A Recurrent Chromosome Breakpoint in Breast Cancer at the NRG1/Neuregulin 1/Heregulin Gene

Large-scale rearrangements of the genome in common epithelial cancers such as breast and colorectal cancer remain poorly characterized. Most interest has focused on regional gains and losses, but it would be surprising if these cancers did not also show rearrangements that break at specific genes, creating gene fusions or promoter insertions. The most familiar examples of such rearrangements are reciprocal chromosome translocations, but gene-specific breaks could also occur at unbalanced translocations, inversions, insertions, or deletions. In all cases they would be detected as recurrent breakpoints at a specific gene. We recently described recurrent breaks at the NRG1 gene in seven cancer cell lines, five from breast and two from pancreatic cancers (1). One of these breaks is at a reciprocal chromosome translocation; five are at unbalanced translocations; and one borders a deletion (1). A plausible interpretation of all these breaks is that they cause abnormal transcription of the NRG1 gene. However, to date, evidence for rearrangements that break in NRG1 comes only from cell lines. We have now devised a screen to detect the breakage of NRG1 in paraffin sections of tumors, and we report here that it occurs in uncultured breast tumors and, indeed, in some other cancers.

The tumors and tissue microarrays made from them are summarized in Table 1. Their use had appropriate local ethical approval. We used a specific arraying device to prepare arrays as described (Beecher Instruments, Silver Spring, MD; refs. 2, 3). Cambridge arrays comprised breast cancers treated at various centers in the United Kingdom between 1991 and 2003. Array Cambridge 1 comprised 18 primary node-positive tumors from the MRC multicenter ABO1 Molecular Markers Clinical Trial, whereas array Cambridge 2 comprised 101 arbitrarily chosen tumors. These arrays had respectively quadruplicate and duplicate cores of each tumor. Of the 109 of these tumors successfully screened (see below), 89 were invasive ductal carcinomas, 11 were invasive lobular carcinomas, and 9 were “other” (5 mixed, 1 tubular, 1 mucinous, 1 atypical medullary, and 1 intracystic papillary). Sixty-three tumors were grade III, 35 tumors were grade II, and 8 tumors were grade I, with 3 not graded. Estrogen-receptor status was available for 75 tumors, of which 72% were positive. The Marseille arrays (Table 2) were from consecutive cases with localized invasive breast carcinoma treated at the Institut Paoli-Calmettes between 1981 and 1999. Five-year survival was available for 177 of the 213 tumors successfully screened by fluorescence in situ hybridization (FISH), including all of the breakpoint-positive cases. The representative quality of the arrays was assessed by comparison with conventional sections for the usual prognostic parameters, including estrogen receptor and ErbB2: the value of the kappa test was 0.95 (3). The Ambion array included assorted carcinomas and was supplied by Ambion Inc. (Austin, TX); the NCI TARP array was an array of epithelial ovarian cancers from the NCI TARP project.

Bacterial artificial chromosome (BAC) clones containing inserts from chromosome arm 8p were selected with the Washington University fingerprint map and the UCSC Golden Path draft human genome sequence.⁵ All were from the RP11 library except where indicated. They were obtained from the Wellcome Trust Sanger Institute (Hinxton, United Kingdom) or from Invitrogen (Paisley, United Kingdom). All BACs used in FISH had been checked for chimerism and location on 8p with conventional FISH on normal metaphases (4).

FISH on 5-μm paraffin sections was as described previously (5). To screen for breakpoints in NRG1, probes were made by pooling overlapping BAC clones as shown in Fig. 1,A, labeled with FITC or digoxigenin, hybridized, and detected respectively with rabbit anti-FITC antibody and FITC-labeled goat antirabbit antibody (both from Sigma, St. Louis, MO) or rhodamine-labeled Fab fragments of sheep antidigoxigenin antibody (Roche, Lewes, United Kingdom). Some tumor cores in the sections could not be screened as they were damaged, failed to hybridize, or had no tumor (Table 1). Areas enriched for tumor cells were identified by reference to near-adjacent sections stained with H&E, and fluorescence was scored on a minimum of 50 nuclei per tumor. To map breakpoints in sections from original paraffin blocks, probes were made by pooling BACs. BACs 97N12 + 212N14, 650B11 + 147C21, or 275E10 + 669B22 were hybridized in green together with CTD-2329M5+GS1–57G24 + 11N9 in red. For tumor T92, additional single BACs GS1-57G24, 11N9, and 317J8 were used to confirm the position of the breakpoint.

For array-comparative genomic hybridization, DNA was extracted, as described previously (6), from paraffin blocks of tumor and a block of normal breast taken at reduction mammoplasty. Tumor DNA was labeled with Cy3-dUTP, normal DNA was labeled with Cy5-dUTP (Amersham), and they were hybridized together as described by Fiegler et al. (7) to a custom 500-clone genomic array. This comprised 58 BAC clones covering 30.9 Mb (RP11-473A17) to 40.5 Mb (RP11-51K12) on chromosome 8 according to human genome sequence NCBI Build 34, with flanking clones at intervals averaging 1 Mb over chromosome 8 and 10 Mb over the rest of the genome. BAC DNAs were spotted in triplicate. BAC hybridization ratios were only plotted where at least two of triplicate spots had ratios within 5% of each other. This was achieved by rejecting spots if their hybridization ratio was more than 5% different from the median of triplicates, then if a minimum of two spots were accepted, plotting the mean of the accepted spots. For the two tumors that gave good hybridizations, no BACs within the NRG1 region failed this criterion; whereas for the other three tumors, at least 35% failed, and the hybridizations were discarded.

The two antipeptide polyclonal antibodies to NRG1, respectively to the α and β isoforms, have been described previously (8). Endogenous peroxidase was blocked with 3% H₂O₂ for 10 minutes, then primary antibody was incubated overnight at 4°C, and detected with biotinylated antirabbit antibody then Streptavidin ABC complex/Horseradish Peroxidase (Dako Corporation, Carpinteria, CA). Other antibodies are detailed in Supplementary Table 1. Slides were evaluated by Jocelyne Jacquemier or William J. Gullick without knowledge of breakpoint status. For NRG1α or β, faint cytoplasmic staining (scored 1) was considered negative, and good cytoplasm staining (scored 2 or 3) with or without membrane staining was considered positive.

We devised a screen to detect breaks at NRG1 in paraffin sections of tumors, with FISH, and applied it to tissue microarrays, i.e., sections that contain multiple small cores of many tumors (9). NRG1 is located on the short arm of chromosome 8 at 8p12. As a result of the rearrangements found in the cell lines (1), distal 8p is lost and is replaced by another chromosome fragment, different in each cell line, whereas most or all of the NRG1 gene is retained (Fig. 1,A). (In the case of the deletion NRG1 appears to be joined to a telomeric fragment of 8p.) Therefore, we screened for the NRG1 break in tissue sections by looking for separation of sequences that flank the breakpoint. Hybridization probes were made for sequences proximal and distal to the breakpoints found in the cell lines, by combining overlapping BAC clones (Fig. 1,A). They were respectively labeled with red or green fluorescence and hybridized together to sections of tumor microarrays (5). Intact copies of the NRG1 region were seen as a pair of adjacent red and green signals, whereas broken copies appeared as isolated red signals (Fig. 1, B and C). The screening assay proved robust: two observers independently scored 110 arrayed tumors and identified the same positive cases.

In our first set of 110 tumors screened (Cambridge arrays 1 and 2; see Table 1), 6 cases were positive for breakage of the NRG1 region (6%).

To map the breakpoints to higher resolution and confirm the screening result with independent probes, the original paraffin blocks were recovered for five positive tumors. Sections were first hybridized with the screening probes, to repeat the screening result, then sections were hybridized with other BACs from the NRG1 region. The observed breakpoints were distributed over the 5′ end of the gene as previously seen in cell lines (Fig. 1,A). The properties of the five tumors are summarized in Table 3.

Other details of the breaks paralleled the cell line breaks. In one of the breast cancer cell lines, HCC1937, there is a reciprocal translocation of NRG1, i.e., the reciprocal product is also present. Similarly, in some of the tumors, notably T21 (Table 3), isolated green signals were seen in addition to isolated red signals, indicating a balanced rearrangement such as a reciprocal translocation. In three cell lines, ZR-75-1, Suit-2, and SUM-52, the NRG1 breakpoint is present in duplicate or multiple copies. Similarly, in some tumors, e.g., T92, the red signals were in pairs or several copies, suggesting duplication of the breakpoint junction.

To confirm the presence of the breaks by a completely independent method, we used array-comparative genomic hybridization. Array-comparative genomic hybridization would detect the breaks as a decrease in copy number of sequences distal to the break, except where they were reciprocal translocations. DNA was recovered from paraffin sections of the five positive tumors, directly labeled and hybridized to a genomic DNA microarray that includes 58 BACs that cover 9.6 Mb of 8p12 (Fig. 1,A). DNA from paraffin sections often performs poorly in such hybridizations, and we were not surprised that only two of the five samples gave hybridizations of acceptable quality. Array-comparative genomic hybridization confirmed the FISH results: there was a decrease in copy number distal to the breakpoint mapped by FISH. Tumor T92 showed a copy number change at about 32.3 Mb (Fig. 1 A) and T70 at 31.4 Mb (data not shown).

The prevalence of the NRG1 break, and relation to other features of breast cancers, was then determined in an independent set of unselected cases of localized invasive breast cancer, most with greater than 5 years follow-up (Marseille arrays 1 and 2; Table 1). Using the same FISH screening strategy, we detected the breakpoint in 6% (13 of 213) of tumors that gave satisfactory hybridization, the same frequency as in the Cambridge tumors (bringing the total of positive tumors to 19 of 323). A high proportion of the breakpoint-positive tumors were histologic grade III: 65% compared with 28% of negative cases (significant at P < 0.05 by Spearman log-rank test, without correction for multiple tests; Table 2). Otherwise, breakpoint-positive tumors were not clearly different from negative tumors, although, as numbers were low, it would have been difficult to reach statistical significance for associations with most parameters. Positive tumors were, like breast tumors in general, mainly ErbB2/HER2 negative (11 of 13 low) and estrogen receptor positive (11 of 13 positive). There was no significant association with 5-year survival (although only one positive patient died, at 5.1 years), histologic type (1 was lobular, 11 ductal, and 1 other), node status, or antibody staining for epidermal growth factor receptor (EGFR)/ErbB, ErbB2/HER2, ErbB3, ErbB4, or estrogen receptor. However, there was a significant correlation with strong staining for FGFR1 and TACC1, the genes of which are close together at about 38Mb on 8p11-12, suggesting an association of NRG1 breakage with amplification of this region (1).

To detect expression of NRG1 products, three of the four arrays (Marseille array 2 had been used up) were stained with antibodies specific for α and β isoforms of Heregulin/NRG-1. All nine breakpoint-positive tumors in the arrays stained gave clear cytoplasmic staining with at least one antibody. However, many negative tumors also stained: overall 47% (93 of 197) of tumors stained for NRG1-α (as reported previously; ref. 10) and 63% (126 of 201) for NRG1-β.

We used FISH on arrays of other carcinomas to perform a preliminary screen for NRG1 breakage in other tumor types: we detected it in one of two squamous-cell lung carcinomas and two of 35 ovarian carcinomas (Table 1).

Our FISH screen shows that the NRG1 locus is broken in clinical breast cancers and not merely in cell lines. The breakpoint appears to be present in several other types of epithelial cancers, because we detected it in one squamous-cell lung carcinoma and two ovarian adenocarcinomas, and we had previously detected it in two pancreatic cancer cell lines (1). All of these cancer types can have abnormalities of the ErbB family.

By analogy with the cell lines, it is possible that most of these breaks represent chromosome translocations, but it was not feasible to test directly for the translocations in paraffin-embedded material: we would have to know what the translocation partners are, and they are different in each of the cell lines tested thus far (1). Some of the breakpoints detected could be because of interstitial deletion rather than translocation, as in one of the seven cell lines with the breakpoint. However, functionally this can still have the same consequences: both translocation and deletion can lead to fusion of the 3′ end of NRG1 to other sequences.

One attractive hypothesis is that these breaks in NRG1 cause expression of normal or abnormal NRG1 transcripts and so create an autocrine loop. Indeed, there is already evidence for this in one of the cell lines, MDA-MB-175. NRG1 encodes a family of growth factor-like proteins, the Heregulins or Neuregulins 1, that bind to the receptor tyrosine kinases ErbB3/HER3 and ErbB4/HER4, and heterodimers of these receptors with ErbB2/HER2. The breaks we have described are within the NRG1 gene, or immediately upstream of its 5′ end, and they probably all preserve the receptor-binding domain of NRG1, which is encoded by the last cluster of exons at 32.7 Mb (Fig. 1 A). In one cell line, MDA-MB-175, it is already known that the chromosome translocation creates a fusion transcript that includes the 3′ end of NRG1 (11, 12). The resulting fusion protein is secreted and has heregulin-like activity (13). Receptors of the ErbB family are normally present on mammary epithelial cells, and Neuregulins stimulate mammary epithelial growth and differentiation (14, 15). Overactivity of the ErbB2 pathway is particularly potent at driving mammary tumor development in mice (16, 17). The breakage of NRG1 might, however, have more complex effects, because there are multiple splice forms of NRG1 with different activities, and modification of the 5′ end can dictate which splice forms are made (1, 14).

The implication of our findings is that chromosome rearrangements in common epithelial cancers can consistently break at specific genes, and so could result in gene fusions or promoter insertions, as they do in many less common cancers (18, 19). There may be many more consistent breakpoints like this in common cancers, because relatively little effort has gone into mapping such breaks to the gene level. This has been, in part, for technical reasons: it is difficult to obtain metaphase chromosomes from surgical material, and the chromosome rearrangements are often too complex and variable for classical techniques to identify them. But also, there has been a perception that rearrangements are selected merely because they result in loss or gain of regions that contain critical genes. There may also have been an expectation that consistent breakpoints that target a particular gene would occur mainly in reciprocal translocations between the same two chromosomes and would be present in most cases of a given type of cancer. This is based on the chronic myeloid leukemia paradigm, where 90% of cases show the same t(9;22) translocation. However, even other leukemias show much more variability (20). The NRG1 breaks in the cell lines are quite different: they are mostly in unbalanced rather than reciprocal translocations; the broken NRG1 is joined to a wide variety of partner chromosome fragments; and they are present only in a modest proportion of cases (1). Many more gene-specific breakpoints may therefore remain to be discovered.

We thank Edith Blackburn and Peter Sibley for staining and Jim Neal and Jeannine Geneix for tissue array construction.