Quantitative assessment of invasive mena isoforms (Mena^calc) as an independent prognostic marker in breast cancer

This study was conducted using data from two cohorts of breast cancer patients. The first cohort consists of 501 patients who underwent surgery at Yale University Cancer Center/Yale New Haven hospital between1962 and 1982 and had formalin-fixed, paraffin-embedded (FFPE) primary invasive breast tumors available for study. Cohort 2 consists of 296 patients who had surgery for breast cancer at Yale University Cancer Center/Yale New Haven hospital between1976 and 2005 and for whom FFPE tissue was available. Tissue microarrays were constructed in two-fold redundancy for each cohort. Both cohorts have been described previously [17, 18]. In both, follow-up information on cases was obtained from the Yale New Haven Tumor Registry, the Yale-New Haven Hospital medical records and the Connecticut Death Records. Tissues were collected in accordance with consent guidelines using protocol number 9500008219 issued to DLR from the Yale Institutional Review Board, most recently reapproved in May 2012.

The arrays were deparaffinized first by melting at 60°C in an oven equipped with a fan for 20 minutes followed by 2x xylene treatment for 20 minutes each. Slides were then rehydrated and antigen retrieval was done in citrate buffer (pH 6.0) at 97°C for 20 minutes in a PT module (Labvision, Kalamazoo, MI, USA). Endogenous peroxidase was blocked by using 0.3% hydrogen peroxide in methanol for 30 minutes in the dark followed by incubation of slides in a blocking buffer (0.3% bovine serum albumin in TBST (0.1 mol/L of TRIS-buffered saline (pH 7.0) containing 0.05% Tween-20)) for 30 minutes at room temperature. Slides were incubated with a cocktail of mouse anti-pan-Mena (1:1,000 dilution; BD Biosciences, San Jose, CA, USA, catalog number 610693) mixed with rabbit anti-Mena11a (1:500 dilution of 1 mg/ml stock; generated in the laboratory of FBG) in the blocking buffer overnight at 4°C. After washing away the primary antibodies, slides were incubated with secondary antibody (goat anti-rabbit conjugated to horseradish peroxidase, Jackson ImmunoResearch Laboratories Inc., West Grove, PA, USA) to target Mena11a for one hour. After washing, slides were incubated with biotinylated tyramide (Perkin Elmer, Waltham, MA, USA) diluted as 1:50 in amplification buffer (Perkin Elmer) for 10 minutes. After washing, peroxidase activity was quenched by 2x treatment with benzoic hydrazide (100 mM in PBS; Sigma-Aldrich, St. Louis, MO, USA) with 50 mM hydrogen peroxide for seven minutes each. After washing, slides were incubated for an hour with goat anti-mouse envision (DAKO, Carpinteria, CA, USA) followed by treatment with a chicken anti-Pan cytokeratin (1:100, generated in house) for 2 h at room temperature. Slides were washed and then incubated with goat anti-chicken conjugated to Alexa546 (Invitrogen, Grand Island, NY, USA) to visualize cytokeratin ) and streptavidin conjugated to CY7 (750 nm, Invitrogen, Grand Island, NY, USA) to visualize Mena11a for an hour. After washing, slides were treated with CY5 conjugated tyramide (1:50 dilution; Perkin Elmer, Waltham, MA, USA) in amplification buffer for 10 minutes to visualize pan-Mena. Slides were mounted with ProLong gold mixed with DAPI (Molecular Probes, Grand Island, NY, USA). Serial sections of the index array used for assay standardization [18] were stained alongside each cohort to assess the assay reproducibility. An additional serial section of the index array was stained with each experiment with no primary antibodies as a negative control.

The AQUA technology (HistoRx, Branford, CT, USA) allows quantitative measurement of biomolecules in sub-cellular compartments as described previously [19, 20]. Briefly, a series of monochromatic images for each histospot was captured using a PM-2000 microscope (HistoRx, Branford, CT, USA) equipped with an automated stage. A binary 'tumor mask' was created using cytokeratin staining of the histospot representing only epithelial cells and excluding stromal features. AQUA scores for both pan-Mena and Mena11a were calculated by dividing the signal intensity by the area of the specific compartment (in this case within the tumor mask area). Normalized AQUA scores for both targets (pan-Mena and Mena11a) were used to calculate the Mena^calc fraction for each histospot by subtracting the z score of Mena11a from the z score of pan-Mena as described below.

Pearson's correlation coefficient (R) was used to assess the reproducibility of the multiplexed assay between near-serial sections of the index array. Based on extensive experience in our laboratory, an R ² value greater than 0.4 was considered acceptable for both inter- and intra-array reproducibility. For both cohorts, pan-Mena, Mena11a AQUA scores, and Mena^calc values from two independent cores for each histospot were averaged and the averages were used for final analysis. Both cohorts were independently divided into quartiles based on the distribution of Mena^calc values in the respective cohorts. Within each cohort, the survival rates in the lowest three quartiles were overlapping so these three quartiles combined into a single group (Q1 to 3, data not shown). The top quartile (Q4) was compared with the lowest three quartiles for disease-specific survival differences using the log-rank test. A multivariate Cox proportional hazards analysis was used to examine the prognostic value of Mena^calc while controlling for other common prognostic factors as well as potential confounders. Since similar differences between Mena^calc groups were observed in both cohorts, we combined the two cohorts for the multivariate analysis to gain efficiency while stratifying on the cohort to allow the two cohorts to have different baseline survival functions, as the Kaplan-Meier survival curves suggested that Cohort 2 had better survival. All P-values were two-sided, and P-values <0.05 were considered to be statistically significant. All statistical analyses were done using SAS 9.2 (SAS Institute, Cary, NC, USA).

We developed and validated a method of MQIF to assess the relative abundance of total Mena and the anti-metastatic Mena11a by multiplexing pan-Mena antibody with antibody against Mena11a, and then subtracting the Mena11a from the total Mena as measured by a "pan-Mena" antibody (that recognizes all known Mena protein isoforms). Pan-cytokeratin staining was used to define the epithelial component of the tumor area in which the signal intensity of total Mena and Mena11a was assessed. As expected, a cytoplasmic localization pattern was observed for both pan-Mena and Mena11a (Figure 1A). Inter-array reproducibility of the MQIF assay was evaluated by staining serial sections of an index array TMA (tissue microarray) consisting of a small subgroup of breast cancer patients used as a control array for each experiment. This yielded Pearson's R ² values of 0.841 for pan-Mena and 0.725 for Mena11a, indicative of good reproducibility (Figure 1B, C). Further, cross-reactivity of the two antibodies was determined using linear regression between the AQUA scores of both antibodies for the same index slide. As shown in Figure 1D, there was no correlation (R ² = 0.059) between pan-Mena and Mena11a, suggesting assay isoform specificity that total Mena expression levels are independent of the frequency of 11a exon inclusion.

The different isoforms of Mena are shown schematically in Figure 2A. The relative amount of Mena lacking 11a was calculated by first converting the AQUA scores of both pan-Mena and Mena11a to Z scores to normalize the AQUA scores to the same scale. Then Mena^calc, a measure of the abundance of Mena lacking its anti-metastatic Mena11a isoform, was calculated by simple subtraction of Z scores of Mena11a from the total Mena (Figure 2B). The multiplexed assay was applied to the two primary breast cancer cohorts on TMAs. Tumor heterogeneity was evaluated by comparing Mena^calc values in two different cores (histospots) from the same tumor blocks. This showed significant correlation between histospots (R ² = 0.653 for cohort 1 and 0.626 for cohort 2), as shown in Figure 2C, D. Based on our experience with TMA and AQUA technology, two independent 0.6 mm cores from each patient has been considered representative of a whole tissue section. While this is clearly a function of the heterogeneity of the marker considered and the tumor type, the use of two cores is often the standard for TMA based studies [21, 22].

Clinicopathological characteristics of both cohorts and of the combination of the two are presented in Table 1. Compared to cohort 1, the tumors in cohort 2 were smaller in size and more likely to be ER⁺, PR^- and Her2^-.

Mena^calc was analyzed using averaged AQUA scores of two independent cores for both cohorts. We evaluated the prognostic impact of Mena^calc in Cohort1 using 20 years of follow-up for survival of patients by dividing the population (501 patients) into quartiles. The highest quartile (Q4) of Mena^calc was strongly associated with poor outcome (log rank P = 0.0004, Figure 3A). The lowest three quartiles (Q1 to 3) did not differ between each other in their association with survival (data not shown) and, therefore, we combined them for further analysis. Pan-Mena and Mena11a quartiles alone were not associated with survival (Figure 3A) (log rank P = 0.0749 and P = 0.5255, respectively). Cohort 2, consisting of 296 patients, yielded similar results. Again, the highest quartile of Mena^calc, but not pan-Mena or Mena11a, was able to stratify patients for poor outcome (log rank P = 0.0321, Figure 3B). Cohort 2 consists of patients from 1976 to 2005 and thus shows fewer events (53 for Cohort 2 vs 244 for Cohort 1) as shown in Figure 3A, B and thus is less well-powered. Additionally, we assessed the impact of Mena^calc on survival within subgroups of both cohorts. Mena^calc was prognostic in the node positive subset with a log rank P-value of 0.0039, but not in node negative patients (Figure 4A). Likewise, Mena^calc showed prognostic value among the ER negative patient group (log rank P = 0.0004), but not in ER positive patients (log rank P = 0.5651). We evaluated the same subgroups in Cohort 2 but failed to see any significant differences within these subgroups (Figure 4B).

The cohorts were combined for multivariate analysis, in which a multivariate Cox proportional hazards analysis was used to examine the prognostic value of Mena^calc while controlling for other common prognostic factors as well as potential confounders. Mena^calc retained its prognostic value such that patients in the highest quartile had a 60% increase in risk of breast cancer death compared to those in the lowest three quartiles (hazard ratio, HR = 1.597; 95% CI = 1.20 to 2.13; P = 0.0015, Table 2). The linear trend in risk across Mena^calc was statistically significant: HR = 1.211 (1.08 to 1.36), P = 0.00164. In addition to Mena^calc, age, nodal status, nuclear grade and tumor size were significantly associated with risk of death from breast cancer (Table 2 and Additional file 1, Table S1).