In the present study, cell lines representative of cancer types were selected as described previously (16): MCF7 and T47D for luminal A type; BT474 for luminal B type; and MDA-MB468, MDA-MB231, HCC70 and HCC1143 for TNBC. The human breast cancer cell lines were obtained from the Riken Cell Bank (MCF7; Riken BioResource Research Center, Tsukuba, Japan) and from the American Type Culture Collection (T47D, BT474, MDA-MB468, MDA-MB231, HCC70 and HCC1143; Manassas, VA, USA). Cells were cultured in RPMI-1640 medium (Wako Pure Chemical Industries, Ltd., Osaka, Japan) supplemented with 100 U/ml of penicillin, 100 U/ml of streptomycin and 10% fetal bovine serum (FBS; Gibco; Thermo Fisher Scientific, Inc., Waltham, MA, USA) in a humidified atmosphere containing 5% CO₂ at 37°C.

Four genes were identified as candidates for molecular target therapy in TNBC using whole transcriptome analysis and a lentiviral shRNA-screening library. For transcriptome analysis, total RNA was prepared from cell lines (MCF7, BT474, T47D, MDA-MB468, HCC70 and HCC1143) using NucleoSpin RNA (Takara Bio, Inc., Shiga, Japan). The quality of the RNA was assessed using the RNA integrity number (RIN) obtained by the Agilent RNA6000 Pico Kit and the Agilent 2100 Bioanalyzer (both Agilent Technologies, Santa Clara, CA, USA). Samples used for transcriptome analysis had, on average, an RIN value of 9.4 and a minimum RIN value of 8.9. The library was prepared using the TruSeq RNA Sample Prep Kit v2 (Illumina, Inc., San Diego, CA, USA) from 1 µg of total RNA according to the manufacturer's protocol. The resulting libraries were subjected to single-end sequencing of 76-bp reads using the NextSeq 500 High Output v1 Kit on the Illumina NextSeq 500 system (both from Illumina, Inc.). Data processing and analyses were performed using the TopHat version 2 alignment, cufflinks assembly and differential expression apps (all from Illumina, Inc.). Briefly, the reads were filtered, trimmed and aligned in the UCSC reference human genome 19 (hg19) using the Tophat2 (v2.0.7) and Bowtie1 (0.12.9) pipelines. The transcripts were assembled using Cufflinks 2.1.1, and differentially expressed transcripts were detected using Cuffdiff 2.1.1. Genes with a false discovery rate-adjusted q-value of <0.05 and log₂-fold change (TNBC/non-TNBC) of >5 were defined as significantly upregulated genes in TNBC cells.

MISSION LentiPlex Human Pooled shRNA Library (SHPH01-1SET; Sigma-Aldrich; Merck KGaA, Darmstadt, Germany) is a genome-wide shRNA library that covers ~15,000 genes, with 80,000 shRNA clones. In order to identify shRNAs that targeted genes able to specifically destroy TNBC cell lines compared with non-TNBC cell lines, a mixture of lentiviral particles (SHPH01-1SET; Sigma-Aldrich; Merck KGaA) was introduced to the breast cancer cell lines of non-TNBC (MCF7, T47D and BT474) and TNBC (MDA-MB468, HCC70 and HCC1143) subtypes at multiplicity of infections that yielded 30-50% infected cells according to the manufacturer's protocol. The infected cells were subsequently selected using puromycin for 7 days, and genomic DNA with integrated shRNA was isolated from the cell lines. Thirty cycles of polymerase chain reaction (PCR) was performed using KAPA HiFi HotStart ReadyMix (2X) (KAPA Biosystems; Roche Diagnostics, Indianapolis, IN, USA). The following primers were used for the PCR of the shRNA vector: 5′-ATTTCTTGGCTTTATATATCTTGTGGAAAG-3′ (sense) and reverse primer, 5′-TGTGGATGAATACTGCCATTTGT CTC-3′ (antisense). The PCR were performed on 3-µg genomic DNAs. PCR conditions consisted of initial denaturation at 95°C for 3 min, followed by 30 cycles of denaturation at 98°C for 20 sec, annealing at 60°C for 15 sec and elongation at 72°C for 30 sec. Sequencing adaptors (BIOO Scientific, Austin, TX, USA) were ligated to the PCR amplicons with 8 cycles of PCR and Illumina sequencing was performed for 150 cycles on an Illumina NextSeq500 sequencer with 10% PhiX control DNA (Illumina, Inc.). Data processing was performed in accordance with the manufacturer's protocol. Briefly, FASTQ files were used to trim and align the adapter sequences to the reference shRNA sequences using Bowtie2. The Bowtie output files were converted to count files through a python script 'countBowtieHits.py' and 'RTable.py', which were obtained from the manufacturer's protocol. TCC-edgeR was used for normalization and statistical analysis as described previously (17).

The Reference Expression dataset (RefEx: http://refex.dbcls.jp) was used to examine the expression levels of CPA4 in several noncancerous tissues using RNA sequencing methods (http://refex.dbcls.jp). Specific low-expression genes in 40 normal organs were picked up as maximum fragments per kilobase of transcript per million mapped reads of <1.0. The present transcriptome data were submitted to a public repository, the Gene Expression Omnibus (accession no. GSE113653).

The Cancer Genome Atlas database (cBioPortal Breast Cancer: METABRIC, Nature 2012 and Nat Commun 2016: http://www.cbioportal.org) was used to validate the prognostic significance of CPA4 in patients with breast cancer.

Tumor specimens from 221 patients with primary breast cancer who underwent surgery for excision of a primary tumor between January 1999 and October 2010 at the Gunma University Hospital (Maebashi, Japan) were retrospectively analyzed. The median follow-up period for survivors was 118 months. Some data from patients with ductal carcinoma in situ, preoperative chemotherapy, preoperative hormone therapy and male breast carcinoma from the present tissue microarray preparations were excluded. Tumor staging was based on the Union for International Cancer Control TNM classification (seventh edition) (18). Nuclear grades (NGs) were defined as the sum of scores for the nuclear atypia, as described previously (19). The present research conformed to the tenets of the Declaration of Helsinki and to the guidelines of the Gunma University Ethical Review Board for Medical Research Involving Human Subjects (approval no. 1297).

For tissue microarray, clinical formalin-fixed paraffin-embedded samples were stored in the archives of the Clinical Department of Pathology, Gunma University Hospital (Maebashi, Japan). For each patient, one paraffin block containing representative non-necrotic tumor areas was selected. Breast cancer tissue cores (2.0-mm diameter per tumor) were punched out from the representative areas near the invasive front and transferred into the paired recipient paraffin block using a tissue array instrument (Beecher Instruments, Inc., Silver Spring, MD, USA).

For IHC, a 4-µm section was cut from the sample paraffin blocks. Each section was mounted on a silane-coated glass slide, deparaffinized and soaked for 30 min at room temperature in 0.3% H₂O₂/methanol to block endogenous peroxidases. The sections were subsequently heated in boiled 10 mM citrate buffer (pH 6.0) at 98°C for 30 min. Non-specific binding sites were blocked by incubating with 0.25% Casein/1% bovine serum albumin (Code 81-003-3; EMD Millipore, Kankakee, IL, USA) for 30 min at room temperature. A rabbit polyclonal anti-CPA4 antibody (HPA021030; Sigma-Aldrich; Merck KGaA) was applied at a dilution of 1:400 for 24 h at 4°C. A MAX-PO secondary antibody from the Histofine Simple Stain MAX-PO (Multi) Kit (Nichirei Corporation, Tokyo, Japan) was used for 30 min at room temperature according to the manufacturer's instructions. The chromogen 3,3-diaminobenzidine tetrahydrochloride (Dojindo Molecular Technologies, Inc., Kumamoto, Japan) was applied as a 0.02% solution containing 0.005% H2O2 in 50 mM Tris-HCl buffer (pH 7.6). The sections were lightly counterstained with Mayer's hematoxylin and mounted. Negative controls were established by omitting the primary antibody. Other IHC was performed using the following primary antibodies: Anti-ER (ready to use; SP1), anti-PgR (ready to use; 1E2), anti-HER2 (ready to use; 4B5), anti-Ki67 (ready to use; 30-9) (Ventana Medical Systems, Inc., Tucson, AZ, USA), anti-EGFR (ready to use; 31G7; Nichirei Corporation), anti-CK5/6 (1:50; 5/16 B4; Dako; Agilent Technologies, Inc., Santa Clara, CA, USA), anti-E-cadherin (ready to use; 36; Ventana Medical Systems, Inc.) anti-ALDH1 (1:400; 46/ALDH; BD Biosciences, San Jose, CA, USA), anti-CD44 (1:50; DF1485; Dako; Agilent Technologies, Inc.), anti-CD24 (1:40; SN3b; Thermo Fisher Scientific, Inc.), AR (1:200; AR441; Thermo Fisher Scientific, Inc.), Claudin1 (1:50; 1449527A; Invitrogen; Thermo Fisher Scientific, Inc.), Claudin3 (1:100; GR253635-1; Abcam, Cambridge, UK), Claudin4 (1:50; 3E2C1; Invitrogen; Thermo Fisher Scientific, Inc.), Claudin7 (1:50; GR212894-5; Abcam) and Vimentin (ready to use; V9; Dako; Agilent Technologies, Inc.). A Ventana BenchMark XT system (Roche Diagnostics, Basel, Switzerland) was used for ER, PgR, HER2, Ki67 and E-cadherin. All sections were examined under a BX43 light microscope (Olympus Corporation, Tokyo, Japan).

CPA4 expression was deemed positive in cells with stained cytoplasms. In addition, the cutoff value for CPA4 positivity was 20% (14). The cutoff value for ER and PgR positivity was 1%. HER2 expression was scored according to the American Society of Clinical Oncology/College of American Pathologists guideline (20). Ki67 labeling index was used to calculate the percentage of cells with high nuclear expression in ~1,000 cells per sample, as described previously (21). EGFR expression was scored in the same manner as HER2 expression; scores of 0 and 1+ were considered as negative, and 2+ and 3+ as positive. A cutoff value of 10% for CK5/6, ALDH1, CD44, CD24, claudin and AR was used (19,22). The cutoff value for high E-cadherin expression was set as >70% (23).

Based on the IHC results, the breast cancer subtypes were defined as follows: Luminal A-like (ER or PgR+ and HER2 0/1+/2+), luminal B-like (ER or PgR+ and HER2 3+), HER2-like (ER-, PgR- and HER2 3+) and TNBC-like (ER-, PgR- and HER2 0/1+/2+).

The sections were prepared, and endogenous peroxidase was blocked as described above. Antigen retrieval was performed by heating samples in boiled Immunosaver Antigen Activator (Nisshin EM Co., Ltd., Tokyo, Japan) at 98°C for 45 min and stripping was performed by heating in boiled 10 mM citrate buffer (pH 6.0) at 98°C 15 min in a microwave. Nonspecific binding sites were blocked as described above (19). The sections were incubated with primary antibodies CPA4 (1:400) and E-cadherin (1:500) overnight at 4°C or for 1 h at room temperature, respectively. The secondary antibody was used as described for the protocol for IHC. Multiplex covalent labeling was performed (CPA4; Fluorescein, E-cadherin; Cyanine 3) with tyramide signal amplification (Opal 3-Plex Kit; PerkinElmer, Inc., Waltham, MA, USA) according to the manufacturer's protocol. All sections were counterstained with 4′,6-diamidino-2-phenylin-dole at room temperature (Opal 3-Plex Kit) and examined under an All-in-One BZ-X710 fluorescence microscope (Keyence Corporation, Osaka, Japan).

CPA4-specific siRNAs #1 (5′-CCAAAGAACAUCUGAGAUGtt-3′) and siRNA #2 (5′-CAGCAAAUCUUGUAGGGAUtt-3′) and negative-control siRNA were purchased from GeneDesign, Inc. (Osaka, Japan) and Bonac Corporation (Fukuoka, Japan). MDA-MB231 and MDA-MB468 at a density of 1×10⁶ cells/well were seeded in 100 µl of Opti MEM Reduced Serum Medium (Invitrogen; Thermo Fisher Scientific, Inc.). In total, 20 nM of CPA4-specific siRNAs 1 and 2, and scrambled siRNA (as a negative control) were used to treat cells; siRNAs was transfected using an electroporator CUY-21 EDIT II (Bex Co., Ltd., Tokyo, Japan) according to the manufacturer′s instructions. The subsequent experiments were performed following 24 h of transfection.

The proteins were extracted using lysis buffer [10 mM Tris-HCl (pH 7.5), 1 mM ethylene-diaminetetraacetic acid, 10% glycerol, 0.5% NP-40 detergent, 400 mM NaCl, 4 µg/ml of aprotinin, phenylmethylsulfonyl fluoride and dithiothreitol]. The BCA protein assay was used for the protein determination. Total protein (10 µg loaded per lane) was separated using SDS-PAGE (on a 10% polyacrylamide gel), at 300 mA for 90 min and transferred on a nitrocellulose membrane (Invitrogen; Thermo Fisher Scientific, Inc.). Blocking of the membrane was performed using 4% skimmed milk for 60 min at room temperature. The protein expression levels of CPA4, E-cadherin and β-actin were assessed using western blot analysis. These proteins were detected using specific antibodies to CPA4 (1:200; HPA021030; Sigma-Aldrich; Merck KGaA), E-cadherin (1:1,000; M106; Takara Bio, Inc.), and β-actin (1:1,000; #3700; Cell Signaling Technology, Inc.). The membrane was incubated in primary antibodies for overnight at 4°C. β-actin served as a loading control. ECL anti-mouse or anti-rabbit IgG, peroxidase-linked whole antibody was used for secondary antibody (GE Healthcare Life Sciences, Little Chalfont, UK). The signals were detected using the ECL Western Blotting Detection System and an Image Quant LAS 4000 machine (GE Healthcare Life Sciences).

Cell proliferation analyses were performed using Cell Counting Kit-8 (CCK-8; Dojindo Molecular Technologies, Inc., Kumamoto, Japan). Breast cancer cell line cells were plated onto 96-well plates in 100 µl of RPMI-1640 medium containing 10% FBS, at ~3,000 cells per well following siRNA transfection. MDA-MB231 and MDA-MB468 cells were evaluated at 48 and 72 h post-CPA4 siRNA treatment, respectively. For quantifying cell proliferation, 10 µl of CCK-8 solution was added to each well and incubated at 37°C for 2 h. Following this, the absorbance of each well at 450 nm was detected using a plate reader (Bio-Rad Laboratories, Inc., Hercules, CA, USA).

Cell migration was examined using MDA-MB231 and MDA-MB468 cells transfected with a negative control or CPA4 siRNAs. The transfected breast cancer cells were grown in 6-well plates until 100% confluence, and a uniform straight wound was produced in the monolayer in each well using a pipette tip. Wells were washed with phosphate-buffered saline to remove all cell debris Cells were cultured in an atmosphere containing 5% CO₂ at 37°C. Closure or filling of the wound was evaluated at 3 and 24 h in MDA-MB231 and MDA-MB468 plates, respectively.

Statistical analyses were performed using the Wilcoxon test for continuous variables, and the χ² test and one-way analysis of variance for categorical variables. Survival curves were generated according to the Kaplan-Meier method. The differences between survival curves were examined using the log-rank test. Cox proportional hazard regression was used to examine the independent prognostic contribution of standard prognostic variables, including CPA4 expression, age, tumor factor, NG, lymph node metastasis (LNM), venous invasion and adjuvant therapy. P<0.05 was considered to indicate a statistically significant difference. JMP software version 12 (SAS Institute, Cary, NC, USA) was used to analyze all statistical analyses.

Among 23,609 genes, 171 that were highly expressed in TNBC compared with their expression in non-TNBC cell lines were selected for transcriptome analysis (Fig. 1A). Following this, 86 of the 171 genes, in which suppression was indicated to have inhibited the viability of TNBC cell lines according to the shRNA library, were selected (Fig. 1B). Finally, 4 genes that were not expressed in normal organs according to a public database were selected (Fig. 1C). It was deduced that these 4 candidates were associated with TNBC viability because they were highly expressed in the TNBC cells compared with non-TNBC cells and normal organs. In the present study, CPA4 was investigated in order to discover a novel molecular target against TNBC.

Cytoplasmic expression of CPA4 in normal breast tissues was decreased compared with breast cancer tissues (Fig. 2). Out of the 221 breast cancer samples, 153 specimens (69.2%) were assigned to the low CPA4 expression group (Fig. 2B) and 68 specimens (30.8%) were assigned to the high CPA4 expression group (Fig. 2C and D).

The association between CPA4 expression and clinicopathological characteristics of the patients were indicated (Table I). Notably, no association was identified between CPA4 expression and clinicopathological characteristics in the 221 breast cancer cases. Furthermore, the association between CPA4 expression levels with breast CSCs markers (ALDH1, CD44 and CD24), EMT markers (vimentin and E-cadherin) and the AR was examined using immunohistochemical staining. An association between high CPA4 expression and high ALDH1 expression was noted (Table II, P=0.027). However, only a few cases exhibited high ALDH1 expression.

Clinicopathological factors were significantly different in 22/68 TNBC cases in the high CPA4 expression group (Tables III and IV). TNBC with high CPA4 expression was associated with low Ki67 expression and the expression of CD44/CD24 (Table III, P=0.011; and Table IV, P=0.016, respectively). The association between CPA4 and E-cadherin, a representative epithelial marker (7), was also assessed. High CPA4 expression was associated with low E-cadherin expression (Table IV, P=0.016). The association between high CPA4 and low E-cadherin expression was validated in representative TNBC sections using IHC (Fig. 2E and F). No other significant differences in the other evaluated factors were indicated.

The overall and disease-free survival were not significantly associated with CPA4 expression (Fig. 3A and B; P=0.19 and P=0.49, respectively). However, the overall and disease-free survival intervals of high-CPA4-expressing TNBC cells were worse compared with those of low-CPA4-expressing TNBC cells (Fig. 3C and D; P=0.004 and P=0.017, respectively).

Using a public database cohort, the prognostic significance of CPA4 expression on overall survival was validated (Fig. 4). As expected, the patients with TNBC and high CPA4 expression exhibited poorer prognoses compared with those with low CPA4 expression; however, the differences were not significant (Fig. 4A and B; P=0.720 and P=0.078, respectively).

The association between CPA4 expression and other factors was also explored, including the NG and LNM (Fig. 5). Among NG1-2 and LNM-negative cases, the overall and disease-free survival were not significantly associated with CPA4 expression (Fig. 5A, B, E and F; P=0.39, P=0.66, P=0.29 and P=0.16, respectively). However, among NG3 cases and LNM-positive cases, the overall and disease-free survival intervals of highly expressed CPA4 TNBC cells were worse than those of low CPA4 TNBC cells (Fig. 5C, D, G and 5H; P=0.0002, P=0.019, P=0.0004 and P=0.015, respectively). To clarify the prognostic significance of CPA4, multivariate analysis was performed for survival, and results indicated that high CPA4 expression was an independent prognostic factor for poor survival (Table V; P=0.001).

The role of CPA4 on cell viability and migration ability was assessed using RNA interference. Western blot analysis was performed to validate the CPA4 knockdown experiments in TNBC cell lines treated with specific CPA4 siRNAs (Fig. 6A and B). In addition, it was identified that cell viability in the CPA4 siRNA group was significantly inhibited compared with that in the negative-control group (Fig. 6C, P<0.05). Furthermore, CPA4 knockdown suppressed cell migration in comparison with the negative-control cells (Fig. 6D, P<0.05).

The association between CPA4 and the expression of E-cadherin was examined. The expression of E-cadherin was increased in CPA4 siRNA groups (Fig. 6A and B). These data were consistent with the inverse association identified between CPA4 and E-cadherin expression in clinical TNBC samples.