Serial analysis of gene expression of lobular carcinoma in situ identifies down regulation of claudin 4 and overexpression of matrix metalloproteinase 9

Lobular carcinoma in situ (LCIS) is characterised by small, discohesive epithelial cells that fill, distend and distort the terminal duct lobular units of the breast [1, 2]. LCIS cells, which are cytologically identical to those of invasive lobular carcinoma, frequently contain mucin vacuoles, imparting a signet-ring cell appearance. At the immunohistochemical level, the hallmark of LCIS is loss of the expression of the E-cadherin protein, which results in the loss of cohesion of the cells. Unlike ductal carcinoma in situ (DCIS), a localised proven precursor to invasive breast carcinoma, LCIS tends to be multifocal and bilateral, and typically is neither calcified on mammography nor mass forming on clinical or gross pathological examination. Instead, LCIS is almost always an incidental microscopic finding identified by the pathologist. Less well-developed examples of the same process, in which there is insufficient distention and distortion of the terminal duct lobular units, are classified as atypical lobular hyperplasia (ALH).

Based on several classic studies that showed the invasive carcinomas that follow LCIS are often invasive ductal carcinomas (IDC), and that the risk of breast cancer was almost equal in each breast [3, 4], LCIS has traditionally been viewed and managed as a marker of bilateral breast cancer risk. However, this concept is difficult to reconcile with several other clinical, morphological and molecular observations that suggest that LCIS is instead a cancer precursor. First, several other follow-up studies have shown that the majority (three of four) of cancers that follow ALH and LCIS are in fact ipsilateral [5‐9]. Second, while many of the carcinomas that follow LCIS are IDCs, invasive lobular carcinoma is over-represented in these cases. It is possible that the frequent occurrence of IDC in patients followed for LCIS may be explained by the frequent co-existence of LCIS and DCIS in these patients, with the DCIS acting as the precursor to the IDC. In fact, when cases of pure LCIS unassociated with concurrent DCIS are studied, the invasive carcinoma that follows is almost always invasive lobular carcinoma [10]. Third, it is not uncommon for LCIS to be associated with microinvasive lobular carcinoma [11], a morphology that strongly suggests that the LCIS gives rise to the invasive lobular carcinoma.

Finally, at the genetic level, multiple studies have shown similarities between LCIS and invasive lobular carcinoma. Identical activating mutations of the E-cadherin gene have been identified in concurrent LCIS and invasive lobular carcinoma [12, 13]. Array Comparative Genomic Hybridisation (CGH) [14] and mitochondrial DNA analyses have demonstrated marked similarities between matched LCIS and invasive lobular carcinoma, further supporting a clonal relationship. Moreover, LCIS demonstrates methylation of the same cancer specific genes found in DCIS, IDC and invasive lobular carcinoma [15]. Hence, many now view LCIS as a low-risk direct precursor to breast cancer that tends to be bilaterally distributed [16, 17]. Under this view, the more frequent bilateral distribution of LCIS accounts for the greater proportion of contralateral cancers that follow LCIS as compared with DCIS.

Given the growing view that LCIS represents a cancer precursor, the identification of molecular alterations that suggest potential therapeutic and chemopreventive targets within this lesion is of great interest. At this point, studies have been limited by the typically small, microscopic extent of LCIS, which precludes obtaining frozen tissue with intact, high-quality DNA and RNA. At the DNA level, CGH studies have consistently demonstrated loss of the long arm of chromosome 16, where the E-cadherin gene is located [18, 19]. RNA-based expression profiling analyses have not thus far been feasible.

In this study, we report the gene expression profile of a florid case of LCIS by serial analysis of gene expression (SAGE). SAGE provides an unbiased and quantitative assessment of the cellular transcriptome, and it is a reliable tool for identifying differentially expressed transcripts in cancer. In the remarkable case studied, the LCIS formed a mass lesion, such that frozen tissue could be obtained for genetic analysis. We validate several up and down regulated genes at the RNA and protein level, and identify dysregulated pathways in this unusual cancer precursor.

SAGE is an unbiased and quantitative method of identifying differentially expressed transcripts in human cancer, enabling discovery of cancer biomarkers, therapeutic and chemopreventive targets. For example, in 2001, SAGE was used to first demonstrate that mesothelin [34] and prostate stem cell antigen (PSCA) [35] were overexpressed in pancreatic adenocarcinoma. Mesothelin and PSCA have subsequently been utilised as adjuncts in diagnosis, as candidate serum biomarkers and as targets of pancreatic cancer immunotherapy. We report herein the first SAGE analysis of LCIS. Such studies have previously not been possible because of the microscopic nature of LCIS.

The unique nature of the index case of this study, specifically the fact that it formed a mass, allowed the lesion to be identified grossly such that a small amount could be frozen for gene expression studies. Although our study is somewhat limited by the fact that the analysis is only of a single LCIS case with some unusual features (mass-forming lesion, central necrosis), the LCIS LSAGE library demonstrated differential expression of several genes (E-cadherin, cyclin D1), which was concordant with the published literature on these genes in LCIS. Moreover, we validated the results immunohistochemically for novel genes that are both up regulated (MMP9) and down regulated (claudin 4) relative to normal ductal epithelium on a larger series of more typical LCIS cases. Therefore, we believe our results are more globally applicable to LCIS and that alterations identified in other pathways not specifically validated in this study are nonetheless very likely to be significant.

One key theme of our SAGE analysis was the observed down regulation of cellular junctional proteins in LCIS. Such down regulation should have been expected. Epithelial cell junctions are comprised of both tight and adherens type junctions. Inactivation of the E-cadherin protein, a key component of the adherens junction that interacts with the actin cytoskeleton via binding of alpha, beta and delta catenins, is a hallmark of LCIS. Previous studies have demonstrated inactivation of the entire E-cadherin complex in LCIS, including loss of expression of alpha-catenin and membrane to cytoplasmic translocation of p120 catenin [22‐25]. Therefore, the down regulation of E-cadherins that we observed in the index case and the four cases in the validation set is consistent with the published literature, and supports the concept that the adherens junction complex is inactivated in LCIS.

The loss of the tight junction protein claudin 4, validated by immunohistochemistry in this study, has not been previously reported in LCIS. However, these results are concordant with current knowledge of claudin biology and the known expression pattern of claudin 4 in related carcinomas. First, at the cellular level, tight junction formation is thought to depend on intact adherens-based intercellular contacts [36]. Therefore, because of the absence of E-cadherin and therefore the absence of a functional adherens junction in LCIS, the absence of tight junctions (and therefore claudin 4) should be expected. The striking parallel expression of E-cadherin and claudin 4 on serial sections of the remarkable case of DCIS analysed in this study (Figure 4) further supports the notion that the expression of E-cadherin and claudin 4 proteins is tightly linked.

Second, although no previous study has examined LCIS, several studies have examined expression of claudin 4 in invasive lobular carcinoma. Two studies reported intact claudin 4 expression in invasive lobular carcinoma [37, 38]; however, these studies did not compare expression with the normal duct epithelium, and used as little as 10% labelling as a criterion for positivity. The one study that compared labelling to that of the normal duct epithelium reported down regulation of claudin 4 in two of two cases of invasive lobular carcinoma [39]. Moreoever, it has been reported that diffuse gastric-type adenocarcinoma, which cytologically and immunohistochemically resembles invasive lobular carcinoma in that it is composed of discohesive signet-ring cells that lack E-cadherin protein expression, demonstrates down regulation of claudin 4 protein [40, 41]. Hence our results with LCIS are consistent with those seen in related invasive carcinomas.

Importantly, these results have therapeutic implications. Our group has shown that claudin 4 is up regulated in IDC, and is a proposed target for therapy and chemoprevention [31]. Our results show that such an approach may not be applicable to LCIS or to invasive lobular carcinoma, because these lesions typically underexpress claudin 4 relative to normal breast epithelia.

In contrast, we were surprised by the overexpression of MMPs in LCIS, and therefore we thought that it was important to validate and verify this result using multiple methods. Indeed, we show herein that MMP9 is over expressed at the RNA and protein level in LCIS by quantitative RT-PCR and immunohistochemistry, respectively. No previous study has systematically evaluated MMP9 expression in LCIS. However, one study did find overexpression in invasive lobular carcinoma relative to IDC, and noted that a single case of concurrent LCIS also showed MMP9 expression [42].

Moreover, recent data in ovarian and pulmonary carcinomas has suggested a causal inverse relationship between E-cadherin and MMP9 expression, which would fit our finding of high MMP9 expression in LCIS that characteristically lacks E-cadherin [43, 44]. MMPs are postulated to be important in the degradation of basement membrane and extracellular matrix during cancer cell invasion. In particular, MMP9 and MMP2 (also know as gelatinases) are involved in the degradation of type 4 collagen that comprises basement membranes, a process thought to be important in the development of invasive carcinoma and metastasis. Although degradation of basement membranes is not a function that is concordant with the usual clinical behaviour of LCIS (a non-invasive, low-risk cancer precursor), we suspect that the activity of MMP9 in LCIS must be tightly regulated. Indeed, MMP2 and MMP9 are secreted as zymogens and cleaved to their active form, and their activity is tightly regulated by several mechanisms, including the actions of tissue inhibitors of metalloproteinases (TIMPS). It is certainly conceivable that the expression of activated MMP9 helps effect the characteristic, permeative, 'Indian file' growth pattern of invasive lobular carcinoma. Our study indicates that MMP9 RNA and protein is already present at the precursor (LCIS) stage of invasive lobular carcinoma, and raises the possibility that factors leading to its activation may play a role in the development of invasive disease. MMP9 therefore represents an interesting therapeutic [32] and chemopreventive target for patients with lobular neoplasia, both invasive and non-invasive.

Finally, we note that the down regulation of cell junction proteins (such as E-cadherin and claudin 4) and up regulation of MMP9 that we observed may be taken by some to be evidence of so-called epithelial-mesenchymal transition in LCIS. Indeed, the transcription factor Twist is known to down regulate E-cadherin expression in invasive lobular carcinoma, and is postulated to be a master regulator of epithelial-mesenchymal transition and metastasis [45, 46]. Although we agree that the abnormalities we have elucidated in LCIS in this study do overlap with those described in vitro in epithelial-mesenchymal transition, LCIS remains a cytokeratin-positive, epithelial neoplasm.

In conclusion, an unbiased analysis of the LCIS transcriptome by SAGE has identified MMP9 overexpression and claudin 4 underexpression. This LSAGE library is publicly available at the Cancer Genome Anatomy Project SAGE website [20], and should facilitate further, much needed research on this enigmatic lesion.