Homeobox proteins are potential biomarkers and therapeutic targets in gastric cancer: a systematic review and meta-analysis

Background An increasing number of studies have described the aberrant expression of homeobox (HOX) proteins in gastric cancer (GC), which is critically associated with the prognosis and clinicopathological characteristics of GC. This study was conducted to investigate the clinical value and action mechanisms of HOX proteins in GC. Methods A comprehensive search of PubMed, Embase, Web of Science and Cochrane Library was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement. The pooled hazard ratio (HR) with its 95% confidence interval (95% CI) and the pooled odds ratio (OR) with its 95% CI were used to assess the effect of HOX protein expression on the prognosis and clinicopathological features of GC, respectively. Results Nineteen studies containing 3775 patients were selected for this study. Heterogeneity among HRs of overall survival (OS) was markedly high (I2 = 90.5%, p = 0.000). According to the subgroup analysis, increased expression of HOX protein in the downregulated subgroup was associated with a good prognosis for patients with GC (pooled HR: 0.46, 95% CI: 0.36–0.59, I2 = 3.1%, p = 0.377), while overexpression of HOX protein in the upregulated subgroup was correlated with a reduced OS (pooled HR: 2.59, 95% CI: 1.79–3.74, I2 = 73.5%, p = 0.000). The aberrant expression of HOX protein was crucially related to the TNM stage, depth of tumour invasion, tumour size, lymph node metastasis, distant metastasis, vascular invasion, histological differentiation and Lauren classification in patients with GC. In addition, the molecular mechanisms by which HOX proteins regulate tumorigenesis and development of GC were also explored. Conclusions HOX proteins play vital roles in GC progression, which might serve as prognostic markers and therapeutic targets for GC.


Background
Gastric cancer (GC) is one of the most common cancers. Although the incidence of GC is decreasing, it remains the sixth most common malignancy and accounts for 8.2% of global cancer-related deaths, according to the most recent global cancer statistics reported in 2018 [1]. GC is highly heterogeneous, and patients with GC are usually diagnosed in an advanced stage. Despite significant developments in surgical techniques and adjuvant therapy, the overall prognosis of patients with GC is still poor. Therefore, it is urgent to identify molecular markers to predict the prognosis and evaluate the therapeutic effect of GC.
Homeobox (HOX) genes encode a highly conserved family of 39 transcription factors that are divided into four clusters (HOXA, HOXB, HOXC, and HOXD). HOX proteins play crucial roles in the early development of embryos and organs, including vertebrae, external genitalia, gastrointestinal tract and central nervous system [2]. Because embryogenesis and tumorigenesis share similar characteristics such as growth and differentiation, the deregulation of HOX protein has been observed in abnormal development and malignancy [3]. HOX proteins function as oncogenes or tumour suppressors, depending on the context [4]. An increasing number of HOX proteins have been investigated in various tumours, including acute myeloid leukaemia [5], breast cancer [6], lung cancer [7], and digestive tract neoplasms [8][9][10]. Currently, the implications of HOX proteins in tumorigenesis and development of GC have been reported in many studies. Nevertheless, the roles of HOX proteins in GC remain controversial.
Therefore, considering the conflicting conclusions of current researches, we conducted this systematic review and meta-analysis to explore the prognostic and clinicopathological value of HOX proteins in GC and summarized the molecular mechanisms by which HOX proteins regulate tumorigenesis and development of GC.

Methods
This study was conducted according to the PRISMA guidelines [11].

Selection criteria
All authors jointly deliberated and set the selection criteria. The following inclusion criteria were established: (1) GC was pathologically and histologically confirmed; (2) HOX protein expression was detected using immunohistochemical (IHC) staining in GC tissues and paired noncancerous mucosae; (3) studies evaluated the correlation between HOX protein expression and the outcome of GC; and (4) studies supplied sufficient information for calculating the HR with its 95% CI for survival and the OR with its 95% CI for clinicopathological parameters. The following exclusion criteria were used: (1) overlapping or duplicate data; (2) reviews, letters, case reports, conference abstracts, retracted articles, editorials, and full texts not published in English; (3) studies of cancer cells or animal models, or irrelevant studies; and (4) studies without adequate information for calculating HRs and 95% CIs.

Quality assessment
Two researchers (YM and JW) assessed the quality of studies using the Newcastle-Ottawa Quality Assessment Scale (NOS) [12]. The NOS consists of three aspects of evaluation: selection, comparability and outcomes between the case group and the control group. Studies that scored ≥6 points were considered high quality. Any disagreement was resolved by discussion.

Data extraction
Two investigators (XJ and LD) independently reviewed all included studies and extracted the available data. The following information was collected: (1) publication details, including the first author's name, publication year and origin of the studied population; (2) characteristics of the studied population, including HRs with 95% CIs and clinicopathological features. In the studies that reported HRs from both univariate and multivariate models, we extracted the HR from the latter model that had been adjusted for potential confounders. If the HR was not reported, it was extrapolated using Engauge Digitizer v.11.1 software, Tierney's spreadsheet [13], and (3) a cut-off value.

Statistical analysis
HR and its 95% CI from each study were used to calculate the pooled HR and its 95% CI. The heterogeneity of the combined HR was determined using Cochran's Q test and Higgins' I-squared statistics. A p value< 0.05 was considered significant. We used the random effects model if heterogeneity was observed (I 2 ≥ 50%). The fixed effects model was used in the absence of heterogeneity (I 2 <50%) [14]. A subgroup analysis was conducted based on the expression level of HOX proteins in patients with GC. The sensitivity analysis was conducted to evaluate the stability of the results, after excluding each study. Publication bias was assessed using the Begg's funnel plots and Egger's tests, and a p value< 0.05 was considered significant. Statistical analyses were performed using Stata statistical software (version 15.0).

Literature search
Our search strategy preliminarily identified 329 potential records. One hundred seventy-three articles remained after the removal of duplicated studies. Forty-eight of these studies were removed after perusing the titles and abstracts. Then, reviews, editorials, letters, conference abstracts, retracted articles, full texts not published in English, and studies of cancer cells or animal models were excluded. Subsequently, 18 studies lacking insufficient data were rejected. Finally, 19 studies including 3775 patients with GC were included in this analysis. The selection process is shown in Fig. 1.

Correlation of HOX protein expression with the prognosis
This meta-analysis included a total of 19 articles containing 14 HOX proteins. HOXB9, HOXD10 and HOXA5 were expressed at low levels in GC and acted as tumour suppressors. In contrast, HOXA13, HOXC6, HOXB7, HOXA1, HOXC9, HOXC10, HOXD4, HOXA11 and HOXD9 were expressed at high levels and functioned as tumour promotors in patients with GC. In addition, HOXA10 expression was increased in GC, but its role in predicting the prognosis of GC was unclear. In a pooled analysis including all studies with data on the prognostic effects of HOX proteins in GC, considerable heterogeneity among pooled HRs for OS was observed. A subgroup analysis stratified by the expression level was performed, and the results revealed different trends between the downregulated subgroup and the upregulated subgroup. High expression of HOX proteins in the downregulated subgroup was associated with a good prognosis for patients with GC (pooled HR: 0.46, 95% CI: 0.36-0.59, I 2 = 3.1%, p = 0.377), while the overexpression of HOX proteins in the upregulated subgroup was correlated with a poor OS (pooled HR: 2.59, 95% CI: 1.79-3.74, I 2 = 73.5%, p = 0.000) (Fig. 2a). The explanation for the high level of heterogeneity of the upregulated subgroup might be that HOXA10 had different prognostic values in the existing studies. The result of the analysis of the upregulated subgroup after excluding HOXA10 suggested that overexpressed HOX proteins significantly indicated a poor prognosis (pooled HR = 3.03, 95% CI: 2.45-3.74, I 2 = 16.5%, p = 0.283) (Fig. 3). DFS was reported in 6 studies involving 5 HOX proteins.

Correlation of HOX protein expression with clinicopathological characteristics
Seventeen studies with 2899 patients were included to detect the relationship between HOX protein expression and tumour stage. As shown in Fig. 4a, increased expression of HOXB9 and HOXD10 was significantly correlated with an earlier TNM stage (HOXB9: OR = 0.22, 95% CI: 0.12-0.41, HOXD10: OR = 0.21, 95% CI: 0.14-0.31), while increased expression of HOXA13, HOXB7, HOXA1, HOXA9, HOXC9, HOXC10, HOXA11 and HOXD9 was notably associated with an advanced TNM stage (I 2 = 92.6%, p = 0.000). Due to the high level of heterogeneity, we performed a subgroup analysis based on the expression levels of HOX proteins. The heterogeneity of the upregulated group was decreased but still at a high level (I 2 = 75.8%, p = 0.000) (Fig. 4b). A subsequent analysis showed that the studies of HOXA10 contributed a considerable amount of heterogeneity (data not shown). In addition, the difference of scoring systems for assessing expression levels of HOX proteins in the included studies was also one of the main sources of heterogeneity. The pooled analysis of the relationship between HOX proteins and the depth of tumour invasion showed that HOXD10 indicated a low T category (HOXD10: OR = 0.20, 95% CI: 0.09-0.41), while HOXA13, HOXC6, HOXB7 and HOXA1 were related to a high T category (HOXA13 (2013) (Fig. 5b). Similarly, the heterogeneity was significantly reduced by conducting a subgroup analysis according to expression levels of HOX proteins (Fig. 5c). Sixteen  (Fig. 6b). In addition, the upregulation of HOXB7 promoted vascular invasion in patients with GC (HOXB7 (2017): OR = 5.12, 95% CI: 3.18-8.23) (Fig. 6c). Moreover, HOXB9, HOXD10, HOXA5 and HOXC9 were factors contributing to good or moderate histological differentiation (HOXB9: OR = 0.17, 95% CI: 0.09-0.33, HOXD10:  (Fig. 7b). None of the HOX proteins included in the pooled analysis exhibited significant associations with age ( Fig. 8a), sex (Fig. 8b) or tumour location (Fig. 8c). Additionally, the relationships between HOXA5, HOXA10, HOXA13 and HOXB7 expression and clinicopathological characteristics were all explored in several studies. As shown in Fig. 9, HOXA5 expression predicted a smaller tumour size (OR = 0.22, 95% CI: 0.10-0.45) (Fig. 9a), but there is no correlation between HOXA10 expression and clinicopathological features (Fig. 9b). The overexpression of both HOXA13 (Fig. 9c) and HOXB7 (Fig. 9d)

Sensitivity analysis
A sensitivity analysis was performed to verify the robustness of our results. As shown in Fig. 10, the pooled HR was not significantly altered when each study was removed, which confirmed the reliability of overall results for the OS of patients with GC.

Publication bias
Begg's test and Egger's test were performed to evaluate publication bias. The results did not reveal substantial publication bias (Fig. 11: Begg's test: p = 0.576, Egger's test: p = 0.166).

Discussion
GC is a main cause of cancer-related mortality. Currently, radical gastrectomy combined with adjuvant chemotherapy is recognized as the most effective treatment for GC. Nevertheless, many patients with GC are usually diagnosed in an advanced stage, missing the opportunity for radical surgical resection. Based on the current situation, it is important to identify factors which is helpful to improving prediction accuracy and promoting curative effect of GC. Most of the HOX genes are generally activated and expressed during embryogenesis, and many of these proteins are aberrantly expressed during tumorigenesis. According to the literatures, HOX proteins are related to the prognosis and clinicopathological features of GC, but the results are controversial. We conducted this study to further clarify the effects of HOX proteins on the prognosis and clinicopathological characteristics of GC and describe the molecular mechanisms by which HOX proteins regulate tumorigenesis and development of GC. The present systematic review and meta-analysis enrolled 19 eligible studies containing 3775 patients. In the pooled analysis of the effects of HOX proteins on the GC prognosis, HOXB9, HOXD10 and HOXA5 were correlated with a good prognosis in patients with GC, while HOXA13, HOXC6, HOXB7, HOXA1, HOXC9, HOXC10, HOXD4, HOXA1 and HOXD9 were related to a poor prognosis. However, Kato et al. identified positive HOXB9 expression in GC as a marker of a poor prognosis. Unfortunately, the study by Kato was not included in this meta-analysis due to the lack of an analysis of HOXB9 expression in paired noncancerous mucosae [58].
The relationship between HOX proteins and clinicopathological features of GC were also analysed in this study. The results revealed correlations between the expression of HOX proteins and TNM stage, T category, tumour size, lymph node metastasis, distant metastasis, vascular invasion, histological differentiation, and Lauren classification in GC. Based on the results of the meta-analysis described above, we speculated that HOX proteins might predict the prognosis of patients with GC, which was also confirmed in each included original study. Therefore, we inferred that combined detection of the expression of various HOX proteins might provide a novel perspective for predicting the prognosis of patients with GC.
Currently, some clinicopathological parameters such as age, sex, tumour stage, depth of invasion, lymph node metastasis, distant metastasis, and resection margins, have been proven to be prognostic indicators of GC [59,60]. At the same time, several molecules are under investigation as predictors of survival, such as gene mutations, DNA methylation, RNAs, and proteins [61]. Regrettably, many studies have only explored the individual relationship between clinicopathological characteristics or molecular markers and the prognosis of patients with GC, although a few studies have established prognostic models [62]. Bria et al. combined clinicopathological parameters (sex, age, Lauren classification, stage, margins, grade, site, size, and resected nodes) with molecular markers (HER2, FHIT, and APC) to construct a risk stratification of GC, establishing a scientific model to determine its prognosis. In addition, the authors conducted a large prospective validation with a larger sample size to eliminate all sources of bias in the retrospective study [63]. GC is highly heterogeneous, and even similar clinicopathological features result in different outcomes, suggesting that a more reasonable classification system is needed for predicting the prognosis and therapeutic effect of GC. A novel classification system with four molecular subtypes was developed by The Cancer Genome Atlas (TCGA) [64]. Besides, Sohn et al. developed a scoring system (TCGA risk score) based on TCGA to predict prognosis and adjuvant chemotherapy outcomes in patients with GC, which was validated as an independent prognostic factor for GC in multivariate Cox regression analyses [65]. Analogously, Lin et al. established a novel prognosis scoring system based on TCGA and Gene Expression Omnibus to predict the prognosis of GC, which comprised signatures of tumour protein-coding genes (P), tumour noncoding genes (N) and immune/ stroma cells in the tumour microenvironment (M) (PMC score). Furthermore, the combination of PNM scores with American Joint Committeeon Cancer (AJCC) staging significantly increased its predictive value [66]. In addition, Tahara et al. investigated the prognosis and clinicopathological characteristics of GC by combining genetic and epigenetic abnormalities. The CpG island methylator phenotype (CIMP) and TP53 hot spot mutation status (R175, G245, R248, R273, and R282) were sufficient to predict the prognosis and clinicopathological features of GC. Among these features, patients with the CIMP − TP53 hot spot + subtype presented the worst overall survival [67]. Moreover, Ooi et al. selected three oncogenic pathways (NF-κB, Wnt/β-catenin, and proliferation/ stem cells) by analysing a GC pathway heatmap and combined them to predict its prognosis, which was validated in vitro [68].
The development of GC is determined by both genes and environmental factors, which has been confirmed in mouse models. Microbial infections, particularly Helicobacter pylori (H. pylori) and Epstein-Barr virus, are important environmental factors and have been confirmed as prognostic factors for GC [69,70]. Although H. pylori infection is the strongest risk factor for GC, very few H. pylori-infected populations develop GC. This outcome is attributed to the duration of infection, strain type and host genetic signatures [71]. The crucial effects of genetic factors on GC development have been revealed using progress in genetic technology, including the construction of genetically engineered mice via recombinant DNA technology to achieve molecular overexpression or deficiency, as well as gene mutations, clarifying the pathogenesis of GC and the interactions between various factors. For example, INS-GAS transgenic mice on the FVB genetic background that overexpress gastrin develop intramucosal carcinomas with submucosal and   [72,73]. However, INS-GAS mice on a C57BL/6 background infected with H. felis do not progress to GC [74]. Surprisingly, gastrin knockout mice (GAS−/− mice) are also confirmed to be susceptible to GC and exhibit antral GC, in contrast to INS-GAS mice, which develop corpus cancers [75]. Moreover, GAS−/− mice are more susceptible to antral cancer induced by MNU, a gastric carcinogen used in mouse models, compared to WT mice on the same genetic background [76]. Taken together, these studies reveal important roles of genetic signatures in the development of GC, and the external factor such as infection is also indispensable. Thus, the establishment of a comprehensive and detailed scoring system containing the most basic clinicopathological parameters, molecular markers, gene expression profiles, microbial infections, etc., might be more accurate in predicting the prognosis of patients with GC than a single factor. Our manuscript analysing the effects of HOX proteins in GC development aimed to predict the prognosis and provide therapeutic targets for GC. The results of this metaanalysis recommend the inclusion of HOX proteins in the model predicting the prognosis of GC.
Several limitations of this systematic review and meta-analysis should be noted. First, several HRs and their 95% CIs for OS were extracted from the survival curves, which might affect the reliability of the results. Second, the sample size of each study was not large enough, which might affect the accuracy of the results. Third, IHC methodologies including the primary antibody used, antibody dilutions, and the scoring system applied, differed, which might partially contribute to the heterogeneity. Finally, all patients included in our study were Asians, which might restrict the applicability of our results to other races.

Conclusions
This systematic review and meta-analysis firstly generalized and evaluated the significance of HOX proteins in modulating the prognosis and clinicopathological characteristics of GC. We also summarized the molecular mechanisms by which HOX proteins regulate tumorigenesis and development of GC. Based on these findings, HOX proteins might serve as biomarkers and therapeutic targets for GC. Considering the limitations of this study, further large-scale prospective and high-quality studies are required to confirm the potential values of HOX proteins in GC.