Abstract
Circulating tumor cells (CTCs) can reliably be identified in cancer patients and are associated with clinical outcome. Next‐generation “liquid biopsy” technologies will expand CTC diagnostic investigation to include phenotypic characterization and single‐cell molecular analysis. We describe here a rare cell analysis platform designed to comprehensively collect and identify CTCs, enable multi‐parameter assessment of individual CTCs, and retrieve single cells for molecular analysis. The platform has the following four integrated components: 1) density‐based separation of the CTC‐containing blood fraction and sample deposition onto microscope slides; 2) automated multiparameter fluorescence staining; 3) image scanning, analysis, and review; and 4) mechanical CTC retrieval. The open platform utilizes six fluorescence channels, of which four channels are used to identify CTC and two channels are available for investigational biomarkers; a prototype assay that allows three investigational biomarker channels has been developed. Single‐cell retrieval from fixed slides is compatible with whole genome amplification methods for genomic analysis. © 2018 The Authors. Cytometry Part A published by Wiley Periodicals, Inc. on behalf of International Society for Advancement of Cytometry.
Keywords: liquid biopsy, circulating tumor cells, cell separation, immunofluorescence microscopy, molecular imaging, image processing, computer assisted, single‐cell analysis, DNA amplification techniques
“Liquid Biopsy”—Requiring More from Less
Over the past century, the medical approach to accessing tumors for analysis has evolved from highly invasive to increasingly less invasive: from surgical resection to excisional biopsy, to core needle biopsy, and to fine needle aspirate. Yet as biopsy sample volume has decreased, the diagnostic information required from the sample has increased. The least invasive access to tumor cells is by “liquid biopsy,” a sampling method to collect rare circulating tumor cells (CTCs) in the blood 1, 2. The promise of liquid biopsy is that real‐time blood sampling can provide the depth of data necessary for 21st century clinical investigation and patient care. The current challenge is to extract the required biological information from these few rare cells.
The CellSearch® system established that CTCs can be reliably identified in the blood of cancer patients and that the number of CTCs is strongly associated with clinical outcome 3, 4, 5, 6, 7, 8, 9, 10. Despite its groundbreaking technology, CellSearch has historically been limited by the immuno‐magnetic method it employs to collect and identify EpCAM‐expressing CTCs and by the lack of an integrated single‐cell isolation capability.
Traditional tissue biopsy samples are processed using formalin fixation and paraffin embedding. Sections are cut from the paraffin block and placed onto a microscope slide for analysis. The tissue slide provides information at the following three levels: 1) identification of cancer; 2) phenotypic characterization of cancer; and 3) molecular analysis of cancer. The future of cell‐based liquid biopsy will involve expanding the breadth and depth of CTC‐derived information from the first level to the second and third levels. The rare cell analysis platform presented here provides such information.
Working Principles
The RareCyte platform comprises buffy coat expansion and spreading onto slides, automated multiparameter fluorescence staining, automated microscopic imaging and analysis, and integrated single‐cell retrieval. The working principle that enables the comprehensive collection of CTCs is the differential density of the nucleated cell buffy coat layer from red blood cells (RBCs) and plasma. The method is based on buffy coat “expansion,” introduced by Levine and Wardlaw as the quantitative buffy coat (QBC) system 11. In the QBC, a float with defined density is placed within a capillary tube, which is filled with a blood sample. After centrifugation of the tube, the buffy coat is “expanded” in the space surrounding the float; the number of platelets and leukocytes can be calculated by determining the linear space occupied after imaging.
The same “expansion” concept can be scaled for larger blood volumes. The platform uses this principle to separate nucleated cells by density and, then, transfer them to microscope slides. The four components of the system are described below; associated methods have been published in detail 12, 13, 14, 15.
Sample Preparation System
The sample preparation system enables comprehensive and reproducible collection and transfer of nucleated cells from blood to microscope slides (Fig. 1A–D). Since CTCs are nucleated cells, their density falls within the range of white blood cells (WBCs), which are denser than plasma, but less dense than RBCs. The system collects cells by density, without regard to size or protein expression. This overcomes limitations of filters and microfluidic devices that will fail to collect CTCs that are not larger than WBCs, and surface immuno‐capture methods that rely on proteins that may not be expressed on all CTCs.
Figure 1.
RareCyte platform CTC sample preparation, staining, and analysis workflow. (A) Blood is collected into the AccuCyte BCT that preserves samples for processing up to 72 h. (B) Diagram of steps in density‐based collection of nucleated cells and transfer to microscope slide. (C) Photograph of AccuCyte separation tube after initial centrifugation, demonstrating fractionation of blood into plasma (top, yellow), red blood cell layer (bottom, red), and nucleated cell layer containing CTCs (gray‐white band around black float). (D) Diagram of subsequent CTC analysis workflow: automated staining of slides, automated scanning image acquisition and machine learning image analysis by the CyteFinder system, candidate CTC review and confirmation, and report generation. (Note: retrieval of individual CTCs by the CytePicker module is not shown.)
Whole blood is collected in the RareCyte blood collection tube (BCT) and is stable for up to 72 h (Fig. 1A). Blood is placed in a separation tube containing an internal float having density similar to nucleated blood cells. After centrifugation, the buffy coat is “expanded” within the space between the tube and the float, above the RBC layer, and below the plasma. An external sealing ring is applied that clamps the tube against the float to create a physical barrier between the RBC and buffy coat layers. The plasma is collected and may be used for cell‐free DNA analysis. A displacement fluid (~1.13 g/cm3) is added above the float, and a collector is inserted that connects the contents of the separation tube to an isolation tube containing a slightly less dense fluid (~1.10 g/cm3) that further separates the blood components. The collector allows movement of fluid and cells between the separation tube and the isolation tube. During a second centrifugation, the desired nucleated cell layer, which is less dense than the isolation fluid, is buoyantly displaced into the isolation tube, to collect the nucleated cells in a small volume (~120 μl). A transfer fluid containing a nonformalin fixative is added, and the resulting mixture is spread onto eight microscope slides using a device that draws the fluid into a monolayer smear by surface tension between its spreading blade and the slide (Fig. 1B).
The system is designed to collect platelets along with nucleated cells while removing 99.5–100% of RBCs. The absence of RBC lysis or wash steps minimizes cell loss. The processing time from blood to slide is about 75 min. Prepared slides are air dried for 30 minutes and can be stored frozen for 1 year or longer before staining. Components of the system are shown in Supporting Information Figure S1.
Automated Fluorescence Staining of Microscope Slides
The prepared slide contains CTCs, if present, in a background of WBCs. Once on the slide, CTCs may be identified, characterized by phenotype, and analyzed at a molecular level. The RareCyte platform workflow uses automated staining instruments (made by Ventana, Dako, and Leica) to stain slides by immunofluorescence. This increases laboratory throughput, eliminates variability, and increases control of the staining conditions.
The four canonical markers used for the identification of epithelial CTCs are a nuclear dye, the WBC exclusion marker CD45, and the epithelial CTC markers EpCAM and cytokeratin (CK). Since the imaging system can analyze six fluorescence channels, two or three additional investigational biomarkers may be simultaneously assessed on the identified CTCs, depending on whether the epithelial markers are in separate channels or combined into one (see Fig. 3).
Figure 3.
Six‐parameter breast cancer CTC assay incorporating investigational biomarkers ER, Her2, and Ki‐67. Breast cancer cell lines representing various ER and Her2 phenotypes were spiked into normal donor blood and processed to slides using the AccuCyte system. Model CTCs were identified by positive staining for epithelial markers against CK and/or EpCAM and negative staining for CD45. The ER/Her2 staining pattern of the cell lines matched reported phenotypes. Ki‐67‐positive cells are shown.
Imaging System
Stained slides are placed into the platform's automated fluorescence scanning microscope and scanned at 10× objective magnification. The microscope stage employs a kinematic mount that establishes highly reproducible positioning of the slide; X–Y displacement upon reloading is approximately 2–3 μm. Scanning of each slide in four channels takes about 12 min inclusive of image plane determination (Supporting Information Fig. S3); scanning all eight slides from a 7.5‐ml sample thus takes ~1.5 h. To increase imaging throughput, an automated slide loader for up to 80 slides is being developed (Supporting Information Fig. S2).
Slide images are analyzed with integrated image analysis software (Fig. 1D and Supporting Information Fig. S2) that automatically analyzes the images to find and score CTC candidates in about 10 min as follows. Objects with signal in CK or EpCAM are identified. These are then screened by the software for CD45 signal, which, if present, causes the object to be rejected. This excludes virtually all WBCs and results in a set of CTC candidate objects. Candidate objects are ranked by a machine‐learning model based on a large set of image features that determines a probability score (0–100) for being a “true” CTC. The model is trained on a data set of >10,000 CTCs from various cancer types that have been observer confirmed to have a “true” epithelial CTC phenotype. Candidate object images are presented to the reviewer in order of their score, from high to low, ending at a threshold cutoff score of 15 that was selected to retain approximately 99% of “true” CTCs while excluding approximately 90% of “false” candidate objects in the data set. Probability score ranking increases the concordance between reviewers and decreases review time: low‐scoring objects may be screened quickly, since CTCs are rarely present at end of the list. A typical slide generates ~300 candidate cell objects that are displayed for reviewer confirmation; per‐slide review time is 3–5 min.
Final classification of a cell as a CTC requires reviewer confirmation of nuclear fluorescence signal, signal for either CK or EpCAM or both, and absence of CD45 signal. Single‐cell molecular studies have identified mutations that demonstrate the malignant origin of CTCs so classified (see below). Nuclear morphology may be used to discriminate between CTCs and WBCs if dim CD45 signal is observed on a cell with epithelial marker staining.
Single Cell Retrieval System
The imaging instrument incorporates a fluid‐coupled picking system above the slide stage for retrieval of individual CTCs (Fig. 1E). Its needle tip mechanically dislodges the cell from the slide surface and deposits it into an imaging tube for visual confirmation. The system is highly automated and is designed so that it does not require high technical skill; the rate of successful cell retrieval is 80–90%. Because slides may be stored and archived after staining, single CTC can be retrieved for molecular analysis retrospectively. Retrieved cells are suitable for whole genome amplification by methods that are compatible with fixed samples. (A method for preparation and single‐cell analysis of live cells using chamber slides has been developed but is not discussed here.) Prepared DNA may be used for various types of analysis including PCR, mass spectrometry, comparative genomic hybridization, and next‐generation sequencing 12, 15, 17. Specific gene sequencing of identified CTCs has confirmed mutations present in tumor tissue samples. Whole exome sequencing has been used to investigate genomic evolution of CTCs during the course of therapy (Supporting Information Figs. S5 and S6) 18.
System Performance
Blinded studies using precisely counted spike‐in samples of high‐ and low‐EpCAM expressing cell lines yielded highly linear (R 2 = 0.98) recovery of spiked‐in cells with mean greater than 90%. Single‐digit spike‐in recovery was greater than 80%, and solitary spiked‐in cells could be detected 12. In blinded comparison against CellSearch, equivalent recoveries were observed in lines that express high EpCAM high levels (MCF7 and LNCaP). However, in low/variable EpCAM‐expressing cell lines (PC3 and A549) RareCyte counts were approximately two‐ to threefold higher than CellSearch counts.
In similar blinded comparison testing of paired breast (n = 24), prostate (n = 17), and lung (n = 12) cancer samples, 55% had equivalent counts with both platforms; 28% had higher RareCyte counts than CellSearch counts by more than 50% (mean approximately twofold, maximum approximately threefold); and 17% had CellSearch counts less than 5 and RareCyte counts 5 or more 19. There was strong correlation between the RareCyte and CellSearch counts in breast (R 2 = 0.94) and prostate (R 2 = 0.83) cancers. However, there was no correlation in lung cancer (R 2 = 0.03), where samples often had no CTCs found by CellSearch; this is consistent with recognized absence of EpCAM in many lung cancers. When the epithelial phenotype of confirmed clinical CTCs was investigated, the majority of CTCs identified with the RareCyte platform were both CK positive and EpCAM positive, ~25% of cells were CK positive and EpCAM negative, and ~20% of cells were CK negative and EpCAM positive (unpublished). In a triple‐negative breast cancer case study, intensive longitudinal monitoring of CTC count was shown to be a sensitive and dynamic correlate of therapeutic response to anticancer therapies, supporting the definition of CTCs employed with the platform (Supporting Information Fig. S5) 20.
Concordance between CTC counts in paired patient samples was assessed in a blinded study (Fig. 2). Linear regression demonstrated very high correlation between sample pairs (R 2 = 0.99). The 1000× random permutation resampling of paired sample order yielded a set of tightly distributed slopes (mean 0.98) and intercepts (mean 0.11) supporting the conclusion that test results for the same patient are highly correlated across the sample set (Supporting Information Fig. S4).
Figure 2.
Concordance of CTC counts in paired blood samples. Paired blood samples were collected from 21 patients with advanced cancers (prostate, breast, lung, and Merkel cell). The samples were processed and analyzed using the RareCyte platform laboratory by operators blinded to the sample identity and CTC count result. Samples were randomly assigned as “Tube A” and “Tube B.” (A) Individual paired sample data demonstrate minimal variability between paired samples. (B) Linear regression analysis yielded extremely high correlation between the Tube A and Tube B samples sets. Random permutation resampling analysis yielded additional evidence that test results are highly correlated across the sample set (Supporting Information Fig. S4).
False‐positive CTCs are rarely observed. The platform infrequently finds a single cell that fits the classification of a CTC in blood from healthy volunteers. Assessment of CTCs in persons with noncancer diseases has not been performed. Clinical outcome studies to demonstrate the prognostic significance of RareCyte CTC counts are planned but have not been completed. However, sequencing studies of CTCs have provided molecular evidence that cells identified as CTCs by the RareCyte platform are in fact malignant. For example, in a sample from a patient with prostate cancer with known mutation in the TP53 gene, EpCAM‐positive and CK‐negative CTCs were retrieved for gene‐specific PCR and Sanger sequencing; in five of the eight cells, the mutation was identified (unpublished).
Multi‐Parameter Phenotypic Characterization CTCs
The diagnostic value of multiparameter investigation of CTCs is enhanced by the noninvasive nature of blood sampling, which allows access to tumors in patients at risk for procedural complications of biopsy and enables monitoring for dynamic assessment of cancer phenotype. Phenotypic panels can include markers that may be rationally used to inform treatment approach. Such “actionable” information could be provided by assays for proliferation, mesenchymal transition, resistance, and lineage differentiation.
In breast cancer, the primary approaches to therapy remain directed at the estrogen and Her2 pathways 16. Figure 2 shows images generated using a novel six‐parameter CTC assay that assesses the expression of estrogen receptor (ER) and Her2 as well as the proliferation marker Ki‐67 on breast cancer cell line model CTCs. We have also demonstrated the assessment of potentially actionable markers on prostate CTCs, including androgen receptor (AR) and its splice variant ARv7, and prostate‐specific membrane antigen (PSMA). A multiparameter assay developed for the investigation of immune checkpoint inhibitor therapeutic response includes both the well‐established marker PD‐L1 and the interferon‐gamma induced transcription factor interferon regulatory factor 1 (IRF1). We are deploying multiparameter phenotypic analysis in companion diagnostics development to identify the presence of surface protein drug targets on identified CTCs. This has the potential for immediate clinical utility and is not achievable by plasma nucleic acid analysis.
Application Beyond CTCs
Using similar methods of sample preparation, multiparameter staining, imaging and analysis, and cell retrieval, the platform has been applied to other rare cell investigations. Areas include fetal cell analysis for noninvasive prenatal genetic testing 17, minimal residual disease monitoring of liquid tumors, identification of rare immune cell subsets in blood, and pathological analysis of fine needle aspirate, other cytology samples, and tissue sections.
Supporting information
Figure S1. Components of AccuCyte® sample preparation system. (A) separation tube, (B) sealing ring, (C) collector clip, (D) collector, (E) float, (F) separation tube cap, (G) isolation tube, (H) excess fluid shield tube, (I) CyteSpreader base assembled with slide (J) and blade (K). Assembled separation tube/collector is shown in (L). Within the collector is a needle that pierces the rubber septum base of the isolation tube in order to allow movement of fluids and cells between the separation tube and the isolation tube.
Figure S2. CyteFinder® imaging system. (A) The CyteFinder system is an automated digital microscope with software that performs high speed imaging of slides in six spectrally discrete fluorescent channels. The system performs whole slide scanning in 12 minutes (4‐channels, 10x magnification, 4.5 cm2 scan area). It has a two‐slide stage, positional repeatability of less than 5 μm, a high resolution 2560 x 2160‐pixel sCMOS camera for quantitative imaging, and an automated five‐position objective turret with standard 10x and 40x objectives. (B) CyteFinder® HT is a high throughput CyteFinder system with 80‐slide capacity and is expected to be available at the end of 2018. Slides are loaded in cassettes holding 20 slides each (inset).
Figure S3. CTC identification. Image files generated by CyteFinder are automatically analyzed to identify and rank CTC candidates, and then reviewed by the pathologist (or trained reviewer) for CTC confirmation using CyteMapper® software. The machine learning algorithm scores candidate CTCs, reducing user review time and increasing consistency between reviewers. The algorithm evaluates more than 500 image features that are extracted from each candidate target cell. After completed review, a report is generated that includes images of all confirmed CTCs. In a set of more than 100 patient samples, the total initial object count per slide was ~3000; after application of a score threshold of 15 the number of candidate objects for review was reduced to ~350. (A) CyteMap of stained and scanned slide smear. Rectangular outlines (white) define panels within the CyteMap that contained reviewer‐confirmed CTCs. (B) Collage view of confirmed CTCs. Note, some CTCs are present in clusters.
Figure S4. Random permutation resampling linear regression analysis. The classification of a sample as “Tube A” or “Tube B” in paired samples in Figure 2 are arbitrary, as there is no known difference in experimental condition. Therefore, we performed 1000x resampling with random permutation of the Tube A and Tube B labels for each sample. This yielded a distribution of linear regression slopes and intercepts. The mean slope was 0.9765 (nearly 1) and the mean intercept was 0.1074 (nearly 0). The slope was tightly distributed around the mean, providing additional evidence that test results for the same patient are highly correlated across the sample set. The minimal variability present may be in part due to sampling differences in rare cell number.
Figure S5. Serial monitoring and whole exome sequence analysis of individual CTCs and correlation with therapeutic response in patient with triple‐negative breast cancer. (A) CTC counts over the course of treatment with anti‐cancer drugs shown. Decreasing counts correlate with response to treatment and increasing counts with progression of disease. (B) Cancer driver or predicted deleterious sequence variants identified after whole exome sequencing of individual CTCs at six time points during study course (N = 28; per time point range = 3–6). Bioinformatics filters are described in Figure S6. Cancer driver/deleterious variant number increased after response to treatment and decreased with progression. CTCs with higher numbers of cancer driver variants may represent cells that are resistant to the current therapy. Panel (A) has been previously published (13). The study was performed in accordance with University of Washington institutional review board policies, and according to the Declaration of Helsinki of 1975, as revised in 2008.
Figure S6. Individual CTC sequencing bioinformatics analysis pipeline. A multi‐level approach was used to identify genetic variants in individual CTCs from blood collected at six time points from a patient with triple‐negative breast cancer. To eliminate false positive results due to introduction of sequence errors by the whole genome amplification or sequence library preparation process, a variant was required to be present in at least two CTCs, or in a plasma DNA sample from one of the time points. A confidence filter was then applied to the variant call. Common variants were removed using information from the Whole Genomes Project, as were germline variants found in single white blood cells sequenced from the patient. Predicted deleterious variants were identified and tallied by use of publicly available software tools. Finally, genes reported to be cancer drivers in the scientific literature were identified and tallied.
Literature Cited
- 1. Alix‐Panabières C, Pantel K. Circulating tumor cells: Liquid biopsy of cancer. Clin Chem 2013;59(1):110–118. [DOI] [PubMed] [Google Scholar]
- 2. Mader S, Pantel K. Liquid biopsy: Current status and future perspectives. Oncol Res Treat 2017;40(7–8):404–408. [DOI] [PubMed] [Google Scholar]
- 3. Cristofanilli M, Budd GT, Ellis MJ, Stopeck A, Matera J, Miller MC, Reuben JM, Doyle GV, Allard WJ, Terstappen LWMM, et al. Circulating tumor cells, disease progression, and survival in metastatic breast cancer. N Engl J Med 2004;351(8):781–791. [DOI] [PubMed] [Google Scholar]
- 4. Allard WJ, Matera J, Miller MC, Repollet M, Connelly MC, Rao C, Tibbe AG, Uhr JW, Terstappen LW. Tumor cells circulate in the peripheral blood of all major carcinomas but not in healthy subjects or patients with nonmalignant diseases. Clin Cancer Res 2004;10(20):6897–6904. [DOI] [PubMed] [Google Scholar]
- 5. Riethdorf S, Fritsche H, Muller V, Rau T, Schindlbeck C, Rack B, Janni W, Coith C, Beck K, Janicke F, et al. Detection of circulating tumor cells in peripheral blood of patients with metastatic breast cancer: A validation study of the Cell Search system. Clin Cancer Res 2007;13(3):920–928. [DOI] [PubMed] [Google Scholar]
- 6. de Bono JS, Scher HI, Montgomery RB, Parker C, Miller MC, Tissing H, Doyle GV, Terstappen LWWM, Pienta KJ, Raghavan D. Circulating tumor cells predict survival benefit from treatment in metastatic castration‐resistant prostate cancer. Clin Cancer Res 2008;14(19):6302–6309. [DOI] [PubMed] [Google Scholar]
- 7. Riethdorf S, O'Flaherty L, Hille C, Pantel K. Clinical applications of the CellSearch platform in cancer patients. Adv Drug Deliv Rev 2018;125:101–121. pii:S0169‐409X(18)30011‐5. [DOI] [PubMed] [Google Scholar]
- 8. Scher HI, Heller G, Molina A, Attard G, Danila DC, Jia X, Peng W, Sandhu SK, Olmos D, Riisnaes R, et al. Circulating tumor cell biomarker panel as an individual‐level surrogate for survival in metastatic castration‐resistant prostate cancer. J Clin Oncol 2015;33(12):1348–1355. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Lorente D, Olmos D, Mateo J, Bianchini D, Seed G, Fleisher M, Danila DC, Flohr P, Crespo M, Figueiredo I, et al. Decline in circulating tumor cell count and treatment outcome in advanced prostate cancer. Eur Urol 2016;70(6):985–992. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Heller G, Fizazi K, McCormack R, Molina A, MacLean D, Webb IJ, Saad F, de Bono JS, Scher HI. The added value of circulating tumor cell enumeration to standard markers in assessing prognosis in a metastatic castration‐resistant prostate cancer population. Clin Cancer Res 2017;23(8):1967–1973. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Wardlaw SC, Levine RA. Quantitative buffy coat analysis. A new laboratory tool functioning as a screening complete blood cell count. JAMA 1983;249(5):617–620. [DOI] [PubMed] [Google Scholar]
- 12. Campton DE, Ramirez AB, Nordberg JJ, Drovetto N, Clein AC, Varshavskaya P, Friemel BH, Quarre S, Breman A, Dorschner M, et al. High‐recovery visual identification and single‐cell retrieval of circulating tumor cells for genomic analysis using a dual‐technology platform integrated with automated immunofluorescence staining. BMC Cancer 2015;15:360. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Werbin JL, Nordberg JJ, Tzucker J, Varshavskaya P, Stilwell JL, Kaldjian EP. RareCyte® CTC analysis step 2: Detection of circulating tumor cells by CyteFinder® automated scanning and semiautomated image analysis. Methods Mol Biol 1634;2017:173–180. [DOI] [PubMed] [Google Scholar]
- 14. Ramirez AB, U'Ren L, Campton DE, Stewart D, Nordberg JJ, Stilwell JL, Kaldjian EP. RareCyte® CTC analysis step 1: AccuCyte® sample preparation for the comprehensive recovery of nucleated cells from whole blood. Methods Mol Biol 1634;2017:163–172. [DOI] [PubMed] [Google Scholar]
- 15. Stilwell JL, Varshavskaya P, Werbin JL, Nordberg JJ, Ramirez AB, Quarre S, Tzucker J, Chow J, Enright B, Kaldjian EP. RareCyte® CTC analysis step 3: Using the CytePicker® module for individual cell retrieval and subsequent whole genome amplification of circulating tumor cells for genomic analysis. Methods Mol Biol 1634;2017:181–192. [DOI] [PubMed] [Google Scholar]
- 16. Moffat FL. Chapter 28: Clinical and pathologic prognostic and predictive factors In: Harris JR, Lippman ME, Morrow M, Osborne CK, editors. Diseases of the Breast. 5th ed. Philadelphia, PA: Lippincott Williams & Wilkins, 2014. [Google Scholar]
- 17. Breman AM, Chow JC, U'Ren L, Normand EA, Qdaisat S, Zhao L, Henke DM, Chen R, Shaw CA, Jackson L, et al. Evidence for feasibility of fetal trophoblastic cell‐based noninvasive prenatal testing. Prenat Diagn 2016;36(11):1009–1019. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Howard Kellie, Austin Sharon, Lo Fang Yin, Ramirez Arturo, Boles Debbie, Pruitt John, E Mahen, H Collins, A Leonti, L Maasel, et al. Meta‐analysis of genomic aberrations identified in CTCs and ctDNA in triple negative breast cancer. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research, April 16–20, 2016. New Orleans, LA/Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 498.
- 19. Stilwell Jackie L., Drovetto Nick, Ramirez Arturo B., Campton Daniel, Nordberg Joshua, P Varshavskaya, A Clein, S Quarre, B Freimel, D Sabath, et al. Clinical performance of the AccuCyte® ‐ CyteFinder® system, a dual‐technology platform for comprehensive collection and high resolution imaging of circulating tumor cells. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research, April 18–22, 2015; Philadelphia, PA: AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 1601.
- 20. Blau CA, Ramirez AB, Blau S, Pritchard CC, Dorschner MO, Schmechel SC, Martins TJ, Mahen EM, Burton KA, Komashko VM, et al. A distributed network for intensive longitudinal monitoring in metastatic triple‐negative breast cancer. J Natl Compr Canc Netw 2016;14(1):8–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1. Components of AccuCyte® sample preparation system. (A) separation tube, (B) sealing ring, (C) collector clip, (D) collector, (E) float, (F) separation tube cap, (G) isolation tube, (H) excess fluid shield tube, (I) CyteSpreader base assembled with slide (J) and blade (K). Assembled separation tube/collector is shown in (L). Within the collector is a needle that pierces the rubber septum base of the isolation tube in order to allow movement of fluids and cells between the separation tube and the isolation tube.
Figure S2. CyteFinder® imaging system. (A) The CyteFinder system is an automated digital microscope with software that performs high speed imaging of slides in six spectrally discrete fluorescent channels. The system performs whole slide scanning in 12 minutes (4‐channels, 10x magnification, 4.5 cm2 scan area). It has a two‐slide stage, positional repeatability of less than 5 μm, a high resolution 2560 x 2160‐pixel sCMOS camera for quantitative imaging, and an automated five‐position objective turret with standard 10x and 40x objectives. (B) CyteFinder® HT is a high throughput CyteFinder system with 80‐slide capacity and is expected to be available at the end of 2018. Slides are loaded in cassettes holding 20 slides each (inset).
Figure S3. CTC identification. Image files generated by CyteFinder are automatically analyzed to identify and rank CTC candidates, and then reviewed by the pathologist (or trained reviewer) for CTC confirmation using CyteMapper® software. The machine learning algorithm scores candidate CTCs, reducing user review time and increasing consistency between reviewers. The algorithm evaluates more than 500 image features that are extracted from each candidate target cell. After completed review, a report is generated that includes images of all confirmed CTCs. In a set of more than 100 patient samples, the total initial object count per slide was ~3000; after application of a score threshold of 15 the number of candidate objects for review was reduced to ~350. (A) CyteMap of stained and scanned slide smear. Rectangular outlines (white) define panels within the CyteMap that contained reviewer‐confirmed CTCs. (B) Collage view of confirmed CTCs. Note, some CTCs are present in clusters.
Figure S4. Random permutation resampling linear regression analysis. The classification of a sample as “Tube A” or “Tube B” in paired samples in Figure 2 are arbitrary, as there is no known difference in experimental condition. Therefore, we performed 1000x resampling with random permutation of the Tube A and Tube B labels for each sample. This yielded a distribution of linear regression slopes and intercepts. The mean slope was 0.9765 (nearly 1) and the mean intercept was 0.1074 (nearly 0). The slope was tightly distributed around the mean, providing additional evidence that test results for the same patient are highly correlated across the sample set. The minimal variability present may be in part due to sampling differences in rare cell number.
Figure S5. Serial monitoring and whole exome sequence analysis of individual CTCs and correlation with therapeutic response in patient with triple‐negative breast cancer. (A) CTC counts over the course of treatment with anti‐cancer drugs shown. Decreasing counts correlate with response to treatment and increasing counts with progression of disease. (B) Cancer driver or predicted deleterious sequence variants identified after whole exome sequencing of individual CTCs at six time points during study course (N = 28; per time point range = 3–6). Bioinformatics filters are described in Figure S6. Cancer driver/deleterious variant number increased after response to treatment and decreased with progression. CTCs with higher numbers of cancer driver variants may represent cells that are resistant to the current therapy. Panel (A) has been previously published (13). The study was performed in accordance with University of Washington institutional review board policies, and according to the Declaration of Helsinki of 1975, as revised in 2008.
Figure S6. Individual CTC sequencing bioinformatics analysis pipeline. A multi‐level approach was used to identify genetic variants in individual CTCs from blood collected at six time points from a patient with triple‐negative breast cancer. To eliminate false positive results due to introduction of sequence errors by the whole genome amplification or sequence library preparation process, a variant was required to be present in at least two CTCs, or in a plasma DNA sample from one of the time points. A confidence filter was then applied to the variant call. Common variants were removed using information from the Whole Genomes Project, as were germline variants found in single white blood cells sequenced from the patient. Predicted deleterious variants were identified and tallied by use of publicly available software tools. Finally, genes reported to be cancer drivers in the scientific literature were identified and tallied.