IGFBP2 modulates the chemoresistant phenotype in esophageal adenocarcinoma.

Esophageal adenocarcinoma (EAC) patients commonly present with advanced stage disease and demonstrate resistance to therapy, with response rates below 40%. Understanding the molecular mechanisms of resistance is crucial for improvement of clinical outcomes. IGFBP2 is a member of the IGFBP family of proteins that has been reported to modulate both IGF and integrin signaling and is a mediator of cell growth, invasion and resistance in other tumor types. In this study, high IGFBP2 expression was observed in a subset of primary EACs and was found to be significantly higher in patients with shorter disease-free intervals as well as in treatment-resistant EACs as compared to chemonaive EACs. Modulation of IGFBP2 expression in EAC cell lines promoted cell proliferation, migration and invasion, implicating a role in the metastatic potential of these cells. Additionally, knockdown of IGFBP2 sensitized EAC cells to cisplatin in a serum-dependent manner. Further in vitro exploration into this chemosensitization implicated both the AKT and ERK pathways. Silencing of IGFBP2 enhanced IGF1-induced immediate activation of AKT and reduced cisplatin-induced ERK activation. Addition of MEK1/2 (selumetinib or trametinib) or AKT (AKT Inhibitor VIII) inhibitors enhanced siIGFBP2-induced sensitization of EAC cells to cisplatin. These results suggest that targeted inhibition of IGFBP2 alone or together with either the MAPK or PI3K/AKT signaling pathway in IGFBP2-overexpressing EAC tumors may be an effective approach for sensitizing resistant EACs to standard neoadjuvant chemotherapy.


INTRODUCTION
The incidence of esophageal adenocarcinoma has been increasing steadily in industrialized countries over the past four decades and continues to rise [1]. Despite advances in endoscopic surveillance and multimodality treatment regimens, the prognosis for EAC patients remains poor, with an overall 5-year relative survival rate of less than 20% [2]. While understanding of the progression from Barrett's esophagus to EAC has improved, patients commonly present with advanced stage disease and demonstrate resistance to conventional therapy, with complete response rates to trimodality therapy (chemotherapy, radiation therapy and esophageal resection) below 40% [3][4][5]. Only complete responders (those patients who are restaged as T0N0 or T1N0) have improved survival as compared to incomplete or nonresponders. Understanding the molecular mechanisms for chemoresistance is crucial in order to identify strategies to address such resistance, and thus improve clinical outcome.
The insulin growth factor-1 (IGF1) and IGF1 receptor (IGF1R) signaling pathway, implicated in esophageal cancer growth and progression, has been shown to play pivotal roles in cell growth, differentiation, survival, transformation and metastasis [6][7][8][9]. Overexpression of IGF1R is prevalent in esophageal adenocarcinoma (EAC), as high as 75% [10], and has been associated with the progression of Barrett's to EAC [8]. Overexpression of IGF1R also appears to be an independent predictor of survival [10]. Serum IGF-1 level correlates with pathological stage, depth of invasion and prognosis in esophageal cancer patients [11]. Additionally, Doyle et al. (2011) reported a dose-dependent increase in proliferation in response to IGF1 in EAC cells and significantly higher serum IGF1 levels in EAC patients compared to those with Barrett's esophagus or controls [12]. Function-altering polymorphisms in IGF1R [13] as well as IGF1 microsatellite repeats and single strand nucleotide polymorphisms [14,15] have been shown to affect risk for developing Barrett's esophagus and/or EAC.
Insulin-like growth factor (IGF) binding protein 2 (IGFBP2) is a member of the IGFBP family of proteins that binds both IGF1 and IGF2 in circulation with higher affinity than ligand-receptor interactions and can inhibit or stimulate the growth-promoting effects of IGFs by altering their bioavailability, distribution, stability and/ or interactions with cell surface receptors [6,[16][17][18]. IGFBP2 also contains arginine-glycine-aspartic acid (RGD) and heparin-binding motifs that bind to integrins and extracellular matrix, respectively, to mediate its role in cell detachment, migration and invasion [17][18][19][20][21].
Because IGFBP2 has been shown to modulate both the IGF and integrin pathways and is a mediator of cell growth, invasion, and resistance in other tumor types [22-37, 39, 40], we evaluated the functional role of IGFBP2 in EAC progression and chemoresistance. We show that modulation of IGFBP2 expression affects proliferation, motility, and chemosensitization of EAC cells in a serumdependent manner. Silencing of IGFBP2 affects both AKT and ERK activity and addition of targeted pharmacologic inhibitors of these pathways enhances siIGFBP2-induced cisplatin (CDDP) chemosensitization. IGFBP2 is a potential mediator of chemoresistance in a subset of EACs and its modulation in overexpressing EAC tumors may be an effective approach to sensitizing resistant tumors to standard of care chemotherapy.

Differential expression of IGFBP2 in human esophageal tissues
IGFBP2 expression levels in a progression series of human esophageal tissues including Barrett's metaplasia, low-grade dysplasia, high-grade dysplasia and EAC were examined by Affymetrix HG-U133A oligonucleotide microarray ( Figure 1A). Although many EACs expressed very low levels of IGFBP2, a subset of primary tumors expressed higher levels of IGFBP2. Expression values for this cohort were confirmed by qRT-PCR (R 2 = 0.80) (Supplementary Figure S1A). We analyzed the association between IGFBP2 expression and stage, node status or differentiation. None of these clinical parameters were associated with IGFBP2 expression (p = 1.00, 1.00 or 0.57, respectively, using Fisher's exact test). Immunoblot analysis of protein extracts from 5 paired samples of esophageal adenocarcinoma and associated Barrett's metaplasia indicated increased expression in three of five tumor samples and decreased expression in one of five tumor samples relative to Barrett's metaplasia ( Figure 1B). Variable IGFBP2 protein expression was observed among the EAC tissue samples and correlated highly with IGFBP2 tissue mRNA expression, as noted by both oligonucleotide microarray and qRT-PCR. Immunohistochemistry of tissue microarrays confirmed variable levels of IGFBP2 expression in EACs, ranging from undetectable to high expression at the brush border of several patient sections ( Figure 1C). Expression was detected in approximately 23% of tumors but did not correlate with pathologic tumor stage.
Affymetrix HG-U133A oligonucleotide microarray analysis of a second cohort of twenty EACs, obtained from esophagectomy patients who subsequently received chemotherapy, identified IGFBP2 as one of the genes with lower expression among disease-free patients as compared to patients with recurrent disease (Figure 2A). Expression levels for this cohort were confirmed by qRT-PCR (R 2 = 0.80) (Supplementary Figure S1B). In this cohort, patients with higher IGFBP2 expression had worse overall survival, even when controlled for cancer stage (p = 0.033, log rank test) ( Figure 2B). Multivariable Cox analyses were performed for stage, node positivity and differentiation. Stage and node positivity were correlated (concordance = 0.703, R 2 = 0.292), and both of these variables were significantly correlated to overall survival (p = 0.014 and 0.0317, respectively). There was no concordance between stage and IGFBP2 expression, and the IGFBP2 correlation to overall survival was stronger than stage (p = 0.0069). Differentiation was not significant due to lack of power (only 8 of 20 samples had known differentiation status).
To address the hypothesis that IGFBP2 is involved in chemoresistance, we performed real-time PCR on 200 EAC samples obtained from previously untreated patients ("chemonaïve") and 16 treatment-resistant EAC samples to compare their levels of IGFBP2 expression relative to Barrett's esophagus and normalized to GAPDH. Baseline clinical data on all subjects included in this study are provided in Table 1, and chemoradiation therapy regimens for treated patients are detailed in Supplementary Table S1. Although there was considerable variability within the two groups, www.impactjournals.com/oncotarget expression (K_202718_at) in a progression series of human esophageal tissues ranging from Barrett's metaplasia, low-grade dysplasia, high-grade dysplasia, to adenocarcinoma relative to the Barrett's metaplasia mean; B. Immunoblot analysis of protein extracts from 5 micro-arrayed EACs paired with their matching Barrett's metaplasia; C. Representative immunohistochemistry of EAC tissue microarrays at both 10× and 20× magnification (0, no staining; 1+, < 10% staining; 2+, 10-50% staining; 3+, > 50% staining). www.impactjournals.com/oncotarget Figure 2: Association of IGFBP2 expression with chemoresistance in EACs. A. Affymetrix HG-U133A oligonucleotide microarray of IGFBP2 expression (K_202718_at) in a cohort of disease-free versus recurrent disease chemonaïve EACs relative to the overall tumor mean; B. Kaplan-Meier Survival curve of IGFBP2 expression in this cohort of chemonaïve EACs, grouped according to high (n = 11) and low (n = 9) IGFBP2 expression relative to the mean (p = 0.0330, log rank test); C. Box plot representation of real-time PCR analysis of IGFBP2 expression in 200 chemonaïve EACs and 16 treatment-resistant EACs relative to Barrett's esophagus mean and normalized to GAPDH. Differences in IGFBP2 levels were statistically significant as determined by nonparametric Mann-Whitney analysis (p = 0.0097) and categorical binary analysis using a 1.5-fold threshold (p = 0.006). www.impactjournals.com/oncotarget IGFBP2 expression was significantly higher in resistant EACs as determined by nonparametric Mann-Whitney analysis (p = 0.0097) ( Figure 2C). A categorical binary analysis was also performed using a 1.5-fold threshold, which would be equivalent to a 2-fold change assuming approximately 70% tumor content for each tissue analyzed. Expression of IGFBP2 differed significantly between untreated and treated samples (p = 0.006).

IGFBP2 expression in EAC cell lines
The three EAC cell lines Flo-1, OE19 and OE33 were utilized in functional assays to investigate the role of IGFBP2 in EAC tumorigenesis and chemoresistance. Real-time PCR analysis of the EAC cell lines Flo-1, OE19 and OE33 revealed high IGFBP2 expression in Flo-1 cells but minimal expression in OE19 and OE33 cells (Supplementary Figure S2A). Although IGFBP2 protein expression was consistent with the levels of IGFBP2 mRNA expression in Flo-1 and OE33, OE19 cells expressed relatively high levels of IGFBP2 protein, suggesting a potential post-transcriptional or post-translational modification to stabilize the protein (Supplementary Figure S2B).

Effect of IGF1 and IGFBP2 on proliferation of EAC cells
Since IGFBP2 may rely on its interaction with IGFs to exact its role in chemoresistance, we evaluated the proliferative effects of exogenous human recombinant IGFs in EAC cell lines. IGF1 promoted dose-dependent growth in Flo-1 and OE33 cells in serum-free conditions but had little to no effect in serum-replete conditions (Supplementary Figure S3A & S3B), suggesting that the proliferative impact of IGF1 is masked in serum-replete conditions. In contrast to what has been previously reported [12], OE19 cells did not significantly respond to IGF1 in the presence or absence of serum (Supplementary Figure S3C), emphasizing its self-sufficiency in growth signaling. Additionally, the proliferative response to exogenous IGF1 was blunted in Flo-1 cells treated with siRNA to IGFBP2 in the absence of serum (data not shown), suggesting that IGFBP2 may potentiate IGF1dependent proliferation in these cells.
Treatment with siIGFBP2 alone reduced proliferation of Flo-1 cells in serum-free conditions, suggesting an IGFindependent role of IGFBP2 in the proliferation of these cells. In most instances, this IGFBP2 effect was seen in the presence or absence of IGFs, but addition of IGFs also enhanced this reduction, possibly due to increased proliferation in IGFBP2-expressing controls. This inhibitory effect of IGFBP2 knockdown was not consistently observed in serum-replete conditions, suggesting that IGFBP2 may have differential roles or significance based on the availability of additional growth factors.
An IGFBP2 expression construct was transfected into OE33 EAC cells with baseline endogenous IGFBP2 expression and stable cells derived from single cell clones were selected for their increased IGFBP2 expression levels (Supplementary Figure S4A). Differential clonal proliferation was observed among the individual OE33/ IGFBP2 cells, independent of their IGFBP2 expression levels (Supplementary Figure S4B), indicating that these increased levels of IGFBP2 expression were not responsible for the altered proliferative potential and that an inherent heterogeneity exists in OE33 cells. We did, however, observe a statistically significant proliferative advantage for OE33 cells treated with exogenous human recombinant IGFBP2 in an IGF1-dependent manner in serum-replete conditions. This effect was not observed in serum-free cultures (Supplementary Figure S4C).

Correlation of IGFBP2 and EMT-related gene expression
SiRNA knockdown of IGFBP2 in Flo-1 was associated with increased expression of several epithelialmesenchymal transition-related genes, including SNAI1, ZEB1, VIM, MMP9, and MMP1 (Supplementary Figure  S5A). Treatment with CDDP resulted in a marked increase in SNAI1, SNAI2 and FN1 expression while CDH1 and VIM expression decreased (Supplementary Figure S5B). CDH1 levels decreased while CDH2 levels increased in siIGFBP2/CDDP-treated Flo-1 cells as compared to mocktreated or siNon-targeting controls (Supplementary Figure  S5C), indicating an "E-cadherin to N-cadherin switch" that has been previously shown to contribute to tumor cell motility and invasion [41,42]. Expression of MMP1, MMP9 and SNAI1 was higher in siIGFBP2-treated cells as compared to the non-targeting siRNA control, suggesting a potential relationship between IGFBP2 and these genes in CDDP-treated cells (Supplementary Figure S5B).

IGFBP2-mediated migratory and invasive capacity of EAC cells
To determine the role of IGFBP2 in EAC cell motility, we used siIGFBP2-treated Flo-1 cells and IGFBP2 stably-transfected OE33 cells in wound healing assays. IGFBP2-expressing OE33 stable cells migrated further than empty vector controls by 8 hours post-wound and fully closed the wound by 20 hours (Supplementary Figure  S6A). Conversely, Flo-1 cells treated with siIGFBP2 appeared to migrate into the wound more slowly than the non-targeting siRNA control (Supplementary Figure S6B). Boyden chamber assays using IGFBP2-expressing Flo-1 cells showed only marginal Matrigel invasion differences upon knockdown of IGFBP2 (Supplementary Figure S6C). These subtle differences suggest that although IGFBP2 might not be the driver gene for metastasis, this protein does contribute to EAC cell migration and invasion.

Role of IGFBP2 in chemoresistance in vitro
In order to evaluate the role of IGFBP2 as a potential modulator of chemoresistance in vitro, IGFBP2-expressing Flo-1 cells were pretreated with siRNA to IGFBP2 while baseline IGFBP2-expressing OE33 cells were transfected with a pEGFP-IGFBP2 expression construct followed by a 3-day treatment with low doses of CDDP or 5-FU. The dose ranges utilized in these assays were clinically relevant as determined by previously reported achievable in vivo drug plasma concentrations [43,44]. Inhibition of IGFBP2 expression sensitized Flo-1 cells to CDDP but not 5-FU in serum-replete conditions ( Figure 3A). Addition of either exogenous IGF1 or IGF2 did not appreciably alter this effect ( Figure 3B). In serum-free cell cultures, IGFBP2 knockdown did not enhance the response to chemotherapy.
Acute 24 hr treatments with high doses of CDDP following IGFBP2 knockdown were also performed. In contrast to 3-day low-dose CDDP treatments, siIGFBP2treated Flo-1 cells were only slightly sensitized to CDDP (Supplementary Figure S7A), suggesting that the chemosensitization effect of IGFBP2 knockdown may be dependent on its modulation of proliferation and would require longer incubation periods to detect differences. Neither stable overexpression of IGFBP2 (Supplementary Figure S7B) nor exogenous IGFBP2 (Supplementary Figure S7C) altered the chemotherapy sensitivity of OE33 cells, suggesting that this ERBB2-amplified [45] cell line is not dependent on IGFBP2 expression for cell survival. The chemosensitization effect of knocking down IGFBP2 was also examined in endogenously-expressing OE19 cells. This more CDDP-resistant EAC cell line also has ERBB2 amplification [45] and was not as responsive to siIGFBP2 treatment as Flo-1 cells (Supplementary Figure S7D), suggesting that IGFBP2 is not a driver of chemoresistance in these cells.

IGFBP2-dependent effects of IGF1 on ERK and AKT activation
IGFBP2 has been shown to stimulate PI3K and MAPK activity in transformed cells by both IGFdependent and IGF-independent mechanisms [19,23,32,33,35,46]. To further explore the mechanism of IGFBP2induced chemoresistance, we assessed whether inhibition of IGFBP2 could modulate IGF1-induced ERK and/or AKT activation ( Figure 4A). Addition of recombinant IGF1 only transiently activated ERK and this activation at 15 minutes was attenuated by siIGFBP2 treatment. IGF1 inhibited ERK activation at 1 hour in an IGFBP2independent manner. In contrast, AKT was significantly activated by exogenous IGF1 or IGFBP2 knockdown at each measured time point as compared to the non-targeting siRNA control. SiIGFBP2-induced AKT activation was further enhanced by the addition of IGF1.

Effects of chemotherapy on IGFBP2 and its effectors
To better understand the role of IGFBP2 in chemoresistance, we examined the effects of CDDP and 5-FU treatment alone on IGFBP2 and these critical signaling pathways. All three EAC cell lines used in this study were sensitive to CDDP. In Flo-1 cells, short-term treatment with higher doses and longer-term treatment with lower doses of CDDP or 5-FU both resulted in decreased levels of IGFBP2 ( Figure 4B). OE33 cells were more sensitive to 3-day treatment with CDDP and also expressed much lower levels of IGFBP2 than Flo-1 cells ( Figure 4B). Interestingly, in both Flo-1 and OE33 cell cultures, cells surviving repeated acute IC90 CDDP treatments had decreased expression of IGFBP2 (data www.impactjournals.com/oncotarget siRNA SMARTpool (siIGFBP2 pool), siNon-Targeting control (siNonT) or lipofectamine alone (Mock), Flo-1 cells were mock-treated or treated with 1 μg/mL (3.3 μM) CDDP, 2.5 μg/mL (8.3 μM) CDDP, 2.5 μg/mL (19.2 μM) 5-FU or 5 μg/mL (38.4 μM) 5-FU for 3 days and analyzed for cell viability using Cell Proliferation Reagent WST-1. Concurrent qRT-PCR was performed to verify IGFBP2 knockdown. B. Following 24-hour treatment with individual ON-TARGETplus IGFBP2 siRNAs, siNon-Targeting controls or lipofectamine alone, Flo-1 cells were mock-pretreated or pretreated with 200 ng/mL IGF1 or IGF2 for 1 hour followed by mock-treatment or treatment with 1 or 2.5 μg/mL (3.3 or 8.3 μM) CDDP in serum-free or 10% serum DMEM for 3 days. Cell viability was analyzed using Cell Proliferation Reagent WST-1. Columns and error bars are the mean ± SD of 3 or more wells in each experiment. Concurrent qRT-PCR was performed to verify IGFBP2 knockdown. (1 P, 1 μg/mL CDDP; 2.5 P, 2.5 μg/mL CDDP; 2.5 5-FU, 2.5 μg/mL 5-FU; 5 5-FU, 5 μg/mL 5-FU). www.impactjournals.com/oncotarget  C. Following 24-hour treatment with ON-TARGETplus IGFBP2 siRNA SMARTpool or siNon-Targeting control, Flo-1 cells were mockpretreated or pretreated with 200 ng/mL IGF1 for 1 hour followed by mock-treatment or treatment with 1 μg/mL (3.3 μM) CDDP in serum-free or 10% serum DMEM for 3 days. D. Following 24-hour treatment with individual ON-TARGETplus IGFBP2 siRNAs, siNon-Targeting controls or lipofectamine alone, Flo-1 cells were mock-pretreated or pretreated with 200 ng/mL IGF1 for 1 hour followed by mock-treatment or treatment with 1 μg/mL (3.3 μM) CDDP in serum-free or 10% serum DMEM for 3 days. Protein lysates were collected and Western blotting was performed to analyze changes in IGFBP2 expression and activated and total cellular levels of ERK and AKT. (I, 200 ng/mL IGF1; P, 1 μg/mL CDDP). not shown). OE19 cells were the most resistant to CDDP treatment. In contrast to Flo-1 and OE33 cells, IGFBP2 mRNA expression levels in OE19 cells increased from lower baseline levels following repeated acute IC90 CDDP treatments (data not shown), while no difference in protein levels were detected when cells were treated with lower doses of CDDP for 3 days (Supplementary Figure S8). These data suggest that these 3 patient-derived EAC cell lines may utilize different mechanisms when developing resistance to chemotherapeutic agents such as CDDP.
Treatment of both Flo-1 and OE33 cells with high doses of CDDP for 24 hrs led to dose-dependent ERK activation, with much higher levels observed in OE33 cells. Treatment with 5-FU at higher doses dramatically induced ERK activation in OE33 but not Flo-1 cells. Acute CDDP treatment induced AKT activation in Flo-1 cells but inhibited activation in OE33 cells. Interestingly, total AKT expression was reduced in both CDDP-treated cell lines. Exposure of Flo-1 cells to low doses of either CDDP or 5-FU for 72 hours did not significantly alter ERK activation, but the highest dose of CDDP and both doses of 5-FU analyzed significantly increased AKT activation. These results suggest that EAC cell lines utilize different signaling pathways in their response to chemotherapy.

ERK and AKT status during siIGFBP2mediated chemosensitization
We then determined the impact of IGFBP2 inhibition on ERK or AKT signaling in IGF1-and/or CDDP-treated EAC cells ( Figure 4C). IGFBP2-silenced Flo-1 cells were treated in the presence or absence of IGF1 and CDDP for 3 days. In serum-free conditions, ERK activation decreased following IGFBP2 knockdown. CDDP induced ERK activation in both serum-free and serum-replete conditions but this response was attenuated in cells treated with siIGFBP2. Only subtle changes in AKT activation were observed, suggesting that AKT activation may not be a sustained response to this combination treatment. Further examination of these signaling pathways using the individual siRNAs to IGFBP2 in the SMARTpool confirmed that IGFBP2 knockdown reduced CDDP-induced ERK activation ( Figure 4D). CDDP similarly induced ERK activation in OE19 cells (Supplementary Figure S8). IGFBP2 knockdown with the most effective individual siIGFBP2 from the SMARTpool reduced ERK activation in the presence or absence of both IGF1 and CDDP. Effector differences among individual siRNAs in a SMARTpool have been previously reported [45] and may reflect differences in sequence-specific thermodynamics that translate into differences in effectiveness between oligos.

Effect of small molecule inhibitors on siIGFBP2mediated chemosensitization
In order to detect differences between single agents and combinations of agents in chemosensitization experiments, selected drug concentrations were optimized to capture additive/synergistic effects. Because CDDP treatment induced ERK phosphorylation in all 3 EAC cell lines, we assessed whether the MEK1/2 inhibitor, selumetinib, or the more potent MEK1/2 inhibitor, trametinib, would sensitize Flo-1 cells to CDDP following IGFBP2 knockdown. Doses that inhibited ERK1/2 activation effectively increased IGFBP2 expression ( Figure 5A and 5B), reduced cell viability and enhanced CDDP sensitivity in both serum-free and serum-replete conditions. When combined with knockdown of IGFBP2, both selumetinib and trametinib increased sensitivity to CDDP in serum-replete ( Figure 5C and 5D) but not in serum-free culture conditions (Supplementary Figure S9A and S9B).
Since knockdown of IGFBP2 induced AKT activation in Flo-1 cells, we hypothesized that inhibition of this pathway would further sensitize cells to CDDP. Doses of the potent, selective AKT Inhibitor VIII that completely abrogated AKT activation decreased IGFBP2 expression ( Figure 6A) and dramatically reduced Flo-1 cell viability (Supplementary Figure S10). For combined inhibitor analyses, lower doses of inhibitor were necessary to capture additive/synergistic effects. In serum-free conditions, sensitization to CDDP was enhanced by addition of AKT Inhibitor VIII to siIGFBP2-treated cells ( Figure 6B), suggesting that simultaneous modulation of AKT and IGFBP2 may be a useful treatment strategy. Although there was a slight dose-dependent decrease in cell viability following combined AKT Inhibitor VIII and siIGFBP2 treatment, the addition of AKT Inhibitor VIII did not further enhance siIGFBP2-induced CDDP cytotoxicity in serum-replete conditions (Supplementary Figure S9C).
While IGFBP2 expression varied across the spectrum of disease represented by the progression from dysplastic Barrett's esophagus to EAC, a subset of EAC samples expressed very high levels of IGFBP2. We explored whether IGFBP2 in such a subset of tumors might be associated with tumorigenesis or treatment resistance. www.impactjournals.com/oncotarget  increasing doses of AKT Inhibitor VIII (AI) for 24 and 72 hours. Protein lysates were collected and Western blotting was performed to analyze AKT activation and IGFBP2 expression. B. Following 24-hour treatment with individual ON-TARGETplus IGFBP2 siRNAs, siNon-Targeting controls or lipofectamine alone, Flo-1 cells were pretreated with 200 ng/mL IGF1 for 1 hour followed by mock-treatment or treatment with 1 μM AKT Inhibitor VIII and, lastly, mock-treatment or treatment with 1 or 2.5 μg/mL (3.3 or 8.3 μM) CDDP in serumfree DMEM for 3 days. WST analyses were performed to assess viability of treated cells. Columns and error bars are the mean ± SD of 3 or more wells in each experiment. (1 AI, 1 μM AKT Inhibitor VIII; 1 P, 1 μg/mL CDDP; 2.5 P, 2.5 μg/mL CDDP). www.impactjournals.com/oncotarget In a cohort of patients who received chemotherapy following esophagectomy, IGFBP2 expression was significantly higher in the resected EACs of patients with shorter disease-free intervals. Additionally, real-time PCR analysis showed that IGFBP2 expression was higher among tumor specimens obtained from patients who had treatment-resistant EACs as compared with patients who had not yet received chemotherapy. The broad range of IGFBP2 expression within the chemonaïve group, coupled with our finding that higher IGFBP2 levels correlated with chemoresistance, suggests that a subset of these tumors may have exhibited resistance if treated. This idea is consistent with the observation that 20-30% of treated EACs show a high degree of resistance [3,4]. It is also probable that mechanisms of treatment resistance in some of the tumors were independent of IGFBP2 expression. As pretreatment biopsies were not available for the patients found to have treatment-resistant EACs, it is not clear whether higher IGFBP2 levels in these patients were treatment-induced or inherent in patient tissues prior to treatment. We have shown, however, that CDDP and 5-FU treatments did not induce but rather inhibited IGFBP2 expression in Flo-1 and OE33 cell lines. Additionally, in a study of childhood acute lymphoblastic leukemia, elevated IGFBP2 serum levels remained high from diagnosis through intensive chemotherapy treatment for relapsed patients, while levels were significantly reduced in patients in remission [47], suggesting that high IGFBP2 level is prognostic of worse treatment response.
The three EAC cell lines Flo-1, OE19, and OE33 were evaluated in functional assays to investigate the role of IGFBP2 in EAC tumorigenesis and chemoresistance. Effective knockdown of IGFBP2 expression was obtained with IGFBP2-targeting siRNA in Flo-1 and OE19 cells while IGFBP2 re-expression was accomplished by stable transfection or exogenous supplementation in OE33 cells. Since serum deprivation is a common environmental condition in tumors and different pathways can be utilized based on serum status [48,49], our studies were performed in both serum-free and serum-replete conditions. Additionally, as IGFBP2 is a known modulator of IGF signaling, we attempted to eliminate serum factors that could mask or interfere with our experimental manipulations. It has been previously reported that expression of downstream targets of the PI3K-AKT and EGFR-MAPK pathways increased following serum starvation in gliomas [50]. Serum starvation in vitro has been shown to reduce growth factor levels and has been reported to lead to a proliferation-quiescent status in normal cells while activating the cellular DNA damage response pathway in cancer cells [51]. Additionally, serum deprivation up-regulated invasion and the Na+/ H+ exchanger (NHE), the principle pH-regulating mechanism in tumor cells, while serum replete conditions inhibited NHE activity and invasion of breast cancer cells in a PI3K-dependent manner [52]. Comparison of fifty fibroblast cultures derived from ten anatomical sites that were cultured in 10% serum versus 0.1% serum revealed a shared stereotypical fibroblast core serum response (CSR) that strongly correlated with a typical wound healing response [53]. The wound-like phenotype predicted increased risk of metastasis and death. Examination of this CSR in different tumor types found that although almost all prostate and hepatocellular carcinomas had the seruminduced signature, the CSR signature was evident in some breast, lung, and gastric carcinomas but absent in others, reemphasizing the importance of establishing response in the context of both serum-deplete and serum-replete conditions.
Silencing IGFBP2 in Flo-1 EAC cells resulted in significant sensitization to CDDP in serum-supplemented conditions, independent of the addition of IGF1 or IGF2, but was not as effective in serum-free conditions. This difference may be due to the fact that rapidly dividing cancer cells are more sensitive to chemotherapy and/or modulation but also that IGFBP2 may affect different pathways depending on the presence or absence of additional growth factors. Modulation of IGFBP2 expression was less effective in OE19 and OE33 cells, suggesting that IGFBP2 is not a driver of resistance in these ERBB2 (HER2/neu)-amplified [45] EAC cell lines.
Subtle differences in migration and invasion assays suggest that this gene may contribute to the metastatic potential of EAC cells. We also observed, however, that IGFBP2 knockdown alone and CDDP treatment itself induced the expression of several EMT-promoting genes in Flo-1 cells. This may warrant further exploration as this may confound the utility of this gene as a target to combat EAC chemoresistance.
IGFBP2 has been shown to reduce PTEN expression and enhance AKT activation in response to IGF1 through its heparin-binding domain [19]. It plays a key role in AKT pathway activation during glioma development and progression [32] and its knockdown has been shown to decrease IGF1-stimulated AKT activation in vascular smooth muscle cells [54] and acute leukemia cells [35].
In contrast to what has been previously reported, IGFBP2 knockdown activated AKT in Flo-1 EAC cells, and the effect was further enhanced by the addition of IGF1. We also observed an inverse correlation between IGFBP2 expression and AKT activation following acute high dose and 3-day low dose treatment with both CDDP and 5-FU. These observations and previously published data showing enhanced efficacy of receptor tyrosine kinase inhibitors and cytotoxic agents with AKT inhibition [55] prompted us to evaluate whether Akt inhibition might increase tumor cell response to CDDP. In the absence of serum but in the presence of IGF1, AKT inhibitor VIII further enhanced the effects of a 3-day treatment with CDDP and was even more effective following IGFBP2 knockdown, suggesting that the acute AKT activation following IGFBP2 knockdown may have been a compensatory survival mechanism. In fact, a compensatory activation of insulin receptor-AKT signaling has been previously reported with IGF-1R monoclonal antibody therapy in human tumor cells [56,57]. In the presence of serum, however, inhibition of AKT phosphorylation did not improve response to chemotherapy over IGFBP2 knockdown alone, indicating that these mediators may be acting through the same signaling pathway or that other signaling pathways predominate in serum-replete conditions.
Modulation of IGFBP2 levels has been shown to affect ERK activation as well. Human recombinant IGFBP2 activated the ERK signaling pathway and stimulated proliferation in glioblastoma cells (through an integrin β1-dependent mechanism) [33] and ovarian cancer cells [58] and was shown to increase ERK as well as AKT activation in breast cancer cells [23]. Inhibition of the ERK pathway with PD98059 and U1026 abrogated IGFBP2-induced activation [33,58]. While lentivirusbased shRNA silencing of IGFBP2 decreased activation of ERK and AKT in human leukemia cells [35], it resulted in the inability of p53 to inhibit IGF1-induced ERK phosphorylation in PC-3 cells [59]. We show that IGFBP2 knockdown also led to decreased ERK activation in Flo-1 and OE19 EAC cells, suggesting that ERK may be a downstream target of IGFBP2 in this tissue type as well. Interestingly, previous studies have shown that resistance to the IGF-1R inhibitor NVP-AEW541 may be driven by its inability to inhibit RAS/RAF/ERK activity in colon [60], biliary tract [61] and esophageal cancer [62]. In light of this and because CDDP treatment inhibited IGFBP2 expression but induced ERK activation in EAC cells, we tested whether MEK1/2 inhibitors could enhance siIGFBP2-induced CDDP cytotoxicity in Flo-1 cells. As opposed to AKT Inhibitor VIII, selumetinib and trametinib enhanced siIGFBP2-induced CDDP sensitization in serum-replete conditions but not in serum-free conditions. These data support the idea that activation of different signaling pathways can be dependent on serum status and that the effectiveness of targeted therapy may be dependent on the tumor microenvironment.
Despite promising data in preclinical models and a strong rationale for targeting the IGF signaling pathway in cancer, clinical efficacy of select IGF-1R inhibitors has been disappointing in multiple studies across many different tumor types [57]. Lack of targeted specificity, dependence on IGF1 levels and activation of compensatory pathways have all been implicated in the failure to improve current standard of care treatment. Of note, a phase I clinical trial using combination cixutumumab and selumetinib in advanced solid tumors showed preliminary evidence of clinical benefit [63]. IGFBP2 modulation is a promising alternative to IGF-1R inhibition due to its multi-faceted nature as a modulator of the IGF pathway, integrin signaling, cell-matrix interactions and transcription [16,64]. Currently, a NCIsponsored Phase I clinical trial is in progress to test the safety and immunogenicity of an IGFBP2-encoding DNA plasmid-based vaccine in patients with advanced stage or recurrent ovarian cancer, with the hopes of generating an IGFBP2-specific Th1 immune response (NCT01322802). Additional clinical trials are utilizing IGFBP2 as a potential immunotherapy target (NCT00821964) as well as a marker for cancer risk (NCT02450097), for colorectal cancer survival (NCT00503685) and for monitoring breast cancer treatment response (NCT01293032). Due to the established importance of both the IGF and integrin signaling pathways in esophageal cancer, IGFBP2 would be an ideal therapeutic target in the treatment of EAC. Our results linking increased IGFBP2 expression to chemoresistance in a subset of primary EACs and increased chemosensitization with its down-regulation in EAC cells makes it a strong candidate to combat resistant disease that so often claims the lives of patients.
In summary, oligonucleotide microarray and realtime PCR analyses implicated IGFBP2 as a potential mediator of chemoresistance in a subset of EACs. Further in vitro analyses of its role in EAC progression and chemoresistance showed that modulation of IGFBP2 expression affected proliferation, motility, and chemosensitization of EAC cells in a serum-dependent manner. Silencing of IGFBP2 affected both AKT and ERK activity and addition of targeted pharmacologic inhibitors of these pathways enhanced siIGFBP2induced CDDP chemosensitization. Modulation of IGFBP2 in overexpressing EACs may be an effective approach to sensitizing resistant tumors to standard of care chemotherapy. Identification of such potential mechanisms of chemoresistance might have direct relevance for subjects currently being enrolled into therapeutic clinical trials, as these studies frequently include platinum-based therapy. While further in vitro studies and clinical validation will be needed, our findings suggest that characterization of IGFBP2 expression could inform treatment selection for subjects with esophageal adenocarcinoma, particularly for the majority of patients whose tumors do not demonstrate HER2/neu overexpression.

Patients and tissues
Patients were reviewed by a multidisciplinary thoracic oncology tumor board and deemed eligible for chemoradiation and esophagectomy in terms of functional status, performance status (0-1) and tumor resectability. Written subject consent and approval of the Institutional Review Board were obtained to collect specimens from patients undergoing esophagectomy at the University of Michigan Medical Center (Ann Arbor MI). Specimens were transported to the laboratory in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Carlsbad CA) on ice. A portion of each sample was frozen in optimal cutting temperature (OCT) compound (Miles Inc., Elkhart IN) for cryostat sectioning. The remainder was frozen in liquid nitrogen and stored at -80°C. All tissue specimens were obtained from esophagectomies at the University of Michigan between 1991 and 2012 and were histologically classified as adenocarcinoma arising from either the gastroesophageal junction or the esophagus. Table 1 contains clinical information for all subjects whose tumors were analyzed in this study. Data were obtained primarily through our institutional cancer registry but supplemented with information from our prospective database and medical records.

Cell lines and cell culture
Flo-1 was derived in our laboratory from a resected stage IIb EAC and has been described previously [65]. OE33 and OE19 were derived from stage IIa and III EAC patient tumors respectively, and are maintained by The European Collection of Cell Cultures (Sigma-Aldrich, Corp.; St Louis MO). Cell lines were grown in DMEM (Flo-1) or RPMI (OE33 and OE19) supplemented with 10% fetal bovine serum (FBS; Atlanta Biologicals, Norcross GA) and 1% penicillin/streptomycin/fungizone (Invitrogen) at 37°C in 5% carbon dioxide/95% air.

Oligonucleotide microarray analysis
Total RNA was isolated from histologically normal esophageal squamous tissue, Barrett's esophageal tissue and EACs as previously described [66] using Trizol (Invitrogen) followed by RNeasy column purification (Qiagen, Valencia CA) per the manufacturers' instructions. cRNA was generated and hybridized to GeneChip HG-U133A oligonucleotide microarrays (Affymetrix, Santa Clara CA). Image analysis was performed by the University of Michigan DNA Microarray Core Facility. To normalize the microarray data, a summary statistic was calculated using the robust multichip average method [67] as implemented in the Affymetrix library of Bioconductor (version 1.3, www.bioconductor.org) which provides background adjustment, quantile normalization, and summarization. Normalized expression data was deposited to Gene Expression Omnibus (GEO), accession GSE37203 [68].

Real-time PCR analysis
Total RNA from treated cells was isolated and column purified using the RNeasy Mini Kit (Qiagen) per the manufacturer's instructions. RNA was eluted from the spin column using RNase-free dH 2 Table S2) was performed on the Corbett Rotor-Gene 6000 (Qiagen) or the ABI PRISM 7900HT Sequence Detection System (Life Technologies) through the University of Michigan Sequencing Core. Annealing temperatures were optimized using the Cepheid SmartCycler (Cepheid, Sunnyvale, CA). Melt curves were monitored for nonspecific SYBR Green signals. Significant differences of relative quantification were determined using the 2 -ΔΔCt method [69] with normalization to GAPDH or β-actin.

Construction of IGFBP2 stable cell line
An IGFBP2 mammalian expression construct was created and stably tranfected into OE33 cells. IGFBP2 was PCR-amplified from the cDNA of endogenously expressing Flo-1 cells using the GC-Rich PCR System (Roche, Indianapolis, IN) and primers containing EcoRI and BamHI restriction sites for directional cloning into the pEGFP-C1 vector (Clontech, Mountain View, CA). The pEGFP-IGFBP2 or empty vector construct was transfected into OE33 cells using FuGENE 6 transfection reagent (Roche) per the manufacturer's instructions and selected with 1000 μg/mL Geneticin (Invitrogen). Colonies were ring-cloned, expanded and then maintained in growth media containing 200 μg/ml Geneticin.

Proliferation/cytotoxicity assay
The cell proliferation reagent WST-1 (Roche) was used for spectrophotometric quantification of cell proliferation, viability, and chemosensitivity in accordance with the manufacturer's directions. Relative proliferation rates of EAC cell lines and chemosensitivity analyses in stably transfected OE33 cells were calculated as a percentage of the initial T 0 reading within each cell line. T 0 readings were measured 24 hours post-seeding and marked the initiation of respective treatments. Relative cell viability was calculated as OD T /OD C × 100, where OD T represented the absorbance of the treatment group and OD C represented the absorbance of the control group.

Wound healing assay
Cells were densely plated in triplicate and allowed to grow to 100% confluence. Following serum-starvation for 24 hours, cell monolayers were wounded with sterile p200 tips. Digital images of predetermined locations in each well were taken at both 4× optical magnification on an Olympus CK2 inverted microscope using a SPOT Idea 1.3 MP camera and analyzed with SPOT Basic software (Diagnostic Instruments, Sterling Heights, MI) immediately after wounding and subsequently until the wounds were closed. The percentage of initial wound width was determined by averaging measurements per well at each time point. These percentages were then averaged per group and compared across groups.

Matrigel invasion assay
Ice-cold matrigel was diluted with coating buffer to a final concentration of 250 μg/mL. Diluted matrigel (100 μL) was added to each upper chamber of a 24-well transwell plate and allowed to gel at 37°C for 6 hours. Flo-1 cells that had been pretreated with siIGFBP2 or Nontarget siRNA control (Dharmacon) were trypsinized and resuspended in siRNA-containing serum-free DMEM to a final concentration of 40,000 cells in 500 μL and added to each transwell in duplicate. Pretreated cells were also added to uncoated transwells to serve as internal controls for each treatment group. Lower chambers were filled with 750 μL 10% FBS DMEM and plates were incubated at 37°C for 36 hours. Transwells were removed and stained with Diff-Quick solution. Noninvaded cells were scraped from the inside of the transwell with a cotton swab. Invaded cells were counted in 5 fields per transwell by light microscopy and percent invasion was calculated as I C /I UNC × 100, where I C represented total invaded cells counted per coated transwell and I UNC represented total invaded cells counted per uncoated transwell for each treatment group.

ACKNOWLEDGMENTS
We thank David Erdody for technical assistance in figure preparation. This work was supported by the NIH through U54CA163059 (DGB), 5K08CA127212 (ACC) and the Thoracic Surgery Foundation for Research and Education (ACC).