Exploring the causality and pathogenesis of Type 2 diabetes in gastric cancer based on Mendelian randomization and transcriptome data analyses

doi:10.21203/rs.3.rs-4171771/v1

Download PDF

Research Article

Exploring the causality and pathogenesis of Type 2 diabetes in gastric cancer based on Mendelian randomization and transcriptome data analyses

https://doi.org/10.21203/rs.3.rs-4171771/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Introduction: Gastric cancer (GC) is the fifth most common type of cancer in the world, and the relationship between type 2 diabetes mellitus (T2DM) and cancer risk has been a focus of attention. However, the causal relationship between the two and its pathogenesis are still highly controversial. Methods: Two-sample bidirectional Mendelian randomization (MR) and meta-analysis were used to explore the potential causal relationship between T2DM and GC. The sensitivity of the MR analysis was assessed by Cochran's Q test and the leave-one-out test. Next, transcriptome analyses of T2DM and GC patients were performed using The Cancer Genome Atlas and Gene Expression Omnibus databases to identify genes common to T2DM and GC and to construct a risk score and a prediction model for GC risk in combination with clinical features. KM curves and Cox regression were used for survival analysis, and the AUC values of the ROC curves were used for model evaluation. Mechanisms were explored by gene enrichment analysis and tumor immune microenvironment analysis. RESULTS: MR analysis revealed a causal relationship between T2DM and GC at the genetic level. In addition, the meta-analysis and sensitivity analysis confirmed the robustness of the current MR results. We constructed a prognostic risk score consisting of three T2DM-related genes (CST2, PSAPL1, and C4orf48). KM curves and Cox regression analyses suggested that OS was worse in patients with high risk scores (p < 0.05), and the constructed nomogram model had good predictive ability (training set: AUC = 0.786 and validation set: AUC = 0.757). Conclusion: Bidirectional MR two-sample analysis revealed that T2DM was a high-level risk factor for GC. Transcriptome data analysis revealed that the risk score, which reflects the expression of T2DM-related genes, can be used as a new marker to predict the prognosis of patients with coexisting T2DM. KEYWORDS: Gastric cancer, Type 2 diabetes, Mendelian randomization, Transcriptomic data, prevention

Gastric cancer

Type 2 diabetes

Mendelian randomization

Transcriptomic data

prevention

Gastric cancer (GC) is the fifth most common type of cancer globally (5.6% of total cases) and the fourth leading cause of cancer deaths worldwide(Sung et al., 2021). Despite the overall decline in global GC incidence and mortality due to the implementation of various tumor screening methods and anticancer strategies, reports over the past decade indicate that the incidence of GC remains high or is on the rise in certain regions and populations and that the global burden of this malignancy is projected to increase by 62% by 2040(Guan et al., 2023; Huang et al., 2023) (Thrift et al., 2023). The treatment of GC is mainly radical surgery supplemented with radiotherapy, chemotherapy and drug therapy. However, research on the treatment of GC is still in progress, with the aim of further identifying the risk factors for GC intervention, clarifying the pathogenesis of GC, and improving the diagnosis and treatment of GC(Guan et al., 2023; Lordick et al., 2022; Machlowska et al., 2020; Thrift & El-Serag, 2020). Current studies have shown that Helicobacter pylori infection is the main cause of distal GC, and nearly 90% of patients are infected with Helicobacter pylori before developing GC. Other potential risk factors for GC include EBV infection and autoimmune gastritis(Machlowska et al., 2020; Thrift et al., 2023). In addition, the development of GC may be associated with T2DM(Chen et al., 2017; Choi et al., 2022; Hosseini et al., 2019).

Type 2 diabetes mellitus (T2DM) has become a common public health burden worldwide, with a global prevalence of 536 million adults with T2DM in 2021, and the prevalence is projected to reach 784 million by 2045(Sun et al., 2022) (Eisenberg et al., 2023; Sun et al., 2022). Several studies have shown that people with T2DM have an increased risk of developing multiple cancers(Saarela et al., 2019), and Hans Scherübl et al. showed that people with T2DM have an increased risk of gastrointestinal adenocarcinoma(Scherübl, 2023). However, researchers have failed to reach a consensus on T2DM as a potential risk factor for GC, and studies exploring the correlation between the two remain controversial. A meta-analysis exploring the relationship between diabetes and GC risk conducted by Guo et al. suggested that T2DM may increase the risk of GC by 14% (Guo et al., 2022). Similarly, a large Finnish cohort study conducted by Saarela et al. revealed a significant association between T2DM and the incidence of GC(Saarela et al., 2019). However, several observational studies have yielded conflicting results, and a meta-analysis of data from the Stomach Cancer Pooling (StoP) Project by Dabo et al. showed no association between diabetes and the risk of GC and GC(Dabo et al., 2022). A cohort study of Chinese men and women by Xu et al. also showed no association between T2DM and an increased risk of GC in either men or women(Xu et al., 2015). As the results of observational studies are susceptible to many potential confounding factors, more appropriate research methods are needed to explore the causal relationship between the two factors(Saarela et al., 2019; Xu et al., 2015).

Mendelian randomization (MR), a widely used method of causal inference in which exposure-related genetic variants can be used as instrumental variables (IVs) to determine the causal effect of exposure on outcomes, is not affected by common confounders such as the acquired environment and lifestyle behavioral habits and can maximize its effectiveness in minimizing reverse causal effects(Qi et al., 2023; Yu et al., 2020). Although there have been MR studies revealing causal associations between GC and DM pathogenesis(Xu et al., 2023), there is a lack of valid evidence exploring the causal associations between GC and the pathogenesis of DM subtypes (T2DM). In addition, most of the existing studies have focused only on the association between GC and T2DM based on limited clinical findings and the disease itself, without further research on the association between T2DM and GC prognosis and its genetic level.

Therefore, the aim of this paper was to explore the possible causal relationship between GC and T2DM through MR analysis of pooled statistics from genome-wide association studies (GWASs) and to investigate the biological mechanisms and disease prognosis through the common differentially expressed genes (DEGs) of GC and T2DM.

The general design of our work is shown in Fig. 1. The data sources for the MR and transcriptome data are publicly available online.

I. Mendelian randomization analysis

01. Data sources

The GWAS data for patients with T2DM (ebia-GCST90018926, bbj-a-77) and GC (ebi-a-GCST90018849) were obtained from the IEU Open GWAS project (https://gwas.mrcieu.ac.uk/). In the T2DM dataset, bbj-a-77 was used as the validation cohort. The final data populations obtained were derived from European and Asian populations, for which the summary information is provided in Table 1. Informed consent for the relevant data had already been obtained in the original study; therefore, this aspect of the ethical approval was waived in this study.

Table 1

Brief information on the GWAS database in the MR study
Data source	Phenotype	Sample size	Cases	Population	Adjustment
ebi-a-GCST90018926	Type2diabetes	490089	38841	European	NA
bbj-a-77	Type2Diabetes	191764	36614	East Asian	Males and Females
ebi-a-GCST90018849	Gastric cancer	476116	1029	European	NA

02. Conditioning of SNPs as instrumental variables (IVs)

All MR methods are based on three core assumptions to minimize the impact of bias on MR estimation (VanderWeele et al., 2014): assumption (i): genotypes must be associated with the exposure factor to be studied; assumption (ii): genotypes must be independent of confounding factors; and assumption (iii): genotypes can only be associated with the outcome by influencing the exposure factor to be studied (Figure S1).

03. SNP Screening

Significant SNPs were screened from the GWAS pooled data of exposure factor T2DM (with P < 5 × 10^− 8 as the screening condition, hypothesis (i)); the chain disequilibrium coefficient r² was set to 0.001, and the width of the chain disequilibrium region was set to 10,000 kb to ensure that individual SNPs were independent of each other and to exclude the influence of genetic pleiotropy on the results; SNPs associated with confounding factors and outcomes were excluded by PhenoScanner (http://www.phenoscanner.medschl.cam.ac.uk/) to exclude SNPs associated with confounders and outcome (hypotheses (ii) and (iii)); in addition, instrumental variables with F-statistics greater than 10 were used to reduce weak instrumental variable bias. The above screened associated SNPs were extracted from the GWAS pooled data of the outcome variables, and a minimum r² > 0.8 was set. Information from the above two datasets was summarized, and the same steps were applied in reverse causality to screen for associated SNPs.

04. Verification of causality

Random-effects inverse-variance weighted (IVW) regression and MR‒Egger regression were used to validate the causal relationship between exposure and outcome, using SNPs as instrumental variables. The random-effects inverse variance-weighted method was used as the primary outcome to be reported, and the MR‒Egger regression method was used as a supplementary analysis method. These methods were implemented using the Two Sample MR package in R4.1.0 software with a test level of α = 0.05.

05. Heterogeneity, pleiotropy and sensitivity analysis

SNP heterogeneity was determined by calculating the Cochran's Q score; if P < 0.05, the results were considered heterogeneous (Bowden et al., 2017). I² (I-squared) is another statistic used to measure heterogeneity, and I²>50% indicates that there is a certain degree of heterogeneity; it is calculated as I² = (Q -Q_df)/Q (Bowden et al., 2016). The pleiotropy test is mainly used to test whether there is horizontal pleiotropy in multiple IVs; the MR‒Egger intercept term is commonly used for pleiotropy analysis, and the MR‒Egger regression model of the intercept term does not differ significantly from 0 (P > 0.05), indicating that the IVs do not have pleiotropy(Bowden et al., 2018). The leave-one-out sensitivity test is mainly used to calculate the MR results of the remaining IVs after removing the IVs one by one; if the difference between the estimated MR results of the other IVs and the total results after removing a certain IV is very large, the MR results are sensitive to that IV. Leave-one-out analysis was performed on the IVs by gradually removing each SNP and reanalyzing the remaining SNPs to observe the magnitude of the effect of each SNP on the analysis results(Cheng et al., 2021; Y. Gao et al., 2023).

06. Meta-analysis

Meta-analysis of the ebi-a-GCST90018926 and bbj-a-77 datasets for T2DM with MR estimates for GC was performed using the meta-package for R.

II. Transcriptomic analyses

01. Data acquisition

Transcriptome RNA-Seq data of tissue samples (32 nontumor tissue samples and 375 tumor tissue samples) were obtained from the TCGA database. A total of 297 GC patients whose clinical and RNA-seq data were available were screened for prognostic analyses; microarray expression and clinical data from the GSE20966-Diabetes dataset (10 T2DM patients and 10 controls) were downloaded from the Gene Expression Omnibus (GEO) database.

02. Identification of differentially expressed genes (DEGs)

T2DM and GC deg (PMCID: PMC4402510) were identified by the "limma" R package, with a threshold |log2FC| >1 and adjusted p value < 0.05. Subsequently, weighted gene coexpression network analysis (WGCNA) was used to identify gene modules with similar expression patterns and to correlate these modules with the phenotypic data(Langfelder & Horvath, 2008).

03. Establishment and validation of a prognostic model related to gastric cancer

First, all patients were randomized into training and validation sets, and the DEGs between patients with T2DM and those with GC were calculated separately by limma; then, the intersection was taken. The module genes with the highest correlation were selected by WGCNA of the intersecting genes. Genes associated with prognosis were screened by one-way Cox regression. Patients were randomly grouped at a ratio of 6:4 in the training set and further screened for prognosis-related genes by LASSO-Cox regression, and a risk score model was constructed based on the regression coefficients. The risk score was calculated as follows: RiskScore= ∑ (gene expression level * corresponding coefficient (estimated value)). The optimal cutoff value of 0.9316 was derived according to the "survminer" software package, the samples were divided into high- and low-risk groups, and the difference in survival between the two groups was assessed by KM curve and Cox regression analyses. The clinical variables were screened by stepwise regression with the AIC as the criterion, and the included variables were characterized by selecting the smallest AIC information statistic and constructing nomogram models; column line plots of 1-, 2- and 3-year survival probabilities were generated via a nomogram. Subject operating characteristic (ROC) curves were used to further analyze the performance of the models.

04. Functional enrichment analysis and mutation and drug sensitivity analyses

To further investigate the biological mechanism of gene variants, the expression matrix of 197 patients with GC in the TCGA cohort was analyzed using the R package "limma" pairs of types to calculate the pathway enrichment scores of KEGG pathway gene sets and hallmark gene sets in each sample via GSVA, and the top 50 pathways were visualized with |t|=1 as the threshold (KEGG pathway enrichment score). The pathway enrichment score was used to analyze the difference between the high- and low-risk groups by visualizing the top 50 pathways with |t|=1 as the threshold (186 pathways in the KEGG pathway gene set analysis and 50 pathways in the hallmark gene set enrichment analysis). Subsequently, we downloaded the mutation data of TCGA-STAD patients from the TCGA data portal (https://portal.gdc.cancer.gov/) and combined these data with the pathohistology data of 297 samples with a sample size of 290. We analyzed the mutation data using the R package maftools to visualize the top 15 mutated genes in patients with STAD. Finally, we downloaded the IC50 values of 198 drugs from the GDSC database (http://www.cancerrxgene.org/) and used the R package "OncoPredict" to predict the IC50 values of each sample based on RNA-seq data and analyzed the relationships between the high- and low-risk groups in the prognostic risk score model.

05. Analysis of the tumor immune microenvironment and immunotherapeutic response

We uploaded the gene expression matrices of GC samples to the ImmuCellAI database (http://bioinfo.life.hust.edu.cn/ImmuCellAI/#!/) to calculate the immune cell infiltration of each sample and analyzed the correlation between the RiskScore and the degree of immune cell infiltration using Spearman correlation. Subsequently, we used the Wilcoxon rank sum test to analyze the variability in immune gene expression (Xie et al., 2022) between the high- and low-risk groups in the prognostic risk score model. Finally, the expression matrix of TCGA-STAD was uploaded to the TIDE database (ttp://tide.dfci.harvard.edu), and TIDE, MSI Expr Sig, Dysfunction, and Exclusion were assessed by the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm to analyze the relationships between risk score and immunotherapy response.

Statistical analysis

All data analyses in this study were performed using GraphPad Prism 8.0.1 software (GraphPad Software e, Inc.) and R software (version 4.1.3), and all MR analyses were performed with the "TwoSample MR" package. The classification performance of the model was evaluated based on sensitivity, specificity, accuracy, the subject operating characteristic (ROC) curve and the area under the ROC curve (AUC).

I. Mendelian randomization analysis

01. SNP screening

When T2DM–GC–causality was explored, 182 SNPs were ultimately included as instrumental variables (Table S1-1). In reverse causality, to explore the GC–T2DM–causality association, 8 and 7 SNPs were ultimately included as instrumental variables from ebi-a-GCST90018926 and bbj-a-77, respectively. The basic information for the SNPs is shown in Tables S1-2 and S1-3. The distribution of the F-statistics corresponding to all the included SNPs ranged from 29.61 to 353.66, which indicated that the causal association was less likely to be affected by weak instrumental variable bias.

02. Arguments for a causal relationship between diabetes and gastric cancer

T2DM-GC MR analysis was performed on the ebi-a-GCST90018926 dataset, and the results of the IVW analysis showed a causal relationship between T2DM-GC MR (OR = 0.916, 95% CI: 0.850–0.986, P = 0.020; Table 2, Fig. 2A, B). The results of the validation cohort bbj-a-77 dataset showed that there was also a causal association between T2DM and GC (OR = 0.882, 95% CI: 0.814–0.956, P = 0.002; Table 2, Fig. 2A, C). In contrast, in reverse causality, the results of the IVW analysis of the ebi-a-GCST90018926 dataset when GC was the exposure factor, as well as the results of the IVW analysis when GC was analyzed with the validation cohort bbj-a-77 dataset, showed no causal relationship between GC and T2DM(Table 2, Fig. S2).

Table 2

MR Regression Causal Association Results
Exposure	Data source	Classification	Nsnp	Methods	OR(95%CI)	P-value
T2DM-GC MR analysis
ebi-a-GCST90018926	ebi-a-GCST90018849	Type 2 diabetes	125	MR Egger	0.938(0.746–1.178)	0.583
ebi-a-GCST90018926	ebi-a-GCST90018849	Type 2 diabetes	125	IVW	0.916(0.850–0.986)	0.020
bbj-a-77	ebi-a-GCST90018849	Type 2 diabetes	57	MR Egger	0.947(0.770–1.165)	0.609
bbj-a-77	ebi-a-GCST90018849	Type 2 diabetes	57	IVW	0.882(0.814–0.956)	0.002
GC-T2DM MR analysis
ebi-a-GCST90018849	ebi-a-GCST90018926	Gastric cancer	8	MR Egger	1.320(1.178–1.478)	0.003
ebi-a-GCST90018849	ebi-a-GCST90018926	Gastric cancer	8	IVW	0.979(0.892–1.075)	0.658
ebi-a-GCST90018849	bbj-a-77	Gastric cancer	7	MR Egger	1.258(0.961–1.647)	0.156
ebi-a-GCST90018849	bbj-a-77	Gastric cancer	7	IVW	0.926(0.818–1.048)	0.222

03. Heterogeneity, pleiotropy and sensitivity analysis

When T2DM was used as an exposure factor, MR analysis based on the ebi-a-GCST90018926 dataset revealed no heterogeneity between the SNPs associated with T2DM-GC (Cochran's Q = 167.303, I² = 26%, Table 3). MR‒Egger results showed no statistically significant difference between the intercept term and 0 (P = 0.829), and it was concluded that there was no horizontal pleiotropy of SNPs. After sequentially excluding each SNP for T2DM by the leave-one-out test, the results of the analysis of the remaining SNPs were similar to those of the inclusion of all SNPs (see Figure S3), and no SNPs were found to have a large impact on the causal association estimates, suggesting that the MR results of the present study were robust. The results of the validation cohort bbj-a-77 dataset also showed no heterogeneity (Cochran's Q = 82.119, I² = 32%, Table 3) or horizontal pleiotropy (MR‒Egger P = 0.467) between SNPs associated with T2DM and GC.

In reverse causality, when GC was used as the exposure factor, there was heterogeneity in the SNPs associated with GC T2DM based on the ebi-a-GCST90018926 dataset (Cochran's Q = 51.994, I² = 87%, Table 3). The MR‒Egger leave-one-out test (see Figure S2) confirmed the robustness of the MR results of the study. In addition, the validation cohort was based on the bbj-a-77 dataset, and there was heterogeneity of SNPs associated with GC-T2DM (Cochran's Q = 33.333, I² = 82%, Table 3). The MR‒Egger and leave-one-out tests (see Figure S2) also confirmed the robustness of the MR results of the present study.

Table 3

Results of MR Heterogeneity and Horizontal pleiotropy Analysis
Data source	Classification		Heterogeneity		Horizontal pleiotropy
		Nsnp	I²(%)	Cochran's Q	Egger intercept	SE	P-value
T2DM-GC MR analysis
ebi-a-GCST90018849	Type 2 diabetes \|\| id:ebi-a-GCST90018926	125	26	167.303	-0.001	0.006	0.829
ebi-a-GCST90018849	Type 2 diabetes \|\| id:bbj-a-77	57	32	82.119	-0.006	0.009	0.467
GC-T2DM MR analysis
bbj-a-77	Gastric cancer \|\| id:ebi-a-GCST90018849	7	82	33.333	-0.053	0.022	0.063
ebi-a-GCST90018926	Gastric cancer \|\| id:ebi-a-GCST90018849	8	87	51.994	-0.056	0.010	0.001

04. Meta-analysis

To further validate our results, we combined T2DM as an exposure factor in the ebi-a-GCST90018926 and bbj-a-77 datasets estimated by meta-analysis. P = 0.5 for Cochran's Q in the heterogeneity test indicated that there was no heterogeneity between the effect estimates of the different datasets. According to the fixed effect model (random effect model) (OR = 0.900, 95% CI = 0.852–0.951, P < 0.001), there was a significant causal relationship between T2DM and GC (Fig. 2D).

II Transcriptomic analyses

01. Identification of T2DM − related DEGs in patients with gastric cancer

A total of 1335 DEGs were screened from the TCGA GC dataset, 2870 DEGs were screened from the T2DM dataset (GSE20966) (Fig. 3A, B), and 169 DEGs common to both the GC- and T2DM-related datasets were identified (Fig. 3C). A total of 80 genes in the lime-green module with the highest correlations were further selected by clustering via WGCNA (Figs. 3D, E, F).

02. Risk modeling

Three genes associated with prognosis were identified using one-way Cox regression (Fig. 4A), and the aforementioned genes were further screened by the LASSO model. Finally, three genes (CST2, PSAPL1, and C4orf48) were identified to construct the gene risk score prognostic model. The risk score was calculated as RiskScore = Expression level of CST2*(0.0637) + Expression level of PSAPL1*(0.1973) + Expression level of C4orf48*(0.1742) (Fig. 4B, E). The samples were divided into high- and low-risk groups according to the optimal cutoff value of 0.9316 obtained from the "survminer" package, and the data were randomly divided into training and validation sets at a ratio of 6:4 (Table 4). There were 178 patients in the training set and 119 patients in the validation set, and the P value for each variable in the between-group analysis of variance was > 0.05, indicating that the baseline conditions of the patients in the training set and the validation set were close to each other and comparable between the groups. In the training set, the median survival time of patients in the high-risk group according to the prognostic risk score was 22.3 months, and that of patients in the low-risk group was 46.9 months. K‒M curves revealed that the prognostic risk score of the high-risk group was significantly correlated with a decrease in OS (p = 0.0089) (Fig. 4C). In the validation set, the median survival time for the high-risk group according to the prognostic risk score model was 18.47 months, and the median survival time for the low-risk group was 60.37 months. Kaplan‒Meier curves showed that the prognostic risk score model was significantly associated with worsening OS (p = 0.00042) (Fig. 4D).

In addition, in the training set, high risk was a statistically significant risk factor for OS in the univariate analysis (HR = 1.914, 95% CI 1.167–3.138, p = 0.010) (Fig. 5A). According to the multifactorial analysis, high risk (HR = 2.040, 95% CI 1.157–3.596, P = 0.014) was a risk factor for OS after multifactorial adjustment (Fig. 5C). In the validation set, the high-risk group was a significant risk factor for OS in the univariate analysis (HR = 2.976, 95% CI 1.577–5.615, P < 0.001) (Fig. 5B). According to the multifactorial analysis, after multifactorial adjustment, high risk (HR = 4.957, 95% CI 2.114–11.624, P < 0.001) was a significant risk factor for OS (Fig. 5D). The ROC curves showed that the risk score could be used as a sensitive indicator for predicting OS in GC patients, as shown in the figure. The AUC values of the column-linear graph model were 0.786, 0.779, and 0.744 for years 1, 2, and 3 of the training set, respectively (Fig. 5E, F), and the AUC values of the validation set were 0.757, 0.808, and 0.73 for years 1, 2, and 3, respectively (Fig. 5G, H).

Table 4

Baseline data table: information on prognostic risk score model groupings and clinical information on patients with GC
Variables	Total (n = 297)	Training cohorts (n = 178)	testing cohorts (n = 119)	p
Risk_Group, n (%)				0.776
high	173 (58)	102 (57)	71 (60)
low	124 (42)	76 (43)	48 (40)
Age, n (%)				0.773
~ 65	133 (45)	78 (44)	55 (46)
66~	164 (55)	100 (56)	64 (54)
Gender, n (%)				0.995
Female	111 (37)	66 (37)	45 (38)
Male	186 (63)	112 (63)	74 (62)
Pathologic_stage, n (%)				0.999
I	37 (12)	22 (12)	15 (13)
II	98 (33)	59 (33)	39 (33)
III	133 (45)	80 (45)	53 (45)
IV	29 (10)	17 (10)	12 (10)
Histologic_grade, n (%)				0.325
G1/G2	111 (37)	62 (35)	49 (41)
G3	186 (63)	116 (65)	70 (59)
Radiotherapy, n (%)				0.511
NO	258 (87)	157 (88)	101 (85)
YES	39 (13)	21 (12)	18 (15)
Histologic_type, n (%)				0.994
SA Diffuse Type	52 (18)	30 (17)	22 (18)
SA NOS	96 (32)	59 (33)	37 (31)
SA Signet Ring Type	9 (3)	6 (3)	3 (3)
SIA Mucinous Type	17 (6)	11 (6)	6 (5)
SIA NOS	59 (20)	35 (20)	24 (20)
SIA Papillary Type	4 (1)	2 (1)	2 (2)
SIA Tubular Type	60 (20)	35 (20)	25 (21)
Residual_tumor, n (%)				0.437
R0	251 (85)	153 (86)	98 (82)
R1	12 (4)	6 (3)	6 (5)
R2	11 (4)	8 (4)	3 (3)
RX/Unknown	23 (8)	11 (6)	12 (10)
Tumor_location, n (%)				0.981
Antrum/Distal	113 (38)	67 (38)	46 (39)
Cardia/Proximal	42 (14)	26 (15)	16 (13)
Fundus/Body	109 (37)	66 (37)	43 (36)
Gastroesophageal Junction	33 (11)	19 (11)	14 (12)
Chemotherapy, n (%)				0.741
NO	160 (54)	94 (53)	66 (55)
YES	137 (46)	84 (47)	53 (45)

03. Analysis of functional characteristics

To further understand the potential function of DEGs in GC in the high/low-risk groups according to the GC Prognostic Risk Scoring Model, we performed GSVA enrichment analyses of the KEGG and hallmark gene sets. According to the KEGG gene set, the low-risk group was significantly enriched in apoptosis (APOPTOSIS), cell cycle (CELL_CYCLE) and other signaling pathways (Fig. 6A); according to the hallmark gene set, the low-risk group was significantly enriched in P53 (P53_PATHWAY) and PI3K-AKT-MTOR (PI3K_AKT_MTOR_SIGNALING) (Fig. 6B). Subsequently, we utilized the available somatic mutation data to assess the mutation profiles of high- and low-risk GC patients. Missense_Mutation was the most frequently occurring mutation type. The mutation rate of ARID1A was greater than 10% in both the high- and low-risk groups, and in the low-risk group, the mutation rate of ARID1A was greater than that in the high-risk group (Fig. 6C). Subsequently, we performed a drug sensitivity analysis showing a total of 147 drugs with statistically significant differences in drug sensitivity between the high- and low-risk groups (P < 0.05). The box-and-line plot demonstrated the three most significant drugs with significantly greater IC₅₀ in the high-risk group than in the low-risk group, suggesting that these drugs may be sensitive drugs for the treatment of GC patients in the high-risk group.

04. Tumor immune cell infiltration analysis, immunogenetic analysis and immunotherapy response prediction

We analyzed immune cell infiltration in GC via the ImmuCellAI database, and Spearman correlation analysis revealed that the risk score was negatively correlated with the degrees of infiltration of multiple cell types, including CD8⁺ T cells and exhausted cells (P < 0.05), and positively correlated with the degrees of infiltration of monocytes, DCs, and NKT cells, among others (P > 0.05) (Fig. 7A). Subsequently, we used Wilcoxon rank sum test analysis to show that the gene expression levels of CD276, HHLA2 and other genes were significantly elevated in the high-risk group according to the prognostic risk scoring model (P < 0.001, Fig. 7B). Finally, we used the Wilcoxon rank sum test to analyze the associations between the high- and low-risk groups according to the prognostic risk score and response to immunotherapy, and the MSI Expr Sig value was greater in the low-risk group according to the prognostic risk score (Fig. 7C).Figure 7 Correlation analysis of prognostic risk scores with immunization A: Correlation analysis of prognostic risk scores with immune cell infiltration. b: Immune infiltrating cell abundance based on box line plot between high and low risk groups. c: Difference in immunotherapy between high and low risk groups.

T2DM has become a global health problem(Ahmad et al., 2022). Several recent studies have suggested that there may be a potential relationship between T2DM and cancer incidence or mortality(Huo et al., 2023; Suh & Kim, 2019; Zhang et al., 2021). However, the relationship between T2DM and the development of GC is unclear, and the clinical features and pathophysiology of overlapping cases of GC with T2DM are unknown(Jenkins et al., 2024). In this study, we assessed the causal relationship between T2DM and GC risk by Mendelian randomization analysis. Our MR results suggest that genetic susceptibility to T2DM is associated with an increased risk of individuals developing GC. The T2DM-associated DEG risk score can classify GC patients into high- and low-risk categories for predicting their survival. The correlations between the T2DM-associated DEG risk score and the level of tumor-infiltrating immune cells, drug sensitivity analysis, and the prediction of immunotherapy response were further validated.

There is growing evidence that T2DM is associated with the incidence of cancer in multiple organs; previous studies have suggested that T2DM may be associated with the development of bladder, breast, colorectal, endometrial, liver, and pancreatic cancers and non-Hodgkin's lymphoma, among other diseases(Georgescu et al., 2023; Jenkins et al., 2024; Sung et al., 2021). A recent large-scale prospective cohort study based on the UK Biobank revealed that T2DM and elevated blood glucose levels may be risk factors for the development of upper gastrointestinal tumors (esophageal and gastric cancers) and that individuals with a high genetic risk of T2DM have a significantly increased risk of developing upper gastrointestinal tumors (Cao et al., 2024). These results are to some extent consistent with our findings that T2DM is a risk factor for GC. Compared with the results of previous observational studies, our MR study may provide stronger evidence. Several potential confounders in observational studies, including time period, environmental exposure, and population-specific genetics, may affect the results to a greater or lesser extent. However, in MR studies, SNP loci of genes are considered natural instrumental variables because they are not altered by external environmental or behavioral factors(Davey Smith & Ebrahim, 2005). The use of SNP loci associated with exposure factors can avoid the influence of confounding factors and ensure the accuracy of the results; moreover, the application of MR controls the influence of confounding factors and reverse causation on the estimation so that reliable causal effect estimations can be obtained based on observational studies(Bastarache et al., 2022; Xu et al., 2022).

In addition to MR analysis, we also performed transcriptome analysis, a genetic validation method used to analyze the relationship between two diseases. We identified 169 DEGs associated with T2DM and GC by differential expression analysis and then used the LASSO algorithm to identify three key therapeutic targets (CST2, PSAPL1, and C4orf48). Among the three genes mentioned above, CST2 and PSAPL1 were shown to be correlated with the prognosis of patients with GC (Li et al., 2022) (Luo et al., 2023). The present study supports the above experimental findings, but a correlation between C4orf48 and GC has not been previously reported. We also constructed a risk score to predict the prognosis of patients with GC combined with T2DM based on T2DM-related DEGs and nomograms. Univariate and multivariate Cox models, KM survival analysis and ROC curve analysis further confirmed the predictive accuracy of the risk score. Recent studies have shown that ARID1A-deficient GC is a unique type of GC that is sensitive to both para-chemotherapy and anti-PD-1 immunotherapy. In this study, the rates of ARID1A mutation were greater than 10% in both the high- and low-risk groups, indicating that ARID1A may be a potential target for treating GC combined withT2DM.

An increasing number of studies have confirmed that the development of GC is closely related to the TME(Zhang et al., 2023). Our study revealed that the risk score was negatively correlated with the degrees of cellular infiltration of CD8⁺ T cells, CTLs and other cells, which act as tumor suppressors, and the current results of TME cellular infiltration are consistent with the results of the survival analyses of the above molecules. We then performed an analysis of immune genes and found that the expression levels of genes such as CD276 and HHLA2 were significantly elevated in the high-risk group according to the prognostic risk score model. CD276 promotes the migration of GC cells, and HHLA2 overexpression is a novel biomarker of malignant status and poor prognosis in GC, similar to the results of this study (Q. Gao et al., 2023; Mansorunov et al., 2022).

There are several flaws and shortcomings in this study. First, the database of patients with GC in our study included only Europeans. Although we compared the results of data from populations with diabetes in East Asia and Europe, and no significant differences were observed in the study outcomes and meta-analyses of the two populations, the heterogeneity of the exposure and outcome populations may still introduce an ethnographic bias. Second, T2DM is associated with an increased risk of developing GC, but the underlying mechanisms are unclear and require further study. Third, similar to other studies, individual-level data were not obtained for further exploration based on the pooled data analysis used by GWAS.

In this study, we explored the bidirectional causal relationship between T2DM and GC using MR analysis and performed a meta-analysis to increase the reliability of our results. The results showed that T2DM was associated with an increased risk of GC. In addition, the analysis of transcriptomic overlapping hub genes provides a new perspective for the precise diagnosis and treatment of patients with T2DM combined with GC and provides a theoretical basis for future studies of common underlying mechanisms and potential therapeutic targets.

Competing interests

The authors declare no competing interests.

Conflict of interest

Data were collected from public databases. And there weas no ethical approval necessary

Consent for publication

Not applicable

Funding

This work was supported by Supports for Shandong First Medical University Research Grant Fund (342601).

Author Contribution

JKZ, JYM and YG conceived and designed the study. JYM, SFH and YG collected the data. JYM and YG analyzed the data and results. JYM and SFH completed the writing of the manuscript. JKZ, SFH, SCC, NM, ZWZ and WYL reviewed and edited the manuscript. All authors contributed to the article and approved the submitted version.

Acknowledgements

Our thanks go out to the original GWAS participants and investigators for sharing and managing the summary statistics.

Data availability

The original data presented in the study are included in the article/Supplementary, and further inquiries can be directed to the corresponding author([email protected]).

Ahmad, E., Lim, S., Lamptey, R., Webb, D. R., & Davies, M. J. (2022). Type 2 diabetes. Lancet, 400(10365), 1803-1820. https://doi.org/10.1016/s0140-6736(22)01655-5
Bastarache, L., Denny, J. C., & Roden, D. M. (2022). Phenome-Wide Association Studies. Jama, 327(1), 75-76. https://doi.org/10.1001/jama.2021.20356
Bowden, J., Del Greco, M. F., Minelli, C., Davey Smith, G., Sheehan, N., & Thompson, J. (2017). A framework for the investigation of pleiotropy in two-sample summary data Mendelian randomization. Stat Med, 36(11), 1783-1802. https://doi.org/10.1002/sim.7221
Bowden, J., Del Greco, M. F., Minelli, C., Davey Smith, G., Sheehan, N. A., & Thompson, J. R. (2016). Assessing the suitability of summary data for two-sample Mendelian randomization analyses using MR-Egger regression: the role of the I2 statistic. Int J Epidemiol, 45(6), 1961-1974. https://doi.org/10.1093/ije/dyw220
Bowden, J., Spiller, W., Del Greco, M. F., Sheehan, N., Thompson, J., Minelli, C., & Davey Smith, G. (2018). Improving the visualization, interpretation and analysis of two-sample summary data Mendelian randomization via the Radial plot and Radial regression. Int J Epidemiol, 47(4), 1264-1278. https://doi.org/10.1093/ije/dyy101
Cao, L., Wang, T., Li, H., El Hafa, F., Zhu, X., Yu, Y., Yan, C., Du, L., Zhu, M., & Jin, G. (2024). Type 2 diabetes, glycaemic traits and upper gastrointestinal cancer risk: a prospective cohort study. J Epidemiol Community Health. https://doi.org/10.1136/jech-2023-221351
Chen, Y., Wu, F., Saito, E., Lin, Y., Song, M., Luu, H. N., Gupta, P. C., Sawada, N., Tamakoshi, A., Shu, X. O., Koh, W. P., Xiang, Y. B., Tomata, Y., Sugiyama, K., Park, S. K., Matsuo, K., Nagata, C., Sugawara, Y., Qiao, Y. L., . . . Potter, J. D. (2017). Association between type 2 diabetes and risk of cancer mortality: a pooled analysis of over 771,000 individuals in the Asia Cohort Consortium. Diabetologia, 60(6), 1022-1032. https://doi.org/10.1007/s00125-017-4229-z
Cheng, J., Dekkers, J. C. M., & Fernando, R. L. (2021). Cross-validation of best linear unbiased predictions of breeding values using an efficient leave-one-out strategy. J Anim Breed Genet, 138(5), 519-527. https://doi.org/10.1111/jbg.12545
Choi, Y. S., Yi, J. W., Shin, W. Y., & Heo, Y. (2022). Oncometabolic surgery in gastric cancer patients with type 2 diabetes. Sci Rep, 12(1), 11853. https://doi.org/10.1038/s41598-022-15404-2
Dabo, B., Pelucchi, C., Rota, M., Jain, H., Bertuccio, P., Bonzi, R., Palli, D., Ferraroni, M., Zhang, Z. F., Sanchez-Anguiano, A., Thi-Hai Pham, Y., Thi-Du Tran, C., Gia Pham, A., Yu, G. P., Nguyen, T. C., Muscat, J., Tsugane, S., Hidaka, A., Hamada, G. S., . . . Luu, H. N. (2022). The association between diabetes and gastric cancer: results from the Stomach Cancer Pooling Project Consortium. Eur J Cancer Prev, 31(3), 260-269. https://doi.org/10.1097/cej.0000000000000703
Davey Smith, G., & Ebrahim, S. (2005). What can mendelian randomisation tell us about modifiable behavioural and environmental exposures? Bmj, 330(7499), 1076-1079. https://doi.org/10.1136/bmj.330.7499.1076
Eisenberg, D., Shikora, S. A., Aarts, E., Aminian, A., Angrisani, L., Cohen, R. V., de Luca, M., Faria, S. L., Goodpaster, K. P. S., Haddad, A., Himpens, J. M., Kow, L., Kurian, M., Loi, K., Mahawar, K., Nimeri, A., O'Kane, M., Papasavas, P. K., Ponce, J., . . . Kothari, S. N. (2023). 2022 American Society of Metabolic and Bariatric Surgery (ASMBS) and International Federation for the Surgery of Obesity and Metabolic Disorders (IFSO) Indications for Metabolic and Bariatric Surgery. Obes Surg, 33(1), 3-14. https://doi.org/10.1007/s11695-022-06332-1
Gao, Q., Cui, L., Huang, C., Chen, Z., Wang, X., Wen, S., Zhao, Y., Wang, M., Shen, B., & Zhu, W. (2023). Gastric cancer-derived mesenchymal stem cells promote gastric cancer cell lines migration by modulating CD276 expression. Exp Cell Res, 422(1), 113414. https://doi.org/10.1016/j.yexcr.2022.113414
Gao, Y., Zhou, M., Xu, X., Ma, J. Y., & Qin, M. F. (2023). Body composition and risk of gestational diabetes mellitus: A univariable and multivariable Mendelian randomization study. J Diabetes Investig. https://doi.org/10.1111/jdi.14115
Georgescu, D., Lighezan, D. F., Rosca, C. I., Nistor, D., Ancusa, O. E., Suceava, I., Iancu, M. A., & Kundnani, N. R. (2023). NASH/NAFLD-Related Hepatocellular Carcinoma: An Added Burden. Life (Basel), 14(1). https://doi.org/10.3390/life14010025
Guan, W. L., He, Y., & Xu, R. H. (2023). Gastric cancer treatment: recent progress and future perspectives. J Hematol Oncol, 16(1), 57. https://doi.org/10.1186/s13045-023-01451-3
Guo, J., Liu, C., Pan, J., & Yang, J. (2022). Relationship between diabetes and risk of gastric cancer: A systematic review and meta-analysis of cohort studies. Diabetes Res Clin Pract, 187, 109866. https://doi.org/10.1016/j.diabres.2022.109866
Hosseini, L., Alinezhad, F., Kharazi, U., & Badalzadeh, R. (2019). The Association of Type 2 Diabetes and Site-Specific Cancers: Linking Mechanisms. Crit Rev Oncog, 24(3), 259-267. https://doi.org/10.1615/CritRevOncog.2019031108
Huang, B., Liu, J., Ding, F., & Li, Y. (2023). Epidemiology, risk areas and macro determinants of gastric cancer: a study based on geospatial analysis. Int J Health Geogr, 22(1), 32. https://doi.org/10.1186/s12942-023-00356-1
Huo, Q., Wang, S., Hou, Y., Gorczynski, R. M., Shen, Y., Wang, B., Ge, H., & Li, T. (2023). Editorial: The relationship between diabetes and cancers and its underlying mechanisms, volume II. Front Endocrinol (Lausanne), 14, 1357577. https://doi.org/10.3389/fendo.2023.1357577
Jenkins, D. J. A., Willett, W. C., Yusuf, S., Hu, F. B., Glenn, A. J., Liu, S., Mente, A., Miller, V., Bangdiwala, S. I., Gerstein, H. C., Sieri, S., Ferrari, P., Patel, A. V., McCullough, M. L., Le Marchand, L., Freedman, N. D., Loftfield, E., Sinha, R., Shu, X. O., . . . Sievenpiper, J. L. (2024). Association of glycaemic index and glycaemic load with type 2 diabetes, cardiovascular disease, cancer, and all-cause mortality: a meta-analysis of mega cohorts of more than 100 000 participants. Lancet Diabetes Endocrinol, 12(2), 107-118. https://doi.org/10.1016/s2213-8587(23)00344-3
Langfelder, P., & Horvath, S. (2008). WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics, 9, 559. https://doi.org/10.1186/1471-2105-9-559
Li, J., Han, T., Wang, X., Wang, Y., Chen, X., Chen, W., & Yang, Q. (2022). H19 may regulate the immune cell infiltration in carcinogenesis of gastric cancer through miR-378a-5p/SERPINH1 signaling. World J Surg Oncol, 20(1), 295. https://doi.org/10.1186/s12957-022-02760-6
Lordick, F., Carneiro, F., Cascinu, S., Fleitas, T., Haustermans, K., Piessen, G., Vogel, A., & Smyth, E. C. (2022). Gastric cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Ann Oncol, 33(10), 1005-1020. https://doi.org/10.1016/j.annonc.2022.07.004
Luo, P., Chen, G., Shi, Z., Yang, J., Wang, X., Pan, J., & Zhu, L. (2023). Comprehensive multi-omics analysis of tryptophan metabolism-related gene expression signature to predict prognosis in gastric cancer. Front Pharmacol, 14, 1267186. https://doi.org/10.3389/fphar.2023.1267186
Machlowska, J., Baj, J., Sitarz, M., Maciejewski, R., & Sitarz, R. (2020). Gastric Cancer: Epidemiology, Risk Factors, Classification, Genomic Characteristics and Treatment Strategies. Int J Mol Sci, 21(11). https://doi.org/10.3390/ijms21114012
Mansorunov, D., Apanovich, N., Kipkeeva, F., Nikulin, M., Malikhova, O., Stilidi, I., & Karpukhin, A. (2022). The Correlation of Ten Immune Checkpoint Gene Expressions and Their Association with Gastric Cancer Development. Int J Mol Sci, 23(22). https://doi.org/10.3390/ijms232213846
Qi, H., Wen, F. Y., Xie, Y. Y., Liu, X. H., Li, B. X., Peng, W. J., Cao, H., & Zhang, L. (2023). Associations between depressive, anxiety, stress symptoms and elevated blood pressure: Findings from the CHCN-BTH cohort study and a two-sample Mendelian randomization analysis. J Affect Disord, 341, 176-184. https://doi.org/10.1016/j.jad.2023.08.086
Saarela, K., Tuomilehto, J., Sund, R., Keskimäki, I., Hartikainen, S., & Pukkala, E. (2019). Cancer incidence among Finnish people with type 2 diabetes during 1989-2014. Eur J Epidemiol, 34(3), 259-265. https://doi.org/10.1007/s10654-018-0438-0
Scherübl, H. (2023). [Type-2-diabetes and gastrointestinal cancer screening]. Z Gastroenterol, 61(6), 683-689. https://doi.org/10.1055/a-1821-9108 (Typ-2-Diabetes mellitus und gastrointestinale Krebsvorsorge.)
Suh, S., & Kim, K. W. (2019). Diabetes and Cancer: Cancer Should Be Screened in Routine Diabetes Assessment. Diabetes Metab J, 43(6), 733-743. https://doi.org/10.4093/dmj.2019.0177
Sun, H., Saeedi, P., Karuranga, S., Pinkepank, M., Ogurtsova, K., Duncan, B. B., Stein, C., Basit, A., Chan, J. C. N., Mbanya, J. C., Pavkov, M. E., Ramachandaran, A., Wild, S. H., James, S., Herman, W. H., Zhang, P., Bommer, C., Kuo, S., Boyko, E. J., & Magliano, D. J. (2022). IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract, 183, 109119. https://doi.org/10.1016/j.diabres.2021.109119
Sung, H., Ferlay, J., Siegel, R. L., Laversanne, M., Soerjomataram, I., Jemal, A., & Bray, F. (2021). Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin, 71(3), 209-249. https://doi.org/10.3322/caac.21660
Thrift, A. P., & El-Serag, H. B. (2020). Burden of Gastric Cancer. Clin Gastroenterol Hepatol, 18(3), 534-542. https://doi.org/10.1016/j.cgh.2019.07.045
Thrift, A. P., Wenker, T. N., & El-Serag, H. B. (2023). Global burden of gastric cancer: epidemiological trends, risk factors, screening and prevention. Nat Rev Clin Oncol, 20(5), 338-349. https://doi.org/10.1038/s41571-023-00747-0
VanderWeele, T. J., Tchetgen Tchetgen, E. J., Cornelis, M., & Kraft, P. (2014). Methodological challenges in mendelian randomization. Epidemiology, 25(3), 427-435. https://doi.org/10.1097/ede.0000000000000081
Xie, J., Chen, L., Tang, Q., Wei, W., Cao, Y., Wu, C., Hang, J., Zhang, K., Shi, J., & Wang, M. (2022). A Necroptosis-Related Prognostic Model of Uveal Melanoma Was Constructed by Single-Cell Sequencing Analysis and Weighted Co-Expression Network Analysis Based on Public Databases. Front Immunol, 13, 847624. https://doi.org/10.3389/fimmu.2022.847624
Xu, H. L., Tan, Y. T., Epplein, M., Li, H. L., Gao, J., Gao, Y. T., Zheng, W., Shu, X. O., & Xiang, Y. B. (2015). Population-based cohort studies of type 2 diabetes and stomach cancer risk in Chinese men and women. Cancer Sci, 106(3), 294-298. https://doi.org/10.1111/cas.12597
Xu, J., Zhang, S., Tian, Y., Si, H., Zeng, Y., Wu, Y., Liu, Y., Li, M., Sun, K., Wu, L., & Shen, B. (2022). Genetic Causal Association between Iron Status and Osteoarthritis: A Two-Sample Mendelian Randomization. Nutrients, 14(18). https://doi.org/10.3390/nu14183683
Xu, R., Zheng, T., Ouyang, C., Ding, X., & Ge, C. (2023). Causal associations between site-specific cancer and diabetes risk: A two-sample Mendelian randomization study. Front Endocrinol (Lausanne), 14, 1110523. https://doi.org/10.3389/fendo.2023.1110523
Yu, Z., Coresh, J., Qi, G., Grams, M., Boerwinkle, E., Snieder, H., Teumer, A., Pattaro, C., Köttgen, A., Chatterjee, N., & Tin, A. (2020). A bidirectional Mendelian randomization study supports causal effects of kidney function on blood pressure. Kidney Int, 98(3), 708-716. https://doi.org/10.1016/j.kint.2020.04.044
Zhang, F., de Haan-Du, J., Sidorenkov, G., Landman, G. W. D., Jalving, M., Zhang, Q., & de Bock, G. H. (2021). Type 2 Diabetes Mellitus and Clinicopathological Tumor Characteristics in Women Diagnosed with Breast Cancer: A Systematic Review and Meta-Analysis. Cancers (Basel), 13(19). https://doi.org/10.3390/cancers13194992
Zhang, J., Hu, C., Zhang, R., Xu, J., Zhang, Y., Yuan, L., Zhang, S., Pan, S., Cao, M., Qin, J., Cheng, X., & Xu, Z. (2023). The role of macrophages in gastric cancer. Front Immunol, 14, 1282176. https://doi.org/10.3389/fimmu.2023.1282176

No competing interests reported.

SupplementaryMaterials.pdf

Download PDF

Version 1

posted

You are reading this latest preprint version

Exploring the causality and pathogenesis of Type 2 diabetes in gastric cancer based on Mendelian randomization and transcriptome data analyses

Status:

Version 1

Abstract

Figures

Introduction

Method

I. Mendelian randomization analysis

01. Data sources

02. Conditioning of SNPs as instrumental variables (IVs)

03. SNP Screening

04. Verification of causality

05. Heterogeneity, pleiotropy and sensitivity analysis

06. Meta-analysis

II. Transcriptomic analyses

01. Data acquisition

02. Identification of differentially expressed genes (DEGs)

03. Establishment and validation of a prognostic model related to gastric cancer

04. Functional enrichment analysis and mutation and drug sensitivity analyses

05. Analysis of the tumor immune microenvironment and immunotherapeutic response

Statistical analysis

Results

I. Mendelian randomization analysis

01. SNP screening

02. Arguments for a causal relationship between diabetes and gastric cancer

03. Heterogeneity, pleiotropy and sensitivity analysis

04. Meta-analysis

II Transcriptomic analyses

01. Identification of T2DM − related DEGs in patients with gastric cancer

02. Risk modeling

03. Analysis of functional characteristics

04. Tumor immune cell infiltration analysis, immunogenetic analysis and immunotherapy response prediction

Discussion

Conclusions

Declarations

Funding

Author Contribution

Acknowledgements

Data availability

References

Additional Declarations

Supplementary Files

Status:

Version 1