Quantifying Argonaute 2 (Ago2) expression to stratify breast cancer

Human breast cancer cell lines (MCF10A, T47D, BT-474, SK-BR3, BT-20) were cultured at 37 °C and 5% CO2, in media as directed by the American Type Culture Collection repository (ATCC). All cell lines were originally purchased from ATCC.

Whole cell lysates were prepared using RIPA buffer (Sigma) containing a protease inhibitor cocktail (Roche). Whole cell lysates were re-suspended in reducing sample buffer, boiled and separated by SDS-PAGE. Proteins were transferred to nitrocellulose membranes (Life Technologies) by standard procedures. Indicated proteins detected using anti-Ago2 antibody (ab57113, Abcam) and anti-Actin by (A2066-Sigma).

Cells were prepared and stained as using standard techniques. Briefly, cells were fixed in 4% paraformaldehyde followed by permeabilisation in 0.05% Triton-X-100 and then blocked in PBS/1% BSA. Cells were incubated with anti-Ago2 (ab57113, Abcam) for 2 h at 37 °C, washed and incubated with Alexa Fluor® 594-conjugated anti-mouse secondary antibody and counterstained with 1R,2R- diaminocyclohexane(trans-diacetato) (dichloro) platinum(IV) (DAPI) (blue). Images were acquired with using an Olympus BX61Research System Microscope, with F-View® Imaging System. Olympus CellSens™ Dimension software was used for digital image capture and analysis. Z-stack images (10–15 layers) 0.2um apart were captured and merged using Maximum intensity projection. All images were captured using the same sensitivity and exposure times across cell lines.

Cell lines for immunohistochemical staining were fixed using − 20 °C Methanol. Cells were blocked in 2.5% Normal Horse Serum and probed using Ago2 (ab57113, Abcam) primary antibody. Cells were then stained using the Anti-Mouse Ig ImmPRESS™ Excel Staining Kit (Vector Laboratories) used as per manufacturers instructions. Haematoxylin was used as a counterstain and cells fixed in DPX (Sigma-Aldrich). The TMA was stained for Ago2 using the ImmPACT/ImmPRESS DAB Immunohistochemistry as per manufacturers instructions.

Clinical breast tissue samples comprised core biopsies, wide local excisions and mastectomy specimens received by the Galway University Hospital pathology department (1999–2005) which were used to construct a consecutive tissue microarray, based on breast cancer diagnosis and availability of biopsy tissue in the paraffin block. A core (0.6 mm diameter) of formalin-fixed paraffin-embedded tissue was used to construct the TMA, as previously described [9‐12]. Tumour cells in each section were confirmed by haematoxylin and eosin staining by a clinical pathologist. Pathological data was collected from the clinical pathology reports. Images of the Ago2 stained sections were captured using an Olympus VS120 Digital Scanner with a 40x objective and images processed using OlyVIA software (v2.8).

This study group consists of consecutively collected breast cancer patients treated at a tertiary referral unit (Galway University Hospital) entered into a prospectively maintained database (1999–2005). Only patients with a definitive subtype were included. Multiple clinical-pathological details were selected as indicated and used for further analysis. Tumours were staged according to the International Union against Cancer’s Tumour-Node-Metastasis (TNM) classification and histologically subtyped according to WHO guidelines. A total of 328 patients had Ago2 staining results with matched complete clinical information, including survival and outcome data for 327 patients. Table 1 describes the collected clinicopathological characteristics of the cohort.

Breast cancer molecular subtypes were defined based using standard accepted markers: Luminal A [ER and/or PR positive, HER2 negative, Ki-67 low (< 20%)]; Luminal B [ER and/or PR positive, HER2 positive or [ER and/or PR positive, HER2 negative, Ki-67 high (> 20%)]; HER2-overexpressing (ER and PR negative, HER2 positive); Triple negative (ER, PR and HER2 negative). The HER2 receptor status was identified by immunohistochemistry with any inconclusive results confirmed using a FISH test.

The scoring system developed (in collaboration with clinical pathologists) was based on the observed pattern of Ago2 cytoplasmic staining (in control and breast cancer specimens), with staining intensities placed in categorises (Negative, Weak, Moderate or Strong) (See Fig. 1a and Additional file 1: Figure S1). TMA images (Ago2 stained) were scored in an independent blinded analysis (by two independent researchers, with further validation performed by a trained practicing clinical pathologist). Independent analysis of the TMA scoring was performed by the study biostatisticians.

TaqMan assays were used as per manufacturers instructions for the relative quantification PCR (RQ-PCR) of our target gene Ago2 (Hs01085579_m1), and the endogenous control genes used were Mitochondrial Ribosomal Protein L19 (MRPL19) (Hs00608519_m1) and Peptidylprolyl Isomerase A (PPIA) (Hs99999904_m1) [13‐15]. Assays were performed using an AB7900HT (Applied Biosystems), using standard conditions as per the manufactures instructions. mRNA expression levels were normalized using endogenous controls PPIA and MRPL-19. Raw fluorescence data from RQ-PCR was exported into the software package qBase plus and relative quantification performed.

The generated data (Ago2 TMA staining, clinicopathological and survival data) were statistically analysed, associations between categorical variables were evaluated by applying Cochran-Mantel-Haenszel tests (as appropriate), measured via Goodman and Kruskal’s Lambda statistic and Cramer’s V statistic. ANOVA or non-parametric alternative Kruskal-Wallis H tests were applied to test for differences in continuous variables by categories of staining. S series of Mann-Whitney tests were used with Bonferroni correction to control Type I error rate to calculate the effect size of significant differences. Significance was reached if p ≤ 0.05. To assess the relationship between Ago2 staining and the indicated clinicopathological variables for Overall Survival response and Disease Free Survival response Complete-Case Cox PH regression models were used. The R statistical program, and packages, were used for analysis.

As previously described [35], the prognostic value of Ago2 mRNA expression was investigated using Breast Cancer database of The Kaplan-Meier Plotter (incorporating gene expression and clinical data) (http://​kmplot.​com/​analysis/​) [16‐19]. At use, data on 5143 breast cancer patients was available (December 2017). Specific parameters used as previously described [35]: Gene: EIF2C2 (Search: AGO2) (Affy ID: 225827_at); Split patients by: median; Survival: Disease Free Survival (DFS n = 1402) or Overall Survival (OS, n = 626); Follow up threshold: all; Censure at threshold, selected; Use only JetSet best probe set, selected; Quality control, Remove redundant samples: checked; Array quality control: exclude biased arrays selected; Intrinsic subtype, selected as indicated.

As previously described [35], somatic mutations and copy number changes to the Ago2 gene were investigated using the cBioPortal database [The Cancer Genome Atlas, Breast Cancer dataset] (http://​www.​cbioportal.​org/​) [20‐24]. Specific Search parameters (as previously described [35]): Select Studies: Breast Cancer; Studies Select all: As of December 2017 9 studies available [Adenoid Cystic Carcinoma of the Breast (MSKCC, J Pathol. 2015; 12 samples), Breast Cancer (METABRIC, Nature 2012 & Nat Commun 2016; 2509 samples), Breast Invasive Carcinoma (British Columbia, Nature 2012; 65 samples), Breast Invasive Carcinoma (Broad, Nature 2012; 103 samples), Breast Invasive Carcinoma (Sanger, Nature 2012; 100 samples), Breast Invasive Carcinoma (TCGA, Provisional; 1105 samples), Breast cancer patient xenografts (British Columbia, Nature 2014; 29 samples), Mutational profiles of metastatic breast cancer (France, 2016; 216 samples), The Metastatic Breast Cancer Project (Provisional, October 2017; 103 samples)]; Select Data Type Priority: Mutation and CAN selected; Enter Gene Set: Ago2. Results returned following query submission were: copy number variation (CNV) and genomic alterations (including somatic mutations).

Ethical approval granted by University College Hospital Galway and National University of Ireland Galway (approvals C.A.151, C.A.1012). All patients included in the study had breast cancer confirmed histologically. A prospectively maintained breast cancer database provided the clinicopathological data. The Galway University Hospitals Clinical Research Ethics Committee approved use of patient material (2008 meeting; approval C.A.151; 2014 approval C.A.1012).

The produced datasets are available, from the corresponding author, on reasonable request.

To investigate Ago2 expression in breast cancer, cell lines representing the four common molecular subtypes: T47D (Luminal A), BT-474 (Luminal B), SK-BR-3 (HER2 positive), BT-20 (Basal/Triple negative), and one non-cancer non-tumourigenic control breast epithelial cell line (MCF-10A) were profiled. The cell lines were assayed for Ago2 mRNA expression (Additional file 1: Figure S1A). T47D (Luminal A) was found to express Ago2 mRNA at significantly higher levels than either the SK-BR-3 (HER2 positive) or BT-20 (Triple Negative) cell lines (Anova p = 0.005). To further characterize Ago2 expression, the levels of Ago2 protein in a panel of subtype representative breast cancer cell lines was profiled (Fig. 1a). In agreement with the mRNA levels, the BT-20 line displayed distinctly lower total Ago2 protein expression than the other cell lines (MCF-10A, T47D, BT-474, SK-BR-3). Interestingly, only the SKBR-3 cell line displayed disagreement between the low mRNA and high protein expression levels indicating the presence of a highly stable Ago2 protein in these cells. In addition to profiling the total levels of Ago2, localisation of Ago2 protein in the breast cancer cell lines was explored. According to the Human Protein Atlas Ago2 localises predominantly to the nucleus and cell-cell junctions in 3 distinct cancer types, with breast cancer localisation unknown [9, 25‐27]. Investigating Ago2 protein intensity and localisation using IHC in breast cancer cell lines, overall all breast cancer cell lines displayed stronger Ago2 staining, compared to MCF-10A (with very weak staining) (Fig. 1b, Additional file 1: Figure S1B). The Luminal cell lines (T47-D and BT-474) displayed predominantly cytoplasmic Ago2 staining, of low-moderate intensity. The SK-BR3 (HER2 positive) and BT-20 (Triple Negative) cell lines displayed stronger cytoplasmic Ago2 staining (Fig. 1b). Intriguingly, the BT-474, SK-BR-3 and BT-20 lines displayed a sub-population with strong nuclear staining Fig. 1b). Ago2 localisation was explored further using the more sensitive immunofluorescent staining method. In non-dividing cells Ago2 was observed to stain almost exclusively in the cytoplasm in the control MCF10A cells. In non-dividing breast cancer lines there was strong cytoplasmic staining, with additional nuclear staining observed (Additional file 1: Figure S2). Surprisingly, we observed Ago2 localizing to the midbody of dividing cells in MCF10A, T47D, BT-474 and SK-BR-3 (but not in the basal line BT-20: images shown representative of 20 observed mitotics for each cell line), a localisation not previously observed before (Fig. 1c).

To further characterize Ago2 localisation in breast cancer, a tissue microarray (TMA) consisting of 328 breast cancer biopsies (Table 1) was stained for Ago2. The intensity of the Ago2 staining observed was categorised (Negative, Weak, Moderate, Strong) (Fig. 2a, Additional file 1: Figure S3). Stratifying the Ago2 stained sections by subtype, we find that the TMA loosely follows the distribution of subtypes observed clinically: Luminal A (52.7%), Luminal B (12.8%), Her2 positive (8%), with a slight overrepresentation of the Basal/Triple Negative (26.5%) subtype (Fig. 2b). Investigating the samples with both subtype and Ago2 staining (n = 226), the predominant staining intensity observed was Weak (33.2%, n = 75), followed closely by Negative (31%, n = 70), Moderate (19.5%, n = 44) and Strong (16.4%, n = 37) (Fig. 2c). As observed in the breast cancer cell lines, the Ago2 staining pattern observed in the TMA was mainly cytoplasmic (of varying intensities), with nuclear staining observed within the strong staining pattern (Additional file 1: Figure S3, Strong).

To explore the relevance of Ago2 staining as a prognostic or diagnostic marker, associations between Ago2 intensity and key clinicopathological factors were tested. Testing the association between Ago2 staining intensity and breast cancer subtype, there is evidence to suggest a general association in the population (p ≈ 0.000) (Fig. 3a-b). Importantly, Goodman and Kruskal’s Lambda (λ) measure for association (a measure of association between two nominal categorical variables) suggests knowledge of Ago2 staining intensity improves the ability to predict breast cancer subtype by 20% (probability distribution analysis of observed and expected counts, determining key variables contributing more to subtype prediction). Testing the association between Ago2 staining intensity and breast cancer receptor status, there is evidence to suggest a general association between the Estrogen Receptor (ER) and Ago2 staining in the population (p ≈ 0.000). Furthermore using the λ measure for association knowledge of Ago2 staining intensity improves the ability to predict ER status by 15.7% (Fig. 3c). There is evidence to suggest a general association between the Progesterone Receptor (PR) and Ago2 staining in the population (p ≈ 0.000). In addition, the λ measure for association suggests knowledge of Ago2 staining improves the ability to predict PR status by 17.5% (Fig. 2d). Interestingly, a higher proportion of ER positive samples with weak or negative staining were observed [combined 73.6% (65 + 80)/197] (Fig. 3c), with a similar trend seen in PR positive samples [combined 74.2% (63 + 72)/182] (Fig. 3d). Intriguingly, if we consider only the ER or PR negative groups (Fig. 3c-d), the average of the Ago2 staining intensities would be relatively similar, averaging 8.85% (±1.3%) and 9.93% (±0.87%) respectively, suggesting a possible link between ER/PR positivity and Ago2 regulation.

Exploring the relationship between Ago2 staining intensity and clinicopathological variables, the Kruskal-Wallis test (a rank-based nonparametric test used to determine any statistically significant differences between two or more populations) suggests a significant difference in the tumour size distribution, when comparing Ago2 staining intensities (p = 0.0267) (Fig. 4a). The observed data gave no evidence of an association between Ago2 staining intensity with the other commonly tested clinicopathological variables (T-score, N-score, M-score, NPI, DCIS, Her2, Stage or age) (Data not shown). There is no evidence of an effect of Ago2 staining intensity on Overall Survival (OS) outcome, comparing the different Ago2 staining patterns in the population (p = 0.7) (Fig. 4b). Additionally, there is no evidence of an effect of Ago2 staining intensity on Disease-Free Survival (DFS) (Data not shown).

To further analyze the relationship between Ago2 staining intensity and the common clinicopathological variables two types of saturated models (with are as many estimated parameters as data points) were utilized: Complete-case and Imputed-dataset. Complete-case uses and analyzes only data from patients with complete covariate data set, which can be considered inefficient as not all data in the cohort is utilized. However, the Imputed-dataset modeling can incorporate data from all cohort patients (featuring a predictive model based on observed data to estimate missing covariate values). Investigating the fitting of complete-case saturated and Imputed-dataset saturated models for DFS, the models fitted 177 (of 308) complete-case individuals across the modelled variables [Ago2 staining intensity, Subtype, Nottingham Prognostic Index (NPI), Tumour size, disease stage, patient age] (Additional file 1: Figure S4A and B). As expected in this model Basal Subtype and Stage III show a significant difference in DFS outcome (p = 0.02 and 0.02 respectively). In these models comparing different Ago2 staining patterns in the population found no significance.

It was previously reported that Ago2 protein and mRNA levels were differentially regulated in melanoma [32]. Therefore, the expression levels of Ago2 mRNA in breast cancer were investigated [16, 23, 28‐32]. Ago2 mRNA expression levels were correlated this with key clinicopathological characteristics DFS and OS [16, 33, 34]. Investigating Ago2 mRNA expression and DFS, we find evidence of a difference in DFS comparing Ago2 expression Low and High, when considering all breast cancers (p = 0.0012, n = 1764) (Fig. 5a). Investigating individual subtypes, no evidence of a difference in DFS comparing Ago2 expression Low and High was found [Luminal A, p = 0.22; Luminal B, p = 0.94; Her2 positive, p = 0.73; Basal, p = 0.14] (Fig. 5b-e). Comparing the distribution of OS between High and Low Ago2 expression, considering all breast cancers (including all breast cancer subtypes, n = 626) there is no evidence of a difference (p = 0.5608) (Additional file 1: Figuere S5A). Furthermore, there was no evidence of a difference in OS between High and Low Ago2 expression within each individual subtype [Luminal A, p = 0.5941; Luminal B, p = 0.8123; Her2 positive, p = 0.2194; Basal, p = 0.4965] (Additional file 1: Figure S5B-E).

Previous work found that increased Ago2 mRNA expression was associated with increased AGO2 gene copy number, and this was significantly associated with a poorer disease outcome in high-risk multiple myeloma [36]. Using the cBioPortal [20, 29] we investigated copy number variations (CNVs) of the Ago2 gene in breast cancer. Amplification of the Ago2 gene was found in 18.25% of cases (650/3562 samples), in data from seven independent breast databases available (with more than 100 samples), ranging between ~ 27–10% (in individual studies, where amplification was observed) (Additional file 1: Figure S6A). Importantly, no significant Ago2 gene deletion was observed in this breast cancer cohort (4/3562 samples, 0.11%). Interestingly, a very low Somatic Mutation Frequency (missense) of 0.31% (n = 11) was observed (Additional file 1: Figure S6B).