Protein arginine methyltransferase-6 (PRMT6) regulates steroid-dependent transcription and alternative splicing and is implicated in endocrine system development and function, cell death, cell cycle, gene expression and cancer. Despite its role in these processes, little is known about its function and cellular targets in breast cancer. To identify novel gene targets regulated by PRMT6 in breast cancer cells, we used a combination of small interfering RNA and exon-specific microarray profiling in vitro coupled to in vivo validation in normal breast and primary human breast tumours. This approach, which allows the examination of genome-wide changes in individual exon usage and total transcript levels, demonstrated that PRMT6 knockdown significantly affected i) the transcription of 159 genes and ii) alternate splicing of 449 genes. The PRMT6-dependent transcriptional and alternative splicing targets identified in vitro were validated in human breast tumours. Using the list of genes differentially expressed between normal and PRMT6 knockdown cells, we generated a PRMT6-dependent gene expression signature that provides an indication of PRMT6 dysfunction in breast cancer cells. Interrogation of several well-studied breast cancer microarray expression datasets with the PRMT6 gene expression signature demonstrated that PRMT6 dysfunction is associated with better overall relapse-free and distant metastasis-free survival in the oestrogen receptor (ER (ESR1)) breast cancer subgroup. These results suggest that dysregulation of PRMT6-dependent transcription and alternative splicing may be involved in breast cancer pathophysiology and the molecular consequences identifying a unique and informative biomarker profile.
Protein arginine methylation is a common post-translational modification that was first described over 40 years ago (Paik & Kim 1967). It is catalysed by a family of 11 human protein arginine methyltransferases (PRMTs) and results in the transfer of either one or two methyl groups from the methyl donor S-adenosyl-l-methionine to guanidino nitrogen atoms in arginine residues of the target protein. This alters the function and activity of the protein and impacts on a diverse range of cellular processes including differentiation, DNA repair, transcription, RNA processing, signal transduction, cell localisation and apoptosis (Bedford & Richard 2005). As would be expected of proteins that regulate such diverse processes, their aberrant expression and activity have been linked to a number of human diseases, including cancer (Aletta & Hu 2008).
PRMTs regulate transcription by functioning as co-activators for a range of transcription factors, including steroid/nuclear receptors, p53, YY1 and NF-κB; E2F1; MEF2C and β-catenin (Bedford & Clarke 2009, Lee & Stallcup 2009). We have recently shown that PRMT6 can also co-activate steroid hormone receptors (SHRs), including the ERα, progesterone receptor and glucocorticoid receptor (GR (NR3C1)), in a methylation-dependent manner (Harrison et al. 2010). This appears to contrast with the studies demonstrating that PRMT6 can repress transcription by methylating histone 3 at arginine 2 (H3R2), which prevents the mixed lineage leukaemia methyltransferase complex binding to and methylating histone 3 at lysine 4 (H3K4; Guccione et al. 2007, Hyllus et al. 2007, Iberg et al. 2008). Therefore, whether PRMT6 functions as a transcriptional enhancer or repressor in relation to gene expression remains controversial.
PRMTs also regulate alternate splicing in a methylation-dependent manner. Co-activator-associated arginine methyltransferase 1 (CARM1) can methylate a number of proteins involved in pre-mRNA processing and promotes skipping of both a splicing reporter and the endogenous CD44 gene (Cheng et al. 2007). PRMT6, which can methylate the splicing factors RDA288 and heterogeneous nuclear ribonucleoprotein D-like (Cheng et al. 2007), promotes exon skipping of vascular endothelial growth factor (VEGF (VEGFA)) and spleen tyrosine kinase (SYK) genes in MCF-7 and T-47D breast cancer cells (Harrison et al. 2010).
Initial studies have identified several PRMT proteins as having a key role in SHR-mediated prostate cancer, breast cancer, gastric cancer, lymphoma and leukaemia (Teyssier et al. 2010). Many proteins involved in RNA splicing/processing are methylated and/or associated in complexes with PRMTs. The distinct possibility arises that PRMTs are able to directly methylate and regulate proteins involved in RNA processing events and alternative splicing. Exon profiling in the framework of transcriptomic analysis is an emerging area in disease diagnosis and prognosis (Dutertre et al. 2010). The fact that certain PRMTs are associated with RNA processing and gene expression makes them a clear choice to examine their ability to regulate alternative splicing and transcription in breast cancer.
Materials and methods
Cell culture and transfection
MCF-7 cells were maintained in DMEM nutrient mixture F-12 plus 10% FBS. T-47D cells were maintained in RPMI-1640 medium plus 10% FBS, 5 μg/ml insulin, 10 mM HEPES and 1 mM sodium pyruvate. Both breast cancer cells lines were maintained at 37 °C and 5% CO2. For small interfering RNA (siRNA) experiments in MCF-7 and/or T-47D cell lines, cells were transfected for 48 h with PRMT6-specific siRNA or control siRNA duplexes at a final concentration of 10 nM using RNAiMAX (Invitrogen, Victoria, Australia), with six biological replicates for each treatment regime, as described previously (Harrison et al. 2010). siRNA sequences were PRMT6-siRNA-1, sense 5′-CGGGACCAGCUGUACUACGTT-3′, PRMT6-siRNA-1, antisense 5′-CGUAGUACAGCUGGUCCCGTT-3′, PRMT6-siRNA-2, sense 5′-GCACUUGUAUUUCCGUAUATT-3′, PRMT6-siRNA-2, antisense 5′-UAUACGGAAAUACAAGUGCTT-3′. A siRNA control, Silencer Select negative control #1 (Ambion, Texas, USA), was used as a negative control (control siRNA), which has a proprietary unreleased sequence.
Exon microarray data analysis
Labelling and hybridisation of the Affymetrix Exon 1.0 ST Arrays were performed by the Ramaciotti Centre for Gene Function Analysis (UNSW, Sydney, Australia) following the recommended Affymetrix protocols. Exon array data were analysed using XRAY excel array expression analysis software (Biotique Systems, Inc., Nevada, USA). Normalized data and raw gene expression files have been submitted to the NCBI Gene Expression Omnibus (GEO; Barrett et al. 2005), accession number: GSE36542.
Input files were normalised with full quantile normalisation to remove variation produced amongst samples during array labelling, hybridisation and scanning. Principal component analysis was performed using GeneSpring GX version 11 (Agilent, Victoria, Australia) on each sample to confirm similarity between samples with the same treatment. Principal component analysis scores were used to form a three-dimensional plot of each sample.
Probe sets then underwent filtering (to ensure that only high quality probe sets were included in the analysis), GC content-based probe screening, log transformation and background correction using GC-matched probes. Individual probe sets that are not significantly expressed above background can produce false positives for alternative splicing and so were excluded from tests of alternative exon usage. Probe sets are ranked as Core, Extended or Full depending on their reliability. Only Core probe sets, which are derived from high-quality genomic features like RefSDefault (www.ncbi.nlm.nih.gov) or Ensembl (www.ensembl.org), were used in the analysis. The expression score for each probe set was then derived via Tukey's median-polish and Mixed Model, and Nested ANOVA was then used to identify changes in gene expression or alternative splicing between groups. Alternative splicing results were corrected using the Benjamini and Hochberg False Discovery Rate to remove false positives. Genes were considered to be differentially expressed or alternatively spliced when P<0.01 (Supplementary Tables 2 and 3).
To determine splice events within significantly alternatively spliced genes, probe sets showing ‘non-parallel’ lines when average group expression is plotted over the probe sets were identified. Probe sets displaying ‘non-parallel’ expression were assigned to a particular exon using X:map (http://xmap.picr.man.ac.uk/). Changes in exon-specific splicing were then compared to known alternative splicing events using X:map (built using the Ensembl database (www.ensembl.org)) and FAST DB (http://www.fast-db.com/fastdb2/frame.html), built by combining databases within www.ensembl.org, http://genome.ucsc.edu and www.ncbi.nlm.nih.gov.
RNA isolation, reverse transcription and quantitative real-time PCR
Total RNA was extracted from cells using Trizol Reagent (Invitrogen) and purified using RNeasy RNA columns, with on-column DNase treatment (Qiagen, Victoria, Australia). RNA was normalised using u.v. spectrophotometry and agarose gel electrophoresis. RNA integrity was determined using RNA 6000 Nano chips (Agilent) on a 2100 BioAnalyser (Agilent) according to the manufacturer's instructions. cDNA was synthesized from 600 ng of total RNA using TaqMan reverse transcription reagents, according to the manufacturer's instructions (Applied Biosystems, Victoria, Australia).
Gene expression levels were determined by 384-well quantitative real-time PCR (Q-RT-PCR) on a 7900HT Fast RT-PCR System (Applied Biosystems) as described previously (Harrison et al. 2010). RT-PCR was performed using Assay on Demand TaqMan primer/probes (Applied Biosystems) or Sybr green; PRMT6 (Hs00250803_s1) CARM1 (Hs00406354_m1), PRMT1 (Hs01587651_g1), NCOA4 (Hs00428331_g1), FKBP4 (Hs00427038_g1), HIPK3 (Hs00178628_m1), ARL6IP1 (Hs00760013_s1), YWHAE (Hs00356749_g1), MINA (Hs00262155_m1), PTEN (Hs02621230_s1), IGFBP3 (Hs00181211_m1), THEM4 (Hs00940013_g1), BRCA2 and CDKN1A interacting protein α (BCCIPα; Hs01567098_m1) and BCCIPβ (Hs01555220_m1). All expression levels were normalised to 18S RNA or RLP0 expression levels using TaqMan primer/probe sets (Applied Biosystems). Target cDNA levels were analysed by Q-RT-PCR in 10 μl reactions using TaqMan PCR master mix (Applied Biosystems), TaqMan probe/primer sets and cDNA (5% of the starting 600 ng of RNA). PCR was initiated at 95 °C for 10 min to activate Amplitaq Gold DNA polymerase followed by 45 cycles of 95 °C for 15 s and 60 °C for a 1 min two-step thermal cycling. Relative changes in gene expression were calculated using the ΔΔCt method. For each experiment, six replicate wells of cells were used for each experimental condition and all experiments were repeated twice.
Alternative splicing validation of ninein (GSK3B interacting protein) (NIN), catenin (cadherin-associated protein)-delta 1 (CTNND1) and thioesterase superfamily member 4 (THEM4) was performed by Q-RT-PCR on a 7900HT Fast RT-PCR System (Applied Biosystems) in a 10 μl reaction using SYBR-green master mix (Applied Biosystems) using manufacturer's instructions. Primers, which were designed to specifically amplify the alternatively spliced isoforms, were NIN inclusion – forward 5′-GAGTCACTATTTTAGTTAAGC-3′, NIN inclusion – reverse 5′-AGATTCATATCTCAGAAGTTC-3′, NIN skipping – forward 5′-GAACTAGAGCAGTTTCACCAG-3′, NIN skipping – reverse 5′-CAGATTCATATCTTCCCTCCT-3′, CTNND1 inclusion – forward 5′-ACACCCTTGATGCAGGACGAG-3′, CTNND1 inclusion – reverse 5′-AATCTTCTGCATGGAGGGGTA-3′, CTNND1 skipping – forward 5′-ACAACTATTCCACACCAAATG-3′, CTNND1 skipping – reverse 5′-AATCTTCTGCATCAAGGGTGT-3′, THEM4 inclusion – forward 5′-CATTTTCTTGCCTGGTTAATG-3′, THEM4 inclusion – reverse 5′-GTTGGTCAGGCTGGTCTTGAA-3′ and THEM4 skipping – forward 5′-ACAAGGACCTAAGACTGCTCT-3′, THEM4 skipping – reverse 5′-AAGCTTTGGGTCAAGAAAATG-3′. RT-PCR was initiated at 95 °C for 10 min to activate Amplitaq Gold DNA polymerase followed by 45 cycles of 95 °C for 15 s and 56 °C for a 1 min two-step thermal cycling. Relative changes in gene expression were calculated using the ΔΔCt method. For each experiment, six replicate wells of cells were used for each experimental condition and all experiments were repeated twice.
Melt curve analysis was performed to ensure that only one product was amplified in the Q-RT-PCRs. To determine the specificity of the amplified product, PCR fragments amplified from MCF-7 cell cDNA using the above primer sets were sub-cloned into pCR2.1 using the TOPO-TA cloning kit (Invitrogen) and sequenced. We also verified that the Q-RT-PCR assay produces a linear standard curve by plotting the log of the starting quantity of the template against the Ct value after serially diluting the pCR2.1-cloned PCR products and analysing them by Q-RT-PCR using the appropriate primers. We also verified that primer sets designed to detect inclusion do not detect the pCR2.1-cloned skipping spliced product, and primer sets designed to detect skipping do not detect the pCR2.1-cloned PCR product including the variable exon in Q-RT-PCR assays. Statistical analysis of RT-PCR data was performed using either one- or two-tailed Student's t-tests as appropriate.
Human breast tissue samples
Data and de-identified breast tissue samples (either fresh frozen tissue or purified total RNA) were obtained from the tissue banks listed in the Acknowledgements section. The breast tissue samples used were from individual cases of primary invasive ductal carcinoma (IDC) or normal breast samples, with no known history of breast disease, were collected following breast reduction mammoplasty or from women who had volunteered normal breast biopsy tissue. This study was approved by the human research ethics committees of all participating institutions.
Q-RT-PCR of human breast tissue samples
Commercial micro-fluidic cards, the TaqMan Low-Density Arrays (TLDA, Applied Biosystems) and Q-RT-PCR using 384-well 7900HT Fast RT-PCR System were used to profile gene expression for breast tissue samples. The geNorm software imbedded within the ABI/intergromics StatMiner V4.1 software package was used to compute least expression variation and select the most appropriate, stable and robust combination of internal control genes with which to normalise the expression data (against the mean of the most stable controls). TLDAs were used according to the manufacturer's instructions and analysed as described previously (Myers et al. 2009, Raichur et al. 2010). The TaqMan Gene Expression Assays used as controls were MRPL19 (Hs00608519_m1), PGK1 (Hs99999906_m1), PPIA (Hs99999904_m1), TFRC (Hs99999911_m1) and UBC (Hs00824723_m1) median. The TaqMan Gene Expression Assays used were AKR1C2;AKR1C1 (Hs00413886_m1), SLIRP (Hs00364015_m1), EDF1 (Hs00610152_m1), FKBP4 (Hs00427038_g1), HIPK3 (Hs00178628_m1), NCOA4 (Hs00428328_m1), NQO1 (Hs00168547_m1), PA2G4 (Hs00854538_g1), PRKCD (Hs00178914_m1), PTEN (Hs00829813_s1) and TRIP13 (Hs00188500_m1). For each sample, 1.5 μg of total RNA was reverse transcribed using random hexamers with SuperScript III reverse transcriptase (Invitrogen) in a total volume of 45 μl. A volume of DNase/RNase-free water and an equal volume of TaqMan universal master mix was added to each TLDA fill reservoir. Each GCT sample was run once whilst the cell lines were run three times. Four reservoirs per sample were filled. The TLDA includes all gene assays and endogenous controls in triplicate. After sealing the plate, it was run on an ABI 7900HT Real-Time instrument.
Examination of alternative splicing in breast tissue samples was done by 384-well Q-RT-PCR as described earlier, using IPO8 (Hs00183533_m1) and TBP (Hs00427620_m1) median as the controls for that specific cohort of samples. TaqMan primer/probes (Applied Biosystems) used were BCCIPα (Hs01567098_m1), BCCIPβ (Hs01555220_m1), NINEIN-inclusion – forward: 5′-CACGTGCAGTGAGATGCAGC-3′, NINEIN-inclusion – reverse: 5′-CTTCCTGAGATCCATTTAATGTCC-3′, NINEIN-inclusion-Probe: 5′-6-FAM-TGAACTTCTGAGATATGAAT-MGB-3′, NINEIN-skipping – forward: 5′-GAGCTCTTGGAAAAGCACCAA-3′, NINEIN-skipping – reverse: 5′-CTTCCTGAGATCCATTTAATGTCC-3′, NINEIN-skipping-Probe: 5′-6-FAM-CAGGAGGGAAGATATGAATC-MGB-3′, THEM4-inclusion – forward: 5′-ACACCTACTGAATGGATTCAAGACT-3′, THEM4-inclusion – reverse: 5′-AGGGATCCGCCTGCCTTG-3′, THEM4-inclusion-Probe: 5′-6-FAM-CATTTTCTTGCCTGGTTA-MGB-3′, THEM4-skipping – forward: 5′-ACGTACACCTACTGAATGGATTC-3′, THEM4-skipping – reverse: 5′-CTGGGCCTGTGACATTTGTTC-3′, THEM4-skipping-Probe: 5′-6-FAM-CATTTTCTTGACCCAAAGC-MGB-3′. TaqMan primer/probes (Applied Biosystems) for SYK have been described previously (Harrison et al. 2010).
The TLDAs and Q-RT-PCR of breast tissue samples were analysed using the relative quantification method of Ct and the normalised Ct values (ΔCt) of the replicates between the control/calibrator and the target sample (normal breast) were compared as described previously (Myers et al. 2009, Raichur et al. 2010).
Analysis of breast cancer microarray datasets
Statistical analysis was performed in R (http://www.r-project.org/). Gene expression datasets were retrieved from Bioconductor (http://bioconductor.org). The gene signature used comprised the 159 genes that were differentially expressed (i.e. up- and down-regulated) in a significant manner relative to controls (≥1.3-fold, P<0.01) as identified in exon array analysis (Supplementary Table 2, see section on supplementary data given at the end of this article). Genes listed in Supplementary Table 2 that did not have a fold change ≥1.3 but did have a probability of differential expression P<0.01 (highlighted in grey) were not included in the signature. Experimental data packages examined include BreastCancerUPP (Miller et al. 2011), BreastCancerNKI (van de Vijver et al. 2002) and BreastCancerTRANSBIG (Desmedt et al. 2008). PRMT6 signature scores were computed using the sig.score function from the Genefu package (Haibe-Kains et al. 2010a). This function computes a signature score from a gene list (aka gene signature), i.e. a signed average as previously published (Sotiriou et al. 2006, Haibe-Kains 2009). Briefly, positive or high signature scores are indicative of PRMT6 dysfunction (and by inference low PRMT6 expression levels), whereas negative or low signature scores are indicative of normal PRMT6 function (and by inference high PRMT6 levels). Survival analysis was performed using the SurvComp package (Haibe-Kains et al. 2008, 2010b). PRMT6 signature scores were split according to their 33% and 66% quantiles. Upper and lower quantiles were used for survival analysis. We considered relapse-free survival (RFS) as the survival endpoint. When RFS was not available, we used distant metastasis-free survival (DMFS) data. All survival data were censored at 10 years. Survival curves were based on Kaplan–Meier estimates.
siRNA depletion of PRMT6 and detection of genome-wide exon expression
MCF-7 breast cancer cells were transfected with a validated siRNA targeting PRMT6 (PRMT6-siRNA-1) or a control siRNA, as previously reported (Harrison et al. 2010). PRMT6 siRNA-1 specifically reduced mRNA levels of PRMT6 by ∼80% but did not affect the levels of PRMT1 or CARM1 (Fig. 1A). Alterations in gene transcription and alternative splicing were determined by hybridisation of labelled cDNA onto GeneChip Human Exon 1.0 ST Arrays (Affymetrix). Following full quantile normalisation, principal component analysis was performed, with exon arrays clustering into groups depending on the treatment regime (Fig. 1B). This result demonstrates that replicates of the same siRNA treatment produced a similar exon expression profile but a distinctly different profile in relation to the siRNA treatment based on clustering distance. The data were filtered by removing less reliable probes, probes with a low or high GC content, and probes that were not significantly expressed above background. The effects of the filtering are summarised in Supplementary Table 1, see section on supplementary data given at the end of this article. Of the different alternatively spliced products (referred to as transcript clusters) available for analysis following background correction, 3316 were expressed in both treatment groups.
PRMT6 influences both gene activation and repression
Analysis from genome-wide exon expression profiling of whole transcripts identified 159 genes that had a fold change ≥1.3 and were significantly differentially expressed compared to controls (P<0.01; Supplementary Table 2). Of these 159 genes, 121 were down-regulated and 38 were up-regulated in response to PRMT6 knockdown. This demonstrated that attenuation of PRMT6 expression resulted in the activation and repression of downstream gene expression. The differentially expressed genes were analysed by Ingenuity Pathway Analysis in order to identify functional groups regulated by PRMT6. Several functional groups were found to be over-represented, including endocrine system development and function, small molecule biochemistry, cell death and cancer, indicating that PRMT6 may regulate these processes (Fig. 2A).
In order to validate the gene expression results obtained from the exon array, eight genes were chosen for re-examination by Q-RT-PCR in the ERα+ MCF-7 and T-47D cell lines. These genes, taken from the list of 159 differentially expressed genes, comprised six genes with decreased expression and two with increased expression following PRMT6 knockdown. The genes were chosen for their biological significance and had relevance to either steroid hormone-dependent gene expression (FK506 binding protein 4, 59 kDa: FKBP4; homeodomain interacting protein kinase 3: HIPK3 and nuclear receptor co-activator 4: NCOA4) (Yeh & Chang 1996, Moilanen et al. 1998, Riggs et al. 2003) and/or cancer (NCOA4, ADP-ribosylation factor-like 6 interacting protein 1: ARL6IP1; MYC-induced nuclear antigen: MINA; tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, epsilon polypeptide: YWHAE; phosphatase and tensin homolog:PTEN and insulin-like growth factor binding protein 3: IGFBP3) (Eng 2003, Hu et al. 2004, Teye et al. 2004, Cimino et al. 2008, Peng et al. 2008, Guo et al. 2010, Ingermann et al. 2010). PRMT6 levels were decreased in MCF-7 or T-47D breast cancer cells by siRNA for 48 h. We conducted further validation, utilising two (different) non-overlapping siRNAs targeting PRMT6 (individually) to decrease the likelihood of observing non-specific off-target effects. These two siRNAs have been previously validated and do not affect PRMT1 or CARM1 RNA levels (Fig. 2B and C; Harrison et al. 2010). Q-RT-PCR following knockdown of PRMT6 with both siRNAs gave similar results to the exon array, with the expression of all eight genes approximating the fold changes observed in the exon array in both MCF-7 and T-47D cells (Fig. 2D and E).
PRMT6-dependent alternative splicing
Analysis of genome-wide exon expression profiling of individual exons identified 449 genes as being alternatively spliced compared with controls after multiple testing correction with the Benjamini and Hochberg False Discovery Rate (P<0.01; Supplementary Table 3). Analysis identified that 91 of these genes were also significantly differentially expressed, demonstrating that PRMT6 can regulate both transcription and pre-mRNA processing of the same gene (Fig. 3A). The 449 differentially alternatively spliced genes identified in the array were analysed by Ingenuity Pathway Analysis to determine cellular pathways that may be regulated by PRMT6-dependent alternative splicing. Several pathways were over-represented, including cell death, cell cycle and cancer (Fig. 3B).
In order to validate the alternative splicing array data, four genes that indicated differential exon alternative splicing events were chosen from the list of 449 alternatively spliced genes by their biological relevance to cancer, namely NIN, CTNND1, THEM4 and BCCIP. NIN plays a role in the cell cycle and is involved in preventing microtubule aster-organisation defects during microtubule nucleation (Abal et al. 2002, Dammermann & Merdes 2002, Stillwell et al. 2004). CTNND1 is thought to have a metastatic promoting role in breast cancer (van Roy & McCrea 2005), THEM4 is involved in apoptosis (Parcellier et al. 2009) and BCCIP is involved with inhibiting breast cancer cell growth (Meng et al. 2004). The details of the genes chosen for validation are shown in Table 1. PRMT6 knockdown coupled with exon-specific Q-RT-PCR was used to examine the alternative splicing of these genes. Results are described in the sections below.
Genes analysed by Q-RT-PCR to validate differential alternative splicing observed in exon array. The P value (corrected) is the probability of differential alternative splicing (corrected by the Benjamini and Hochberg False Discovery Rate), and rank relates to the position of the gene amongst all genes in the array when ranked for P value (corrected)
|Gene symbol||Gene name||P value (corrected)||Rank|
|NIN||NINEIN (GSK3B interacting protein)||2.1×10−34||2|
|CTNND1||Catenin (cadherin-associated protein) delta 1||4.03×10−14||10|
|BCCIP||BRCA2 and CDK1NA interacting protein||3.18×10−8||52|
|THEM4||Thioesterase superfamily member 4||1.09×10−7||60|
PRMT6-dependent regulation of NIN exon skipping
Comparison of probe sets within the NIN gene indicates that NIN may be alternatively spliced following PRMT6 knockdown. Seven probe sets (probe sets 24–30) displayed higher than expected expression following PRMT6 knockdown, indicative of potential differential alternative splicing (Fig. 4A). The NIN gene is composed of 34 exons, and all 7 ‘non-linear’ probe sets map to exon 19, indicating that knockdown of PRMT6 leads to preferential inclusion of exon 19. Two transcripts within the Ensembl database display skipping of NIN exon 19 (ENST00000389868 and ENST00000382043), and this underscores alternative splicing of exon 19 and is an identified biologically relevant RNA processing event.
In order to validate the regulation of NIN alternative splicing by PRMT6, we designed primers to detect inclusion and skipping of NIN exon 19 for use in Q-RT-PCR (Fig. 4B). Following reduction of PRMT6 levels by PRMT6-siRNA-1 or PRMT6-siRNA-2 in MCF-7 and T-47D breast cancer cells, we saw ∼3- to 3.3-fold increase in inclusion and a ∼2.3-fold decrease in skipping of NIN exon 19, leading to an increase in the inclusion: skipping ratio of approximately sevenfold (Fig. 4C, D, E and F). This is in agreement with the increase in inclusion of NIN exon 19 observed in the exon array following PRMT6 knockdown and confirms that a cellular function of PRMT6 is to promote NIN exon 19 skipping.
PRMT6 regulates a variety of alternative splicing events
Examination of median probe set scores from the exon expression profiling indicates that PRMT6 regulates the inclusion of CTNND1 exon 20, alternative splicing of exon 3 and overall transcript levels of THEM4, as well as BCCIP isoform production by alternative terminal exon choice (Supplementary Figures 1A, 2A, B and 3A, see section on supplementary data given at the end of this article). The alternative splicing of specific exons within the CTNND1, THEM4 and BCCIP genes identified from the exon array is supported by alternative transcripts present in the Ensembl databases. This indicates that these alternative splicing events are common and previously identified biologically relevant RNA processing events.
We designed Q-RT-PCR assays to detect these various alternative splicing events for CTNND1, THEM4 and BCCIP in the MCF-7 and T-47D breast cancer cell lines (Supplementary Figures 1B, 2C and 3B). Knockdown of PRMT6 and Q-RT-PCR analysis confirmed a ∼2.2-fold increase in skipping of CTNND1 exon 20, a >1.5-fold decrease in expression and preferential inclusion of THEM4 exon 3 an approximate twofold increase in the BCCIPβ and a concomitant twofold decrease in the levels of BCCIPα (Fig. 5A, B and C). These results show significant changes in the alternative splicing of specific exons within the CTNND1, THEM4 and BCCIP genes (Supplementary Figures 1C, D, E, 2D, E, F and 3C, D and E). This indicates that one cellular function of PRMT6 is to influence a variety of selective alternative splicing events, which may have pathophysiological consequences in several cancer-related signalling pathways.
Differential expression of PRMT6 transcriptional and alternatively spliced genes in primary breast cancer tumours
We assessed the expression of the mRNA encoding PRMT6 in a primary breast IDC cohort (n=66, ER+ and ER−) relative to a normal breast tissue cohort (n=50). We examined PRMT6 mRNA levels by Q-RT-PCR, and observed significantly decreased PRMT6 mRNA expression in the IDC breast cancer cohort relative to normal breast (Fig. 6A). Furthermore, the expression of the (in vitro identified) PRMT6-dependent breast cancer-related genes (Table 2) (SRA stem-loop interacting RNA binding protein: SLIRP; endothelial differentiation-related factor 1: EDF1; FKBP4, HIPK3, NCOA4, NAD(P)H dehydrogenase, quinone 1: NQO1; proliferation-associated 2G4: PA2G4; protein kinase C, delta: PRKCD; PTEN and tudor domain containing 3: TRIP13) demonstrated significant and differential expression, as observed after PRMT6 mRNA attenuation in breast cancer cell lines (Fig. 6B). The only exceptions were PTEN and TRIP13, which showed a significant yet opposite expression pattern in comparison with the in vitro results (see Fig. 2).
Genes identified as being transcriptionally regulated by PRMT6 and examined in human breast tissue: cancer relevance
|Gene symbol||Cancer relevance||Reference|
|AKR1C2; AKR1C1||Progesterone metabolism||Ji et al. (2004)|
|SLIRP||Nuclear receptor co-repressor||Hatchell et al. (2006)|
|EDF1||Sensitivity to tamoxifen resistance||Mendes-Pereira et al. (2012)|
|FKBP4||Breast cancer prediction marker||Yang et al. (2012)|
|HIPK3||Resistance to Fas receptor-mediated apoptosis||Curtin & Cotter (2004)|
|NQ01||Nitrosamine metabolism, dysregulated in nasopharyngeal carcinoma||Dodd et al. (2006)|
|PA2G4||Inhibition of heregulin-mediated MCF-7 breast cancer cell growth||Zhang et al. (2008)|
|PRKCD||Pro-proliferative factor in oestrogen-dependent breast cancer cells||McCracken et al. (2003)|
|PTEN||Tumour suppressor gene||Chow & Baker (2006)|
|TRIP13||Interacts with thyroid hormone receptor, lung cancer marker||Kang et al. (2008)|
We then examined the four cancer-related genes we identified and validated to be alternatively spliced upon PRMT6 knockdown in MCF-7 and T-47D breast cancer cell lines. Examination of the IDC breast cancer cohort (n=10) and normal breast tissue (n=12) by Q-RT-PCR revealed changes in the exon usage of three of our previously identified alternatively spliced genes in the IDC cohort relative to normal breast tissue (Fig. 7A, B and C). We found increased NIN exon 19 inclusion, decreased skipping of THEM4 exon 3, increased levels of the BCCIPβ isoform and no significant change in CTNND1 exon usage (data not shown). Our previous study (Harrison et al. 2010) demonstrated PRMT6-dependent alternative splicing of the breast cancer tumour suppressor SYK to be regulated by PRMT6 in breast cancer cell lines. In that study, we found reduction of PRMT6 in breast cancer cells resulted in an increase in the full-length SYK transcript (SYK-L) over the shorter SYK (SYK-S) transcript, which skips exon 7. Interestingly, we found similar increase in the SYK-L mRNA transcript expression in our IDC breast cancer samples (Fig. 7D).
Expression of the PRMT6-dependent gene signature predicts disease outcome
To investigate and expand upon the clinical relevance of our findings, we evaluated a PRMT6 transcriptionally regulated gene signature in the context of clinical outcome associations in three independent publically available breast cancer datasets with available clinical outcomes (van de Vijver et al. 2002, Desmedt et al. 2008, Miller et al. 2011). To determine whether PRMT6-dependent gene expression is associated with disease recurrence (RFS) and/or metastasis (DMFS), we utilised a PRMT6-dependent gene expression signature based on the 159 genes that were differentially expressed in a significant manner relative to controls (≥1.3-fold, P<0.01; see Supplementary Table 1). PRMT6 gene expression signature scores were calculated for each patient tumour sample. The PRMT6-dependent gene expression signature is calculated as a signed average, with the sign of the PRMT6 gene expression signature score providing an indication of PRMT6 dysfunction (and by inference the level of PRMT6 expression). For example, a negative (i.e. low) PRMT6 gene expression signature score indicates that the genes comprising the PRMT6 gene expression signature and having a negative weight (i.e. the genes significantly down-regulated in the PRMT6 knockdown cells) are up-regulated relative to the PRMT6 signature genes having a positive weight (i.e. the genes significantly up-regulated in the PRMT6 knockdown cells). Accordingly, a negative or low PRMT6-dependent signature score is indicative of normal to high PRMT6 expression levels and a positive or high PRMT6-dependent signature score is indicative of PRMT6 dysfunction (i.e. low PRMT6 expression levels). In keeping with this notion, we observed a significant inverse correlation between the PRMT6 gene expression levels and PRMT6-dependent signature scores in the NKI and UPP breast cancer datasets (Supplementary Figure 4A and B, see section on supplementary data given at the end of this article).
Patient samples were stratified according to their PRMT6-dependent gene signature scores (high scores ≥66% quantile; low scores ≤33% quantile). Kaplan–Meier survival analysis for RFS or DMFS (Fig. 8A, B and C, Supplementary Figures 5 and 6, see section on supplementary data given at the end of this article) clearly demonstrated that patients falling within the ER+ subgroup and having low PRMT6 signature scores (i.e. indicative of normal PRMT6 function/normal to high PRMT6 expression levels) exhibited significantly lower (P<0.05 in all cohorts analysed) probabilities of RFS and metastasis-free survival when compared to patients falling within the ER+ subgroup and having high PRMT6 signature scores (i.e. indicative of PRMT6 dysfunction/low PRMT6 expression levels). Accordingly, the PRMT6-dependent gene signature is significantly associated with clinical outcome in the ER+ subgroup and suggests the PRMT6-dependent genetic program plays a significant role in breast cancer disease progression.
We used Affymetrix exon array in combination with RNA interference technology to investigate the global effects of PRMT6 on transcription and alternative splicing in breast cancer cells and human breast cancer. Before this study, the role of PRMT6 in regulating transcription was controversial. We previously identified PRMT6 as a SHR co-activator capable of promoting the transcription of oestrogen- and progesterone-dependent genes (Harrison et al. 2010). However, earlier studies demonstrated a role for PRMT6 in repressing transcription by methylating H3R2 (Guccione et al. 2007, Hyllus et al. 2007, Iberg et al. 2008). In order to resolve this controversy, we examined the effects of knocking down PRMT6 on global transcription in breast cancer cells. We found that PRMT6 plays a dual role and can both co-activate and co-repress transcription. This is in agreement with a recent study in lung and bladder cell lines, in which PRMT6 functioned primarily as a transcriptional enhancer but could also repress transcription of a few genes (Yoshimatsu et al. 2011). The ability of PRMT6 to co-activate or co-repress transcription of a given gene, as demonstrated in our knockdown studies, may be due to direct or indirect effects. Regardless of the mode of action, our study highlights the fact that the PRMT6-dependent pathways do not exclusively modulate gene expression in a positive or negative fashion. Therefore, there is no conflict between the previous studies investigating the role of PRMT6 in gene expression, with the PRMT6-dependent transcriptional outcome likely to be gene dependent.
While the exact mechanism of PRMT6-dependent gene activation is still unknown, it has been shown that PRMT6 can methylate histone H4R3, a target that is also methylated by PRMT1 during oestrogen-induced gene activation (Strahl et al. 2001, Wagner et al. 2006, Hyllus et al. 2007). H4R3 methylation is read by tudor domain containing 3, an effector molecule that has transcriptional co-activator properties (Yang et al. 2010). It is possible that chromatin architecture at promoter/enhancer regions or previously laid down chromatin modifications determine whether the PRMT activity enhances or inhibits transcription.
Pathway analysis of PRMT6-dependent transcriptionally regulated genes indicates that PRMT6 may play a role in many cellular pathways, including those implicated in tumourigenesis. PRMT6 has been shown to be required for the G1-S transition and proliferation of lung and bladder cancer cells and is aberrantly regulated in a number of cancers, including breast cancer (Yoshimatsu et al. 2011). In addition, we have previously demonstrated that PRMT6 is required for maximal oestrogen-stimulated proliferation of MCF-7 breast cancer cells (Harrison et al. 2010).
We validated the results from the exon array by examining PRMT6-mediated changes in gene expression of eight genes, six of which have known roles in cancer. NCOA4, YWHAE, ARL6IP1 and MINA are overexpressed in a number of different cancers (Hu et al. 2004, Teye et al. 2004, Cimino et al. 2008, Peng et al. 2008, Guo et al. 2010). The up-regulation of YWHAE in breast cancer is associated with disease-free and overall survival, and therefore, the down-regulation of YWHAE from reduced PRMT6 levels may be unfavourable in relation to breast cancer prognosis (Cimino et al. 2008). Conversely, down-regulation of PRMT6 expression and corresponding low expression of ARL6IP1 and MINA may be favourable to decreased breast cancer growth.
In our study, PTEN and IGFBP3 expression were increased when we knocked down PRMT6 in the MCF-7 and T-47D cell lines. PTEN is a tumour suppressor gene that inhibits the PI3K pathway (Chow & Baker 2006). Loss of PTEN leads to the activation of a number of kinases and consequent cell cycle progression (Gera et al. 2004, Guertin & Sabatini 2005). IGFBP3 has an anti-proliferative effect and induces apoptosis in breast cancer cells (Kim et al. 2004, 2010). This indicates that lower PRMT6 expression may result in increased PTEN and IGFBP3 expression, decreased cell cycle progression and increased breast cancer cell apoptosis.
We have demonstrated that PRMT6 has roles in steroid hormone-dependent transcription (Harrison et al. 2010), and pathway analysis demonstrated that PRMT6 regulates the transcription of genes involved in endocrine system development and function. HIPK3 and NCOA4 are AR co-activators (Yeh & Chang 1996, Moilanen et al. 1998), and FKBP4 promotes binding of hormone to the GR (Riggs et al. 2003). This indicates that in addition to co-activation of steroid hormone-dependent genes, PRMT6 is involved in the expression of genes that promote steroid hormone signalling. The reduction of PRMT6 in the MCF-7 and T47D ER+ cell lines results in a decrease in HIPK3 and NCOA4, two transcriptional co-activators primarily associated with androgen receptor (AR) signalling. Interestingly, AR signalling in ER+ breast cancer is thought to be associated with favourable outcomes (Castellano et al. 2010, Park et al. 2011), and the reduction of HIPK3 and NCOA4 may have a negative effect in relation to AR signalling and outcomes in ER+ breast cancer. In contrast, reduction of the full-length isoform of NCOA4 in MCF-7 cells is associated with increased breast cancer metastasis (Wu et al. 2011), while knockdown of NCOA4 in MCF-7 cells results in decreased cell proliferation (Hua et al. 2008). The exact role NCOA4 plays in breast cancer is not fully defined and requires further investigation.
PRMT6 functions as a tertiary co-regulator, and from our microarray data, is able to modulate the expression of a large number of genes. In the context of breast cancer, the expression of several of the examined PRMT6-regulated oncogene/tumour suppressor genes indicates that reduction of PRMT6 and corresponding PRMT6-dependent gene changes may be involved in breast carcinogenesis.
Analysis of the exon array data identified 449 genes that are alternatively spliced in response to PRMT6. Of these genes, 121 were also regulated transcriptionally by PRMT6. Therefore, PRMT6 can have a major impact on the function of particular genes by simultaneously regulating both their expression levels and final protein composition. However, the fact that not all genes overlap in the two groups demonstrates that PRMT6-regulated transcription and alternative splicing are not necessarily inter-dependent processes.
Pathway analysis of the alternatively spliced genes regulated by PRMT6 indicates that PRMT6 may play roles in cancer, gene expression, cell death and cell cycle progression. Therefore, it would appear that dysregulation of PRMT6 may lead to tumourigenesis by causing aberrant alternative splicing as well as by deregulating transcription. NIN plays a key role in centrosome function, which may be regulated by PRMT6-directed alternative splicing. PRMT6 promotes skipping of NIN exon 19, which removes a region important in the centrosomal targeting ability of NIN. Therefore, removal of this region is likely to interfere with NIN chromosomal targeting and γ-tubulin binding, resulting in mitotic abnormalities (Abal et al. 2002, Dammermann & Merdes 2002, Stillwell et al. 2004, Lin et al. 2006).
CTNND1 exon 20 encodes a functional leucine-rich nuclear export signal (Keirsebilck et al. 1998, van Hengel et al. 1999), and so PRMT6-directed inclusion of this exon will lead to cytoplasmic CTNND1. CTNND1 is often located in the cytoplasm but rarely in the nucleus of tumour cells, and cytoplasmic CTNND1 has been linked to a more invasive form of breast cancer (Sarrio et al. 2004, Shibata et al. 2004, Soubry et al. 2005). Whilst the metastatic-promoting role of cytoplasmic CTNND1 is unknown, it is possibly due to its binding to microtubules and/or activation of Rac (Franz & Ridley 2004). In addition, CTNND1 may sequester Kaiso in the cytoplasm, preventing Kaiso-regulated repression of tumour suppressor genes (van Roy & McCrea 2005), leading to a more oncogenic version of CTNND1 being expressed.
THEM4 is a mitochondrial protein that sensitises cells to apoptosis by binding to heat-shock protein 70 and promoting caspase-3 cleavage (Parcellier et al. 2009, Piao et al. 2009). Currently, nothing is known about the effect of inclusion/skipping of exon 3 on THEM4 structure and activity. Considering the importance of THEM4 in the apoptotic pathway, this PRMT6-regulated alternative splicing event is certainly worth further investigation in relation to breast cancer.
BCCIP is a BRCA2 and CDKN1A (Cip1/waf1/p21) interacting protein that plays an important role in cancer aetiology (Ono et al. 2000, Liu et al. 2001). BCCIP is alternatively spliced to form two isoforms (-α and -β) and PRMT6 knockdown leads to the preferential production of BCCIPβ in breast cancer cells. Whilst both BCCIP isoforms inhibit cell growth by interacting with p21, BCCIPβ has a much greater effect on cell growth inhibition, particularly in breast cancer cells (Meng et al. 2004).
We found that PRMT6 expression was substantially down-regulated in IDC compared with normal breast tissue. The recent study by Yoshimatsu et al. (2011) found PRMT6 to be significantly increased in ∼33% of breast cancers. Knowing that breast cancer is not a homogeneous disease, it is not surprising that individual breast cancer tumours demonstrate dissimilar expression of PRMT6. Given our findings and that of Yoshimatsu et al.'s (2011), PRMT6 may be abnormally regulated up or down in breast cancer. Many of the genes we identified as transcriptionally regulated after attenuation of PRMT6 in MCF-7 and T-47D cells also demonstrated changes in expression in IDC breast cancer compared to normal breast tissue, including SLIRP, EDF1, FKBP4, HIPK3, NCOA4, NQ01, PA2G4, PRKCD and PTEN. Similarly, genes we have identified as alternatively spliced by PRMT6 in our array and previous studies (NIN, THEM4, BCCIPβ and SYK) demonstrate changes in alternative splicing in IDC compared to normal breast tissue.
Through the identification of genes regulated by PRMT6 at the gene expression level, we were able to investigate the significance of a PRMT6-dependent gene expression signature in three published breast cancer cohorts (Fig. 8A, B and C, Supplementary Figures 5 and 6). We found that low PRMT6 expression and the corresponding high PRMT6-dependent gene signature correlated with increased probability of relapse-free or metastasis-free survival in ER+ breast cancer. This would indicate that the role of PRMT6 in breast cancer is associated with ER-mediated signalling but extends beyond SHR transcriptional regulation.
PRMT6 can determine transcriptional gene regulation and alternative splicing decisions, including both exon retention and skipping. Therefore, PRMT6 can be considered a master modulator of gene expression, capable of determining both transcript abundance and protein structure. Our findings indicate that PRMT6 mRNA is aberrantly down-regulated in breast cancer and the genes we found regulated by PRMT6 in human breast cancer cells are over-represented by genes involved in important aspects of carcinogenesis. In addition, our PRMT6-dependent gene signature is significantly associated with clinical outcome in ER+ breast cancer and suggests that the PRMT6-dependent genetic program plays a significant role in breast cancer disease progression.
This is linked to the online version of the paper at http://dx.doi.org/10.1530/ERC-12-0100.
Declaration of interest
D H Dowhan and George E O Muscat are co-Chief Investigators on a 2012/2013 grant submission. All other authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.
This work was supported by a National Health and Medical Research Council of Australia (NHMRC) project grant (Grant#511153) to D H Dowhan. M J Harrison was the recipient of an ANZ Trustees PhD Medical Research scholarship. In addition, this research was supported by a grant from the National Breast Cancer Foundation (NBCF), Australia.
Author contribution statement
D H Dowhan conceived the study, performed breast tissue experimental studies, data analysis, participated the study design, coordination and manuscript preparation. M J Harrison participated the study design, performed cell line experimental studies, data analysis and manuscript preparation. N A Eriksson performed breast tissue RNA extraction, experimental studies and data analysis. P Bailey and M A Brown performed public data set mining, Kaplan–Meier survival analysis and data analysis. M A Pearen carried out PCA and Ingenuity analysis on the exon array data. P J Fuller, J W Funder, E R Simpson, P J Leedman, W D Tilley, C L Clarke and G E O Muscat provided breast tissue samples and were involved in the study, design and analysis of transcriptional profiling studies in breast cancer. P Bailey, P J Fuller, J W Funder, E R Simpson, P J Leedman, W D Tilley, M A Brown and C L Clarke assisted in manuscript preparation. G E O Muscat participated in the study design, coordination and manuscript preparation. All authors read and approved the final manuscript.
Breast cancers and normal tissues were provided by Australian Breast Cancer Tissue Bank (BCTB), which is supported by the NHMRC of Australia, the Cancer Institute NSW and the NCBF, or by the Victorian Cancer BioBank Australia, which is supported by the Victorian Government. BCTB tissues and samples were made available to researchers on a non-exclusive basis. Normal breast biopsies were obtained from the Susan G Komen for the Cure Tissue Bank at the IU Simon Cancer Center. The authors thank contributors to the Susan G Komen for the Cure Tissue Bank, including Indiana University, who collected the samples used in this study, as well as patients and their families, whose help and participation made this work possible. The authors would like to thank the Ramaciotti Centre for Gene Function Analysis (University of NSW), for RNA conversion, exon array hybridization and data collection and Dr Paul Leo for initial microarray analysis.
CastellanoIAlliaEAccortanzoVVandoneAMChiusaLArisioRDurandoADonadioMBussolatiGCoatesAS2010Androgen receptor expression is a significant prognostic factor in estrogen receptor positive breast cancers. Breast Cancer Research and Treatment124607–617. doi:10.1007/s10549-010-0761-y.
CiminoDFusoLSfiligoiCBigliaNPonzoneRMaggiorottoFRussoGCicatielloLWeiszATavernaD2008Identification of new genes associated with breast cancer progression by gene expression analysis of predefined sets of neoplastic tissues. International Journal of Cancer1231327–1338. doi:10.1002/ijc.23660.
DoddLESenguptaSChenIHden BoonJAChengYJWestraWNewtonMAMittlBFMcShaneLChenCJ2006Genes involved in DNA repair and nitrosamine metabolism and those located on chromosome 14q32 are dysregulated in nasopharyngeal carcinoma. Cancer Epidemiology Biomarkers & Prevention152216–2225. doi:10.1158/1055-9965.EPI-06-0455.
DutertreMLacroix-TrikiMDriouchKde la GrangePGratadouLBeckSMillevoiSTaziJLidereauRVagnerS2010Exon-based clustering of murine breast tumor transcriptomes reveals alternative exons whose expression is associated with metastasis. Cancer Research70896–905. doi:10.1158/0008-5472.CAN-09-2703.
GeraJFMellinghoffIKShiYRettigMBTranCHsuJHSawyersCLLichtensteinAK2004AKT activity determines sensitivity to mammalian target of rapamycin (mTOR) inhibitors by regulating cyclin D1 and c-myc expression. Journal of Biological Chemistry2792737–2746. doi:10.1074/jbc.M309999200.
Haibe-Kains B 2009 Identification and Assessment of Gene Signatures in Human Breast Cancer. PhD thesis. Universite Libre de Bruxelles
Haibe-Kains B Bontempi B Quackenbush J & Sotiriou C 2010a genefu: Relevant Functions for Gene Expression Analysis Especially in Breast Cancer. R package version 1.0.9. http://CRAN.R-project.org/package=genefu.
Haibe-Kains B Sotiriou C & Bontempi G 2010b survcomp: Performance Assessment and Comparison for Survival Analysis. R package version 1.1.6. http://CRAN.R-project.org/package=survcomp.
IngermannARYangYFHanJMikamiAGarzaAEMohanrajLFanLIdowuMWareJLKimHS2010Identification of a novel cell death receptor mediating IGFBP-3-induced anti-tumor effects in breast and prostate cancer. Journal of Biological Chemistry28530233–30246. doi:10.1074/jbc.M110.122226.
KimHSIngermannARTsubakiJTwiggSMWalkerGEOhY2004Insulin-like growth factor-binding protein 3 induces caspase-dependent apoptosis through a death receptor-mediated pathway in MCF-7 human breast cancer cells. Cancer Research642229–2237. doi:10.1158/0008-5472.CAN-03-1675.
LinCCChengTSHsuCMWuCHChangLSShenZSYehHMChangLKHowngSLHongYR2006Characterization and functional aspects of human ninein isoforms that regulated by centrosomal targeting signals and evidence for docking sites to direct gamma-tubulin. Cell Cycle52517–2527. doi:10.4161/cc.5.21.3404.
MillerTWBalkoJMGhazouiZDunbierAAndersonHDowsettMGonzalez-AnguloAMMillsGBMillerWRWuH2011A gene expression signature from human breast cancer cells with acquired hormone independence identifies MYC as a mediator of antiestrogen resistance. Clinical Cancer Research172024–2034. doi:10.1158/1078-0432.CCR-10-2567.
SarrioDPerez-MiesBHardissonDMoreno-BuenoGSuarezACanoAMartin-PerezJGamalloCPalaciosJ2004Cytoplasmic localization of p120ctn and E-cadherin loss characterize lobular breast carcinoma from preinvasive to metastatic lesions. Oncogene233272–3283. doi:10.1038/sj.onc.1207439.
SotiriouCWirapatiPLoiSHarrisAFoxSSmedsJNordgrenHFarmerPPrazVHaibe-KainsB2006Gene expression profiling in breast cancer: understanding the molecular basis of histologic grade to improve prognosis. Journal of the National Cancer Institute98262–272. doi:10.1093/jnci/djj052.
(D H Dowhan, M J Harrison, N A Eriksson and P Bailey are co-first authors)
(M J Harrison is now at School of Biomedical Sciences, The University of Queensland, Brisbane, Queensland 4072, Australia)