Prognostic value of Aurora kinase A (AURKA) expression among solid tumor patients: a systematic review and meta-analysis

Abstract

Objective

The prognostic significance of Aurora kinase A expression in cancer patients is currently under debate. Here, we conducted the first comprehensive meta-analysis of the prognostic relevance of Aurora kinase A associated with survival in solid tumors.

Methods

Pubmed was searched for studies evaluating the Aurora kinase A expression and survival in solid tumors through 10 October 2014. The main outcome analyzed was overall survival and secondary outcomes were progression-free survival, disease-free survival and cancer-specific survival. Pooled hazard ratio and 95% confidence intervals were calculated to assess the association. Subgroup and sensitivity analyses were also conducted.

Results

Thirty-nine eligible studies enrolling 5523 patients were identified. The high Aurora kinase A expression level was significantly associated with shorter overall survival (hazard ratio = 2.31; 95% confidence interval, 1.81–2.95). Further subgroup analyses showed that our results were stable irrespective of the disease sites, stages and methods of Aurora kinase A expression. The high Aurora kinase A expression level was also associated with progression-free survival (hazard ratio = 2.68; 95% confidence interval, 1.96–3.69), disease-free survival (hazard ratio = 1.91; 95% confidence interval, 1.45–2.51) and cancer-specific survival (hazard ratio = 2.13; 95% confidence interval, 1.38–3.29).

Conclusions

Our present meta-analysis indicates that the Aurora kinase A is an effective prognosticator in solid tumors patients. Further studies are required to explore the clinical utility of Aurora kinase A in solid tumor.

Aurora kinase A, prognosis, solid tumor, therapy, meta-analysis

Introduction

Cancers figure among the leading causes of death worldwide, according to the statistics, there were 14.1 million new cases and 8.2 million deaths in 2012 (1). More than 80 percent of all cancers are caused by solid tumor. Therapeutic decisions are usually based on clinical and histopathologic variables like tumor-node-metastasis stage and tumor grading, which, however, often fail to predict patient outcome. Therefore, there is an urgent need to look for effective biomarkers to lead to an improved stratification between higher-risk and lower-risk patients, which can be treated in a more selective and individualized manner.

Aurora kinase A (AURKA/STK15/BTAK) is a member of the Aurora/Ipl1p family of cell cycle-regulating serine/threonine kinases and associated with chromosomal distribution and its up-regulation induces centrosome amplification, chromosomal instability and oncogenic transformation (2–5). Studies have demonstrated amplification and/or over-expression of AURKA in several solid tumors, including ovarian and breast cancer (6,7). Recent studies have also shown the up-regulation of AURKA was associated with tumorigenesis, clinical aggressiveness and tumor progression of several cancers (8–13). These data indicate that AURKA possesses oncogenic activity and may be a valuable therapeutic target in cancer therapy (14,15). Consequently, several small-molecule Aurora A kinase inhibitors have been developed and are currently undergoing advanced clinical trials (16,17).

Recently, the amount of studies that have been conducted to investigate the relationship between AURKA expression and survival in solid tumor patients, however, the role of AURKA as a prognostic indicator is controversial. Some studies showed that high AURKA expression was significantly associated with shorter survival (13,18–20). In contrast, other studies failed to show such an association between high AURKA expression and worse prognosis (21–23). These disparate results suggest that the consistency and magnitude of the prognostic impact of AURKA are unclear. To clarify this issue, and because of the need for more accurate prognostic markers in solid tumors, we conducted the first comprehensive meta-analysis of published literature on this topic to derive a summary estimate of the prognostic significance of the AURKA for survival.

Patients and methods

Search strategy and eligibility criteria

This analysis was conducted in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (24). Pubmed was searched for studies focused on evaluating the AURKA expression and survival in solid tumors. The search strategy included the following keywords variably combined: ‘Aurora Kinase A,’ ‘AURKA,’ ‘Aurka,’ ‘ Aurora A,’ ‘ STK15,’ and ‘prognos*.’ Furthermore, reference lists of retrieved articles and review articles were reviewed manually to implement our search. Only studies published in peer reviewed journals were included, data from letters and meetings abstracts were not eligible.

The studies were included in the meta-analysis if they reported survival data in solid tumors patients stratified by AURKA expression (high/positive and low/negative, AURKA mRNA expression or AURKA protein expression), provided sufficient data for determining an estimate of hazard ratio (HR) and a 95% confidence interval (CI). In case of multiple publications from the same institution with identical or overlapping patient cohorts, the most informative report or the most complete one, was included in the analysis. Although no language restrictions were imposed initially, for the final analysis only articles written in English were included.

We did not assign each study a quality score, because no such score has received general agreement for use in a prognostic meta-analysis, especially of observational studies. Instead, we conducted subgroup and sensitivity analyses because they are widely recommended.

Data extraction and statistical analysis

We recorded the following information about each eligible trial: author's names, journal and year of publication, country, number of patients analyzed, disease site, tumor stage [non-metastatic, metastatic, mixed (non-metastatic and metastatic)], methods for detection of AURKA expression and survival analysis method (multivariate or univariate). We also recorded overall survival (OS), progression-free survival (PFS), Disease-free survival (DFS), cancer-specific survival (CSS), survival curves, HR, and 95% CI if available. OS was the primary outcome of interest, CSS, PFS, and DFS were secondary outcomes. The HR was measured by comparing the high AURKA expression and low AURKA expression arms. Data for multivariate survival analysis reported in the article were included in the present systematic review and meta-analysis; if these data were not available, then univariate analysis data were included.

HR of each study was either directly collected from the original article or calculated as suggested by Parmar et al. (25). The pooled HR for survival was calculated by fixed and random-effects models. And if heterogeneity among the studies was observed, only random-effects estimates were presented.

Heterogeneity between studies was evaluated with the Cochran's Q test as well as the I² index. Potential causes of heterogeneity were explored by subgroup analyses.

Publication bias was evaluated using the funnel plot and the Egger's and Begg's test. The effect of publication bias on the pooled effect was assessed by the ‘trim-and-fill’ method.

To evaluate the stability of the results, a one-way sensitivity analysis was conducted. The scope of this analysis was to evaluate the influence of individual studies by estimating the average HR in the absence of each study.

Extracted data were combined into a meta-analysis using RevMan 5.2 analysis software (Cochrane Collaboration, Copenhagen, Denmark). Estimates of hazard ratios were weighted and pooled using the generic inverse-variance and random-effect model (26). Analyses were conducted for all studies, and differences between the subgroups were assessed using methods described by Deeks et al. (26).Publication bias and sensitivity analysis were conducted with the software STATA version 12.0. All statistical tests were two-sided, and statistical significance was defined as P < 0.05.

Results

Characteristics of identified studies

Two hundred twenty-one records were identified by the primary computerized literature search. However, after screening of the titles and abstracts, 171 studies were excluded because they were either non-solid tumor, review articles, laboratory studies, written in non-English or irrelevant to the current study. Fifty articles were further reviewed in detail. Eleven studies were further excluded because of multiple publications or no survival data. Finally, 39 studies were identified as eligible for inclusion in the meta-analysis (9,18–23,27–58) (Fig. 1).

Figure 1.

Study identification flowchart.

We identified a total of 39 eligible studies that included 5523 solid tumors patients. The median study sample size was 99 patients (range, 40–645 patients). All eligible studies were published between 2005 and 2014. All TNM Stages (I–IV) were considered in 18 studies (46.2%) (9,18,20,27,28,30,34–38,40,43,44,50,55,56,58), non-metastatic solid tumors (including Stage I–III or localized disease, based solely on single-stage disease or mixture of tumor stages) were considered in 12 (30.8%) studies (22,29,31,33,39,41,42,46,48,51,53,54); metastatic solid tumors (Stage IV) patients were considered in four studies (10.2%) (23,45,49,57); disease stage was not available were considered in five studies (12.8%). Four studies analyzing AURKA mRNA expression used real-time reverse transcription PCR (RT-PCR) (20,34,35,45); one study analyzing AURKA gene amplification by fluorescence in situ hybridization (FISH) (54); the other one used immunohistochemical (IHC) staining and RT-PCR (47); and the remaining 33 studies applied a immunohistochemistry detection method (9,18,19,21–23,27–33,36–44,46,48–53,55–58). Eight studies were conducted in patients from European countries (19,21,28,32,36,40,55,57), four in patients from USA (18,20,37,45), three in Oceania countries (33,46,49), one in South America (50), and twenty-three in Asian countries (9,22,23,27,29–31,34,35,38,39,41–44,47,48,51–54,56,58). One or more oncologic outcome parameters were analyzed on multivariate analysis in 33 studies (9,18–21,29–44,46–56,58). A significant association between high AURKA expression and the poor prognosis of patients was reported in 71.4% of studies on OS (9,18,31,32,34–37,39,42–45,48,50–53,55,58). The main features of the eligible studies are summarized in Table 1.

Table 1.

Characteristics of 39 studies included in the meta-analysis

Disease site	Author	Year	Country	Sample size	Stage	Detection method	Survival analysis	Outcomes
Lung cancer
1	Xu	2014	China	102	M0	IHC	M	OS
2	Zeng	2014	China	142	M0	IHC	M	OS, DFS
3	Takeshit	2008	Japan	157	M0, M1	IHC	M	OS
4	Ogawa	2008	Japan	189	M0	IHC	M	OS
Colorectal cancer
5	Goktas	2014	Turkey	40	MI	IHC	U	OS
6	Goos	2013	The Netherlands	343	MI	IHC	M	OS
7	Belt	2012	The Netherlands	386	M0	IHC	M	DFS
8	Dotan	2012	USA	61	M1	RT-PCR	U	OS
9	Burum-Auensen	2008	Norway	55	NA	IHC	M	OS
Breast cancer
10	Yamamoto	2013	Japan	78	NA	RT-PCR, IHC	M,U	DFS, CSS
11	Dedic	2013	Croatia	215	NA	IHC	M	OS, DFS
12	Xu	2013	China	122	M0	IHC	M	OS, PFS
13	Loddo	2009	UK	182	NA	IHC	M,U	OS, DFS
Ovarian carcinoma
14	Yang	2011	USA	51	M0, M1	IHC	M	OS, DFS
15	Lassus	2011	Finland	645	M0, M1	IHC	M	DFS
16	Mendiola	2009	Spain	68	M0, M1	IHC	M	OS, PFS
17	Kulkarni	2007	UK	143	M0, M1	IHC	U	DFS
18	Landen	2007	USA	70	M0, M1	IHC	M	OS
Esophageal carcinoma
19	Hsu	2014	Taiwan	97	M0, M1	IHC	M	OS
20	Tanaka	2005	Japan	142	M0, M1	IHC	M	OS
Nasopharyngeal carcinoma
21	Liu	2012	China	208	M0,M1	IHC	M	OS
22	Wan	2012	China	144	M0	IHC	M	OS, PFS, DFS
Bladder cancer
23	Lei	2011	China	96	M0	IHC	M	OS
24	Behnsawy	2011	Japan	161	M0	IHC	U	DFS
25	Bruyere	2008	Australia	73	M0	IHC	M	DFS
Neuroblastoma
26	Ramani	2014	UK	88	M0, M1	IHC	M	OS,DFS
27	Shang	2009	USA	62	M0, M1	RT-PCR	M	PFS
GIST
28	Yeh	2014	Taiwan	99	NA	IHC	M	OS, PFS
29	Yen	2012	Taiwan	142	M0	IHC	M	DFS
Gastric cancer
30	Shinmura	2014	Japan	271	M0	FISH	M	OS
31	Wang	2011	China	89	M0, M1	IHC	M	DFS, CSS
Duodenal adenocarcinoma
32	Chen	2014	China	140	MO, M1	IHC	M	OS
Adrenocortical tumors
33	Borges	2013	Brazil	60	MO, M1	IHC	U	OS
Chondrosarcoma
34	Liang	2012	China	72	M0, M1	IHC	M	OS
Renal cell carcinoma
35	Miyake	2009	Japan	40	M1	IHC	U	CSS
Cervical carcinoma
36	Zhang	2009	China	74	M0, M1	RT-PCR	M	OS
Hepatocellular carcinoma
37	Wang	2009	China	46	M0, M1	RT-PCR	M	OS, DFS
Prostate cancer
38	Furukawa	2007	Japan	97	M0	IHC	M	DFS
Endometrial carcinoma
39	Kurai	2005	Japan	73	M0, M1	IHC	U	OS

M0, non-metastatic disease; M1, metastatic disease; NA, not available; IHC, immunohistochemistry; FISH, fluorescence in situ hybridization; RT-PCR, real-time reverse transcription PCR; M, multivariate analysis; U, univariate analysis; OS, overall survival; DFS, disease-free survival; PFS, progression-free survival; CSS, cancer-specific survival; GIST, gastrointestinal stromal tumors.

Analysis of AURKA expression effects on overall survival

The HRs for OS was available in 28 studies accounting for 3407 patients. The pooled HR showed a significantly increased risk of mortality in patients with high AURKA expression level (HR, 2.31;95% CI, 1.81–2.95; P < 0.00001, df = 27, random effects) (Fig. 2A). As the heterogeneity among studies was significant (P < 0.00001, I² = 83%), random-effects model was applied. To explore potential sources of heterogeneity, we performed subgroup analysis in the following subgroups: disease site, tumor stage and methods for detection of AURKA expression.

Figure 2.

(A) Forest plots of combined hazard ratios (HRs) of overall survival (OS) in studies of solid tumor patients associated with Aurora kinase A (AURKA) expression level. (B) Forest plots showing combined HRs of OS by disease sites subgroups. (C) Forest plots showing combined Hazard ratios of OS by disease stage subgroups. (D) Forest plots showing combined HRs of OS by methods for detection of AURKA expression subgroups.

The effect of AURKA expression on OS among disease site subgroups is shown in Fig. 2B. The prognostic effect of AURKA was highest in lung cancer (HR, 2.52; 95% CI, 1.76–3.60), followed by ovarian cancer (HR, 2.31; 95% CI, 1.52–3.52), colorectal cancer (HR, 1.97; 95% CI, 1.41–2.74), nasopharyngeal carcinoma (HR, 1.95; 95% CI, 1.21–3.15), breast cancer (HR, 1.68; 95% CI, 0.73–3.89), esophageal cancer (HR,1.35;95% CI, 0.64–2.85), the hazard ratio for the subgroup of other unselected solid tumors was 2.94 (95% CI, 1.97–4.37). For the seven disease-site subgroups analyzed, there was statistically significant heterogeneity among all different disease-sites (P < 0.00001), whereas differences between disease subgroups were not statistically significant (P for subgroup difference = 0.5).

The effect of AURKA expression on OS among different tumor stages is shown in Fig. 2C. The hazard ratios were 2.21 (95% CI, 1.60–3.04) for non-metastatic disease, 2.05(95% CI, 1.29–3.28) for metastatic disease, 2.31 (95% CI, 1.77–3.01) for a mixed group consisting of studies that included both metastatic and non-metastatic patients, and 2.08 (95% CI, 0.91–4.75) for disease stage was not available. There was statistically significant heterogeneity among all different tumor stages (P < 0.00001), whereas this difference was not statistically significant (P for subgroup difference = 0.98).

The effect of AURKA expression on OS among different methods for detection of AURKA expression is shown in Fig. 2D. The hazard ratios were 2.22(95% CI, 1.71–2.88) for immunohistochemistry, 2.78(95% CI, 2.07–3.72) for RT-PCR, and 2.11 (95% CI, 0.90–4.93) for FISH. There was statistically significant heterogeneity among all different methods for detection of AURKA expression (P < 0.00001), whereas this difference was not statistically significant (P for subgroup difference = 0.5).

We also performed the Begg's and Egger's test to assess the publication bias in 28 studies evaluating the AURKA expression and overall survival, and publication bias existed (P_Begg = 0.737; P_Egger = 0.00; Supplementary data, Figs S1 and S2). The trim-and-fill analysis revealed that 10 studies might be missing and that if these were published, the adjusted HR would be 1.817 (95% CI, 1.473–2.241; P = 0.00; df = 37; random effects; Supplementary data, Fig. S3).

One-way sensitivity analysis confirmed the stability of our results (Supplementary data, Fig. S4).

Analysis of AURKA expression effects on progression-free survival

HRs for PFS was available in five studies accounting for 495 patients. The estimated pooled HR for all studies showed a significantly increased risk of disease progression in patients with high AURKA expression level (HR, 2.68; 95% CI, 1.96–3.69; P < 0.00001, df = 4, fixed effects) (Fig. 3A). The heterogeneity among studies was moderate.

Figure 3.

(A) Forest plots of combined HRs of progression-free survival (PFA) in studies of solid tumor patients associated with AURKA expression level. (B) Forest plots of combined HRs of DFS in studies of solid tumor patients associated with AURKA expression level. (C) Forest plots of combined HRs of cancer-specific survival in studies of solid tumor patients associated with AURKA expression level.

Analysis of AURKA expression effects on disease/recurrence-free survival

HRs for DFS was available in 16 studies accounting for 2882 patients. Overall, high AURKA expression level was associated with a hazard ratio for DFS of 1.91(95% CI, 1.45–2.51; P < 0.00001,df = 15, random effects) (Fig. 3B). The heterogeneity among studies was significant (P < 0.00001, I² = 87%), and publication bias existed (P_Begg = 0.392; P_Egger = 0.00; Supplementary data, Figs S5 and S6). The trim-and-fill analysis revealed that three studies might be missing and that if these were published, the adjusted HR would be 1.728 (95% CI, 1.344–2.220; P = 0.00; df = 18; random effects; Supplementary data, Fig. S7).

One-way sensitivity analysis confirmed the stability of our results (Supplementary data, Fig. S8).

Analysis of AURKA expression effects on cancer-specific survival

HRs for CSS was available in three studies accounting for 407 patients. Overall, high AURKA expression level was associated with a hazard ratio for CSS of 2.13(95% CI, 1.38–3.29; P = 0.0007, df = 2, fixed effects). The heterogeneity among studies was absent (P = 0.29, I² = 20%) (Fig. 3C).

Discussion

The present analysis is the first comprehensive meta-analysis of published literature on the prognostic relevance of AURKA expression in solid tumors, which is based on a large pool of clinical studies (5523 patients). Here, we identified 39 studies that assessed the prognostic value of AURKA expression by immunohistochemistry as well as the RT-PCR methods. It provides evidence that the high AURKA expression is significantly associated with poorer prognosis both in OS and PFS/DFS/CSS in solid tumors, even after adjustment for publication bias. In addition, we found a consistent effect of high AURKA expression on OS among various disease subgroups, disease stages and methods for detection of AURKA expression. The pooled results are fairly stable and not influenced by the methods for detection. Of interest, this meta-analysis of pooled data confirmed that the high AURKA expression represents a significant risk factor for both non-metastatic and metastatic solid tumors.

In our subgroup analysis, among all seven disease sites of solid cancer, a high AURKA expression level was related to poor prognosis in lung cancer, ovarian cancer, colorectal cancer and nasopharyngeal carcinoma, and pooled HR was not statistically significant in breast cancer (HR, 1.68; 95% CI, 0.73–3.89) and esophageal cancer (HR, 1.35; 95% CI, 0.64–2.85). However, in the Phase 2 study of Aurora kinase A inhibitor alisertib, 9 (18%, 95% CI 9−32) of 49 women with breast cancer showed an objective response (59). In addition, the Aurora kinase A inhibitor MLN8237 could enhances cisplatin-induced cell death in esophageal adenocarcinoma cells (60). We deduced that the heterogeneity might be the main cause for this discrepancy. In addition, only a few eligible studies measuring AURKA expression in breast cancer and esophageal cancer included in this meta-analysis. These results indicate that AURKA may be a useful ‘prognosticator’ identifying a subgroup of cancer patients with a poor prognosis. However, it is desirable that future adequately designed prospective studies should be conducted within the tumor-specific cohorts.

The prognosis of patients with non-metastatic and metastatic solid tumor is obviously quite different. However, in this meta-analysis, there are 14 studies that analyzed patients without separation into non-metastatic and metastatic disease, 4 studies that analyzed patients without an accurate staging. When these 18 studies were omitted, heterogeneity was decreased or disappeared. This confirmed that non-metastatic and metastatic solid tumor patients should be analyzed separately to obtain more accurate results. However, the difference of pooled HR was not significant between non-metastatic disease (HR, 2.21; 95% CI, 1.60–3.04) and metastatic disease (HR, 2.05; 95% CI, 1.29–3.28). These results indicate that future studies should be conducted within the stage-specific cohorts, and the difference between early and metastatic disease in AURKA expression should also be focused on so that patients who will benefit from early treatment change and/or more investigational approaches.

In our subgroup analysis, the effect of AURKA expression on OS among different methods for detection of AURKA expression has its own significance. The hazard ratios were 2.22 (95% CI, 1.71–2.88) for immunohistochemistry, 2.78 (95% CI, 2.07–3.72) for RT-PCR, and 2.11 (95% CI, 0.90–4.93) for FISH. However, immunohistochemistry is an universal, effective, readily available and inexpensive detection technique compared with the other two methods.

There is significant heterogeneity among included studies in this meta-analysis, although we used random-effects models during pooling data of subgroup. The heterogeneity in these studies could be explained by different characteristics of included patients, or differences in the methods used to detect alterations in AURKA expression, including antigen retrieval methods, choice of AURKA antibody, dilutions of the antibodies, and revelation protocols. What's more, different sample types including tissue microarray (TMA) and the whole section might also contribute to the heterogeneity because it is possible that more false-negative cases are obtained in TMA than the whole section. Finally, the differences of methodology among included studies also were sources of heterogeneity and caused selection biases potentially. Therefore, validation of the prognostic power of AURKA should be conducted through large multicenter prospective studies based on homogeneous populations.

Some limitations of this meta-analysis need to be discussed. First, our meta-analysis is based on data from trials whose results have been published, and we did not obtain updated individual patient data. Use of individual patient data may further enhance the accuracy and reduce the uncertainty of the estimates. Second, publication bias is a concern. We tried to identify all relevant data, but it is unavoidable that some data could still be missing. Missing information may reflect negative results that could reduce the prognostic power of AURKA. However, we conducted the trim-and-fill analysis and found that even if these missing studies were published the association of AURKA expression and poorer survival was still significant. Third, immunohistochemistry was the most commonly applied method for detecting AURKA in situ, RT-PCR method had also been used for the evaluation of the levels of AURKA gene or mRNA expression in tumor tissue. However, only a few eligible studies measuring AURKA gene or mRNA level by RT-PCR included in this meta-analysis. Moreover, the score determining AURKA over-expression level was defined differently in these studies, leading to between-study heterogeneity. Thus we had adopted random effect model and subgroup sensitivity analyses to adjust for the shortcomings. However, it is desirable that future studies on the clinical utility of AURKA use standardized detection methods.

In conclusion, the present results support the notion of a strong prognostic value of AURKA expression in solid tumors patients. In the future, with the development of AURKA inhibitors, the detection of AURKA might serve as a tool to guide treatment in cancer patients. To achieve clinical utility of AURKA expression in solid tumors, more intervention trials in which therapy decision making is based on AURKA expression results need to be initiated.

Funding

This study was supported in part by the grants from the Key Technology R&D Program of Public Health in Tianjin (No.14KG141).

Conflict of interest statement

None declared.

Acknowledgements

The authors thank all the participants of this study.