurance and quality control program required performance of spike-in experiments with MCF-7, MDA-MB-231, and SKBR3 cells in normal control blood samples.^12,16 A minimum of two slides of a positive control were routinely included in each autostainer run with slides from patients’ blood samples and subsequently scanned and evaluated. Cell enumeration results in these positive control slides were used to construct a Levey-Jennings chart.

Genomic Characterization of Single Cells from Liquid Biopsies

Isolation of Single Cells

CTCs identified in the 24- and 48-hour matched samples were relocated on the glass slides and reimaged at an objective of 40× for detailed cytomorphometric analysis. For cell extraction, an Eppendorf Transfer Man NK2 micromanipulator (Eppendorf AG, Hamburg, Germany) was used to capture the cell of interest inside a 25-degrees jagged micropipette by applying fluid suction. Once the cell of interest was captured inside the micropipette, the cell was deposited inside a 0.2-mL PCR tube containing 1 μL of Tris-EDTA buffer. The sample was then immediately frozen and stored at −80°C until further processing. The cell-containing vials were transferred on dry ice to the sequencing laboratory, where the lysed cell mixture was thawed and subjected to whole-genome amplification (WGA) and sequencing library construction.

CTCs from blood samples collected from nonmetastatic, treatment-naïve patients and metastatic breast cancer patients were used to compare genome-wide CNV profiles derived from low-depth sequencing of amplified single-cell DNA from the 24- and 48-hour TTAs, as previously described.⁹

WGA of Single Cells

WGA of single cells was performed using the WGA4 Genomeplex Single Cell Whole Genome Amplification Kit (catalog number WGA4; Sigma-Aldrich). Briefly, the cells were lysed by adding 1.5 μL lysis buffer to each single thawed cell (1:1 solution of 100 mmol/L dithiothreitol and 400 mmol/L KOH) and subsequent incubation for 2 minutes at 95°C. A master mix containing 6.5 μL 10 mmol/L Tris-HCl-EDTA, pH 8.0, per reaction and 1 μL of the 10× Single Cell Lysis & Fragmentation Buffer (provided in the WGA4 Genomeplex kit) was added to the cold reaction. The samples were incubated during 4 minutes at 99°C. Library preparation and amplification were performed according to the manufacturer's protocol, with the exception of amplification performed with 23 instead of 25 PCR cycles. Final sample volume was 75 μL.

Successful amplification was confirmed by gel electrophoresis (1.5% agarose gel), loading 10 μL total volume of each sample. A smear of approximately 200 to 1200 bp was expected. Furthermore, DNA was purified using a QIAquick PCR Purification Kit (catalog number K210012; Thermo Fisher Scientific). DNA was eluted in 50 μL of Tris-EDTA buffer. Concentration of amplified and purified DNA was quantified with Qubit Fluorometric Quantification (Thermo Fisher Scientific). Amplified DNA was sheared using sonication (Covaris S2/E210 focused ultrasonicater; Covaris, Woburn, MA) with the microtube setup and the 200-bp target size protocol for DNA shearing. Volume and concentration depended on the planned downstream experiments (compare input DNA for library construction for targeted sequencing or/and for CNV analysis).

Copy Number Variation Analysis of Single Cells

CNV analysis of the CTCs at the 24-hour TTA was compared with that of the 48-hour TTA. For this variable, 50 ng of amplified and sonicated DNA from single cells (for details, see WGA of Single Cells) was used for library construction using the NEBNext Ultra DNA Library Preparation Kit for Illumina (catalog number E7370L; New England Biolabs, Ipswich, MA). The constructed library DNA concentration was quantified with Qubit, and the expected library size distribution of 300 to 500 bp was confirmed using the Agilent 2100 Bioanalyzer (High-Sensitivity DNA Assay and Kit; catalog number 5067-4626; Agilent Technologies). The individual libraries from single cells with identifiable and distinguishable indexes were pooled (approximately 6 ng per library; 50 to 70 single cells). The pooled libraries were cleaned using AMPure XP Beads (catalog number A63882; Beckman Coulter Inc., Brea, CA). Adaptors and primers were eliminated from Illumina library using 0.8× dilution of magnetic beads.

Libraries were sequenced using the Illumina NextSeq 500 or the HiSeq2500 SR50, generating FASTQ files. A total of 30 bp was trimmed off the ’5 end of each read to remove the WGA4 adapter sequence before alignment to the hg19 reference genome using the Bowtie algorithm. The resulting BAM file was sorted, and PCR duplicates were removed using SAMtools. The number of reads falling into each of 5000 bins, comprising the entire University of California, Santa Cruz, reference genome, was calculated using a previously published Python script.¹⁷ Finally, an R script using the Bioconductor package, DNAcopy_1.26.0 (http://bioconductor.org/packages/DNAcopy, last accessed August 6, 2019), was used to normalize and segment the bin counts across each chromosome, generating a genome-wide CNV profile.

CNV profile comparison was conducted using a calculated quality score using equation 1. With fewer reads, there is greater noise in the data, so the variance of the low ratio around the segment value must be adjusted for the total amount of data. A high-quality score indicates a large variance due to greater noise; hence, the lower the quality score, the more precise the CNV profile.

$$var\left(\frac{x\$lowratio}{x\$seg.mean.LOWESS}\right)\times \surd \left(\frac{sum\left(x\$bincount\right)}{1000}\right)$$ (1)

The genomic instability was calculated as the ratio to the median using a hyperbolic tangent function. Using the logarithmic function allows for the score of one complete deletion to be equivalent to the score of one complete amplification. The further the ratio to the median is from 1, the higher the score. The calculation of genomic instability is a summation of equation 2 for every bin.

$$\tanh \left(3\times \left|\log \left(\frac{x}{median\left(x\right)}\right)\right|-2\right)+\text{tanh}\left(2\right)$$ (2)

Single-Nucleotide Variation of Single Cells

Targeted sequencing of the CTCs at the 24-hour TTA was compared with that of the 48-hour TTA. The NEBNext Direct Cancer HotSpot Panel for Illumina (catalog number E7000X; New England Biolabs) was employed for target enrichment from amplified single-cell DNA. The kit uses hybridization-capture with biotinylated single-strand DNA as baits, and targets 190 hot spots from 50 cancer-associated genes (oncogenes and tumor suppressor genes) encompassing approximately 40 kb of sequence and including >18,000 Catalogue of Somatic Mutations in Cancer (COSMIC) database features.

A total of 100 ng of amplified and sonicated DNA from single cells (for details, see WGA of Single Cells) was used to construct libraries, according to the manufacturer's instructions. The constructed libraries were amplified over 23 cycles. DNA concentration was quantified with Qubit, and the expected library size distribution of 300 to 350 bp was confirmed using the Agilent 2100 Bioanalyzer. The individual libraries from single cells with identifiable and distinguishable indexes were pooled (5 ng per library) together to fulfill the Illumina high-throughput capacity and further cleaned using 0.8× AMPure XP Beads to remove remaining adaptors. The pooled libraries were sequenced using Illumina HiSeq2500 PE100, generating FASTQ files.

FASTQ data were uploaded and analyzed on PartekFlow, a licensed NGS data analysis platform broadly used at University of Southern California. Briefly, original FASTQ reads were first trimmed from 5′ end with 30 bp to remove single-cell amplification primers; then, they were trimmed from 3′ end to remove all reads with low quality (<20). The BWA-MEM alignment program (https://github.com/lh3/bwa, last accessed August 6, 2019) was used to generate BAM files against hg19 (http://hgdownload.cse.ucsc.edu/goldenPath/hg19/bigZips, last accessed August 6, 2019). The target region was filtered with E7000baitedregionsGRCh37hg191.bed, provided by the manufacturer, before variant calling. This BED-format file, which contains a single track of the genomic regions of interest for this custom Direct Cancer Panel, was used to assess the exact regions to cover during bioinformatic analysis. The mpileup command in the SAMtools suite was used as the variant caller, and vcf files were generated for annotation with GENCODE Genes-release 19 and dbSNP_146.

Characterization of Circulating cfDNA from Liquid Biopsies

The cfDNA was isolated from matched fresh and frozen plasma samples from breast cancer patients. The impact of freezing was compared in the matched samples using CNV analysis and targeted sequencing of a panel of genes. To isolate the plasma, whole blood samples were centrifuged at 2000 × g for 10 minutes for collection of at least 2 mL of plasma per blood tube. Initial total whole blood volume was reconstituted with phosphate-buffered saline, and blood underwent red blood cell lysis for subsequent CTC analysis. Plasma was transferred into a new vial and centrifuged twice at 14,000 × g for 10 minutes. One of the matched pairs was immediately used for cfDNA isolation (fresh), whereas the second sample was stored at −80°C and cfDNA was extracted 14 days after time of draw (frozen).

Isolation of cfDNA

Cell-free DNA purification from ≥2 mL of fresh and frozen plasma samples using the QIAamp Circulating Nucleic Acid Kit (catalog number 55114; Qiagen, Germantown, MD) was conducted, as indicated in the manufacturer's protocol. If available plasma per paired sample was >5 mL, the protocol was performed in duplicate. DNA of each column was eluted in 55 μL of elution buffer. Frozen samples were thawed on ice for 30 minutes, equilibrated to room temperature, and processed as for fresh sample.

Copy Number Variation Analysis of cfDNA

A total of 5 ng of input DNA was used for library construction with the NEBNext Ultra II DNA Library Prep Kit for Illumina (catalog number E7645L; New England Biolabs), according to manufacturer's instructions. After adaptor ligation, a cleanup without size selection of adaptor-ligated DNA was performed using AMPure Beads before PCR enrichment. Seven PCR cycles were performed to generate a library of sufficient yield. The constructed library DNA concentration was quantified with Qubit, and the expected library size distribution of 300 to 500 bp was confirmed using the Agilent 2100 Bioanalyzer.

The individual libraries from each plasma cfDNA sample were pooled at equimolar concentrations and cleaned again using AMPure XP Beads. The pooled libraries were sequenced using the Illumina NextSeq 500 or the HiSeq2500 PE100. The resulting FASTQ files were aligned to the hg19 reference genome using the Bowtie algorithm. The resulting BAM file was sorted, and PCR duplicates were removed using SAMtools. The number of reads was counted in each bin using a Python script.¹⁷ Finally, an R script using the Bioconductor package DNAcopy_1.26.0 (http://bioconductor.org/packages/DNAcopy, last accessed August 6, 2019) normalized the bin count and segmented the bin counts into a CNV profile.

Single-Nucleotide Variation Analysis of cfDNA

A total of 56 pairs of FVF cfDNA were tested with QIASeq Target DNA Breast Cancer Panel (catalog number DHS-001Z; Qiagen, Germantown, MD), followed with Illumina HiSeq2500 PE100 run. QIASeq Targeted DNA panel was used to construct targeted, molecularly barcoded libraries from cfDNA for digital sequencing with NGS using a minimum of 20 ng cfDNA. The constructed molecular barcoded library was quality checked and controlled with Qubit, and the expected library size distribution of 300 to 450 bp was confirmed using the Agilent 2100 Bioanalyzer and real-time PCR. The Illumina HiSeq 2500 PE100 run was performed by the University of Southern California University Park Campus Genome Core facility, generating FASTQ files.

FASTQ data were uploaded and analyzed with QIAseq DNA V3 Panel Analysis plugin on Biomedical Genomics Workbench v3, provided by Qiagen. This QIAseq DNA V3 Panel Analysis plugin is a ready-to-use workflow that can identify and annotate variants in targeted amplicon sequencing data generated with QIAseq Targeted DNA V3 Panels. This panel integrates molecular barcode technology into a highly multiplexed PCR-based target enrichment process, enabling accurate variant calling at low frequency. In brief, all sequenced reads were imported into the workbench; then, the reads were trimmed to remove Nextera sequencing adapters and QIAseq Targeted DNA V3 universal adapters, which were introduced into sequences during library construction. Alignment was performed against hg19. Aligned reads were further filtered with the Qiagen-provided BED file DHS-001Z.BED. Low-frequency variant detection was performed under 1.0% required significance (the cutoff value for the statistical test for the variant not being due to sequencing errors), and only positions in the regions specified in the BED files will be inspected for variants. Only variants in regions covered by at least 30 reads are called, and only variants that are present in at least three reads are called.

Statistical Analysis

Graphical representation and statistical analyses were performed using GraphPad Prism software version 5 (GraphPad Software, Inc., La Jolla, CA) or RStudio (R version 3.2.4; RStudio Team; 2015; RStudio: Integrated Development for R; RStudio, Inc., Boston, MA). Wilcoxon matched-pairs signed rank test or U-test was used to compare various assay results in BCT and TTA-matched samples. FVF analysis results were analyzed by Wilcoxon signed-rank test and two-way analysis of variance. P ≤ 0.05 was used to declare a significant difference between groups. Effect size for nonparametric analysis was calculated using equation 3 and interpreted using Cohen's effect size estimates (small, 0.2; medium, 0.50; large, 0.80).¹⁸

$$r=\frac{Z}{\sqrt{n_x+n_y}}$$ (3)

Morphometric Density Plots

Single-parameter density plots were generated in R using the stats library to inform on the morphometric differences underlying patients categorized as nonmetastatic or metastatic by the clinicians at each institution. Computational bias was reduced by optimizing the ratio of cells per patient and total patients. For each selected parameter, the two highest and two lowest values were selected as representatives of that particular patient and combined with other patients of the two categories.

Data Availability

The single-cell morphometric and genomics data and cfDNA genomics data that support the findings of this study are available at the National Cancer Institute's Center for Strategic Scientific Initiatives Data Coordinating Center (https://cssi-dcc.nci.nih.gov/cssiportal/view/59cd0c0c34b81e4633fd309b, last accessed August 6, 2019).

Results

Study Cohort

A total of 163 patients were accrued for this study, consisting of 98 (60.1%) treatment-naive, nonmetastatic patients and 65 (39.9%) metastatic patients (Figure 1). Patient demographics are provided in Supplemental Table S1. The total sample set consisted of 689 blood draws from these patients, with a total of 595 BCTs processed and analyzed for rare cell enumeration.

CTC enumeration of all samples resulted in a median of 0 (range, 0 to 2382) HD-CTCs/mL, in which 98 (60.1%) patients had detectable CTCs in at least one blood draw. The treatment-naive, nonmetastatic patient samples had a median of 0 (range, 0 to 503.4) HD-CTCs/mL, with 62 (63.3%) patients presenting with at least one sample positive for CTCs. The metastatic patient population had a median of 0 (range, 0 to 2382.7) HD-CTCs/mL, in which 36 (55.4%) patients were positive for CTCs in at least one blood draw. HD-CTC enumeration was not significantly different between treatment naive patients with local disease and metastatic breast cancer patients (U-test; P = 0.6966). The distribution of HD-CTCs/mL, identified by draw number, is shown for metastatic and nonmetastatic patients in Supplemental Figure S1.

The previously defined CTC candidate subpopulations were also detected by the HD-SCA workflow.^11,12,19,20 The distribution of each candidate CTC subclassification by stage is shown in Supplemental Figure S1. There was no significant difference between metastatic and nonmetastatic patients in CTC low CK, CTC small, and CTC cfDNA producing cell enumerations (U-test; P = 0.1633, P = 0.9883, and P = 0.5092, respectively).

Quality Control/Quality Assurance

Shipping

Shipping events were tracked and monitored throughout the study. Each shipper received from blood collection sites was visually inspected for container integrity on unpacking at the laboratory. When components showed damage, the dysfunctional parts were replaced, or the entire box was discontinued. Temperature maintenance tests of 10% of the shippers was completed using XpressPDF temperature labels (PakSense). Blood samples were shipped at an average temperature of 22.51°C ± 1.62°C to the central laboratory via FedEx priority overnight (Supplemental Figure S2A). These results indicated that the thermal insulation of the shippers maintained the temperature within the acceptable range, which is ±10°C from original temperature at time of box closure. In general, this range falls between 15°C and 25°C.

During the period spanning November 2015 to January 2017, five sample shipments (0.7%) were received from sites participating in the study that were flagged as deviations. Three sample sets were delayed arrivals to the processing laboratory, one sample set arrived with temperature inside the shipper that was below the acceptable range, and one sample set was a difficult draw and only two BCTs could be obtained. These patient draws were excluded from any further analyses.

Assessment of the Variability between Autostainer Runs

At various points in the HD-SCA workflow, quality assurance and quality control checkpoints are used to ensure that test methods are accurate. Cell enumeration results from positive control slides were used to construct a Levey-Jennings chart to assess potential variability between autostainer runs. Results indicated no out-of-range controls using the 2 SD × 1 rule (ie, when control limits are set as the means ± 2 SDs) (Supplemental Figure S2B). This indicates that the autostainer process was meeting the intended specifications for rare cell detection.

CellSearch Correlation

A preliminary study was conducted to correlate CTC enumeration, measured by the CellSearch system, and HD-SCA workflow from matched patient blood draws. The CellSearch system uses the cell surface antigen epithelial cell adhesion molecule for immunomagnetic enrichment, followed by immunofluorescence staining for CK. The CTC counts for the two methods obtained were strongly correlated (r = 0.9992; P < 2.2 × 10¹⁶). The HD-SCA workflow detected significantly more CTCs compared with CellSearch in a higher proportion of patients (P = 0.0028) (Supplemental Figure S3A). Fifteen samples evaluated by the HD-SCA were CTC positive, whereas only five were found to be CTC positive using the CellSearch assay (>5 CTCs/7.5 mL) (Supplemental Figure S3B). Agreement in sample CTC positivity between both detection techniques was observed in 30 (75.0%) blood samples. Of the 40 total blood samples, 31 (77.5%) had HD-CTC counts equivalent or higher than the CTCs/mL identified by CellSearch. Among the 35 CTC^negative samples using CellSearch detection, 10 (25.0%) were found positive by HD-SCA. The HD-SCA workflow exhibited higher sensitivity, as indicated by the higher CTCs/mL, in a greater proportion of patients compared with the CellSearch assay in matched samples. The results indicate the HD-SCA workflow can predict CTC positivity in the CellSearch system. At a threshold of 5 CTCs/7.5 mL, with a sensitivity of 80%, the specificity is 82.86% to 97.14% (Supplemental Figure S3C).

The HD-SCA workflow has previously been shown to demonstrate robust performance in cell lines, as well as a variety of patient samples.^{9,10,12,13,19,}21, 22, 23, 24, 25, 26, 27, 28 The higher sensitivity of this platform is not surprising given that the approach is inclusive of all CK-positive and CD45-negative cells. Although this preliminary study was not specifically powered to compare clinical outcomes between the two approaches, it can be concluded that both systems provide comparable results in this type of cohort. The HD-SCA workflow, although not relying on enrichment of the CTC population, is an all-inclusive system for rare cell detection and subsequent downstream genomic and proteomic analysis at the single-cell level, which is a more comprehensive approach for liquid biopsy analysis.^29,30 The following study was conducted to validate the reproducibility and scalability of the HD-SCA workflow with the primary objective of enabling evidence-based scientific discovery and clinical utility.

Analysis I: Effect of Blood Collection Tube Type in Rare Cell Enumeration

In a previous study, the Streck BCT was shown to provide more sensitivity and the least amount of imaging artifacts and cellular debris in the HD-SCA workflow compared with EDTA, acid-citrate-dextrose, and heparin BCTs.¹¹ To complete the analysis of the most commonly used BCTs and complement the previous study, the CellSave Preservative BCT was employed to evaluate differences in rare cell detection compared with the Streck BCT with the goal of determining the optimal BCT for the HD-SCA assay. Quantitative comparison was conducted on 33 sets of two BCTs (CellSave and Streck) (Figure 2) processed at 24 hours. Cell retention, as determined by the total DAPI nuclear stain count, was significantly different between the CellSave and Streck BCTs, with the CellSave BCT yielding approximately 10% fewer cells (P = 0.0005). There was no significant difference in the enumeration of HD-CTCs/mL between BCT types. In the previous publication, cell retention was not different between Streck, EDTA, and citrate BCTs, but there were 30% fewer cells from the heparin BCT. In addition, the highest number of HD-CTCs/mL was detected from the Streck BCT, whereas there was no difference in cell counts among EDTA, citrate, and heparin BCTs.¹¹

Figure 2.

Comparison of BCTs from a subset of 33 matched blood samples. A: Number of total retained cells, as estimated by the DAPI nuclear stain, from patient samples collected in either a CellSave or a Streck BCT and processed within 24 hours. Median and interquartile range are indicated in red. Wilcoxon matched-pairs signed rank test was used to compare BCT-matched pair results. P = 0.0005. B: Distribution of HD-CTCs/mL, as determined by the HD-SCA workflow. P = 0.1075.

The morphometric parameters of the leukocytes (WBCs) and the HD-CTCs per sample were analyzed to determine if the cellular morphology varied between BCTs. The difference in cell retention was obvious when comparing the number of WBCs detected per slide. An average of 2.217 × 10⁶ (median, 2.245 × 10⁶; range, 9.26 × 10⁵ to 2.919 × 10⁶) WBCs were analyzed per slide from samples collected in the CellSave tube, whereas the Streck BCT samples had an average of 2.463 × 10⁶ (median, 2.492 × 10⁶; range, 1.114 to 2.917 × 10⁶) WBCs per slide. There was a significant difference in the nuclear morphology of the WBCs between BCT types, in which the CellSave BCT WBCs had a less circular, more elliptical nuclei than the Streck BCT WBCs (Supplemental Figure S4A and Supplemental Table S2). No significant differences were identified in the other morphometric parameters for the WBCs by BCT type.

There was a significant difference in the nuclear morphology of the HD-CTCs identified from each BCT (Supplemental Figure S4B and Supplemental Table S3). The CellSave samples had a smaller nuclear area, and thus a lower nuclear area ratio to the surrounding WBCs, with greater concavity and less circular, more elliptical nuclei compared with the HD-CTCs. Additional parameters related to nuclear morphology were examined and revealed the same trend (nuclear aspect ratio, nuclear area local ratio; P < 0.0001). From this preliminary study, it can be concluded that the Streck BCT outperformed the CellSave BCT in the HD-SCA workflow by providing more efficient cellular adhesion and retaining a greater quality of cellular morphology. This may be due to the differences in preservatives between BCTs, potentially causing overfixation, altering the adherent properties of the cells and thus the results of the assay.

Overall, the results of the five BCTs indicate that the Streck BCT showed the best performance at 24 hours in rare cell detection using the HD-SCA workflow. Therefore, samples collected in the Streck BCT were evaluated for the four TTA analyses.

Analysis II: Effect of TTA in CTC Enumeration

Analysis was conducted to evaluate a complete set of four TTAs (24, 48, 72, and 96 hours) using 87 matched sets of peripheral blood samples to unify the preanalytical data associated with the HD-SCA workflow. TTA-matched blood samples were obtained from 35 unique patients enrolled in the study at participating clinical sites. Cell retention and the distribution of CTCs per TTA in this subset are shown in Figure 3. Cell retention was shown to be compromised by prolonged TTA. At 48 hours, there was an approximate loss of 8% of cells compared with 24 hours, whereas 72 and 96 hours showed a loss of 29% and 42%, respectively. The results support the finding from the previous study comparing 24- and 72-hour TTA.¹¹ In the four TTA analysis presented herein, there is a significant difference between 24- and 48-hour TTA in the total number of candidate cells/mL and HD-CTC/mL, as well as all subclassifications of CTCs detectable by the HD-SCA workflow. Overall, the 24-hour TTA provided the most utility of the four TTAs analyzed because of the best overall cell retention and high efficacy in the identification of rare cells.

Figure 3.

Comparison of four TTAs from a subset of 87 matched blood samples. A: Number of total retained cells, as estimated by the DAPI nuclear stain. B–F: Total number of candidate cells (B), HD-CTCs (C), CTC small cells (D), CTC low CK cells (E), and CTC cfDNA producing cells (F), as determined by the HD-SCA workflow. Median and interquartile range are indicated in red. Wilcoxon matched-pairs signed rank test was used to compare TTA-matched pair results. *P < 0.05, **P < 0.001, and ****P < 0.0001.

To further evaluate the different effects of 24- versus 48-hour TTA in the enumeration of rare cells, a larger sample set of 167 matched draws from 87 unique patients was screened for CTC positivity, which was defined by the detection of at least one HD-CTC in a test of two slides. Ninety-three draws were found to be CTC negative at 24-hour TTA. The remaining 74 draws were CTC positive at 24-hour TTA (44.3%), with 50 of these draws additionally found CTC positive at 48 hours. There was no significant difference in the total number of candidate cells detected at 24 and 48 hours in this subset of 24-/48-hour positive matched samples (Figure 4A), whereas cell retention was decreased (6.5%) at 48 hours (Figure 4B). The ratio of total candidate cells detected per million cells retained across two slides was not different (Figure 4C). In addition, the distribution of HD-CTC and marginal cell population (CTC small, CTC low CK, and CTC cfDNA producing) was similarly comparable (Figure 4, D–G). Supplemental Table S4 provides an overview of the different metrics calculated in this set of 50 HD-CTC–positive, TTA-matched samples. The lack of significant difference in the 24-/48-hour TTA set compared with the four TTA set may be due to the larger sample size and use of only matched TTA positive samples. The HD-SCA workflow showed precision in the detection of CTCs from breast cancer patient samples up to a 48-hour TTA.

Figure 4.

Comparison of 24-/48-hour TTAs from a subset of 50 HD-CTC positive matched blood samples. A: Total number of candidate cells detected. B: Total number of cells available. C: The ratio of total number of candidate cells detected per total million cells in a test of two slides in a set of 50 TTA-matched samples processed at 24 and 48 hours. D–G: CTC candidate cells enumerated per test included HD-CTCs (D) and marginal CTC populations: CTC small (E), CTC low CK (F), and CTC cfDNA producing (G). Wilcoxon matched-pairs signed rank test was used to compare matched samples.

The subset of draws that were found CTC positive at 24 hours, but CTC negative at 48 hours, was further explored. Median HD-CTC per test at 24-hour TTA in this subset was lower [2.5; interquartile range (IQR), 7.0] compared with the previous subset of 50 CTC-positive draws (6.0; IQR, 37.3). Although HD-CTCs were not detected at 48 hours, associated cell subpopulations were detected (Supplemental Figure S5). A comparison of the TTA-matched results in this subset showed a significantly lower cellular enumeration at 48 hours for all categories (P < 0.05). This is not surprising given the lack of HD-CTCs detected at 48 hours, suggesting a lower frequency of rare cells in general. Supplemental Table S5 provides an overview of the different metrics calculated in this set of 24-hour CTC-positive, 48-hour CTC-negative TTA matched samples.

Single-Cell NGS: Copy Number Profiling and Targeted Sequencing

Despite a lack of significant difference in the enumeration of HD-CTCs between 24-/48-hour TTA matched samples, the detectable cells were further analyzed to understand the effects of TTA on single-cell genomic analysis.

Copy Number Variations

Single cells were isolated from matched 24-/48-hour TTA samples and analyzed by CNV to determine potential effects of TTA at the genomic level. A total of 762 (77.4%) successful CNV profiles were generated, of which 590 (77.4%) correspond to HD-CTCs. The failure rate was not significant between 24- and 48-hour TTA. The remaining 172 single-cell CNV profiles correspond to cells identified as one of the three other candidate CTC classifications (CTC low CK, CTC small, or CTC cfDNA producing) detectable by the HD-SCA workflow or represent normal WBCs. Specific parameters reported include the DNA concentration (amplification yield) before library preparation, total read count, sequence coverage, and the quality score (Figure 5A). The total read count does not reflect the original DNA quality or region of interest; rather, it reflects the effort of sequencing the DNA fragments with an inherent bias based on machine and operator variability. The percentage aligned reads is the proportion of reads that align with the reference genome (hg19). The quality score for CNV analysis was calculated as the variance of the unsegmented value over the segment value, adjusted for the total amount of data.

Figure 5.

TTA comparison of single-cell genomic parameters. A: Analysis of the DNA quantity and quality of single cells isolated from 24-/48-hour matched samples. DNA concentration, percentage of aligned reads, total number of reads, and quality score from single-cell copy number variation analysis. The y axis is a logarithmic scale for total reads and quality score. B: TTA comparison of single-cell single-nucleotide variation (SNV) sequencing statistics. Analysis of the sequencing quality of single cells isolated from 24-/48-hour matched samples. Percentage coverage, variant count, total reads, and total alignments of reads from SNV analysis. The U-test was performed. ****P < 0.0001.

The total number of reads and percentage aligned reads detected at both TTAs were different (P = 0.0062 and P = 0.0055, respectively). At 24-hour TTA, the average total read count was 1.509 × 10⁶ (SEM, ±1.919 × 10⁵; median, 6.258 × 10⁵; IQR, 3.586 to 9.624 × 10⁵), with a percentage aligned reads of 47.92% (SEM, ±0.69%; median, 51.29%; IQR, 41.07% to 56.42%). The 48-hour TTA had an average total read count of 6.316 × 10⁵ (SEM, ±3.226 × 10⁴; median, 4.755 × 10⁵; IQR, 1.576 to 9.848 × 10⁵) and a percentage aligned reads of 49.48% (SEM, ±0.78%; median, 53.50%; IQR, 42.10% to 59.41%). This does not reflect the original DNA quality or region of interest. Although the parameters presented herein are necessary to understand the complexity of the genomic data, additional experimentation and analysis will likely be necessary to better understand the effects of TTA at a single-cell genomic level.

Interestingly, a significant difference in DNA concentration and CNV quality score was observed between TTAs (P < 0.0001). At 24-hour TTA, the initial average DNA concentration was 15.35 (SEM, ±0.92; median, 7.07; IQR, 3.32 to 21.20) ng/μL, whereas 48-hour TTA had an average of 26.37 (SEM, ±1.48; median, 18.40; IQR, 6.39 to 37.60) ng/μL. The average quality score of the CNV profiles at 24-hour TTA was 3.04 (SEM, ±0.38; median, 3.04; IQR, 0.709 to 2.272). At 48-hour TTA, the average quality score was 1.28 (SEM, ±0.11; median, 1.28; IQR, 0.519 to 1.255). The significantly lower DNA concentration observed at 24-hour TTA is theorized to be due to an on average longer time between immunofluorescence staining and cell isolation, indicating a greater potential for DNA degradation (Supplemental Figure S6). The time between staining and cell isolation was not controlled for, but analysis suggests this parameter should be considered as a significant contributor to the quality of genomic analysis. Overall, there is insufficient evidence to indicate an outstanding difference in single-cell CNV quality between 24-/48-hour TTA.

Single-Nucleotide Variations

Illumina sequencing libraries were constructed for SNV analysis of 126 single cells (62 cells from the 24-hour TTA and 64 cells from the 48-hour TTA) from five unique patients. Initial results indicated that libraries constructed from single cells required 100 ng of 200-bp amplified DNA fragments. Analysis of the sequencing statistics is shown in Figure 5B. Because SNV analysis uses pair end reading, each read is analyzed twice (once from each direction) and provides double the total reads. Total alignment count is the number of reads that align with the reference genome (hg19), whereas the alignments on the reference genome can be either on target or off target. On-target alignments are those reads that align to the region of interest designated by the kit. The off-target alignments are the reads that align to the reference human genome, but do not align to the region of interest. Percentage coverage for SNV has the off-target alignments filtered out, leaving the on-target alignments only for determining what percentage of the whole human genome they cover. Variant count is the number of reads with base pair changes (up to 3 bp) compared with the reference genome.

Total number of reads and total alignment were not significantly different between the TTAs (P = 0.0714 and P = 0.0786, respectively). The percentage coverage and total variant count were significantly greater in the 48-hour TTA (P < 0.0001). This may be because of the faster processing between thawing for immunofluorescence staining and single-cell isolation for whole genome amplification of the 48-hour cells (Supplemental Figure S6). Further analysis was conducted on the individual variant calls within each patient sample to determine if they were unique to either TTA or identified in both TTAs (Figure 6). The samples analyzed from one patient (BC-OPT-070) showed a greater number of unique variants at the 48-hour TTA compared with those identified in both TTAs and the 24-hour TTA alone. Despite this, neither the number of unique variant calls nor the gene region (intronic versus exonic) affected by the variant identified at 24 and 48 hours was significantly different overall (Supplemental Table S6). Together, this indicates single-cell SNV analysis is not significantly affected by an additional 24 hours (ie, 48-hour TTA).

Figure 6.

TTA comparison of single-cell single-nucleotide variation variant calls. A: Variant calls are identified as unique to either 24 or 48 hours, or identified in both TTAs from each patient (1 to 5). B: The region of interest affected is shown for each patient sample.

Analysis III: Effect of FVF on cfDNA

It is often necessary to freeze plasma samples for several weeks between initial tube processing and eventual extraction of cfDNA. To test the effects of freezing plasma at −80°C, incoming plasma samples were split, and DNA isolated from fresh aliquots was compared with matched aliquots frozen for approximately 2 weeks (actual range, 9 to 24 days, with a median of 14 days).

Copy Number Variations

DNA was isolated from a total of 60 FVF paired blood samples and prepared for sequencing and CNV analysis (see Materials and Methods). Average length and total amount of DNA after initial column purification of all samples showed the expected peak of approximately 166 bp for extracted DNA and approximately 300 bp for Illumina libraries. Samples had an average of 4.3 × 10⁶ reads, and all met the minimum requirement of 500,000 reads, with all but three samples having >1 × 10⁶ reads. As shown in Figure 7A, no significant statistical difference was observed between the cfDNA yield, the total number of aligned reads, the percentage of aligned reads, and the CNV quality score in the FVF matched pairs. This indicates that freezing the plasma samples does not significantly affect the yield or quality of the extracted DNA.

Figure 7.

Fresh versus frozen (FVF) comparison of cfDNA genomic parameters. A: DNA quantity and quality of FVF copy number variation matched plasma samples: DNA concentration, percentage of aligned reads, total number of reads, and quality score. B: Analysis of the sequencing quality of cfDNA isolated from FVF plasma samples: variant count, total reads, on-target percentage, and uniformity percentage. Wilcoxon signed-rank test was performed.

Single-Nucleotide Variations

Specific sequence changes in isolated cfDNA from paired FVF plasma samples from the same patient were compared (Figure 7B) using the methods described above for the single-cell TTA analysis. A strong linear correlation in the mutation frequencies for FVF was identified (R² = 0.8714; P < 0.0001), suggesting that the freezing process does not significantly affect the results of SNV analysis (Figure 8A). Likewise, there was no significant difference in the types of variant calls when compared by the region of interest affected (exonic versus intronic), type of mutation (missense, synonymous, or frameshift), and functional prediction (tolerated, damaging, or activating) (Figure 8, B–E).

Figure 8.

Fresh versus frozen (FVF) comparison of cfDNA single-nucleotide variation (SNV) variant calls. A: FVF comparison of cfDNA SNV mutational frequency. Linear regression analysis: R² = 0.8714; P < 0.0001. B–E: Variant calls are identified as unique to either the fresh or frozen condition, or identified in both conditions. The region of interest affected, translational impact, and functional prediction are shown for each patient sample.

Clinical Significance

Previous publications have reported clinical validity of rare cells detected by the HD-SCA workflow in multiple cancer types. Initial work by Marrinucci et al²³ showed that detectable HD-CTCs from a patient with stage IIIB well-differentiated lung adenocarcinoma had cytomorphology consistent with the primary tumor. Next, through the use of longitudinal sampling of the liquid biopsy, the time frame of tumor evolution in response to therapy was monitored through genomic analysis of HD-CTCs detected in an individual prostate cancer patient.⁹ In addition, the HD-SCA workflow can use bone marrow and primary tumor tissue samples to elucidate tumor evolution through proteogenomic evaluation of individual cells.³¹ The study presented herein was not specifically designed for clinical validation of the HD-SCA workflow, but some interesting trends in the data indicative of clinical utility can be found.

Through a retrospective analysis of the HD-CTC morphologic parameters collected from imaging analysis, three parameters were determined to vary by the stage of disease involvement, either treatment naïve/nonmetastatic or metastatic. A density plot of each parameter was generated from an equal number of HD-CTCs per patient to identify cellular features unique to each stage of disease involvement (Figure 9). Interestingly, patients with metastatic disease presented with two distinct populations of cells that varied by nuclear area and CK signal intensity. Nonmetastatic patients presented with a less heterogeneous population of HD-CTCs, consisting of cells with a small, round nucleus and low CK expression. These observations suggest that HD-CTC morphometrics could serve as a staging biomarker to assist clinical diagnostics or as a biomarker of progression to metastatic involvement in breast cancer.

Figure 9.

Density plot of the morphometric parameters of individual HD-CTCs detected in nonmetastatic and metastatic breast cancer patients. Nuclear area (A), nuclear roundness (B), and CK SD over the mean (SDOM; C) for four HD-CTCs per patient, stratified by stage of disease involvement.

Next, the morphometric parameters (nuclear area, nuclear roundness, and CK signal intensity) and a genomic instability score for each individual HD-CTC were used to evaluate a potential relationship with the patient's clinical response to treatment, determined by the reported disease status at the most recent follow-up visit. Clinical response was stratified into two categories: active disease (stable or progressive disease) and complete response.

Most patients with treatment-naïve, nonmetastatic breast cancer generally have a favorable prognosis. However, despite curative surgical resection, a significant number of patients experience tumor recurrence and cancer-related death.32, 33, 34 A prognostic stratification and an appropriate selection of those patients that will have limited response to primary therapy alone are major challenges to overcome. Morphometric analysis of HD-CTCs from nonmetastatic, treatment-naïve patients indicated that those individuals who responded to primary therapy have nuclei that were smaller and less circular at baseline, with cells that had a lower CK signal intensity at follow-up draws compared with patients with active disease (Figure 10). In addition, the genomic data were used to determine the relative genomic instability of each individual cell. Statistical analysis indicates that nonmetastatic responders presented with HD-CTCs at both time points with a greater overall genomic instability compared with nonresponders.

Figure 10.

Single-cell morphometrics and genomic instability of HD-CTCs in relation to patient clinical response. HD-CTCs from nonmetastatic (A–D) and metastatic (E–G) breast cancer patients were analyzed by nuclear area (A and E), nuclear roundness (B and F), CK SD over the mean (SDOM) before and after primary therapy (C), and genomic instability score in relation to clinical response (D and G). Clinical response was stratified into two categories: active disease (AD) and complete response (CR). Effect size for nonparametric analysis was calculated using equation 3 and interpreted using Cohen's effect size estimates. Kruskal-Wallis rank sum test was used.

Some studies have demonstrated the prognostic utility of CTCs in metastatic breast cancer patients. In this study, the metastatic patients who responded to therapy presented with HD-CTCs with larger, but again less circular, nuclei than patients who did not respond to treatment (Figure 10). Interestingly, metastatic patients who responded to treatment had a lower genomic instability at follow-up draws compared with those patients with active disease. This supports the need for additional work to prove clinical utility of HD-CTCs. Specifically, a larger sample set with survival follow-up is needed to determine if the preliminary associations between HD-CTC morphology and genomic instability and patient response are sustainable for potential clinical application. By expanding the cohort, whether the HD-SCA workflow is sensitive enough to demonstrate clinical validity can be determined.

Discussion

The importance of the liquid biopsy in the metastatic process and the potential use as a noninvasive route for cancer detection, characterization, and monitoring warrant the development of robust, reproducible CTC detection and characterization technologies for use in the clinical settings. The HD-SCA technology is a validated detection platform for rare cell characterization. As a member of the Blood Profiling Atlas Commons, we are making available standard operating procedures and data to the scientific community to advance the development of clinically useful tests by standardizing the methods for analysis of CTC morphology, genomics, and proteomics, with concurrent cell-free genomics.³⁵ The goal with the Blood Profiling Atlas Commons is to accelerate the development and validation of liquid biopsy assays to improve the outcomes of cancer patients. In the clinical realm, standard guidelines for reproducibility must be followed to allow for clinical decision making and enable robust, evidence-based science. The identification and elimination or minimization of specific preanalytic variables can increase the reliability, and most likely the reproducibility, for an unbiased assessment. This is of upmost importance in the investigation of biomarkers for clinical medicine.

Herein, preanalytical variables were tested to ascertain the HD-SCA workflow's capacity to detect CTCs and perform multianalyte genomic analysis in the context of accuracy, reproducibility, and specificity. This study, in combination with previous analyses, completes the evaluation of specific preanalytical variables in relation to the HD-SCA workflow: five BCTs, four TTAs, and FVF plasma preparation. Together, these findings demonstrate the reproducibility and scalability of the HD-SCA workflow in the detection of CTCs and genomic analysis of all major analytes of the liquid biopsy from breast cancer patients. This is an essential step toward clinical implementation of fluid or liquid biopsies that seek to recover and characterize whole cells and plasma.

Blood samples collected in the Streck BCT but processed at 72 hours instead of at 24 hours were shown to have fewer candidate cells detectable using the HD-SCA workflow, with the larger effect observed in the HD-CTC and CTC small subpopulations.¹¹ From matched positive samples, no difference was observed between 24- and 48-hour TTAs, preserving candidate cell enumeration. Thus, the conclusion can be drawn that HD-SCA performance is optimized for high detection levels of rare tumor cells when blood is collected into the Streck BCT and processed within 48 (±8) hours of collection. At 48 hours, TTA indicates that international shipments could be used to cover large, rich, and diverse populations, allowing for coordinated parallel and pooled analyses to better understand cancer from a global perspective.

In the current genomic era, different types of genome-scale data from the same individual are increasingly available. Development of applications to allow meaningful integration of genomic information is imperative to reveal essential disease-implicated mutational profiles at individual scale. Because of the varying purity of malignant cells in tissue samples that can dilute the signals of focal amplifications and homozygous deletions, effective molecular diagnostic tests require reproducible coverage over a broad dynamic range. Herein, a high efficiency and capacity for both CNV and SNV analyses at the single-cell and cfDNA levels of the liquid biopsy were demonstrated through the utility of the HD-SCA workflow. The performance of genomic analysis as part of the HD-SCA workflow was assessed using quality control metrics for analysis aimed to improve the repeatability and reproducibility of genomic analysis. The variability in basic sequencing metrics, such as the number of total reads and coverage of a particular location, was generally noted. Because the number of total reads directly influences coverage, it is the most easily quality-controlled factor that could affect reproducibility. Having a measure of confidence in genomic quality can help gauge the accuracy of the results. Overall performance assessment showed significant overlap between 24-/48-hour TTAs in CNV and SNV sequencing. CNV profile comparison was conducted using a calculated quality score metric indicating the variance due to sequencing noise. The 48-hour TTA had a higher initial average DNA concentration and better overall quality score. Significant differences in SNV metrics, such as percentage coverage and variant count, in which a broader range of values was seen, were also determined for the 48-hour TTA. Analysis suggests that the major factor impacting sensitivity was the time from immunofluorescence staining and candidate cell classification to cell isolation and subsequent genomic amplification. We conclude that a 48-hour TTA is acceptable for CNV and SNV profiling of the complete liquid biopsy. The genomic analysis described herein is applicable widely to liquid biopsy platforms that can isolate DNA from the single cell or plasma.

To assess how the sensitivity of SNV calling depended on the variant base quality and sequencing depth, the variant detection rate was analyzed. Because of concerns regarding reliability and usability of SNV findings, the variance count was used as a metric for confidence level. If multiple cells from the same sample have the same identical variant, then there is a high confidence in data. If only a single cell shows a variant, there is less confidence in the data. It was observed that most mutations in most patient samples were repeatable across TTAs. The type of nucleotide substitution and genomic location of the variant, coverage level, variant allele count, and variant allele frequency were not affected by TTA. Several factors may limit the conclusions drawn herein. First, cancer is highly heterogeneous, and so somatic SNVs might show up in one cell, but not another, from the same sample. Second, if the DNA sample is compromised to some extent, this will interfere with the ability to recover a full DNA profile from the sample. Of note, all of the variants are identified using high stringency settings defined through the following: consideration of hg19 data set of the human genome and minimum requirement of 50× sequencing depth as a useful threshold of variance calling for each examined position. Relaxing either of these requirements would substantially increase the number of the calls. Given the nature of the assay, finding mutations is not significant in the context of this study; rather, identifying differences in the unique alignment reflects the original amplicon coverage that may be affected by TTA preanalytical parameters.

The pairing of FVF plasma samples for each evaluated parameter was highly effective, indicating that the samples were similar between the FVF pairs, but not between patients. The lack of an observable significant difference in the cfDNA concentrations between FVF indicated neither a loss of DNA by freezing of the plasma samples nor a gain of additional DNA through the damage of potential cells remaining in the plasma during the freezing process. In addition, the number and percentage of aligned reads were comparable between FVF samples, indicating no change in quality of the DNA fragments. The variability in the days spent in the freezer did not have an appreciable effect on the DNA isolated or the quality of the genomic results, as shown by the CNV profiles and variant calls. We conclude that the freezing process of the plasma does neither diminish nor increase the DNA quantity, nor does it affect the quality of sequencing.

To optimize clinical management for individuals, underlying genetic causes would help determine prognosis, guide treatment, and conduct disease surveillance. Monitoring the emergence and alterations within the genome allows for personalized targeted therapy. Consistent with previous work,^9,10 the current study showed that genomic profiles of CTCs may be successfully generated, with the added integration of cfDNA genomic analysis. Variations in gene sequence or copy number may result in complete loss of function, partial decrease or increase in enzyme activity, or an altered affinity for substrates, which may in turn significantly impact a drug's efficacy. In the evolution of disease, cancer cells with gains and losses in copy numbers of specific chromosome regions are selected for metastases, ultimately leading to disease progression. Individual genomic analysis of single cells and cfDNA has enormous potential value for both the clinical and research communities.

It is important to consider that outputs will depend on the pipelines used and their experimental and algorithmic boundary conditions. Notably, most genomic platforms are designed for diploid genomes, and the tumor genomes often present with genome-wide ploidy alterations.^36,37 Alignment software is expected to preserve the ploidy alterations present in the sample while using generic definitions of diploid genomes, but loci different from diploid need to be treated consciously without bias. There is a strong demand for techniques that accurately measure mutational consequences and a scarcity of experimental data around preanalytical variation corresponding to direct downstream effects of the genomic analysis. A pipeline capable of processing individual patients and comparing the results across patient cohorts is highly desirable. By integrating genomic analysis into the HD-SCA workflow, single-cell genomics are anticipated to accelerate the ability to measure and interpret the functional consequences of genetic variation in relation to disease progression and treatment failure.

Any reliable diagnostic or prognostic clinical method needs to be highly reproducible. The enumeration of CTCs in the liquid biopsy of breast cancer patients is not prognostic in this study cohort, but in various clinical contexts of use may provide critical insights with specific analysis. Current technology does not indicate use of the liquid biopsy as a clinically diagnostic test; rather, it serves more as an important biospecimen noninvasively collected for cancer research. Ongoing studies aim to optimize the efficiency of single-cell genomic analysis, integrate proteomic analysis, and maximize clinical correlation. However, the results presented provide an important framework for understanding how preanalytic variables can impact the fluid biopsy test results. The HD-SCA performance quantified herein may be used as a guide for site-specific implementation into patient care and/or research biorepository processes. Statistical significance of the HD-CTCs in relation to patient outcome was shown as an insight into the potential clinical utility of the HD-SCA workflow to emphasize the need for further study to verify the clinical validity of these findings.

The work conducted herein focused on breast cancer patient samples, and certain aspects of this study should be repeated in other disease settings, but the preanalytical deviations that may be expected with rare cell detection are now understood. Previous publications have shown that the HD-SCA workflow is suitable for a broad repertoire of cancer types.^10,12,26,31 Recently, using a similar method as presented herein, Epic Sciences has released the only predictive and prognostic test for patients with metastatic castrate resistance prostate cancer using the androgen receptor splice variant 7 as a biomarker. In the androgen receptor splice variant 7 scenario, the detection of even just one CTC with nuclear androgen receptor splice variant 7 translates to failure of AR-targeted therapy and sensitivity to the taxanes in metastatic castrate resistance prostate cancer.6, 7, 8 The clinical utility of this signal relies on the initial identification of the candidate cell by morphologic characterization, followed by determination of nuclear androgen receptor splice variant 7 expression. Because of the analogous technology used, this supports the data, justifying further study into the clinical utility of the HD-SCA workflow for the identification of a biomarker in breast cancer patients.

We conclude that employment of the Streck BCT up to 48 hours after sample collection is optimal for the detection of CTCs and high-quality sequencing data in routinely collected peripheral blood samples using the HD-SCA workflow. This study provided evidence of the reliability and reproducibility of the workflow in preserving cellular morphology and molecular integrity of the liquid biopsy, while highlighting its significant but largely untapped potential for clinical utility. Determining the potential diagnostic and prognostic relevance of the liquid biopsy goes beyond rare cell enumeration; we hope to identify patients with specific morphologic or genomic signatures that may prove to be clinically relevant. In moving toward clinical validation, we continue to measure the presence of CTCs with additional downstream genomic characterization of the complete liquid biopsy.

Supplemental Data

Supplemental Figure S1.

Enumeration of HD-CTCs in breast cancer at 24-hour TTA. A: Distribution of HD-CTCs in nonmetastatic and metastatic breast cancer patients. B: CTC counts in nonmetastatic breast cancer patients before and after primary therapy. C: Identification of HD-CTCs by blood draw in metastatic breast cancer patients. D–F: Distribution of CTC small (D), CTC low CK (E), and CTC cfDNA producing candidate (F) cells in breast cancer patients. The y axis is a logarithmic scale. Outliers are noted with an adjacent red asterisk. Data were collected from 24-hour TTA analysis. Each data point represents one peripheral blood sample.

Supplemental Figure S2.

Quality assurance and quality control analysis of blood samples processed for the detection of HD-CTCs using the HD-SCA workflow. A: Temperature maintenance tests of 42 (13%) of the shippers was also completed using XpressPDF temperature labels (PakSense). The thermal insulation of the shippers maintained the temperature within the acceptable range, which is ±10°C from original temperature at time of box closure. B: Levey-Jennings plots for autostainer runs. Results of enumeration of positive control slides generated by spiking MDA-MB-231 (top panel), MCF-7 (middle panel), or SKBR3 (bottom panel) cells in healthy blood donor samples. Staining dates are plotted along the x axis. The mean (red line) and two SD limits (black dashed lines) are marked on the y axis.

Supplemental Figure S3.

CTC counts by CellSearch and HD-SCA in study subjects. A: Comparison of CTC enumeration for each method in matched samples where both assays detected CTCs (zero values are not present). P = 0.0028 by Wilcoxon matched-pairs signed rank test. B: Classification of patients according to the numbers of CTCs detected by CellSearch and HD-SCA. For both assays, CTC positivity was defined as >5 CTCs/7.5 mL (>0.667 CTCs/mL). C: Receiver-operating characteristic curve indicating ability of the HD-SCA workflow to predict a positive outcome from the CellSearch system. Area under curve was 0.9571 (95% CI, 0.8797–1).

Supplemental Figure S4.

Morphometric parameters of white blood cells (A) and HD-CTCs (B) detected from the liquid biopsy of patient samples collected in either a CellSave or a Streck tube and processed within 24 hours. Median and interquartile range are indicated in red. Wilcoxon matched-pairs signed rank test was used to determine P value. ****P < 0.0001.

Supplemental Figure S5.

Cell enumeration in a set of 24 draws found CTC positive at 24 hours and CTC negative at 48 hours TTA. CTC candidate cells identified at 24 hours (A) and 48 hours (B) included HD-CTCs and marginal CTC populations (CTC small, CTC low CK, and CTC cfDNA producing). Median and interquartile range are indicated in red. Wilcoxon matched-pairs signed rank test was used to compare TTA-matched results. P values for 24-hour compared with 48-hour TTA for this subset are provided for each category below the figures, from left to right: candidate cells’ P = 0.0001, HD-CTC P < 0.0001, CTC small P = 0.0103, CTC low CK P = 0.0018, and CTC cfDNA P = 0.0403.

Supplemental Figure S6.

TTA comparison of time to single-cell isolation after thawing. A–C: Analysis of the time between thawing and immunofluorescence staining of each slide to single-cell isolation for genomic analysis from 24-/48-hour matched samples. The U-test was performed. Significance defined by P < 0.05. D and E: Correlation between the time between thawing and immunofluorescence staining of each slide to single-cell isolation and either percentage coverage (D) or variant count from single-nucleotide variation analysis (E). Spearman correlation: r = −0.2467 and r = −0.4916 (P = 0.0057 and P < 0.0001, respectively). ****P < 0.0001.