The transcription factor ATF3 acts as an oncogene in mouse mammary tumorigenesis

As described previously [19], transgenic BK5.ATF3 mice were obtained by embryo injection in FVB/N mice. The construct for these experiments contained the human ATF3 cDNA sequence [23] in an expression vector containing the bovine keratin 5 promoter [24]. This promoter is active in basal epithelial cells in several tissues, including skin, thymus, and mammary gland (vide infra). Five independent lines that transmitted the transgene were established; four of these displayed an easily observable sparse hair phenotype. None of these lines showed significant neo-natal mortality, and only line 2 produced significantly fewer transgenic than non-transgenic progeny (4.1 ± 2.2 non-transgenic pups at weaning, 2.2 ± 2.0 transgenic pups at weaning, p = 0.00075). We have been unable to demonstrate transgene expression in the fifth line (line 6, no observable hair phenotype) in either skin or mammary gland, and this line was not studied further.

Genotyping was performed using genomic DNA purified from tail snips as template. For most of these studies, a previously described PCR assay [19] was used. Alternatively, a real time PCR assay for the coding region of human ATF3 (Hs00231069_m1, Applied Biosystems, Foster City, CA) was utilized, with analysis on an Applied Biosystems 7900 HT. A positive control assay for an endogenous murine gene (Atf3, Gapdh or Rash) was also performed with each DNA sample.

Tissues obtained at necropsy were fixed in buffered neutral formalin and paraffin embedded. Sections were prepared and stained (either with hematoxylin and eosin or for IHC) by the Tissue Processing Facility Core of the Center for Research on Environmental Disease. Tissue microarrays containing normal human mammary samples and breast cancer samples were the kind gift of Dr. A.J.P. Klein-Szanto (Fox Chase Cancer Center), or were purchased from Zymed (San Francisco, CA). For IHC, antigen retrieval was performed as described [19]. Binding of primary antibodies to tissue sections was visualized with a chromogenic substrate, using a secondary antibody coupled to horseradish peroxidase as described [19]. The sources of the primary antibodies were as follows: CK5, CK6, CK10, Covance (Berkeley, CA); CK8, Developmental Studies Hybridoma Bank (U. of Iowa, Ames, IA); Ki67, Dako (Carpinteria, CA); ATF3, ERα, ErbB2, Santa Cruz (Santa Cruz, CA). Polyclonal antibodies specific for inner root sheath keratins mIRSA2 and mIRSA3.1 [25] were generously provided by Dr. R.M. Porter (University of Cardiff). The blocking peptide for ATF3 IHC (sc-188p) was obtained from Santa Cruz (Santa Cruz, CA). IHCs for each marker were prepared from at least 10 mammary tumors obtained from 10 different animals. For quantitative analyses of cytoplasmic expression of CK10, over 1000 cells/tumor were evaluated in three micrographs prepared from each of the 10 tumors.

The fourth and fifth mammary glands were obtained from virgin female mice, either non-transgenic FVB, or from BK5.ATF3, lines 1–4. Nuclear extracts were prepared using the NE-PER kit (Pierce). 60 μg of nuclear extract were electrophoresed on 4–20% SDS-polyacrylamide gels and transferred onto Immobilon-FL membranes (Millipore); RAW 264.7 whole cell lysate (Santa Cruz, sc-2211) was used as a positive control. Rabbit polyclonal anti-ATF3 or anti β-actin antibodies (Santa Cruz) and ECL Plus (Amersham) were used to detect the respective proteins.

The incidence of squamous metaplastic lesions in nulliparous females was compared between transgenic and non-transgenic animals by Fisher's exact test. Differences in survival between genotypes and groups were analyzed by logrank test. Quantitative differences in litter size and transmission of the transgene were analyzed by t-test.

As described elsewhere [19], transgenic BK5.ATF3 animals exhibit several epidermal phenotypes, including sparse hair, marked hyperplasia of the outer root sheath cells of the hair follicle, inability to complete the hair cycle, and mild chronic hyperplasia of the interfollicular epidermis. Late in life, hyperplasia, dysplasia and frank neoplasia is seen in the oral cavity, but no epidermal tumors have been observed.

The ducts of the mammary gland are composed of two layers of epithelial cells: an outer layer of myoepithelial cells that express CK5 (Figure 1a, red) and an inner layer of luminal cells that express CK8 (Figure 1a, green). Thus, the myoepithelial cells of BK5.ATF3 mice might be expected to express ATF3, and this was confirmed by IHC in mammary gland sections of 8 week old BK5.ATF3 mice (Figure 1b,d). The immunohistochemically-detected ATF3 protein appeared to be concentrated in the nuclei of the myoepithelial cells, as expected for a transcription factor. No ATF3 expression was detectable by IHC in the mammary glands of wild type littermates (Figure 1c,e), although the antibody used cross-reacts with ATF3 of both human and murine origin [26]. This is not unexpected, since ATF3 expression is typically only detected after treatments that induce DNA damage or other conditions of cellular stress [19‐22]. Analysis of nuclear extracts by immunoblotting revealed strong expression of the protein in BK5.ATF3 line 1 (Figure 1f, lane 1), weaker expression in lines 2–4 (lanes 2–4) and barely detectable expression in non-transgenic animals (lane WT), presumably due to endogenous ATF3.

When mammary glands were examined in somewhat older virgin females (14–32 weeks), numerous dysplastic lesions were found in all four independently derived BK5.ATF3 transgenic lines. For many of the affected glands, multiple ducts exhibited squamous metaplasia (Figure 2a, arrows), and a sizeable fraction of the gland was abnormal. The lumens of the affected ducts were filled with keratin, and, depending on the plane of the section, often appeared as cysts. In BK5.ATF3 line 1, squamous metaplastic lesions were seen in 15 of 21 transgenic females examined; no lesions were seen in 18 non-transgenic animals examined (Table 1). This difference in incidence is statistically significant (Fisher's exact test, p = 2.84 × 10^-6). The other three transgenic lines examined exhibited squamous metaplastic mammary lesions in from 46.2 to 66.7% of the animals examined (Table 1); in all cases the difference between transgenic and non-transgenic animals was statistically significant (Fisher's exact test, p < 0.001). In transgenic lines 1 and 2, over 20% of the glands contained one or more lesions; this fraction was somewhat lower in the other two lines (Table 1). Expression of CK5 in these lesions was most intense in the outermost, basal-like layer of cells, and in all lesions multiple layers of CK5-immunoreactive cells were seen (arrow, Figure 2b). Cytokeratin CK6, not normally observable in mammary ducts of post-pubescent animals, was strongly expressed throughout the lesions (Figure 2c). Expression of CK8 within the lesions was variable; when present, the cells stained intensely (Figure 2d, arrows), but other areas appeared to completely lack CK8 expression (Figure 2d, arrowhead). The basal-like outer layer of cells (denoted by an asterisk in Figure 2e &2f) routinely expressed nuclear ATF3 (Figure 2e, arrow) and the Ki67 proliferation marker (data not shown). Strong staining for the interfollicular epidermal keratin CK10 [27] was also seen in some lesions (Figure 2f, arrow), and when present was noted to be expressed supra-basally. An additional marker of squamous differentiation, keratohyaline granules, could be seen in most lesions (Figure 2e, arrowhead).

Initially, BK5.ATF3 females were used as breeders to maintain the lines by mating with non-transgenic FVB/N males. However, litter sizes at weaning tended to be smaller for transgenic dams (6.4 ± 2.2 pups) compared to non-transgenic dams (8.5 ± 2.9 pups), the pups had lower weights, and the pups (both transgenic and non-transgenic) were apparently unable to suckle from several teats (data not shown). Because of these problems, it was decided to retire the transgenic dams, and maintain the line by mating BK5.ATF3 males to FVB/N dams. Unexpectedly, three of three retired BK5.ATF3 line 1 female breeders developed large mammary tumors between 6 months and 1 year of age.

To better understand this finding, cohorts of non-transgenic FVB/N and BK5.ATF3 line 1 females were either maintained as virgins for 16 months, or mated and allowed to raise pups to weaning twice between the ages of 6 and 13 weeks. All animals were monitored weekly for palpable mammary masses and sacrificed when such masses reached 1.5 cm in diameter or when the animals became moribund. Surviving animals were sacrificed at 16 months. The survival of BK5.ATF3 and non-transgenic, biparous female mice is shown in Figure 3a. No suspicious masses were observed in the biparous wild-type mice (red squares), and none were seen in virgin non-transgenic or BK5.ATF3 mice during the 16 month observation period (data not shown). Mammary tumor incidence reached 67% within the biparous BK5.ATF3 group (blue triangles) by 12 months of age. Logrank analysis of the survival curves indicated a significant difference between the biparous BK5.ATF3 group and the biparous non-transgenic group (p = 0.0005).

At necropsy, the masses were found to be confined to the mammary glands. About 80% of the tumor-bearing animals (12 out of 15; this includes 3 transgenic tumor-bearing mice examined after the original survival study was completed) had masses in more than one gland. All mammary masses occurring in the biparous BK5.ATF3 mice were examined by hematoxylin and eosin staining of paraffin sections and found to be carcinomas with squamous differentiation; examples of the histopathology are shown in Figure 3b–h. The bulk of the tumors consisted of large, cystic lesions with a well-defined epithelium surrounding keratinaceous debris, dead cells and infiltrating inflammatory cells (Figure 3b &3c). Significant atypia and dysplasia could be observed in all tumors. In some areas, the epithelial component was relatively well organized (Figure 3d), with a basal layer one-two cells in thickness, and suprabasal components that mimicked squamous differentiation, including the appearance of keratohyaline granules adjacent to the keratinaceous core (arrows). In other areas, the epithelial component became more hyperplastic (Figure 3e), more anaplastic with increasing signs of atypia (Figure 3f), and finally transitioned into frank neoplasia, exhibiting features of squamous cell carcinoma (Figure 3g &3h). Atypia was noted in all layers (Figure 3h), and atypical, enlarged nuclei were evident (arrow). The stroma of the tumors was extremely cellular, with abundant inflammatory infiltrates. Signs of stromal invasion were seen in most lesions (for example, Figure 3g).

Immunohistochemistry indicated that the outermost, basal layers of tumor cells (that is, furthest removed from the keratinaceous core) typically exhibited nuclear expression of the ATF3 transgene (Figure 4a), and were enriched for dividing cells as shown by expression of the Ki67 proliferation marker (Figure 4b). CK5 immunoreactivity, characteristic of basal epithelial cells, was high in both basal and supra-basal layers of the metaplastic lesions virtually throughout all tumors examined (Figure 4c). In rare areas, CK5 expression was confined to the most basal cell layer (not shown). Stromal components (S) were not immunoreactive for CK5 except for scattered "nests" of tumor cells (Figure 4c, arrowheads). Interestingly, CK8, characteristic of luminal mammary epithelial cells, was also expressed in both basal and supra-basal cells in 14 of 14 tumors examined (Figure 4d). Co-expression of CK5 and CK8 throughout all layers of the epithelial component was the predominant expression pattern in 9 of these tumors. In the remaining 5 tumors, approximately half of the epithelial component of the tumors showed this expression pattern. The remaining areas either exhibited no expression of CK8, or expression was much stronger in basal cells than in supra-basal cells. The widespread, intense expression of CK8 confirms that the tumors arise from mammary epithelium.

In most human breast tumors that express both CK5 and CK8, these two cytokeratins are generally confined to distinct sub-populations of tumor cells. Thus, our finding from the standard IHC studies that the majority of tumor cells in the ATF3-induced murine tumors stained for both CK5 and CK8 was somewhat unusual. To confirm this finding, we repeated the IHC analysis using anti-CK5 antibody tagged with a red fluorescent dye, and anti-CK8 antibody tagged with a green fluorescent dye, and analyzed the results by fluorescence microscopy. Figure 5, panels a and b shows a representative portion of a BK5.ATF3 mammary tumor photographed with filters for the individual dye-tagged antibodies, with the two signals merged in panel c. Groups of fluorescently-labeled tumor cells in a background of unlabeled stroma were seen with both antibodies, and the merged image indicates that these groups are in many cases approximately co-extensive. Within a group, the fluorescent intensities in individual cells varied over a broad range for both antibodies. For example, in the tumor cell island marked with a white asterisk in panel b, highly fluorescent and barely fluorescent CK8-labeled cells were seen, and the physical extent of the CK5 labeling in this island suggests the presence of CK5-stained cells that do not stain for CK8 (asterisk in panel a). Conversely, the white arrowhead in panel b marks a cell that appeared to be CK8-positive and CK5-negative. However, the most common pattern seen was expression of both markers in individual cells, albeit at different levels of intensity. This was particularly evident in the appearance of yellow cells (white arrows) in the merged image of panel c. The predominant pattern of co-expression of cytokeratins characteristic of both the myoepithelial lineage (CK5) and the luminal lineage (CK8) suggests that the tumors may arise from progenitor cells capable of contributing to both major lineages.

Cytokeratin CK6 is generally not found in adult ductal mammary gland epithelium, but has been suggested to be a marker of mammary progenitor cells [14]. Indeed mammary tumors with phenotypic characteristics similar to those described here have been suggested to arise from mammary progenitor or stem cells [14]. Essentially all cells in the suprabasal layers of the BK5.ATF3 mammary tumors clearly expressed CK6 (Figure 4e, arrows). However, CK6 expression was generally absent in the basal, rapidly dividing cell layer (Figure 4e, asterisk), which is expected to contain the least differentiated, most "primitive" cells. Scattered cells within the basal layers did exhibit CK6 expression (Figure 4e, arrowheads). It is worth noting that supra-basal expression of CK6 is characteristically found in hair follicles [28].

Further analysis of the BK5.ATF3 mammary tumors revealed deeper anomalies in cytokeratin expression that are not easily explained solely by mammary stem cell origin. Because the tumors exhibit squamous differentiation, the possibility existed that epidermal differentiation markers would be expressed in the tumors. As basal keratinocytes differentiate and move into the suprabasal layers of the epidermis, cytokeratin expression shifts from CK5 and CK14 to CK1 and CK10, two epidermal-specific keratins [27]. Analysis of ATF3-induced mammary tumors revealed widespread expression of CK10 in 10 of 10 tumors examined (Figure 6a). On average, about 46.6% (range 38–61%, standard deviation 6.3%) of the tumor cells expressed CK10. Generally, CK10 expression was not seen in basal cells (indicated in Figure 6a by asterisks), but was confined to the suprabasal layers (Figure 6a, arrowheads). Uninvolved mammary ducts of tumor-bearing animals were generally negative for CK10 expression, although scattered light background staining of stromal elements was seen (Figure 6b). We also looked for markers of hair follicle differentiation. Type I and type II keratins that are apparently only expressed in the supra-basal inner root sheath cells of hair follicles have been described [25]. IHC analysis with antibodies specific for the inner root sheath-specific type I keratin mIRSa.2 showed widespread, medium intensity staining of supra-basal tumor cells in 10 of 10 tumors examined (Figure 6c, arrowheads). Generally, the basal-most 2–3 cell layers (Figure 6c, asterisks) were unstained. The generalized, supra-basal staining was somewhat more intense than background staining seen in stromal elements of uninvolved, adjacent mammary gland tissue with this antibody (Figure 6d, arrow); no unequivocal staining of ductal epithelial cells was seen in uninvolved ducts. In addition to this generalized, medium intensity staining, scattered single cells and groups of cells stained intensely for mIRS.a2 in all tumors examined (Figure 6c, arrows). A second inner root sheath-specific type I cytokeratin, mIRS.a3.1, gave essentially the same results in 10 of 10 tumors examined (Figure 6e). The generalized, supra-basal staining pattern was weaker, but strong expression in scattered groups of tumor cells was seen in all tumors examined, comparable to the mIRS.a2 staining intensity (Figure 6e, arrows). Again, apparent background staining in uninvolved mammary ducts was seen in scattered stromal elements, and no unequivocal staining of ductal epithelium was seen (Fig. 6f). Thus, not only differentiation markers of two mammary cell lineages, but also differentiation markers of several epithelial lineages of the epidermis are expressed in these tumors. This suggests that over-expression of ATF3 results in fundamental alterations in the differentiation status of the affected tumor cells, including expression of markers characteristic of multiple tissues and epithelial lineages.

Several markers commonly used to classify human breast tumors were analyzed in these tumor specimens to begin to understand the genesis of the ATF3-induced tumors. Expression of ERα is an important prognostic factor in human breast cancer: tumors that are ERα⁺ have a more favorable prognosis. Over 90% (13 of 14 tumors examined) of the mammary tumors that developed in parous BK5.ATF3 female animals showed significant nuclear expression of ERα (Figure 7a). In about 40% of the tumors, expression was widespread, with apparent expression in both basal (arrows) and supra-basal (arrowheads) tumor cell nuclei. In the remaining tumors, expression was high in most regions of the tumor, but other regions were negative for ERα. Within ERα⁺ regions of ten tumors examined in detail, an average of 74% of the cells expressed ERα (range 54–88%). In uninvolved regions of the mammary gland adjacent to the tumors, 50–60% of the ductal cells were ERα⁺ (Figure 7b). In contrast, no tumors examined were markedly positive for ErbB2, a well-known mammary oncogene that is often amplified and overexpressed in human breast tumors. No expression was seen in 4 of 10 tumors examined, and the remaining tumors were marginally positive, with less than 10% of the tumor cells stained. As seen in Figure 7d, even in marginally positive tumors ErbB2 staining was relatively weak (compare with staining in a tumor derived from a transgenic MMTV.neu mouse, Figure 7e). Interestingly, both of these markers were expressed at moderate to high levels in most of the metaplastic lesions seen in mammary glands of virgin BK5.ATF3 females (Figure 7c,f).

Almost 70% of human breast tumors exhibit amplification of parts of the q arm of chromosome 1 [29‐32]. The ATF3 gene is located at 1q32.1, in a region often overrepresented as detected by comparative genomic hybridization studies. Direct measurements of ATF3 copy number by quantitative PCR indicated some amplification in ~80% of human breast tumors [33]. As part of a more extensive study of gene expression signatures in human breast tumors, expression of ATF3 was measured in a series of 28 early (stage 1–2) breast tumor samples. Although ATF3 expression is barely detectable in normal mammary cells, expression at high levels (at least 4-fold higher than the mean of all genes assayed) was detected in many of the samples (manuscript in preparation). Furthermore, overexpression of ATF3 protein in 50% of human breast tumors has recently been reported by Hai and co-workers [33].

To determine whether the overexpression detected at the RNA level and by immunoblotting was reflected in protein expression levels in the epithelial component of human tumors, paraffin-embedded sections were obtained from 18 early breast tumors, similar to those studied at the RNA level, and analyzed for ATF3 protein by IHC. Sections derived from non-tumorous mammary tissue (mammoplasty specimens) were also analyzed. Adjacent sections were analyzed in parallel by IHC for CK5 or after staining with hematoxylin and eosin to better define the cellular elements of the tumors. Many of the tumors exhibited low levels of cytoplasmic expression of ATF3, as well as scattered positive nuclei (Figure 8a, arrows). In addition, almost all of the tumors (17 of 18) exhibited diffuse staining of stromal elements, including adipocytes, fibroblasts and infiltrating inflammatory cells; a representative view is shown in Figure 8a (arrowheads). When a blocking peptide was added prior to addition of the ATF3 primary antibody, staining of both stromal and epithelial components was abolished, confirming the specificity of IHC staining (Figure 8b). About 40% of these tumors also contained significant, well-defined regions (i.e., foci of malignant epithelial cells surrounded by stromal elements and containing at least 200 tumor cell nuclei in the section examined) that exhibited strong nuclear staining in at least 30% of the nuclei. An example is shown in Figure 8c. To identify the epithelial portion of the tumor, an adjacent section was stained for CK5 (Figure 8d), and the observed boundaries of the epithelial components are indicated by a red line in both panels. In many cases, the regions with high levels of staining appeared to be ductal carcinoma in situ (DCIS) (e.g., Figure 8c and 8e), located within the tumor. In all cases examined, the tumor regions that exhibited ATF3 nuclear staining were also positive for cytoplasmic CK5 (e.g. Figure 8d) and CK8 (data not shown). None of the tumors examined appeared to have strong nuclear ATF3 expression in all tumor cells. In addition, clear nuclear ATF3 expression was seen in hyperplastic ducts located around the periphery of several tumors (Figure 8f); both basal and supra-basal cells in these abnormal ducts were immunoreactive for ATF3.

None of the normal mammary tissues examined exhibited staining for ATF3 in the stromal elements, and the staining of epithelial nuclei was generally much lower than in the ATF3-positive tumors (Figure 8g–h). In some cases, diffuse cytoplasmic staining was seen in scattered acini. A small fraction of large, apparently normal, ducts in both normal mammoplasty tissues and in tumor-adjacent normal tissue exhibited strong nuclear ATF3 expression in the basal layers (not shown).

To confirm the overexpression of ATF3 protein in human mammary tumors, tissue microarrays were obtained and analyzed by the same method. Overall, a total of 127 cores representing mammary tumors and 13 normal mammary glands were analyzed. Strong nuclear expression of ATF3 in the mammary tumor cells was detected in 18% of the tumors overall; none of the normal samples in these arrays showed nuclear ATF3 expression.