Patterns of Chromosomal Alterations in Breast Ductal Carcinoma In situ

Breast cancer likely develops through a series of genetic events, with neoplastic conversion to invasive cancer occurring sometime during the preinvasive histological phases of usual hyperplasia, atypical hyperplasia, and ductal carcinoma in situ (DCIS). In the past 15 years, the diagnosis of DCIS has increased dramatically, largely due to the more widespread use of screening mammography (1). From 1983 to 1992, the incidence of DCIS had grown 500% and now accounts for >25% of all mammographically detected breast neoplasms (1). Because DCIS is thought to be a precursor for invasive cancer, current therapy consists of complete surgical extirpation to free margins by either lumpectomy or mastectomy. Surgical treatment is often combined with radiation to the ipsilateral breast and/or tamoxifen in those patients undergoing lumpectomy. Natural history studies suggest that <50% of all DCIS will progress to invasive cancer if treated with biopsy alone (2, 3). Nevertheless, because of limited prognostic utility of clinical and biological markers in DCIS, most patients are treated as if they would progress, often with treatments as aggressive as those offered to patients with early, localized invasive disease.

DCIS is thought to be a nonobligate precursor to invasive cancer. A number of studies have shown that DCIS and invasive ductal cancer (IDC) exhibit similar genetic abnormalities (4, 5, 6). These studies characterized genetic alterations at a small number of genomic loci and were able only to estimate the overall burden of genomic instability in these lesions. Furthermore, many previous articles characterized DCIS that were found in the presence of synchronous invasive cancer, rather than lesions occurring alone. Thus, genomic changes unique to DCIS, as well as analysis of changes unique to different subtypes of DCIS, have not been fully described.

To study chromosomal alterations occurring in pure DCIS, we used the technique of comparative genomic hybridization (CGH) to characterize genomic changes in 34 cases of DCIS occurring in the absence of invasive cancer. We compared these chromosomal alterations to 12 cases of synchronous DCIS and IDC, as well as to 146 cases of invasive ductal or lobular cancer. Genomic differences between high- and low/intermediate-grade DCIS were also analyzed to establish whether genomic differences between individual subtypes of DCIS could be determined.

Formalin-fixed, paraffin-embedded tissue blocks were identified from 46 patients from the University of California San Francisco, California Pacific Medical Center, University of Basel, or University of Vermont pathology archives. Institutional Review Board approval was obtained at all institutions participating in this study. Of these 46 samples, 34 patients were diagnosed with DCIS without invasion (pure DCIS), and 12 cases contained DCIS as well as synchronous invasive cancer either in the same or adjacent blocks. Seventeen of these cases have been described previously (7). The changes identified in the 34 cases of pure DCIS were compared with a combined set of 146 cases of invasive cancers of ductal, lobular, micropapillary, and tubular cancers reported previously (8, 9, 10, 11). CGH profiles for these 146 cases were reviewed and rescored for the current study.

H&E-stained sections were reviewed to define areas of DCIS, invasive tumors (if present), and normal ducts or lobules. Nuclear grade of the DCIS was recorded as low, intermediate, or high and the histological pattern as comedo or noncomedo (solid, cribriform, and micropapillary) types (12). Comedo-type DCIS was defined as solid-type DCIS with high nuclear grade and moderate or extensive necrosis. Tumors exhibiting a mixture of histological grades were classified by the highest grade seen. DCIS were also classified based on estimated size of tumor and degree of necrosis. The size of the lesions was categorized as small (<1 cm in greatest diameter), medium (1–2 cm), and extensive (>2 cm). Necrosis was defined as negative when none was present, small when limited to an occasional duct or small amounts in multiple ducts, moderate when significant necrosis was present in more than two to three ducts, and extensive when present in the majority of ducts.

Tissue preparation, DNA extraction, and CGH were performed as described previously (7). Briefly, using the adjacent H&E-stained slide for orientation, two to three methyl-green-stained 5-μm sections were microdissected to isolate DNA from DCIS, IDCs, and normal ducts. DNA was isolated by a 3-day proteinase K digestion. In five cases, regions of DCIS were sufficiently large (>2 cm, >70% tumor cells) so that CGH could be performed without prior PCR amplification. For the remaining cases, 1–2 μl aliquots of microdissected DNA were amplified in duplicate by degenerate oligonucleotide primer PCR. PCR amplified normal reference DNA (Promega) was labeled by nick translation with Texas Red-5-dUTP or with Fluorescein-12-dUTP (Dupont). The MPE600 cell line was labeled with fluorescein or with digoxigenin-11-dUTP (Boehringer) and served as a positive control, with clearly defined chromosomal alterations, including gains on 1q, 11q13-14, 13q, and 17q and losses on 9p, 11qter, and 16q. Forty μl of amplified DNA from paraffin sections were used per 50-μl reaction, and DNA was labeled with fluorescein or digoxigenin. For the five larger tumors, 1 μg of DNA was labeled by nick translation. Samples were labeled in duplicate with fluorescein and Texas Red. Samples were hybridized onto normal male metaphases. Each sample was hybridized in duplicate with different fluorochromes. Digoxigenin-labeled samples were stained with anti-digoxigenin rhodamine (Boehringer). Texas Red labeling was not used for microdissected PCR-amplified samples because of inconsistent results and lower intensities.

Successful hybridizations showed good intensity signals, with smooth, homogeneous staining over entire metaphases. At least five metaphase spreads were acquired for each case. Acquisition was performed using our Quantitative Image Processing System (13). Two to three metaphases/sample were analyzed in each color. Gains and losses were defined if both the mean ratio and its SD were >1.2 or <0.85. Inverse CGH pairs were examined together, and all changes must have been seen in both hybridizations. Interpretations of changes at 1pter, 19, and 22 (GC-rich areas) were interpreted more stringently, requiring the threshold to be exceeded in both forward and reverse hybridizations for inclusion in the analysis. Overall, 94% of the PCR-amplified cases were successfully hybridized using these methods. When present in sufficient amounts, normal breast tissue was microdissected from the DCIS cases (both pure and synchronous) concurrent with microdissection of the DCIS. Of these normal samples, 20 contained sufficient DNA for evaluation. CGH of these specimens showed no chromosomal alterations compared with normal reference DNA.

Genetic alterations were tabulated as a single change if they (a) consisted of a single regional change, (b) consisted of a whole arm change, or (c) were comprised of either two gains or two losses on one chromosomal arm. If both a gain and a loss were present on the same arm, two events were recorded. Frequencies of genetic changes between different tumor groups were compared with a χ² statistic calculated for each 2 × 2 contingency table. No adjustment for multiple comparisons was done.

Percent concordance (7) was defined as

Concordance was used to determine the extent to which two specimens shared the same genomic changes, with higher concordance indicating a greater number of shared changes.

Patient and tumor characteristics for 34 cases of DCIS without invasive cancer (pure DCIS) are shown in Table 1. The mean age at diagnosis was 52 years. DCIS was 50% high grade, 44% intermediate grade, and 6% low grade. The extent of necrosis was categorized as extensive in 35%, moderate in 41%, and small or absent in the remainder. Four cases lacked necrosis. For the 12 cases of paired DCIS and concurrent invasive cancer (Table 2), the mean age at diagnosis was also 52 years. Only two patients in this group had an invasive tumor measuring <2 cm. Four patients had node-positive disease. Of the 34 patients with DCIS alone, 17 cases had no known recurrence within 5 years, whereas the remaining 17 cases had recurrences of DCIS (13 within 5 years and 4 additional cases with known recurrence in >5 years).

Thirty-four cases of pure DCIS were characterized by chromosomal CGH (Fig. 1). The mean number of chromosomal alterations was 8.0 (average of 4.7 losses and 3.3 gains). All cases had at least one chromosomal CGH aberration (range, 1–20). The prevalence of specific CGH alterations within groups was analyzed by lesion type (Table 3); any genomic alteration exceeding a frequency > 25% for at least one lesion type is shown. The most common changes in pure DCIS were gains on chromosome arms 1q (59%), 17q (53%), and 8q (38%), and losses on 8p (59%), 17p (44%), and 16q (38%). Loss of 8p occurred in two predominant patterns, consisting of either whole arm loss (38%) or regional loss (21%). For pure DCIS, all cases with partial loss of distal 8p21-pter also showed gain of the most proximal region of 8p (8p11-p12).

Separation of the pure DCIS cases into low/intermediate-grade (n = 17) and high-grade (n = 17) groups demonstrated a similar number of alterations (7.7 versus 8.2; P = 0.39) in the two groups. The low/intermediate-grade lesions showed an increased frequency of losses of 16q (65 versus 12%; P = 0.002) and 17p (53 versus 35%; P = 0.25) compared with high-grade DCIS but were less likely to have gains of 17q (41 versus 65%; P = 0.15) and losses of 8p (41 versus 76%; P = 0.04). In the high-grade DCIS, loss of 8p occurred predominantly as whole arm loss (65%), whereas the low/intermediate-grade lesions showed a more common pattern of 8p loss as a partial arm loss combined with proximal gain (29%), rather than as whole arm loss (12%).

CGH alterations associated with comedo and noncomedo histologies were similar to high-grade and low/intermediate-grade lesions, respectively, because 15 of 17 cases of high-grade DCIS exhibited comedo necrosis. Comedo-type DCIS had fewer losses of 16q than the noncomedo DCIS (13 versus 57%; P = 0.009), as well as more 8p losses (80 versus 47%; P = 0.055), and 17q gains (67 versus 47%; P = 0.219).

Pure DCIS cases showed slightly fewer chromosomal changes than the invasive cancers, although this difference was not statistically significant (8.0 versus 8.8, P = 0.26). Many of the most common changes were similar in the pure DCIS and the invasive cancers, including gain of 1q (59% in DCIS versus 57% in invasive; P = 0.53), gain of 8q (38 versus 48%; P = 0.22), and loss of 16q (38 versus 34%; P = 0.41). Significant differences were seen in the frequency of 11q gain (0% in DCIS versus 18% in invasive cancer; P = 0.003), 17q gain (53 versus 29%; P = 0.011), and of 8p loss (59 versus 33%; P = 0.008). The concurrent loss of 8p21-pter with a gain of 8p11-p12 occurred in 21% of the DCIS but only in 6% of IDC (P = 0.026).

The degree of genomic alteration was similar between pure DCIS and node-negative IDC (8.0 versus 7.7, P = 0.26). However, node-positive cancers as a whole demonstrated a greater number of alterations (mean of alterations = 11.0; P = 0.01). Although 8p loss was seen in both pure DCIS and invasive cancer, the latter group was more likely to exhibit loss of the entire arm of 8p than either node-negative or node-positive invasive ductal cancers (59% in DCIS versus 19 and 17%, respectively, P = 0.004 and P = 0.036). Chromosome14q loss was also seen more frequently in DCIS (29%) than either node-negative or node-positive invasive tumors (14% in both; P = 0.07 and P = 0.12, respectively).

Genomic changes most characteristic of node-positive versus node-negative invasive cancers was loss of 18p and 18q (P = 0.001 for both). However, other regions of genomic alteration significantly different between node-positive and node-negative cancers were gain of 17q (48 versus 19%; P = 0.005) and gain of 16p (35 versus 10%; P = 0.005). Other changes between node-positive and node-negative cancers approaching significance included losses of 1p (P = 0.049), 3p (P = 0.043), 4q (P = 0.026), 6q (P = 0.028), 9p (P = 0.17), 9q (P = 0.026), and 11q (P = 0.036), all showing greater frequency of loss in the node-positive group.

Among the invasive cancers, there were distinct genomic changes characterizing some individual histological subtypes. Invasive lobular cancers (ILCs) had a high prevalence of 1q gain (89%) and 16q loss (72%), as well as a low likelihood of 17q gain (17%). Interestingly, this pattern of 1q gain and 16q loss was shared with tubular cancers as well as with low/intermediate-grade DCIS. Very few cases of ILC showed 8p loss (17%), but all of these were characterized by loss of the entire arm of 8p. In contrast, all of the invasive micropapillary cancers (n = 16) had loss of 8p, with 63% of these cases showing regional, rather than whole arm loss. This IDC subtype also demonstrated a high prevalence of 8p gain and 8q gain, 56 and 88%, respectively.

Twelve pairs of synchronous DCIS and invasive cancer were separately microdissected and characterized by chromosomal CGH (Table 4). The mean number of aberrations was 8.1 for the DCIS and 7.8 for their concurrent IDC (P = 0.42). Although the sample size was small, this set of DCIS showed a significant difference in number of alterations between high-grade (n = 5) and low/intermediate-grade (n = 7) lesions (11.8 versus 5.4; P = 0.002).

The mean concordance (see “Materials and Methods”) between the DCIS and the paired invasive cancers was 77% (range, 17–100%). The majority of the pairs (9 of 12 cases) showed a strong similarity between the DCIS and IDC with few differences (mean concordance of 91%; Table 2), whereas the remaining 3 cases appeared to be less genomically related (mean concordance of 35%). Four DCIS/IDC pairs were isolated from the same archival paraffin block, whereas eight were from different blocks in the same surgical tissue specimen. All three cases with lower concordance were from separate blocks.

This study clearly confirms that DCIS is a genetically advanced lesion, with almost the same degree of genomic instability as invasive cancer. The extent to which the genome is altered in DCIS indicates that genomic instability most likely precedes phenotypic evidence of invasion. Genomic data have supported several possible models of breast cancer progression. One hypothesis suggests that there exist genetically distinct subgroups of DCIS, only some of which ever have the potential to progress to invasion (14). An alternate dedifferentiation theory proposes that DCIS progresses from lower to higher grade and then to invasive cancer with progressive accumulation of genomic changes (15, 16, 17). Perhaps the most compelling model is one in which there are distinct genomic progression pathways leading from any DCIS to invasive cancer. Different histological grades are thus associated with specific genomic changes, and each DCIS pathway may carry a distinct invasive potential (18).

In this latter model, specific genomic changes are selected for from random genomic alterations arising from an unstable environment. Phenotype, particularly histological grade, is likely the result of distinct and different genomic changes. It has been shown that nuclear grade of DCIS is one of the most important clinical determinants of biological behavior, with high-grade tumors more likely to recur than low-grade tumors, even after optimal therapy (19, 20, 21). Prognosis is therefore likely to be a direct consequence of tumor genetics. In our study, when pure DCIS was divided into low/intermediate- versus high-grade groups, clear genetic differences between grades were apparent.

Lower grade DCIS was more likely than high-grade DCIS to show loss of 16q. Three small loss of heterozygosity studies have reported that loss of 16q is among the most frequent genomic changes seen in DCIS (6, 22, 23). CGH studies have confirmed and expanded on these results by analyzing gains as well as losses along the entire genome. Our data substantiates previous smaller studies that indicate loss of 16q is an early event most often found in lower grade DCIS lesions (24, 25, 26). We have previously found that loss of 16q can even be detected even at the stage of atypical ductal hyperplasia (27). Thus, loss of 16q appears to be a critical event in early cancer progression.

An important strength of our study is that the entire set of in situ and invasive cancers were evaluated under the same experimental conditions and with identical analytical techniques, thus allowing direct comparison between subsets of invasive cancer and DCIS. Comparison of genomic changes between preinvasive and invasive tumors showed that loss of 16q, typical of low/intermediate-grade DCIS, was also seen in some groups of invasive cancers. Chief among these were the tubular cancers, a subtype known to have a very favorable prognosis and characterized by low grade. Loss of 16q was seen in 78% of tubular cancers versus 34% in nontubular invasive ductal cancers. This corroborates observations by other groups who have also found that 16q alterations appear to be specific to low-grade invasive tumors (28).

Another subgroup of invasive cancers commonly demonstrating 16q loss were those of lobular histology. We have previously reported that 16q is also lost in most lobular carcinoma in situ (29). In invasive lesions, loss of 16q was highly prevalent in ILC when compared with IDC. In addition to increased prevalence, the pattern of 16q loss differed markedly between lobular and ductal cancers. Few IDCs had concomitant 16q loss and 1q gain, compared with over half of ILCs and tubular cancers. Fluorescence in situ hybridization analysis has shown that translocation of the long arms of chromosomes 1 and 16 often occurs in ILC and tumors of low grade (30, 31, 32). This is possibly a result of reciprocal translocation at centromeric heterochromatin. The lower frequency of 1q+/16q− translocation in IDCs than in lobular and tubular tumors suggests that the selective pressures resulting in 16q loss in tubular and lobular cancers (including lobular carcinoma in situ) may target different genetic loci when compared with IDC. Furthermore, this implies that lobular carcinomas in situ, ILCs, and tubular cancers may share a common progression pathway, a suggestion that is supported by the frequent coexistence of tubular and lobular patterns (33).

Several candidate genes mapping to 16q may be important in cancer progression. Perhaps best studied is E-cadherin (CDH1), located on 16q22.1. This glycoprotein has been shown to be critical for cell adhesion and morphogenesis and is markedly absent by immunohistochemistry in ILCs and lobular carcinomas in situ, as well as in some tubular cancers. Perturbation of normal E-cadherin function either through promoter methylation or gene mutation has been implicated in lobular tumorigenesis (28). Buerger et al. (26) has correlated CGH findings in DCIS with mitotic index and has found that loss of 16q was associated with low tumor proliferation rate. Because regulation of cell turnover is not typically attributed to CDH1, it is possible that for ductal tumors, there may be other important regions of chromosome 16q subjected to selective pressure. Such candidate genes include E2F4 (16q22.1), a cell cycle transcription factor, and WWOX (16q23.3-24.2), which appears to be a potent tumor suppressor gene in breast cancer as well as in esophageal malignancies (34).

8p was another region frequently showing alterations in low-grade DCIS. Interestingly, we observed several distinct patterns of 8p alteration in DCIS. The loss of 8p involved either the whole arm or was a partial loss in the 8p21-pter region often accompanied by a concomitant gain at 8p11-12, a pattern that may reflect a targeted translocation. Forty-one percent of low/intermediate-grade DCIS showed 8p loss. Of these, the majority had a partial loss of the distal arm of 8p, whereas almost all high-grade DCIS with 8p loss exhibited loss of the entire chromosomal arm. In our dataset of invasive cancers, the low-grade tubular cancers more often exhibited partial, rather than whole arm loss, whereas invasive cancers in general were equally likely to have either partial or whole arm loss.

The chromosomal region 8p21-23 is the site of the observed break point in these tumors and has previously been described in breast cancers (26, 35). Fibroblast growth factor receptor 1 (8p11.2-11.1) appears to be just proximal to the breakpoint region and codes for a tyrosine kinase, which is thought to lead to an increased cell proliferation and progression to invasion (36). Heregulin (NRG1; 8p22–11) is a glycoprotein which binds to ERBB4 to regulate cell growth and proliferation (37), and maps just distal to the breakpoint region. Our studies suggest that fibroblast growth factor receptor 1 is gained and NRG1 is lost in those cases with the partial chromosomal loss on 8p. Such distal loss just adjacent to a proximal gain (or amplification) has been noted in other chromosomal regions, and is postulated to be related to a breakage/fusion/bridge mechanism of amplification. With ongoing mapping studies, we hope to more clearly define this breakpoint region at higher resolution.

Some genomic alterations were seen at greater frequency in high-grade DCIS than in lower grade lesions. Most apparent among these was gain of 17q. Gain of 17q has been associated with amplification of ERBB2, located at 17q12 (38). Studies have shown that the erbB2/her2/neu protein product is overexpressed in 60–70% of DCIS by immunohistochemistry (39, 40), although in only 20–30% of invasive breast cancers. The higher prevalence of 17q gain in DCIS versus invasive ductal cancer is somewhat counterintuitive given the precursor-product relationship between them. It is unlikely that an acquired genomic alteration such as gain of 17q and amplification of ERBB2 would be consistently reversed during cancer progression. One possible explanation for this seeming paradox is detection bias, in that high-grade DCIS with ERBB2 overexpression may be more likely to have in situ necrosis and microcalcifications and thus be more apparent on mammography. The increased necrosis might be associated with a faster growth rate and greater hypoxia caused by the physical restraint of the intact basement membrane in DCIS. Alternatively, the higher prevalence of erbB2 positivity in DCIS when compared with invasive cancer could also be explained if not all DCIS progresses to invasive cancer. In addition to ERBB2, chromosome 17q also contains GRB7 (17q21-22), involved in cell cycle activation, and TBX2 (17q23), an immortalizing gene that has been identified in breast cancers (41). It is possible that selected gain of these genes also confers some survival advantage, particularly in high-grade DCIS.

17p loss was seen frequently in both low- and high-grade DCIS. This genetic alteration is seen in many subtypes of breast cancer and is detected in over a third of all DCIS and invasive cancers, although interestingly, it is uncommon in tubular cancers. It has been suggested that p53 (17p13.1) is the target for this deletion, releasing cells from its checkpoint function controlling apoptosis, proliferation, cytoskeletal regulation, and DNA repair (42, 43, 44). Our findings suggest that loss of p53 may be important early in the development of many preinvasive cancers, conferring a survival benefit during this time of increased proliferation, and that 17p loss may be an important mechanism for cancer progression in DCIS regardless of grade.

The majority of cases in our small set of paired DCIS and synchronous invasive cancer were clonally related by CGH. Loss of heterozygosity studies have reported similar findings for DCIS and concurrent IDCs (4, 5, 6). The synchronous DCIS and IDCs in our study appear to be much more similar than different, as indicated by the high degree of concordance in 9 of 12 cases (mean concordance was 91%). The remaining three cases had a mean concordance of 35%, suggesting that in these cases, the IDC either had progressed from an entirely different tumor clone than the DCIS component or the invasive component separated from the DCIS at a very early stage of tumor evolution.

Given the heterogeneity of genomic alterations seen in both DCIS and invasive cancer, it is unlikely that there exists only a single pathway for breast cancer progression. Low/intermediate-grade DCIS manifests many of the same genetic changes as low-grade invasive cancers such as gain of 1q, loss of 16q, and distal loss of 8p. Similarly, high-grade DCIS and many invasive ductal cancers share whole arm loss of 8p and 18q, as well as gain of 17q. Our paired cases clearly show that when synchronous invasive and in situ cancer coexist, both components are likely to share virtually identical genomic changes. These findings are consistent with a model in which some lower grade DCIS may progress to low-grade invasive cancer, and some high-grade DCIS is likely to progress to high-grade invasive cancer. Thus, genetic alterations occurring early during in situ tumor development determines the phenotype of the DCIS as well as the subsequent invasive cancer. Tumor genetic signature therefore drives biological behavior, clinical behavior, and ultimately patient outcome. These divergent genetic characteristics of in situ tumors will help to establish molecular classification schemes, to serve as potential prognostic markers for behavior of preinvasive cancer, and eventually to identify targets for individualized therapy for this heterogeneous disease.