Reduced Expression of SMAD4 Is Associated with Poor Survival in Colon Cancer Free

Colorectal cancer remains a major public health problem in the Western world with an estimated 136,830 new cases and 50,310 deaths occurring in 2014 in the United States alone (1). Despite the progress made in the management of metastatic colorectal cancer over the last few years, with the incorporation in combination chemotherapy of two monoclonal antibodies targeting the EGFR (2–8) and one targeting the VEGF (9–12), in the adjuvant setting, after introduction of oxaliplatin/fluoruracil (5-FU)/leucovorin(LV) era, innovative approaches are eagerly awaited (13, 14). One approach is the search for prognostic and predictive markers, which might serve in selecting patients who are likely to benefit from adjuvant chemotherapy. To date, the most important prognostic factor and treatment determinant remains disease stage (15). Identification of tissue biomarkers capable of improving outcome through better patient stratification and selection for specific treatment is gaining momentum, as reflected in studies on microsatellite instability (MSI), loss of heterozygosity (LOH) of 18q, copy-number aberrations (CNA) and the mutation status of KRAS, BRAF, and TP53 (16–19). In spite of increasingly detailed molecular mapping of colorectal cancer, only few markers have shown some promise in clinical practice in terms of capability to predict disease course.

The detailed unraveling of molecular abnormalities that characterize oncogenesis of colon cancer is of profound importance for understanding the biology and clinical behavior of this disease. As previously noted, one of the most important and frequently encountered events is allelic loss on chromosome 18q (20, 21), where SMAD4 (also called deleted in pancreatic carcinoma 4, DPC4) gene is located (22–24). The SMAD4 tumor-suppressor gene (TSG) codes for the common intracellular mediator of the TGFβ superfamily signaling pathway, one of the most commonly altered cellular signaling pathways in human cancers (25–27) and involved in the regulation of cell proliferation, differentiation, apoptosis, and cell migration (27–30).

In mouse models, selective loss of SMAD4-dependent signaling in T lymphocytes has been shown to induce the development of spontaneous epithelial cancers throughout the gastrointestinal tract, suggesting that SMAD4 mediates crosstalk between epithelial and stromal inflammatory/immune-effector cells (31). Furthermore, experimental data obtained in vitro and in vivo in mice provided substantial evidence that SMAD4 mutation/silencing is involved in gastrointestinal tumorigenesis (32–34).

Germline mutations of SMAD4 have been identified in a subset of patients with familial juvenile polyposis, an autosomal-dominant disease characterized by a predisposition to hamartomatous polyps and cancers of the gastrointestinal tract (35, 36). Furthermore, SMAD4 has been found deleted or mutated in approximately 5% of adenomas and 20% of sporadic colorectal cancer, consistent with its role in colorectal cancer progression (29, 37). In addition, retrospective trial data suggest that tumors expressing high SMAD4 levels show significantly better overall and disease-free survival (38, 39). However, detailed documentation of the prognostic role of SMAD4 expression and its association with other clinicopathologic and molecular parameters in large patient cohorts are still lacking. As in a previous article, we reported that SMAD4 loss was the most powerful predictor of RFS in a multivariable analysis (after N and T stage; ref. 40), therefore, here we explored in more detail the significance of SMAD4 immunohistochemical (IHC) expression using the large randomized phase III trial PETACC-3, in which the addition of irinotecan to 5-FU/LV was explored in the adjuvant setting on 3,278 patients with stages II and III colon cancer (17, 18, 41). The predictive role of SMAD4 expression and its association with tumor characteristics, including molecular parameters, were evaluated in terms of relapse-free survival (RFS) and overall survival (OS).

From 1,564 of the 3,278 patients with stages II and III colon cancer enrolled in PETACC-3 (Pan-European Trials in Alimentary Tract Cancers; ref. 35) tissue specimens were prospectively collected (17, 18). Trial details, including patient characteristics have been reported previously (35). The median follow-up of the patients was 69 months (IQR, 66–74 months). Endpoints were OS, defined as the time from random allocation until death and RFS, defined as the time from the date of random allocation to the first date of local, regional, or distant relapse, the occurrence of a second primary colon cancer or death. The trial obtained ethical approval by the EORTC (European Organization for Research and Treatment of Cancer) and informed consent was obtained from included patients for the translational studies.

Formalin-fixed, paraffin-embedded (FFPE) material was collected from all participating centers in the Institute of Pathology of the University Hospital in Lausanne (Switzerland), in the form of 20 to 25 five-μm-sections of one tissue block per patient. As a rule each tissue block contained tumor and normal mucosa and each case was reclassified and graded by one expert pathologist (R. Fiocca) according to WHO criteria (42). The worst grade was retained regardless of extent but for G3-4 the poorly differentiated component had to exceed 5%.

Colorectal adenomas (n = 34) with tubular, villous, or mixed morphology and low- and high-grade dysplasia were collected at the Institute of Pathology, University Hospital in Lausanne and 10 pairs of primary colon cancer with liver metastasis at the Department of Clinical Pathology, University Hospital of Geneva.

IHC for SMAD4 was performed in the laboratory of the Department of Pathology, DISC, University of Genova. In brief, deparaffinized sections were immersed in 3% H₂O₂ for 10 minutes to abolish endogenous peroxidase activity and washed in 0.05 M TBS (pH 7.3) for 15 minutes. Antigen retrieval was performed in citrate buffer (pH 6.0) in a microwave oven with one cycle at 900 W for 3 minutes followed by a second cycle at 360 W for 13 minutes. Sections were then incubated with anti-SMAD4 mouse monoclonal antibody (clone B-8; Santa Cruz Biotechnology; 200 μg/mL, diluted 1/1,000) for 1 hour at room temperature. Negative control consisted of omission of the primary antibody. A colon cancer with confirmed focal SMAD4 loss served as positive control, in addition to adjacent normal tissue (internal positive control). The slides were subsequently rinsed three times in TBS and incubated for a further 20 minutes with Multilink Biotinylated anti-Ig (Biogenex) diluted 1:20, followed by 3 minute incubation with 0.05% 3-3′-diaminobenzidine-tetrahydrochloride in 0.02% H₂O₂. Sections were finally counterstained with Mayer's hematoxylin, dehydrated and mounted. With this antibody SMAD4 staining is nuclear as well as cytoplasmic.

Assessment of SMAD4 staining was carried out independently by two expert pathologists (R. Fiocca and P. Ceppa). Any discordance was solved through consensus discussion. Normal mucosa, which shows without exception cytoplasmic expression, was used as positive control and internal standard; in few samples lacking non-neoplastic mucosa, SMAD4-reactive intra- and peri-tumor lymphocytes were alternatively used as internal control. Semiquantitative assessment of SMAD4 staining intensity was as follows: intense (strong) expression as in non-tumor mucosa; decreased (weak) expression less intense than normal mucosa; absence of staining (loss). In each tumor, the percentage of tumor cells showing strong, weak or loss of expression was separately scored. In an initial phase, cutoff point effect on survival was studied using 5% increments. As between any loss (5%) and any other higher cutoff point the differences in survival were negligeable (data not shown), we chose any loss (5% or more) as cutoff point for this study. In view of the reported association of SMAD4 loss with poor prognosis in prior publications (38, 39), we hypothesized that survival might be influenced (i) by the presence of cell clusters with loss of SMAD4 expression and (ii) by reduced SMAD4 expression level as reflected in weak staining intensity. Consequently, cases with 5% or more of neoplastic cells with no SMAD4 expression were classified as SMAD4 loss. Cases with more than 95% of cells with strong expression were considered strong expressors and all others weak expressors.

To assess to what extent SMAD4 expression by IHC reflects SMAD4 mRNA expression, we used previously reported expression data on the PETACC-3 trial (43). This was done on tissue samples collected from paraffin-embedded tissue blocks. RNA of sufficient quantity and quality was extracted from 895 samples, and gene expression was measured on the ALMAC Colorectal Cancer DSA platform (Craigavon)—a customized Affymetrix chip with 61,528 probe sets mapping to 15,920 unique Entrez Gene Ids.

Tumor status regarding KRAS and BRAF mutation, expression status of thymidilate synthetase (TS), and TP53, LOH of 18q, MSI status (MSI-high, MSI-H or microsatellite stable, MSS), and CNAs were published earlier (16–19).

Survival curves were determined using Kaplan–Meier methods and compared using the log-rank test. Frequencies were compared using Fisher's exact and Pearson's χ² tests. Continuous mRNA expression was compared by SMAD4 expression status using the Wilcoxon's rank sum test. HRs with 95% confidence intervals (CIs) were computed with uni- and multivariable proportional hazards models. The HRs are from univariable models unless specified otherwise. Multivariable models were adjusted for tumor stage and site, treatment group, as well as MSI, BRAF, and KRAS status. The proportional hazards assumption was tested using the scaled Schoenfeld residuals. Interactions were assessed by likelihood ratio tests. P values are two-sided, not adjusted for multiple testing and considered significant if <0.05. Analyses were performed using the free R software package (www.r-project.org) version 2.13.0 or later. As the analyses were exploratory in nature, correction for multiple testing was not done.

SMAD4 expression by IHC was informative in 1,381 of the 1,564 available cases. Cases were not informative for a variety of reasons, which include absence of immunoreactivity in adjacent normal mucosa (negative internal control), high background staining (false positive) and loss of tissue section and absence of tumor tissue in the available block. Staining patterns are illustrated in Fig. 1A–D. Normal mucosa stained intensely for SMAD4 in the normal colonic epithelium and in mesenchymal stromal cells, with both nuclear and cytoplasmic staining. Of the 1,381 cases, 293 (21%) showed SMAD4 loss, 73 of which were stage II and 220 stage III (P = 0.03; Table 1). Loss of SMAD4 expression presented a clonal pattern and seemed to involve whole glands or groups of glands, but never single dispersed tumor cells (Fig. 1D). SMAD4 loss was not significantly different between right and left colon, whereas a marginally significant difference in weak versus strong staining was found (Table 1). Lack of standardization of site definition precluded analysis of continuous differences in SMAD4 expression along the colon. No particular distribution of glands with loss of SMAD4 expression within tumor areas was observed, nor were these over-represented in the invasion front. Of the 34 adenomas, 2 showed complete loss (100%) of SMAD4 expression and 2 focal loss (>5% of cells) with a clonal pattern.

Expression of SMAD4 mRNA showed a gradual decrease from strong staining to weak staining to loss. The differences in mRNA level between strong/weak and strong/loss were highly significant (P < 0.001) whereas those between weak/loss were not significant (P = 0.13). These data indicate that individual discrepancies exist, which is not surprising, given the presence of contaminating SMAD4-positive mesenchymal cells, not removed by macrodissection (Supplementary Fig. S1).

Patients with a colon cancer showing SMAD4 loss (293/1,381 cases) had statistically significantly poorer survival in RFS and OS in univariable models (Fig. 2A and B), which was confirmed in multivariable models (Table 2).

For stage II patients the impact of SMAD4 loss (73/293 cases) was not statistically significant in RFS or in OS (Fig. 2C and D). However, for stage III patients, SMAD4 loss (220/293 cases) showed a statistically significant negative prognostic impact both in RFS and OS (Fig. 2E and F). The impact of SMAD4 loss was stronger for stage III N2 (84/220 cases) than stage III N1 (136/220 cases) patients for RFS as well as OS (Fig. 2G–J). These results were confirmed in multivariable models (Table 2).

In both FOLFIRI and 5-FU/FA treatment arms SMAD4 loss associated with poorer survival for RFS as well as OS with a nonsignificant tendency for a stronger effect in the FOLFIRI arm (p_interaction = 0.02 for RFS and p_interaction = 0.05 for OS; Fig. 3A and B; Supplementary Table S1).

The effect of SMAD4 loss on survival suggested that weak SMAD4 expression might also impact on survival. Of the 1088 cases without SMAD4 loss, 530 (49%) showed weak expression and 558 (51%) strong expression. Of the 530 weak expressors, 154 were stage II and 376 stage III patients. Of the 558 strong expressors, 190 were stage II and 368 stage III patients.

Overall, the difference between SMAD4 weak and strong expressors (Fig. 4A–D), significant for OS and RFS in univariable analysis, was maintained in multivariable analysis only for RFS in stage II patients (Table 2). In stage III, we found no difference between weak and strong SMAD4 expression level in RFS or in OS (Fig. 4E and F), which was confirmed by multivariable models (Table 2).

For liver metastasis, our working hypothesis was that if SMAD4 negative cell clones would determine metastatic propensity, liver metastasis of a colorectal cancer with any SMAD4 loss would mostly be SMAD4 negative. In contrast, we found liver metastases to follow mostly the phenotype of the primary tumor: the liver metastases of 4 primary tumors with strong homogeneous SMAD4 expression generated metastases with equivalent expression and rarely zone of weaker or absent expression while the 4 cases with loss of or weak SMAD4 expression gave rise to metastases with a similar pattern of SMAD4 expression.

Of our cohort, 799 patients had an MSS and 178 an MSI-H tumor. Loss of SMAD4 was found in 249 MSS and in only 11 MSI-H tumors. For patients with a tumor with MSS as well as SMAD4 loss, RFS as well as OS were significantly worse (Supplementary Fig. S2A and S2B). This result was confirmed in multivariable models (Table 2). The effect seemed even stronger in MSI-H tumors (Supplementary Fig. S2C and S2D), but with higher uncertainty due to small numbers which also precluded analysis in multivariable models (Table 2).

SMAD4 expression status was not associated with treatment group, grade, T-stage, BRAF mutational status or p53 expression status. Marginally significant associations were found between SMAD4 expression and N-stage, primary tumor location, KRAS mutational status and age. In contrast, SMAD4 loss showed a highly significant positive association with loss of 18q23 (P < 0.001). A similar association was found with loss of 4q22.3. In contrast, SMAD4 loss was associated with gains of 9q13 and 21q21.2. A negative association was found between SMAD4 loss and MSI-H (P < 0.001; Table 1).

We found loss of SMAD4 expression in colon cancer in cell clusters or tumor areas rather than in diffusely scattered cells, which we have interpreted as a clonal pattern. This occurs also in a small percentage of adenomas and (be it on a small sample) not more often in metastases than in primary colon cancer. We found SMAD4 loss to be strongly associated with poor survival in RFS and OS in stage III patients. Other authors have also reported that the incidence of SMAD4 loss increases from adenoma to carcinoma stage II to carcinoma stage III to liver metastasis, providing strong support for the notion that SMAD4 loss is involved in tumor progression (21, 29, 44). Also, weak SMAD4 expression in stage II patients was associated with poor survival, but only in RFS.

SMAD4 is a key element in the TGF-β signaling pathway as SMAD2 and SMAD3, phosphorylated upon type II TGF-β receptor activation, bind to SMAD4 and as a complex translocate to the nucleus and interact with transcription factors in a cell specific manner to regulate transcription of a number of TGF-β responsive genes (45–47). TGF-β signaling has been shown to regulate normal colonic epithelial differentiation (48) and to inhibit normal intestinal epithelial cell proliferation (49). In cancer development and progression, the TGFβ–SMAD signaling pathway has been found to function as a tumor suppressor in early and as a tumor enhancer in late tumorigenesis (47). Although the molecular mechanisms responsible for the differences in TGFβ response between normal and transformed intestinal epithelial cells are not yet clearly elucidated, we know that inactivation of SMAD genes by mutation or deletion is one of the involved mechanisms (49). Mutations in the MH1 and MH2 domains of the SMAD4 gene have been identified in both primary colon cancer and liver metastasis, accompanied by LOH (37). MH1 mutations result in a protein with a markedly reduced ability to bind a consensus SMAD-binding element and MH2 mutations result in a protein with reduced DNA binding capacity (50). Missense mutations on the MH1 domain of SMAD4 have been found to physically increase the intramolecular interaction with its MH2 domain, which inactivates SMAD4 function (51). Such observations suggest that different types of somatic mutations of SMAD4 gene may play a role in the SMAD4 protein expression level. To understand how the clonal pattern of SMAD4 loss arises, it will be necessary to study the gene in areas with SMAD4 loss and SMAD4 expression, which can be done by microdissection of (premalignant and malignant) colorectal neoplasms.

A number of additional observations are worth to be briefly mentioned. The predictive value of SMAD4 expression for RFS and OS tended to be stronger in FOLFIRI than in 5FU/FA-treated patients. It has been suggested (39) that patients with low SMAD4 expression might be good candidates for combined treatment with the addition of irinotecan or oxaliplatin to 5-FU/FA, whereas our results could be taken to suggest that response to irinotecan-based combination chemotherapy might be less in patients with a colon cancer with SMAD4 loss. However, this notion needs to be further explored. The SMAD4 loss effect was stronger for stage III N2 than for N1 patients. We hypothesize that this is a general reflection of disease progression.

Interesting is also the correlation between SMAD4 expression and MSI status. Strong SMAD4 expression has been noted in MSI colorectal cancer, which fits with a contributing role of both SMAD4 and TGFβIIR to tumor-suppressive effects of the TGF-β pathway (52, 53). Our data suggests that the MSS-SMAD4 loss group carries a poor prognosis, whereas there is also a trend for the SMAD4 loss effect to be worse in MSI-H patients. Interactions between MSI status and SMAD4 expression status merit to be studied in larger patient cohorts. A significant positive association was found between SMAD4 loss and the loss of 18q23 and 4q22.3, as well as gain of 9q13 and 21q21.2. Although the association found between SMAD4 loss and loss of 18q23 confirms the results of earlier studies (54), we did not include 18q LOH in our multivariable models, as this information was missing in more than 40% of cases and multiple imputations to circumvent this problem would not have improved reliability. An interesting question is the possibility of continuous changes in molecular marker changes along the colon, which have been found for MSI and CIMP (55). We found marginal differences in SMAD4 staining between left and right colon, but lack of standardization of site definition precluded detailed analysis of this question.

Our study has several limitations. First of all, although our cohort of patients with colon cancer is large, the number of adenomas and liver metastases studied was rather limited. Our data are, therefore, compatible with a notion of increasing SMAD4 loss along progression of colon cancer, but do not provide new solid evidence. Second, no correction for multiple testing was done. We could have included post hoc Bonferroni corrections but, as somewhat arbitrary choices would have had to be made, this would not have improved the clarity of the results. Third, our study was conducted with a retrospective approach, even though the tissue samples were prospectively collected. Further confirmation of our findings awaits truly prospective studies. Finally, even though we found SMAD4 immunohistochemical staining robust and scoring results reliable, an intrinsic problem of immunohistochemical assays is technical variability and intra- and interobserver variability of assessment of staining results. In this context, it might have been of interest to compare the performance as prognostic parameter of SMAD4 IHC with that of loss of 18q. However, the missing information on 18q status in a large proportion of our cases precluded this.

We conclude that SMAD4 expression in colon cancer varies in terms of intensity (strong vs. weak) and in pattern (clonal pattern of complete loss) for which the governing mechanisms remain to be elucidated. The incidence of SMAD4 loss increases from adenoma to carcinoma stage II to carcinoma stage III, which supports the notion that SMAD4 loss is involved in tumor progression. Loss of SMAD4 expression strongly correlates with survival in terms of OS and RFS in stage III patients, which is maintained in multivariate analysis. The same is true also for reduced expression but only in stage II patients. The robustness of the used methodology and the HRs justify validation of SMAD4 expression as a prognostic parameter anticipating its use in clinical decision making.

No potential conflicts of interest were disclosed.

Conception and design: Z. Saridaki, A. Roth, S. Tejpar, M. Delorenzi, F.T. Bosman, R. Fiocca

Development of methodology: P. Ceppa, T.A. McKee, S. Tejpar, R. Fiocca

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Ceppa, T.A. McKee, F.T. Bosman, R. Fiocca

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Yan, D. Klingbiel, Z. Saridaki, P. Ceppa, T.A. McKee, A. Roth, S. Tejpar, M. Delorenzi, F.T. Bosman

Writing, review, and/or revision of the manuscript: P. Yan, D. Klingbiel, Z. Saridaki, T.A. McKee, A. Roth, M. Delorenzi, F.T. Bosman, R. Fiocca

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Klingbiel, Z. Saridaki, M. Curto, M. Delorenzi

Study supervision: S. Tejpar, M. Delorenzi, F.T. Bosman, R. Fiocca

The authors thank the patients, support staff, and all investigators who are not coauthors. The authors also thank Pfizer for supporting and funding the study.

This study was supported and funded by Pfizer. Z. Saridaki was a recipient of a research fellowship from the Hellenic Society of Medical Oncology (Hesmo).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.