SOX2 contributes to invasion and poor prognosis of gastric cancer: A meta-analysis

Background:

The sex-determining region Y-box 2 (SOX2) has been identified to be involved in tumor progression and prognosis in patients with gastric cancer (GC). However, its action is paradoxical. Thus, we conducted the first meta-analysis based on eligible studies to evaluate the clinical utility of SOX2 in GC only.

Methods:

A thorough electronic search was performed to collect eligible studies. The hazard ratios (HRs) or odds ratios (ORs) with 95% confidence intervals (CIs) were generated from included studies to assess the strength of the association between SOX2 and prognosis and clinicopathological characteristics in GC.

Results:

A total of 10 studies comprising 1321 patients with GC were identified for the meta-analysis. The pooled results revealed that high SOX2 expression was significantly associated with poor overall survival compared to low SOX2 expression (pooled HR = 1.485; 95% CI: 1.022–2.160; 𝑃 = .04). The statistical significance between SOX2 expression and overall survival was also established in univariate analysis (pooled HR = 1.606; 95% CI: 1.134–2.274; 𝑃 < .01), as well as recruitment time exceeding 2010 (pooled HR = 1.873; 95% CI: 1.041–3.371; 𝑃 = .04), follow-up time more than 5 years (pooled HR = 1.642; 95% CI: 1.066–2.527; 𝑃 = .02), and cutoff value of more than 5% of cells stained (pooled HR = 1.730; 95% CI: 1.162–2.577; 𝑃 < .01). Moreover, we verified that positive SOX2 expression was correlated with advanced tumor invasion depth (pooled OR = 0.494; 95% CI: 0.362–0.675; 𝑃 < .01) and positive vascular invasion (pooled OR = 1.515; 95% CI: 1.078–2.130; 𝑃 = .02).

Conclusion:

SOX2 could not only be an independent prognostic marker in GC but might also be a novel target for cancer therapy.

1. Introduction

Gastric cancer (GC) is the third leading cause of cancer deaths worldwide and represents an active topic of clinical and basic concern. Owing to advances in prevention, diagnosis, and therapy, there has been a persistent and steady decline in incidence and mortality in recent years.^[1] To date, clinical prognostic factors of GC, including TNM stage, depth of invasion, lymph node (LN) or distant metastasis, age, and recurrence, have been well recognized.^[2,3] Otherwise, molecular characterization for illuminating the mechanisms involved in tumor initiation and development of GC has been progressively published. However, the prognosis of patients with GC is still poor, and the 5-year survival rate is not satisfactory.^[4] Based on this information, it is vital to find reliable and effective prognostic markers to help patients avoid being enslaved to GC.

The sex-determining region Y-box 2 (SOX2) is known as a stem cell-related factor, which is essential for the establishment and maintenance of embryonic stem cells and the reprogramming of the pluripotency of terminally differentiated cells.^[5] Furthermore, SOX2 also acts as a transcription factor in regulating embryonic stem cell-specific expression.^[6] Recently, cancer stem cells (CSCs) have been identified to be involved in tumor progression and recurrence.^[7] In line with this idea, many convincing studies have gradually focused on the oncogenic and prognostic roles of SOX2 in several cancers, especially GC. However, its action is paradoxical. Downregulated expression of SOX2 is observed in GC and is associated with a worse prognosis.^[8] The arrest in G1 of the cell cycle and decrease in cyclin D1 and increase in p27Kip1 protein levels and the orchestration of downstream phospho-Akt dephosphorylation could partially clarify the anticancer effect of SOX2 in GC cells.^[9,10] In obvious contrast, a high SOX2 level was correlated with poor survival and unfavorable prognostic factors, including tumor size, tumor invasion, and LN metastasis of GC, according to recent studies.^[11,12] Functionally, SOX2 enhances the tumorigenicity and chemoresistance of GC stem-like cells and regulates the expression of genes associated with oncogenic factors.^[13] Meanwhile, inhibition of gastric tumor growth was found in a xenograft mouse model with blockade of SOX2.^[14] Motivated by this inconsistency, the prognostic value of SOX2 in GC needs to be clarified urgently. To date, some meta-analyses have been performed to assess the overall value of SOX2 in the prognosis of various cancers.^[15,16] However, we conducted the first meta-analysis based on eligible studies to evaluate the clinical utility of SOX2 in GC only.

2. Materials and Methods

2.1. Search strategy and study selection

A thorough electronic search (last updated in December, 2021) of published English-language literature was performed in PubMed, PMC, EMBASE, and EBSCO using the following keywords: “(SOX2, SRY box-2, Sex determine region Y box 2, SRY-Related HMG-Box Gene 2) AND (cancer or tumor or malignancy or neoplasm or carcinoma or adenocarcinoma) AND (gastric or stomach or ventriculi) AND (prognosis or prognostic or survival or outcome).” Additionally, reference lists of relevant studies were also screened to find additional work. The meta-analysis was in accordance with the reporting checklist as part of the preferred reporting items for systematic reviews and meta-analyses statement.^[17]

2.2. Inclusion and exclusion criteria

All studies had to meet the following predefined criteria to be included: studies reporting the relationship between SOX2 expression and clinical parameters and overall survival (OS) in patients with GC; studies reporting a reliable and stable cutoff value based on SOX2 expression in tumor tissue; patients were divided into SOX2-positive (high) or SOX2-negative (low) groups; studies with more than 60 patients; sufficient published data were provided to obtain the hazard ratio (HR) and 95% confidence interval (CI) information; patients underwent surgical treatment, and received standard subsequent therapy after surgery; and scientific and valid studies. The following studies were excluded: laboratory articles, abstracts, letters, editorials, meeting abstracts, case reports, and reviews.

2.3. Data extraction and quality assessment

Two authors independently extracted the data from studies identified in the literature searches. Data regarding baseline characteristics (first author, publication date, country, age, recruitment time, sample size, detection method, follow-up period, cutoff values, analysis method, etc), and outcomes (OS, HRs, CIs, etc) were extracted using a predefined template. It should be noted that HRs and the corresponding 95% CIs provided by multivariate analysis were preferred for subsequent data analysis. Otherwise, the HR could be extracted from the univariate analysis or calculated using the Kaplan–Meier survival curves for data analysis.^[18] Disagreements were solved by consensus. The quality of the included studies was assessed by the Newcastle-Ottawa Scale (NOS).^[19] The total NOS score (range 0–9) was based on three categories: selection, comparability, and exposure. A high-quality study was defined as a score ≥ 6.

2.4. Statistical analysis

Data analyses were performed using STATA 14.0 software (STATA Corporation, College Station, TX). The odds ratios (ORs) and 95% CIs were combined to estimate the relationship between SOX2 expression and clinicopathological features, and a Z-test was used to determine the statistical significance of the association between SOX2 expression and OS based on the result of the combined HR and 95% CI. Heterogeneity was examined through the chi-squared test based on the Q statistic; meanwhile, the random models were used for all analyses for consistency. Sensitivity analysis was conducted by sequentially omitting studies to validate the stability of the pooled outcomes. Meta-regression analysis was carried out to explain the potential source of heterogeneity among studies. Publication bias was assessed by Begg and Egger’s tests.^[20] All statistical tests were 2 sided. 𝑃 < .05 was considered statistically significant.

3. Results

3.1. Characteristics of the included studies

As illustrated in Figure 1, a total of 404 studies were retrieved in the initial search. After a careful review of the title and abstract, 381 articles were excluded because they were obviously unrelated or of other types. Then, the remaining 23 full-text articles were reviewed for eligibility. Thirteen of them were further excluded (excluded for animal studies [n = 2], cell studies [n = 3], no definition of cutoff value [n = 3], and no survival data or small size [n = 5]). Finally, a total of 10 studies comprising 1321 patients with GC were identified for the meta-analysis. The main characteristics of the 10 eligible studies are summarized in Table 1. Among these, 6 studies were conducted in China,^[8,21–25] 2 in Japan,^[11,26] and the other studies were conducted in Portugal^[27] and Iran.^[12] Immunohistochemistry was used to evaluate the expression of SOX2 in 9 studies, while RT-PCR was applied in 1 study. The cutoff value for defining positive SOX2 expression could be retrieved from all eligible studies. Nine studies clearly defined the recruitment time. In addition, 8 studies reported the follow-up period. Five studies contained HRs from multivariate analysis; meanwhile, the HRs in 8 studies were extracted through univariate analysis or survival curves. The total NOS scores confirmed the high quality of the included articles.

Figure 1.

Flow chart of study selection in the meta-analysis.

Table 1.

Main characteristics of studies included in meta-analysis.

Author	Year	Study region	Recruitment time	Sample size	Detection method	Follow up period	Cutoff values (positive/negative)	Analysis method	OS, HR estimation	Quality score
Matsuoka J	2010	Japan	NS	253	IHC	NS	Score ≥ 5*	Univariate/multivariate analysis	2.37 (1.19–4.72)	7
Peng W	2013	China	1998.3–2011.5	64	IHC	NS	Positive if 5% or more of the tumor cells†	Univariate analysis	1.95 (1.61–2.37)	7
Camilo V	2014	Portugal	1988–2010	176	IHC	62 mo (median)	Positive if 5% or more of the tumor cells	Univariate analysis	2.33 (1.14–4.75)	7
Li N	2015	China	2008–2009	69	IHC	35 mo (median)	Score ≥ 5*	Univariate analysis	1.34 (0.96–1.86)	7
Chen X L	2016	China	2007.1–2012.10	264	IHC	Up to 2015.1	Score ≥ 8‡	Multivariate analysis	0.616 (0.421–0.901)	8
Zhang X W	2016	China	2005.1–2011.8	94	IHC	Up to 2011.11	Positive if 10% or more of the tumor cells†	Univariate analysis	3.24 (1.97–5.34)	7
Chen Y S	2016	China	1990.5–1995.6	77	IHC	5 yr (maximum)	Score > 0‡	Univariate/multivariate analysis	0.37 (0.20–0.70)	8
Lang Y	2017	China	2010–2013	116	IHC	26 mo (median)	Positive if 5% or more of the tumor cells	Univariate/multivariate analysis	2.02 (1.20–3.40)	8
Hashimoto I	2017	Japan	2001.1–2006.6	108	IHC	56 mo (median)	Score ≥ 3*	Univariate analysis	0..96 (0.77–1.20)	7
Basati G	2020	Iran	2014.1–2017.12	100	qRT-PCR	4 yr (maximum)	Median	Univariate/multivariate analysis	3.71 (1.47–8.85)	8

HR = hazard ratio, IHC = immunohistochemical, NS = data were not shown, OS = overall survival.

^*The sum of the intensity and distribution scores.

^†Cells showed moderate or strong staining intensity.

^‡The score for intensity was multiplied by the score for extent of staining.

3.2. Correlation between SOX2 expression and prognosis in GC

The combined results based on the included datasets are summarized in Table 2. Remarkable heterogeneity (𝐼² = 88.6%; 𝑃_h < .01) was found in the analysis, and the random model showed that high SOX2 expression was significantly associated with poor OS compared to low SOX2 expression (pooled HR = 1.485; 95% CI: 1.022–2.160; 𝑃 = .04, Fig. 2). In addition, the statistical significance between SOX2 expression and OS was also established in univariate analysis (pooled HR = 1.606; 95% CI: 1.134–2.274; P < .01) but not in multivariate analysis (pooled HR = 1.275; 95% CI: 0.576–2.823; P = .55). Notably, the results revealed that high SOX2 expression significantly led to worse OS in the subgroups with recruitment times exceeding 2010 (pooled HR = 1.873; 95% CI: 1.041–3.371; P = .04) and follow-up times longer than 5 years (pooled HR = 1.642; 95% CI: 1.066–2.527; P = .02). Moreover, the combined analysis indicated that the high expression of SOX2 was still an unfavorable predictor of OS when the cutoff value was more than 5% of cells stained (pooled HR = 1.730; 95% CI: 1.162–2.577; P < .01). Furthermore, this association also existed in the non-east Asian population (pooled HR = 2.790; 95% CI: 1.596–4.877; P < .01). However, no significant relationship was found in the pooled analysis of other subgroups (Table 2).

Table 2.

Meta-analysis of SOX2 expression and prognosis in gastric cancer.

Categories	Studies (patients)	HR (95% CI)	I2 (%)	P h	Z	P
OS	10 (1321)	1.485 (1.022–2.160)R	88.6	<.01	2.07	.04
Analysis method
Multivariate analysis	5 (810)	1.275 (0.576–2.823)R	89.2	<.01	0.60	.55
Univariate analysis	10 (1321)	1.606 (1.134–2.274)R	90.1	<.01	2.67	<.01
Recruitment time
Exceeding 2010	5 (638)	1.873 (1.041–3.371)R	89.7	<.01	2.09	.04
Up to 2010	5 (683)	1.180 (0.729–1.910)R	82.9	<.01	0.67	.50
Follow up period
>5yr	7 (1075)	1.642 (1.066–2.527)R	89.5	<.01	2.25	.02
≤5 yr	3 (246)	1.181 (0.402–3.468)R	90.0	<.01	0.30	.76
Cutoff value
More than 5% of cells stained	7 (1036)	1.730 (1.162–2.577)R	84.9	<.01	2.70	<.01
Less than 5% of cells stained	2 (185)	0.625 (0.246–1.583)R	87.4	<.01	0.99	.32
Research region
East Asia	8 (1045)	1.312 (0.874–1.970)R	90.3	<.01	1.31	.19
Non-east Asia	2 (276)	2.790 (1.596–4.877)R	0.0	.43	3.60	<.01

CI = confidence interval, HR = hazard ratio, OS = overall survival, P = P-value for statistical significance based on Z test, P_h = P-value for heterogeneity based on Q test,

^R= pooled HRs were derived from random-effect model.

Figure 2.

Forest plot of HR of overall survival for patients with gastric cancer. CI = confidence interval, HR = hazard ratio.

The heterogeneity among the included studies was assessed by the Q statistic. As noted in Table 2, extreme heterogeneity existed in assessing the association of SOX2 expression and OS. Even though the analysis was performed based on the subgroups of analysis method, recruitment time, follow-up period, cutoff value, and research region, the heterogeneity was still irreducible. In addition, meta-regression analysis indicated that sample size (P = .57), detection method (P = .27), analysis method (P = .80), recruitment time (P = .45), follow-up period (P = .35), cutoff value (P = .24), and research region (P = .85) were not responsible for the source of heterogeneity for OS.

3.3. Correlation between SOX2 expression and clinicopathological characteristics in GC

The relationship of SOX2 expression with clinicopathological characteristics is evaluated in Table 3. Based on the ORs derived from each available study, we demonstrated that positive SOX2 expression was correlated with certain phenotypes of tumor aggressiveness, such as advanced tumor invasion depth (pooled OR = 0.547; 95% CI: 0.354–0.847; P < .01) and positive vascular invasion (pooled OR = 1.515; 95% CI: 1.078–2.130; P = .02). Meanwhile, the results revealed that high SOX2 expression was associated with male sex (pooled OR = 1.628; 95% CI: 1.156–2.293; P < .01). Furthermore, no significant association was found between SOX2 and any other clinicopathological characteristics, including age, Lauren classification, LN metastasis, distant metastasis, and TNM stage (P > .05, Table 3).

Table 3.

Meta-analysis of SOX2 expression classified by clinicopathological parameters.

Study covariates	Studies (patients)	OR (95% CI)	I2 (%)	P h	Z	P
Gender (male/female)	5 (776)	1.628 (1.156–2.293)	0.0	.41	2.79	<.01
Age (≥60/<60)	3 (498)	1.086 (0.746–1.582)	0.0	.57	0.43	.67
Lauren classification (intestinal/diffuse)	4 (619)	1.078 (0.779–1.490)	0.0	.92	0.45	.65
LN metastasis (positive/negative)	5 (997)	1.105 (0.534–2.283)	82.2	<.01	0.27	.79
Distant metastasis (positive/negative)	2 (384)	1.792 (0.510–6.294)	56.2	.13	1.91	.36
Tumor invasion (1–2/3–4)	6 (1065)	0.547 (0.354–0.847)	36.4	.16	2.71	<.01
TNM stage (I–II/III–IV)	6 (1061)	0.800 (0.425–1.504)	79.9	<.01	0.69	.45
Vascular invasion (positive/negative)	3 (796)	1.515 (1.078–2.130)	0.0	.80	2.39	.02

CI = confidence intervals, LN = lymph node, OR = odds ratio, P = P-value for statistical significance based on Z test, P_h = P-value for heterogeneity based on Q test.

3.4. Sensitivity analysis and publication bias

To gain insights into the stability and scientificity of the combined results, sensitivity analysis was performed to evaluate the influence of each study on the pooled HR. After omitting each study in turn, no point estimate of the pooled HR was outside the 95% CI, which suggested that the results were reliable and stable (Fig. 3).

Figure 3.

Sensitive analysis of overall survival for patients with gastric cancer. CI = confidence interval.

There were no obvious publication biases in Egger’s tests (P = .80, Fig. 4) or Begg tests (P = .72) for OS of SOX2 expression, which indicated that the result of this meta-analysis was robust and scientific.

Figure 4.

Effect estimate for potential publication bias of included studies in Egger’ test. CI = confidence interval.

4. Discussion

The prognosis of tumors has always been the focus and difficulty of medical science research. Notably, the OS of GC is not optimistic when compared with other types of tumors.^[28] Clinical characteristics such as TNM and LN cannot completely reveal the outcome of GC, which leads to delayed treatment measures and inevitably results in poor prognosis. In recent years, biological mechanisms have gradually been discovered as a potential factor for patient outcome in GC. Consequently, the identification of molecular biomarkers is of critical importance to facilitate clinical decision-making. As this meta-analysis reveals the importance of SOX2 expression in the prediction of GC, SOX2 detection appears to be an independent prognostic factor for the survival of patients with GC.

Cancer stem cells, as the potential source of maintaining tumor growth, recurrence, and chemotherapy resistance, were initially identified as displaying characteristics similar to normal stem cells.^[29] Recent studies have demonstrated that CSCs are also involved in tumor initiation and invasion.^[30] Therefore, effort has been made in medical research on specific markers of CSCs in GC tissues to determine their potential clinical application in the past decade. Notably, SOX2 was found to be the most elevated factor in CSCs and could establish a continuum between tumor initiation and progression by regulating the function of CSCs.^[31,32] Specifically, SOX2 not only regulates embryonic stem cell fate and activity as well as tissue homeostasis and regeneration but also participates in gastric differentiation and progression.^[9,33] Indeed, aberrant expression of SOX2 acts as an oncogene and initiates carcinoma by synergizing with inflammation-induced STAT3 activation.^[34] Moreover, SOX2 also directly interacts with cancer-related genes to accelerate tumorigenesis.^[13] Meanwhile, a high SOX2 level was correlated with poor survival.^[21,27] In obvious contrast, it has been shown that SOX2 is frequently downregulated in GC and plays a role in tumor inhibition, thereby leading to a worse prognosis.^[24] Wang et al^[8] reported that SOX2 expression is progressively reduced during gastric carcinogenesis. Consistent with this result, aberrant DNA methylation, overexpression of miR-126, and H. pylori were reported to be correlated with a decrease in SOX2 expression and further regulated downstream molecules to stimulate carcinogenesis.^[10,35,36] Taken together, the specific role of SOX2 in tumorigenesis is controversial, and as a consequence, it results in the instability of the correlation between SOX2 expression and the prognosis of GC. In view of this, we specifically conducted this meta-analysis to explore the prognostic value of SOX2 in GC.

Our meta-analysis is the first to evaluate the association between SOX2 expression and the prognosis of patients with GC only. A total of 10 studies comprising 1321 patients with GC were included in our meta-analysis. The pooled results confirmed that upregulated expression of SOX2 was related to poor prognosis, which suggested that SOX2 might be an independent marker of prognosis in GC. Similar results have been reported in meta-analyses of other cancers.^[15,16] Based on the fact that there was no single study to govern the results and no publication bias was found, the stability and scientificity of the combined results were further verified by sensitivity analysis and bias tests. However, despite the robustness of the pooled results, the extreme heterogeneity among the included studies was unexpectedly extracted by the forest plots and I². As previously described in Table 2, even when we conducted subgroup analyses according to analysis method, recruitment time, follow-up period, cutoff value, and research region, significant heterogeneity was still present. We hypothesized that the source of the heterogeneity might come from the sample size, detection method, analysis method, recruitment time, follow-up period, cutoff value, and research region. Nevertheless, meta-regression analysis indicated that none of the assumptive factors could completely explain the heterogeneity. In addition, according to our stratified analysis, a significant relationship between SOX2 expression and OS was found in subgroups of recruitment time (exceeding 2010), follow-up time (>5 years), and cut off value (more than 5% of cells stained), suggesting that studies that enrolled more recent patients with longer follow-up and stricter cutoff values would obtain more stable results, which precisely proved that the treatment of GC was gradually improving and stabilizing. However, the results differed among some other subgroups, and the reasons for these differences may lie in the paradoxical role of SOX2 in GC. Current studies have demonstrated that SOX2 can act not only as a carcinogen but also as a tumor suppressor, which inevitably leads to different results in clinical studies. Fortunately, our meta-analysis revealed the prognostic prediction effect of SOX2 in GC, even though the data were still insufficient. Of course, these results need to be confirmed by further research.

Aberrant SOX2 expression has been strongly linked to clinicopathological factors in cancers. Therefore, we further analyzed the association between SOX2 expression and clinicopathological factors based on the included studies. Evidence has postulated that high expression of SOX2 is closely related to LN, distant metastasis and advanced TNM stage in head and neck and colorectal cancer.^[15,16] However, regarding the role of SOX2 in GC metastasis, it has been found that SOX2 inhibits metastasis by regulating CCND1 and PARP,^[37] while some clinical studies have found that enforced SOX2 promotes metastasis.^[11,27] Interestingly, this uncertainty was confirmed in the current meta-analysis. In addition, according to the pooled results, although only 6 studies have provided the relevant data, which makes the conclusions seem to be incomplete, there is no doubt that SOX2 was associated with tumor invasion, both in tumor depth and in vessels. Since both tumor invasion and vascular invasion are the key events of tumor growth and play a critical role in promoting gastric tumorigenesis, we have no reason to doubt that overexpression of SOX2 could stimulate invasion and lead to tumor progression, thereby contributing to poor prognosis in patients with GC.

Chemotherapy has been internationally and generally accepted and used in the past century. In the initial stage of chemotherapy, tumor progression, and metastasis can be effectively suppressed, and even complete tumor regression can be induced. However, cancer cells eventually become resistant, impairing the effects of chemotherapy.^[38] Currently, downregulation of SOX2 with siRNA obviously reduced doxorubicin efflux and cisplatin, thus weakening chemoresistance in GC.^[13] Furthermore, Gangemi et al^[39] reported that SOX2 silencing stopped the proliferation and eliminated the tumorigenicity of glioblastoma cells. Similarly, knockdown of SOX2 in colorectal cancer cells inhibited cell migration and invasion as well as attenuated colony forming capacity and tumorigenicity.^[40] Moreover, experimental evidence from the GC cell model significantly demonstrated that inhibition of SOX2 reduced cell proliferation and migration and blocking of SOX2 also reduced the tumorigenic potential of AZ-521 cells in vivo.^[14] These results are consistent with our meta-analysis and support that SOX2 might be a crucial therapeutic target in patients with GC.

The limitations of this meta-analysis should be stressed. First, although we abandoned some studies with too few samples, there were still 4 articles with fewer than 100 samples. Meanwhile, the total number of articles included was relatively small. Second, based on the available web search, the language of the included literature was limited to English. It is very important to take other language reports into consideration in future analyses. Third, due to the small number of included studies, we failed to comprehensively explore the association between SOX2 expression and clinicopathological factors. In particular, we found that high SOX2 expression was associated with male sex, but there was not enough evidence to explain this association. Fourth, some conditions, such as age, follow-up time, and cut-off value, could not be unified in this meta-analysis, which might lead to the heterogeneity and weakened the strength and extensionality of conclusion. Finally, the current meta-analysis only focused on the prognostic value of SOX2, although some other studies have shown great interest in the diagnostic value of SOX2 in patients with GC. Thus, more well-designed studies with larger cohorts of patients in different ethnicities are warranted to further explore the role of SOX2 in GC.

5. Conclusions

In conclusion, this meta-analysis supported that SOX2 enhanced tumor invasion and vascular invasion in GC, and high SOX2 expression was significantly associated with poor OS in patients with GC. Moreover, SOX2 could not only be an independent prognostic marker in GC but might also be a novel target for cancer therapy. For the limitations presented above, more studies are warranted to support our findings.

Author contributions

Conceptualization: Kaifeng Tang, Jingting Liu, Jianhua Liao.

Data curation: Jingting Liu, Chunyan Meng, Jianhua Liao.

Formal analysis: Baoqing Liu, Chunyan Meng.

Software: Kaifeng Tang, Baoqing Liu.

Writing – original draft: Jianhua Liao.

Writing – review & editing: Kaifeng Tang, Jingting Liu, Jianhua Liao.