Synchronous Detection of Circulating Tumor Cells in Blood and Disseminated Tumor Cells in Bone Marrow Predicts Adverse Outcome in Early Breast Cancer

Recurrence of breast cancer after initial treatment with surgery and adjuvant therapies remains the major cause of mortality from this disease (1). Mechanisms involved in the persistence of breast cancer cells and their spread to distant sites are not fully understood (2). Accumulated evidence demonstrates that the presence of cancer cells in hematopoietic compartments (blood and bone marrow) is associated with poor clinical outcome (3–6). Methods for reliable detection of these tumor cells have been actively pursued in the last decade (7, 8).

Cancer cells in blood and bone marrow, referred to as circulating tumor cells (CTCs) and disseminated tumor cells (DTCs), respectively, can be detected using immunocytochemical and nucleic acid–based assays (e.g., RT-PCR; refs. 9–11). Pooled analysis of data from several studies have now shown that the presence of DTCs is a strong predictor of poor outcomes (6). Despite demonstrated clinical significance, testing for DTCs has not yet become a standard component of disease staging. Lack of a standard DTC methodology has been one of the issues hampering adoption (12).

The CellSearch (Veridex LLC) system is currently the only FDA-cleared system for detection of EPCAM-positive CTCs (13). Studies using CellSearch have demonstrated that enumeration (counting) of CTCs can provide prognostic information in patients with early (5, 14, 15) and advanced (13, 16, 17) breast cancer.

We have described an EPCAM-based immunomagnetic enrichment/flow cytometry (IE/FC) for enumeration and isolation of CTCs (18–20) in blood, and have applied this method as well to DTCs (18, 21, 22) in bone marrow.

We hypothesize that CTCs, DTCs, and simultaneous detection of these cell at the time of surgery are associated with worse outcome. To address this hypothesis, we prospectively enumerated CTCs and DTCs from each patient immediately prior to breast cancer surgery. CTC enumeration was performed first by IE/FC and then by CellSearch; DTC enumeration was performed by IE/FC. With long patient follow-up (up to a median of 13.3 years), we analyzed the prognostic significance of CTC and DTC status in these patients. To our knowledge, our study is the first to report on synchronous detection of CTC and DTC in a large cohort using the same quantitative assay system.

Patients with early breast cancer who were scheduled to undergo breast cancer surgery at the University of California San Francisco (UCSF, San Francisco, CA) were recruited to participate in this study. The parent study included prospective collection of samples from both patients with treatment-naïve and neoadjuvant-treated breast cancer. In this article, we excluded the neoadjuvant cohort to rule out potential confounding effects of neoadjuvant therapy on the levels of CTCs and DTCs. The study was performed with Institutional Review Board (UCSF Committee on Human Research) approval, and informed written consent was obtained from each patient. The study was performed in accordance with the Declaration of Helsinki.

Bone marrow samples were collected via a unilateral bone marrow aspiration from the posterior superior iliac crest while patients were under anesthesia prior to surgery. Two 5 mL samples were withdrawn from one site in posterior iliac crest. Peripheral blood was obtained on the same day, either in the preoperative setting or at the same time as bone marrow aspiration. Bone marrow (∼4 mL) and peripheral blood (∼10 mL) samples were drawn into tubes containing ethylenediaminetetraacetic acid for IE/FC (information to follow). Samples were processed within 24 hours after collection.

Enumeration of CTCs (by IE/FC and CellSearch) and DTCs was performed by investigator J.H. Scott who was blinded to the study endpoints.

Blood and bone marrow samples were subjected to the IE/FC assay to enumerate CTCs and DTCs, respectively (18–20, 22). Briefly, two distinct mAbs against EPCAM, one conjugated to immunomagnetic beads (MJ37) and the other conjugated to phycoerythrin (EBA-1) were added to whole blood or bone marrow. The sample was then placed in a magnet to capture cells labeled with the magnetic bead–antibody conjugated. The supernatant containing cells that were unbound (including red blood cells) was aspirated. Magnetic separation was repeated twice to further enrich for EPCAM-expressing cells. A nucleic acid dye (Thioflavin-T, BD Biosciences) and a mAb to the leukocyte-specific marker CD45 (2D1) conjugated to peridinin-chlorophyll-protein-Cy5.5 were added to the sample. The enriched sample was transferred to a BD TruCount (BD Biosciences) tube, and flow cytometric analysis was performed using the BD FACSCalibur (BD Biosciences). CTCs and DTCs were defined as nucleated cells that are EPCAM positive and CD45 negative.

In 2004, the CellSearch system was cleared by the FDA for enumeration of CTCs. We amended our study protocol to utilize CellSearch for CTC enumeration in place of the IE/FC starting August 2005.

Peripheral blood was collected into CellSave preservative tubes (Menarini) for CellSearch analysis. CTCs were enumerated in 7.5 mL of blood using the Circulating Tumor Cell Kit (Menarini) following the manufacturer's instructions without modifications (23). Briefly, the sample was subjected to immunomagnetic enrichment using beads coated with mAb against EPCAM and then CTCs were detected by fluorescence microscopy. CTCs were defined as nucleated cells that are cytokeratin positive and CD45 negative. CellSearch results were expressed as CTC/mL for direct comparison with IE/FC.

Samples were prospectively collected between April 27, 1999 and June 19, 2012. Survival analysis was performed on follow-up data available as of December 30, 2017. The median follow-up times for distant recurrence-free survival (DRFS) and breast cancer-specific survival (BCSS)/overall survival (OS) for all patients in the study were 7.1, and 9.1 years, respectively. In subset analyses, median follow-up duration for BCSS/OS reached 13.3 years (Table 1).

The primary clinical endpoints for the survival analysis included: DRFS, BCSS, and OS. Survival was measured from the date of diagnosis to the corresponding event in question. Patients lost to follow-up were censored at the time of their last visit. Covariates examined in survival models include age at diagnosis, tumor size at surgery, stage, grade, hormone-receptor, HER2, and nodal status.

To determine thresholds for CTC and DTC positivity by IE/FC, cutoffs were initially based on mean CTC/mL and DTC/mL in controls (see above) plus two SDs. To find optimal thresholds for association with outcome, we performed cutoff optimization with Monte–Carlo cross validation. First, half of the cases (balanced for number of events) were subsampled and used to derive a threshold between the 20% and 80% percentile that yielded the maximum Kaplan–Meier curve separation (i.e., minimum log-rank P value) for DRFS. We chose DRFS because we expected this was most likely to reflect breast cancer–specific outcome, while giving us more power to detect outcome differences (i.e., more events) than BCSS within our follow-up period. The threshold was then applied to the remaining half of the cases. The log-rank P values were assessed in the test set. The above procedure was repeated 1,000 times. The log-rank P values for the test set over the 1,000 iterations were combined using the logit method and the threshold with the lowest combined P value was selected.

Univariate and multivariate Cox proportional hazards regression analysis was performed to calculate hazard ratios and 95% confidence intervals (CIs). Kaplan–Meier survival curves were plotted and P values were calculated using the log-rank test. The R package “survival” was used for Cox proportional hazards model, Kaplan–Meier survival analysis, Wald, and log-rank tests.

Of the 1,121 patients with early-stage breast cancer enrolled in the study, 742 were treatment naïve, that is, did not receive neoadjuvant therapy (Supplementary Fig. S1A). The 379 patients who did receive neoadjuvant therapy were excluded from this analysis.

All patients underwent blood and/or bone marrow sampling in the operating room immediately prior to breast surgery (Supplementary Fig. S2A). A total of 71% patients were subsequently found to be lymph node negative, and 87% were hormone receptor positive (Table 1).

The median follow-up times for DRFS and BCSS/OS for all patients in the study were 7.1, and 9.1 years, respectively (Table 1). In subset analyses, median follow-up duration for BCSS/OS reached 13.3 years.

The study initially used only IE/FC analysis of CTCs and DTCs. In 2004, the CellSearch system was granted clearance by the FDA for enumeration of CTCs in breast cancer. On the basis of this, the study protocol was amended to replace IE/FC with CellSearch for CTC enumeration (starting August 2005). This explains the shorter follow-up duration among patients whose CTCs were enumerated by CellSearch. The final analysis therefore includes two separate CTC detection strategies: an EPCAM-positive cytokeratin approach (CellSearch) and a dual epitope EPCAM-based approach (IE/FC).

CTCs in blood and DTCs in bone marrow were enumerated from samples collected immediately prior to surgery. The frequency distribution of CTC and DTC counts per mL are shown in Supplementary Fig. S2B. The mean concentration of tumor cells in the bone marrow was significantly higher than that in blood (23.31 DTCs/mL vs. CellSearch: 0.09 CTCs/mL and IE/FC: 1.01 CTCs/mL; t tests; P < 0.001; Supplementary Table S1). The range of DTCs (0–4743.20, median 6.73 DTC/mL) was similarly larger than CTC range by IE/FC assay or CellSearch (0–33.74, median 0.34 CTC/mL and 0–6.67, median 0 CTC/mL, respectively).

In addition to tumor cells per mL of bone marrow versus blood, we also compared the number of tumor cells per 10⁶ mononuclear cells (MNC) in blood (n = 73 by IE/FC) and bone marrow (n = 184) samples. Comparison of tumor cells/10⁶ MNC data between compartments confirmed a significantly higher tumor cell/MNC ratio in the bone marrow, which was nearly 5-fold higher than in blood (0.23/10⁶ MNCs in bone marrow vs. 0.05/10⁶ MNCs in blood, Wilcoxon rank-sum test; P < 0.001; Supplementary Fig. S3).

Patient samples analyzed by IE/FC were scored as positive for CTCs and DTCs using thresholds based on two SDs above the mean background levels in controls, that is, >0.54 CTCs/mL of blood and >4.16 DTCs/mL of bone marrow. Using these cut-off points, 38% and 68% of patients were considered positive for CTCs and DTCs, respectively (Supplementary Table S1). The percent CTC detection rate by CellSearch was 23% (cutoff ≥ 1 CTC per 7.5 mL of blood).

We performed Monte–Carlo cross-validation to find optimal cutoffs for prognostication in patients analyzed by IE/FC (see Materials and Methods). Threshold optimization yielded the following cutoffs: >0.44 cells/mL for CTCs and >18.61 cells/mL for DTCs. Using these thresholds, percent positivity for CTCs increased from 38% to 41%, while percent positivity for DTCs decreased from 68% to 19% (Supplementary Table S1). For CellSearch, we used the previously validated cutoffs of ≥1 CTC and ≥2 CTC per 7.5 mL of blood (5, 14, 24). Percent positivity decreased from 23% to 9% using the latter cutoff.

No significant association was observed between CTCs/DTCs and standard clinicopathologic variables using the initial thresholds. With the optimized cutoffs, we observed a significant association between CTC positivity (by CellSearch) and HER2 positivity (Fisher exact test; P = 0.011; Supplementary Table S2). We also found that patients who were positive for CTCs (by IE/FC) had numerically larger mean tumor size compared with those who were CTC negative (t test; P = 0.05).

Of the 742 patients in the study, 65 (9%) experienced a distant recurrence and 97 (13%) died during the study; 40 (6%) were breast cancer–specific deaths. (Table 1). We performed univariate Cox regression analysis to evaluate the prognostic significance of established clinicopathologic variables in our study population. As expected, tumor size, nodal status, grade, and pathologic stage were strong predictors for all survival endpoints (Supplementary Table S3). HER2 status was prognostic for DRFS and OS, while age at diagnosis and HR status were prognostic for OS.

We evaluated CTC and DTC levels as continuous variables versus outcome. CTCs by IE/FC was prognostic for BCSS (hazard ratio, 1.25; P = 0.0119), while DTCs were prognostic for BCSS (hazard ratio, 1.2; P < 0.0001) and OS (hazard ratio, 1.19; P < 0.0001; Supplementary Table S4).

Next, we used the initial cutoffs of >0.54 cells/mL for CTCs and >4.16 cells/mL for DTCs to dichotomize patients into positive and negative groups. Univariate Cox regression analysis revealed no significant correlation between DTCs and clinical outcomes. In contrast, patients positive for CTCs by IE/FC had significantly reduced DRFS (hazard ratio, 1.96; P = 0.0420), and BCSS (hazard ratio, 2.73; P = 0.0172; Supplementary Table S4).

To evaluate the prognostic impact of CTCs and DTCs using the optimized thresholds described above, we performed univariate Cox regression and Kaplan–Meier analyses. We also performed multivariate Cox regression analyses to adjust for age, tumor size, nodal status, HR/HER2 status, grade, and pathologic stage. The median follow-up times for each patient subset are shown in Table 1.

Patients positive for CTCs (>0.44 CTCs/mL) had significantly reduced DRFS (hazard ratio, 2.16; P = 0.0189), BCSS (hazard ratio, 3.63; P = 0.0021), and OS (hazard ratio, 1.79; P = 0.0235; Fig. 1A). In multivariate models, CTCs remained prognostic for BCSS (hazard ratio, 3.54; P = 0.0138) and OS (hazard ratio, 1.89; P = 0.0301; Table 2; Supplementary Table S5).

Using the cutoff of ≥1 CTC per 7.5 mL, no significant correlation between CTCs and any of the survival endpoints was observed (Supplementary Fig. S4A). However, when the cutoff of ≥2 CTCs per 7.5 mL was used, we observed significantly shorter DRFS in patients who were CTC positive compared with those who were CTC negative (hazard ratio, 3.12; P = 0.0108; Fig. 1B). Multivariate analyses confirmed the prognostic significance of CTCs in predicting DRFS (hazard ratio, 4.93; P = 0.0067; Table 2; Supplementary Table S5).

Survival analysis suggested a trend toward shorter DRFS in DTC-positive patients (log-rank P = 0.0599; Fig. 1C). Univariate Cox regression analysis also showed a trend toward reduced DRFS in DTC-positive patients compared with those who were DTC negative (hazard ratio, 1.77; P = 0.0634; Supplementary Table S4).

Next, we examined whether simultaneous detection of CTCs and DTCs predicted survival. Paired CTC (by CellSearch) and DTC data were available for 246 patients. Results of the survival analysis was inconclusive due to the relatively small size of the double-positive group [4 CTC+/DTC+ (2%) vs. 183 CTC−/DTC− (74%), 38 CTC−/DTC+ (15%), and 21 CTC+/DTC− (9%)].

Using the optimized cutoffs for IE/FC, we categorized the 273 patients with paired DTC and CTC data (by IE/FC) into four groups: 136 CTC−/DTC− (50%), 26 CTC−/DTC+ (10%), 88 CTC+/DTC− (32%), and 23 CTC+/DTC+ (8%). We found that the CTC+/DTC+ group had the highest proportion of distant recurrence and deaths (Supplementary Fig. S5A). Log-rank tests revealed significant differences in DRFS (P = 0.0048), BCSS (P = 0.0106), and OS (P = 0.0132) among the four groups (Fig. 2A). Multivariate analysis showed that patients who were positive for both CTCs and DTCs (CTC+/DTC+) had inferior DRFS (hazard ratio, 3.09; P = 0.0270), BCSS (hazard ratio, 4.55; P = 0.0205), and OS (hazard ratio, 2.70; P = 0.0168) compared with those in the CTC−/DTC− group (Table 3).

The study cohort consisted of 87% HR-positive (n = 645) and 13% HR-negative patients (n = 92; Table 1; Supplementary Fig. S1B). We evaluated the prognostic impact of CTCs and DTCs in these two subsets using the same analysis approach performed on the entire cohort. The number of distant recurrences and deaths in each group are found in Supplementary Table S6.

For HR-positive patients, we observed the following:

Patients positive for CTCs (>0.44 CTCs/mL) had significantly reduced DRFS (hazard ratio, 2.12; P = 0.0311), BCSS (hazard ratio, 3.78; P = 0.0028), and OS (hazard ratio, 1.88; P = 0.0233; Fig. 3A). After adjusting for potential confounders, CTCs remained prognostic for BCSS (hazard ratio, 2.80; P = 0.0454; Table 2; Supplementary Table S5).

Using the cutoff of ≥1 CTC per 7.5 mL, we observed a significant association between CTC positivity and reduced DRFS (hazard ratio, 2.88; P = 0.0322; Supplementary Fig. S4B). Similarly, CTC positivity using the cutoff of ≥2 CTC per 7.5 mL was significantly associated with shorter DRFS (hazard ratio, 6.23; P = 0.0001) and BCSS (hazard ratio, 6.43; P = 0.0052; Fig. 3B). After multivariate analyses, CTCs remained significant predictors of DRFS (hazard ratio, 21.2; P = 0.0002) and BCSS (hazard ratio, 9.94; P = 0.0204; Table 2; Supplementary Table S5).

No significant prognostic impact was observed for DTCs (Fig. 3C; Table 2; Supplementary Table S5).

Using the optimized cutoffs for IE/FC, we categorized the 235 patients with paired DTC and CTC data (by IE/FC) into four groups: 116 CTC−/DTC− (49%), 20 CTC−/DTC+ (9%), 78 CTC+/DTC− (33%), and 21 CTC+/DTC+ (9%). The CTC+/DTC+ group had the highest proportion of distant recurrence and deaths (Supplementary Fig. S5B). Log-rank tests revealed significant differences in DRFS (P = 0.0145), BCSS (P = 0.0092), and OS (P = 0.0039) among the four groups (Fig. 2B). Multivariate analysis showed that patients who were positive for both CTCs and DTCs (CTC+/DTC+) had inferior DRFS (hazard ratio, 3.05; P = 0.0285), BCSS (hazard ratio, 3.90; P = 0.0355), and OS (hazard ratio, 3.03; P = 0.0091) compared with those in the CTC−/DTC− group (Table 3).

We did not observe significant correlation between CTCs and DTCs versus survival endpoints in the much smaller HR-negative subset (Supplementary Fig. S6).

CTC and DTC assessment can facilitate precision medicine–based management of patients with early breast cancer by identifying those with increased risk of metastatic recurrence. In this study, we used two clinically validated, EPCAM-based rare cell detection platforms for CTC enumeration: CellSearch (13–17, 25, 26) and IE/FC (18–22). Our previous studies in triple-negative metastatic breast cancer have demonstrated high concordance between these methods (20). A major difference between the two is that IE/FC has been validated for detection of both CTCs (18–20) and DTCs (18, 21, 22), while the current configuration of CellSearch only allows for CTC enumeration. One advantage of CTC or DTC detection by IE/FC is that it can be performed in concert with FACS for isolation of the tumor cells. This capability can in turn provide detailed genomic and phenotypic profiling for personalized medicine applications (19, 22, 27–31).

In this study, CTCs and DTCs were simultaneously enumerated in a cohort of patients with early breast cancer. We found 12 studies published between 1997 and 2018 that reported simultaneous CTC and DTC detection at the time of breast surgery (refs. 9, 10, 32–41; Supplementary Table S7). Of these, two assessed the prognostic impact of combined CTC and DTC detection (9, 34); however, these two studies did not compare quantitative results (e.g., tumor cells per mL or tumor cells per 10⁶ leukocytes) for both DTCs and CTCs. A fully quantitative assay, like IE/FC, is of interest because it enables enumeration of tumor cells from each compartment, that is, blood and bone marrow, and can report both CTCs/DTCs per mL, as well as CTCs/DTCs per 10⁶ MNCs.

We found that DTCs were present at generally higher levels than CTCs, including higher mean concentration and larger range. The higher levels of DTCs in marrow relative to CTCs in blood suggests that tumor cell dissemination is not merely stochastic, and that there may be an intrinsic difference in the biology of tumor cell localization to each compartment.

CTC detection by IE/FC was performed at the study outset, and thus the median follow-up for this cohort was particularly long (13.3 years). We found that CTC positivity by IE/FC in all patients, as well as in the HR-positive group was significantly associated with reduced BCSS and OS.

CTC detection by CellSearch, which was implemented later in the study, was significantly associated with poor DRFS (median follow-up 6.4 years) in all patients, as well as the HR-positive subset. Janni and colleagues using the CellSearch system previously demonstrated that CTCs in HR-positive early breast cancer are a significant prognostic factor for OS (5). In addition, Sparano and colleagues (42) recently reported that detection of CTCs by CellSearch 5 years after diagnosis of HR-positive breast cancer is associated with increased recurrence risk.

We observed a DTC positivity rate of 68% before cutoff optimization. In a large study that used immunocytochemistry (ICC)-based assay for detection of DTCs, the detection rate was 31% (6). The higher detection rate by IE/FC compared with the standard ICC method is likely due, at least in part, to the total number of cells analyzed in each assay. The standard ICC protocol for DTC detection typically examines about 4–8 million mononuclear cells per sample (11), while IE/FC examines approximately 176 million mononuclear cells per sample (4 mL of bone marrow); this is a >20-fold larger number of cells analyzed compared with that of the standard ICC assay.

DTC positivity by itself was not significantly correlated with survival in this study. However, when CTC and DTC status (both by IE/FC) were simultaneously considered, we found that positive detection for both CTCs and DTCs (CTC+/DTC+) in all patients, as well as in the HR-positive subset, was significantly associated with poor outcome. CTC+/DTC+ patients had significantly shorter DRFS, BCSS, and OS compared with those who were CTC−/DTC−. These results suggest that assessment of CTC and DTC status at surgery in patients with early breast cancer may help identify those who are at increased risk of distant recurrence and death due to breast cancer.

Our study observed that detection of CTCs in patients with HR-positive early breast cancer was an independent prognostic factor for DRFS (using CellSearch) and BCSS (using IE/FC). Simultaneous detection of DTCs provided additional prognostic power for outcome, including OS. These results are consistent with previous reports in which detection of DTCs (3, 6) and CTCs (5, 14, 15, 24, 43) have separately been demonstrated to have prognostic significance in early breast cancer. Although these methods have not yet become standard clinical tests for early breast cancer, it is possible that they may provide information about metastatic potential that complements existing tumor profiling assays.

Molecular characterization of CTCs and DTCs may provide novel insights into mechanisms of tumor dormancy, metastatic spread, and cancer recurrence. In this regard, we have previously reported strategies for molecular characterization of IE/FC-isolated CTCs and DTCs, and have confirmed the malignant nature of these cells (19, 22, 27–31).

In addition to EPCAM, cytokeratin expression has also been used for detecting cancer cells in the blood (e.g., CellSearch) and bone marrow (e.g., standard ICC; refs. 11, 44). In this study, we used the CellSearch system for CTC detection, which combines anti-EPCAM immunomagnetic enrichment with anti-cytokeratin ICC. In addition, we utilized the IE/FC strategy, which is based on dual epitope EPCAM capture. This approach offers several potential advantages: First, the assay configuration targets EPCAM in both immunomagnetic enrichment and flow cytometric steps, using two independent mAbs. This eliminates the possibility of missing tumor cells that fail to show adequate expression of two different antigens, such as EPCAM and selected cytokeratins, especially because breast cancer cells vary in their cytokeratin expression profile (45). Note that EPCAM-negative tumors will be missed by any strategy relying upon anti-EPCAM enrichment; however, because 90% of invasive breast cancer express EPCAM (46), and primary and metastatic breast cancer cells overexpress EPCAM by 100- to 1,000-fold (47), there is only a small possibility of missing breast cancer tumor cells. The IE/FC assay configuration also obviates the need for a permeabilization step to stain for intracellular cytokeratin antigens. Because detergent-based permeabilization may affect the suitability of cells for downstream analyses, the assay described here minimizes such manipulation by direct staining of intact cells prior to acquisition. Nonetheless, tumor cells that express EPCAM at low levels, for example, those undergoing epithelial–mesenchymal transition, will be missed by IE/FC and CellSearch, and thus represents a limitation of this study. While numerous clinical studies have demonstrated that EPCAM-positive CTCs are unequivocally associated with poor response and survival (5, 14, 15), the clinical relevance of EPCAM-negative tumor cells in circulation remains unclear (48).

In summary, we show that CTC detection either by CellSearch or IE/FC are adverse prognostic factors for distant recurrence and death. We also demonstrate the feasibility of simultaneous enumeration of CTCs and DTCs using the same quantitative IE/FC approach. With long follow-up (up to a median of 13.3 years), we show that detection of CTCs and DTCs at the time of surgery in patients with HR-positive early breast cancer is an independent prognostic factor for distant recurrence and breast cancer–specific death. Given the lack of early endpoints in this low-risk subtype, liquid biopsy may be an important consideration for future studies. Validation in an independent cohort is warranted to confirm the results of this study.

E. Bowlby-Yoder is an employee of and has ownership interests (including patents) at Agendia. J.W. Park reports receiving speakers bureau honoraria from Genetech and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Essermann, J.W. Park

Development of methodology: M.J.M. Magbanua, D.M. Wolf, J.W. Park

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.J.M. Magbanua, J.S. Lee, A. Chattopadhyay, J.H. Scott, E. Bowlby-Yoder, E.S. Hwang, M. Alvarado, C.A. Ewing, J.W. Park

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.J.M. Magbanua, C. Yau, D.M. Wolf, A. Chattopadhyay, L. Essermann, J.W. Park

Writing, review, and/or revision of the manuscript: M.J.M. Magbanua, C. Yau, D.M. Wolf, J.S. Lee, A. Chattopadhyay, E.S. Hwang, L.J. van't Veer, L. Essermann, J.W. Park

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.J.M. Magbanua, E. Bowlby-Yoder, J.W. Park

Study supervision: M.J.M. Magbanua, J.W. Park

Other (advocate review): A.L. Delson

We acknowledge the outstanding assistance of Margot Paisley, Alvina Leung, Alison Lozner, Teresa Seo, Kavitha Krishnan, Laura Petrillo, Amy Moore, Jasmine Wong, Hope Timberlake, and Richard Hwang in coordinating the clinical studies and Ann Griffin and Joseph McGuire for information on patient follow-up. This work was supported by University of California BioStar (S97-49, B99-55); NIH/NCI (U54 CA90788 and Early Detection Research Network U01 CA111234); and Breast Cancer Research Foundation [to M.J. M. Magbanua (BCRF-17-140) and to D.M. Wolf and L.J. van't Veer (BCRF-17-162)].

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.