Breast cancers are thought usually to develop from regions of cellular atypia into clinically evident preinvasive or invasive lesions with subsequent evolution of lethal features such as metastatic spead and therapeutic resistance. Biological and statistical models that guide our thinking about breast cancer genomic progression events that underlie this process include linear, stepwise models, such as those proposed by Vogelstein and coworkers for colon cancer (Fig. 24-1A) and branched, Darwinian evolutionary models, such as that proposed by Nowell, where there is no fixed order of events or steps (Fig. 24-1B) (1, 2). The recent application of next-generation sequencing technology has uncovered immense genomic complexity but has mostly focused on a single time point— the breast cancer genome at diagnosis. Longitudinal studies of individual cancer genomes over time will ultimately be required to completely characterize the genomics of breast cancer progression and are now possible as massively parallel sequencing platforms become more accessible. In this chapter, we will describe the theoretical basis for breast cancer evolution, from the earliest stages to advanced disease. We will review what is known about the dynamic structural changes in the genome that underlie the disease.

The revolution in DNA sequencing technology, with the development of next-generation sequencing instruments, has provided a wealth of new data on the genomic evolution of cancer, and it is worthwhile beginning our discussion with an overview of this technology as many of the discussions in this chapter rely on data derived from these “massively parallel sequencing” (MPS) approaches. The development and commercialization of next-generation DNA sequencing instruments began around 2006 and has made DNA sequencing thousands of times faster and considerably cheaper. With this technological advance, projects to sequence large numbers of cancer genomes became feasible. Here, we will mostly focus on an MPS process called “sequencing-by-synthesis,” as most data sets have used Illumina-based technology. The steps in this process involve generating a DNA library from the sample of interest (Fig. 24-2A) and attaching individal DNA molecules to a glass slide called a flow cell. The individual sequences are then amplified into “clusters” to increase signal intensity (Fig. 24-2B). A sequencing reaction is then performed whereby a different colored fluorophor is activated when each of the four different nucleotides are added to the DNA polymer (Fig. 24-2C) (3, 4). These sequential light reactions are captured by a sensitive light detector and the sequences entered into programs that align the sequences to the reference human genome. The process is remarkably efficient, and sequencing throughput is advancing rapidly. Currently more than 100 billion base pairs (Gigabases, Gb) can be generated per instrument run. Other methodologies and companies for next-generation DNA sequencing include the SOLiD system developed by Applied Biosystems, and single-molecule DNA sequencers, which are still under development (3).

A key advantage of the MPS approach is that the analysis begins with an individual DNA molecule. This is fundamentally different than the original Sanger DNA sequencing method, developed over 35 years ago, which averages sequences across millions of DNA molecules and, as a
result, has limited ability to distinguish variations in DNA sequences. Because MPS provides data on the frequency of a DNA mutation within a population, tumor clonality or heterogenity can be inferred from the variant allele frequency, and this provides new means for studying the clonal progression of cancer. Rare alleles can be detected down to a frequency of between 0.1% and 1% depending on the depth of the sequencing instrument and the fidelity of the enzymes used to generate the original clusters on the flow cell.

Much of the sequencing funded by the National Cancer Institute has focused on the initial diagnostic sample from invasive cancers through The Cancer Genome Atlas Project, with the goal to catalogue all the somatic mutations and structual abnoramlities in breast cancer (5). The application of this approach in the study of breast cancer is still in its infancy. Our current understanding of the genomic alterations that promote preinvasive to invasive breast cancer is still largely based on comparative genomic hybridization (CGH) techniques, loss of heterozygosity (LOH) analysis, gene expression profiling, and selective gene sequencing studies that have been applied to synchronous preinvasive and invasive breast cancers.

To understand breast cancer progression, a description of the anatomy of the breast and the histology of preinvasive versus invasive breast cancer is required; the reader is guided to Chapters 1, 9, 21, and 25 which deal with these issues in depth. In brief, several models have been proposed to describe the development of breast cancer that focus on the relationship between preinvasive and invasive breast cancers. The most widely accepted “linear” multistep model suggests a transition from normal epithelium to invasive breast cancer via non-atypical and atypical hyperplasia and in situ carcinoma through accumulation of genetic mutations (6). In this classic Wellings model, premalignant breast lesions arise from terminal ductal lobular units (TDLUs) and give rise to flat epithelial ayplasia (FEA), atypical ductal hyperplasia (ADH), and ductal carcinoma in situ (DCIS), which subsequently progresses, over a long period of time, to invasive ductal carcinoma (IDC), whereas atypical lobular hyperplasia (ALH) and lobular carcinoma in situ (LCIS) progress to invasive lobular cancer (ILC) (Fig. 24-3) (7). This theory was originally based on the histologic observation of the gradual histologic continuity, but it also has been supported by analyses of genetic alterations that compared preinvasive and invasive breast cancers, especially when they occur in the same breast (8, 9, 10 and 11). Subsequent findings introduced the concept of usual ductal hyperplasia (UDH), in which cells pile up to fill the terminal duct (TD) and acini compared to the single or minimally pseudostratified layer of cells in FEA that distends the TDLU, as the direct precursor to ADH (12, 13). However, UDH as the precursor for ADH has not been supported by recent immunohistochemical and molecular evidence. The LOH pattern observed in UDH is notably different from that associated with ADH and DCIS (8, 14, 15, 16, 17 and 18). This linear model of breast cancer progression provided the rationale for detection methods such as mammography in the hope of diagnosing and treating breast cancer at earlier and preinvasive stages before lethal features of the disease have developed (19). However, the occurrence of a preinvasive lesion is probably not an obligatory event in the development of invasive breast cancer. Although many premalignant lesions progress through the lifespan of some patients, others might stay stable throughout their lives; it remains unclear which lesions have the capacity to progress to invasive cancer. The “nonlinear” or “branched” model builds upon the “linear” model in that it agrees that DCIS is the precurser for IDC, but hypothesizes that different grades of DCIS progress to corresponding grades of IDC. In contrast, the “parallel” model hypothesizes that DCIS and IDC are parallel and independent developments from a common progenitor cell through different grades (20). This is supported by the investigation of gene copy number changes in synchronous DCIS and IDC lesions, which demonstrated changes that are specific to DCIS or IDC (21).

The molecular differences among the preinvasive and invasive breast cancers are largely unknown but have been an area of great research interest with the hope to identify the key events that drive the development and progression of invasive breast cancer. The pathological heterogeneity and the microscopic size of the preinvasive lesions have posed a practical challenge in isolating sufficient material that is devoid of contaminating tissues. The availability of laser capture microdissection (LCM) and genome-wide analysis tools provide a new opportunity to discover genetic events specifically activated or inactivated in the course of breast cancer development.

In gene expression profiling studies of synchronous DCIS and IDC lesions, the greatest alterations were among the different histological grades of DCIS and IDC (26). Notably, the grade I and grade III tumors demonstrated reciprocal gene expression patterns, whereas grade II tumors exhibited a hybrid pattern of grade I and grade III signatures. ADH samples demonstrated a grade I gene expression signature and clustered with the low-grade DCIS and IDC. Similarly, several comparative genomic hybridization studies revealed that the low-grade and high-grade DCIS have distinct gains and losses of genetic material. In the CGH study by Buerger et al. on 38 DCIS and 6 associated invasive breast cancers,
losses of 16q were exclusively seen in well- and intermediately-differentiated DCIS. A higher frequency of gains of 1q and losses of 11q was observed in intermediately-differentiated DCIS. The poorly-differentiated DCIS displayed complex genomic alterations, including loss of 8p, 11q, 13q, and 14q and gains of 1p, 8q, and 17q. This was characterized by a higher frequency of amplifications of 17q12 and 11q13. Analysis of paired IBC revealed a CGH pattern similar to the DCIS counterpart (33). This data provided further evidence that DCIS are precursor lesions of IBC, but progression from low- and high-grade DCIS to IBC are likely from different genetic pathways (34) (Table 24-1).

In addition, FEA and ADH has morphological overlap with low-grade carcinomas, and ADH is accepted as a precursor lesion for low-grade/noncomedo DCIS. Molecularly, FEA has a genetic profile that overlaps with those of synchronous lowgrade DCIS and low-grade invasive carcinomas and has a high rate of LOH at 16q, supporting its precursor role in the evolution to low-grade cancers (11). In another study, FEA exhibited recurrent chromosomal copy number gains and losses (gains on 15p, 16p, and 19; losses on 16q, 17p, and X), which overlap with those observed in both ADH and low-grade DCIS (9).

However, the precursor lesion of high-grade/comedo DCIS is not clear. High-grade DCIS lesions are associated with more complex genetic alterations and overexpression of genes related to mitotic activity and cell cycle processes (35). In a comparison of copy number changes of 21 breast cancer-related genes between laser-microdissected DCIS and adjacent IDC lesions (25), no significant differences existed between DCIS and adjacent IDC. However, low/intermediate-grade DCIS showed on average 6 gains/amplifications versus 8 in high-grade DCIS. Furthermore, alterations of AURKA (aurora kinase A) and CCNE1 (cyclin E1) were exclusively found in high-grade DCIS, and HER2, PRDM14 (PR domain containing 14), and EMSY amplification was more frequent in high-grade DCIS than in low/intermediate-grade DCIS. These data indicate that DCIS is genetically as advanced as IBC and support a model in which different histological grades of DCIS are associated with distinct genomic changes that progress to IDC in different routes.

The significant intratumoral histological and biological diversity within cases of DCIS argues for an evolution of DCIS from lower grade to higher grade (28). In an analysis of 120 consecutive cases of pure DCIS, 45.8% of cases showed areas of diversity in nuclear grades, including 30% with grades I and II, 6.6% with grades II and III, and 9.2% with grades I, II, and III. In addition, about one-third of the cases showing histologic diversity also showed biologic diversity for one or more biomarkers that included ER, GATA3, HER2, CK5/6, CK18, and p53 by immunohistochemistry staining. Similarly, in studies assessing the LOH, DCIS contained many of the same specific genetic defects regardless of histologic differentiation, although the absolute number of defects was found to be higher in higher grade lesions (8). Furthermore, a significant subset of genes expressed at higher levels in grade III DCIS compared to grade I DCIS were further elevated in IDC. In addition, the link between tumor grade and transition from DCIS to IDC is consistent with the clinical observation that grade III DCIS is more likely to be associated with occult invasive disease than grade I DCIS (36).

The evolution of genetic events that drives the process of breast tumorigenesis is poorly understood. As demonstrated in previous studies, many genetic alterations present in the invasive breast cancer already exist in the earliest phenotypically identifiable lesions such as ADH. However, gene expression profiling, CGH, and candidate gene approach, which are the main technologies used in these studies, are limited in their capacity for detailed genomic interrogation. A comprehensive genomic sequencing study of synchronous preinvasive and invasive cancer in comparison to the normal breast epithelial cells is needed but has not been reported. In this section, we will focus on available data investigating molecular markers that may predict the prognosis of DCIS and potential candidate drivers in cancer progression.