Ski-related novel protein N (SnoN), a Negative Controller of Transforming Growth Factor-β Signaling, Is a Prognostic Marker in Estrogen Receptor-positive Breast Carcinomas¹,²

TGF-β⁴ is a negative regulator of proliferation of mammary gland epithelial cells. It regulates normal ductal and alveolar development of the mammary gland and is involved in postlactational involution (1, 2, 3, 4). High levels of TGF-β expression in vivo prevent mammary tumor formation (5), suggesting that it is an important physiological controller of the mammary gland. TGF-β signals are transduced by its type I and II receptors to the intracellular signaling mediators Smad2 and Smad3. After their phosphorylation and association of co-Smads, they migrate into the nucleus, where they modulate the transcription of a large number of genes (6, 7, 8).

SnoN (SKIL) and Ski are negative controllers of TGF-β signaling (9, 10, 11). Ski was originally discovered as an oncogene present in the avian Sloan-Kettering virus (12), followed by identification of Sno and its isoforms SnoN, SnoA, and SnoI (13, 14). SnoN and Ski are structurally and functionally highly homologous proteins. They are able to induce transformation in chicken embryonic fibroblasts as well as induce muscle differentiation in quail embryonic cells (15, 16, 17). Furthermore, snoN+/− or ski+/− heterozygous mice display increased susceptibility to tumorigenesis (18, 19), suggesting these proteins also have tumor suppressive function. The diverse effects of SnoN/Ski may relate to their abilities to modulate transcription. They bind directly to nuclear hormone receptor corepressor, silencing mediator of retinoid and thyroid receptors (SMRT) corepressor, and mSin3A to form a complex with the histone deacetylase, which allows them to act as transcriptional repressors (20). Thus, overexpression or alternatively absence of SnoN/Ski may disrupt multiple transcriptionally regulatory pathways in the cells.

SnoN functions like a “switch” in TGF-β signal transduction; in the absence of TGF-β, it interacts directly with Smad2/Smad3-Smad4 complexes and recruits nuclear hormone receptor corepressor/mSin3A/histone deacetylase complex to Smads, hence inhibiting their transactivation capability and repressing TGF-β signal transduction (9, 21). On TGF-β treatment, SnoN is rapidly degraded via the ubiquitin–proteasome pathway, mediated by anaphase-promoting complex or Smurf2 E3 ligases (22, 23, 24, 25). This leads to dissociation of SnoN from the Smads and allows TGF-β signal to pass through. However, longer TGF-β treatment increases SnoN through an induction of SnoN mRNA (10). This may exert a negative feedback to limit TGF-β effects. Importantly, overexpression of Ski/SnoN results in the loss of certain TGF-β inducible signals leading to growth arrest of the cells (9), suggesting a potential mechanism for SnoN/Ski-mediated oncogenesis. The regulation of SnoN thus plays an important role in controlling TGF-β activity corresponding to the environmental cues of the cells.

Cell lines isolated from breast tumors are generally highly resistant to TGF-β growth inhibition. In some cases, this is attributable to alterations in either TGF-β type I or II receptors, or in Smad2 or Smad4, but these represent a minority (26). This suggests that the normal function of TGF-β in breast epithelial cells, i.e., control of proliferation, is abrogated downstream of the signaling cascade. This is corroborated by a recent tissue microarray study of 456 breast tumors, which indicated that in the vast majority of the tumors (94%), phosphorylated SMAD2 was present in the tumor cells, indicating the presence of TGF-β activity (27). To address the question how TGF-β signaling is turned off in breast carcinoma, we undertook a tissue microarray study of breast carcinomas to evaluate the expression of SnoN and its potential prognostic value as an inhibitor of TGF-β signaling.

As reported previously (28, 29, 30), the FinProg Breast Cancer Database comprises 2842 breast carcinoma patients identified using the files of the nationwide Finnish Cancer Registry. The database includes information on >50 clinical characteristics extracted from the hospital records, including a series of established prognostic markers. In this study, patients with in situ carcinomas, distant metastasis at the time of diagnosis, synchronous or metachronous bilateral breast cancer, malignancy other than breast cancer in history except for basal cell carcinoma or cervical in situ carcinoma, and patients who did not undergo breast surgery were excluded. After these exclusions, 2032 patients were eligible in the final data set, and in 1478 of these patients, tissue array core biopsies were available. The preparation of tissue microarray is described in Ristimäki et al. (29). Paraffin-embedded tumor samples were extracted and represented in the array slides as core biopsies of 0.6 mm in diameter. Sections of 5 μm were cut and processed for staining. After excluding cores detached during the staining procedure or containing no identifiable tumor cells, 1122 primary breast cancer samples were used for further analysis. The median follow-up time for the unrelapsed patients was 9.5 years. Specimens from reduction mammoplasties (n = 10) were obtained from the Department of Pathology, University of Helsinki.

Normal tissue microarray (catalogue no. 3170) was purchased from AMBION, Inc. (Austin, TX), which contains 85 samples from 21 different tissue types, including adrenal, thyroid, tonsil, lymph node, spleen, liver, pancreas, stomach, colon, bladder, kidney, prostate, testis, lung, heart, brain, ovary, breast, myometrium, muscle, and parotid.

Polyclonal goat-anti human SnoN (K-20) antibody and its blocking peptide were purchased from Santa Cruz Biotechnology (Santa Cruz, CA). Biotinylated rabbit-anti goat IgG (H+L) and Vectastain avidin-biotin peroxidase complex kit were purchased from Vector Laboratories (Burlingame, CA). 3-amino-9-ethylcarbazole was obtained from Lab Vision Co. (Fremont, CA). Mayer’s hemalum was from Merck Eurolab (Espoo, Finland).

Samples were deparaffinized in xylene and rehydrated in a series of graded alcohols, and the antigen was retrieved in 0.01 m sodium-citrate buffer (pH 6.0) using a microwave oven. The sections were then treated with 0.6% hydrogen peroxide in methanol for 30 min to exhaust endogenous peroxidase activity. After 1-h preincubation in a blocking buffer (0.01 m Tris-HCl, 0.1 m MgCl₂, 0.5% Tween 20, 1% BSA, and 5% normal rabbit serum) to prevent unspecific staining, the samples were incubated with goat-anti human SnoN polyclonal antibody (2 μg/ml) in the blocking buffer at 4°C overnight and then 2 h at room temperature. The sections were thereafter treated with biotinylated rabbit-anti goat immunoglobulin (7.5 μg/ml) followed by incubations with avidin-biotin peroxidase complex solution and 3-amino-9-ethylcarbazole solution for 30 min at room temperature, respectively. The counterstaining was carried out using Mayer’s hemalum.

To verify the specificity of the antibody, tissue array slides and sections of breast tumors were stained in the presence of a blocking peptide (2 μg/ml). The blocking peptide abolished both cytoplasmic and nuclear SnoN signals, demonstrating the specificity of the antibody (Supplemental Fig. 1).

SnoN immunohistochemical staining was examined under a Leica DMBL microscope, and the staining intensity and subcellular localization were evaluated in a blinded manner by two investigators (F. Z. and Ma. L.) independently using a preagreed scoring standard. Cytoplasmic and nuclear staining were scored separately using the following scoring criteria: (a) 0–1, negative or low staining intensity in >50% of tumor cells or moderate to high in <50% of the cells (hereafter referred to as low); and (b) 2–3, moderate to high staining intensity in >50% of tumor cells (hereafter referred to as high).

The associations between SnoN expression and other clinicopathological markers were analyzed using the χ² test. DDFS was calculated from the date of the diagnosis to the occurrence of metastases outside the locoregional area or death from breast cancer, whichever came first (28, 30). Life tables were calculated according to the Kaplan-Meier method. Survival of the groups was compared with the Log-rank test. Multivariate survival analyses were performed with the Cox proportional hazards model, entering the following covariates: (a) SnoN expression (0–1 versus 2–3, in cytoplasm or nucleus); (b) age (≤50 year versus >50 year); (c) tumor size in centimeters (continuous); (d) the number of metastatic axillary lymph nodes (continuous); (e) histological grade (I, II, and III); (f) ER (negative versus positive); (g) PgR (negative versus positive); (h) p53 expression (negative-low versus high); (i) Ki-67 expression (low-moderate versus high); (j) Cox-2 expression (0–1 versus 2–3); and (k) HER-2 gene amplification (negative versus positive). Cox regression was done using a backward stepwise selection of variables, and a P of 0.05 was adopted as the limit for inclusion of a covariate. All statistical tests were two sided.

SnoN mRNA is present in different tissues (14, 31), but there are few examples of its expression levels or subcellular localization. Nuclear staining of SnoN is detected in mink lung epithelial cells and spleen T cells of Sno +/− mice (18, 32). To first analyze the expression of SnoN in normal mammary tissues, SnoN immunohistochemical staining was carried out in samples of reduction mammoplasties (n = 10). Cytoplasmic as well as nuclear SnoN were present at moderate levels in epithelial cells lining the terminal ductuli units (Fig. 1, A and B), whereas epithelial cells lining larger mammary ducts showed moderate to high nuclear SnoN (Fig. 1, C and D). The expression of SnoN was also examined in other human tissues in a tissue microarray containing multiple samples from 21 different normal tissues (for details, see “Materials and Methods”). SnoN was detected in the epithelial cells of the thyroid, kidney, pancreas, and breast. In pancreas, the staining was cytoplasmic, whereas in thyroid, kidney, and breast, SnoN was present in both localizations (Supplemental Fig. 2). We thus conclude that only a subset of human epithelial tissues expresses SnoN and that normal mammary epithelial cells show cell- and structure-dependent expression of SnoN.

SnoN immunohistochemical staining was evaluated in 1122 primary breast carcinomas included in the FinProg Database. Scoring of the immunohistochemical stainings indicated 31.2% of the breast tumors were negative or weakly positive for nuclear SnoN (scored as 0–1), whereas 68.8% of the tumors showed high nuclear SnoN expression (scored as 2–3). No or low cytoplasmic SnoN expression was seen in 29.5% cases, whereas the cytoplasm was moderately or strongly stained in 70.5%. In addition, SnoN expression was absent or low in both nucleus and cytoplasm in 10.2% of the tumors (Fig. 2,A), whereas 49.5% of the tumors had moderate to high staining of SnoN in both localizations (Fig. 2,D). We observed that 21% of the tumors had high cytoplasmic SnoN staining without nuclear SnoN expression (Fig. 2,C), whereas 19.3% showed no or low cytoplasmic expression with moderate to high nuclear SnoN expression (Fig. 2 B). Thus, the tumors showed variations both in the SnoN expression levels and localization patterns.

SnoN was found to be expressed in both nucleus and cytoplasm in human breast carcinomas and show varying degrees of intensity (Fig. 2). To evaluate the significance of the expression pattern, we analyzed the association of SnoN expression with other clinical and pathological markers of breast cancer, provided in the FinProg Database. SnoN expression pattern was found to associate with certain subtypes of breast cancer (Table 1). Nuclear expression of SnoN was significantly more frequent in lobular tumors (P = 0.0232), grade I and II tumors (P = 0.0004), ER-positive tumors (P = 0.0014), and tumors with low levels of p53 (P = 0.003), Ki-67 (P < 0.0001), and Cox-2 (P < 0.0001). In contrast, cytoplasmic expression of SnoN was significantly more frequent in ductal tumors (P = 0.0019), grade III tumors (P = 0.0101), ER- and PgR-negative tumors (P = 0.0332 and 0.0168, respectively), and high levels of p53 (P < 0.0001), Ki-67 (P = 0.0009), and Cox-2 (P < 0.0001) and in tumors with HER-2 amplification (P = 0.0021).

SnoN expression was not significantly associated with patients’ age (year 50 was used as a cutoff, P = 0.2364 and 0.2436 for cytoplasmic and nuclear SnoN) nor with tumor size (2 cm as a cutoff, P = 0.9459 and 0.2877 for cytoplasmic and nuclear SnoN) and did not correlate with the presence of lymph node metastases (P = 0.9381 and 0.4993, respectively, for cytoplasmic and nuclear SnoN). There was no significant association between nuclear and cytoplasmic expression levels of SnoN (P = 0.1289).

To estimate the prognostic value of SnoN in breast carcinomas, we analyzed the DDFS in all patients with different expression patterns of SnoN. Univariate analyses in all tumors showed that low expression of SnoN (negative-low in both cytoplasm and nucleus) indicated a better survival (RR 1.61; 95% CI, 1.02–2.54; P = 0.0381; Table 2; Fig. 3 A). In all tumors, low nuclear SnoN showed no statistically significant difference on survival, whereas the statistic significance of low cytoplasmic SnoN approached borderline (P = 0.0581, data not shown).

Cox multivariate analysis showed that tumor size (RR = 1.22; 95% CI, 1.14–1.3; P < 0.0001), number of axillary node metastases (RR = 1.15; 95% CI, 1.12–1.18; P < 0.0001), and histological grade (grade 2 versus grade 1 RR = 2.59; 95% CI, 1.79–3.75; P < 0.0001 and grade 3 versus grade 1 RR = 3.19; 95% CI, 2.16–4.7; P < 0.0001) were independent prognostic factors, whereas nuclear, cytoplasmic, or combined expression of SnoN, age, histological type, ER or PgR status, p53 expression, Ki-67 expression, Cox-2 expression, or HER-2 gene amplification did not add independent prognostic information.

Subgroup analysis showed that low SnoN expression in both cytoplasm and nucleus as compared with high levels either in nucleus or cytoplasm, or both, indicated a statistically significant better outcome in several patient subgroups, including patients with tumors positive for ER (RR = 3.27; 95% CI, 1.44–7.41; P = 0.0027) or PgR (RR = 2.68; 95% CI, 1.09–6.58; P = 0.0255), tumors of the ductal type (RR = 1.69; 95% CI, 1.03–2.77; P = 0.0349), of grades I–II (RR = 2.32; 95% CI, 1.08–4.97; P = 0.0265), or with a low level of Cox-2 (RR = 2.14; 95% CI, 1.05–4.37; P = 0.0331; Table 2).

As SnoN had a highly significant prognostic value in ER-positive tumors (P = 0.0027; Fig. 3,B) but not in ER-negative tumors (RR = 0.9; 95% CI, 0.52–1.6; P = 0.6985; Fig. 3,C; Table 2), we further analyzed the prognostic effect of low expression level of SnoN either in cytoplasm or nucleus in the ER-positive patients. A low level of cytoplasmic SnoN associated significantly with longer DDFS (RR = 1.66; 95% CI, 1.14–2.4; P = 0.0076), whereas a low level of nuclear SnoN did not (RR = 1.28; 95% CI, 0.88–1.85; P = 0.1965), as compared with high levels of SnoN in cytoplasm or nucleus, respectively. When the DDFS was graphed according to different combinations of SnoN expression levels and localizations (low in both versus high in any location) in the ER-positive tumors, it showed that even in the presence of a low level of SnoN in the cytoplasm, its combination with high nuclear SnoN associated with worse DDFS (Fig. 3,D). This indicates that the reduction of SnoN in both localizations is important for better outcome (P = 0.0125; Fig. 3 D).

TGF-β potently inhibits the growth of mammary epithelial cells and regulates mammary development in vivo (3). Mammary-specific overexpression of TGF-β in transgenic mice can induce mammary hypoplasia and inhibit tumorigenesis (2, 5, 33). Although normal human mammary epithelial cells are exquisitely sensitive to TGF-β (34), breast cancer cells require higher levels of TGF-β to produce an antiproliferative response, and most of the cell lines show a complete loss of this effect (35). Breast cancers usually have intact TGF-β signal transducers (27), which is in contrast to many other tumors, like gastrointestinal, lung, and ovarian cancers and lymphomas, which carry deletions or inactivating mutations of the TGF-β signaling components (26, 36). One particular alteration in the TGF-β resistance of a breast cancer cell line is the lack of down-regulation of c-myc mRNA, whereas several other genes maintain TGF-β-targeted regulation (37). Whether this is the sole mechanism how breast tumor cell lines evade TGF-β arrest is unknown. It is interesting to note, however, that overexpression of Ski specifically abrogates TGF-β-mediated down-regulation of c-myc, whereas other transcriptional events are intact (9). Here, we provide evidence that a TGF-β signaling inhibitor, SnoN, may contribute to the loss of TGF-β antimitogenic action in breast carcinoma.

To address the involvement of SnoN in mediation of TGF-β-resistant phenotype in breast cancer, we examined SnoN protein levels in 1122 human breast cancers by immunohistochemistry and analyzed the association of SnoN expression with other clinical and pathological features, as well as its prognostic significance. We found highly variable expression patterns of SnoN in respect to its level and cellular localization. Although SnoN is alluded as a nuclear protein according to its activity as inhibitor of Smads, we observed that 70% (791 cases) of the analyzed breast carcinomas have cytoplasmic SnoN staining. Similar cytoplasmic staining was observed in epithelial cells residing on the luminal side of the ductules in normal breast. The cytoplasmic location was significantly more frequent in ductal cancers and correlated with worse prognosis indicators, such as grade III; lack of estrogen and PgRs; high levels of p53, Ki-67, and Cox-2; and HER-2 amplifications. On the other hand, nuclear SnoN was observed specifically in epithelial cells lining the large ducts in normal breast, whereas in tumors, the nuclear location of SnoN was more frequently associated with lobular tumors and tumors of lower grade, expression of hormone receptors, and low levels of p53, Ki-67, and Cox-2.

The expression patterns of SnoN observed in the tumors may reflect different cell type-specific origins of the tumors. Alternatively, its levels and subcellular localization may be affected by other autocrine or paracrine factors. Although breast tumors are believed to originate from epithelial cells of the terminal duct units, alterations in differentiation programs of the breast cells are likely to contribute to the phenotype of the tumors (38). Recent genomic analyses of gene expression profiles of breast tumors have clearly indicated that the breast tumors can be classified to separate entities, including tumors of basal and luminal cell origins (39, 40). The subcellular localizations of SnoN observed in normal breast epithelial cells corroborate those observed in tumors to some extent. SnoN expression in terminal ducts was mostly cytoplasmic, and this correlated with the higher fraction of cytoplasmic SnoN in ductal tumors. However, distinct nuclear staining of SnoN in epithelial cells lining the large ducts correlated with the phenotype of the lobular tumors. The highly distinct expression patterns of SnoN in these cell types may reflect different signals determining its location and subsequently affect its function. Reed et al. (41) reported that the subcellular localization of Ski, the analogue of SnoN, changed from being nuclear in preinvasive melanomas (melanoma in situ) to nuclear and cytoplasmic in primary invasive and metastatic melanomas in vivo. Furthermore, their in vitro study suggested that Ski/Smad association in cytoplasm may prevent the nuclear localization of the Smads in response to TGF-β, revealing another pathway on how Ski/SnoN can block TGF-β signal transduction. This is corroborated by the recent structural analysis of Ski interaction with Smad4, suggesting that Ski/SnoN blocks the interaction between Smad4 and the effector-Smads 2 and 3 (42). Therefore, cytoplasmic SnoN could block the association between Smad2/3 and Smad4, inhibiting the entry of the Smads into the nucleus, whereas nuclear SnoN can directly interfere with the Smad-transcriptional complex. It cannot be excluded that cytoplasmic SnoN has also other regulatory functions besides governing the Smad pathway.

A low level of SnoN expression as compared with high in either nucleus or cytoplasm, or both, was a significant prognostic indicator in all tumors analyzed (P = 0.0381). Furthermore, in ER-positive tumors, low SnoN level was a significant good prognostic indicator (P = 0.0027). Similarly, it had prognostic value in ductal and grade I–II tumors, tumors positive for PgR, and tumors with low levels of Cox-2. In contrast, SnoN had no prognostic significance within the subgroups of patients with lymph node-negative versus -positive disease. The finding that SnoN is an independent prognostic factor in ER-positive tumors is intriguing. That TGF-β and estrogen pathways cross-talk is suggested by several studies showing that in breast cancer cells, tamoxifen induces TGF-β mRNA and protein and increases TGF-β secretion (43, 44, 45, 46). Tamoxifen treatment may therefore inhibit proliferation of ER-positive tumors through TGF-β. Therefore, low levels of SnoN would maintain a positive feedback loop of TGF-β signaling and lead to inhibition of cell cycling. Our observation that reduction in SnoN correlates with better prognosis in ER-positive patients gives another indication that TGF-β and estrogen pathways may cross-talk.

We report here the first in vivo analysis of SnoN expression in human breast carcinomas. SnoN was found to be expressed in breast tumor cells in distinct but variable subcellular localizations. The localizations reflected those observed in normal breast epithelial cells but showed a greater degree of variation in expression levels and subcellular sites. Additional work in experimental models will be required to determine the significance of the variations of SnoN levels on TGF-β signaling and especially on the proliferative responses. A confounding element in this assessment is the highly frequent mutational events in the cell cycle machinery proteins observed in breast carcinomas (47), which may prohibit an adequate growth arrest despite functional signaling. In this study, a low level of SnoN was found to be a potential prognostic indicator, especially in patients with ER-positive tumors, suggesting that lack of a negative regulator of TGF-β signaling is beneficial for breast cancer outcome. On the basis of the data, we therefore propose that SnoN is a novel prognostic marker for ER-positive breast cancers.

We thank Maija Hälikkä, Marilotta Turunen, Harri Sihto, and Tuija Hallikainen for excellent technical assistance and advice.