This study was approved by North West 12 Research Ethics Committee (Lancaster) in accordance with the Medical Research Council guidelines (project reference no. Gosden, 10/H1015/2). From 20 primary breast cancer specimens, frozen tissue sections were cut and stained with H&E to assess morphology. Human breast tissues were obtained from Liverpool Tissue Bank as frozen tissues with the full record of the grade of carcinoma, age and hormonal status (ER, PR and HER2) of the samples (Table I).

Frozen tissues (30 mg) from each of the 20 specimens were extracted using RLT buffer containing 2-ME (1 ml buffer and 10 μl of 2-ME) added to the tissues following evaporation of liquid nitrogen.

After centrifugation, RNA, DNA and protein were extracted from the supernatant. The quantity of RNA and DNA was measured by NanoDrop (Labtech, Ringmer, UK). The quality of RNA for each sample was assessed by 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) using Agilent RNA 6000 Nano kit. The RNA integrity number (RIN) for all the samples was >7.0. The quality of DNA and protein was assessed using gel electrophoresis and Bradford assay, respectively.

Total RNA was further cleaned using the RiboMinus concentration modules (cat. no. K1550-05; Invitrogen) according to manufacturer's instructions. Depletion of rRNA from each sample was performed using the RiboMinus Eukaryote kit for RNA-Seq (cat. no. A1083708; Invitrogen) according to manufacturer's instructions. Successful removal of rRNA was confirmed using the Bioanalyzer.

Libraries of rRNA-depleted RNA suitable for sequencing using the SOLiD platform were created using the SOLiD Total RNA-Seq kit (part no. 4445374; Applied Biosystems). In each instance, 100 ng of RNase III digested rRNA depleted RNA was used as input into the library creation and 15 cycles of amplification were employed to produce the final libraries. These libraries were sequenced on the SOLiD platform from Applied Biosystems. Initially sequencing of the 10 bp library barcode was performed followed by 35 and 50 bp paired-end sequencing. Fragments were subjected to paired-end sequencing using 35 and 50 bp reads.

The sequenced paired-end tags were mapped to the reference human genome (hg19) using TopHat (24). This mapping approach ensures that a read, which spans an exon/exon boundary on the mRNA, is mapped to the genome such that it flanks the intervening intronic region (called junctions by TopHat). After alignment, two rounds of filtering were performed, the first to remove any low quality reads (mapping quality <10) and the second to remove any reads that mapped to ribosomal genes. Mappings were examined visually using SAMtools (25) and IGV (26). FPKM measures (gene expression quantified using the fragments per kilobase of exon per million reads mapped-FPKM) were calculated using the following definition: ‘a fragment was counted if both paired end tags of the fragment F3 and F5 are observed or if only one of them (F3 or F5) is observed’. RefSeq co-ordinates were used to define exonic and intronic regions of the gene. Thus two FPKM values were calculated for each gene to measure exonic and intronic expression.

Genes were screened for differential exonic expression between a tumour sample and the paired normal tissue. Normal tissue samples were grouped together to calculate a standard deviation of the FPKM for each gene. A one-sampled t-test was applied for each gene, comparing its expression in a single cancer against the pooled normal tissue. Genes with low p-values indicated a high differential expression. The q-value package in R was applied to take account of multiple testing. The false discovery rate (FDR) was set to 0.001. This was applied to each tumour sample to give a profile of the differentially expressed genes. Of the genes with significantly different exonic expression we distinguished whether they are up or down regulated in the tumour. This entire process was repeated to screen for genes with significant differential intronic expression.

In certain cancer specimens, significantly upregulated genes were clustered together at certain chromosomal locations. These were hypothesised to be the consequence of amplicons. The length of these amplicons was quantified using a method we devised that involves the minimization of a binomial probability. For a total of X genes on a chromosome, then parameter p is defined as Y/X where Y is the number of genes on that chromosome with significantly upregulated expression (intronic, exonic or both). Then, we consider n consecutive genes encompassing the cluster, and count how many of those are significantly upregulated (intronic or exonic or both), let this number be k.

We calculate the binomial probability in the usual way with n trials and k successes. This is performed for different sets of consecutive genes encompassing the cluster; the set with the lowest binomial probability is chosen to represent the amplicon.

The protein level of ERBB2 was evaluated with western blot analysis using two ERBB2 antibodies: mouse monoclonal antibody (ab16901) against 1242–1255 aa in C-terminus and rabbit polyclonal antibody (ab11717) (both from Abcam) against 651–660 aa. Western blotting was performed as described previously (27). The dilution of antibodies are, 1/1,000 for polyclonal and 1/50 for monoclonal antibody. To confirm the presence of different splice variants, we employed PCR (FGFR2), IHC (p53) and western blotting (ERBB2). The presence of small numbers of mutations in our genes of interest such as p53 and BRCA2 was validated using Sanger sequencing.

ERBB2 an oncogenic receptor tyrosine kinase (RTK), is amplified in ~20% of breast cancers (28). Of the 20 breast cancers examined, 5 contained ERBB2 amplification identified by FISH or IHC. The transcriptional consequences of ERBB2 amplification were assessed using the FPKM measure defined as ‘fragments per kilobase of exon per million reads mapped’. Stringent statistical analysis was employed to compare transcriptional expression in exons, introns, UTRs and overlaps on positive and negative strands in cancers with amplified and non-amplified ERBB2 and 835 genes covering the entire chromosome 17 in cancers of all three tumour sub-types and controls. Major differences were identified between the tumours with ERBB2 amplification at 17q11.2-12 and matched controls for ERBB2 expression (Fig. 1). Those with ERBB2 amplification exhibited high levels of intronic as well as exonic expression, whereas all controls were negative. Intronic expression in tumours was not limited to ERBB2. Detailed analysis of 835 genes located on whole chromosome 17 showed intronic tumour-specific sequence overexpression to occur 3.7 times more frequently than that for exons (Table III). Rigorous statistical analysis showed the mean intronic expression between tumours compared with controls was 48.7 (range, 1–208) (FDR 0.001 and >2-fold-change). The mean number of genes with FDR <0.001 exonic expression of the 835 genes on chromosome 17 in each of the tumour subtypes, was 12.7 (range, 0–41). Comparatively few genes had concurrent exonic and intronic expression, and for those genes with both exonic and intronic overexpression the mean was 5.1 (range, 0–32) (Table III).

To investigate whether intronic expression was specific to chromosome 17, or occurred in other breast tumours (including those with ESR1 amplification at 6q25 or 8q24), intronic and exonic gene expression was assessed for chromosomes 6 and 8 in each breast cancer subtype (Table III). Analysis of 2,119 genes spanning the whole of chromosomes 6, 8 and 17 revealed >2-fold-changes for intronic and/or exonic expression, in each of the different cancers and subtypes, but not controls (Table III). Intronic overexpression also occurred in TNCs and was not restricted to ERBB2 and ESR1 amplified tumours (Table III). There was overlap of transcripts between genes on the sense and antisense strands in the ERBB2 amplicon, but occurred only in tumours with 17q amplification. Paired end analysis of sequence overlap between ERBB2 and adjacent gene C17ORF37 on the opposite strand revealed transcriptional overlaps between sense and antisense strands when alternatively spliced ERBB2 and C17ORF37 variants with longer 5′ and 3′UTRs were expressed. These cancer-specific variants extended beyond normal RefSeq variants, creating sequence overlaps that did not occur in controls or tumours without ERBB2 amplification (Fig. 1). The amplicons and breakpoints involving ERBB2 at chromosome 17q11-12 were identified by stringent analysis of exonic and intronic expression of 835 genes on chromosome 17 in all three cancer subtypes and controls (Fig. 1).

ERBB2 amplified tumours were characterised by >2-fold intronic and exonic overexpression of ERBB2; neighbouring genes including MED1, CDK12, STARD3, PGAP3, ERBB2, C17orf37, GRB7, GSDMB, ORMDL3, PSMD3, MED24, MSL1 and CASC3 also showed intronic and exonic overexpres-sion (Fig. 1). Most ERBB2 amplified tumours contained only one amplicon that included ERBB2 at 17q11.2-q12. However, two cancers contained both an ERBB2-amplicon and a second amplicon adjacent to the centromere (221T and 121T). Both amplicons were characterised by genes with >2-fold exonic and intronic expression. Between the two amplicons was a non-amplified region in which none of the genes including the TSG, NF1 overexpressed or amplified (Fig. 1).

Our data revealed 17q amplification to be cancer-specific, occurring only in tumours but not matched controls (Fig. 1). The amplified 17q amplicons varied in size, involved genes with elevated (>2-fold) levels of intronic/exonic expression. Each individual amplicon comprised different breakpoints and genes, indicating a predisposition to amplicon formation in proximal 17q, although this was not site-specific (Fig. 1). In contrast, none of the TSG on chromosome 17 (TP53 at 17p13, NF1 at 17q11.2), nor BRCA1 at 17q21 showed >2-fold exonic or intronic expression in any tumour or control (Fig. 2). The most extreme case was that for the TSG and NF1 located in the region between the two chromosome 17q amplicons. Analysis of exonic and intronic expression in the two tumours with two 17q amplicons revealed that although genes proximal and distal to NF1 in the pericentromeric and ERBB2 amplicons showed >2-fold intronic or exonic overexpression, NF1 was not overexpressed in any tumour or control. Mutation analysis, SNP expression, alternative splicing and subcellular localisation assessed for each of the three TSGs TP53, BRCA1 and NF1 located on chromosome 17 (Fig. 2) revealed that alternative splicing of TP53 generated only 1–2 RefSeq splice variants. The findings for TP53, NF1 and BRCA1 reveal no >2-fold intronic or exonic expression and only RefSeq isoforms were detected in any tumour or control. This was consistent in other major TSGs including TP63/73, BRCA1/2, CDKN2A/2B and RB1/RBL1/ RBL2 and it was in marked contrast to amplified oncogenes including ERBB2 and ESR1 that generated multiple truncated tumour-specific isoforms. Using mutation analysis, a non-synonymous TP53 mutation was detected resulting in the amino acid change Cys106Ser in cancer 086T which was validated and confirmed using Sanger sequencing (Fig. 2). TP63 showed higher expression in controls than tumours, the expected pattern for TSGs (Fig. 2). TP63 expressed RefSeq variant b, an isoform with nuclear localisation and tetramerisa-tion domains conferring nuclear localisation and confirmed by IHC (Fig. 2). TP73 was not expressed in these breast carcinomas.

In normal tissues, BRCA2 expression was significantly lower than in the malignant counterparts (Table IV and Fig. 3). Although no intronic or exonic expression >2-fold was found for BRCA2, FPKM values were higher for exon 11, involved in DNA repair (29), than other exons. BRCA1, expression was significantly lower in 16/20 carcinomas than the controls, with low intronic and exonic expression. Mutations were detected in BRCA2, in contrast to BRCA1 where no mutations were identified (Table IV). Thus a novel and important finding of this study is that TSGs express no more than two RefSeq splice variants in any cancer or control, in contrast to the multiple alternatively spliced transcripts in oncogenes and hormone receptors, suggesting that the transcriptional integrity of TSGs is tightly regulated in cancers, although this offers no protection against mutations or other changes.

Transcriptome analyses for ERBB2 were validated using qPCR to confirm transcript sequences, FISH to identify amplification, IHC to determine subcellular localisations and western blotting to confirm translation of alternatively spliced transcripts into proteins and assess the numbers and sizes of truncated protein isoforms in tumours and controls (Fig. 1). Several carcinomas expressed >30 copies of ERBB2 when compared with matched controls and non-amplified tumours containing only two copies (Fig. 1). Chromosome 17 copy numbers in each cell were characterized using a concurrent second FISH probe to detect chromosome 17 centromeric sequences and hence identify chromosomal losses, gains and ploidy changes. In tumours with both ERBB2 and pericentromeric ampli-cons, FISH revealed chromosome 17 centromeric regions in tumours to be larger than controls (Fig. 1), indicating amplification of the satellite and pericentromeric sequences, in agreement with studies demonstrating transcription of satellite, repeat, retrotransposon and other RNAs in tumouri-genesis (30–33). We investigated intronic overexpression in relation to alternative splicing (AS) by calculating FPKM values for each exon, alternative exon and intron, aligning these with AceView (NCBI HG 19 Human AceView 2010) IGV and UCSC Blat. In carcinomas, when ERBB2 was amplified, multiple highly truncated alternatively spliced variants are transcribed, resulting in the expression of a larger proportion of the genome, particularly intronic sequences not expressed in normal controls. The splice variants contained initiation sites, exons, introns, 5′ and 3′UTRs, differing from the normal RefSeq variants of control samples, many being highly expressed. The transcriptional start and end points were validated using RT-PCR. To demonstrate that NMD pathways were not activated in these shortened transcripts, we used western blot analysis and subcellular localisation analysis to validate that in ERBB2 amplified breast tumours, multiple alternatively spliced truncated transcripts were translated into highly truncated proteins, preponderantly smaller and varying in size, corresponding to the alternatively spliced ERBB2 transcripts identified according to AceView (Fig. 1). In contrast, in normal breast and non-ERBB2 amplified tumours, the full length ERBB2 isoform (149 kDa) was predominant (Fig. 1).

Five of the 20 breast carcinomas expressed ESR1 protein identified by IHC, although only three showed >2 exonic fold-change (Fig. 3). FPKM analysis of exonic and intronic expression of ESR1 at 6q25 and a further 794 genes on chromosome 6 detected a small amplicon at 6q25 comprising the three genes ESR1, C6orf211 and C6orf9 in an ESR1-positive tumour sample, associated with both intronic and exonic overexpression in ESR1 (Figs. 3 and 4). ESR1 amplification was found to be associated with intronic overexpression and alternative splicing (Fig. 5). In the absence of estrogen, ESR1 functions as a nuclear transcription factor that promotes breast cell growth. The receptors are inactive and cytoplasmic. However, on activation by estrogen, they dimerise and translocate to the nucleus. Thus, changes in subcellular localisation of ESR1 from cytoplasm to the nucleus will effect its function.

We also analysed the transcriptional profiles of the other three major hormone receptors ESR2, progesterone receptor (PGR) and androgen receptor (AR). For PGR, 18/20 cancers showed much lower exonic and intronic expression than the controls. The remaining 2 positive cancers were also positive by IHC. AR expression was negative in 12/20 samples but positive in 8/20. All 12 samples negative for AR were triple-negative subtypes.

The 8q24 region is frequently amplified in breast cancers and may be associated with MYC oncogene overexpression (34). In a tumour with 8q24 amplifi-cation, we studied intronic and exonic expression in 488 genes over the whole of chromosome 8, comparing tumours in each of the three subtypes with controls. Many genes contained exonic or intronic overexpression (Table III). Surprisingly, no exonic or intronic overexpression (>2-fold) was detected for MYC in the 8q24 amplified tumour. This finding was striking, because the adjacent gene, MTSS1 (at 8p22), showed >2-fold exonic and intronic expression of 4.1 and 3.7, respectively, whereas MYC did not (Table III and Fig. 4). It was assumed axiomatic that 8q24 amplification would involve MYC overex-pression. However, our studies show that although 2.5% (range, 0.2–17.8%) of the 488 genes on chromosome 8 demonstrated exonic overexpression, 15.4% (range, 0–24.8%) intronic over-expression and 5.4% (range, 0–20.1%) both exonic and intronic overexpression, MYC showed no overexpression. To further characterise oncogene expression, we investigated whether overexpression occurred for other oncogenes including EGFR family members (EGFR, ERBB3 and ERBB4) in each tumour subgroup and controls. No tumour or control showed EGFR amplification. For ERBB3, one sample with overexpressed ERBB2 (286T) showed a >2-fold-change of both exonic and intronic expressions. For ERBB4, two ERBB2 3⁺ patients showed >2-fold-changes (221T entrance and 286T exonic) and one triple-negative patient (085T) showed >2-fold exonic and intronic overexpression. Other major oncogene families investigated included epidermal, fibroblast, vascular endothelial and platelet derived growth factor receptors.

To investigate whether intronic overexpression was associated with particular introns and sequences, we calculated FPKM values for each exon, alternative exon and intron. We aligned these with the AceView alternative splicing database, using IGV, UCSC Blat and Repeat Masker to identify locations of repeat sequences, retrotransposons and other repeat elements (NCBI HG19 UCSC Blat, IGV and Repeat Masker). The upregulated intronic sequence expression found in cancers was not random, with high specificity for genes, introns and sequences involved associated with retrotransposons and other repeat sequences (data not shown). For ERBB2, the two introns with highest FPKM values were those immediately before the transmembrane domain. Predictions indicated that these would give rise to shorter, alternatively spliced variants and truncated protein isoforms lacking transmembrane and intra-cellular domains, no longer localised to the cell membrane, and instead generating soluble receptors (Fig. 5). Intronic LINES, Alus and other retroelements (enigmatic dark matter) can induce alternative splicing by mechanisms including exonisation of cryptic splice sites, forcing exonisation and cryptic polyadenylation or exon skipping (35–39). In this way, intronic or intergenic sequence expression may contribute to the acquisition of new exons and alternatively spliced variants.

We confirmed that TNCs contained no ERBB2 or ESR1 amplification, but for TNCs our results demonstrate that for VEGFA, 10/13 exhibited >2-fold exonic or intronic expression, but 3/13 (23%) showed no overexpression, emphasising the need for personalised profiling to increase effectiveness and reduce toxicity (Figs. 3 and 5). VEGFA undergoes alternative splicing, producing variants having either stimulatory or inhibitory functions (40), highlighting that transcriptome profiling may enable targeting of stimulatory isoforms. We have also shown a number of synonymous or non-synonymous mutation in VEGFA (Table V). In this study, TNC produced only inhibitory isoforms. VEGF receptors are also potential therapeutic targets, but our results showed no significant elevated expression for VEGFR1 (FLT1) or VEGFR2 (KDR) (Fig. 5).

For ERBB2 amplified tumours, intronic overexpression, multiple alternatively spliced transcripts and protein isoforms of tumours differ markedly from those of non-malignant samples and have implications for diagnosis, prognosis, targeted therapies and biomarkers. Although full length RefSeq ERBB2 found in normal cells is localised to cell membrane, in tumours, truncated ERBB2 isoforms have differing sizes, and subcellular localisations. Some isoforms may lack transmembrane and intracellular domains and no longer localise to cell membrane, becoming soluble receptors, but still have extracellular domains that bind to targeted drugs, forming soluble-receptor drug complexes, potentially partitioning toxic drug-soluble receptor complexes into serum (Fig. 5). Oncogenic proteins with altered or ablated drug target domains with altered monoclonal antibody or TK inhibitor binding, may render therapies ineffective or tumours unresponsive (Figs. 1 and 5). The estrogen receptor ESR1 normally localises to the nucleus, functioning as a nuclear transcription factor. However, our results demonstrate that ESR1 amplification may result in cytoplasmic rather than nuclear localisation, affecting nuclear transcription factor function (Figs. 3 and 5). The results suggest transcription of intronic ‘dark matter’ in cancers plays a key role in generating novel transcriptional landscapes, that provides insights into tumour biology and potential pitfalls and new opportunities for personalised cancer therapies (41).

This study revealed the extent of genomic structural alterations occurring during the evolution of human breast cancers and has identified some of the key changes to be subtype-specific and mutually exclusive. The findings have fundamental implications for tumour-specific breast cancer therapy. In non-neoplastic cells, ~2–3% of genomic ‘exonic’ DNA primes synthesis of protein-coding mRNA and transcription is tightly regulated (42). Here, we show that in cancers, massive genomic rearrangements occur with pervasive cancer-specific transcriptional changes that include intronic sequence overexpression generating alternatively spliced transcripts and truncated protein isoforms cannot be revealed by exome analysis. First, RefSeq exome sequencing cannot detect intronic overexpression and tumour-specific alternative splicing. Second, the relevance of these findings to current breast cancer treatment is that they show the functional correlates of tumour-specific alterations, revealing new biologically relevant therapeutic targets. Third, alternatively spliced tumour-specific transcripts and truncated proteins are likely to alter drug target domains. These structural changes within the genome modify the therapeutic effectiveness of targeted drugs such as Herceptin (trastuzumab) or small molecule TK inhibitors, designed to target external or TK domains of full length RefSeq ERBB2 (43). Truncation may ablate or alter drug target domains and explain lack of response, ineffectiveness or resistance to targeted therapies. Finally, GWAS have identi-fied genetic risk loci for breast cancers and SNPs are a key resource in cancer genetics. However, many high-risk SNPs are intronic, suggesting the functional involvement of intronic sequences in cancers. More than 95% of breast cancer susceptibility variants are found in non-exonic regions previously thought to be non-coding. Our studies provide a framework for tumour-specific analysis of introns and other non-exonic regions to assess whether high risk SNPs may be associated with tumour-specific transcription.

The data herein suggested that some chromosomal regions including 17q, 6q and 8q24 appear predisposed to amplicon formation. However, each amplicon has differing breakpoints, indicating the uniqueness and complexities of individual amplicon formation. Relationships between chromosomal amplification and oncogene overexpres-sion may also be complex. Prior to these studies, it appeared axiomatic that 8q24 amplification inevitably involved MYC overexpression. However, for 8q24 amplification and MYC our data show that oncogene overexpression and chromosomal amplification are not invariably linked. Absence of MYC overexpression in 8q24 amplification was surprising given that 15.4% of chromosome 8 genes showed intronic overexpres-sion, 2.5%, showed exonic overexpression and 5.4% exhibited both. The findings illustrate the necessity for detailed gene expression studies to analyse functional correlates of chromosome amplification, rather than relying solely on surrogates such as FISH.

In contrast to oncogene/hormone receptor amplification and intronic overexpression, no TSG showed intronic or exonic overexpression or the generation of truncated alternatively spliced variants. The comparisons were stark, given that in some tumours, TSGs including BRCA1 and NF1 were located immediately adjacent to 17q amplicons in which oncogenes including ERBB2 showed extensive intronic overexpression, alternative splicing and generated truncated protein isoforms. The expression profiles of TSGs suggest that, by their resistance to transcriptional disruption and intronic overexpression, these genes retain their normal function and highlight the abnormalities and extent of tumour-specific changes in amplified oncogenes. TSGs play key roles to prevent tumourigenesis in breast and other epithelia. The implications of our findings are that in tumours there are either mechanisms leading to intronic sequences overexpression and triggering tumour specific alternative splicing or mechanisms that prevent transcription of intronic sequences, in TSGs even when closely mapping oncogenes are amplified and overexpressed. Suggestions that mutations in TP53 may be linked with catastrophic chromosomal tumour-specific amplifications and rearrangements are intriguing and bring new perspectives in relation to TSGs and mechanisms of amplification (44). Analysis of TP53 mutations in relation to 17q, 6q and 8p amplification revealed no direct links between TP53 mutations and amplicons, although other loci, amplicons and mutations in other types of tumour should be analysed.

The human genome contains 40–50% of repetitive sequences derived from retrotransposable elements, with multiple copies of long and short interspersed nucleotide elements (LINES and SINES), LTRs and other repeat sequences (31,45–47). This study provides additional new information about the extent of intronic transcription and alternative splicing that illuminates the accumulating evidence in cancer, transcription of intronic, UTR satellite and transposon sequences occurs and extends the repertoire of alternatively spliced transcripts (31). Intronic L1s, Alus and other retroelements (enigmatic dark matter) affect transcription and gene expression via exon skipping and exonisation using cryptic splice sites, forcing exonisation and cryptic polyadenylation so genes acquire new exons from intronic or intergenic sequences thereby generating novel alternatively spliced variants (41,48). Recently, we described intronic exonisation in human prostate cancer, suggesting that this phenomenon is likely to be common to many different malignancies (49). In normal somatic cells, this sequence compartment is attenuated by epigenetic silencing mechanisms involving DNA methylation and chromatin-mediated repression to maintain genomic integrity (45,50,51). In tumours, there are suggestions that transcription from retrotransposons occurs subsequent to loss of methylation (50). Loss or inactivation of TSGs in cancers, has led to the suggestion that demethylating agents may be utilised to promote expression of lost or inactivated TSGs. While demethylating drugs may enhance TSG expression, there are potential risks of exacerbating or increasing retrotransposon transcription, that may potentially result in oncogene or other tumour promoting gene upregulation or overexpression, accelerating tumourigenesis rather than suppressing growth (52).

In this study, we demonstrate that intronic regions harbour sequences transcribed only in cancers and not in normal tissues. Since such tumour-specific sequences encompass transcripts not included in RefSeq exons or revealed by exome analysis, studies should be extended beyond the limits of individual exomes if the full extent of cancer transcription and exploitation of intronic and intergenic sequences is to be uncovered. Such previously unrecognised transcription and translation from amplified or disrupted oncogenes will generate RNA transcripts and proteins with altered size structures, functions and subcellular localisations similar to those detected in prostate cancer (49). Drug development has been predicated on targeting normal full-length RefSeq proteins with domains such as external, transmem-brane and TK domains in RTKs. Divergent protein isoforms could potentially affect function through diverse mechanisms ranging from ablation of drug target domains to alterations such as those to soluble rather than transmembrane receptors for RTKs (53,54), nuclear transcription factors localised to the cytoplasm for ESR1 (55), mesenchymal rather than epithelial signalling for FRGR2 (56,57) and inhibitory rather than stimu-latory functions for VEGFA (58). For amplified ERBB2 these findings have major implications for targeted therapies.

Despite the promise of personalized cancer medicine, therapeutic options are available only for breast cancers with amplified/overexpressed ERBB2 or ESR1. TNCs are the most aggressive tumours but lack therapeutic options, having no targetable ERBB2 or ESR1 overexpression. Our findings suggest transcriptome profiling may identify alternative physiological mechanisms as therapeutic targets. Non-selective genome-wide demethylation is unlikely to be therapeutically advantageous while anti-angiogenic therapies may be ineffective, or potentially toxic, without therapeutic benefit, thus explaining the disappointing results of the anti-VEGFA monoclonal bevacizumab and FDA withdrawal of approval for use in breast cancer (FDA 2011). This study suggests transcriptional profiling may be an appropriate approach to individual cancers by providing a comprehensive identification of new therapeutic targets for effective personalised cancer therapies.