Purpose: Molecular evolution of tumors during progression, therapy, and metastasis is a major clinical challenge and the main reason for resistance to therapy. We hypothesized that microRNAs (miRNAs) that exhibit similar variation of expression through the course of disease in several patients have a significant function in the tumorigenic process.

Experimental design: Exploration of evolving disease by profiling 800 miRNA expression from serial samples of individual breast cancer patients at several time points: pretreatment, posttreatment, lymph nodes, and recurrence sites when available (58 unique samples from 19 patients). Using a dynamic approach for analysis, we identified expression modulation patterns and classified varying miRNAs into one of the eight possible temporal expression patterns.

Results: The various patterns were found to be associated with different tumorigenic pathways. The dominant pattern identified an miRNA set that significantly differentiated between disease stages, and its pattern in each patient was also associated with response to therapy. These miRNAs were related to tumor proliferation and to the cell-cycle pathway, and their mRNA targets showed anticorrelated expression. Interestingly, the level of these miRNAs was lowest in matched recurrent samples from distant metastasis, indicating a gradual increase in proliferative potential through the course of disease. Finally, the average expression level of these miRNAs in the pretreatment biopsy was significantly different comparing patients experiencing recurrence to recurrence-free patients.

Conclusions: Serial tumor sampling combined with analysis of temporal expression patterns enabled to pinpoint significant signatures characterizing breast cancer progression, associated with response to therapy and with risk of recurrence. Clin Cancer Res; 22(14); 3651–62. ©2016 AACR.

Translational Relevance

Delineating the evolutionary molecular path of breast cancer in each patient is challenging, as tumors are rarely diagnosed at their initial stages of development. Alternatively, to investigate how individual breast cancer disease evolves over time, we can analyze the molecular changes in tumors at different stages of progression for the same patient. In this study, we profiled miRNA expression of serial samples that represent the disease progression stages of breast cancer patients, from diagnosis through therapy until recurrence. Acquisition of such a dataset, accompanied by clinical data, has the potential for the development of predictive and prognostic signatures.

One of the main difficulties hindering advances in breast cancer treatment is the dynamic molecular evolution of tumors across the course of disease. Tumors may evolve in time during tumor progression and therapy, as well as in space, across different tumor clones and within metastases (1). The dynamics of tumor evolution is individual for each patient, shaped by intrinsic genetic factors along with extrinsic selective forces, such as cancer therapy. Most comprehensive studies, aimed at profiling the molecular players involved in breast cancer, were based on the primary tumor sample. However, the molecular evolution toward recurrence is key to understanding the processes leading to metastases.

Ideally, to understand tumor evolution at the individual patient level, analysis of molecular changes in tumors at different progression stages for the same patient should be performed. This approach of serial assessments requires long-term maintenance of clinical databases and availability of biologic samples along this timeline. Neoadjuvant therapy (preoperative therapy) is an ideal setting for this purpose as we can compare samples at diagnosis (prior to therapy) and at surgery (following therapy). Neoadjuvant systemic therapy is currently an accepted standard approach for women with locally advanced breast cancer and has been shown to reduce tumor size and allow for breast-conserving surgery. Neoadjuvant therapy can provide prognostic information and may also provide in vivo assessment of tumor sensitivity to therapy (2, 3). While some patients exhibit pathologic complete response (pCR) with complete disappearance of the tumor, others respond only partially, or do not respond at all (4). Although pCR is an important prognostic factor, the risk of recurrence is not always in full concordance with it. The physician predicts the risk of recurrence based solely on statistics, taking into account the pathologic response, with no considerations of molecular modulations. The treatment stage is considered as a “black box” for each individual patient, and our knowledge and understanding of the molecular processes and changes that occur in cancer cells in response to treatment is very limited.

As longitudinal sampling can span several years over which preservation of archived tissues may vary, it is important to profile stable molecules, such as miRNAs, to obtain reliable information. miRNAs are a highly conserved group of short noncoding RNAs (∼22 bp) that play a major role in posttranslational regulation of gene expression and are critical regulators of oncogenic pathways (5). miRNAs have been widely associated with breast cancer tumorigenesis (6), as well as with prognosis and response to treatment in breast cancer (7, 8). A recent study by Dvinge and colleagues (9) describes the landscape of miRNA expression in 1,302 primary breast tumors accompanied by matched clinical information and genomic and transcriptomic data (10). This important dataset provides comprehensive information on the function of miRNA in breast cancer regulation. Nevertheless, analyzing miRNA modulations from serial samples prior to therapy, following therapy, and at recurrence may highlight important molecular mechanisms that may be underestimated when analyzing only primary tumors.

In this study, we applied a longitudinal approach to identify expression signatures characterizing individual breast cancer progression. The analysis approach was to classify temporal modulation patterns observed through the course of disease. We hypothesized that expression modulations in individual patients throughout the course of their disease can directly identify miRNA sets that are differentially expressed at the various stages of the disease and are associated with disease progression or response to therapy and may have prognostic value.

Tumor specimen collection

A prospectively maintained database that contains clinical data on ∼600 cases who have undergone neoadjuvant treatment at the Sheba Medical Center from 2003 was screened. Importantly, all patients received a uniform treatment: four cycles of dose-dense doxorubicin (60 mg/m2) and cyclophosphamide (600 mg/m2) every 2 weeks followed by dose-dense paclitaxel (175/m2) two weekly for four cycles or 12 weekly cycles of paclitaxel (80/m2). The database was analyzed to select a cohort of patients having maximal available formalin-fixed paraffin-embedded (FFPE) samples throughout the course of disease at the pathologic archive [tumor pretreatment (C), lymph node pretreatment (CL), tumor posttreatment (T), lymph node posttreatment (TL), and recurrence (R); Fig. 1A]. The main cohort consists of 10 patients with disease recurrence; of them, 8 patients had samples from all 3 time points (C, T, and R), and 2 patients showed pCR (C and R only; Fig. 1B). A second independent cohort consists of nine patients who remained disease free (C and T samples), with median follow-up of 10 years (Supplementary Fig. S1). Lymph node samples for both cohorts were analyzed when available. For control normal breast, we utilized tumor-adjacent normal epithelium for four patients as well as specimens from breast reduction (n = 5). This study was approved by the Institutional Review Board (IRB).

Figure 1.

Identifying temporal miRNA expression patterns through the course of disease. A, the longitudinal collection of samples: from each individual patient, core biopsy (C), and lymph nodes (CL) are collected at diagnosis before the neoadjuvant treatment; tumor sample (T) and lymph node (TL) are collected at surgery after treatment; for patients with recurrent disease (local or distant metastasis), recurrent tissue (R) is also collected. B, the types of samples that were collected for each patient; when present, the box is colored as in A. C, 8 possible varying expression patterns (P1–P8) across the three selected time points: C, T, and R. The number of miRNAs that were classified to each pattern is indicated. D, a flow chart for assigning an miRNA to a pattern (actual data for miR-125b-5p is shown as an example): first, linear mixed-effect model is used to select all miRNAs that are differentially modulated over time. Then, a corresponding pattern is assigned for each patient using a threshold of 1.5-fold change. Next, if the miRNA expression follows the same pattern in three patients or more, miRNA is assigned to this pattern (see Supplementary Methods for miRNAs that are assigned to more than one pattern).

Figure 1.

Identifying temporal miRNA expression patterns through the course of disease. A, the longitudinal collection of samples: from each individual patient, core biopsy (C), and lymph nodes (CL) are collected at diagnosis before the neoadjuvant treatment; tumor sample (T) and lymph node (TL) are collected at surgery after treatment; for patients with recurrent disease (local or distant metastasis), recurrent tissue (R) is also collected. B, the types of samples that were collected for each patient; when present, the box is colored as in A. C, 8 possible varying expression patterns (P1–P8) across the three selected time points: C, T, and R. The number of miRNAs that were classified to each pattern is indicated. D, a flow chart for assigning an miRNA to a pattern (actual data for miR-125b-5p is shown as an example): first, linear mixed-effect model is used to select all miRNAs that are differentially modulated over time. Then, a corresponding pattern is assigned for each patient using a threshold of 1.5-fold change. Next, if the miRNA expression follows the same pattern in three patients or more, miRNA is assigned to this pattern (see Supplementary Methods for miRNAs that are assigned to more than one pattern).

Close modal

Tissue processing and histology

All specimens were stored as FFPE blocks at the Sheba Pathology Institute archive. Tissue slides were examined by expert breast pathologists to include a minimum of 70% cancer cells; otherwise these were macro-dissected to eliminate contamination of stromal components. The extent of pathologic response was evaluated by a pathologist, examining the residual tumor burden based on reduction in tumor cellularity, similar to Ogston and colleagues (Miller and Payne grade; ref.11). In addition, tumor and lymph node slides were evaluated to calculate the Residual Cancer Burden (RCB) class, as defined by Symmans and colleagues (ref.12; Supplementary Table S1).

Each tumor sample was sectioned, and total RNA was extracted using nucleic acid isolation kit (AllPrep Qiagen).

Immunohistochemistry

Slides were immunostained for Ki67 on the Ventana Discovery autostainer (Ventana) using commercial Ki67 antibody (clone SP6; Thermo Scientific). Ki67 levels were assessed blinded to the clinical data using the automated Virtuoso image analysis algorithm (Ventana). Several representative regions of the tumor bed were selected for counting a total of at least 1,000 tumor cells. The percentage of Ki67-positive cells to the total number of evaluated cells was calculated.

miRNA expression profiling

miRNA expression was assayed by NanoString nCounter digital RNA transcript counting that assays 800 miRNAs. Fifty-eight unique samples were profiled for 19 patients. In order to estimate the technical and biologic noise, we profiled three technical and three biological repeats (consecutive sections from the same tumor sample).

Validation of the microarray results by qRT-PCR

The results were validated by qRT-PCR for 21 candidate miRNAs. We found a very good agreement between the nCounter analysis and the qRT-PCR (Supplementary Fig. S2). cDNA generated from purified total RNA (100 ng) by the miScript Reverse Transcription Kit was quantified by qRT-PCR using the miScript SYBR Green PCR Kit (Qiagen) and commercial primers (Qiagen). Reactions were performed in duplicates. RNU6 was used for normalizations. The relative expression levels were quantified using the 2−DDCt method.

qRT-PCR of targeted genes

Primers for BUB1B and CDC25A were designed using Primer Express and were commercially synthesized (Biosearch). All genes were normalized by three endogenous control genes: GAPDH, TBP, and HPRT1. cDNA was synthesized from total RNA by the qSCript cDNA Synthesis Kit (Quanta Biosciences). To receive reliable results of mRNA from FFPE samples preserved across several years, we amplified the cDNA before the qPCR reactions using PerfeCTa PreAmp SuperMix (Quanta; see Supplementary Methods).

Data analysis

See Supplementary Methods for full description; in brief, data were normalized using a Lowess multi-array algorithm (13). A linear mixed-effect model was used to identify miRNAs significantly different across the course of disease. These miRNAs were classified to eight varying expression patterns (Fig. 1C) between C, T and R samples based on two thresholds: (i) call increase/decrease for fold change ≥ 1.5 and (ii) pattern observed in at least 3 patients (Fig. 1D). Differential expression of the identified miRNAs in tumors versus normal samples or high-grade versus low-grade tumors was calculated on the METABRIC dataset (9). The CoSMic algorithm (14) with the METABRIC miRNA and mRNA expression data (9, 10) were used to identify targets for our 14 P1 candidate miRNAs. We applied the CoSMic algorithm with miRanda (15), TargetScan (16), or PITA (17) sequence-based prediction algorithms.

Pathway analysis was performed using the Pathifier (18) algorithm, which infers a pathway deregulation score for each tumor sample and pathway, on the basis of mRNA expression data. We correlated the expression levels of our candidate miRNAs with the pathway deregulation scores of all KEGG pathways (19), calculated using Pathifier on the METABRIC dataset for samples with both mRNA and miRNA expression data (10). Gene Set Enrichment Analysis (GSEA; ref.20) was used to calculate the enrichment of an epithelial-to-mesenchymal (EMT) core list (21) in genes that are positively correlated with P5 miRNA expression levels in the METABRIC dataset.

Identifying temporal miRNA expression patterns through the course of disease

We assembled a unique cohort of patients with all available matched samples through their course of disease: tumor sample before therapy, residual tumor after therapy, lymph nodes and recurrent tumors, to depict molecular modulations for each individual patient (a total of 58 unique samples from 19 patients; Fig. 1A–B and Supplementary Fig. S1). Clinical information is summarized in Table 1. 

Table 1.

Clinical and pathologic characteristics of the cohorts

Cohort 1 (recurrent disease) N = 10Cohort 2 (recurrence-free disease) N = 9
Mean age (range) years 53 (40–69) 51 (40–67) 
Subtype (N, %)   
 Hormone positive 5 (50) 6 (67) 
 Triple negative 2 (20)  
 HER2+/hormone negative 1 (10)  
 HER2+/hormone positive 2 (20) 3 (33) 
Treatment regimen (N, %)   
 AC-T 9 (90) 9 (100) 
 AC-TH 1 (10)  
Pathologic response (N, %)   
 CR 2 (20)  
 PR 7 (70) 8 (89) 
 SD 1 (10) 1 (11) 
Median RFS (range) years 2.5 (1–6) NAa 
Cohort 1 (recurrent disease) N = 10Cohort 2 (recurrence-free disease) N = 9
Mean age (range) years 53 (40–69) 51 (40–67) 
Subtype (N, %)   
 Hormone positive 5 (50) 6 (67) 
 Triple negative 2 (20)  
 HER2+/hormone negative 1 (10)  
 HER2+/hormone positive 2 (20) 3 (33) 
Treatment regimen (N, %)   
 AC-T 9 (90) 9 (100) 
 AC-TH 1 (10)  
Pathologic response (N, %)   
 CR 2 (20)  
 PR 7 (70) 8 (89) 
 SD 1 (10) 1 (11) 
Median RFS (range) years 2.5 (1–6) NAa 

Abbreviations: AC-T, doxorubicin (Adriamycin), cyclophosphamide and paclitaxel (Taxol); AC-TH, as AC-T with trastuzumab (Herceptin); CR, pathologic complete response; PR, pathologic partial response; SD, stable disease; RFS, recurrence-free survival.

aMedian follow-up time, 11 (10–12) years.

We hypothesized that expression modulations in individual patients through their course of disease and therapy can identify miRNAs that play a dynamic role in cancer progression. We therefore searched for miRNAs that exhibit modulation patterns that are shared between patients. There are eight possible patterns that describe expression modulations between three time points across the course of disease: pretreatment (C), posttreatment (T), and recurrence (R; Fig. 1C). To classify a particular miRNA to one of these patterns, we first filtered all miRNAs using the linear mixed-effect model to select only miRNAs that are significantly modulated over time. We then applied a threshold of 1.5-fold to define expression modulation between pairs of time points and considered only patterns that were observed in at least 3 patients (see Materials and Methods). For example, miR-125b exhibits pattern P1 in four of eight patients (having C, T, R samples), and therefore it was assigned to pattern P1 (Fig. 1D).

The miRNA sets classified for each pattern are presented in Fig. 2. As can be seen from the patterns table and Fig. 1C, the number of miRNAs classified to each pattern varied considerably. We focused on the main four patterns (P1, P2, P5, and P6). To investigate the potential involvement of these miRNAs in tumor progression, we examined their expression in an available dataset of miRNA expression in breast cancer (METABRIC; ref. 9). For example, the most abundant pattern, P1, is mainly composed of miRNAs that are downregulated in breast tumors relative to normal samples according to the METABRIC dataset, while in pattern P2, which is the inverse complementary pattern to P1, most miRNAs (three of four) are upregulated in breast tumors (Fig. 2). Furthermore, we performed a detailed and comprehensive search in the literature for evidence of the role of these miRNAs in breast cancer (Supplementary Table S2). Finally, pathway analysis was used to map each miRNA to the most correlated biologic pathway in breast cancer (refs.18, 19; see Materials and Methods). The entire information is summarized in Fig. 2 and Supplementary Table S3.

Figure 2.

List of miRNAs that were classified to each pattern. Results obtained from the METABRIC dataset: red/green arrows indicate whether the miRNA is upregulated/downregulated (FDR = 1%) in tumor versus normal tissue (column 3) and in grade 3 relative to grade 1 tumors (column 4). miRNAs that are not expressed in the METABRIC dataset are indicated by gray entries. Evidence from the literature: whether the miRNA is upregulated (red arrow) or downregulated (green arrow) in breast tumors versus normal samples (column 5), and whether its upregulation or downregulation is associated with metastasis (column 6) or resistance to therapy (column 7). Columns 8–17 indicate whether miRNA expression levels in the METABRIC dataset are significantly correlated (1% FDR) with the pathway deregulation scores of the corresponding KEGG pathways (see the text for details). Green/red entries represent negative/positive correlation. Columns 18–25 indicate for each miRNA whether it is expressed, in each patient, according to its assigned pattern (black entries if consistent with fold-change threshold of 1.5, gray for 1.4).

Figure 2.

List of miRNAs that were classified to each pattern. Results obtained from the METABRIC dataset: red/green arrows indicate whether the miRNA is upregulated/downregulated (FDR = 1%) in tumor versus normal tissue (column 3) and in grade 3 relative to grade 1 tumors (column 4). miRNAs that are not expressed in the METABRIC dataset are indicated by gray entries. Evidence from the literature: whether the miRNA is upregulated (red arrow) or downregulated (green arrow) in breast tumors versus normal samples (column 5), and whether its upregulation or downregulation is associated with metastasis (column 6) or resistance to therapy (column 7). Columns 8–17 indicate whether miRNA expression levels in the METABRIC dataset are significantly correlated (1% FDR) with the pathway deregulation scores of the corresponding KEGG pathways (see the text for details). Green/red entries represent negative/positive correlation. Columns 18–25 indicate for each miRNA whether it is expressed, in each patient, according to its assigned pattern (black entries if consistent with fold-change threshold of 1.5, gray for 1.4).

Close modal

miRNAs in pattern P1 are correlated with response to treatment

Notably, our method of pattern classification (shared by a minimum of three patients) was applied to a heterogeneous cohort of patients with distinct subtypes and diverse courses of disease. Therefore, it is reasonable that distinct groups of miRNAs can be assigned to the same pattern. Indeed, examining miRNAs in pattern P1 across all patients revealed that pattern P1 is composed of two distinct groups (Fig. 2). Most of the miRNAs assigned to P1 (upper P1 panel: 14 miRNAs) followed this pattern mainly in a subset of patients (patients 3, 4, 5, and 9), while few other miRNAs followed pattern P1 in other patients (lower P1 panel; patients 2, 7, 8 and 10). We noted that the majority of miRNAs in the first P1 group are downregulated in breast cancer according to the literature and the METABRIC dataset (Fig. 2). We therefore redefined P1 miRNAs to include only this subset of 14 miRNAs as a more coherent group that exhibit the P1 pattern across the same group of patients and are downregulated in breast tumors.

Next, we calculated the average expression levels of P1 miRNAs in each sample (defined as P1 score) and presented it for each patient individually (Fig. 3A). Examining P1 scores of each patient across the course of disease (Fig. 3B) revealed that P1 miRNAs show distinct patterns in different patients. We suspected that this can be associated with the extent of response to therapy. Therefore, we independently scored the level of pathologic response to treatment by examining the reduction in tumor cellularity for each patient, by a blinded pathologic examination (Miller and Payne—MP grade; ref.11; Supplementary Table S1). Surprisingly, there was full agreement between the expression pattern of P1 miRNAs and the pathologic response to therapy (Fig. 3B). Patients with moderate response to treatment (MP grade = 3) showed an increase in P1 miRNAs posttreatment, as opposed to patients with minor (MP = 2) or no response (MP = 1). The log ratios of P1 between C and T were significantly different among the various response rates (ANOVA test, P = 0.011), suggesting an association of P1 score with response to treatment.

Figure 3.

miRNAs in pattern P1 are correlated with response to treatment and disease progression. A, expression matrices of the subset of 14 miRNAs from P1 (see text), presented for each patient across all its available samples. Colors indicate expression levels after centering and normalizing each miRNA (row) across all patients, with red denoting relatively high expression and blue relatively low expression (see color bar at the bottom right). The mean expression level of the miRNAs (“P1 score”) is presented below each matrix. In all matrices, miRNAs are ordered as shown for patient 2. B, log ratios of P1 scores, relative to core biopsy (C), are plotted for each patient. Pathologic response to treatment was scored by Miller and Payne grade (MP) and is denoted by colors (see legend). Log ratios of P1 between C and T are significantly different among various MP grades (ANOVA test, P = 0.011). C, box plot of P1 scores in core biopsies pretreatment (C), tumor samples posttreatment (T), recurrent tissues (R), lymph node before treatment (CL), and lymph node after treatment (TL). Each box contains P1 scores of patients for which the corresponding information is available. *, P < 0.05; **, P < 0.01.

Figure 3.

miRNAs in pattern P1 are correlated with response to treatment and disease progression. A, expression matrices of the subset of 14 miRNAs from P1 (see text), presented for each patient across all its available samples. Colors indicate expression levels after centering and normalizing each miRNA (row) across all patients, with red denoting relatively high expression and blue relatively low expression (see color bar at the bottom right). The mean expression level of the miRNAs (“P1 score”) is presented below each matrix. In all matrices, miRNAs are ordered as shown for patient 2. B, log ratios of P1 scores, relative to core biopsy (C), are plotted for each patient. Pathologic response to treatment was scored by Miller and Payne grade (MP) and is denoted by colors (see legend). Log ratios of P1 between C and T are significantly different among various MP grades (ANOVA test, P = 0.011). C, box plot of P1 scores in core biopsies pretreatment (C), tumor samples posttreatment (T), recurrent tissues (R), lymph node before treatment (CL), and lymph node after treatment (TL). Each box contains P1 scores of patients for which the corresponding information is available. *, P < 0.05; **, P < 0.01.

Close modal

miRNAs in pattern P1 are differentially expressed across the various disease stages

Remarkably, for most patients the expression levels of P1 miRNAs were lower in the recurrence tissue than in the primary tumor at diagnosis (Fig. 3A–B). This led us to speculate that P1 miRNAs are indicative of disease progression. We compared P1 scores of all patients across the course of disease (Fig. 3C). Interestingly, there was a significant difference in the absolute values of P1 scores across the entire cohort. We found a significant difference between P1 scores of primary tumors before and after treatment and between posttreatment tumors and recurrent tissues (Fig. 3C; P = 0.019 and 1.1 × 10−5, respectively). Moreover, P1 scores at recurrence were significantly lower than at all other disease stages. All the recurrence samples had low P1 scores, independently of tumor subtype or metastatic niche. Thus, P1 miRNAs also differentiate between the stages of the disease and are associated with disease progression.

To find out to what extent these miRNAs return to “normal” expression levels posttherapy, we quantified the expression levels of two miRNAs assigned to pattern P1 (miR-125b-5p and miR-100-5p) in normal breast tissue, either tumor-adjacent normal epithelium or normal epithelium from healthy patients. Interestingly, although tumors exhibit higher posttreatment than pretreatment expression levels, the expression level of adjacent normal tissues posttreatment was higher (Supplementary Fig. S3). Importantly, expression levels of normal breast from healthy patients (breast reduction, n = 5), although varied, were similar to the levels of normal breast tissue posttreatment (no statistical difference between the expression of normal breast epithelium of healthy individuals versus normal epithelium posttreatment; Supplementary Fig. S3).

Next, we investigated P1 scores of involved axillary lymph nodes before and after treatment. We found that P1 scores were not significantly different between lymph nodes before and after treatment (P = 0.47; Fig. 3C), in contrast to the significant difference observed in the primary tumors before and after therapy. While there was no significant difference between primary tumors and lymph nodes before treatment (C vs. CL; P = 0.88), average expression levels were significantly lower in lymph nodes posttreatment compared with primary tumors posttreatment (T vs. TL; P = 0.04; Fig. 3C). Examination of individual patient's profile of miRNAs from pattern P1 across their entire samples (Fig. 3A) emphasizes these results, showing similar expression levels in the lymph nodes before and after treatment, suggesting reduced chemosensitivity in lymph nodes.

miRNAs in pattern P1 are associated with proliferation

As shown in Supplementary Table S2, the downregulation of P1 miRNAs is related to higher proliferation and invasion of breast cancer. Furthermore, by pathway analysis we found that P1 miRNAs are significantly negatively correlated to the cell-cycle pathway and to other proliferative pathways, such as the MAPK and JAK–STAT signaling pathways (1% FDR; see Materials and Methods and Fig. 2). We further quantified miR-10b* (not included in the NanoString panel), which is a known inhibitor of the cell cycle in breast cancer cells (22) by qRT-PCR; its expression modulation exhibited a strong P1 pattern in patients 3 and 9 (Supplementary Fig. S4).

Next, we searched for putative target genes, which are connected to the cell cycle and to proliferative pathways, and are targeted by several miRNAs from the P1 group. We used the CoSMic algorithm (14), which predicts context-specific target genes of an miRNA, utilizing sequence-based prediction scores as well as mRNA and miRNA expression data measured from the same samples (METABRIC). We identified several genes that are related to the cell cycle (8 genes), MAPK (10 genes), and JAK/STAT (6 genes) pathways and are predicted targets of several miRNAs from P1 (Supplementary Table S4). In addition, RAF1, which is an upstream activator of MAPK pathway, is a predicted target of miR-125b-5p (Supplementary Table S4).

We checked the expression levels of two candidates, BUB1B and CDC25A, known for their role in breast cancer progression (23, 24). BUB1B is a kinase essential for the mitotic checkpoint and required for normal mitosis and a putative target of let-7c, miR-199a-3p, miR-199a-5p, and miR-125b-5p. CDC25A is a phosphatase required for the progression from G1 to the S phase and is a putative target of miR-100-5p and miR-145-5p (Supplementary Fig. S5A). We quantified the mRNA levels of these genes by qPCR, specifically calibrated for quantification of mRNAs from archived samples. Expression levels of BUB1B and CDC25A were measured in several patients and were found to be inversely expressed to their predicted regulatory miRNAs (Fig. 4A–B). At recurrence, their levels were only partially inversed to miRNA levels (Supplementary Fig. S5B–C).

Figure 4.

miRNAs in pattern P1 are associated with proliferation and cell-cycle. A, expression levels of P1 miRNAs that target BUB1B—let-7c, miR-125b-5p, miR-199a-3p, and miR-199a-5p. Presented are the log-ratio expression levels (NanoString) of tumor posttreatment relative to core biopsy pretreatment in patients 9 (black), 5 (gray), and 2 (white); and for BUB1B in the same samples and patients as measured by qRT-PCR. B, same as A for miR-100-5p and miR-145-5p, which target CDC25A. C, representative images of Ki67 immunostaining in patients 2, 3, and 9 in C, T, R samples. D, box plot for %Ki67-positive cells in C, T, and R samples. E, comparison between %Ki67-positive cells (blue; left y-axis) and P1 scores (green; right y-axis) in individual patients (indicated in the titles) across C, T, and R samples.

Figure 4.

miRNAs in pattern P1 are associated with proliferation and cell-cycle. A, expression levels of P1 miRNAs that target BUB1B—let-7c, miR-125b-5p, miR-199a-3p, and miR-199a-5p. Presented are the log-ratio expression levels (NanoString) of tumor posttreatment relative to core biopsy pretreatment in patients 9 (black), 5 (gray), and 2 (white); and for BUB1B in the same samples and patients as measured by qRT-PCR. B, same as A for miR-100-5p and miR-145-5p, which target CDC25A. C, representative images of Ki67 immunostaining in patients 2, 3, and 9 in C, T, R samples. D, box plot for %Ki67-positive cells in C, T, and R samples. E, comparison between %Ki67-positive cells (blue; left y-axis) and P1 scores (green; right y-axis) in individual patients (indicated in the titles) across C, T, and R samples.

Close modal

To further support the association of P1 miRNAs with proliferation, we measured the level of the Ki67 proliferation marker in our cohort by immunostaining. Ki67 levels significantly varied between patients, mainly in pretreatment samples and in recurrent samples (Fig. 4C and D). Ki67 levels posttreatment were always lower compared with pretreatment. We found a general agreement between the P1 pattern in each patient and the modulations in Ki67 levels over time for individual patients (Fig. 4E), and there was an overall moderate anticorrelation between P1 scores and Ki67 levels for all samples (cc = −0.26; Supplementary Fig. S6). However, P1 scores better differentiated between samples across disease progression, were associated with response to treatment, and were less variable across all patients than Ki67 levels (Fig. 3C). Importantly, in recurrence samples average P1 scores were the lowest, whereas average Ki67 levels were not the highest, and exhibited high variability.

Overall, the above observations suggest that P1 miRNAs are inhibitors of the cell cycle in breast cancer cells and are highly informative regarding the proliferative state of the tumor.

P1 miRNAs differentiate between recurrence and recurrence-free patients

The observations that different stages of the disease are significantly different on their average P1 miRNA expression levels, and that average expression levels at recurrence are the lowest, indicate that this set of miRNAs may have a prognostic value. We calculated average expression levels of P1 miRNAs for an independent cohort of patients that did not experience recurrence for at least 10 years. We found significantly lower expression levels of P1 miRNAs in pretreatment tumors of patients with recurrence compared with recurrence-free patients (P = 0.03; Fig. 5A). Lower P1 miRNA expression levels were also observed for posttreatment tumors and lymph nodes of recurrent patients, but the differences were not significant. Importantly, there was no significant difference between RCB classes of both cohorts (Supplementary Table S1). Moreover, Kaplan–Meier and Cox proportional hazard models identified significant association between high P1 pretreatment expression levels and better recurrence-free survival (P = 0.016 and 0.045, respectively; Fig. 5B).

Figure 5.

P1 miRNAs differentiate between recurrence and recurrence-free patients. A, comparison between P1 scores in recurrent (REC; patients 1–10) and recurrence-free (REC FREE; patients 11–19) for three different types of tissues: C, T, and TL (see text). *, P < 0.05. B, Kaplan–Meier analysis for the association between pretreatment expression levels of P1 scores and recurrence-free survival; done on both cohorts together (19 patients). HR, hazard ratio; CI, confidence interval, as calculated by the Cox proportional hazard model.

Figure 5.

P1 miRNAs differentiate between recurrence and recurrence-free patients. A, comparison between P1 scores in recurrent (REC; patients 1–10) and recurrence-free (REC FREE; patients 11–19) for three different types of tissues: C, T, and TL (see text). *, P < 0.05. B, Kaplan–Meier analysis for the association between pretreatment expression levels of P1 scores and recurrence-free survival; done on both cohorts together (19 patients). HR, hazard ratio; CI, confidence interval, as calculated by the Cox proportional hazard model.

Close modal

In addition, we tested the prognostic value of P1 miRNAs on the METABRIC dataset. Interestingly, P1 scores are significantly lower in patients who developed metastasis, relative to metastasis-free patients (P = 0.001; Supplementary Fig. S7A), and higher levels of P1 scores are significantly associated with better disease-specific survival (P = 5.9 × 10−4; Supplementary Fig. S7B).

Because pattern P2 is the inverse complementary pattern to P1 and contains miRNAs that are upregulated in breast tumors relative to normal samples, we also calculated average expression levels of P2 miRNAs. As can be seen from Supplementary Fig. S8A, although the pretreatment expression does not significantly differ between patients with and without recurrence, posttreatment expression levels in tumor and lymph nodes are significantly different between patients with and without recurrent disease (P = 0.001 and 0.023, respectively). Moreover, Kaplan–Meier and Cox analysis identified significant association between low expression levels of P2 posttreatment and better recurrence-free survival (P = 0.0066 and 0.0175, respectively; Supplementary Fig. S8B).

Importantly, because the recurrence-free group consisted of hormone-positive patients only, we further analyzed P1 and P2 using solely hormone-positive patients from both cohorts. We found that P1 scores are still lower in patients with recurrence relative to recurrence-free patients (Supplementary Fig. S9A), although the result was not significant, probably due to the small sample size (seven vs. nine patients). P2 was significant despite the small sample size, both in the posttreatment tumor samples and in the posttreatment lymph node samples. Kaplan–Meier analyses performed on the hormone-positive patients alone (16 patients) were significant both for P1 pretreatment expression levels and for P2 posttreatment expression levels (Supplementary Fig. S9B). Similarly, Kaplan–Meier analysis by P1 scores on the METABRIC dataset was also significant when taking into account only hormone-positive samples (Supplementary Fig. S9C).

Thus, in our dataset, P1 and P2 miRNA expression levels pre- and posttreatment can significantly differentiate between patients experiencing recurrence and recurrence-free patients.

miRNAs in patterns P5 and P6 are associated with the metastatic process

miRNAs from patterns P5 and P6 exhibited a considerable modulation mainly at recurrence and not between pre- to posttreatment, which hints at their potential function during the invasive process. Importantly, these patterns were observed independently of the metastatic site (e.g., bone, lung, etc.) Most P5 miRNAs, defined as low at recurrence, are associated in the literature with metastasis in their downregulated state (Fig. 2 and Supplementary Table S2). Interestingly, some P5 miRNAs were upregulated in primary breast cancer relative to normal breast but downregulated in high-grade tumors according to the METABRIC dataset (Fig. 2). This opposite expression in primary and advanced tumors indicates a dual function at different stages of cancer. These miRNAs were found to be correlated with O-glycan biosynthesis; O-glycosylation has an important functional role in EMT induction (25, 26). In line with these results, we found that genes that are positively correlated with P5 miRNAs are enriched by genes that are upregulated during EMT (ref.21; Supplementary Fig. S10).

As most available studies are performed on primary tumors, we found less evidence for the miRNAs of pattern P6 playing any potential role in cancer. Similarly, in the METABRIC dataset, most miRNAs from this pattern (which is characterized by miRNAs that are overexpressed specifically in the recurrent tissue) are not expressed at all.

Next, we checked the mean expression levels of P5 and P6 miRNAs across the course of disease. Interestingly, the mean expression levels of P5 miRNAs in the lymph nodes after treatment (TL) were decreased toward their expression in the recurrent tissue (Supplementary Fig. S11A). We observed the same trend also for P6 miRNAs, where expression levels in the TL were increased toward their expression in the recurrent tissue (T vs. TL P = 0.01; Supplementary Fig. S11B). This observation might hint that the lymph nodes posttreatment already acquired some invasive characteristics.

In this study, we profiled miRNA expression of serial samples that represent the disease progression stages of breast cancer patients, from diagnosis through therapy till recurrence. The neoadjuvant setting is an ideal platform for such a study, enabling the collection of matched pretreatment biopsies and posttreatment residual tumors. Little is known about the evolution of solid tumors in individual patients, as assembly of such a longitudinal cohort that spans several years till recurrence requires comprehensive analysis of clinical databases and exploration of pathologic archives. To our knowledge, no comprehensive temporal analysis that includes matched samples also from distant metastasis for several patients has been performed. The long timespan across the different disease progression stages can result in large variation in archived samples preserved at various years. In this case, miRNAs profiling is currently the most reliable method for such analysis, as miRNA molecules are very stable.

Our approach for analysis of the longitudinal dataset utilizes modulation patterns instead of comparing two time points. Theoretically, each pattern can be associated with a particular scenario in tumor progression, such as proliferation or invasion. Interestingly, some patterns were more abundant, hinting that some scenarios are more likely than others. Only few miRNAs were classified to patterns P3, P4, P7, and P8; thus, these patterns are either not abundant (i.e., observed only for very few miRNAs) or are specific to too few patients.

P1, the most abundant pattern, exhibited an increase in expression after therapy, followed by a decrease at recurrence. The current paradigm is that the highly proliferative cells from the bulk of tumor are eliminated at chemotherapy, and the residual tumor is composed of dormant and/or chemo-resistant cells with a low proliferation rate. Because P1 miRNAs are associated with cell proliferation, this pattern can be considered as a “normalization pattern” in which the overall expression level in the residual tumor approaches normal breast tissue levels. Indeed, the majority of P1 miRNAs are downregulated in primary breast cancer versus normal breast tissue. Importantly, we showed that expression levels in response to therapy are increasing toward expression levels of adjacent normal breast tissue but not reaching this level, hinting at the possibility that albeit the fraction of proliferating cells was reduced, they were not entirely eliminated. Moreover, upregulation of P1 miRNAs following treatment was associated with better response to treatment. We assume that residual tumors of responders will be less proliferative, reflected by increased expression levels of P1 miRNAs, while patients not responding to treatment will not show an increase in P1 miRNA expression. Moreover, we identified several target genes of P1 miRNAs that are associated with the cell cycle or MAPK pathways. Interestingly, some of P1 miRNAs were recently identified in a miRNA signature associated with hyperactive MAPK and poor outcome (27). Most of these shared miRNAs were also correlated to the MAPK pathway in our pathway analysis. Several studies comparing pretreatment to posttreatment samples were previously performed, identifying several genes and pathways that are differentially expressed between pre- and posttreatment samples (28–31). Magbanua and colleagues analyzed gene expression data of serial breast tumor samples during neoadjuvant chemotherapy (31). In accordance with our results, they found that most differentially expressed genes are downregulated following neoadjuvant chemotherapy and are associated with the cell cycle and with response to treatment.

In the neoadjuvant treatment, a complete eradication of the tumor (pCR) is a strong predictor of long-term disease-free survival; however, for partial response the prognostic power is very limited. Previous studies indicated that RCB-I class have also a prognostic value (12); however, there was no significant difference between the RCB class of both cohorts. Interestingly, although P1 miRNAs were identified by the pattern dynamics using only the recurrence cohort, there was a significant difference in the average expression levels of the pretreatment samples between patients experiencing recurrence and recurrence-free patients. Importantly, any attempt to identify miRNAs that significantly differentiate recurrence and recurrence-free patients by standard t tests was not successful; at 25% FDR, none of the miRNAs differentiated recurrence from recurrence-free patients at pre- or posttreatment samples. Our data suggest that low pretreatment expression levels of P1 miRNAs are indicative of a higher probability of a patient to develop a recurrent event and may have a prognostic value. Kaplan–Meier and Cox analysis also supported this conclusion.

As we suggest that P1 miRNAs are a signature for tumor proliferation capacity, it is important to compare the results to Ki67, the gold-standard proliferation marker. The prognostic value of Ki67 score is controversial, due to its high variability and lack of reproducibility (32). Previous neoadjuvant studies demonstrated that only posttherapy Ki67 is a significant independent prognostic factor (33, 34). Our findings, that the change in Ki67 correlates with the P1 pattern, are in accordance with those of a recent study (35) showing that the change in Ki67-positive cells is an independent predictor of treatment outcomes rather than its absolute level. In contrast, another study found that Ki67 is a significant predictive and prognostic marker in pretreatment biopsies over a wide range of cutoff points, but raises the point that the large variability of this marker may impair its usefulness as a prognostic and predictive marker (36). In our dataset, posttreatment Ki67 levels were reduced in all patients, but were not correlated to the residual tumor burden, as was shown for P1 levels. Ki67 levels were much more variable than P1 values, mainly in the recurrence samples, where P1 score was low in all samples, irrespective of the metastatic niche or subtype, in contrast to the variability observed in Ki67. This suggests that the integrated P1 score is a more informative proliferation marker than Ki67.

Exclusive to our study, the cohort contains an additional time point, at recurrence, enabling a broader analysis of molecular changes over the course of disease. Most studies involve the expression of primary tumors and only few studies compared expression levels between primary tumors and metastasis, mostly LN metastasis, showing a high concordance between primary and metastatic tissue (8, 37, 38). In contrast to our approach, most studies search for differential expression between the different groups rather than longitudinal analysis in each patient. Previous studies comparing proliferation rates of primary tumors and metastases found either no difference or up to twofold higher rates for metastases (39–41). However, proliferation rates of matched primary and metastases were rarely estimated and, to our knowledge, compare only primary with matched lymph node metastases but not distant metastases. Pence and colleagues found no significant difference between matched primary breast tumor and lymph node metastases by Ki67 score (42). In agreement, we found that P1 scores of primary and lymph node metastases are not significantly different. However, our data indicate that recurrent metastatic samples are lower than the matched primary pretreatment tumors, suggesting that tumor clones that survived and escaped chemotherapy have a similar or higher proliferation capacity.

The modulation observed in the primary tumor following treatment was not observed in the lymph nodes compared with pretreatment lymph node, suggesting that these cells are either resistant to therapy or are protected from the chemotherapy effect in the lymphatic site. Moreover, for P5 and P6 miRNAs, we observed that posttreatment lymph node levels approach the recurrence expression levels, while the levels of tumor and lymph node pretreatment were similar.

The METABRIC dataset, as most other available datasets of breast cancer, includes only primary breast cancer samples and not metastatic samples. Therefore, the information on miRNAs that are expressed only at invasive stages, such as in patterns P5 and P6, is limited. Similarly, the identification of pathways and genes that are correlated to metastatic events is limited. Notably, most miRNAs from pattern P5 were upregulated in initial stages of breast cancer but downregulated in high-grade tumors. This dual behavior is reminiscent of TGFB, known to exert a dual function in cancer. In normal cells and early carcinoma, it acts as a tumor suppressor, while in aggressive and invasive tumors it promotes cancer progression, migration, and invasion (43, 44). TGFB in its pro-metastatic arm regulates EMT transformation that involves the transition from adherent epithelial cells to motile mesenchymal cells, a process necessary for the primary tumor in order to migrate and invade. The reverse process is MET—mesenchymal–epithelial transition, shown to be involved in the metastatic process, by promoting epithelial properties, thereby facilitating their settlement in distant organs. Indeed, most P5 miRNAs are correlated with O-glycan biosynthesis. O-glycosylation has an important functional role in EMT induction; in addition, our GSEA analysis revealed that most P5 miRNAs are enriched by genes upregulated during EMT.

In summary, identifying molecular players that modulate in the various stages of the disease can pinpoint pathways that are alternately activated, depending on the disease context. Enlargement of our cohort and validation of our findings in additional patients will strengthen our conclusions; moreover, performing the analysis on each molecular subtype will identify processes that are unique to each subtype. However, as discussed earlier, it is very hard to assemble such unique cohorts of patients with matched pretreatment, posttreatment and recurrent samples. Remarkably, notwithstanding the small size of our cohort and its heterogeneity, we identified sets of miRNAs that behave similarly across several patients and have implications for tumor progression, response to treatment, and recurrent disease.

No potential conflicts of interest were disclosed.

Conception and design: M. Dadiani, N. Bossel Ben-Moshe, S. Paluch-Shimon, A. Yosepovich, R. Berger, I. Barshack, E. Domany, B. Kaufman

Development of methodology: M. Dadiani, N. Bossel Ben-Moshe, G. Perry, A. Pavlovski, D. Morzaev, I. Barshack, E. Domany

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Dadiani, S. Paluch-Shimon, G. Perry, A. Yosepovich, I. Barshack, B. Kaufman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Dadiani, N. Bossel Ben-Moshe, G. Perry, I. Barshack, E. Domany, B. Kaufman

Writing, review, and/or revision of the manuscript: M. Dadiani, N. Bossel Ben-Moshe, S. Paluch-Shimon, A. Yosepovich, E.N. Gal-Yam, R. Berger, I. Barshack, E. Domany, B. Kaufman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Dadiani, N. Bossel Ben-Moshe, G. Perry, I. Marin, S. Kahana-Edwin, B. Kaufman

Study supervision: M. Dadiani, R. Berger, E. Domany, B. Kaufman

The authors thank Dr. Nava Epstein and Dr. Libbat Tirosh for helpful discussions, Dr. Camila Avivi for her assistance in the immunohistochemistry, and Dr. Adi Zundelevich for technical help with the experiments and for providing the BT474 cell lines.

This research was supported by a grant from Susan G. Komen for the Cure and by a grant from the Israel Cancer Association. E. Domany and N. Bossel Ben-Moshe were supported by the Leir Charitable foundation.

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.

1.
Polyak
K
. 
Breast cancer: origins and evolution
.
J Clin Invest
2007
;
117
:
3155
63
.
2.
Paluch-Shimon
S
,
Wolf
I
,
Goldberg
H
,
Evron
E
,
Papa
MZ
,
Shabtai
M
, et al
High efficacy of pre-operative trastuzumab combined with paclitaxel following doxorubicin & cyclophosphamide in operable breast cancer
.
Acta Oncol
2008
;
47
:
1564
9
.
3.
Wolmark
N
,
Wang
J
,
Mamounas
E
,
Bryant
J
,
Fisher
B
. 
Preoperative chemotherapy in patients with operable breast cancer: nine-year results from National Surgical Adjuvant Breast and Bowel Project B-18
.
J Natl Cancer Inst
2001
;
30
:
96
102
.
4.
Untch
M
,
Loibl
S
,
Konecny
GE
,
von Minckwitz
G
. 
Neoadjuvant clinical trials for the treatment of primary breast cancer: the experience of the German study groups
.
Curr Oncol Rep
2012
;
14
:
27
34
.
5.
He
L
,
Hannon
GJ
. 
MicroRNAs: small RNAs with a big role in gene regulation
.
Nat Rev Genet
2004
;
5
:
522
31
.
6.
Iorio
M V
,
Casalini
P
,
Tagliabue
E
,
Ménard
S
,
Croce
CM
. 
MicroRNA profiling as a tool to understand prognosis, therapy response and resistance in breast cancer
.
Eur J Cancer
2008
;
44
:
2753
9
.
7.
Volinia
S
,
Galasso
M
,
Elena
M
,
Wise
TF
,
Palatini
J
,
Huebner
K
, et al
Breast cancer signatures for invasiveness and prognosis defined by deep sequencing of microRNA
.
Proc Natl Acad Sci U S A
2012
;
109
:
3024
9
.
8.
Cascione
L
,
Gasparini
P
,
Lovat
F
,
Carasi
S
,
Pulvirenti
A
,
Ferro
A
, et al
Integrated MicroRNA and mRNA signatures associated with survival in triple negative breast cancer
.
PLoS One
2013
;
8
:
1
13
.
9.
Dvinge
H
,
Git
A
,
Gräf
S
,
Salmon-Divon
M
,
Curtis
C
,
Sottoriva
A
, et al
The shaping and functional consequences of the microRNA landscape in breast cancer
.
Nature
2013
;
497
:
378
82
.
10.
Curtis
C
,
Shah
SP
,
Chin
S-F
,
Turashvili
G
,
Rueda
OM
,
Dunning
MJ
, et al
The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups
.
Nature
2012
;
486
:
346
52
.
11.
Ogston
KN
,
Miller
ID
,
Payne
S
,
Hutcheon
AW
,
Sarkar
TK
,
Smith
I
, et al
A new histological grading system to assess response of breast cancers to primary chemotherapy: prognostic significance and survival
.
Breast
2003
;
12
:
320
7
.
12.
Symmans
WF
,
Peintinger
F
,
Hatzis
C
,
Rajan
R
,
Kuerer
H
,
Valero
V
, et al
Measurement of residual breast cancer burden to predict survival after neoadjuvant chemotherapy
.
J Clin Oncol
2007
;
25
:
4414
22
.
13.
Ballman
K V
,
Grill
DE
,
Oberg
AL
,
Therneau
TM
. 
Faster cyclic loess: normalizing RNA arrays via linear models
.
Bioinformatics
2004
;
20
:
2778
86
.
14.
Bossel Ben-moshe
N
,
Avraham
R
,
Kedmi
M
,
Zeisel
A
,
Yitzhaky
A
,
Yarden
Y
, et al
Context-specific microRNA analysis: identification of functional microRNAs and their mRNA targets
.
Nucleic Acids Res
2012
;
40
:
10614
27
.
15.
Enright
AJ
,
John
B
,
Gaul
U
,
Tuschl
T
,
Sander
C
,
Marks
DS
. 
MicroRNA targets in Drosophila
.
Genome Biol
2003
;
5
:
R1
.
16.
Lewis
BP
,
Burge
CB
,
Bartel
DP
. 
Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets
.
Cell
2005
;
120
:
15
20
.
17.
Kertesz
M
,
Iovino
N
,
Unnerstall
U
,
Gaul
U
,
Segal
E
. 
The role of site accessibility in microRNA target recognition
.
Nat Genet
2007
;
39
:
1278
84
.
18.
Drier
Y
,
Sheffer
M
,
Domany
E
. 
Pathway-based personalized analysis of cancer
.
Proc Natl Acad Sci U S A
2013
;
110
:
6388
93
.
19.
Livshits
A
,
Git
A
,
Fuks
G
,
Caldas
C
,
Domany
E
. 
Pathway-based personalized analysis of breast cancer expression data
.
Mol Oncol
2015
;
9
:
1471
83
.
20.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
21.
Groger
CJ
,
Grubinger
M
,
Waldhor
T
,
Vierlinger
K
,
Mikulits
W
. 
Meta-analysis of gene expression signatures defining the epithelial-to-mesenchymal transition during cancer progression
.
PLoS One
2012
;
7
:
e51136
.
22.
Biagioni
F
,
Bossel Ben-Moshe
N
,
Fontemaggi
G
,
Canu
V
,
Mori
F
,
Antoniani
B
, et al
miR-10b*, a master inhibitor of the cell cycle, is downregulated in human breast tumours
.
EMBO Mol Med. England
2012
;
4
:
1214
29
.
23.
Scintu
M
,
Vitale
R
,
Prencipe
M
,
Gallo
AP
,
Bonghi
L
,
Valori
VM
, et al
Genomic instability and increased expression of BUB1B and MAD2L1 genes in ductal breast carcinoma
.
Cancer Lett
2007
;
254
:
298
307
.
24.
Mehdipour
P
,
Pirouzpanah
S
,
Sarafnejad
A
,
Atri
M
,
Shahrestani
ST
,
Haidari
M
. 
Prognostic implication of CDC25A and cyclin E expression on primary breast cancer patients
.
Cell Biol Int
2009
;
33
:
1050
6
.
25.
Freire-de-lima
L
,
Gelfenbeyn
K
,
Ding
Y
,
Mandel
U
,
Clausen
H
,
Handa
K
, et al
Involvement of O-glycosylation de fi ning oncofetal fi bronectin in epithelial-mesenchymal transition process
.
Proc Natl Acad Sci U S A
2011
;
108
:
17690
5
.
26.
Ding
Y
,
Gelfenbeyn
K
,
Freire-de-Lima
L
,
Handa
K
,
Hakomori
S
. 
Induction of epithelial-mesenchymal transition with O-glycosylated oncofetal fibronectin
.
FEBS Lett
2012
;
586
:
1813
20
.
27.
Miller
P
,
Clarke
J
,
Koru-Sengul
T
,
Brinkman
J
,
El-Ashry
D
. 
A novel MAPK-microRNA signature is predictive of hormone-therapy resistance and poor outcome in ER-positive breast cancer
.
Clin Cancer Res
2015
;
21
:
373
85
.
28.
Almendro
V
,
Cheng
Y-K
,
Randles
A
,
Itzkovitz
S
,
Marusyk
A
,
Ametller
E
, et al
Inference of tumor evolution during chemotherapy by computational modeling and in situ analysis of genetic and phenotypic cellular diversity
.
Cell Rep
2014
;
6
:
514
27
.
29.
Korde
LA
,
Lusa
L
,
McShane
L
,
Lebowitz
PF
,
Lukes
L
,
Camphausen
K
, et al
Gene expression pathway analysis to predict response to neoadjuvant docetaxel and capecitabine for breast cancer
.
Breast Cancer Res Treat
2010
;
119
:
685
99
.
30.
Koike Folgueira
MAA
,
Brentani
H
,
Carraro
DM
,
De Camargo Barros Filho
M
,
Hirata Katayama
ML
,
Santana de Abreu
AP
, et al
Gene expression profile of residual breast cancer after doxorubicin and cyclophosphamide neoadjuvant chemotherapy
.
Oncol Rep
2009
;
22
:
805
13
.
31.
Magbanua
MJM
,
Wolf
DM
,
Yau
C
,
Davis
SE
,
Crothers
J
,
Au
A
, et al
Serial expression analysis of breast tumors during neoadjuvant chemotherapy reveals changes in cell cycle and immune pathways associated with recurrence and response
.
Breast Cancer Res
2015
;
17
:
73
.
32.
Polley
M-YC
,
Leung
SCY
,
Gao
D
,
Mastropasqua
MG
,
Zabaglo
LA
,
Bartlett
JMS
, et al
An international study to increase concordance in Ki67 scoring
.
Mod Pathol
2015
;
28
:
778
86
.
33.
Jones
RL
,
Salter
J
,
A'Hern
R
,
Nerurkar
A
,
Parton
M
,
Reis-Filho
JS
, et al
The prognostic significance of Ki67 before and after neoadjuvant chemotherapy in breast cancer
.
Breast Cancer Res Treat
2009
;
116
:
53
68
.
34.
Von Minckwitz
G
,
Schmitt
WD
,
Loibl
S
,
Muller
BM
,
Blohmer
JU
,
Sinn
BV
, et al
Ki67 measured after neoadjuvant chemotherapy for primary breast cancer
.
Clin Cancer Res
2013
;
19
:
4521
31
.
35.
Matsubara
N
,
Mukai
H
,
Fujii
S
,
Wada
N
. 
Different prognostic significance of Ki-67 change between pre- and post-neoadjuvant chemotherapy in various subtypes of breast cancer
.
Breast Cancer Res Treat
2013
;
137
:
203
12
.
36.
Denkert
C
,
Loibl
S
,
Muller
BM
,
Eidtmann
H
,
Schmitt
WD
,
Eiermann
W
, et al
Ki67 levels as predictive and prognostic parameters in pretherapeutic breast cancer core biopsies: a translational investigation in the neoadjuvant GeparTrio trial
.
Ann Oncol
2013
;
24
:
2786
93
.
37.
Weigelt
B
,
Glas
AM
,
Wessels
LFA
,
Witteveen
AT
,
Peterse
JL
,
van't Veer
LJ
. 
Gene expression profiles of primary breast tumors maintained in distant metastases
.
Proc Natl Acad Sci U S A
2003
;
100
:
15901
5
.
38.
Schleifman
EB
,
Desai
R
,
Spoerke
JM
,
Xiao
Y
,
Wong
C
,
Abbas
I
, et al
Targeted biomarker profiling of matched primary and metastatic estrogen receptor positive breast cancers
.
PLoS One
2014
;
9
:
e88401
.
39.
Klein
CA
. 
Parallel progression of primary tumours and metastases
.
Nat Rev Cancer
2009
;
9
:
302
12
.
40.
Frankfurt
OS
,
Greco
WR
,
Slocum
HK
,
Arbuck
SG
,
Gamarra
M
,
Pavelic
ZP
, et al
Proliferative characteristics of primary and metastatic human solid tumors by DNA flow cytometry
.
Cytometry
1984
;
5
:
629
35
.
41.
Kusama
S
,
Spratt
JSJ
,
Donegan
WL
,
Watson
FR
,
Cunningham
C
. 
The gross rates of growth of human mammary carcinoma
.
Cancer
1972
;
30
:
594
9
.
42.
Pence
JC
,
Kizilbash
AM
,
Kerns
BJ
,
Marks
JR
,
Iglehart
JD
. 
Proliferation index in various stages of breast cancer determined by Ki-67 immunostaining
.
J Surg Oncol
1991
;
48
:
11
20
.
43.
Bierie
B
,
Moses
HL
. 
Tumour microenvironment: TGFbeta: the molecular Jekyll and Hyde of cancer
.
Nat Rev Cancer
2006
;
6
:
506
20
.
44.
Lebrun
J-J
. 
The dual role of TGF β in human cancer: from tumor suppression to cancer metastasis
.
ISRN Mol Biol
2012
;
2012
:
1
28
.