Prognostic Value of Cancer Stem Cell Markers in Head and Neck Squamous Cell Carcinoma: a Meta-analysis

Bmi-1, CD133, Nanog and Oct-4 have been reported as cancer stem cell (CSC) markers in head and neck squamous cell carcinoma (HNSCC). However, the prognostic value of them in HNSCC remains controversial. Hence, this meta-analysis was conducted to access the association between the four CSC markers and survival outcome of HNSCC patients. A total of 22 articles with 27 studies met the inclusion criteria and the combined hazard ratio (HR) and 95% confidence intervals (95% CI) were calculated. Data analysis showed that high expression of CSC markers was associated with poor overall survival (OS) (HR = 1.93; 95% CI: 1.46–2.55, P < 0.001) and disease free survival (DFS) (HR = 4.78; 95% CI: 2.95–7.75, P < 0.001) but not disease specific survival (DSS) (HR = 1.17; 95% CI: 0.74–1.84, P = 0.50) of HNSCC patients. Subgroup analysis indicted that high expression of CD133 (HR = 2.33, 95%CI: 1.42–3.83, P < 0.001), Oct-4(HR = 2.10, 95%CI: 1.36–3.22, P = 0.007) and Nanog (HR = 2.49, 95%CI: 1.66–3.72, P < 0.001) could predict poor OS in HNSCC patients respectively whereas overexpression of Bmi-1 was not related to the reduced OS in HNSCC patients (HR = 1.32, 95%CI: 0.66–2.65, P = 0.43). Therefore, we concluded that CSC markers, especially CD133, Nanog and Oct-4, might be predictive factors in HNSCC patients.


Discussion
This meta-analysis of 27 studies including 2143 patients assessing the prognostic value of 4 CSC markers in HNSCC showed that a high expression level of CSC markers was a promising prognostic factor for lower DFS and OS in HNSCC patients. However, the expression of CSC markers had no obvious influence on DSS of HNSCC patients. According to the current results, CSC markers may play an important role in the relapse of HNSCC rather than death from HNSCC. However, considering that the sample sizes about DSS and DFS are relatively limited, our results need to be cautiously interpreted.
Bmi-1 was related to multiple human cancers which played an indispensable role in maintaining the self-renewal ability of both normal and malignant cancer stem cells. It has been reported that high expression of Bmi-1 could significantly lead to a poor OS in gastric cancer patients 36 . But in our subgroup analyses, Bmi-1 could not influence the OS of HNSCC patients. However, poor DSS in HNSCC patients was significantly associated with high expression of Bmi-1 with no heterogeneity (I 2 = 0%, P h = 0.47). It is worth noting that there was high heterogeneity between OS and HNSCC (I 2 = 81%, P h < 0.001). We tried to find what caused this. When we ruled out the study of Chen 2013 15 , the association of poor OS and high expression of Bmi-1 was changed to be significant, meanwhile the heterogeneity was lower than before. This results suggested that we need more studies to get a conclusion whether the expression of Bmi-1 can influence the OS of HNSCC patients.
CD133, also named prominin-1, has been verified to be a CSC marker in many cancers. Previous meta-analyses showed that high expression of CD133 was responsible for the reduced OS of ovarian cancer, gastric cancer, non-small cell lung cancer and hepatocellular carcinoma patients [37][38][39][40] . Our subgroup analysis revealed that a high expression of CD133 was related to a lower OS in HNSCC patients which was in keeping with those previous analyses.
Nanog and Oct-4, also known as embryonic stem cells markers, both have the ability in maintaining the self-renewal capacity in embryonic stem cells 28 . Moreover, studies have showed that the co-expression of Oct-4 and Nanog could be found in hepatocellular carcinoma and lung adenocarcinoma 41,42 . Our results suggested that the high expression of Nanog and Oct-4 could reduce the OS of HNSCC patients, respectively. In the research of Chiou et al. 16 , patients with high expression of both Nanog and Oct-4 were associated with lower OS than those with high expression of Nanog or Oct-4 alone 16 . This reminded us that more studies about the relationship of co-expression of Nanog and Oct-4 and the survival outcome of HNSCC should be conducted.
In the subgroup analysis of tumor location, poor OS was related to CSC markers in OSCC, LSCC and NEPC except ESCC. Interestingly, when we tried to find the origin of the heterogeneity in CSC markers and ESCC, we found that, as same as the result of the association of Bmi-1 and OS, the heterogeneity was lower and the result was significantly changed (HR = 2.03, 95%CI: 1.47-2.82, P < 0.001, I 2 = 10, P h = 0.34) if we discarded the study of Chen 2013 15 . Taking this into consideration, we could not simply conclude that poor OS was not related to high expression of CSC markers in ESCC because the number of included studies was a little small and we needed more powerful evidences.

Figure 3. Forest plot of the association between CSC markers and DSS (A)/DFS (B) in patients with HNSCC.
Scientific RepoRts | 7:43008 | DOI: 10.1038/srep43008 Considering the subgroup of ethnicity, the OS of the Asian and Caucasian was the same to the overall OS. The DSS of the Asian was consistent with the overall DSS while the DSS of the Caucasian was changed to be significant without heterogeneity. In the subgroup analysis of sample size, we used the median as the boundary because most of the sample size of our included studies were relatively small. As showed in Table 2, poor OS of large sample size was not significantly associated with high expression of CSC markers with high heterogeneity because Chen 2013 15 was in this group. Without study of Chen 2013 15 , the results would be changed (HR = 2.09, 95%CI: 1.68-2.60, P < 0.001, I 2 = 30, P h = 0.10).
This meta-analysis has the following limitations that must be taken into consideration. First, both the number of included studies about each CSC markers and the number of included HNSCC patients in each study are relatively small, which, to some extent, may reduce the power and precision of our subcategory analyses. Second, most of the included studies were conducted in Asian and a few studies were about the Caucasians but no studies were about the black populations, which may produce potential population selection bias. Third, nonuniform cut-off value defining high and low expression of CSC markers may impact the results of this meta-analysis. Despite these limitations, we provided a comprehensive analysis of the association between CSC markers and OS/ DFS/DSS of HNSCC patients. To the best of our knowledge, this meta-analysis is the first to systematically assess the association of Bmi-1, Oct-4, Nanog and CD133 expression with survival outcome of HNSCC. So far, there are some meta-analyses showed that high expression of CSC markers, including ALDH1, CD44 and SOX2, could predict poor OS/DFS in head and neck cancer patients [10][11][12] . Our results were in line with these previous analyses.
In summary, our meta-analysis revealed that high expression of CSC markers was significantly associated with poor OS and DFS but not DSS of HNSCC patients. However, because of certain limitations, different subgroup showed to some extent inconsistent results, which prompted future large-sample, well-designed with long-term follow-up to confirm and update the findings of this. Nevertheless, our study still gave some hints that CSC markers have prognostic value in HNSCC patients.   and neck squamous cell carcinoma OR HNSCC OR ((oral OR laryn* OR pharyn* OR tongue OR oropharyn* OR nasopharyn* OR hypopharyn* OR trachea OR laryngopharyn* OR cervical tracheal OR cervical esophagus) AND (cancer* OR tumor* OR carcinoma* OR neoplasm*))). Two reviewers (F. Z. N and L. M. X) inspected all candidate articles independently. Discrepancies were resolved in consensus. The inclusion criteria were: 1) the diagnosis of HNSCC was made based on pathological examination; 2) the expression of CD133 or Bmi-1 or Oct-4 or Nanog with OS/DSS/DFS about HNSCC was reported; 3) HRs and 95% CIs were provided in text or sufficient data was provided for the calculation of HRs and 95% CIs; 4) articles published as original research. In order to avoid duplicate inclusion of data, we selected only the more recent or complete article when multiple reports described the same population.
Data extraction. Two reviewers (C. X. B and W. J) independently extracted the following information from included studies: author, year of publication, study country, sample size, clinicopathological parameters, cut-off value of CD133 or Bmi-1 or Oct-4 or Nanog, survival data and the tumor location. Disagreements were resolved by discussion.
Quality assessment. Two reviewers (L. X. Y and W. H. F) independently assessed the quality of included studies by the NOS 35 . Studies with NOS scores of more than 7 were defined as high quality. Consensus was reached by discussion when there were inconsistent results.
Statistical Analysis. Hazard ratio (HR) was used as a summary statistic for survival outcomes as described by Parmar et al. 43 . An HR greater than 1 represented poor prognosis in HNSCC. Heterogeneity among primary studies was assessed by the Cochrane's Q statistic and I 2 statistic. Cochran Q test' P value < 0.10 or I 2 > 50% indicated large heterogeneity between studies and we used random effects models to calculate the pooled HR and 95% CI. Otherwise, the fixed effects model was used. We used the median as the boundary between the large and small sample size. Studies with sample size >72 were regarded as large sample size, otherwise was considered as small sample studies. Subgroup analyses stratified by different stem cell marker, ethnicity, sample size and tumor location were carried out. Sensitivity analysis were applied to high-quality studies (NOS > 6). The Begg's funnel plots were used to evaluate publication bias. All statistical analysis were performed by using Review manager 5.3 software (Cochrane Collaboration, Oxford, UK) and Stata 12.0 statistical software (Stata Corporation, College Station, TX, USA). A two-tailed P < 0.05 was considered statistically significant.