FOXO transcriptional activity is associated with response to chemoradiation in EAC

In this study we aimed to investigate signaling pathways that drive therapy resistance in esophageal adenocarcinoma (EAC). Paraffin-embedded material was analyzed in two patient cohorts: (i) 236 EAC patients with a primary tumor biopsy and corresponding post neoadjuvant chemoradiotherapy (nCRT) resection; (ii) 66 EAC patients with resection and corresponding recurrence. Activity of six key cancer-related signaling pathways was inferred using the Bayesian inference method. When assessing pre- and post-nCRT samples, lower FOXO transcriptional activity was observed in poor nCRT responders compared to good nCRT responders (p = 0.0017). This poor responder profile was preserved in recurrences compared to matched resections (p = 0.0007). PI3K pathway activity, inversely linked with FOXO activity, was higher in CRT poor responder cell lines compared to CRT good responders. Poor CRT responder cell lines could be sensitized to CRT using PI3K inhibitors. To conclude, by using a novel method to measure signaling pathway activity on clinically available material, we identified an association of low FOXO transcriptional activity with poor response to nCRT. Targeting this pathway sensitized cells for nCRT, underlining its feasibility to select appropriate targeted therapies.


Introduction
Over the past three decades, incidence rates of esophageal adenocarcinoma (EAC) have been progressively increasing [1]. This increase is up to six fold in Western countries and still rising, characterized by a striking rise in incidence of males. For patients with locoregional EAC disease the preferred treatment strategy is neoadjuvant chemoradiation therapy (nCRT) or perioperative chemotherapy followed by an esophagectomy [2], but prognosis remains poor [3]. Thus far, no targeted treatment options are available in the curative setting. Treatment strategies directed against signaling pathways that drive treatment resistance, could improve therapy outcomes.
EAC is characterized by a complex network of aberrant signal transduction pathways, making therapeutic targeting challenging [4,5]. In order to develop novel targeting strategies, these aberrant pathways must be identified. Up until now, methods to examine such pathways have relied on assessments of protein (over-)expression or activation, copy number variations, gene mutations, and full transcriptome through RNA-sequencing approaches [6]. However, these modalities have limited value for implementation in clinical practice. The fresh frozen tissue required for many of these analyses is usually not available, and often the expression of a single protein is assessed, not fully capturing relevant tumor biology. Here, we address these issues by a novel previously described computational Bayesian approach to interpret Page 2 of 11 Creemers et al. Journal of Translational Medicine (2022) 20:183 mRNA expression levels of gene sets [6][7][8]. Target gene mRNA expression levels are used to infer transcription complex activity of the corresponding signal transduction pathway and thereby the activity of the signal transduction pathway. The advantage of such analysis is that a more complete and therefore reliable read-out of the signal transduction pathway is obtained, compared to an mRNA measurement as a proxy for merely its corresponding protein activity [6,7]. Importantly, mRNA extracted from a small amount of formaldehyde fixedparaffin embedded (FFPE) tissue, which is readily available in clinical practice, suffices for the analysis. The aim of this study was to identify aberrant signaling pathways that drive therapy resistance in EAC. To reveal candidate signaling pathways as targets for targeted therapy, we investigated key signal transduction pathways in material available from daily clinical routine. Tissue specimens before and after nCRT, and for primary tumor and recurrent disease were assessed and compared to clinicopathological outcome data. Using patient-derived cell lines, candidate signaling pathways were targeted to study their potential therapeutic relevance.

Patient selection
Medical records of a national referral center for esophagogastric cancer, the Amsterdam UMC, location Academic Medical Center, were systematically searched for patients with histologically proven EAC, including gastric esophageal junction (GEJ), treated between June 2004 and May 2013. (i) Resectable disease: patients treated with curative intent by an esophagectomy as single treatment or by the CROSS regimen: nCRT with paclitaxel (50 mg/m 2 body-surface area), carboplatin (AUC 2 ml/ min), 41.4 Gy radiotherapy in 23 fractions of 1.8 Gy, followed by resection. As part of a randomized phase II study [9], panitumumab, an anti-EGFR monoclonal antibody, was added to the nCRT regimen. Sensitivity analysis showed no significant differences between nCRT only or nCRT + panitumumab, thus in following analysis groups were merged. Cases with available pre-treatment biopsy of the primary tumor site and the corresponding resection specimen were selected. (ii) Recurrent disease: patients with available resection specimen of the primary tumor site and a corresponding metachronous recurrence. Medical records were extracted by a trained physician using standardized data extraction concerning the location of primary tumor, TNM stage of pathological resections reports (pTNM), Mandard response score, received treatment and survival. Confirmation of adenocarcinoma histological subtype, response to therapy and non-cancerous healthy tissue were assessed by a trained pathologist.

RNA analysis
Using hematoxylin and eosin (H&E) stained slides of included biopsies and resection specimens, FFPE tumor blocks containing the highest tumor percentage were selected by a trained pathologist. Selected H&E slides were scanned using Philips Ultra-Fast Scanner (UFS; Philips, Eindhoven, the Netherlands). Tumor areas were marked by a trained pathologist using the Digital Pathology Images portal (Philips, Eindhoven, The Netherlands). A minimal area of 2 mm 2 was annotated. In case of diffuse tumor distribution, areas with a high tumor density were selected. Using a custom-built device, digital annotations of whole slide H&E scans were transferred to an adjacent H-stained slide and deparaffinized. Marked tumor areas were scraped off for RNA extraction with the RNeasy kit (Qiagen). qPCR data from extracted RNA were subjected to stringent quality checks (QC) prior to determining signaling pathway activity scores of the androgen receptor (AR), estrogen receptor (ER), Phosphatidylinositol 3 Kinase (PI3K)-Forkhead Box O (FOXO), Hedgehog (HH), Transforming growth factor receptor Beta (TGF-β) and Wingless-Integrated (WNT) pathways by applying the computational model described below [6,7,[10][11][12].

Measuring activity of signal transduction pathway activity
Measuring signal transduction pathway activity on Affymetrix U133 Plus2.0 data has been described extensively before. Pathway activity was inferred from target gene mRNA levels of six key oncogenic signaling pathways that play a role in tumor growth and metastasis [6] using a Bayesian network, inferring the odds for pathway activity, as published earlier [6,7,10,11]. Affymetrix assays were converted to qPCR-based assays, developed and performed by Philips (Philips Molecular Pathway Dx, Eindhoven) [13]. Samples that failed quality control (QC) were removed from analysis. PI3K pathway activity is inversely related to the measured FOXO transcription factor activity, on the premise that no cellular oxidative stress is present. Therefore, FOXO activity score was interpreted together with SOD2 target gene expression level to distinguish between oxidative stress-and growth control-induced FOXO activity, as described before [7,11].

Western blot
Pre-treatment cells were lysed in RIPA buffer (Cell Signaling) containing phosphatase and protease inhibitor cocktail (Cell Signaling). Protein levels were determined by BCA (Pierce). Samples were heated for 5 min at 95 ºC loaded on 4-20% polyacrylamide precast gels (Bio-Rad) and transferred to PVDF membranes. Samples were blocked with 5% BSA (Lonza) in Tris buffered saline with 0.1% Tween-20 (TBS-T), and incubated overnight at 4 °C with primary antibodies (Additional file 3: Table S3). All were used at 1:1000. (HRP)-conjugated secondary were used at 1:5000 and incubated for 2 h at room temperature. Proteins were imaged using a FuijFilm LAS 4000 imager (Fuji), using ECL Western blotting substrate (Pierce). Western blot bands were quantified using Image J by dividing protein of interest per lane by the housekeeping protein.

Cell viability assay
Cell viability was determined using a Cell Titer-Blue Cell Viability Assay kit (G8081; Promega, Madison, WI). Cells were seeded into 96-well plates in triplicates. After cell adhesion overnight, cells were treated. After one week, cell viability was measured by adding 20 μL of Cell Titer-Blue reagent to each well followed by three hours incubation. Plates were read at 560/590 nm in a cytofluormeter (BioTek Instruments). Viability was calculated from values from CRT cells with or without PI3K inhibitors, minus baseline cell viability.

Imaging based proliferation assay
Proliferation was determined using IncuCyte ™ live cell imaging system (Essen BioScience), quantitatively detecting live cells. Cells were imaged after one week of treatment.

Apoptosis assay
Apoptosis was similarly assessed using the IncuCyte system, by incubating cells in 0.33 mg/mL annexin V-FITC, administered simultaneously with drugs. Cells were imaged after one week of treatment. Apoptotic fraction was calculated by the ratio of FITC-positive cell area, divided by confluence (total cell area).

Statistical analyses
Pathway activity between normal and EAC tissue were calculated using the Mann-Whitney test to compare ranks. In case of paired samples-between matched pretreatment biopsies and resection specimen or resection specimen and recurrence, the Wilcoxon matched-pairs signed ranked test was used. Correlation of pathway activities was assessed using Pearson's correlations. Sensitivity analyses were performed for patients receiving panitumumab [9]. Survival analyses were performed using Kaplan-Meier and multivariable Cox proportional hazard regression analysis, including clinically relevant clinicopathological variables. Statistical analyses were performed in R. A p-value of p < 0.05 was regarded statistically significant. In all in vitro experiments Spearman correlation tests were performed using GraphPad Prism 8. Error bars in bar graphs indicate the mean ± SD. A p-value of p < 0.05 was considered statistically significant.

Study population selection
In order to identify aberrant signaling pathways that drive therapy resistance in EAC, clinically obtained material from the Amsterdam UMC that was stored in FFPE was retrospectively screened for eligibility. Two EAC patient cohorts were assembled: a resectable disease cohort (i), and a recurrent disease cohort (ii). For the resectable disease cohort (i), 236 EAC cases with matched biopsies of the primary tumor site and post-nCRT resection specimens were identified (Fig. 1A). The majority of patients was male (N = 154 of 181, 85.1%, Table 1) and presented with advanced stage disease (T3-4, N = 154, 85.1%). nCRT was given in 175 cases (83.4%). A complete pathological response (pCR, Mandard score 1, no residual tumor) was attained in 11 (6.3%) of the 175 nCRT treated patients. For analysis also tissue samples with very low or no residual tumor cells were taken along as the tumor microenvironment plays an essential role in EAC therapy resistance [12]. Recurrence was detected in 52.5% (N = 92) of the patients. Median annotated tumor area of included biopsies was 9.4 mm 2 and 38.4 mm 2 for resection specimens. In total, 67 biopsies and 73 resection specimens were excluded due to reasons of availability or insufficient tissue and could therefore not be processed for pathway analysis. After stringent quality control, 242 specimens of 181 patients were included, comprising 92 biopsies and 150 resections. Biopsy and resection could be matched for 67 patients, rendering 134 matched and 108 unique samples. Of the 92 biopsies, 79 later received nCRT, of which 77 a mandard score could be determined after surgery. Matched material of resected primary tumor site and metachronous recurrence for the recurrent disease cohort (ii), was available for 66 patients (Fig. 1B)  Median annotated tumor area of included resections was 144.7 mm 2 and 20.9 mm 2 for recurrences.

Low FOXO transcriptional activity associates with poor response to nCRT
In order to reveal candidate signaling pathways in EAC patients associated with response to nCRT, pathway activity of the primary tumor before and after nCRT was assessed in the resectable disease cohort (i). Pathway signal transduction activity of six key oncogenic signal transduction pathways [6] was measured and compared to clinicopathological outcome data. Comparing pre-treatment biopsies and post-nCRT resection specimens (i.e. resections minus biopsies), FOXO transcriptional activity was found to remain lower in poor nCRT responders compared to good nCRT responders (Mandard 4-5 vs 1-3, p = 0.002, respectively, Fig. 2A). Both AR and TGF-β pathway pathway activities remained lower in these groups as well, albeit not statistically significant (TGF-β) or with a small difference between the medians (AR) (AR: p = 0.038, TGF-β: p = 0.066). These data show that patients with a worse pathological response to nCRT have persistently lower FOXO activity in both biopsy and resection specimens compared to patients with a good response to nCRT. Unfortunately, EAC pre-treatment biopsies showed no significant differences in activity of the assessed pathways (Fig. 2B).
When comparing pre-treatment biopsies and post-nCRT resections, FOXO (p < 1•10 -3 ), TGF-β (p < 1•10 -3 ) and HH (p = 0.003) activities were increased after nCRT, compared to no difference or a small decrease in patients who received surgery only (Additional file 1: Fig. S1A). This shows that the nCRT regimen rather than progression over time influenced pathway activity scores. Although tumor tissue is characterized by high activation of oncogenic signal transduction pathways, which pathways is cancer type specific. To assess if any pathway activity scores from EAC tumor tissue are differentially expressed from the healthy tissue from which it is derived, non-treated EAC tissue (N = 84) was compared to adjacent normal healthy esophageal tissue (N = 20). EAC tumor tissue was characterized by lower FOXO activity (p < 3.3•10 -6 ; Fig. 2C). This was accompanied by higher WNT, HH and lower TGF-β pathway activity (p = 2•10 -7 , p = 0.0023, and p = 0.011, respectively).
As FOXO and TGF-β pathway activities often showed the same trend between groups, we investigated whether these two pathways were correlated with each other. When exclusively examining post-nCRT resection specimens, FOXO and TGF-β activity were highly correlated in Mandard high (4-5) patients, while this was not the case in Mandard low (1-2) patients, supporting the hypothesis that combined low activity of both FOXO and TGF-β pathway after nCRT is related to poor prognosis (Mandard low R = 0.47, p = 0.24, Mandard high R = 0.73, p < 1•10 -8 , Additional file 1: Fig.S1B). Although a similar correlation was observed in pre-treatment samples, this was less evident and not associated with pathological response (Mandard low R = 0.11, p = 0.7, Mandard high R = 0.52, p = 0.039, Additional file 1: Fig.S1C). Moreover, a numerically lower median disease-free survival of 17 versus 97 months was seen in patients with combined low versus combined high pathway activities in post-nCRT resections (cut-off by median pathway activity  Fig.S1D). Hence, in EAC patients combined low FOXO and TGF-β pathway activity after nCRT in resection specimens is associated with poor prognosis. Next, to examine if a similar pathway activity profile could be observed in resection versus recurrent EAC specimens, we investigated the recurrent cohort (ii). In patients who had recurrent disease after nCRT and esophagectomy, reduced combined FOXO and TGF-β activity was seen in recurrent sites compared to matched resection specimens (FOXO p = 0.0007 and TGF-β p = 0.0027, respectively, Fig. 2D). This implies that low combined FOXO and TGF-β pathway activity is a profile associated with recurrent disease which is further accentuated when the disease progresses.
In conclusion, in patient subsets with a poor response to nCRT and patients with recurrent disease, we consistently found combined low activity of FOXO transcription factors, and the TGF-β pathway. Therefore, targeting these pathways additionally to nCRT could possibly improve clinical outcome in poor responders. Of all individual pathways, FOXO activity scores showed the highest discriminative power for the response to nCRT ( Fig. 2A and D). Additionally, as the net contributions  of tumor-suppressive and tumor-promoting actions of TGF-β are currently not fully understood in the context of EAC [18], we decided to assess whether targeting the PI3K-FOXO pathway could have a beneficial effect in chemoradiation treatment.

PI3K-FOXO pathway can be targeted to improve CRT response in EAC cell lines
FOXO transcription factors are known to be under negative control by the PI3K pathway, by activity of kinases AKT and S6 [19][20][21][22][23] and are well-known to be inversely correlated to PI3K signaling [21,24,25]. To investigate whether PI3K pathway inhibitors could have a beneficial additive effect in combination with nCRT, a panel of eight EAC cell lines was used, of which five derived from EAC patients treated at our hospital (EAC007B, EAC031M, EAC058M, EAC081R, EAC289B) and three publicly available EAC cell lines (Flo1, OE19, OE33). Cell lines were both assessed for FOXO transcriptional activity based on the Bayesian method, as well as baseline PI3K pathway activity by Western blot for phospho-ERK, phospho-AKT and phospho-S6 Kinase (pS6K) (Fig. 3A). When baseline FOXO transcriptional activity from these cell lines was correlated to PI3K pathway protein activity, no significant association could be found. (Additional file 2: Fig. S2A). We decided to continue with PI3K pathway assessment by Western blot, allowing a direct assessment of phospho pathway activity in vitro. Subsequently, these cell lines were exposed to our previously established in vitro CRT regimen [16] and the surviving fraction after CRT was measured. In line with our patient data, high PI3K pathway activity determined by pS6K, was associated with poor response to CRT in vitro. pS6K was found to most strongly correlate with response to CRT ( Fig. 3B; R = 0.478, p = 0.045). High pS6K cell lines were relatively resistant to CRT (EAC007B, EAC031M, EAC058M and EAC081R), whereas low pS6K cell lines were more sensitive (289B, OE19, OE33 and OE33). We take these data to indicate that in vitro, pre-treatment PI3K pathway activity as determined by pS6K, associates with, and potentially drives poor response to CRT. Aiming to improve anti-tumor efficacy of CRT in poor responder cell lines, the additive effect of PI3K inhibition to the CRT regimen was assessed. Four PI3K inhibitors with positive Phase I clinical results were selected; LY3023414, Alpelisib, Pictilisib and Idelalisib (Additional file 5: Table S2). First, primary cell lines EAC031M (poor CRT responder) and 289B (good CRT responder) were exposed to CRT alone, or with PI3K inhibitors to validate inhibition of the PI3K pathway. Indeed, combined CRT with PI3K inhibition resulted in lower PI3K pathway activity determined by pS6K, in which the poorresponder cell line EAC031M demonstrated a more pronounced response (Additional file 2: Fig.S2B, C). Next, the full panel of cell lines was exposed to a concentration range of PI3K inhibitors and CRT. Of note, a higher sensitivity to PI3K inhibitors was seen in the poor CRT responder cell line EAC031M, compared to the good CRT responder cell line 289B (Fig. 3C). Additional cell lines showed that poor CRT responder cell lines (EAC007B, EAC058M and EAC081R) as well had a higher sensitivity to PI3K inhibitors. (Additional file 2: Fig.S2D). PI3K inhibitor LY3023414 demonstrated the highest potency to inhibit cell viability in nearly all cell lines tested.
In patient-derived EAC cell lines, our group has previously shown CRT to induce epithelial-to-mesenchymal transition (EMT), characterized by a mesenchymal cellular morphology and therapy resistance [16]. Indeed, the PI3K signature was strongly associated with EMT gene expression programs in publicly available data sets (R = 0.554, p < 0.000 in GSE96669 [17], and R = 0.273 p = 0.009 in the EAC samples in TCGA-ESCA [4]; Additional file 3: Fig.S3A). In concordance, the addition of LY023414 to CRT resulted in a reduction of mesenchymal cell morphology induced by CRT (Fig. 3D) and EMT marker expression (CXCR4) [32], most notably in poor responder EAC031M cells (Additional file 3: Fig. S3B). This demonstrates that PI3K inhibition in poor CRT responder cell lines can serve to thwart resistance mechanisms possibly by counteracting mesenchymal tumor cell programs. To further demonstrate that combined targeting of PI3K with CRT results in cell death rather than merely reduced cell growth, we measured apoptosis in poor-and good-CRT responder cell lines. Poor responder cell lines such as EAC031M showed the most robust increase in apoptosis when exposed to PI3K inhibitors LY3023414, Idelalisib and Alpelisib (Fig. 3E, additional cell lines in Additional file 4: Fig. S4A). To conclude, these exploratory results indicate that Phase I PI3K inhibitors could be a promising addition to chemoradiation regimens, but the successful application of such inhibitors hinges on patient selection based on baseline PI3K-FOXO pathway activity.

Discussion
Using clinically available FFPE material we were able to determine activity of key cancer driving pathways before and after nCRT using small amounts of EAC tissue. Persistent low FOXO transcriptional activity was associated with poor response to nCRT in EAC patient samples. This poor nCRT responder profile was also seen in recurrences of nCRT-pretreated patients. In addition, heterogeneity in high PI3K activity, inversely linked with FOXO, was seen in patient-derived cell lines, providing a valuable tool to experimentally target candidate signaling pathways, such as PI3K-FOXO. Our exploratory data demonstrate that PI3K inhibition can indeed sensitize nCRT-resistant cell lines. As EAC is known to be a heterogeneous tumor, the full characteristics of the tumor might not be represented by the analysis of single tumor sites. This could be problematic, as large intratumoral heterogeneity might hamper patient selection for targeted therapy. In addition, not merely tumor cells but also stromal and immune cells influence EAC tumor biology [15]. These environmental stimuli are only partially captured in our analyses. Moreover, a potential lack in our study is the absence of post-treatment healthy esophageal tissue, to compare with pre-treatment healthy tissue as a control for treatment effects of nCRT on FOXO or other transcriptional activities. Additionally, as merely one FFPE tissue slide was used, sometimes insufficient mRNA was extracted in good nCRT responders (post-nCRT Mandard 1 score). By using two or more FFPE slides, this issue could be circumvented. Nevertheless, we demonstrate that in a single tissue sample from a possibly heterogeneous and effectively treated tumor, clinically relevant sample characteristics can be determined.
Here we described that low FOXO transcriptional activity is associated with worse clinical outcome, sometimes combined with low TGF-β pathway activity. TGF-β signaling has a complex dual role in human cancer [18,33]. Via the canonical Smad pathway TGF-β signaling has tumor suppressive effects in early carcinomas. As tumors develop, these protective effects of TGF-β are often lost and TGF-β signaling switches to promote cancer progression, as we have shown before using flow cytometry analysis [16]. Nevertheless, based on our mRNA based pathway activity analysis TGF-β signalling activity in EAC samples seems to be predominantly tumor suppressive. Given that the dual role of TGF-β has not been fully elucidated in EAC, to design a new treatment strategy, we focused on targeting the PI3K-FOXO pathway.
The phosphatidylinositol 3-kinase (PI3K) pathway is one of the main cellular growth factor signaling pathways frequently hyperactivated in cancer [34]. Because PI3K pathway activity negatively regulates forkhead box-O (FOXO) transcription factor activity, FOXO target gene expression is inversely correlated with PI3K activity. Additionally, numerous studies have revealed that high PI3K and low FOXO activity is associated with a poor prognosis [25,35,36], although little research has been done in EAC [35,36]. As our results show an association with poor response and metastatic behavior, in line with other cancer types, PI3K inhibition appears a promising candidate for therapeutic targeting in EAC patients [10,37].
To experimentally test novel therapeutic strategies as PI3K pathway inhibition, preclinical cancer models representative of patients' tumor biology are essential. We found that patient-derived cell lines show heterogeneity in PI3K pathway activity, allowing their use as a tool to test predictive signals for response to PI3K pathway inhibition. In line with our patient data, both good and poor nCRT responder cell lines were identified with corresponding PI3K pathway activity on phosphorylated protein level. In the relatively small number of cell lines we could not find and association of nCRT on FOXO transcriptional activity level. However, as phosphorylation plays an essential role in activation of the comprehensive PI3K pathway, we were able to detect a relation on protein phosphorylation level. Although these experimental data are encouraging, caution must be taken drawing conclusions as additional experiments should be performed in 3D models.
Nonetheless, we identified a PI3K inhibitor that could give a beneficial additional effect to nCRT; LY3023414, a novel PI3K/mTOR inhibitor, with potent activity in EAC in rat models [38]. Interestingly, LY3023414 demonstrated a manageable safety profile in a phase 2 clinical trial [39], as improved clinical responses of all four used inhibitors [26-29, 39, 40]. Hence, we believe it is of value to investigate the use of PI3K pathway inhibition in patients with EAC, either combined with nCRT, or as adjuvant therapy.
To conclude, we were able to determine the activity of key cancer driving pathways before and after nCRT in clinically attainable amounts of EAC tissue. The poor nCRT responder profile was detected in tissue of the primary site, recurrences of nCRT-pretreated patients and patient derived cell lines. Hence, the pathway activity model described here may be used to identify patients irresponsive to nCRT and select for appropriate targeted therapies.

Novelty and impact
By using a novel Bayesian inference method to measure signaling pathway activity on clinically available material, we identified an association of low FOXO transcriptional activity with poor response to nCRT. Targeting this pathway sensitized cells for nCRT, underlining its feasibility to select appropriate targeted therapies.