In modern years, the research of low-penetrance allelic variants was conducted mainly through GWAS. These studies use a large number of common genetic SNPs to identify associations with disease that rely upon patterns of linkage disequilibrium in the human genome [22]. The power of GWAS is to evaluate the association of genetic variants at different loci on different chromosomes in large series of cases versus controls, analyzing a panel of 100,000 SNPs simultaneously, to identify new alleles of susceptibility to BC [23]. In the human genome, it has been estimated that there are 7 million common SNPs that have a minor allele frequency (m.a.f.), 45 %, and because recombination occurs in different hot spots, the nascent polymorphisms are often strongly correlated. These studies therefore supply a powerful tool to recognize novel markers for susceptibility and prognosis of disease [46–48].

In the GWA studies, the accumulation of a large number of data is crucial. Houlston and Peto have estimated the number of cases required to identify low-penetrance alleles conferring a relative risk of two both in an unselected population and in families with first-degree relatives affected [47]. In an unselected population, the identification of a susceptibility allele with a frequency of 5 % requires over 800 cases. In the same population, the identification of a susceptibility allele with a frequency of 1 % requires over 3,700 unselected cases, whereas about 700 would be enough if three affected families are selected. Therefore, the power of association studies can be significantly increased using selected cases with a family history of cancer because fewer cases are required to demonstrate the association with the disease [47].

The potential of the association studies of cases with a family history to identify low-penetrance alleles conferring a relative risk of 2 has been demonstrated by the mutation CHEK2 1100delC in patients with BC. This variant carried by 1 % of the population confers an increased risk of 1.7-fold. The frequency was not significantly increased in unselected cases (1.4 %), but it was strongly increased in familial cases without BRCA1 and BRCA2 mutations (5.1 %) [49].

In the past several years, several novel risk alleles for BC were identified by four recent GWA studies: Breast Cancer Association Consortium, Cancer Genetic Markers of Susceptibility, DeCode Islanda, and Memorial Sloan–Kettering Cancer Center [29, 30, 43, 50]. In each of them, the association study was shared in three phases: the first phase identifies the common SNPs in cases and controls, the second phase evaluates how many of the above SNPs are common to a greater number of cases and controls, and, finally, the third phase aims to identify new alleles of susceptibility to BC. Easton et al., in their study, identified five independent loci associated with increased susceptibility to BC (Po10_7) [30]. This multistage study involved in the first stage 390 BC cases with a strong family history and 364 controls and 3,990 cases and 3,916 controls in the second stage. To define the risk associated with the 30 most significant SNPs, a third stage of the study was conducted involving 21,860 cases and 22,578 controls from 22 additional studies in the Breast Cancer Association Consortium. These combined analyses allowed the observation that the SNPs showing a stronger statistical evidence of association with an increased familial risk were rs2981582 in intron 2 of FGFR2, rs12443621 and rs8051542 within TNRC9, rs889312 in a region that contains the MAP3K1 gene, rs3817198 in intron 10 of lymphocyte-specific protein 1 (LSP1), and rs2107425 within the H19 gene.

In brief, GWAS analyze thousands of cases and controls (huge numbers are necessary to reach sufficient statistical power) in order to compare SNPs, which makes it possible to identify genetic variants associated with disease risk. More than 100 GWAS have related a slight increase in cancer risk [51]. In breast cancer, at least 18 variants have been identified, with a 1.1- to 1.5-fold increase in risk. Interestingly, these SNPs might not only provide useful information on the risk of breast cancer, but they could also be linked to specific molecular subtype of breast cancer, such as FGFR2-RS2981582 and TNRC9-RS3803662 with positive estrogen receptor [52] and rs8170 in chromosome 19p13 with negative estrogen receptors [53]. However, the clinical impact of these approaches is not clear, and this could illustrate how techniques are developing faster than knowledge. For example, a 1.2-fold risk with one of these variants is comparable to that of delaying the age of the first pregnancy to more than 35 years [51]. In addition, these approaches give no information about other complex factors that can affect risk, and there are no data on the possible effects of combining two or more variants. Consequently, large research consortia are essential if a high number of cases and controls are to be attained. Finally, direct access to some GWAS via the Internet can generate anxiety and dangerous misinterpretation of genetic risk.

Endeavors to discover genes’ contributions to intricate diseases, such as cancer, require new study designs that incorporate an efficient use of population resources and modern genotyping technologies. There are two approaches used for the study of breast cancer, both of which incorporate the use of biobanks. One uses a cancer registry as a source of case information, which is then linked to a biobank on blood DNA. The biobank also provides samples from matched controls. After genotyping, clinical data are retrieved from hospital records, and the results can be presented for genotype-specific cancer risks or similarly for genotype-specific clinical and survival parameters. The second approach uses registered data on cancer in families or among twins. With defined groups of patients, paraffin tissue is collected by contacting the pathology departments of the hospitals where the patients were diagnosed. Tumor and healthy tissue are prepared and used for mutation, the loss of heterozygosity, or copy number analysis. In the era of whole-genome genotyping technologies, the importance of well-characterized sample sets cannot be overemphasized. Samples, rather than technologies, limit the rate of gene discovery in complex diseases [54].

Women with germ line BRCA1 or BRCA2 mutations are estimated to have a 45–70 % risk of breast cancer by age 70 years [55–58]. The identification of BRCA1 and BRCA2 mutations was a major step in personalizing breast cancer risk assessment, screening, and risk reduction strategies. Studies are ongoing to determine whether or not certain subgroups of BRCA mutation carriers may be at a higher risk for breast cancer. It has been proposed that certain BRCA mutations may confer a differential risk of future breast cancer development, suggesting an important genotype–phenotype linking [59, 60].

In a recent kin-cohort study in Ontario, Risch et al. observed a trend of increasing breast cancer risk associated with increasing downstream location of BRCA1 mutation with a continuous linear trend and a 32 % increase in risk associated with each additional 10 % or 559 nucleotides of downstream distance [60]. Over the past few years, a considerable effort has been made to characterize genetic abnormalities in cancer, the general idea being that tumor genotyping would be valuable in defining cancer phenotypes. In a previous study, we showed that it was possible to delineate subsets of breast tumors according to specific combinations of DNA amplifications [61]. The present work allowed us to extend the phenotypic description to prognostic significance. We show here that some of the markers tested presented prognostic significance in specific subsets of patients. This was particularly evident for MDM2 amplification and p53 mutations, which showed a strong prognostic value in the N2 subset of patients, or for the amplification of CCND1, EMS1, and FGFR1 in N1 patients. During the course of this study, we also made some observations that suggest the existence of correlation clustering in other patient subsets, such as MYC in patients under 50 years or MDM2 in ER1 patients (data not shown).

Our data constitute an attempt to delineate tumor subsets according to their genotypic specificity. Knowing the complexity of the genetic rearrangements in breast cancer, the nine events studied here probably correspond to a small portion of the genes involved in tumorigenesis. Genotyping of breast tumors will involve the analysis of an ever-larger number of parameters and sorting of the significance of complex combinations. Because different combinations of genes or genetic anomalies may bear a meaning in different populations of patients, the analysis of specific phenotypic subsets will be necessary, thus leading to an increase of the number of comparisons. This will require the analysis of very large cohorts of patients (several thousand) and consequently the use of high-throughput analytical methods [62] in association with statistical tools, especially devised for multiple-comparison analyses.

A pilot genome-wide sequencing exertion on breast cancers identified a total of 1,137 somatically mutated genes from 11 breast cancers, with an average of 52 non-synonymous mutations per sample. Using gene mutation incidence as the principal criterion, 140 genes were identified as candidate cancer genes that require further assessment to confirm their functions as causal contributors to tumorigenesis [63, 64]. These studies also portrayed the genomic landscape of human breast cancer that consists of a few repeatedly mutated gene “mountains” and a huge number of rarely mutated (usually <5 %) gene “hills” [64].

Besides the detection of novel candidate genes, more recent studies have delineated new aspects of breast cancer exomes. Interrogating luminal-type breast cancer genomes with clinical data revealed that somatic mutations in TP53 signaling pathway, DNA replication, and mismatch repair are associated with aromatase inhibitor resistance [65]. Determination of clonal frequencies by deep sequencing provided new insights into the initiating events of TNBCs [66]. Massively parallel paired-end sequencing technologies enable whole-genome detection of gene rearrangements at the DNA sequence level [67]. An analysis on 24 breast cancers revealed more than 2,000 gene rearrangements, enriched with tandem duplications [68]. Analysis of breast cancers across a variety of subtypes revealed that luminal B and HER2-enriched breast tumors harbor many more structural rearrangements when compared to the luminal A subtype. However, no repeatedly recurrent rearrangements have been discovered in breast cancer by earlier studies except for the MAGI3–AKT3 gene fusion detected in 4 % (9 out of 257) of breast cancers [69].

Like all cancer types, breast cancer progression is thought to be a dynamic multistep Darwinian evolution process. Independent mutations arise in a stepwise fashion, of which those conferring selective advantages promote cell proliferation and clonal expansion [70]. Through deep whole-genome sequencing of 21 breast cancers and analysis of subclonal genetic alterations, Nik-Zainal et al. proposed a model for clonal evolution that many molecular aberrations accumulate in dormant cell lineages before final expansion of the most recent common ancestor, which triggers diagnosis [71]. Integrative breast cancer studies aim at developing new definition of breast cancer subtypes with better prognostic and predictive values. A cluster analysis integrating copy number and gene expression profiles of ~2,000 breast cancers suggested a novel classification system [72]. A recent multiplatform study on hundreds of breast cancers revealed subtype-specific pattern in numerous tumor characteristics including gene mutations, microRNA expression, DNA methylation, copy number changes, and protein expression. Moreover, in whole-exome sequencing of more than 500 tumors, this study also revealed almost all repeatedly altered pathways (PI3K/AKT, TP53, RB) in breast cancer [73].

The quickly evolving sequencing technologies generate massive genomic data at an increasing rate with reduced cost. In recent years in particular, large-scale analyses of cancer genomes have produced a prosperity of information, which greatly expanded our knowledge on human breast cancer, summarized in Table 4.2. The launch of comprehensive cancer genome projects including the Cancer Genome Project (CGP) [78], the Cancer Genome Atlas (TCGA) [79], and the International Cancer Genome Consortium (ICGC) [80] facilitates the compilation of an encyclopedic catalogue of the genomic changes involved in cancer. Whole-genome sequencing studies enable observation of genetic alterations earlier undetectable by protein-coding sequence screens, including mutations in noncoding regions and large rearrangements. Primary breast cancers were reported to harbor ~7,000–10,000 somatic point mutations per genome [65, 69, 81] in which tens to hundreds reside in the protein-coding regions [63, 64, 69, 77], as well as up to hundreds (average 20–50) of somatic structural variants [65, 68, 69, 75, 81]. The minimum number of mutations necessary for tumorigenesis has been estimated to be around 5–6, according to incidence modeling of solid tumors such as breast and colorectal cancers, and this number would be smaller in leukemia and childhood cancers [82]. However, recent systematic mutational screens of cancer genomes suggested a higher number of causal gene mutations in each tumor (range 10–20 genes) [63, 83].

Distinguishing the driver mutations from passengers cannot be accomplished by analyzing genetic data alone but requires functional validation of the cancer-relevant activities. Since most functional assays are relatively labor- and time-intensive, prioritization of the genes for functional studies presents a great challenge in cancer genomic data interpretation. Several measurements have been adopted to identify the most promising driver mutations. First, analyzing the ratio of non-synonymous mutations to synonymous mutations of a given gene would indicate whether the mutations have been under positive selection during tumor development, thus a higher than expected ratio always suggests driver mutation [84, 85]. Second, assessment of the mutation prevalence in genes also identifies drivers that contribute to cancer if they are highly unlikely to be mutated by chance [63]. Third, several tools have been employed to predict the effect of non-synonymous single nucleotide variants on protein function based on phylogenetic conservation and physical considerations (e.g., Sorting Intolerant From Tolerant (SIFT) [86], Polymorphism Phenotyping (PolyPhen) [87], Panther [88], MutationTaster [89], etc.). Last but not least, as the number of pathways involved in cancer is much smaller than that of cancer genes and a variety of mutations in multiple cancer genes from the same pathway would likely to have similar pathological effects [90], evaluation of the combined prevalence of somatic alterations at the pathway level provides strategies for identification of cancer-associated processes [91, 92].

In the past few years, comprehensive mutation interpretation implementing most, if not all, of these measurements has been introduced into cancer genome analyses. For example, Carter and her colleagues developed a computational pipeline for cancer-specific high-throughput annotation of somatic mutations (CHASM), which takes a total of 49 predictive features into account for driver identification [93]. Another example is a package for determination of mutational significance in cancer (MuSiC), designed by Dees et al. MuSiC is the first software suite that integrates clinical data with coverage data and database references to identify drivers from large mutational discovery sets [94].

Although many tools can help to prioritize the candidates of interest for downstream analyses, only the evidence from functional assays and biological studies can fully credential a candidate gene as a bonafide cancer gene. It is clear from cancer genome resequencing efforts that not all cancer genes are mutated at high prevalence. On the contrary, despite conferring selective advantage, the vast majorities of cancer genes are not frequently mutated and are therefore difficult to identify through sequencing of a limited number of samples. In order to discover these infrequent driver mutations, systematic screens of large cohorts of patients are required. For example, it was estimated that 500 tumor samples of a particular tumor type are needed in whole-exome sequencing studies to get an ~80 % detection power of genes with ~3 % true mutation frequency [80].

Gene expression profiling has proven to be a useful and reliable tool for classifying breast cancers into subgroups that reflect different histopathological characteristics as well as differential prognostic outcome. It has been suggested that estrogen receptor-negative and estrogen receptor-positive breast cancers can be subdivided into Her-2 positive basal-epithelial-like, normal breast-like, and luminal-like [95]. The potentially different origins of the tumor cells may signify distinct pathways of tumorigenesis and differences in the clinical course of the disease. Germ line mutations in the BRCA1 and BRCA2 genes together account for a significant portion of hereditary breast cancers. They have been shown to leave a characteristic imprint on the panel of genes expressed by the tumors [96], with BRCA1-dependent tumors exhibiting a transcriptional profile similar to the basal subtype of tumors [97]. These findings suggest that the cellular origin of BRCA1- and BRCA2-mutation-positive tumors may differ or that these tumors traverse down separate pathways in their progression toward malignancy [96]. Furthermore, the molecular subclassification of non-BRCA1/2 familial breast cancers into homogeneous subgroups underscores the potential differences in cellular origin and/or disease progression due to the presence of multiple diverse underlying genetic alterations, which is reflected in the phenotype of the tumors [98].

The diversity of breast cancer has been acknowledged for decades, but recent technological advances in molecular biology have given detailed knowledge on how extensive this heterogeneity really is. Traditional classification based on morphology has given limited clinical value, mostly because the majority of breast carcinomas are classified as invasive ductal carcinomas, which show a highly variable response to therapy and outcome [99]. The first molecular subclassification with a major impact on breast cancer research was proposed by Perou and colleagues where the tumors were subdivided according to their pattern of gene expression [95, 100]. Five groups were identified and named luminal A, luminal B, basal-like, normal-like, and HER-2-enriched subgroups. These intrinsic subgroups have been shown to be different in terms of biology, survival, and recurrence rate [95, 97]. The molecular subgroups have been extended to also include a sixth subgroup, which has been named the claudin-low group, based on its low expression level of tight junction genes (the claudin genes) [101].

Different methods for the assignment of individual tumors to its molecular subgroup are proposed, each based on the expression levels of different sets of genes [97, 102, 103]. The agreement between methods on how to classify individual tumors are not optimal, and how to establish more robust single sample predictors is actively debated [104–107].

Aneuploidy is the presence of an abnormal number of parts of or whole chromosomes and is one feature that clearly separates cancer cells from normal cells. This was proposed as being important in cancer nearly a century ago by Theodor Boveri. With array-based comparative genomic hybridization (aCGH), a genome-wide profile of the copy number alterations in the tumor can be obtained. These patterns are related to the molecular subtypes with distinct differences in the number of alterations between the subtypes [108–111]. These copy number alterations (CNAs) alter the dosage of genes and highly influence the level of expression [112, 113]. This frequently affects the activity in oncogenes and tumor suppressor genes, and in this way CNAs are important for the carcinogenic process.

CNAs in tumors are a result of deregulated cell cycle control and of DNA maintenance and repair [114]. Different patterns of copy number alterations have been identified with distinct differences; simplex profiles are characterized by few alterations and complex genomic profiles have extensive changes [115]. Complex genomic rearrangements are areas with high-level amplifications and have prognostic value in breast cancer even when they do not harbor known oncogenes, suggesting that the phenotype of defect DNA repair may be associated with more aggressive disease [115, 116]. Alterations in the expression pattern are caused by changes at the genomic level, and a robust classification of breast cancer for clinical use should probably take these more into account. Changes at the genomic level include point mutations, changes in copy number, and epigenetic events. These are characteristics that enable and drive carcinogenesis together with tumor-promoted inflammation [117].

Breast cancers progress through multiple genomic and epigenomic steps. However, the evolutionary process is unlikely to be the result of a linear and more or less constant rate of successive genomic and epigenomic aberration accretion. It seems more likely that cancer progression varies between individuals and over time within an individual. The breast progenitor cell in which the tumor arises likely determines the spectrum of aberrations needed to enable progression. Evidence for this comes from studies showing that the spectrum of genomic aberrations accumulated is strongly influenced by the normal progenitor cell type in which the cancer arises [108, 206]. In general, the events associated with early aspects of breast tumor appear to be epigenomic in nature, wherein stepwise DNA methylation changes enable escape of telomerase-negative epithelial cells from proliferation barriers [207]. This leads to proliferation in the absence of telomerase and culminates in a period of high genome instability owing to checkpoint deregulation [208, 209] and/or entry into telomere crisis when telomeres become critically short [206, 210]. This barrier is highly effective but not perfectly so. As a consequence, most cells become genomically unstable and die. The increase in genome instability during breast tumor progression is illustrated in Fig. 4.1. Rarely, a single cell accumulates genomic or epigenomic alterations that reactivate telomerase and confer a proliferative advantage. This cell might be considered the tumor initiation cell and will have multiple characteristics that appear “stem cell like.” The extent to which this is related to normal stem cells remains unclear. However, it is likely that the genomic characteristics of this cell—both transcriptional and genomic—will be reflected in subsequent progeny. This may explain why tumor genomes appear to evolve relatively slowly after telomere crisis and why metastases that develop years after immortalization usually retain the genomic characteristics of the primary tumor from which they were derived [212].

That said, not all breast tumors progress this way. A recent analysis of cancer progression termed Sector-Ploidy-Profiling (SPP) demonstrates that breast tumor evolution may evolve as a single major clonal subpopulation or as multiple clonal subpopulations [213]. The latter model may explain why a small percentage of metastatic cancers do not resemble the primary tumor from which they were derived. Figure 4.1 also shows that the cells that survive telomere crisis remain genomically unstable. Thus, tumors are likely to consist of a large number of cells that are not faithful genomic representations of the tumor-initiating cell and are likely to be biologically compromised. These cells are likely to be much more sensitive to treatment than the tumor-initiating cells from which they were derived. This may partially explain why breast tumors initially respond well to treatment but fail to exhibit a durable response.

To facilitate understanding of the scope of epigenetic modifications that occur in normal and cancer cells, a range of gene-specific or genome-wide technological approaches have been developed. We present an overview of recent technological developments and discuss the merits and the limitations of these approaches with respect to studies on cancer cells.