Introduction
Colon cancer remains the second deadliest cancer in the United States with an estimated 136,830 new cases and 50,310 deaths in 2014 [
1]. The overall incidence, as well as cancer-related mortality, have both decreased over the past 10 years, which has been attributed to enhanced screening and early detection. However, once colon cancer has metastasized, the five year survival remains poor [
1]. Further, and disturbingly, the number of young patients with metastatic disease is increasing [
1]. Understanding the switch to metastatic behavior and developing therapeutic strategies to target metastatic signaling are key unmet clinical challenges. This understanding will lead to the generation of functional biomarkers to better predict patient risk and potential treatment response to individual pathway inhibition.
Recent efforts in cancer genome comprehensive sequencing have confirmed key genes whose mutations can drive tumorigenesis [
2] and have solidified components of the Transforming Growth Factor (TGF) β superfamily as drivers of pathogenesis in colon cancer. These include inactivating mutations in the TGFβII receptor (TGFBR2), the activin receptor 2A (ACVR2A) and downstream signaling molecule SMAD4 [
2]. TGFβ and activin are involved in the regulation of cell proliferation, differentiation, migration and apoptosis [
3‐
5]. Activin and TGFβ utilize a specific type I/type II receptor complex for signal transduction [
6‐
8]. In the canonical pathway, ligand binding leads to activation of SMAD2/3/4 proteins, translocation to the nucleus and transcriptional regulation of target genes to affect growth suppression and p21 upregulation. The non-canonical signaling pathway is SMAD4-independent and may engage other signaling pathways [
4,
5]. Activin and TGFβ both have dual and opposing roles in colon carcinogenesis as they may promote growth suppression, as well as migration and metastasis in more advanced colon cancer, also known as the molecular switch [
9‐
12]. TGFβ itself has opposing functions: in early stage colon cancer, the TGFβ super family is growth suppressive, while in advanced disease, high TGFβ serum and stroma levels are associated with poor prognosis [
13,
14]. Similarly, high levels of plasma activin in pancreatic cancer patients are significantly associated with decreased overall survival and increased distant metastases [
15]. Therefore, it is critical to understand the switch from growth suppression to proliferation for TGFβ and activin signaling. In order to identify targets which may directly affect metastatic behavior, we need to understand how the respective pathways intersect with other signaling cascades and how specifically, activin and TGFβ differ in their effects.
We have previously shown that in colon cancer cells, the cyclin dependent kinase (CDK) inhibitor p21 is a downstream target of both activin and TGFβ [
9]. In primary colon cancer tumors, we observed that nuclear p21 localization correlated with TGFBR2 expression while loss of nuclear p21 is associated with ACVR2A expression, respectively [
9]. To understand the contribution of each signaling pathway to the net signaling measured in tumor samples, we further dissected activin and TGFβ signaling in colon cancer cells. Despite shared SMAD2/3/4 signaling of activin and TGFβ, we observed opposing downstream effects on p21; namely TGFβ induces SMAD-dependent upregulation of p21, whereas activin leads to SMAD-independent downregulation of p21 via increased proteasomal degradation [
9]. Additionally, since SMAD4 by itself is frequently mutated and inactivated in colon cancer [
16], it is important to understand the SMAD-independent signaling of activin and TGFβ and the downstream impact on metastatic processes.
While TGFβ induced EMT in cancer has been studied in detail, there are only a few conflicting reports on activin and EMT [
17,
18]. TGFβ has been shown to induce an invasive phenotype and EMT [
17,
19,
20], but the precise mechanism is not well described. Activin’s effects on EMT have not been studied in detail to date. Reports in the literature indicate that increased expression of pAKT is associated with the loss of p21 expression in adenoid cystic carcinomas [
21] while ligand activation of TGFBR2 can activate the MEK/ERK pathway [
5]. In colon cancer, the influence of mitogenic signaling on the regulation of EMT and the effect on a metastatic phenotype in colon cancer through TGFβ is not well known. However, previous reports have implicated PI3K signaling in EMT following TNF ligand stimulation in colon cells [
22]. Phosphatidylinositol-3’-kinase (PI3K) signaling is involved in the regulation of several key cellular processes such as cell growth, survival, motility and proliferation which are involved in tumorigenesis [
23]. The PI3K pathway plays a prominent role in many cancers; in colon cancer specifically, upstream activation or gain of function mutations are common [
24]. Akt, one of the downstream effectors of PI3K signaling, regulates apoptosis and cell cycle progression. Activation of the PI3K/Akt pathway can arise through various mechanisms and is associated with a poor prognosis in a number of cancers [
24]. MEK/ERK mitogenic signaling is also commonly activated in tumors and affects regulation of the cell cycle as well as senescence [
25]. ERK may directly influence p21 localization and stability [
25], but the effects of TGFβ/activin–mediated mitogenic signaling have not been studied.
In order to delineate the respective and unique contributions of activin and TGFβ in anticipation of individual pathway inhibition in metastatic signaling, we first interrogated the association of PI3K/Akt and MEK/ERK signaling with the status of activin and TGFβ receptor expression in colon cancer patients. Then, we further dissected downstream use of mitogenic signaling pathways in activin and TGFβ-specific signaling as well as effects on p21 regulation, EMT, and migration in colon cancer. Better understanding of the distinct effects of activin and TGFβ signaling in metastatic disease is crucial in anticipation of specific pathway inhibition via small molecules. Moreover, we suggest nuclear p21 as a potential therapeutic biomarker that may accurately distinguish whether activin or TGFβ signaling is dominant in a given colon cancer patient. This study adds to the growing complexity of preclinical information necessary to plan much needed trials in advanced colon cancer disease using pathology-guided therapeutics.
Discussion
TGFBR2 and ACVR2A are commonly mutated in microsatellite unstable (MSI) colon cancers, which is the second most common genomic subtype [
30]. The characteristic defects in mismatch repair result in mutations in repetitive sequences termed microsatellites [
31‐
34]. Both TGFBR2 and ACVR2A harbor coding microsatellites and mutations in MSI colon cancers which are associated with loss of protein expression [
35]. In addition, mutations in non-MSI cancers may occur in their common downstream signaling component SMAD4 [
3]. While all this underscores that the TGFβ superfamily undoubtedly is important in colon cancer, the respective contribution of each pathway is still poorly understood. Detailing such contributions and functional net effects in colon cancer is crucial when envisioning treatment with the now emerging TGFβ pathway inhibitors.
It is conceivable that the redundancy in downstream SMAD signaling as exhibited by activin and TGFβ may lead to pathway rescue [
36]; however, given the complex cross talk between TGFβ family members and other signaling pathways, this is likely an oversimplification. We and others have found SMAD-independent signaling is associated with distinct functional effects that occur parallel to SMAD signaling in
SMAD4 wild type colon cancer cells [
9]. In addition, when both signaling cascades are intact, there is downstream divergence in the regulation of p21 [
9], which may remain dominant in some cancers and underscores the complexity of pathway interactions.
It is clear that TGFβ family signaling is modified dependent on cellular context [
3]. Parallel signaling of TGFβ and activin likely both play a role in the development of advanced colon cancer, and net signaling will be affected by selective pathway abrogation as a result of the mutational heterogeneity of each tumor. Understanding this net signaling effect is crucial to the development of therapeutic approaches which target the pro-migratory function in advanced colon cancer while protecting the anti-proliferative functions of TGFβ in early colorectal cancer development. In this report, we have dissected activin and TGFβ signaling divergence and their net functional effects with the aim of identifying a potential therapeutic marker.
Colon cancer treatment is entering the realm of precision medicine. Metastatic colon cancers are routinely tested for mutant
KRAS to determine suitability for biologic adjuvant therapy with the anti-epidermal growth factor receptor (EGFR) antibody, cetuximab [
37], as surprisingly, patients with mutant
KRAS fared worse with anti-EGFR therapy than without treatment [
37]. Small molecule inhibitors for both TGFBR1/2 and ACVR1B are available and conceivably could be used as adjuvant therapies in advanced colon cancer, but biomarkers to determine treatment suitability and specifically to avoid augmentation of oncogenic signaling are lacking. This need is underscored by a recent report that in pancreatic cancer inactivation of PI3K/mTOR signaling led to compensatory increase in mitogenic MEK/ERK signaling [
38]. Here, we have dissected signaling overlap and divergence of activin and TGFβ signaling as well as respective mitogenic signaling and we now suggest nuclear p21 expression as a potential read out for active upstream growth suppressive TGFβ/SMAD signaling. Further, we caution that inhibition of TGFβ in such cancers may lead to increased growth.
Mitogenic signaling pathways are frequently activated in cancers and the subject of intense investigation for therapeutic intervention. The MAPK pathway is commonly activated in tumors by direct mutation or overexpression of upstream molecules such as BCR-ABL and EGFR. TGFβ can activate MAPK and PI3K dependent on cell type and culture condition [
3,
39]. We now observe that in colon cancer cells MAPK is activated by TGFβ by a SMAD-dependent mechanism leading to regulation of p21 expression.
In contrast, we also report that PI3K is an active signaling component of activin/SMAD-independent mediated downregulation of p21 and associated with a metastatic phenotype. Consistent with an activin/PI3K/pAkt–induced p21 downregulation, we now show that in colon tumors of
ACVR2A KO mice pAkt activation is diminished and p21 levels are restored. Activation of PI3K is a potential mechanism of resistance to various cancer therapies [
24,
26,
40] and when constitutively activated, therapeutic targeting of PI3K may be beneficial. In breast cancer cells, for instance, Akt may decrease p21 expression [
41] enhancing to oncogenic behavior [
42], which is consistent with our data regarding activin/PI3K/p21 downregulation.
p21 is a member of the Cip and Kip family of Cdk inhibitors, which includes p21, p27, and p57 [
43]. These inhibit the kinase activity of broad, but not identical, classes of Cdk-cyclin complexes through their N-terminal homologous sequences. p21 arrests cell cycle progression primarily through the inhibition of Cdk2 activity, though it can also mediate p53-dependent G1 growth arrest. Earlier studies support the view that p21 suppresses tumors by promoting cell cycle arrest in response to various stimuli [
44,
45]. While p21 regulation is often compromised in human cancers, its continuous expression, depending on the cellular context, suggests that it can act as either a tumor suppressor or an oncogene [
43]. Deletion of p21 enhanced the rate of Ras- or c-Myc-induced tumorigenesis, and was associated with gene expression profiles and immunohistochemical features of EMT [
29], consistent with our finding of enhanced migration with activin-induced p21 downregulation. Similarly, the loss of p21 enhanced the baseline total migration as well as activin-induced migration in SMAD4 intact cells, [
9]. TGFβ induced p21 upregulation is decreasing over time and associated EMT.
As EMT is a central process in normal development, reactivation in cancer is regarded as dedifferentiation. It is typically characterized by the loss of cell-cell adhesion and apical-basal polarity, and may be induced by many different signaling pathways including TGFβ [
20]. EMT manifests with repression of E-Cadherin often via Snail genes, and the development of a fibroblast-like motile phenotype [
28]. Both SMAD-dependent and SMAD-independent signaling originating from TGFβ receptors have been implicated [
3]. Context-dependent enhancement of EMT by additional mechanisms can include PI3K activation [
46], as well as PDGF, EGF, and VEGF [
20]. The molecular switch of TGFβ from tumor suppressive to oncogenic is also likely context-dependent, but remains poorly understood. We have previously reported a switch similar to that of TGFβ for activin signaling [
9] and now show data implicating activin in EMT as well. TGFβ-induced EMT has been studied in detail but not much is known about the impact of activin signaling. There are some data that suggest that activin does not lead to EMT [
18] or has a low impact on EMT [
17], but experimental approaches were varied. Ansieau et al. report that Twist inhibits p21 leading to EMT and inhibition of oncogene-induced senescence [
47] and our data suggests that TGFβ induced upregulation of p21 is decreasing over time in conjunction with increased EMT. Data from Barrallo-Gimeno et al. suggest long term exposure to TGFβ favors EMT over growth suppression [
48], corresponding to our observations. Similarly, we show that while TGFβ increases p21 expression acutely [
9], long term exposure (72 h and one week respectively) in colon cancer cells leads to downregulation by activin or loss of upregulation by TGFβ of p21 and subsequent induction of cell migration and EMT. We have previously reported that the downregulation of p21 leads to an increase of migration [
9]. Our current data suggests a correlation between the downregulation of p21 and an increase in EMT. Small molecule inhibitors directed at growth factor receptors (such as EGFR) may interfere with EMT, although independent of primary receptor expression [
20]. To forestall the development of autocrine loops, neutralizing antibodies against TGFβ ligand as well as combination therapy directed against EGFR and PI3K are being assessed [
20].
Material and methods
Colon cancer cell lines
SW480 (ATCC, Manassas, VA, USA) were maintained in DMEM, and FET cells (gift from Michael Brattain, University of Nebraska, Omaha, NE, USA) were maintained in DMEM/F12 50:50 (both Corning, Corning, NY, USA) supplemented with 10 % fetal bovine serum and penicillin (100 U/ml)/streptomycin (100 μg/ml) (Invitrogen, Carlsbad, CA, USA). Cells were grown at 37 °C in a humidified incubator with 5 % CO2. All cells were serum starved for 24 h prior to treatment to approximate cell cycle synchronization. Cells were validated by 9 STR (short tandem repeat) profiling using CellCheck 9 Plus and tested for mycoplasma (both IDEXX, Columbia, MO, USA).
Reagents and antibodies
Activin A was reconstituted in PBS; TGFβ1 in 4 mM HCl according to the manufacturer’s instruction (both R&D, Minneapolis, MN, USA). Final concentrations used were 25 ng/ml and 10 ng/ml, respectively, as previously described [
31,
49‐
51]. For inhibition of PI3K, we used LY294002 and for inhibition of MEK1/2, U0126 (both Cell Signaling Technology, Danvers, MA, USA). For immunoprecipitation and Western blotting, we used antibodies against ACVR1B (Santa Cruz Biotechnology, Santa Cruz, CA, USA), ACVR2A (customized by Yenzym, San Francisco, CA, USA), p85 (# 4292), TGFBR1 (# 3712) or TGFBR2 (# 3713, all Cell Signaling), p21 (# sc-65595, Santa Cruz), pan-Akt (# 8805, Abcam, Cambridge, MA, USA), GAPDH (# sc-47724, Santa Cruz), E-Cadherin (# 3195), vimentin (# 3390), pAkt Ser473 (# 4060) and pAkt Thr308 (# 13842, all Cell Signaling). For immunohistochemical analyses, we used p21 (# sc-817, Santa Cruz) ACVR2A (# ab10595), TGFBR2 (# ab78419), pAkt Ser473 (# ab81283, abcam), and pERK1/2 (# ab50011, all Abcam).
Western blotting
Cells were lysed using CHAPS lysis buffer (containing 20 mM Bicine pH 7.6 and 0.6 % Chaps) with added protease and phosphatase inhibitors. Western blots were performed as previously described [
9].
siRNA and transfection
siRNA for SMAD4 (Ambion and Santa Cruz) and Akt1/2 (Santa Cruz) were transfected at a final concentration of 10 nM via electroporation using the AMAXA Nucleofector (Lonza, Basel, Switzerland) in 6-well plates at a density of 1×106 per well according to the manufacturer’s instructions. Transfection efficiency was confirmed using the pmaxGFP™ Control Vector. Forty-eight hours post transfection, colon cancer cells were lysed for protein extraction.
Migration assay
Migration assays were performed as previously described [
31]. Briefly, transwell 12 well plates (8 μm pores, Corning, NY, USA) with fibronectin (Sigma, St. Louis, MO, USA) were seeded with 5 × 10
5 colon cancer cells per well. Cells were then allowed to migrate for 6 h, stained, and imaged. Images from 5 microscopic fields at the center of each well were counted.
Patient samples
110 colon cancer and adjacent normal tissue stains (fixed in formalin and embedded in paraffin) were obtained from Northwestern Memorial Hospital as de-identified, archived tissue samples under IRB approval. Additional File
4: Table S1 contains individual patient information. Gender, age, tumor stage and expression of p21, TGFBR2 and ACVR2 is listed for each patient analyzed.
Immunohistochemistry
Slides containing primary colon cancer and normal tissues were processed as previously described [
52] and stained for AVCR2, TGFBR2, p21, pERK, and pAkt using the Catalyzed Signal Amplification System (CSA) by DAKO (Carpinteria, CA, USA). ACVR2A, TGFBR2, pERK, and pAkt staining was grouped into negative (no or weak signal) and positive (moderate or strong signal) status. The percentage of p21 positive nuclei in each cancer sample was assessed. p21 staining was grouped into nuclear (>50 % nuclei positive) or loss of nuclear (<50 % of nuclei positive). Slides were scored blindly by two investigators.
Immunoprecipitation
1 mg of total protein lysate from various treatments were incubated with 1 μg of ACVR1B or ACVR2A antibody (Yenzym); ligand specificity was confirmed with TGFBR1 or TGFBR2 (Cell Signaling) overnight at 4 °C. Protein A beads (Invitrogen, Carlsbad, CA, USA) were added for 6 h. After denaturation with sample buffer, equivalent protein was fractionated on 4-20 % gradient gels (Biorad, Hercules, CA, USA), transferred to membranes and blotted with antibodies against ACVR1B antibody (Santa Cruz), and p85 (Cell Signaling).
ACVR2A KO in vivo mouse tumor model
ACVR2A KO and ACVR2A wild type (wt) mice were treated with 14 mg/kg Azoxymethane (AOM) intraperitoneal infusion. After 5 days, mice were treated with 3 cycles of 2.5 % dextran sulfate sodium (DSS). Every cycle contained 5 days of DSS and 15 days of water. After day 100, the mice were sacrificed. Tumor and normal tissues were collected and lysed in RIPA buffer (1 % NP-40, 0.1 % SDS, 50 mM Tris–HCl pH 7.4, 150 mM NaCl, 0.5 % Sodium Deoxycholate, 1 mM EDTA) prior to use in Western blot analysis.
Isoelectric focusing immunoassay
Detection of ERK1/2 phosphoisoforms after TGFβ and activin treatment was achieved using an automated capillary based isoelectric focusing immunoassay system (NanoPro1000 assay, ProteinSimple, Santa Clara, CA, USA). Protein isolation, detection, and quantification were done as per the manufacturer’s instructions. In brief, FET colon cancer cells were lysed in CHAPS buffer and 0.2 μg/ml protein was plated in a 384 well plate. For the primary antibody, we used pERK1/2 and ERK1/2 (Cell Signaling) diluted at 1:100. Luminescence and fluorescence images were collected using a charge-couple device (CCD) camera. Peak integration and isoelectric point (pI) marker calibration (for peak alignment) were performed using Compass, version 1.3.7, software (Protein Simple). The difference of isoform expression after TGFβ and activin treatment was compared to the vehicle control.
Statistical analysis
Differences between groups were determined using the Student’s t-test. Probability values less than 0.05 were considered significant. Biological replicates from 3–5 experiments represent the data shown. For associations of IHC staining patterns, we performed Fisher’s exact test calculations.
Competing interests
The authors declare that they have no competing interest.
Authors’ contributions
Conceived and designed the experiments: JB BJ. Performed the experiments: JB OO NA TC DP JS MS LE. Analyzed the data: JB PG BJ. Manuscript preparation: JB BJ. All authors read and approved the final manuscript.