35 blood and urine biomarkers and digestive system cancers: A two-sample Mendelian randomization study

Abstract

Digestive system cancers (DSCs) are globally prevalent and account for a substantial proportion of cancer-related deaths. This study aimed to investigate the potential causal relationships between 35 blood and urine biomarkers and the risk of 4 major DSCs, including esophageal, gastric, colorectal, and pancreatic cancers. A 2-sample bidirectional Mendelian randomization (MR) approach was employed. Genome-wide association studies (GWAS) summary statistics for DSCs were obtained from public databases, and biomarker data were sourced from a GWAS meta-analysis of the UK Biobank. The inverse variance weighted (IVW), MR-Egger, and weighted median methods were used to examine causality. To evaluate robustness, we performed sensitivity analyses, including Cochran Q test for heterogeneity, MR-Egger and MR-PRESSO tests for pleiotropy, and leave-one-out analysis. Furthermore, reverse MR analyses were conducted for all significant causal relationships identified. We identified 14 significant associations: 6 with esophageal cancer, 1 with gastric cancer, 3 with colorectal cancer, and 4 with pancreatic cancer. Elevated serum uric acid was associated with increased pancreatic cancer risk (odds ratio [OR] = 1.230, 95% confidence interval [CI]: 1.015–1.490, P = .035) but decreased esophageal cancer risk (OR = 0.822, 95% CI: 0.683–0.989, P = .038). Increased insulin-like growth factor (IGF-1) levels were associated with higher risks of esophageal (OR = 1.228, 95% CI: 1.035–1.457, P = .019) and colorectal cancer (OR = 1.216, 95% CI: 1.125–1.314, P < .001). Subsequent sensitivity analyses revealed that our results were rarely influenced by pleiotropy or heterogeneity. Reverse MR analysis indicated that there were no causal associations between genetic predisposition to DSCs and the identified biomarkers. The identified biomarkers may serve as potential diagnostic markers, aiding in the early detection of cancer, which is crucial for improving treatment outcomes.

1. Introduction

Digestive system cancers (DCSs), including esophageal, gastric, colorectal, pancreatic, hepatic, and biliary malignancies, are consistently recognized as among the most prevalent cancers diagnosed globally. According to the Global Cancer Observatory estimates by the International Agency for Research on Cancer on cancer incidence and mortality in 2020, DSCs comprised over 26.4% of newly diagnosed cancer cases and over 36.3% of cancer-related deaths, thus posing a substantial medical and economic burden on society and individuals.^[1] Due to the complexity and heterogeneity of the pathogenesis of DSCs, significant challenges arise in the early diagnosis and prevention of these cancers. Employing commonly available biomarkers and laboratory indices for risk assessment holds significant importance in the prevention of DSCs.

Blood and urine biomarkers are routinely used in clinical practice for diagnosing and monitoring a wide range of diseases.^[2–4] For instance, low circulating levels of 25-hydroxyvitamin D (25(OH)D) have been linked to an increased risk of several non-dermatological cancers and cancer-related mortality.^[5] Serum lipids, another key class of biomarkers, are implicated in the development of DSCs through various mechanisms, including influences on cell membrane integrity, signaling pathways, and inflammation.^[6,7] Unlike blood, urine lacks homeostatic mechanisms to stabilize its composition, which makes it an ideal source of biomarkers by more accurately reflecting bodily changes.^[8] However, conventional observational studies are susceptible to confounding factors and reverse causality, limiting the ability to infer causal relationships.^[9]

Mendelian randomization (MR) analysis, a cutting-edge research method, leverages genetic variations as instrumental variables (IVs) to investigate the putative causal link between exposures and specific outcomes. This approach adheres to the fundamental principle of Mendelian inheritance in which alleles are randomly segregated from parents to offspring during conception.^[10] Hence, MR offers a robust method to overcome residual confounding and reverse causality, clarifying causal relationships between blood and urine biomarkers and DSCs. Recently, this research method has also been applied to explore the causal relationship between common blood and urine biomarkers and cancers. Although the genetics of biomarkers such as lipids, glucose profiles, and renal function measurements have been extensively studied, it is noteworthy that the genetic basis of most biomarkers has not yet been fully explored in large-scale population datasets.^[11–14] Sinnott-Armstrong N et al examined the genetic underpinnings of each of the 35 blood and urine biomarkers on an individual basis in the UK Biobank, identifying 51 causal relationships via MR analysis.^[15] These 35 biomarkers encompass a broad range of biological processes, including lipid metabolism, glucose homeostasis, liver and kidney function, inflammation, and hormonal regulation, many of which are implicated in cancer biology. This study provides valuable insights and crucial data source for our research, raising the question of whether MR analysis can be performed to investigate the causal relationship between blood and urine biomarkers and DSCs. Consequently, we designed this 2-sample bidirectional MR study, aiming to explore potential causal links between these common biomarkers and various DSCs. Given the ease of accessibility of these biomarkers, gaining further insights into their role in digestive system cancers could have significant clinical and public health implications.

2. Materials and methods

2.1. Study design

This bidirectional 2-sample MR study was designed to assess the causal relationships between 35 biomarkers and 4 DSCs. The validity of the MR analysis hinges on 3 core assumptions: the genetic instruments are strongly associated with the exposure (biomarker); the instruments are independent of confounders; and the instruments influence the outcome (cancer) only through the exposure, not via alternative pathways.^[16] The specific analytical framework was depicted in Figure 1.

Figure 1.

The framework of 2-sample bidirectional Mendelian randomization analysis. MR = Mendelian randomization, IV = instrumental variable, SNP = single nucleotide polymorphism, LD = linkage disequilibrium, IVW = inverse variance weighted.

2.2. Data sources and participants

Genome-wide association studies (GWAS) is a research approach that examines genetic variations across the entire genome in a large population to identify associations between specific genetic markers and traits or diseases, enabling a comprehensive understanding of their genetic basis.^[17] The phenotypic and genotypic data for 35 blood and urine biomarkers required for this study were derived from a systematic analysis conducted by Sinnott-Armstrong N et al on the genetic structure of over 360,000 individuals from the UK Biobank.^[15] The comprehensive dataset of GWAS for 35 blood and urine biomarkers can be accessed via the following link: https://doi.org/10.35092/yhjc.12355382. The list of these biomarkers was shown in (Table S1, Supplemental Digital Content). All blood and urine samples were assayed during recruitment and at subsequent repeat assessments. For the outcomes, we initially selected 4 types of DSCs: esophageal, gastric, colorectal, and pancreatic cancers. Subsequently, we searched and downloaded the relevant GWAS databases for each cancer. Detailed information regarding the summary statistics can be found in Table 1. All original data were obtained with ethical approval and informed consent. As this study only involved secondary utilization of data, no additional ethical approval was required from our affiliated institutions.

Table 1.

The data resources for digestive system cancers.

Disease	GWAS ID	Consortium	Ancestor	Sample size	Case	Control
Esophageal cancer	ebi-a-GCST90018841	EBI	European	476,306	998	475,308
Gastric cancer	finn-b-C3_STOMACH_EXALLC	FinnGen	European	174,639	633	174,006
Colorectal cancer	ebi-a-GCST90018808	EBI	European	470,002	6581	463,421
Pancreatic cancer	ebi-a-GCST90018893	EBI	European	476,245	1196	475,049

EBI = European Bioinformatics Institute, GWAS = genome-wide association study.

2.3. IVs selection criteria

First, single-nucleotide polymorphisms (SNPs) associated with trait genetics exhibiting genome-wide significance (P < 5 × 10⁻⁸) in GWAS summary data were screened and selected as instrumental variables. Subsequently, to minimize the influence of linkage disequilibrium (LD) on our research outcomes, we established a threshold for LD (R² ≤ 0.001, 10,000 kb) and accordingly conducted further screening of SNPs. The F-statistic serves as an indicator of the strength of the association between IVs and the exposure factor.^[18] We established a threshold of F > 10 for screening SNPs that were strongly associated with the exposure factor, aiming to minimize biases and confounding factors in our analysis.^[17] The F-statistic and R² were calculated using well-established and reliable formulas:^[19,20]

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{F}=\left(\frac{\mathrm{N}-\mathrm{k}-1}{\mathrm{k}}\right)\left(\frac{{\mathrm{R}}^2}{1-{\mathrm{R}}^2}\right)}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle {\mathrm{R}}^2=2\times \mathrm{M}\mathrm{A}\mathrm{F}\times (1-\mathrm{M}\mathrm{A}\mathrm{F})\times {\mathrm{\beta }}^2}\end{array}}$$

R² represents the cumulative explainable variance of the selected SNPs, with N denoting the sample size and MAF indicating minor allele frequency.

We utilized IEU OpenGWAS platform (https://gwas.mrcieu.ac.uk/) to exclude SNPs associated with confounding factors (P < 1 × 10^-5), encompassing dietary behaviors, obesity traits, and body composition features, to ensure a reliable MR analysis of causal relationships.^[21]

2.4. Statistics analysis

Statistical analysis was conducted using both R 4.3.3 software and the R software package, “Two Sample MR.” For the primary analysis, we utilized inverse variance weighted (IVW) analysis, weighted median, and MR Egger. IVW analysis is a statistical approach commonly used in MR studies to estimate the combined effect of genetic variants on complex traits by aggregating the Wald ratios.^[22] We utilized the IVW analysis as our primary tool for estimating causal effects. In the IVW analysis, a P-value <.05 is typically considered significant for the effect estimate, indicating a significant causal relationship between the exposure factor and the outcome. Employing the weighted median approach to investigate the SNP impact with half-effective IVs yields an unbiased estimation, whereas the MR-Egger regression model generates a robust estimate regardless of IV effectiveness.^[23] Given that IVW analysis assumes the effectiveness of all genetic variations, it is prone to pleiotropy bias. Therefore, we employed the weighted median and MR Egger methods as complementary analytical tools to address this issue.

Sensitivity analysis was conducted to assess the robustness of the MR results. Employing the intercept of the MR-Egger regression model to assess pleiotropy, a significant P-value (P < .05) indicated the potential presence of horizontal pleiotropy among genetic IVs.^[24] To assess the robustness of the results, we employed a leave-one-out approach to assess the potential influence of individual SNPs on the results. This method involves systematically excluding each SNP and subsequently performing an MR analysis to determine the effects of each SNP on the overall outcome.^[25] Utilizing the MR-PRESSO method, we could effectively conduct an overall test to assess the presence of horizontal pleiotropy in our data. Once horizontal pleiotropy is confirmed, outlier testing was performed. This step allowed us to identify anomalous values that may contribute to horizontal pleiotropy. After removing outliers with a P-value of <.05, we conducted a distortion test to more accurately evaluate the reliability of the original data. Furthermore, we utilized the Cochran Q test to investigate potential heterogeneity among our data.

3. Results

Our comprehensive MR analysis explored the relationships between 31 blood biomarkers and 4 urine biomarkers with 4 types of DSCs. We identified 14 significant causal associations. As shown in Figure 2, the forest plot illustrates the significant causal effects of biomarkers on DSCs based on IVW analysis. An aggregated funnel plots are provided in Figure 3. An aggregated scatter plot is shown in Figure 4.

Figure 2.

Forest plot showing the significant causal effect of biomarkers on digestive system cancers calculated using IVW methods. DSC = digestive system cancer, SNP = single nucleotide polymorphism, IVW = inverse variance weighted, se = standard error, OR = odds ratio, CI = confidence interval.

Figure 3.

The aggregate funnel plot of the significant casual effect of biomarkers on digestive system cancers. (A) Funnel plot of SNPs associated with albumin and their risk of esophageal cancer. (B) Funnel plot of SNPs associated with cholesterol adjstatins and their risk of esophageal cancer. (C) Funnel plot of SNPs associated with IGF-1 and their risk of esophageal cancer. (D) Funnel plot of SNPs associated with testosterone and their risk of esophageal cancer. (E) Funnel plot of SNPs associated with triglycerides and their risk of esophageal cancer. (F) Funnel plot of SNPs associated with urate and their risk of esophageal cancer. (G) Funnel plot of SNPs associated with alkaline phosphatase and their risk of gastric cancer. (H) Funnel plot of SNPs associated with IGF-1 and their risk of colorectal cancer. (I) Funnel plot of SNPs associated with non-albumin protein and their risk of colorectal cancer. (J) Funnel plot of SNPs associated with phosphate and their risk of colorectal cancer. (K) Funnel plot of SNPs associated with creatinine and their risk of pancreatic cancer. (L) Funnel plot of SNPs associated with non-albumin protein and their risk of pancreatic cancer. (M) Funnel plot of SNPs associated with total protein and their risk of pancreatic cancer. (N) Funnel plot of SNPs associated with urate and their risk of pancreatic cancer. SNP = single nucleotide polymorphism, IGF-1 = insulin-like growth factor-1.

Figure 4.

The aggregate scatter plot showing the significant causal effect of biomarkers on digestive system cancers calculated using the IVW, MR-Egger and weighted median methods. IVW = inverse variance weighted.

3.1. Esophageal cancer

Employing IVW analysis as the primary analytical approach, we identified potential causal associations between the 6 blood and urine biomarkers and esophageal cancer. Among them, blood IGF-1 (odds ratio [OR] = 1.228, 95% confidence interval [CI]: 1.035–1.457, P = .019) and blood testosterone (OR = 1.551, 95% CI: 1.029–2.338, P = .036) were positively associated with the risk of esophageal cancer. Conversely, blood triglycerides (OR = 0.815, 95% CI: 0.679–0.978, P = .028), blood albumin (OR = 0.743, 95% CI: 0.581–0.949, P = .017), blood urate (OR = 0.822, 95% CI: 0.683–0.989, P = .038), and blood cholesterol (OR = 0.728, 95% CI: 0.585–0.906, P = .019) were negatively associated with esophageal cancer risk. These significant effects were corroborated by several alternative MR techniques, including MR Egger and weighted median, which enhanced the robustness of our findings (Table S2, Supplemental Digital Content).

3.2. Gastric cancer

Among all biomarkers, only blood alkaline phosphatase showed a significant causal association with gastric cancer, with higher levels conferring a protective effect (OR = 0.816, 95% CI: 0.702–0.949, P = .008). This result was further validated by the supplemental MR analyses (Table S2, Supplemental Digital Content).

3.3. Colorectal cancer

We found evidence that higher levels of blood phosphate (OR = 1.149, 95% CI: 1.035–1.275, P = .009) and IGF-1 (OR = 1.216, 95% CI: 1.125–1.314, P < .001) increased the risk of colorectal cancer. In contrast, higher levels of blood non-albumin protein were associated with a reduced risk (OR = 0.869, 95% CI: 0.792–0.953, P = .003). The application of MR-Egger and weighted median methods further strengthened the credibility and robustness of these findings (Table S2, Supplemental Digital Content).

3.4. Pancreatic cancer

Through IVW analysis, we identified significant positive causal relationships between pancreatic cancer and blood creatinine (OR = 1.228, 95% CI: 1.006–1.497, P = .043), blood non-albumin protein (OR = 1.352, 95% CI: 1.096–1.668, P = .005), blood total protein (OR = 1.329, 95% CI: 1.028–1.791, P = .030) and blood urate (OR = 1.230, 95% CI: 1.015–1.490, P = .035). We further validated these significant findings using MR-Egger and weighted median methods (Table S2, Supplemental Digital Content).

3.5. Sensitivity analysis

Sensitivity analyses generally supported the robustness of our findings. Cochran Q test indicated no significant heterogeneity for the majority of the associations (P > .05; Table 2). The MR-PRESSO global test detected potential horizontal pleiotropy only in the association between blood total protein and pancreatic cancer (P = .044); however, the causal estimate remained significant after outlier removal. The MR-Egger intercept test showed no significant pleiotropy for the other 13 associations (P > .05). Leave-one-out analysis confirmed that no single SNP was driving the causal estimates (results summarized in main text, details in Figure S1, Supplemental Digital Content). The funnel plot presented a symmetrical distribution of the genetic associations (Fig. 3). This pattern provides no clear evidence of directional pleiotropy and lends support to the robustness of our findings. The scatter plot suggests an observable linear relationship between the genetic associations for the exposure and the outcome (Fig. 4). This overall pattern provides visual support for a potential positive causal effect.

Table 2.

The results of sensitivity analyses of the significant causal effect of biomarkers on digestive system cancers.

Outcome	Exposure	MR-PRESSO Global P	IVW estimates		MR-Egger pleiotropy test
			Cochran Q	P	MR-Egger intercept	P
Esophageal cancer	Albumin	.120	185.93	.126	0.011	.105
	Cholesterol adjstatins	.165	217.81	.159	−0.001	.982
	IGF-1	.225	287.94	.204	−0.001	.864
	Testosterone	.109	79.293	.094	−0.009	.419
	Triglycerides	.137	206.99	.163	0.003	.486
	Urate	.096	227.53	.105	0.001	.850
Gastric cancer	Alkaline phosphatase	.924	228.04	.936	−0.009	.052
Colorectal cancer	IGF-1	.302	236.77	.298	−0.004	.074
	Non-albumin protein	.059	235.08	.067	−0.002	.540
	Phosphate	.124	142.54	.122	−0.001	.954
Pancreatic cancer	Creatinine	.382	273.95	.405	−0.004	.472
	Non-albumin protein	.228	248.36	.234	0.001	.896
	Total protein	.044	235.49	.039	0.012	.099
	Urate	.609	195.72	.611	0.008	.010

IVW = inverse variance weighted.

3.6. Reverse direction analysis

In the reverse MR validation, we adjusted the significance threshold for SNPs selection, ensuring that each set of analyses included at least 3 SNPs for analysis. We conducted reverse MR analyses for all significant causal relationships identified, and the results indicated that there were no causal associations between genetic predisposition to DSCs and the aforementioned identified biomarkers (Table S3, Supplemental Digital Content).

4. Discussion

In this large-scale, bidirectional 2-sample MR study, we systematically evaluated the causal relationships between 35 blood and urine biomarkers and 4 major DSCs. We identified 14 significant associations, most of which were robust in extensive sensitivity analyses. Our findings provide novel insights into the etiological roles of these easily measurable biomarkers. Our rigorous sensitivity analyses of the 14 potential causal relationships have revealed that our findings are largely unaffected by horizontal pleiotropy, with the exception of the MR-PRESSO analysis between total protein and pancreatic cancer, which yielded a P-value of .44. The absence of heterogeneity in all other analyses further strengthens the robustness and reliability of our study.

IGF-1 holds a pivotal role in numerous biological processes including cell proliferation, differentiation, metabolism, apoptosis, and angiogenesis, all of which are potentially linked to cancer risk.^[26] In extensive case-control investigations and comprehensive meta-analyses, pre-diagnostic circulating levels of IGF-I have exhibited a positive correlation with the incidence of colorectal cancer, breast cancer, and prostate cancer.^[27–29] Our research identified a positive causal relationship between IGF-1 and the risk of colorectal cancer, thereby validating this finding. Current evidence indicates that the IGF system represents one of the key pathways involved in the pathogenesis and progression of colorectal cancer. The binding of IGF-1 to the α-subunit of IGF-1R induces phosphorylation of tyrosine residues and activates its intrinsic tyrosine kinase activity, thereby regulates the expression of genes associated with apoptosis, autophagy, and cell proliferation.^[30] Furthermore, studies have shown that the expression level of the IGF system is positively correlated with venous invasion, liver metastasis, and tumor size, which consequently impacts patient survival.^[31,32]

Our study demonstrated a positive causal relationship between higher IGF-1 levels and the risk of esophageal cancer. A prospective investigation revealed heightened blood levels of IGF-1 among patients diagnosed with Barrett esophagus and esophageal adenocarcinoma.^[33] Furthermore, IGF1 levels show significant correlations with both the depth of invasion and pathological stage, and patients with high IGF1 expression exhibit lower survival rates than those with low expression.^[34] Our results provide further support for these earlier findings. In contrast to this finding, a large-scale prospective analysis based on UK Biobank suggested a link between elevated IGF-1 levels and a lower risk of esophageal cancer.^[26] However, this association failed to persist after correction for multiple testing and exhibited bias toward reverse causation.^[26] Nonetheless, further experimental validation is necessary to confirm these findings.

Additionally, our study revealed a positive causal relationship between blood urate levels and the risk of pancreatic cancer. Concordantly, researchers have uncovered a heightened risk of pancreatic cancer among individuals with elevated blood urate levels, which is particularly pronounced among females in gender-stratified analyses.^[35] Existing evidence indicates that serum uric acid is associated with inflammatory stress, which is closely linked to cancer development.^[36] An in vitro experiment demonstrated that cyclooxygenase-2 (COX-2) is positively correlated with microvascular density, promoting the growth of pancreatic cancer cells.^[35] Ohtsubo et al found that blood urate modulates the expression of COX-2 through xanthine oxidoreductase, which may explain the association between blood urate and cancer.^[37] However, it is noteworthy that our study uncovered a causal relationship between elevated urate levels and a decreased risk of esophageal cancer. In fact, a growing body of evidence suggests that maintaining urate homeostasis is crucial to health. Both hypouricemia and hyperuricemia have been linked to an increased cancer risk. Numerous cohort studies have shown a U-shaped association between blood urate levels and increased risk of all-cause mortality, including cancer.^[38,39] Balancing blood urate levels appears to be necessary for reducing the risk of severe diseases, including cancer. This finding provides an intriguing insight into the potential role of uric acid in cancer prevention, although further investigation is necessary to elucidate the underlying mechanisms and clinical implications.

Furthermore, we found a significant causal relationship between lower blood cholesterol levels and decreased risk of esophageal cancer. A Korean cohort study revealed that high HDL-C levels are correlated with esophageal cancer risk.^[40] Additionally, for individuals with a family history of esophageal cancer, high total cholesterol and LDL-C levels are significantly associated with an increased risk of malignant lesions.^[41] The inconsistencies may stem from the opposing effects of LDL and HDL. However, it is noteworthy that our study did not find a significant causal relationship between LDL and HDL levels in esophageal cancer, necessitating further stratified studies to validate our findings.

Moreover, our study revealed a significant correlation between an elevated predisposition to blood albumin levels and a decreased risk of esophageal cancer. Albumin is recognized as playing a pivotal role in the resistance of gastrointestinal mucosa to lipid peroxidation.^[42] Through a prospective analysis, researchers identified a notable inverse association between pre-diagnostic albumin levels and the risk of developing colorectal cancer.^[43] Our conclusions support this finding. Our findings also suggest a potential causal relationship between blood phosphate levels and the risk of colorectal cancer. Blood phosphate is essential for ATP formation, kinase/phosphatase signaling, as well as the synthesis of lipids, carbohydrates, and nucleic acids.^[44] It has been confirmed that cancer cells exposed to an elevated phosphorus environment promote blood vessel formations and endothelial cell migrations.^[45]

The availability of genome-wide resources from Sinnott-Armstrong N et al served as a pivotal foundation for our study, providing invaluable insights and data sources.^[15] The biomarkers identified in this study often reflect fundamental biological processes such as metabolic syndrome (e.g., IGF-1, urate), nutritional status (e.g., albumin), and organ function (e.g., alkaline phosphatase, creatinine). Their causal roles suggest that they are not merely secondary markers but may be involved in the pathogenic pathways of DSCs. These findings hold translational relevance: such biomarkers could enhance risk stratification models to identify high-risk subgroups for tailored screening. For instance, individuals with elevated IGF-1 or urate levels may warrant intensified endoscopic surveillance.

Our study has several limitations. First, the GWAS data were primarily from individuals of European ancestry, which limits the generalizability of our findings to other ethnic populations, such as Asian cohorts where the incidence and genetic architecture of DSCs may differ. Future multi-ancestry studies are essential. Second, we were unable to perform stratified analyses by sex, cancer subtypes, or age due to data availability. Such stratification might reveal more nuanced associations. Third, despite rigorous sensitivity analyses, residual pleiotropy can never be entirely ruled out in MR studies. Finally, the number of biomarkers investigated, while substantial, is not exhaustive, and other important markers may exist.

5. Conclusion

We conducted a 2-sample bidirectional MR study to investigate the association between 35 blood and urine biomarkers and 4 types of DSCs. Our findings reveal 14 significant causal relationships. Clinically, our findings could assist in risk stratification of individuals, enabling more targeted screening and surveillance measures. Additionally, the identified biomarkers may serve as potential diagnostic markers for the early detection of cancer, which is crucial for improving treatment outcomes and patient survival.

Acknowledgments

We extend our gratitude to all the researchers and participants of the UK Biobank and the other GWAS consortia for making their summary data publicly available.

Author contributions

Data curation: Zhilei Zhang, Yuming Jia.

Supervision: Li Peng.

Validation: Changkun Yang, Ben Xing.

Visualization: Changkun Yang, Ben Xing.

Writing – original draft: Changkun Yang.

Writing – review & editing: Changkun Yang, Ben Xing.