Meta-analysis reveals the correlation of Notch signaling with non-small cell lung cancer progression and prognosis

Various studies have assessed the clinicopathological and prognostic value of Notch1 and Notch3 expression in Non-small cell lung cancer (NSCLC), but their results remain controversial. This meta-analysis was conducted to address the above issues by using a total of 19 studies involving 3663 patients. The correlations between Notch1 and Notch3 expression and clinicopathological features and NSCLC prognosis were analyzed. The meta-analysis indicated that higher expression of Notch1 was associated with greater possibility of lymph node metastasis and higher TNM stages. Moreover, patients with Notch1 overexpression and Notch3 overexpression showed significantly poor overall survival (Notch1: HR, 1.29; 95% CI, 1.06–1.57, p = 0.468 and I2 = 0.0%; Notch3: HR, 1.57; 95%CI, 1.04-2.36, p = 0.445 and I2 = 0.0%). Furthermore, there are statistically significant association between overall survival of NSCLC patients and the expression of Notch signaling ligand DLL3 and target gene HES1. Our meta-analysis supports that Notch signaling is a valuable bio-marker to predict progression and targeting Notch signaling could benefit subpopulation of NSCLC patients.

Impact of the expression of other members of Notch signaling on survival for NSCLC. Notch signaling is initiated by ligand binding to Notch receptor; therefore, we further analyzed expressions of Notch ligands and prognosis of NSCLC patients. As shown in Fig. 5, a statistically significant association between DLL4 expression and NSCLC overall survival rate was found (NSCLC: pooled HR = 1.60, 95%CI: 1.18-2.17, p = 0.547 and I 2 = 0.0%; lung adenocarcinoma: pooled HR = 1.66, 95%CI: 1.10-2.50, p = 0.206 and I 2 = 36.7%; lung squamous cell carcinoma: pooled HR = 2.56, 95%CI: 1.56-4.19, p = 0.169 and I 2 = 47%). However, there was no significant relation between the expressions of DLL1 and DLL3 and overall survival of NSCLC patients.
Whole genomic expression profiles provide unbiased quantitative measure of mRNA abundance and have been proved to correlated with histo-pathological characteristics and predict prognosis 24 . To investigate whether HES1 mRNA abundance correlated with NSCLC survival, we analyzed published gene expression database of NSCLC with survival information. Kaplan -Meier survival curve of GSE31210 comprised of 226 NSCLC patients at stage I-II demonstrated that the overall survival rate was lower in patients with higher expression of HES1 (p = 0.007) (Fig. 7A). To explore whether mutation of EGFR interferes with HES1, NSCLC patients were subdivided into EGFR mutation and wild type (wt) group. The overall survival rate of patients with EGFR mutation was correlated with the abundance of HES1 expression (p = 0.008) (Fig. 7B). However, there is no statistical relation between HES1 expression and OS in EGFR wt patients (p = 0.244) (Fig. 7C). Together, analysis results from HES1 mRNA profile are in consisting with protein abundance, both indicating that higher Notch signaling activity predicts worse prognosis.

Discussion
It is generally assumed that Notch1 and Notch3 activity were higher in advanced NSCLC and predicted poor prognosis 16,19 ; However, opposite result was also reported in lung squamous carcinoma 10 . Precisely measuring the prognostic value of Notch1 and Notch3 may help to guide an individual therapy for NSCLC patients 24 . In our study, a combined analysis of 19 eligible clinical studies revealed a predictive value of Notch1 and Notch3 expression in NSCLC patients. The meta-analysis results suggested that Notch1 expression was significantly higher in lung cancer compared with normal tissues, and correlated with lymph node metastasis and TNM stages. However, Notch 3 was only associated with lymph node metastasis. In addition, our meta-analysis suggested that overexpression of Notch1 and Notch 3 could be a prognostic marker for OS. Notch1 is commonly expressed in malignant cells from different types of cancer, participating in multiple functions, including motility, cell-cell connections and cell polarity 25,26 . Thus, Notch1 signaling could affect invasion, lymph node metastasis and TNM stages. Multiple studies have also demonstrated that Notch3 plays a role in the regulation of cellular proliferation, differentiation, apoptosis, as well as tumorigenesis 27 . Moreover, high Notch3 expression in lung cancer represented a higher possibility of being resistant to chemotherapy 28 . In support of clinical-pathological observation, experimental study demonstrated that high Notch activity enhanced tumor sphere formation and knockdown of Notch decreased cell proliferation and induced apoptotic cell death 29 . It has been shown that block Notch activity by GSI reduced xenograft growth of Notch abundant cells and decreased tumor incidence upon re-implantation. Correspondingly, molecular analysis confirmed the reduced expression of downstream effectors of Notch pathway from tumor xenografts of mice treated with GSI 23 , demonstrating a potent antitumor efficacy of Notch1 inhibitor. These findings suggest that GSI may provide novel therapies to improve the efficacy of conventional therapies by directly targeting the CSC, thus delaying tumor recurrence and improving cancer patient survival.
The identification of biomarkers for micro-metastases, disseminated tumors, and residual disease is critical for early detection and treatment of these diseases to prevent metastases and recurrence. To elucidate the correlation between the Notch signaling and NSCLC prognosis, analysis of Notch signaling cascade is necessary. Thus, the ligands and downstream target genes of Notch signaling were evaluated. In previous study, Dll4 was found to positively correlate with VEGFR1. Moreover, by up-regulation of Notch1 and VEGFR1, DLL4 played important roles in tumor angiogenesis and prognosis of lung  30 . In addition, the activation of HES1 in NSCLC leaded to tumor cell growth and tumor progression 31 . Our meta-analysis is in agreement with previous report, showing that DLL4/HES1 were positively associated with poor OS of NSCLC patients.
Currently, new target therapies show a significantly improved progression-free survival (PFS) and relatively less toxicity in comparison with standard chemotherapy 2 . However, the inevitable resistance to target drugs limits theirs clinical efficiency. Identification of molecules cross-talk with target pathways will be critical for further improvements in NSCLC treatment. In vivo experimental studies observed significant antitumor activity in lung xenograft models accompanied with impaired tumor vasculature by reduced expression of several key angiogenic genes after the treatment of GSI RO4929097. Molecular analysis revealed Notch pathway target gene HES1 strongly corresponded with the therapeutic response. Thus, Notch pathway downstream genes could be used to predict the antitumor activity of RO4929097 32 .
EGFR mutations are one of the most important active genetic changes yet discovered in NSCLC. Both EGFR and Notch signaling are known to be deregulated in many human cancers. However, interactions of those two pathways are not well understood. It is observed that in human lung cancer cell line NCI-H292 cells, inhibition of the Notch pathway decreased both the Notch and EGFR/ERK pathways, associated with the reduction of proliferative cells. These results suggest that Notch signaling pathway has crosstalk with EGFR/ERK in human bronchial epithelial cells 33 . It can be proposed that GSIs might synergize with other signaling pathway inhibitors to treat NSCLC. For example, Notch 1 contributed to the acquired resistance to TKI and inhibition of Notch-1 resulted in effective response to EGFR target therapy (gefitinib) 34 .
Heterogeneity test is an essential part in meta-analysis. In this study, evidence of minor heterogeneities was observed; however, there was still some problems worth considering. Firstly, multicenter prospective studies based on large, homogeneous patient populations will be required. Another significant heterogeneity was likely due to the variations in assessing Notch signaling expression. The cutoff value was estimated in 5 studies by using the median Notch1 level measured by RT-PCR, remaining 12 articles measured by IHC. Some studies defined Notch1 expression based upon the percentage of immuno-positive cells or staining intensity, whereas other studies used a scoring system that combing those 2 factors together, causing the cutoff values for judging Notch1 expression diverse.
Lastly, publication bias is worth considering in meta-analysis. In this study, analyses of OS and other clinicopathological parameters did not show big variation. It should be noted that some limitations exist in the present meta-analysis. First, the number of included studies was relatively limited. Second, the applied method for detecting Notch signaling expression and the cutoff values were different in the included studies. Third, publication bias is still a concern. We tried to review all the relevant articles, but in some studies detailed description of raw data were not available.
All together, the meta-analysis provides evidence that Notch signaling is high activated in some NSCLC patients, which is associated with cancer progression and predicts bad prognosis.

Methods
Literature search. We conducted a search of electronic databases PubMed, Embase and Cochrane library published up to October 1, 2014 using the search terms Notch ("Notch Receptors", "Notch Proteins") and NSCLC ("Non-Small-Cell Lung Carcinoma, " "Non-Small Cell Lung Cancer"). The citation lists associated with the retrieved articles were also reviewed to identify additional relevant publications.
Inclusion criteria. This meta-analysis collected data from randomized controlled studies (RCTs) or observational studies (case-control or cohort) that evaluated the relationship between Notch signaling expression and NSCLC parameters such as clinicopathological features and prognostic factors. Studies met the following criteria were included: a) patients recruited in the study were pathologically diagnosed to have NSCLC; b) Notch signaling expression was measured by immunohistochemistry (IHC) or real-time reverse transcription polymerase chain reaction (RT-PCR) in primary NSCLC tissue; c) the hazard ratio (HR) and corresponding 95% confidence interval (CI) were reported or could be statistically extracted from the study. When several articles were from the same patient population, the newest or most complete article was included.
Data extraction. All data were abstracted by using a standardized data collection form, with information recorded as follows: first author's last name, publication year, country of origin, histological type, tumor stage, number of cases and controls, detection methods, cutoff values for Notch1, Notch3, DLL4 and HES1. For articles without HRs, the statistical variables were calculated from published survival curves using methods described by Tierney et al 37 . We also reviewed Oncomine and found 5 independent human NSCLC microarray datasets with Notch signaling expression and survival data. Overall survival (OS) was evaluated by Cox proportional HRs and 95% CIs using these numerical data.
For observational studies, the Newcastle-Ottawa Quality Assessment Scale (NOS) was employed for assessing the quality of these studies. This scale is based on the identification of 8 sources of potential study bias that estimate patient selection, study comparability, and outcomes. Literature search, study selection, and data extraction were performed independently by two reviewers and disagreements among reviewers were resolved by discussion.
Statistical analysis. Statistical analysis was conducted by the guidelines proposed by Meta-Analysis of Observational Studies 38 . The prognosis of NSCLC patients with expressions of Notch1, Notch 3, Notch ligands, Notch downstream target gene HES1 and HEY1was calculated by HRs and 95% CIs, respectively. Clinicopathological parameters included histological type, lymph node metastasis (LNM), TNM stage. Heterogeneity of the odds ratio (OR) and HR was appraised by using the Cochran Q and I 2 test. A random-effect model was applied when p < 0.1 or I 2 > 50%. When heterogeneity was absent, a fixed-effect model was employed. Publication bias was assessed by Begg's rank correlation method and