Stromal Myofibroblasts Are Associated with Poor Prognosis in Solid Cancers: A Meta-Analysis of Published Studies

Objective Published studies have evaluated the impact of stromal myofibroblasts on prognosis in solid cancers. However, the results of these studies remain controversial. We therefore performed a meta-analysis to address this issue. Methods The PubMed, ISI Web of Science and Embase databases were searched through November 30th, 2015 by two investigators, and a total of 17 studies that contained 2606 patients were included. Stromal myofibroblasts were quantified in solid cancers using α-smooth muscle actin staining. Pooled Odds Ratio with 95% Confidence Intervals were calculated, and publication bias was analyzed. Results The results of this study suggest that in solid cancers, a high density of stromal myofibroblasts is significantly associated with poor 3- and 5-year overall survival (pooled odds ratio (95% confidence interval): 1.33 (1.10–1.60) for 3-year overall survival and 1.68 (1.22–2.32) for 5-year overall survival). In addition, a high density of stromal myofibroblasts also predicted poor 3- and 5-year disease-free survival (1.30 (1.05–1.60) for 3-year disease-free survival and 1.36 (1.01–1.83) for 5-year disease-free survival). However, stromal myofibroblasts were not associated with 3- and 5-year cancer-specific survival. No publication bias was found for all analyses. Conclusions The results of this study suggest that a high density of stromal myofibroblasts is associated with poor survival in solid cancers. More studies were required to investigate the prognostic value of stromal myofibroblasts in different types of solid cancers.


Introduction
Solid cancers, such as gastric, liver and lung cancers, have become the leading cause of death around the world [1,2]. Despite the enormous advances that have been made in surgical and radiochemotherapeutic treatments, the prognosis for solid cancer patients remains unfavorable [2]. It has been acknowledged that local recurrence and distant metastasis are the main reasons for the poor prognosis in these patients [3,4]. Therefore, developing methods to efficiently identify patients who are at high risk of a poor prognosis is critical.
In solid cancers, the prognosis is determined not only by the oncological characteristics of the cancer cells but also by the micro-environment of the tumor [5,6]. The tumor micro-environment provides essential nutrients for tumor growth, inhibits immune surveillance against cancer cells and induces cancer cells to take on a more malignant phenotype [5,6]. Targeting the tumor stroma has therefore shown great potential as a cancer treatment [7,8].
The TNM staging system has been widely used to divide solid cancers into early or advanced stages. In this system, the staging is determined by the depth of cancer invasion, regional lymph node involvement and distant metastasis [9]. The TNM staging system undoubtedly provides valuable prognostic information for solid tumors and is also useful for determining the optimal treatment strategy to implement after surgery. However, it has been reported that cancers with the same TNM stage can have distinct prognoses. Thus, the TNM staging system is incompletely adequate for staging cancers [10]. The TNM staging system is partially limited by the fact that it relies on determinations related to the biological characteristics of the tumor cells but ignores the tumor environment. Hence, it is important that we identify supplementary markers that are specific to the cancer environment to assist in predicting prognoses in solid cancers [11].
The cancer stroma is a highly heterogeneous structure that is composed of activated fibroblasts, fibroblast-produced extracellular matrix, inflammatory cells (such as macrophages) and capillaries. Cancer-associated fibroblasts are the main component of the tumor stroma, and these have been paid a large amount of attention because of the prominent roles they play in cancer development, progression and metastasis [5,12]. Cancerassociated fibroblasts are high heterogeneous, and they can originate from residual fibroblasts, vascular smooth muscle cells, endothelial cells and pericytes. Fibroblasts in the tumor stroma transdifferentiate into myofibroblasts that exclusively express α-smooth muscle actin. The myofibroblasts in a cancer stroma represent a subgroup of cancer-associated fibroblasts [13]. It is widely accepted that myofibroblasts have a significant impact on wound healing because they affect extracellular matrix remodeling, secrete growth factors and enhance angiogenesis. Cancer lesions are similar to "unhealed wounds" [14], and myofibroblasts in the cancer stroma express a variety of growth factors and inflammatory chemokines that are involved in the remodeling of the tumor stroma, the regulation of the motility of cancer cells and the induction of tumor cells toward phenotypes that are more resistant to chemotherapy [15].
Because myofibroblasts play significant roles in the cancer stroma and during cancer progression and metastasis, they are viewed as good predictors of cancer prognosis and therefore potential targets for cancer treatments [16][17][18]. Many published studies have evaluated the impact of stromal myofibroblasts on prognoses in solid cancers by using α-smooth muscle actin as a molecular marker. Some studies have reported that stromal myofibroblasts are associated with a poor prognosis in solid cancers [19,20]. However, other reports have not come to the same conclusions [21,22]. The prognostic value of stromal myofibroblasts in solid cancers therefore remains unclear. Hence, in this study, we searched for relevant published studies and performed a meta-analysis to address this issue.

Materials and Methods
This meta-analysis was conducted according to the guidelines of the preferred reporting items for systematic reviews and meta-analyses statement (S1 PRISMA Checklist) [23].

Literature searches
Two investigators (Liu Liu and Lin Liu) searched the PubMed, ISI Web of Science and Embase electronic databases through November 30 th , 2015. The search terms were: (myofibroblast OR (alpha smooth muscle actin)) AND (cancer OR tumor OR carcinoma) AND (survival OR prognosis OR prognostic). The reference lists of included studies and relevant reviews that were published during the past five years were screened to identify additional publications.

Literature selection
The inclusion criteria for this study were: (1) the study investigated the association between stromal myofibroblasts that were positive for α-smooth muscle actin staining and prognoses in solid cancers, such as gastric, colorectal and breast cancer; (2) the expression of α-smooth muscle actin was assessed in the tumor stroma using immunohistochemistry; (3) the article provided sufficient data to obtain an estimated odds risk and a 95% confidence interval; and (4) the article was written in English.
The exclusion criteria applied in this study were: (1) the article reported duplicated data (if two or more articles used the same data, the latest published article was included); (2) the expression of stromal α-smooth muscle actin was determined using other methods, such as reverse transcription-polymerase chain reaction; (3) the article was reported in a non-English language; (4) the report lacked enough information for a combined analysis; and (5) the report was a review, comment or letter.

Endpoints of interest
The primary endpoints in this meta-analysis were 3-or 5-year overall survival, disease-free survival and cancer-specific survival. Patients were classified into positive-or negative-myofibroblast groups according to the cut-off values that were used for stromal α-smooth muscle actin staining in each included study.

Data extraction and collection
Two authors (Liu Liu and Lin Liu) extracted the data from the included studies using a predefined form. The following data were extracted: the name of the first author, the publication year, the country of origin of the included patients, the sample size, the type of cancer, the follow-up period(s), the endpoint(s) of interest, the definition of positively or negatively labeled myofibroblasts, and the number of patients the in positively and negatively labeled myofibroblast groups and their 3-and 5-year overall survival, disease-fress survival and cancer-specific survival. From the published studies [24,25], survival data were extracted from tables or using Kaplan-Meier curves using a digitizing software tool (Engauge Digitizer version 4.1), which converted graphs into data for both groups. One study reported the hazard ratio and its 95% confidence interval for 3-year overall survival and disease-fres survival, and we used these data directly in the subsequent combined analysis [26]. Disagreements regarding extracted data were resolved by consensus.

Quality assessment
The quality of the included studies were independently assessed by two authors [Liu Liu and Lin Liu] using the scale described by Chen H et al [25]. The scale contained 12 items, which were categorized into the five following dimensions: patient features, ascertainment of the cancer, sample size, immunohistochemistry examination and follow-up. The scores on the scale can range from 0 to 10, with higher scores indicating better quality.

Statistical analysis
The odds ratio and its 95% confidence interval were used to present differences in cancer prognosis between positive and negative myofibroblast groups, and P<0.05 was defined as indicating a significant difference. Subgroup analyses were performed by separating the data according to different types of cancers. Between-study heterogeneity was evaluated using the Q test and I 2 test, and P<0.10 or I 2 >50% was defined as suggesting the presence of between-study heterogeneity. A random-effects model was applied using the Der-Simonian and Laird method for the combined analyses because performing that analysis with a random-effects model resulted in a more conservative estimate than performing it using a fixed-effects model. Publication bias was assessed using funnel plots and Egger's test. Visual asymmetry in a funnel plot or P<0.05 in an Egger's test were defined as indicating the presence of publication bias among the included studies.
All statistical analyses were conducted using STATA 10.0 (StataCorp, College Station, TX). All statistical tests were two-sided.
Positive myofibroblasts: a score higher than 2, and negative myofibroblasts: a score between 0 and 2. All of the included studies used α-smooth muscle actin as a specific marker for myofibroblasts in the tumor stroma. The cutoff values for the positive myofibroblast patients depended on the immunohistochemistry staining score and the method that was used in each included study (Table 2). In addition, three studies divided the samples according to their staining intensity for α-smooth muscle actin into low, medium and rich groups [11,16,32]. We therefore grouped the medium and rich levels together into a "positive myofibroblast group", whereas a low level of α-smooth muscle actin staining defined the "negative myofibroblast group".

The impact of stromal myofibroblasts on disease-free survival
Eight studies [17,18,20,21,26,28,31,34] were used in the combined analysis of 3-year diseasefree survival. The estimated odd ratios and its 95%confidence interval were 1.30 (1.05-1.60) with P = 0.016 (Fig 3a). There was no heterogeneity among the included studies (I 2 = 0% and P value for the Q test = 0.764). Six studies [17,18,20,28,31,34] were used in the combined analysis of 5-year disease free survival. The estimated odds ratio and its 95%confidence intervals were 1.36 (1.01-1.83) with P = 0.041 (Fig 3b). There was no heterogeneity among the included studies (I 2 = 0% and P value for the Q test = 0.436). These results suggest that α-smooth muscle actin-labeled stromal myofibroblasts are associated with poor 3-and 5-year disease-free survival in patients with solid cancers.
A subgroup analyses was performed for 3-year disease-free survival for breast cancer (2 studies) and esophageal squamous cell carcinoma (2 studies). The estimated odors ratio and 95%confidence intervals were 1.25 (0.70-2.24) for breast cancer (P = 0.453) and 1.43 (0.92-2.22) for esophageal squamous cell carcinoma (P = 0.109). Subgroup analyses were performed for 5-year disease-free survival for breast cancer (2 studies), and the estimated odds ratio and its 95%confidence interval was 1.33 (0.74-2.40) with P = 0.342. The impact of stromal myofibroblasts on cancer-specific survival Four studies containing five groups of patients [11,16,19,22] were included in the analysis of 3and 5-year cancer-specific survival. The estimated odds ratio and 95% confidence interval for 3-year cancer-specific survival were 1.21 (0.89-1.64) with P = 0.229 (Fig 4a). There was no obvious heterogeneity among the included studies (I 2 = 46.4% and P value for the Q test = 0.113). The estimated odds ratio and 95% confidence interval for 5-year cancer-specific survival were 1.31 (0.90-1.91) with P = 0.155 (Fig 4b). Between-study heterogeneity was found between the included studies (I 2 = 60.8% and P value for the Q test = 0.037).
A subgroup analysis was conducted for oral squamous cell carcinoma (three studies). The estimated odds ratio and 95% confidence interval for 3-year CSS were 1.47 (0.99-2.18) with P = 0.054, and the odds ratio and 95% confidence interval for 5-year cancer-specific survival were 1.68 (1.06-2.65) with P = 0.027.

Assessment of publication bias
Publication bias was assessed for overall survival, disease-free survival and cancer-specific survival. Using funnel plots and Egger's tests, we found that there was no significant publication bias in the analyses of overall survival, disease-free survival and cancer-specific survival (P = 0.09 and 0.818 for Egger's tests of 3-and 5-year overall survival, respectively; P = 0.413 and 0.626 for Egger's test of 3-and 5-year disease-free survival, respectively; and P = 0.832 and 0.503 for Egger's test of 3-and 5-year cancer-specific survival, respectively) (Fig 5A-5F).

Discussion
The TNM staging system is based on the biological features of cancer cells and is used as a foundation for categorizing patients with solid cancers into those with early and advanced stage cancer, which is then used to determine the most reasonable treatment strategy. However, the TNM system is inadequate because patients with same TNM stage can have different prognoses. Introducing a parameter that incorporates information about the tumor microenvironment would significantly supplement the TNM staging system and would be helpful when determining personalized treatment strategies in these patients. To our knowledge, this is the first study to perform a meta-analysis evaluating the association between α-smooth muscle actin-labeled stromal myofibroblasts and prognoses in solid cancers. The results of this study suggest that the abundant presence of stromal myofibroblasts in the cancer stroma is associated not only with poor overall survival but also with unfavorable disease-free survival.
Cancer-associated fibroblasts have a large, plump, and spindle-shaped morphology. Similar to mesenchymal stem cells, cancer-associated fibroblasts have a remarkable capacity to transdifferentiate into cartilage cells and bone cells [35]. Cancer-associated fibroblasts are activated by an increase in the expression of α-smooth muscle actin, which causes them to transdifferentiate into myofibroblasts when they are exposed to inflammatory cytokines from cancer cells [13,36]. Moreover, cancer-associated fibroblasts secrete a number of growth factors and inflammatory chemokines that stimulate proliferation in cancer cells, enhance angiogenesis and epithelial-mesenchymal transition, and eventually accelerate cancer growth in addition to local and distant metastasis [36,37]. Myofibroblasts in the cancer stroma are regarded as a subgroup of cancer-associated fibroblasts [13]. Many experimental studies have suggested that similar to cancer-associated fibroblasts, myofibroblasts promote cancer progression and metastasis by expressing high levels of inflammatory factors and chemokines, such as interleukin-6 and C-X-C motif chemokine [38,39].
An increasing amount of evidence has shown that stromal myofibroblasts promote cancer progression, and this has pushed researchers to investigate whether myofibroblasts can be used as a prognostic marker for solid cancers. Moreover, if stromal myofibroblasts are a main component of the tumor stroma, are they a potential target for cancer treatments? To answer this question, many clinical studies have been conducted. However, no conclusive answer has yet been reached. Ha SY et al [28] assessed α-smooth muscle actin expression in stromal myofibroblasts in 116 cases of esophageal squamous cell carcinoma, and their results suggested that stromal α-smooth muscle actin was expressed at higher levels in larger esophageal squamous cell carcinomas and in advanced T-stage and N-stage esophageal squamous cell carcinomas. In addition, esophageal squamous cell carcinoma patients with higher expression levels of stromal α-smooth muscle actin had lower 5-year overall survival and disease-free survival than patients with lower α-smooth muscle actin expression [28]. Similarly, some studies have reported that stromal fibroblasts are associated with a high risk of recurrence and poor prognosis in other types of solid cancers, such as breast, colorectal and gastric cancer [17,18,33]. Furthermore, targeting stromal myofibroblasts suppressed growth in cholangiocarcinomas and improved host survival in an experimental study [12]. Our study evaluated the impact of stromal myofibroblasts on prognoses in solid cancers. The results suggest that stromal myofibroblasts lead not only to poor overall survival but also to unfavorable disease-free survival. In addition, the results of a stratified analysis also suggested that a high density of stromal myofibroblasts is associated with shorter cancer-specific survial in OSCC. Although the impact of stromal myofibroblasts on poor survival in solid cancers has been described in many studies, some authors do not support the existence of this relationship. Ayala et al reported that in prostate cancer, a low density of stromal myofibroblasts was more correlated with shorter disease-free survival than a high density of cancer-associated fibroblasts [34]. Similarly, Wang WQ showed that αsmooth muscle actin-labeled cancer-associated fibroblasts were not associated with overall survival or disease-free survival in either hepatic or pancreatic cancer [21]. These inconsistent results might be caused by differences in research methods or the number of patients included in the study. Therefore, studies including larger sample size are needed in the future.
This study is meaningful because our data suggest that stromal myofibroblasts are an effective marker for predicting prognoses in patients with solid cancers. Moreover, unlike current chemotherapies that target tumor cells, a therapy that targets myofibroblasts would be a novel avenue for research in cancer therapies in the future. We should mention that there were some limitations to this study. First, the included studies were retrospective, which means they were susceptible to some bias. Second, the sample sizes of the included studies were relatively small. Finally, there was heterogeneity among the included, and this might impair the accuracy of their pooled estimates. To overcome these shortages, we used a random-effects model rather than a fixed-effects model because a random-effect model is more conservative for a combined analyses.
In summary, this study suggests that a high density of stromal myofibroblasts, which were identified using α-smooth muscle actin as a marker, contributed to poor survival in patients with solid cancers. These data could therefore be used to identify high-risk patients who may need more intense therapy.
Supporting Information S1 File.