Single-cell landscape of dynamic changes in CD8⁺ T cells, CD4⁺ T cells and exhausted T cells in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma has a high incidence and poor prognosis. In this study, we investigated the value of T-cell-related genes in prognosis by single-cell sequencing data in hepatocellular carcinoma. Twelve cases of hepatocellular carcinoma single-cell sequencing were included in the study. The high dimensional weighted gene co-expression network analysis (hdWGCNA) was used to identify gene modules associated with CD4⁺ T cells, CD8⁺ T cells and exhausted T cells. Altered signaling pathway activity in exhausted T cells was uncovered by the AUCell algorithm. xCELL, TIMER, QUANTISEQ, CIBERSORT and CIBERSORT-abs were performed to explore immune cell infiltration. Immune checkpoint inhibitor genes and TIDE methods were used to predict immunotherapy response. Finally, immunohistochemistry and real-time PCR were used to verify gene expression. The hdWGCNA algorithm identified 40 genes strongly associated with CD4⁺ T cells, CD8⁺ T cells and exhausted T cells. Seven genes were finally selected for transcriptome data modeling. The results of the three independent datasets suggested that the model had strong prognostic value. Model genes were critical factors influencing CD4⁺ T cell and CD8⁺ T cell infiltration in patients. The efficacy of PD-1 immunotherapy was higher in patients belonging to the low-risk group. Alterations in signaling pathways’ activity within exhausted T cells were crucial factors contributing to the decline in immune function. Differential expression of seven genes in CD8⁺ T cells, CD4⁺ T cells and exhausted T cells were key targets for improving immunotherapy response in HCC.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-88377-7.

Introduction

Hepatocellular carcinoma (HCC) is highly malignant and is the third leading cause of cancer-related deaths¹. The incidence of virus-associated HCC remains high in Asia. However, the incidence of hepatocellular associated with alcoholic liver disease in Europe and the United States is increasing annually^2,3. The extremely low rate of early diagnosis and poor prognosis of HCC make it a huge challenge that endangers human health^4,5.

Chronic liver injury leads to the infiltration of a large number of immune cells, which ultimately leads to the occurrence of HCC^6,7. The tumour microenvironment of HCC patients contains a high proportion of immune cells. The genome of cells in the tumor microenvironment and cell-cell interactions influence the drug resistance and progression of malignant cells⁸. T cells function as immunosurveillance, immunosuppression, and other complex functions in the tumor microenvironment^9,10. CD4⁺ T cells enhance immune surveillance and can be involved in the killing of malignant cells¹¹. CD8⁺ T cells are the primary cells that kill tumor cells¹². However, CD4⁺ T and CD8⁺ T cells would lead to downregulation of immune function as they are regulated by the immunosuppressive components of the tumor microenvironment¹³. Under the continuous stimulation of tumor immune microenvironment, CD4⁺ T and CD8⁺ T cells gradually lost their immune surveillance function and became exhausted T cells. The increase of exhausted T cells would lead to immune evasion, which seriously affects the therapeutic effect and prognosis of the tumor patient¹⁴.

Single-cell RNA sequencing(scRNA-seq) technology has become the state-of-the-art approach for unravelling the heterogeneity and complexity of RNA transcripts within individual cells, as well as revealing the composition of different cell types and functions within highly organized tissues/organs/organisms¹⁵. Single-cell technology has made a huge leap forward in our understanding of the composition and interactions of the tumor immune microenvironment¹⁶. In HCC, the scRNA-seq technology motivates the exploration of genes characterizing T-cell subpopulations and hairy heterogeneous interactions^17,18. High dimensional weighted gene co-expression network analysis (hdWGCNA), a comprehensive framework for analyzing co-expression networks in high-dimensional transcriptomics data such as single-cell and spatial RNA sequencing (RNA-seq), provides functions for network inference, gene module identification, gene enrichment analysis, statistical tests, and data visualization¹⁹. In ulcerative colitis, nonalcoholic fatty liver disease, breast cancer and meningiomas^20–23, hdWGCNA has been shown to have excellent application value. However, hdWGCNA has not yet been used to explore HCC.

In this study, we explored the value of characterized genes among CD4⁺ T, CD8⁺ T and exhausted T cells in the immunotherapy and prognosis of patients with HCC by single-cell sequencing technology and high dimensional weighted gene co-expression network analysis (hdWGCNA).

Method

Data download

Single-cell sequencing data from the tumour cores of 12 patients with HCC were downloaded from GSE189903²⁴ (https://www.ncbi.nlm.nih.gov/geo/). HCC transcriptome data include The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/), GSE14520²⁵ (https://www.ncbi.nlm.nih.gov/geo/) and International Cancer Genome Consortium (ICGC) (https://dcc.icgc.org/).

Data processing

Read 10x Genomics single-cell sequencing data and created Seurat objects with the Seurat function. Application of the harmony package eliminated batch effects and integrated data from 12 single-cell sequencing samples. Cells were annotated by reference to previous literature reports^26–29 and the “SingleR” method. Cells were annotated based on average expression of known cell lineage-specific marker genes of T cells (CD2, CD3E, CD3D, CD3G), B cells (CD79A, SLAMF7), tumor-associated fibroblasts (CAFs) (COL1A2, COL3A1, COL6A1), tumor associated macrophages (TAMs) (CD14, CD163, CD68, CSF1R), tumor associated endothelial cells (TAEs) (ENG, VWF, PECAM1) as well as epithelial cells (KRT18, KRT19, EPCAM). Subset function was used to extract T cell subpopulations and reclustered and re-annotated^26–29. (CD4⁺ T cells: CD3D, CD4), (CD8⁺ T cells: CD3D, CD8A, GZMK), (Exhausted T cells: PDCD1, HAVCR2, LAG3), (Memory T cells: IL17RA), (Proliferative T cells: MKI67), (Regulatory cells: FOXP3).

HdWGCNA

The R package “hdWGCNA” was taken to construct a scale-free network at the single-cell level by high-dimensional weighted gene co-expression network analysis (hdWGCNA). The optimal soft threshold for the scale-free topological model ft was set to > 0.80 when the threshold was 8. The kME values were calculated over the entire single-cell dataset through the ModuleConnectivity function, with higher kME values indicating a higher probability of the gene being hub genes. The topN hub nodes of each module were extracted by GetHubGenes function. Select 20 genes with high kME in each module. The correlation between modules and cell types was shown through bubble plots.

Building and validating machine learning model

Multiple machine learning algorithms were used to construct stable and accurate prognostic models, with the TCGA data being the modeling group. GSE1452 and ICGC data were used as an external validation set to verify the stability of the model. A consensus model was constructed based on algorithms such as Random Survival Forest, Random Survival Forest, CoxBoost. A total of 101 algorithm combinations were performed to match the predictive model based on the leave-one-out cross validation (LOOCV) framework. The area under the receiver operator curves (ROC) were used to assess the judgmental ability of the model. The Kaplan-Meier curves were used to predict prognosis. Consistency index (C-index) was used to assess the accuracy of each model. Module genes of interest were included in machine learning models. Machine learning selected genes were modeled for prognosis by Least absolute shrinkage and selection operator (LASSO) and multifactor Cox regression analysis. Risk score = βgene A × expr gene A + βgene B × expr gene B+… + βgene N × expr gene N, expr is the mRNA expression of the pivotal gene and β is the corresponding regression coefficient in multivariate genetic Cox regression analysis.

Immune infiltration analysis

T cells are the main cell type that exercise the immune function of the body. In this study, we identified the main set of genes affecting T cell function based on HdWGCNA analysis. On this foundation, it was further analyzed whether the model genes reduced the abundance of immune cell infiltration. Variation in immune cell activity was evaluated by single-sample GSEA (ssGSEA). Xcell, timer, quantitative sequence, CIBERSORT, and CIBERSORT-abs methods were used to assess risk modeling and immune cell correlations.

Prediction of immunotherapy response

TIDE predicts patient immunotherapy response through immune dysfunction and rejection. Low TIDE scores indicate a lower likelihood of immune escape and immunotherapy efficacy in patients (http://tide.dfci.harvard.edu). Immune checkpoint inhibitor genes improve patient prognosis by enhancing immune function. Further analysis of the differences in expression of 47 immune checkpoint inhibitor genes between high and low risk groups.

Cellular communication analysis

Subpopulations of T cells interact and influence each other in a number of ways. Ligands and receptors are important mediators of intercellular information transfer. The expression level of genes is an important basis for predicting the pathways of interactions between cells. CellChatDB data collects the receptor and ligand information of cellular communication, which is widely used in the analysis of cellular communication. Extract expression matrix and classification information from harmony and apply “createCellChat” package to create cell chat object. identify overexpressed genes to screen ligand-receptor pairs and project the results to PPI network. Minimum cells = 3 to filter the communication relationship between low quality cells.

Signaling pathway analysis

AUCell calculated the fraction of each cell enriched in the Kyoto Encyclopedia of Genes and Genomes (KEGG) gene set based on the area under the curve. KEGG pathway gene set download from (Gene Set Enrichment Analysis) GSEA (https://www.gsea-msigdb.org/gsea/downloads.jsp). AUCell applies the area under the curve to calculate the enrichment of the input gene set in the cell. We used the AUCell algorithm to assess pathway activity in individual cells. Gene expression in CD4⁺ T cells, CD8⁺ T cells and exhausted T cells were extracted from Harmony objects. Comparison of changes in signaling pathway activity in CD4⁺ T cells, CD8⁺ T cells and exhausted T cells based on AUCell algorithm.

Real-time PCR and immunohistochemistry

Tissues from patients with HCC at the Renmin Hospital of Wuhan University were collected for validation of model gene expression. GAPDH expression was used as an endogenous gene. The cDNA was synthesized from total RNA using NovoScript^® plus an all-in-one first strand cDNA synthesis kit (Novo protein). Human Protein Atlas(HPA) is a comprehensive resource of human protein information designed to provide a detailed description of human gene and protein expression patterns(https://www.proteinatlas.org/). The database integrates proteomic data from multiple tissue and cell types, including immunohistochemistry and high-throughput antibody preparation techniques. We used HPA database to analyze the differences in the expression of model genes in liver cancer tissues and adjacent tissues.

Statistical analysis

The Wilcoxon rank–sum test was used as a backup to analyze differences between two groups. Correlations between variables were explored using pearson or spearman coefficients. Continuous variables fitted to a normal distribution were compared using t-tests. COX regression analysis and LASSO regression analysis were used to screen for prognostic genes. The Kaplan-Meier method and log-rank test were used for prognostic analysis. All data analyses were conducted using R (4.1.2, https://www.r-project.org/) software. P value of < 0.05 was considered to indicate a statistically significant difference. (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001).

Results

Annotated T-cell subpopulations

Single-cell sequencing data from the core region of 12 cases of HCC were integrated using the “harmony” package. After rigorous quality control, a total of 41,184 cells and 22,552 different features were obtained. A total of 9 subpopulations of HCC cells in the core region were identified by “SingleR” and reference to previous literature. The T cell subpopulation was extracted by applying the subset function to obtain 34,070 T cells and 22,552 different features. After downscaling, clustering and re-annotation of the T cells, a total of 7 T cell subpopulations were obtained (Fig. 1A,B).

Fig. 1.

Identify new analysis modules based on the HdWGCNA algorithm. (A) MAP plot of 12 cases of single-cell sequencing data after dimensionality reduction, clustering and annotation. (B) Reduced-dimensional clustering and annotation for T cell subpopulations. (C) Heatmap of correlation between new modules. (D) Correlation analysis of new modules and T cell subsets.

hdWGCNA revealed that cluster 5 and 9 were the target modules

A high-dimensional weighted gene co-expression network analysis was applied to re-cluster the modules of T cell features. The threshold of the scale-free topological model ft exceeded 0.80 when the soft threshold was 8. HdWGCNA algorithm identified a total of 9 different T-cell related gene modules. The kME values for each module and the 10 genes with the largest kME values in each module were shown. The correlation heatmap demonstrated the kME values for each module and the correlation for each T cell subpopulation. Our study focused on modules 5 and 9, as they were found to have a negative correlation with exhausted T cells and a positive correlation with CD4⁺ T cells and CD8⁺ T cells (Fig. 1C,D).

Construction of prognostic signature

The first 20 genes in Modules 5 and 9 were selected for prognostic modeling. 40 genes went through the LOOCV framework of 101 predictive models and the C-index in the modeling and validation groups was calculated to select the optimal features. The 9 best gene combinations were selected by the machine learning algorithm. 9 model genes were analyzed by LASSO and multifactorial Cox regression analysis to establish a prognosis model. Finally, 7 genes were included in the risk model. The survival curves of TCGA, ICGC and GSE14520 datasets showed that the prognosis of the low-risk group was better than that of the high-risk group (Fig. 2D,F,H). Furthermore, the area under the ROC curve confirmed that the model had excellent prognostic ability (Fig. 2). The immunohistochemistry and real-time PCR showed that ZFP36 was low expressed in tumor tissues. However, CACYBP, CKS2, UBB, HSPA8, HSP90AA1 and CXCR4 were highly expressed in tumor tissues (Fig. 3).

Fig. 2.

Building and validating models. (A) Expression of model genes in single cells. (B) TCGA survival curves. 1-, 3- and 5-year time-dependent ROC curves from the TCGA database. (C) GSE14520 survival curves. 1-, 3- and 5-year time-dependent ROC curves from the GSE14520 database. (D) ICGC survival curves. 1-, 3- and 5-year time-dependent ROC curves from the ICGC database.

Fig. 3.

Gene expression analysis. (A–G) CACYBP, CKS2, UBB, HSPA8, HSP90AA1, ZFP36 and CXCR4 relative protein expression. (H) Real-time PCR analysis. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Effect of risk‑score on immune cell infiltration

Multiple algorithms were used to assess differences in immune infiltration in high and low risk groups. Macrophages, CD4⁺ T cells, and CD8⁺ T cells differed in infiltration between the high and low risk groups. Correlation analysis recommended that higher patient risk scores were associated with lower CD8⁺ T cell and CD4⁺ T cell infiltration (Fig. 4).

Fig. 4.

Immune cell infiltration analysis. (A) Spearman correlation analysis shows that risk scores are strongly related to tumor-infiltrating immune cells. (B–I) Correlation of risk models and immune cells based on XCELL, TIMER, QUANTISEQ, CIBERSORT and CIBERSORT-abs.

Immunotherapy analysis

Immunotherapy is increasingly valuable in patients with HCC. TIDE scores of patients showed higher scores in the high-risk group than in the low-risk group suggested that patients in the low-risk group had a higher benefit from immunotherapy. PD-1 was more suitable for patients in the low-risk group. Multiple immune checkpoint genes were differentially expressed in the two risk groups. Meanwhile, the results of correlation analysis confirmed that the model and immune checkpoint genes were highly correlated. The expression of model genes in exhausted T cells correlated extremely highly with PDCD1 (Fig. 5).

Fig. 5.

Immunotherapy analysis. (A,B) Correlation analysis and differential expression analysis between immune checkpoint genes and risk groups. (C) Prediction of immunotherapy for high-risk and low-risk groups. (D) Single-cell sequencing data reveals the correlation analysis results of model genes and PDCD1 expression in T cells. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Analysis of variation in the activity of exhausted T-cell signaling pathways

AUCell algorithm was based on gene expression to assess alterations in vital signaling pathways in exhausted T cells. Compared with CD4⁺ T cells and CD8⁺ T cells, T cell receptor signaling pathway, TGF beta signaling pathway, chemokine signaling pathway, WNT signaling pathway, ubiquitination signaling pathway, VEGF signaling pathway, MAPK signaling pathway and other pathways that were closely related to tumor development and progression were obviously altered in exhausted T cells (Fig. 6).

Fig. 6.

Single-cell sequencing analyzes the metabolic activity in different cell types. (A–C) The difference in activity of MAPK signaling pathway in different types of T cells. (D–F) The difference in activity of P53 signaling pathway in different types of T cells. (G) Differential analysis of CD4⁺ T cell and exhausted T cell signaling pathways. (H) Differential analysis of CD8⁺ T cell and exhausted T cell signaling pathways. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Analysis of cell-cell interactions

Weights and probabilities of intercellular signaling pathways in the tumor microenvironment of HCC were evaluated based on the Cellchat algorithm. Exhausted T cells hold a greater power in the communication between T cell subpopulations(Fig. 7A,B). In particular, the interactions between exhausted T cells and CD4⁺ T cells, CD8⁺ T cells and regulatory T cells were pronounced. MIF signaling pathway and IL16 signaling pathway were the main signaling pathways involved in the interactions between T cell subpopulations (Fig. 7).

Fig. 7.

Cellular communication analysis. (A) The number of interaction in cell-cell communication network. (B) The weight of interaction in cell-cell communication network. (C,D) Heatmap display of intercellular communication weights. (E,F) Cell-cell communication interaction in MIF and IL16 signaling pathway.

Discussion

HCC had a low rate of early diagnosis⁴. Patients with a few liver resections were faced with a high recurrence rate and a high rate of drug resistance after surgery^30,31. It was crucial to explore innovative therapeutic modalities for HCC. The tumor microenvironment of HCC contained tumor cells and non-tumor cells. Inflammatory factors, cytokines, and chemokines secreted by cells in the tumor microenvironment function as essential components in the regulation of distant metastasis, immune escape, and drug resistance^32,33. T cells serve as the main force in the regulation of immune function. Antigens activate precursor T cells to differentiate into effector T cells to perform immune functions^34,35. However, rapid proliferation of tumor cells and long-term antigen exposure lead to a decrease in effector T cell activity and a loss of immune activity to exhausted T cells^36,37. Exploring potential specific targets to activate exhausted T cells may be a new approach for tumor immunotherapy. Single-cell sequencing technology has stimulated the research of cellular subpopulations in the tumor microenvironment.

In this study, we firstly combined scRNA-seq technology with the hdWGCNA to explore the characteristics of T cell subsets in the HCC immune microenvironment. We found that the parts that were inversely related to exhausted T-cells and CD4⁺ T cells, CD8⁺ T-cells were considered as potential targets for reversing exhausted T cells. Through multiple machine learning algorithms, 7 genes (CACYBP, CKS2, UBB, HSPA8, HSP90AA1, ZFP36 and CXCR4) closely related to CD4⁺ T cells, CD8⁺ T cells and exhausted T cells were screened out. Further, the patients were grouped into high and low risk groups by model genes. The results of the TIDE score revealed that patients in the low-risk group were less likely to escape while receiving immunotherapy. This was associated with accelerated immune cell apoptosis, cyclin phosphorylation and ubiquitination in patients in the high-risk group. Model genes and immune checkpoint inhibitor genes had a high correlation. Model genes may be key targets for inhibiting T cell exhaustion.

CACYBP is a cell cycle-regulating calcineurin. It mainly acts with calcium-binding proteins to regulate the cytoskeleton, ubiquitination, and the cell cycle^38,39. Lian et al. revealed that overexpression of CACYBP accelerated the phosphorylation process of P27Kip1, and knockdown of CACYBP resulted in decreased cell cycle protein expression and accelerated apoptosis⁴⁰. Our study further revealed that CACYBP may be overexpressed in functional T cells accelerating the phosphorylation of immune function-related proteins leading to T cell exhaustion. CKS2 was highly expressed in HCC. High expression CKS2 promoted cancer cell proliferation⁴¹. HSPA8 was highly immunogenic and involved in autophagy and immune cell infiltration in high cell carcinoma^42,43. The ubiquitin proteasome system was an essential pathway for the regulation of the cell cycle, apoptosis and immunomodulation. HSP90AA1 was a critical link in the regulation of ubiquitination in HCC⁴⁴. ZFP36 was the key factor regulating cellular autophagy⁴⁵. UBB were able to influence tumor heterogeneity and immune cell infiltration in HCC⁴⁶. Patients with overexpressed CXCR4 had poor sorafenib efficacy. In contrast, CD8⁺ T cell infiltration was increased in the tumor microenvironment of HCC after blockade of the SDF-1α/ CXCR4 pathway^47,48. CXCR4 may be a crucial gene contributing to T-cell exhaustion. The gene expression and metabolism of exhausted T cells were severely altered. TIGIT, LAG3, HAVCR2 and PDCD1 were the marker genes of exhausted T cells. In the exploration of variation in gene expression in exhausted T cells, it was found that the expression of model genes and marker genes in exhausted T cells were highly related. These results indicated that model gene expression variations contribute significantly to the functional evolution of exhausted T cells. Model genes may be potential targets for regulating the functional recovery of exhausted T cells.

Significant changes in tumor-related signaling pathways occurred in exhausted T cells. This study showed that WNT signaling pathway activity was higher in exhausted T-cells than in CD8⁺ T-cells. The active WNT signaling pathway may be a factor contributing to T cell depletion. This study also found that the VEGF signaling pathway or TGF beta signaling pathway was lower in both CD4⁺ T cells and CD8⁺ T cells than in exhausted T cells. Alterations in the activity of the VEGF signaling pathway or TGF beta signaling pathway in exhausted T cells leaded to change in immune function in the tumor microenvironment. The activity of ubiquitination signaling pathway or MAPK signaling pathway was elevated in exhausted T cells compared to CD4⁺ T cells and CD8⁺ T cells. The ubiquitination signaling pathway or MAPK signaling pathway may promote immune escape by increasing exhausted T-cells. This study also found that with the increase of exhausted T-cells, the MIF signaling pathway and IL16 signaling pathway involved in the interactions between T cell subpopulations showed active states. MIF contributed to anti-inflammatory, immune evasive, and immune tolerant phenotypes in both innate and adaptive immune cell types⁴⁹. IL16 can significantly promote the proliferation of liver cancer cells⁵⁰. Model genes may affect the dynamic changes of CD4⁺ T cells, CD8⁺ T cells and exhausted T cells, and change the composition of HCC immune microenvironment through these pathways. Reducing T cell exhaustion and reverting T cell immune function may become the critical for immunotherapy of HCC. We constructed prognostic modeling of T cell exhaustion-related gene modules in HCC patients by multiple machine learning algorithms and provided new targets for patient treatment and prognosis.

The findings of this study were innovative. However, there remained limitations. Firstly, this study only focused on T cells, other immune cells also played an irreplaceable role in tumor cellular immune elimination. Secondly, the findings lack validation at the animal level. Finally, the mechanism of T cell exhaustion needs to be further explored.

In conclusion, the hdWGCNA analysis of single-cell sequencing data from HCC identified gene modules related to CD8⁺ T cells, CD4⁺ T cells and exhausted T cells. These modular genes can predict patient prognosis, guiding immunotherapy and being a potential target for suppressing T-cell exhaustion. However, the results of the study required further experimental validation.

The primers used for RT‒PCR analysis:

CACYBP-F: CTTCTTCTCGCGGAGGCT.

CACYBP-R: CCTCTTCTAGATCTTTCTGTAGCGAT.

CKS2-F: TCTTCGCGCTCTCGTTTCAT.

CKS2-R: TGGGTAACATAACATGCCGGT.

UBB-F: GGCTATGAGGAATTTGGGGCT.

UBB-R: AGATCTGCATTTTGACCCCTCA.

HSPA8-F: CCTACACCCCAGCAACCAT.

HSPA8-R: GTGCTGGAAAACACCCACAC.

HSP90AA1-F: CCCAGAGTGCTGAATACCCG.

HSP90AA1-R: TAACAGGTGCCCTGCTTCTC.

ZFP36-F: ACTGCCATCTACGAGAGCCT.

ZFP36-R: ACTCAGTCCCTCCATGGTC.

CXCR4-F: TACCATGGAGGGGATCAGTGAAAA.

CXCR4-R: AACAAACGGCACCTCCTCC.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

TCGA

The cancer genome atlas program

GEO

Gene expression omnibus

HCC

Hepatocellular carcinoma

UMAP

Uniform manifold approximation and projection

RT-qPCR

Real-time quantitative PCR

ROC

Receiver operating characteristic

ssGSEA

Single sample GSEA

HdWGCNA

high-dimensional weighted gene co-expression network analysis

LASSO

Least absolute shrinkage and selection operator

KEGG

Kyoto encyclopedia of genes and genomes

GSEA

Gene set enrichment analysis

Author contributions

Weixing Wang and Jia Yu contributed to the study’s inception and design. Rongqiang Liu, Jing Ye and Jianguo Wang equally contributed to the literature search, analysis and writing of the manuscript. Other authors contributed to the study design and study supervision. All authors approved the final version of the manuscript.

Data availability

All data was in the manuscript and can be obtained from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.