Prognostic value of MICA/B in cancers: a systematic review and meta-analysis

Purpose MHC class I chain related-proteins A (MICA) and B (MICB) are natural killer group 2D ligands that mediate tumor surveillance. Several studies have suggested that MICA/B levels predict clinical outcomes in patients with cancer; however, this remains contentious. Here, we present a systematic review and meta-analysis of available studies of the prognostic value of MICA/B in cancer. Materials and Methods We searched PubMed, Embase, Clinicaltrials.gov, and Cochrane Library to identify studies published from inception to July 2017 that assessed MICA/B in patients with cancer. The hazard ratio (HR) and 95% confidence interval (CI) of MICA/B were extracted for overall survival (OS) analysis. Results A total of 19 studies comprising 2,588 patients with 10 different types of cancer were included in the study. Low sMICA/B levels were found associated with significantly longer OS (HR = 1.65, 95% CI [1.42–1.92], P < 0.00001). Patients with cancers of digestive system that exhibited high MICA/B expression had significantly longer OS in (HR = 0.56, 95% CI [0.39–0.80], P = 0.002) compared with those with lower MICA/B expression (I2 = 35%, P = 0.18). Conclusions Serum soluble MICA/B represents a potential prognostic marker in various human cancers. High cell-surface MICA/B expression in cancers of the digestive system was found associated with increased survival.


INTRODUCTION
Identifying reliable prognostic biomarkers for cancer patients is a longstanding oncology research goal. To date, the majority of cancer prognosis biomarker findings have focused on cancer cell centric biomarkers, such as cancer cell gene mutations, miRNA and lncRNA secretions, and relative progression protein expressions [1][2][3][4][5]. However, cancer evolution is also greatly influenced by the immune system and complex milieu, indicating that biomarker research should also consider the impact of factors originating from outside of the tumor cells [6].
The MHC class I chain related-protein A (MICA) and B (MICB) are cancer cell-surface molecules that reflect both cancer cell-centric biological behavior and host immune status. The major histocompatibility complex (MHC) class I chain related (MIC) gene family, which lies within the HLA region, consists of five members: MICA, MICB, MICC, MICD, and MICE. Of these five members, MICA and MICB are the only functional genes [7][8][9], which are frequently expressed by carcinomas of the breast, lung, colon, ovary, kidney, prostate, as well as in melanomas, gliomas, and leukemia [10,11]. MICA and MICB, as signals of cellular stress, engage with natural killer group 2D (NKG2D). This engagement actives the cytolytic responses of γδ T cells and NK cells against epithelial tumor cells [12].
In addition to the membrane-bound form, a soluble isoform of MICA/B (sMICA/B) is present in the serum. This serum-soluble form is derived from the proteolytic shedding of the membrane bound molecule. MICA present on the surface forms a complex with a disulfide isomerase/ Meta-Analysis chaperone (ERp5). This complex induces a conformational change, enabling proteolytic cleavage of MICA by a disintegrin and metalloproteinase (ADAM) proteins [13].Then, the interaction of sMICA/B with NKG2D results in the endocytosis and degradation of receptorligand complexes and also suppresses NKG2D-mediated host cancer rejection [14,15]. Therefore, MICA/B and sMICA/B, reflecting the cancer cell centric biological behavior and tumor immune surveillance status, could be a potential prognostic biomarker of cancer patients.
Indeed, this correlation between expression of MICA/B and survival has been reported in a wide range of human carcinomas, including colorectal-, hepatocellular-, pancreatic-, gastric-, and lung carcinoma as well as melanoma [16][17][18][19][20][21][22]. However, some studies also report that high MICA/B expression is associated with unfavorable outcomes in ovarian cancer, non-small cell lung carcinoma, and breast cancer [23]. Previous studies have been carried out across many types of cancer, with widely different sample sizes and outcomes. It is still ambiguous whether the prognostic potential of MICA/B is attributable to the biologic properties of a specific cancer type, whether the prognosis value of MICA/B relies on their soluble isoform, and whether their prognosis effect depends on methodology differences.
Based on current knowledge, low level of sMICA/B in serum and high expression of MICA/B are expected to be associated with less immune escape and stronger cytolytic ability of effector cells, respectively [24][25][26]. And serum soluble MICA/B and cancer cell-surface MICA/B are detected by two different methods. At the same time, these up-or down-regulated molecules on cancer cell surface or in patient serum are all associated with patients' survival. Therefore, we conducted a systematic review and two meta-analysises to establish pooled estimates for survival outcomes based on sMICA/B serum levels and MICA/B cell surface expression in different types of cancers.

RESULTS
A total of 2,588 patients were included in the analysis, with the sample sizes ranging from 48 to 222 between the included studies, and the published years between 2007 and 2016. Of 171 citations, we identified 19 articles (Figure 1) encompassing 10 different cancer types that met the inclusion criteria, including cervical cancer, ovarian cancer, gastric cancer, cholangiocarcinoma, pancreatic cancer, hepatocellular carcinoma, oral squamous cell carcinoma(OSCC), lung carcinoma, melanoma, and multiple myeloma (Tables  1 and 2). Studies of soluble MICA/B comprised a total of 1,482 patients with 7 different types of carcinoma: non-small cell lung, cervical, pancreatic, oral squamous cell, hepatocellular carcinoma, melanoma, and multiple myeloma [27][28][29][30][31][32]. The largest number of studies focused on melanoma (5 studies, 628 patients).In the 10 immunohistochemistry (IHC) studies, we included 1,183 patients with 7 different types of carcinoma: non-small cell lung, cervical, ovarian, gastric, cholangiocarcinoma, pancreatic, and hepatocellular carcinoma [17,19,27,33,34]. The largest number of studies focused on nonsmall cell lung carcinoma (3 studies, 413 patients). Tissue expression of MICA/B was detected by immunohistochemical staining, and all serum sMICA/B studies were performed using ELISA. The Newcastle-Ottawa scale (NOS) confirmed that most of the studies were of high quality (Table 3).

Prognostic effect of serum sMICA/B on OS in various cancer types
Thirteen studies comprising 1,482 patients indicated a correlation between serum sMICA/B and OS. Meta-analysis of all these studies revealed a significant association between serum sMICA/B and survival, with higher sMICA/B in serum associated with a significantly lower OS rate, with a pooled HR of 1.65 (95% CI [1.42-1.92], P < 0.00001). The fixed-effects model was adopted to verify the significance of heterogeneity (I 2 = 45%, P = 0.04; Figure 2). Subgroup analyses were performed to evaluate the effects of various clinical variables on pooled OS (Table 4).

Subgroup analysis of the prognostic effect of sMICA/B in different cancer subtypes
Subgroup analyses and funnel plot of bias were implemented according to cancer type. According to our current included studies, we divided the cancer into three subtypes: digestive system cancers (hepatocellular carcinoma and pancreatic cancer), melanoma, and OSCC. Another two studies in multiple myeloma and non-small cell lung carcinoma were excluded in this subgroup analysis. High sMICA and sMICB levels were associated with significantly longer OS in various cancer types, including melanoma (HR = 1.52, 95% CI

Subgroup analysis according to sMICA level and sMICB level
Significant differences were found between the high level and low level of sMICA/B for cancer patient survival prognosis. For sMICA, the result showed that the pooled HR = 1.60, 95% CI [1.37-1.87], P < 0.00001, with a heterogeneity analysis of I 2 = 47%, P = 0.04. For sMICB, HR = 2.86, 95% CI [1.49-5.51], P = 0.002 with heterogeneity analysis I 2 = 0%, P = 0.78 (Supplementary Figure 2A, 2B). These results suggested that, compared with sMICA, sMICB is a more reliable prognostic marker for patients with cancer.

Subgroup analysis according to multivariate correction and sample size
The association between sMICA/B and OS was significant in studies with multivariate correction (HR = 1.59,

The prognostic effect of MICA/B expression on OS in various cancer types
OS was reported in 10 studies of MICA/B expression in a total of 1,183 cancer patients. The meta-analysis of all these studies revealed a significant association between MICA/B expression and OS. No statistically significant association was observed between high MICA/B expression and longer OS (HR = 0.85, (95% CI [0.49-1.48], P = 0.58). The random-effects model was adopted to determine the significance of heterogeneity (I 2 = 84%, P < 0.00001; Figure 3). Table 5 shows the results of meta-analyses of OS for each subgroup analysis. We divided the cancer types into 3 categories: digestive system cancer (including:cholangiocarcinoma, gastric cancer, hepatocellular carcinoma, pancreatic cancer), respiration system cancer (including:lung adenocarcinoma, and nonsmall cell lung carcinoma), and gynecologic malignant tumors (including: ovarian cancer and cervical cancer). High MICA expression was associated with significantly longer OS in cancers of the digestive system (HR = 0.56, 95% CI [0.39-0.80], P = 0.002; I 2 = 35%), non-small cell lung carcinoma (HR = 2.06, 95% CI [0.86-4.94], P =

Subgroup analysis according to multivariate correction and sample size
We next investigated whether heterogeneity resulted from the difference in statistical methods used between studies. The results of multivariate adjusted analysis for OS were reported in 8 studies; however, in the other 2 studies, 1 study was data extrapolated and 1 was from a univariate analysis. High MICA expression was associated with significantly longer OS in the univariate analysis subgroup    Figure 8A, 8B).

Publication bias and sensitivity analysis
The publication bias of the present meta-analysis was evaluated by funnel plot. The shape of the funnel plot was almost symmetrical. To test the stability of metaanalysis of sMICA/B and OS, we performed a sensitivity analysis by sequentially removing each eligible study. And the present meta-sensitivity analysis did not suggest an undue influence of any single study (Figures 4, 5 and 6).

DISCUSSION
To our our knowledge, the present meta-analysis is the first to comprehensively evaluate the association of immune effector molecule expression (MICA/B) and serum shedding immune effector molecules (sMICA/B) with prognosis in variable types of cancer. Our metaanalysis reviewed a total of 19 studies comprising 2,588 patients to determine the prognositic value of MICA/B and sMICA/B in 10 types of cancer. The overall pooled analysis of all types of cancer suggested that a high level of soluble MICA/B in serum was associated with poor OS, with no significant difference between cancer types. With regards to MICA/B expression as detected by IHC, high MICA/B expression level was associated with longer survival in cancers of the digestive system. However, no statistical significance was observed for MICA/B expression level when all types of cancer were considered. A high level of sMICA/B in serum had a negative prognostic effect on cancer patients. This finding suggested that shedding of NKG2D ligands by cancer cells enable these cells to evade the immune response by escaping detection by NK cells, γδ T cells and CD8+ T cells. Conversely, a high MICA/B expression was associated with a favorable outcome in cancers of the digestive system, supporting the notion that MICA/B is a critical ligand in immune cytotoxicity.
In serum, we observed that sMICA/B was an independent prognosis factor in cancer patients, with moderate heterogeneity. There were no significant differences in various type of cancers. Further, no significant heterogeneity was observed between sMICA and sMICB. Soluble MICA/B was a more reliable prognostic indicator compared with MICA/B cell surface expression, as the former is easier to detect and the results are less affected by individual estimates. To detect MICA/B expression in cacner cell surface, biopsies are necessary; however, cancerous sites are not always reachable. Moreover, not all patients with cancer are tolerant of biopsy. These factors decrease the feasibility of MICA/B detection by IHC. Therefore, soluble MICA/B in serum is more applicable in such situations. Furthermore, the interpretation of immunohistochemical staining is influenced by individual subjective factors, and results are more difficult to quantify. Overall, sMICA/B are more advantaged at cancer patient survival prognosis.
Unlike previous studies that have focused on cancer cell-centric prognosis biomarkers, such as microRNA, lncRNA, or relative progression proteins, the present meta-analysis focused on biomarkers associated with immune cell interaction [3,35,36]. Immune effector molecules MICA/B not only reflect a distinct underlying biology of the tumor, but they are also involved in innate immune activation and cancer immune tolerance [6]. Of the adoptive immunotherapeutic strategies, MICA/B or soluble MICA/B could be a reflection of a pre-existing immune con-texture and of the immune response to NK cell-, T cell-, and γδ T cell-based immunotherapy. Studies of cytokine-induced killer cell (CIK) therapy have reported that cancer patients with high MICA expression experienced a significantly greater survival benefit from CIK treatment [19,21]. Therefore, prognostic markers MICA/B and sMICA/B may additionally serve as predictors of treatment outcomes in patients receiving immunotherapies.
No significant statistical difference was observed for the association between MICA/B expression and survival in all types of cancer. Subgroup analysis was conducted to explore the data further. We found that in respiratory and gynecologic malignant tumors, there was no evidence showing that MICA/B expression was associated with cancer outcome. However, the association between MICA/B expression and survival was significant for digestive system cancers (P = 0.002). The reasons underlying these inconsistent results may be manifold: it is speculated that in respiratory cancers and gynecologic malignant tumors, dedifferentiation results in the expression of MICA; however, in the absence of other pro-inflammatory signals, there was no infiltration of  immune effector cells. Therefore, MICA expression had no influence on cancer survival in these cancer types. This has been observed in ovarian cancer, where high expression of MICA/B significantly correlated with lower degree of NK cells intra-epithelial infiltration [37]. In our studies, IHC data indicated a large proportion of cytoplasmic MICA/B expression instead of cell surface staining in lung cancers and gynecologic malignant tumors. This phenomenon was also reported by another two studies in clear cell renal cell carcinoma and breast cancer [23,38]. However, a widely accepted theory is that MICA/B expressed on the cell surface, but not in the cytoplasm, interacts with NKG2D receptors on NK cells, γδ T cells, and T cells [39]. Alternatively, MICA/B expressed only within the cytoplasm results in the immune evasion by the cancer cell and limited survival for the patient, instead of cytoplasm.
There is no international unification measurement to determine levels of MICA/B, and MICA/B localisation is also not assessed in a standardized manner. In the included studies present in the present analysis, three major evaluation criteria were applied: the expression score was calculated by multiplying the intensity and positivity scores [27], or by adding these scores [19,21], and the third criterion was evaluated according to staining intensity [33]. The heterogeneity associated with the evaluation criteria and cut-off point made the metaanalysis of MICA/B expression by IHC challenging.
Further, the current meta-analysis has several potential limitations: firstly, various factors such as heterogeneity in MICA/B expression IHC scoring strategies, cut-off points, and cancer stages restricted us from obtaining more comprehensive results. As the prognostic role of MICA/B seems to be substantially different according to cancer site, the overall pooled analysis of all types of cancer may have been highly dependent on the relative proportion of each specific type of cancer. Furthermore, in several studies, HR for outcome measures were derived from Kaplan-Meier survival curves when not provided by the original studies directly, and a few of the studies provided an inconsistent HR, 95% CI, and P-value. These inconsistencies were considered to have affected the level of evidence. These limitations mandate caution when interpreting the present results.
In conclusion, despite the limitations described above, the current meta-analysis provides evidence that sMICA/B levels serve as a reliable prognostic marker in various cancers. The expression levels of MICA/B on cancer cell surface were significantly associated with OS in patients with digestive system cancer. In addition to prognostic utility, an improved understanding of MICA/B in various cancers types should enable the development of more precise and effective immunotherapies.

Search strategy and selection criteria
We conducted and reported this systematic review and meta-analysis in accordance with the PRISMA statement. We systematically searched PubMed, Embase, Clinicaltrails.gov, and Cochrane Library(up to July 15, 2017), without language restrictions, for human studies of the prognostic significance of MICA/B and sMICA/B in cancer . The following keywords were used for the search: ("Prognosis" OR "Mortality" OR "Survival") AND ("MICA" OR "MICB" OR "MIC" OR "MICA/B" OR "MICA-B" OR "MHC class I-related chain A" OR "MHC class I chain related gene A protein" OR "MHC class I chain-related protein B" OR "MHC class I chainrelated antigen B") AND ("Neoplasia" OR "Neoplasias" OR "Neoplasm" OR "Tumors" OR "Tumor" OR "Malignancy" OR "Malignancies" OR "Cancer" OR "Cancers" OR "Neoplasms"). We scrutinized the reference lists of the identified reports, reviews, meta-analyses, and other relevant publications to indentify additional pertinent studies.

Inclusion and exclusion criteria
Studies that met the following criteria were included in the meta-analysis: studies must have: (1) been published as original articles; (2) evaluated human subjects; (3) evaluated MICA, MICB or MICA/B in cancer surgical specimens using immunohistochemical method or any other method, or sMICA or sMICB in serum by ELISA and any other method; (4) reported association of high and low MICA/B expression with overall survival (OS); (5) reported association of high and low sMICA/B with OS; and (6) contained the minimum information necessary to estimate the effects (i.e., hazard ratio) and a corresponding measure of uncertainty (i.e., confidence interval, P-values, standard errors or variance). As an additional criterion, when a single population was reported in multiple reports, only the report with the most complete data was included to avoid duplication. Further, when a single report contained more than one unduplicated studies, all of the studies that met the inclusion criteria were included, and these studies were named by author and number. In this review, we additionally included studies that did not distinguish between MICA and MICB by immunohistochemistry, as the MICA and MICB locus encode the proteins of the same size with 83% similar amino acid sequence, and no distinct functions relating to cancer immune response have been found [40]. The eligibility of each study was assessed independently by two investigators (YJ.Z and NF.C). We excluded studies that were not published as full reports, such as conference abstracts and letters to editors, as well as studies that did not report sufficient data for the estimation of survival rates.

Assessment of risk of bias
We used the NOS to assess the risk of bias [41]. The NOS evaluates a high quality study in three domains: selection of participants, comparability of study groups, and the ascertainment of outcomes of interest. We considered studies that received a score of seven or more than that as low risk of bias, and those that scored less than seven as high risk of bias. This cut-off point was chosen according to the distribution of relative quality scores of all included studies.

Data extraction
Two investigators (YJ.Z and NF.C) independently summarized the studies meeting the inclusion criteria and performed data extraction using a predefined form, recording: author, year of publication, MICA or MICB or sMICA or sMICB, sample size, cancer type, median follow-up time, country, scoring protocols to identify MICA/B expression, cut-off point to identify sMICA/B, expression (High: Low), method of study, outcome of univariate and/or multivariate analysis (including P-values, hazard ratios [HR], and 95% confidence intervals [CI]).

Indices
The endpoint used in this meta-analysis was OS. For MICA/B, study-defined binary variables indicating either the presence (versus absence), positive (versus negative), or high (versus low) expression were used and described as "high" or "low" MICA/B expression. For sMICA/B in serum, study-defined binary variables indicating either more (versus less) or high (versus low) were also described as "high " or "low" level.

Statistical analyses
The present meta-analysis and statistical analyses were using Revman software (version 5.3; Cochrane Collaboration, Oxford, United Kingdom). HR and its 95% CI were used to estimate the association between MICA/B and prognosis. For sensitivity analysis, we used Stata software (version 14.2;STATA Corp., College Station, TX, USA). If results of both univariate and multivariate Cox regression analyses were reported, we used estimates from the multivariate Cox regression model for a more direct estimate of the effect of MICA after controlling for potential confounding variables. If a direct reported of HR and 95% CI were not available, survival data from Kaplan-Meier curves were read using Engauge Digitizer version 9.7 (http://digitizer.sourceforge.net/) as described previously. The calculation method for HR was based on Tierney's protocol [42,43]. This work was performed by two independent reviewers (YJ.Z and NF.C) to reduce inaccuracy in the extracted survival rates.
We assessed heterogeneity between studies with the I 2 statistic as a measure of the proportion of total variation in estimates due to heterogeneity, where I 2 values of 25%, 50%, and 75% corresponded to cut-off points for low, moderate, and high degrees of heterogeneity. Subgroup analyses were carried out to investigate potential sources of heterogeneity. Sensitive analysis conducted to assess whether conclusions were sensitive to restricted studies. Subgroups were defined according to different types of cancer, sample sizes, multivariate correction, and whether MICA or MICB were assessed together or individually. Tests for the effects of subgroup interaction were performed.