Diagnosis of Basal-Like Breast Cancer Using a FOXC1-Based Assay

Abstract

Background:

Diagnosis of basal-like breast cancer (BLBC) remains a bottleneck to conducting effective clinical trials for this aggressive subtype. We postulated that elevated expression of Forkhead Box transcription factor C1 (FOXC1) is a simple and accurate diagnostic biomarker for BLBC.

Methods:

Accuracy of FOXC1 expression in identifying BLBC was compared with the PAM50 gene expression panel in gene expression microarray (GEM) (n = 1992) and quantitative real-time polymerase chain reaction (qRT-PCR) (n = 349) datasets. A FOXC1-based immunohistochemical (IHC) assay was developed and assessed in 96 archival formalin-fixed, paraffin-embedded (FFPE) breast cancer samples that also underwent PAM50 profiling. All statistical tests were two-sided.

Results:

A FOXC1-based two-tier assay (IHC +/- qRT-PCR) accurately identified BLBC (AUC = 0.88) in an independent cohort of FFPE samples, validating the accuracy of FOXC1-defined BLBC in GEM (AUC = 0.90) and qRT-PCR (AUC = 0.88) studies, when compared with platform-specific PAM50-defined BLBC. The hazard ratio (HR) for disease-specific survival in patients having FOXC1-defined BLBC was 1.71 (95% CI = 1.31 to 2.23, P < .001), comparable to PAM50 assay-defined BLBC (HR = 1.74, 95% CI = 1.40 to 2.17, P < .001). FOXC1 expression also predicted the development of brain metastasis. Importantly, unlike triple-negative or Core Basal IHC definitions, a FOXC1-based definition is able to identify BLBC in both ER+ and HER2+ patients.

Conclusion:

A FOXC1-based two-tier assay, by virtue of being rapid, simple, accurate, and cost-effective may emerge as the diagnostic assay of choice for BLBC. Such a test could substantially improve clinical trial enrichment of BLBC patients and accelerate the identification of effective chemotherapeutic options for this aggressive disease.

Following the elucidation of unique breast cancer molecular subtypes, the basal-like breast cancer (BLBC) subtype gained much attention because of its poor prognosis and lack of targeted therapy (1). Currently the incidence of BLBC is estimated to be 15% to 20% of all breast cancer cases and recent research suggests that when estrogen receptor–positive (ER+) breast cancers recur, approximately 30% will transform into the more aggressive basal-like phenotype (2,3). BLBC patients often are younger in age, of African American descent, display a high incidence of BRCA1 mutations and high histologic grade, suffer a high rate of metastasis to the brain and/or lung within three to five years of initial presentation, and have poor overall survival (4–7). Standard chemotherapy is not effective against BLBC, which currently lacks personalized, targeted therapeutic options.

One of the major obstacles in developing effective therapeutic options for BLBC has been the inability to accurately identify this molecular subtype using standard histopathological techniques. Most clinical trials have utilized the triple-negative phenotype (TNP)—negative for the expression of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2)—to define BLBC. The fact that BLBC is not synonymous with triple-negative breast cancer has been established by several investigators (3,8,9). Utilizing additional immunohistochemistry (IHC) markers (such as basal cytokeratins CK5/6, CK 14, CK17, and epidermal growth factor receptor [EGFR]) to better define BLBC has proven to be superior to using only the TNP, but they still lack accuracy (2,10,11). One glaring problem with current IHC protocols is their inability to diagnose the known occurrence of BLBC in ER+ and HER2+ tumors. In fact, 20% to 30% of BLBC tumors express ER and/or HER2 markers (Curtis and Parker datasets; Supplementary Figure 1, available online) (12,13). Thus validation of a diagnostic test for the accurate identification of BLBC in the clinic remains a critical bottleneck in efforts directed to personalize therapy for BLBC (14,15). Such a test needs to preserve the accuracy of BLBC prediction observed with the gene expression microarray/multimarker quantitative real-time polymerase chain reaction (qRT-PCR) PAM50 test (that can cost several thousand dollars), but enable performance in the end-user pathology laboratory at less than a tenth of the cost (13). A single marker for identification of basal-like breast cancer would reduce technical errors involved in a multimarker test and be easily integrated into current pathology practice alongside the established ER and HER2 tests.

A functional transcriptomics approach originally led to the identification of Forkhead Box C1 (FOXC1) as a characteristic tissue level biomarker for BLBC (10,16,17). Herein we validate the use of FOXC1 as a diagnostic and prognostic biomarker for BLBC, as compared with the PAM50 panel, to define this aggressive molecular subtype in large human microarray and qRT-PCR datasets (12,13). We report results of synchronous profiling of FOXC1 mRNA and protein expression in an independent cohort of matched human breast cancer samples (Table 1; Supplementary Table 1, available online). Our studies may lead to the development of a pragmatic, inexpensive molecular diagnostic test for BLBC with ready applicability for use in the clinic.

Table 1.

Clinical and pathological characteristics of patient cohorts used to assess diagnostic accuracy of FOXC1*

	Curtis et al. dataset	Parker et al. dataset	Jensen et al. dataset
	(n = 1992) No. (%)	(n = 349) No. (%)	(n = 96) No. (%)
Platform	cDNA Microarray	FFPE qRTPCR	FFPE IHC + qRTPCR
Factor
Age, y
<50	426 (21.4)	56 (16.0)	25 (26.0)
>50	1566 (78.6)	65 (18.6)	71 (74.0)
Unknown	0 (0.0)	228 (65.3)	0 (0.0)
Tumor size, cm
<2	862 (43.3)	129 (37.0)	49 (51.0)
2–5	1008 (50.6)	165 (47.3)	37 (38.5)
>5	102 (5.1)	44 (12.6)	5 (5.2)
Unknown	20 (1.0)	11 (3.2)	5 (5.2)
Lymph node status
0	1042 (52.3)	178 (51.0)	61 (63.5)
1–3	625 (31.4)	127 (36.4)	19 (19.8)
>3	318 (16.0)	9 (2.6)	12 (12.5)
Unknown	7 (0.4)	35 (10.0)	4 (4.2)
Tumor grade
Low	170 (8.5)	157 (45.0)	7 (7.3)
Intermediate	775 (38.9)	71 (20.3)	22 (22.9)
High	957 (48.0)	114 (32.7)	62 (64.6)
Unknown	90 (4.5)	7 (2.0)	5 (5.2)
ER (IHC)
Negative	440 (22.1)	120 (34.4)	59 (61.5)
Positive	1508 (75.7)	206 (59.0)	37 (38.5)
Unknown	44 (2.2)	23 (6.6)	0 (0.0)
PR (IHC)
Negative	943 (47.3)	149 (42.7)	62 (64.6)
Positive	1049 (52.7)	144 (41.3)	34 (35.4)
Unknown	0 (0.0)	56 (16.0)	0 (0.0)
HER2 (IHC, FISH)
Negative	1546 (77.6)	258 (73.9)	73 (76.0)
Positive	441 (22.1)	60 (17.2)	23 (24.0)
Unknown	5 (0.3)	31 (8.9)	0 (0.0)
PAM50-defined BLBC
Negative	1655 (83.1)	242 (69.3)	55 (57.3)
Positive	331 (16.6)	107 (30.7)	41 (42.7)
Unclassified	6 (0.3)	0 (0.0)	0 (0.0)
FOXC1-defined BLBC
Negative	1636 (82.1)	236 (67.6)	56 (58.3)
Positive	356 (17.9)	113 (32.4)	40 (41.7)
Unclassified	0 (0.0)	0 (0.0)	0 (0.0)

* BLBC = basal-like breast cancer; ER = estrogen receptor; FFPE = formalin-fixed paraffin-embedded; FISH = fluorescent in situ hybridization; IHC = immunohistochemistry; PR = progestin receptor; qRT-PCR = quantitative real-time polymerase chain reaction.

Methods

Immunohistochemistry of Formalin-Fixed Paraffin-Embedded Samples

Formalin-fixed paraffin-embedded (FFPE) blocks from 118 patients enriched for ‘triple-negative’ breast cancer were drawn from the archives of the Department of Pathology at Carle Foundation Hospital, Urbana, IL in accordance with the approval of the Carle Institutional Review Board (Table 1; Supplementary Table 1, available online). One hundred and twelve blocks were found to have adequate tissue for processing. Of these, 96 had adequate tissue available to perform both IHC and qRT-PCR profiling (see STARD diagrams in Supplementary Materials, available online) (18,19). Freshly cut sections were prepared at 5 microns and submitted to the University of Southern California, Keck School of Medicine Immunohistochemistry laboratory, after removal of all personal, demographic, and clinical data. Immunohistochemical staining was performed for FOXC1 (Onconostic Technologies, Inc., Champaign, IL; see Supplementary Materials, available online, for details), CK5/6 (D5 & 16B4, Cell Marge, Rocklin, CA), and EGFR (Cat #M3563, Dako, Carpinteria, CA) on a Leica Bond III automated platform (Leica, Buffalo Grove, IL) using the standard H2 protocols. Development of a modified Allred method for FOXC1 IHC scoring is discussed under Results. See the Supplementary Materials (available online) for CK 5/6 and EGFR IHC scoring performed using reported criteria (20,21).

qRT-PCR of FFPE Samples

qRT-PCR assays were performed on mRNA obtained from FFPE samples as described by Mullins et al. and Parker et al. with some modifications (see Supplementary Materials, available online, for details) (13,22).

FFPE Sample Subtyping Analyses

Centroid data for each PAM50-defined molecular subtype described by Parker et al. (13) was obtained from the UNC Microarray Database (https://genome.unc.edu/pubsup/breastGEO/ #G1). Ninety-six FFPE samples yielding sufficient mRNA were initially subtyped based on a nearest centroid analysis using the PAM50 centroids and assigned a molecular subtype after confirmation by PAM reanalysis (see the Supplementary Materials, available online, for details). A heat map of the qRT-PCR data was generated using GenePattern (Broad Institute, Cambridge, MA) hierarchical clustering module with data log transformed, uncentered correlation, pairwise average-linkage (23). Results were imaged with Hierarchical Clustering Image in GenePattern.

Statistical Analysis

Prognostic significance of FOXC1 in predicting disease-specific survival was examined in the Curtis et al. dataset (12). Organ-specific metastasis-free survival was examined in the van de Vijver et al. dataset (24). Kaplan-Meier plots were generated using Prism software (GraphPad Software, La Jolla, CA). Univariate and multivariate analyses were performed using Cox regression model (SAS software, SAS Institute Inc., Cary, NC). Assumptions of proportionality were tested by inspecting plots of log(-log(survival function)) vs log(survival time) under different predictor values, and by testing statistical significance of the interaction effects of predictors and log(survival time) (Supplementary Table 2, available online). If the proportionality assumption was violated, the average effects of hazard ratios over time are provided (Table 2; Supplementary Table 3, available online). For our final results, we did not need to rely on proportionality assumptions. Variables included in the multivariate analysis are age, tumor size, lymph node status, and either PAM50 or FOXC1-defined BLBC status (Table 2). All tests were two-sided, and P values of less than .05 were considered statistically significant.

Table 2.

Univariate and multivariable analysis of 1992 patient dataset (Curtis et al. [12]) used to validate prognostic value of FOXC1 for 10-year follow-up

Characteristic	Univariate analysis			Multivariable analysis (age, tumor size, node status, PAM50-defined BLBC)			Multivariable analysis (age, tumor size, node status, FOXC1-defined BLBC)
	n	P*	HR (95% CI)	n	P*	HR (95% CI)	n	P*	HR (95% CI)
Age	1975	.20	1.00 (0.99 to 1.01)	1944	.91	1.00 (0.99 to 1.01)	1950	.66	1.00 (0.99 to 1.01)
Tumor size (0–2, 2–5, >5)	1955	<.001	1.86 (1.60 to 2.18)	1944	<.001	1.57 (1.34 to 1.85)	1950	<.001	1.57 (1.34 to 1.85)
Tumor grade (low, medium, high)	1887	<.001	1.82 (1.54 to 2.15)	—	—	—	—	—	—
Nodal Status (positive vs negative)	1969	<.001	2.82 (2.32 to 3.44)	1944	<.001	2.49 (2.03 to 3.05)	1950	<.001	2.53 (2.06 to 3.10)
ER status (IHC)	1932	<.001	0.47 (0.39 to 0.57)	—	—	—	—	—	—
PR status (IHC)	1975	<.001	0.49 (0.40 to 0.59)	—	—	—	—	—	—
HER2 status (IHC, FISH)	1970	<.001	1.78 (1.46 to 2.18)	—	—	—	—	—	—
PAM50-defined BLBC	1969	<.001	1.74 (1.40 to 2.17)	1944	.001	1.56 (1.23 to 1.98)	—	—	—
FOXC1-defined BLBC	1975	<.001	1.71 (1.31 to 2.23)	—	—	—	1950	.003	1.55 (1.17 to 2.06)

* Two-sided P values were based on χ² or Fisher’s exact test, whenever appropriate. BLBC = basal-like breast cancer; CI = confidence interval; ER = estrogen receptor; HR = hazard ratio; IHC = immunohistochemistry; PR = progestin receptor.

Results

Generation and Validation of FOXC1 Antibody for IHC

To improve IHC detection of the FOXC1 protein in human breast cancers, we generated a monoclonal mouse antibody against the FOXC1 N-terminal domain. The specificity of this antibody for detection of nuclear FOXC1 was confirmed using immunoblotting and immunofluorescence of MCF-7 breast cancer cells (which expressed low or undetectable FOXC1 levels) transfected with either vector control, Myc-tagged human FOXC1, or Myc-tagged FOXC2, a FOXC1 homolog (Supplementary Figure 2, available online) (10). Both the Myc and the FOXC1 antibodies detected the FOXC1 protein, whereas only the Myc antibody, but not the FOXC1 antibody, recognized the FOXC2 protein (Supplementary Figure 2A, available online), highlighting lack of cross-reactivity of the FOXC1 antibody with FOXC2. This monoclonal antibody promises to enable more sensitive and specific clinical evaluation of FOXC1 expression in FFPE samples than can be achieved with polyclonal antibodies.

FOXC1 Protein Expression in FFPE Tissue Sections as Determined by IHC

This portion of the study was performed in accordance with STARD guidelines (See STARD diagrams in the Supplementary Materials, available online) (18,19). The patterns of IHC staining were scored by three pathologists individually. All three pathologists remained blinded with respect to laboratory (ER, PR, HER2, PAM50 molecular subtype) and clinical data (staging, etc.). Prior to first examination, the three pathologists met and agreed to perform initial scoring by overall intensity and percentage of positive cancer cells (FOXC1 staining is nuclear in location) in increments of 1 below 10%, and thereon in ‘buckets’ of 10%, leading to a scoring process based upon the established ‘H’ and Allred scoring guidelines for ER and PR (25–27). A finalized modified Allred Score for the IHC assessment of FOXC1 positivity, using a 0 to 8 score, was developed by three pathologists (CRT, WYP, XL) who were blinded to the receptor profile and molecular subtype of the matched breast cancer samples (Supplementary Figure 3, available online). The IHC staining results for FOXC1 compared with both CK 5/6 and EGFR for FFPE samples clustered by the PAM50 genes are shown in Figure 1A. Comparatively, FOXC1 protein is specific to basal-like tumors and not merely enriched in this subset.

Figure 1.

PAM50-defined basal-like breast cancer subtype and expression of various associated markers. A) Dendrogram of an independent cohort of 96 patient/tumor samples (vertical columns) hierarchically clustered on the basis of the PAM50 genes (horizontal rows). All samples were subjected to immunohistochemical (IHC) testing for CK 5/6, epidermal growth factor receptor (EGFR), and FOXC1 using matched, whole tissue sections to avoid errors related to tissue heterogeneity. Interpreted CK 5/6 and EGFR IHC scores—positive (red) or negative (blue). Interpreted FOXC1 IHC scores—positive (red), intermediate/indeterminate (purple), or negative (blue). Samples with intermediate/indeterminate FOXC1 on IHC “reflexed” to a FOXC1 real-time polymerase chain reaction test—positive (red) or negative (blue), for final determination of FOXC1 status. B) Representative hematoxylin and eosin staining as well as immunostaining for FOXC1, CK 5/6, and EGFR in a basal-like tumor. EGFR = epidermal growth factor receptor; H&E = hematoxylin and eosin.

FOXC1 Expression as a Predictor of Basal-Like Subtype

The molecular subtype of 96 FFPE samples enriched for triple-negative phenotype was determined using the PAM50 panel of genes. The molecular subtype prediction relative to ER/HER2 marker status is shown in Supplementary Table 1 (available online). Both protein and mRNA expression levels of FOXC1 are statistically significantly higher (P < .001) in basal-like breast tumors compared with other subtypes, with a high correlation observed between FOXC1 mRNA and protein expression (correlation coefficient = 0.68) (Figure 2, A-C). While using the IHC FOXC1 scoring system to predict basal-like subtype defined by PAM50, cutoff values of 4 (sensitivity = 0.84, specificity = 0.79) or 5 (sensitivity = 0.75, specificity = 0.90) maximized the sum of sensitivity and specificity values (Supplementary Figure 4A, available online) in a close range. A single cutoff value for the qRT-PCR test resulted in a higher sum of sensitivity (0.89) and specificity (0.90).

Figure 2.

Expression of FOXC1 mRNA and protein in PAM50-defined molecular subtypes. A) The average value for the Modified Allred immunohistochemistry (IHC) score for each molecular subset defined by PAM50 (*P < .001). B) The average value for FOXC1 mRNA expression normalized to housekeeping genes for each PAM50-defined molecular subtype (**P < .001). C) Increasing expression of FOXC1 RNA in a sample corresponds with higher levels of FOXC1 protein measured by the B2E3 antibody. Error bars represent 99% confidence interval for each value. Two-sided P values are determined using the Student’s t test. HER2 = human epidermal growth factor receptor 2.

The ability of FOXC1 expression to predict the basal-like subtype was further assessed by examining the area under curve (AUC) for receiver operator characteristic (ROC) curves generated for microarray values of FOXC1 in the Curtis dataset and for qRT-PCR generated values of FOXC1 from FFPE tissues in the Parker dataset (Table 1) (12,13). FOXC1 expression proved to be highly diagnostic of BLBC in both the Curtis (AUC = 0.90) and Parker (AUC = 0.88) datasets (Figure 3, A and B). The AUCs for the FOXC1 IHC test (AUC = 0.86) and qRT-PCR test (AUC = 0.92) in the 96 FFPE samples are comparable with those calculated in other datasets (Figure 3, C and D).

Figure 3.

Receiver operating characteristic curves (ROC AUC) curves of FOXC1-defined basal-like breast cancer (BLBC) vs PAM50-defined BLBC generated using (A) gene expression data from a 1992 patient microarray dataset (Curtis et al. [12]), (B) gene expression data from a 349 patient quantitative real-time polymerase chain reaction (qRT-PCR) dataset (Parker et al. [13]), (C) FOXC1 data from an independent dataset of 96 patients used to develop a novel FOXC1-based diagnostic assay for BLBC showing values for protein expression (immunohistochemistry), and (D) the same 96 patient samples measuring FOXC1 with qRT-PCR. Each dataset had their respective, separate PAM50 gene expression data used to define BLBC for the comparison. AUC = area under the curve; BLBC = basal-like breast cancer; FFPE = formalin-fixed, paraffin-embedded; qRT-PCR = quantitative real-time polymerase chain reaction.

Because FOXC1 is included in the PAM50 panel, we repeated the above analysis in our cohort by first classifying tumors using PAM50 minus FOXC1. As expected for the removal of a single gene from the PAM50 analysis, only two designations were altered when FOXC1 was removed from the PAM50 subtyping (both from Basal-like to Luminal B subtype) (13,22). The AUCs for the IHC (AUC = 0.84) and qRT-PCR (AUC = 0.90) tests in the reassigned, FOXC1-excluded (PAM49) dataset are comparable with the values obtained from the full PAM50 assignments.

Based on the above data, if IHC testing alone was to be utilized for detection of FOXC1 expression in human FFPE breast cancer tissue using these protocols, the cutoff that maximizes the sum of sensitivity and specificity would be either 4 or 5 (Supplementary Figure 4A, available online). However, this approach has a lower sensitivity and/or specificity relative to the more-difficult-to-perform qRT-PCR test. To maximize the simplicity and short turnaround time of IHC and retain the performance of qRT-PCR, we examined a novel two-tier reflex strategy in which an initial IHC test would only require supplemental qRT-PCR for intermediate IHC FOXC1 scores; thus balancing ease of implementation with maximal test accuracy. FOXC1 IHC scores of 0–3 were considered to be negative, 6–8 were considered to be positive, and 4–5 were considered to be equivocal for FOXC1 expression. Samples scored 4–5 would automatically “reflex” to the FOXC1-based qRT-PCR assay as the more quantitatively accurate test. Using this model we obtained superior sensitivity and specificity of 0.84 and 0.94, respectively (AUC = 0.88) (Supplementary Figure 4B, available online). This single marker, two-tier testing strategy is comparable with other BLBC definitions such as TNP (sensitivity = 0.93, specificity = 0.79) and core-basal phenotype (sensitivity = 0.91, specificity = 0.85). However, unlike both TNP and core-basal phenotype, FOXC1 is capable of identifying basal-like tumors in both ER+ and HER2+ tumors. FOXC1 expression is capable of identifying basal-like tumors independent of ER or HER2 expression status with high sensitivity and specificity (Figure 4).

Figure 4.

Sensitivity and specificity of FOXC1-based prediction of basal-like breast cancer (BLBC) segregated by immunohistochemical (IHC) marker expression. FOXC1 expression can identify PAM50-defined BLBC with high sensitvity and specificity in all subgroups defined by estrogen receptor and human epidermal growth factor receptor 2 IHC expression in both a microarray dataset (Curtis et al. [12]) and quantitative real-time polymerase chain reaction dataset (Parker et al. [13]). Cutoff levels of FOXC1 for BLBC prediction are optimized for each marker subgroup. Cutoff values are given for expression levels of the entire dataset and not for each individual subgroup. ER = estrogen receptor; HER2 = human epidermal growth factor receptor 2.

Validation of FOXC1 as a Prognostic Biomarker

With the goal of validating the earlier reported prognostic significance of FOXC1 in breast cancer by accurately detecting the BLBC molecular subtype, we examined the 1992-sample Curtis et al. dataset with respect to disease-specific survival (DSS) and the 295-sample van de Vijver et al. dataset with respect to distant metastasis-free survival (DMFS) (10,12,16,24). In univariate analysis, DSS was statistically significantly worse for patients identified as having either PAM50-defined BLBC (Figure 5A) or FOXC1-defined BLBC (P < .001) (Figure 5B). The hazard ratio (HR) for disease-specific survival in patients having FOXC1-defined BLBC was 1.71 (95% CI = 1.31 to 2.23, P < .001), comparable with PAM50 assay-defined BLBC (HR = 1.74, 95% CI = 1.40 to 2.17, P < .001) (Table 2). In multivariate analysis, FOXC1 expression was an independent prognostic indicator of DSS after adjusting for clinical/pathologic variables such as age, tumor size, and lymph node status (top 10th decile cutoff used in this validation dataset based on predetermined cutoff value from earlier training datasets, HR = 1.55 95% CI = 1.17 to 2.06, P = .003), and again was comparable with PAM50 basal-like designation (HR = 1.56, 95% CI = 1.23 to 1.98, P = .001) (Table 2) (10,16). Furthermore, the FOXC1-defined Basal-like designation allowed prognostic stratification of lymph node–negative breast cancer patients similar to the PAM50 basal-like designation (P = .002 and .007, respectively) in the Curtis dataset (Figure 5, C and D). Using multivariate analysis FOXC1-defined Basal-like designation was confirmed to be an independent prognostic indicator even in this lymph node–negative subset of patients (Supplementary Table 3, available online).

Figure 5.

Validation of the prognostic significance of FOXC1 in breast cancer with regard to disease-specific survival (DSS) in an independent 1992 patient Microarray Dataset (Curtis et al. [12]). Kaplan-Meier analysis of DSS in all patients with basal-like breast cancer (BLBC) defined using either (A) PAM50 or (B) FOXC1 with a top 10^th decile cutoff. Kaplan-Meier analysis of DSS in lymph node–negative (LN-) patients with BLBC defined using either (C) PAM50 or (D) FOXC1 with a top 10^th decile cutoff. All P values are two-sided. BLBC = basal-like breast cancer; DSS = disease-specific survival.

Elevated FOXC1 expression was also found to be positively associated with brain (P = .04) and lung (P = .01) metastasis in the van de Vijver dataset, (Supplementary Figure 5, available online), further validating this predictive association reported earlier in another independent dataset (10,24). This correlation was stronger than the PAM50 basal-like association with brain metastasis (P = .31) and comparable with that for lung metastasis (P = .01) (Supplementary Figure 5, available online). While both FOXC1 and PAM50 basal-like designations exhibited a negative correlative trend with bone metastasis, neither correlation was statistically significant (P = .61 and .23 for PAM50 and FOXC1, respectively) (Supplementary Figure 5, available online). The ability to accurately identify BLBC tumors with a single gene biomarker holds promise of greatly increasing the clinical application of BLBC status in assessing patient outcomes.

Discussion

Sequencing of the human genome has improved our understanding of the molecular underpinnings of human disease and accelerated development of tailored personalized therapeutics for complex diseases, including cancer. While several tissue level cancer biomarkers are reported in the literature every year, few survive validation to warrant utilization in clinical management. This attrition may be because of the fact that while strong associations are often initially derived between putative biomarker expression and disease status, candidate markers often fail to have any meaningful and disease-relevant function in clinical laboratory and therapeutic practice.

Using a transcriptomics-driven approach, we initially reported that mRNA overexpression of FOXC1 was a tissue-level biomarker of BLBC and was superior to surrogate biomarker panels (10). We subsequently went on to suggest that FOXC1 protein overexpression also appeared to be a highly characteristic feature of BLBC, albeit in an independent cohort of human breast cancer tissue samples (16). FOXC1 mRNA and protein expression levels were demonstrated to be specific for basal-like breast cancer tissue and/or cell lines (on microarray analysis and immunoblotting). Finally, and most importantly, FOXC1 was demonstrated to be of critical and central importance in orchestrating the aggressive biology and metastatic potential of BLBC (17,28). As such FOXC1 represents a potential therapeutic target and/or surrogate predictive marker of therapeutic efficacy in BLBC. However, with respect to its diagnostic potential, the accuracy of FOXC1 protein expression as measured by IHC had never been compared with the academic gold standard definition of BLBC, namely multimarker gene expression on a microarray platform (Intrinsic Gene List) or multimarker gene expression on a qRT-PCR platform (PAM50). Absent such a study, the cost-effective diagnosis of BLBC based on IHC FOXC1 expression could not be proposed.

In the present study we used a large, publicly available microarray dataset to validate the prognostic importance of FOXC1 expression in breast cancer reported previously using other datasets, thereby confirming its accuracy in predicting BLBC-associated poor prognosis (10,12,16). Importantly, in both the prior and current datasets, the prognostic predictive ability of FOXC1 expression was also confirmed using multivariate analysis (Supplementary Table 3, available online) in the lymph node–negative subset of patients, who might otherwise be expected to have had more favorable outcomes. Prognostic significance of FOXC1 with regard to predicting organ-specific metastases and survival was also validated in an independent dataset. We further demonstrated the proof-of-concept of an in vitro molecular diagnostic test, the FOXC1-based two-tier assay, for the accurate detection and diagnosis of BLBC comparable with the qRT-PCR–based PAM50 assay. This diagnostic test is based on the IHC detection of the FOXC1 transcription factor in FFPE human breast cancer tissues using an epitope-specific monoclonal antibody developed specifically for that purpose. The test has the benefit of having a second tier single-gene qRT-PCR test available to resolve the diagnostic dilemma that arises when equivocal results are obtained on IHC alone.

It has been assumed that BLBC is found only within triple-negative breast cancers and is not found in those tumors that express ER and HER2. Contrary to such an assumption and in agreement with prior reports in the literature, we identified BLBC in non–triple negative breast cancer samples, both in microarray and qRT-PCR datasets as well as FFPE profiled samples (Figure 4; Supplementary Table 1, available online), suggesting that all breast samples, as opposed to only triple-negative breast cancer samples, should be evaluated for harboring a BLBC molecular subtype (29). Assessment of FOXC1 expression meets this need. Other proposed methods of identifying BLBC, including TNP status or core-basal staining (TNP with positive staining for EGFR and/or CK 5/6), by definition will not identify BLBC in ER+ or HER2+ tumors.

The main limitation of the current study is its retrospective design. Also the conclusions cannot be generalized, as all patients were treated either in the United States or Europe at different time points. As the distribution of clinical characteristics might be different in patients from other areas (even within the United States and Europe), the current study should be considered to suffer from biases inherent to such a study design. Our results do however suggest that they merit further large-scale retrospective and ultimately prospective independent validation.

In summary, based on the above results, we propose that a highly sensitive and specific, two-tier testing strategy for FOXC1 expression may be the diagnostic assay of choice for BLBC. This would be akin to the current concept of determining HER2 status in the clinic, which uses a combination of IHC and fluorescent in situ hybridization assays, the latter to aid definitive diagnosis of HER2 expression status in those cases that have equivocal results rendered on initial IHC for HER2. The two-tier FOXC1-based molecular diagnostic assay for BLBC merits further large scale validation ideally in the context of a multi-institutional retrospective study, prior to undertaking prospective validation at independent centers. Such a study is currently underway.

Funding

This work was supported by research grants/contracts from the National Cancer Institute at the National Institutes of Health 261201100089C-0-0-1 (PSR) and 261201300028C-0-0-1 (PSR, TR), 5R01CA151610-04 (XC), an Avon Foundation research grant 02-2014-063 (XC), the David Salomon Breast Cancer Research Fund (XC), Carle Foundation Translational Cancer Research Fund (PSR), and a Career Development Award from the Warren H. and Clara Cole Society (PSR).

The study funders had no role in the design of the study, the collection, analysis, or interpretation of the data, the writing of the manuscript, nor the decision to submit the manuscript for publication.

Authors’ contributions: PSR designed the study. TWJ, TR, JW, XL, WYN, BH, and FB obtained and assembled the data. TWJ, TR, XL, WYN, SPB, AQ, XC, CRT, and PSR analyzed and interpreted the data. TWJ, TR, and PSR wrote the manuscript, which was edited by all the authors who approved the final version. All authors are guarantors of the integrity of the data collection and interpretation.

The authors declare the following conflicts of interest: TWJ, consultant, Onconostic Technologies, Inc.; TR, employee, Onconostic Technologies, Inc.; JW, patent related to FOXC1 in cancer; XL, none; WYN, none; BH, none; FB, none; SPB, patent related to FOXC1 in cancer, stock ownership in and consultant, Onconostic Technologies, Inc.; AQ, none; XC, patents related to FOXC1 in cancer; CRT, stock ownership in and consultant, Onconostic Technologies, Inc.; PSR, patents related to FOXC1 in cancer, stock ownership in and consultant, Onconostic Technologies, Inc.