Screening of keloid immune microenvironment-related genes and their effects on keloid proliferation and migration

doi:10.21203/rs.3.rs-3234024/v1

Download PDF

Research Article

Screening of keloid immune microenvironment-related genes and their effects on keloid proliferation and migration

https://doi.org/10.21203/rs.3.rs-3234024/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 12 Dec, 2023

Read the published version in Biochemical Genetics →

You are reading this latest preprint version

Keloid is a kind of proliferative scar with continuous growth, no restriction and easy recurrence, which cannot be cured and bring serious physical injury and psychological burden to patients. The main reason is that the pathological mechanism is not clear. Therefore, this project is expected to reveal the immune microenvironment-related genes and their functions in keloid progression, and provide effective targets for the treatment of keloid. Firstly, 8 kinds of immune infiltrating cells and 19 potential characteristic genes were identified by immune infiltration analysis, ssGSEA, LASSO regression (glmnet algorithm and lars algorithm) and WGCNA, indicating that keloid is closely related to the changes of immune microenvironment. Then, 4 pathological biomarkers of keloid (MAPK1, PTPRC, STAT3 and IL1R1) were identified by differentially analysis, univariate analysis, LASSO regression (lars algorithm), support vector machine recursive feature elimination (SVM-REF) algorithm, multivariate logical regression analysis and six machine learning algorithms. Based on the 4-characteristic genes, the risk prediction model and nomogram are constructed. Calibration curve and ROC analysis (AUC = 0.930) show that the model has reliable clinical value. Subsequently, consistent cluster analysis was used to find that there were 2 immune microenvironment subsets in keloid patients, of which subgroup Ⅱ was immune subgroup. Multiple independent datasets and RT-qPCR showed that the expression trend of the 4 genes was consistent with the analysis. Cell gain-loss experiment confirmed that 4 genes regulate the proliferation and migration of keloid cells. The above data shows that MAPK1, PTPRC, STAT3 and IL1R1 may be personalized therapeutic targets for keloid patients.

keloid

immune microenvironment

characteristic genes

proliferation and migration

immune subtypes

Keloid is the result of the excessive growth of type III collagen caused by the combined action of multiple cells in the healing process of injured skin. it is characterized by continuous growth, unrestricted, itching and pain, easy recurrence after surgery and invasive to normal skin, which brings serious physical damage and psychological burden to patients (Liu et al., 2022; Macarak et al.,2021; Elsaie, 2021). At present, there are many effective ways to treat keloid, such as corticosteroid injection, surgical resection, radiotherapy and laser therapy, but the prognosis is still not ideal, and there are various adverse reactions and relapses. The main reason are the clinical complexity and individual heterogeneity of keloid patients, the pathological mechanism has not been fully explored, and the individual differences lack effective markers (Naik,2022; Ekstein et al.,2021). At the same time, some emerging treatments (targeted therapy and cell therapy, etc.) lack effective targets. Therefore, it is very important to identify reliable keloid diagnostic markers and targets, develop new molecular classification techniques to identify patient differences, and guide the customized treatment of keloid patients.

More and more evidences show that there are immune cell disorders in keloid scars. It is reported that there is a significant difference in the immune spectrum between keloid and adjacent normal tissues. The proportion of macrophages in keloid is significantly higher than that of surrounding normal tissues, and the proportion of cDC2 cells is significantly reduced, indicating that the growth of keloid is closely related to immunity (Feng et al.,2022). Some researchers have also found that the occurrence and development of keloid is the result of the co-evolution of keloid fibroblasts and their surrounding immune cells. The immune cells infiltrate the microenvironment and regulate the keloid fibroblasts to promote the progress of keloid by secreting cytokines (Shan and Wang, 2022). Chen et al. found that the overexpression and imbalance of collagen in keloid tissue are closely related to Treg disorder (Chen et al.,2019). Shan et al. reported that CD⁸⁺T cells can promote keloid-related immunosuppression and may be a predictor of keloid severity (Shan et al.,2022). Liu et al. found that SPP1 secreted by macrophages can stimulate CTHRC¹⁺fibroblasts, induce excessive collagen deposition in keloids, and promote the deterioration of keloids (Liu et al.,2022). These reports show that both immune system and immune microenvironment play an important role in keloid. Therefore, it is very urgent to screen out characteristic genes related to immune microenvironment to accurately identify immune molecular subtypes and the heterogeneity of keloid patients.

In this project, we comprehensively identified the differentially expressed immune cell spectrum through ssGSEA, LASSO and WGCNA. Then identify keloid immune microenvironment characteristic genes and construct keloid risk prediction model by univariate analysis, LASSO-lars algorithm, support vector machine recursive feature elimination (SVM-REF) algorithm, logical regression (gradient descent method), naive Bayesian classification, support vector machine (SVM) classification, bp neural network classification, K-Nearest Neighbor (KNN) and multivariate logical regression. The risk model was visualized by nomogram and evaluated by ROC and calibration curve. In addition, we identified different subtypes of the immune microenvironment in keloid patients based on the expression of characteristic genes, and investigated the differences of these subtypes in the biological function, pathway, immune cell infiltration and immune characteristics. Finally, in order to further verify the expression of different characteristic genes, RT-PCR analysis and pan-carcinoma analysis were performed. The analysis process of this study is shown in Figure S1.

Data acquisition

All common datasets GSE44270, GSE7890, GSE92566, GSE145725 and GSE83286 are downloaded from the Gene Expression Omnibus database (GEO, https://www.ncbi.nlm.nih.gov/geo/). The details of all datasets are summarized in Table S1. 28 immune cell subtypes were downloaded from TISIDB server (http://cis.hku.hk/TISIDB/download.php) (Ru et al.,2019). Immune-related genes were obtained from Immunology Database and Analysis Portal (ImmPort; https://www.immport.org/home) and InnateDB database (https://www.innatedb.ca/) (Breuer et al.,2013; Bhattacharya et al.,2018).

Analysis of immune cell infiltration

Firstly, the enrichment scores of 28 immune cell subtypes (the relative proportion of each cell subtype in immune cells) in keloid patients were evaluated by single sample gene cluster enrichment analysis (ssGSEA) (Charoentong et al.,2017), and then the difference of immune cell proportion was analyzed by Wilcoxon test. The immune score was then calculated using the "ESTIMATE" tool in Sangerbox 3.0 (http://sangerbox.com/) (Shen et al.,2017).

Screening of key characteristic immune cells

The key characteristic immune cells in the result of immune infiltration were identified by LASSO regression analysis using R package "glmnet" and "lars". Combine the results of the two algorithms for screening. The analytical results were calibrated using a tenfold cross-validation method of 100 iterations. Determine the optimal penalty parameter by the minimum criterion λ, The immune cell subpopulation with non-zero coefficient is considered as the optimal variable.

Weighted correlation network analysis (WGCNA)

The WGCNA network was established by the "WGCNA" tool in Sangerbox 3.0 (http://sangerbox.com/) to screen the gene modules related to the key characteristic immune cell subtypes. Firstly, the outlier genes and samples were removed by the goodSamplesGenes method of R software package WGCNA. Then the optimal soft threshold is determined based on the optimal scale-free topology model, and the weighted adjacency matrix is constructed. Then the hierarchical clustering tree method was used to aggregate the gene module, and the parameters were as follows: minimum size (Gene group) = 30, sensitivity = 3, module merge threshold = 0.30. Finally, Pearson correlation analysis was carried out between gene modules and key characteristic immune cells, and the gene modules significantly related to characteristic immune cells were screened.

Identification of immune microenvironment-related differentially expressed genes

The differentially expressed genes in keloid samples and control samples were screened by "limma" tool in Sangerbox 3.0 (http://sangerbox.com/). The screening criteria are as follows: | log2 (fold change) | > 1, False Discovery Rate (FDR) < 0.05 and p value < 0.05 (Xie et al.,2022). Immune-related genes are then downloaded from Import (https://www.immport.org/home) website and InnateDB (https://www.innatedb.ca/) database. Finally, Venn diagram was used to cross-compare the three results of differentially expressed genes, key gene modules and immune-related genes to obtain differentially expressed genes related to immune microenvironment.

Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG)

The functional enrichment of differentially expressed genes related to immune microenvironment was analyzed by "clusterProfiler" tool in Sangerbox 3.0 (http://sangerbox.com/). The analysis parameters are as follows: the maximum number of visualizations = 20, the minimum gene set = 5, and the maximum gene set = 5000, p value < 0.05 and FDR < 0.1 were considered statistically significant (Yu et al.,2012). GO analysis can be divided into three aspects: biological process (BP), molecular function (MF) and cellular component (CC).

Establishment of risk prediction model

The screening genes were analyzed by online tool STRING (http://string-db.org), and the interaction network was obtained and visualized by Cytoscape software. Then the hub sub-network was filtered by the Molecular Complex Detection (MCODE) plug-in according to the parameters: degree cutoff = 2, K-Core = 2, node score cutoff = 0.2. Then univariate analysis was carried out on the hub sub-network genes. The genes with statistical significance(p < 0.05) were further screened by two machine learning algorithms: LASSO (lars algorithm) regression analysis and support vector machine recursive feature elimination (SVM-REF) algorithm. The R package "lars" was used for LASSO regression analysis, and 10 cross validation was used. SVM-RFE analysis and screening of characteristic genes were carried out using R-packet "caret", and 10-fold cross-validation was used. The key characteristic factors are obtained by integrating the results of the three methods. Finally, the key characteristic factors are analyzed by multivariable logistic regression analysis to build a risk prediction model. The model formula was: risk value = constant + β₁ × expression(gene₁)+ β₂ × expression(gene₂)+……+β_n × expression( gene_n)。β was the logical regression coefficient. The diagnostic effect of the model was evaluated by the ROC curve and the area under the curve (AUC).

Validation of characteristic genes in risk prediction model

5 machine learning algorithms, namely, logical regression (gradient descent method), naive Bayesian classification, support vector machine (SVM) classification, bp neural network classification and K-Nearest Neighbor (KNN) classification, were used to classify and evaluate the characteristic genes in the risk prediction model. When analyzing each algorithm, the dataset was divided into training set and test set according to the ratio of 7:3. Then calculate the accuracy, precision and F1 value of the algorithm. Finally, the clinical effect of the classification method was evaluated by the ROC curve and the area under the curve (AUC).

Establishment and evaluation of nomogram

Based on the characteristic gene expression profile, the R packet "rms" was used to establish nomogram. Then draw a calibration curve to evaluate the accuracy of nomogram, and use ROC curve to evaluate the clinical application of nomogram.

Consistent cluster recognition of different immune microenvironment subtypes

Consensus clustering analysis was carried out by "Consensus Cluster Plus" tool in Sangerbox 3.0 (http://sangerbox.com/). Pearson method was used for clustering distance, and "partitioning around medoid (PAM)" was used for clustering method. The optimal number of subtypes was selected by cumulative distribution function (CDF) curve, relative change of area under CDF curve and consistent clustering score. Then use the "Rtsne" tool for tSNE analysis.

Gene set enrichment analysis (GSEA) and Gene set enrichment analysis (GSVA)

Use GSEA v3.0 (http://software.broadinstitute.org/gsea/index.jsp) for GSEA analysis. Download the c2.cp.kegg.v7.4.symbols.gmt subset from Molecular Signatures Database (MSigDB, http://www.gsea-msigdb.org/gsea/downloads.jsp) to evaluate the pathways and molecular mechanisms related to characteristic genes. The execution parameters were as follows: the minimum gene set = 5, the maximum gene set = 5000, a thousand times of resampling, p < 0.05 and FDR < 0.25 was considered to be significant.

Functional enrichment among immune microenvironment subtypes was evaluated by the "GSVA" tool in Sangerbox 3.0 (http://sangerbox.com/). Download the c2.cp.kegg.v7.4.symbols.gmt and c5.go.v7.4.symbols.gmt subsets from Molecular Signatures Database (MSigDB, http://www.gsea-msigdb.org/gsea/downloads.jsp). The execution parameters are as follows: minimum gene set = 5, maximum gene set = 5000. After obtaining the enrichment score of each sample in the gene set, the "limma" tool was used to screen the significant differences marker pathways and biological functions. The screening criteria were: | log2 (fold change) | > 1, False Discovery Rate (FDR) < 0.05 and p < 0.05.

Cell culture

Normal human skin fibroblasts (NHDF, BNCC358600) and human primary keloid fibroblasts (HKF, BNCC357596) were purchased from BeNa Culture Collection (BNCC, China). The culture medium of NHDF was 90% Dulbecco's modified eagle medium-high (Gibco, USA) + 10% fetal bovine serum (Gibco, USA) + penicillin-streptomycin (Gibco, USA). The culture medium of HKF was special medium (BNCC362693, China), which was purchased from BNCC. Both kinds of cells were cultured in a constant temperature incubator of 5% CO₂, 37 ℃. When the cell density reaches 80%-90%, passage was carried out, and the culture medium was changed regularly.

Real-time RT-PCR

RNA Easy Fast animal tissue/cell total RNA extraction kit was used to extract cell total RNA (TIANGEN, China). Then nanodrop 2000 was used to quantify the extracted RNA. Then Fast King one-step RT-PCR kit (TIANGEN, China) was used to detect RT-qPCR in the samples. GAPDH was used as the internal reference gene for relative expression, and the relative expression was calculated as 2^−ΔΔCt. All the primers for detecting genes were designed by Primer 5, and the sequence of primers can be found in Table S2. Sangon Biotech (shanghai, China) for synthesis. Repeat each sample three times.

Statistical analysis

All statistical analyses were performed using RStudio and SPSS 25.0. The Spearman method was used for correlation analysis. The difference between the two groups was detected by Wilcoxon test or Student's t test. P < 0.05 was considered to be statistically significant.

Analysis of immune infiltration in keloid patients.

Firstly, in order to describe the immune difference between keloid samples and healthy skin samples, we used single sample gene set enrichment analysis (ssGSEA) method to compare the difference of immune cell subsets enrichment scores of each sample in GSE44270. The results showed that T cell (Effector memory CD⁸⁺ T cell and T follicular helper cell), macrophage and monocyte infiltration were higher in keloid samples, while Effector memory CD⁴⁺ T cell, Type 17 T helper cell, Natural killer cell, Immature dendritic cell, Immature B cell and Mast cell were significantly different, but the infiltration was lower than that in healthy skin samples (Fig. 1A, B).In order to further verify the results of immune infiltration, we used the same method to evaluate the difference of 28 immune cell scores in GSE7890.The results showed that there were significant differences in T cells, including Effector memory CD⁸⁺T cell, Central memory CD⁸⁺T cell, Effector memory CD⁴⁺T cell, Activated CD⁴⁺T cell, Type 17 T helper cell, Type 1 T helper cell, Type 2 T helper cell, Natural killer T cell and T follicular helper cell. Other differential immune cells were Activated B cell, Natural killer cell, Immature dendritic cell, Macrophage and Monocyte. Effector memory CD8 T cell, Macrophage, Monocyte and T follicular helper cell immunocytes were also highly infiltrated in keloid samples (Fig. 1C, D). Combining the results of the two analyses, we obtained 9 kinds of differentially expressed immune cells in keloid patients (Effector memory CD⁸⁺T cell, Macrophage, Monocyte, T follicular helper cell, Effector memory CD⁴⁺T cell, Type 17 T helper cell, Activated B cell, Natural killer cell and Immature dendritic cell). Subsequently, we used Glmnet algorithm and Lars algorithm for LASSO regression analysis to further identify the characteristic immune cells associated with keloid progression. Combining the two algorithms, we found that among the 9 immune cell subsets, except T follicular helper cell, the other eight are determined as the optimal variables with non-zero coefficients(λ_glmnet = 0.020; λ_Lars = 0.750, R² = 0.917 and ministep = 8; Fig. 1E-H, Table S3).

Screening of differentially expressed genes (DEG) associated with immune microenvironment.

Through the screening conditions of differentially expressed genes, we identified 333 differentially expressed genes (150 up-regulated and 183 down-regulated) from the GSE92566 dataset (Fig. S2A, Fig. S2B). Then, with the GSE44270 expression profile as the background, we identified the key modules related to 8 characteristic immune cell subtypes in keloid patients by WGCNA. The analysis results showed that the scale-free topological network and connectivity were the most effective when the soft threshold was 3 (Fig. 2A). The hierarchical clustering algorithm combines all genes into six gene modules with different colors (Fig. 2B). The result of correlation analysis shows that turquoise module (1842 genes) has a high correlation with Effector memory CD4 T cell (R = 0.59), Type 17 T helper cell (R=-0.73), Activated B cell (R=-0.44), Natural killer cell (R=-0.68), Imitate dendritic cell (R=-0.37), Macrophage (R=-0.45) and Monocyte (R=-0.55) (Fig. 2C). At the same time, the blue module (74 genes) also has a certain correlation with the Type 17 T helper cell (R=-0.77), Activated B cell (R=-0.69), Natural killer cell (R=-0.37), Automatic differentiation cell (R=-0.42) and Macrophage (R = 0.48) (Fig. 2C). Therefore, we selected the genes in turquoise and blue module for further analysis. Then using Venn diagram to compare key module genes, differentially expressed genes and immune microenvironment related genes (ImmPort dataset and InnateDB dataset), we obtained 19 overlapping immune microenvironment related differentially expressed genes (Fig. 2D).

Correlation and functional enrichment analysis of immune microenvironment-related DEGs.

Firstly, we analyzed the correlation between 19 overlapping differentially expressed genes and 28 immune cell subsets. The results showed that these differential genes were significantly correlated with immune cells, indicating the accuracy of the previous analysis results. At the same time, it also showed that the change of immune microenvironment may be an important physiological mechanism for the occurrence and development of keloid (Fig. 2E). Subsequently, we analyzed the functional enrichment of 19 differential genes. Biological processes (BP) are mainly involved in immune response and glial cell differentiation. Cell composition (CC) is mainly concentrated in the granule cavity, transcription factor complex and receptor complex. In terms of molecular function (MF), it is mainly involved in receptor binding, phosphatase activity and amino acid binding (Fig. 3A-C). KEGG enrichment analysis showed that differentially expressed genes related to these immune microenvironments were closely related to Th17 cell differentiation, Jak-STAT signal pathway, nod-like receptor signal pathway, B/T cell receptor signal pathway, HIF-1 signal pathway, cGMP-PKG signal pathway and a variety of human diseases (cytomegalovirus infection, leukemia virus type 1 infection and immunodeficiency virus type 1 infection, etc.) (Fig. 3D). These results further indicate that these differentially expressed genes play an important role in the immune microenvironment. Finally, in order to evaluate the interaction between these differentially expressed genes, we constructed protein-protein interaction (PPI) network (p = 8.1e-15, Fig. 3E), including 18 nodes and 44 edges. 2 key subnetworks were identified by MCODE plug-in, which included 10 differentially expressed genes (Fig. 3E).

Establishment and evaluation of keloid risk prediction model.

To establish the keloid risk prediction model, we used three methods: univariate analysis, support vector machine recursive feature elimination (SVM-REF) algorithm and LASSO regression (Lars algorithm) for screening feature genes. Merge GSE44270,GSE7890 and GSE92566, and remove the batch effect to create a combined dataset. Univariate analysis results showed that, except for TLR4 and INPP5D, the p values of the other 8 genes were less than 0.05 (Table 1). At the same time, the LASSO regression (Lars algorithm) results showed that 7 genes (IL1R1, TLR4, JAK1, MAPK1, PTPN6, PTPRC and STAT3) were determined as the optimal variables with non-zero coefficients (R² = 0.952, λ = 0.31, ministep = 9; Fig. 4A, B). In addition, the SVM-REF analysis results show that the highest performance model can be obtained using 4 features (RMSE = 0.086, Fig. 4C), namely MAPK1, PTPRC, STAT3 and IL1R1. Based on the results of the 3 methods, we identified MAPK1, PTPRC, STAT3 and IL1R1 as candidate characteristic genes for risk prediction model. Subsequently, we used the expression profiles of these 4 genes as input variables for multivariate logical regression analysis. A risk prediction model was established based on the logical regression coefficients of 4 genes (R² = 0.953, Table 2). The formula of the model was: 0.700+( -0.001) ×expression (IL1R1) + 0.005×expression (MAPK1) + (-0.029) ×expression (PTPRC)+(-0.004) ×expression (STAT3). ROC analysis showed that the AUC value of the model was 96.4% (Fig. 4D), indicating that the model has a strong clinical prediction ability. Subsequently, we used ROC to further evaluate the clinical risk prediction ability of the model on GSE44270, GSE145725 and GSE7890, respectively. The results show that the AUC values of the three datasets were 96% (Fig. 4E), 100% (Fig. 4F) and 100% (Fig. 4G), respectively, which further illustrates the reliability of the risk prediction of the model.

Table 1

Univariate analysis of 10 genes.
	B			Standard error	Wald		Significance		Hazard ratio		95% confidence interval for the Hazard ratio
	B			Standard error	Wald		Significance		Hazard ratio		lower		upper
IL1R1	-0.042			0.013	9.962		0.002		0.958		0.934		0.984
IL6ST	-0.014			0.004	10.408		0.001		0.986		.977		0.994
INPP5D	-0.243			0.201	1.452		0.228		0.785		0.529		1.164
JAK1	-0.032			0.010	11.054		0.001		0.969		0.951		0.987
MAPK1	0.074			0.024	9.071		0.003		1.077		1.026		1.129
PPARG	0.668			0.317	4.445		0.035		1.950		1.048		3.626
PTPN6	0.168			0.062	7.336		0.007		1.183		1.048		1.336
PTPRC	-1.138			0.423	7.231		0.007		0.320		0.140		0.734
STAT3	-0.110			0.032	11.628		0.001		0.896		0.841		0.954
TLR4	-0.026			0.024	1.231		0.267		0.974		0.930		1.020
		Unstandardized Coefficients				Standardization coefficient		t		Significance		95% confidence interval for the Hazard ratio			Collinear statistics
		B	Standard error			β		t		Significance		lower		upper	Allowance	VIF
(Constant)		0.700	0.133					5.279		< 0.01		0.433		0.968
IL1R1		-0.001	0.000			-0.135		-2.273		0.028		-0.002		0.000	0.281	3.557
MAPK1		0.005	0.001			0.435		7.776		< 0.01		0.004		0.006	0.315	3.178
PTPRC		-0.029	0.012			-0.143		-2.368		.022		-0.054		-0.004	0.270	3.701
STAT3		-0.004	0.000			-0.468		-8.584		< 0.01		-0.005		-0.003	0.332	3.009

Table 2

Multivariate logistic regression analysis.
	B	Standard error	Wald	Significance	Hazard ratio	95% confidence interval for the Hazard ratio
	B	Standard error	Wald	Significance	Hazard ratio	lower	upper
IL1R1	-0.042	0.013	9.962	0.002	0.958	0.934	0.984
IL6ST	-0.014	0.004	10.408	0.001	0.986	.977	0.994
INPP5D	-0.243	0.201	1.452	0.228	0.785	0.529	1.164
JAK1	-0.032	0.010	11.054	0.001	0.969	0.951	0.987
MAPK1	0.074	0.024	9.071	0.003	1.077	1.026	1.129
PPARG	0.668	0.317	4.445	0.035	1.950	1.048	3.626
PTPN6	0.168	0.062	7.336	0.007	1.183	1.048	1.336
PTPRC	-1.138	0.423	7.231	0.007	0.320	0.140	0.734
STAT3	-0.110	0.032	11.628	0.001	0.896	0.841	0.954
TLR4	-0.026	0.024	1.231	0.267	0.974	0.930	1.020

Evaluation of Characteristic factors in keloid risk prediction model.

Then, we evaluate the characteristic factors in the risk prediction model. Taking the combined dataset as the background, we use five algorithms: logical regression (gradient descent method), naive Bayesian classification, support vector machine (SVM) classification, bp neural network classification and K-Nearest Neighbor (KNN) classification to verify the classification effect of the classifier established by the 4 feature factors. The sample was divided into training set and test set according to the ratio of 7:3. The results show that the 5 classifiers have significant classification effects, and the classification accuracy, accuracy and F1 value of each classifier are 100% (Fig. 5A-E). The AUC value of each classifier evaluated by ROC was also 100% (Fig. 5F-J). Then, based on the expression profile of the combined dataset, we drew the ROC of each characteristic factor. The analysis results showed that the AUC values of MAPK1, PTPRC, STAT3 and IL1R1 were 97.306%, 92.592%, 97.475% and 96.633 respectively (Fig. 6A), which showed that these 4 genes themselves also have the potential of keloid diagnosis biomarkers. In addition, based on the expression of MAPK1, PTPRC, STAT3 and IL1R1 genes in combined dataset, we constructed the corresponding nomograph of the keloid risk prediction model (Fig. 6B). The result shows that the C index was 0.873, indicating that the model has recognition ability. The calibration curve of nomograph shows a good consistency between the predicted value and the actual value (Fig. 6C). The ROC results showed that the AUC of nomograph model was 0.930, indicating that the nomograph model was highly feasible (Fig. 6D). In order to further understand the potential biological role of these 4 genes in keloid, we performed gene set enrichment analysis (GSEA) for the 4 genes using KEGG gene sets. GSEA results show that these 4 genes are mainly involved in immune cell (B/T) receptor signal pathway, Fc epsilon RI signal pathway, ERBB signal pathway, cancer-related pathway, endocytosis, fatty acid metabolism and phosphorylation regulation (Fig. 7). Combined with the above results, we suggest that these immune microenvironment-related characteristic genes have good diagnostic ability in predicting keloid risk.

Identification of immune microenvironment subtypes in keloid patients.

To further explore the keloid immune microenvironment, we merged the combined dataset (GSE44270, GSE7890 and GSE92566), GSE145725 and GSE83286 into a new dataset containing 74 samples (39 keloids and 35 controls) (Fig. S3A). Then, based on 39 keloid samples, we conducted consensus cluster analysis on keloid samples according to the expression of 4 characteristic genes, and identified different immune microenvironment subtypes of keloid patients. According to the cluster heat map (Fig. 8A), CDF curve (Fig. 8B), the relative change of CDF curve area (Fig. 8C) and the consistent cluster score (Fig. 8D), we selected k = 2 as the optimal value, and divided 39 patients into two different subtypes, including 23 subtype Ⅰ and 16 subtype Ⅱ (Fig. 8A). At the same time, the tSNE dimension reduction analysis showed that there was a significant difference between subtype Ⅰ and subtype Ⅱ samples (Fig. 8E). In addition, the differentially expressed genes of subtype Ⅰ and subtype Ⅱ were analyzed by limma method, and a total of 1292 differentially expressed genes were identified (511 up-regulated and 781 down-regulated (Fig. S3B, Fig. S3C), of which 116 were related to immune response (Fig. 8F), and the expression of the 4 characteristic genes was also significantly different (Fig. 8G).

To understand the relationship between different immune microenvironment subtypes and biology, we carried out Gene set variation analysis (GSVA) on two different subtypes of patients base on all GO gene sets and KEGG gene sets. GO results showed that immune microenvironment subtype Ⅰ mainly involved cell division and chromosome structure. The biological functions related to subtype Ⅱ mainly include interferon production, leukocyte regulation, immune cell receptor regulation and other immune responses (Fig. 8H). KEGG results showed that immune microenvironment subtype Ⅱ was also related to immune response, including GNRH signal pathway, immune cell receptor signal pathway, ERBB signal pathway, p53 signal pathway, TGF β signal pathway, phosphorylation and amino acid metabolism. However, subtype Ⅰ is involved in cell division, homologous recombination and splicing (Fig. 8I). Subsequently, ssGSEA results showed that the immune infiltration efficiency of patients with subtype Ⅱ was significantly higher than that of patients with subtype Ⅰ, especially the infiltration rate of T cells, monocytes and macrophages (Fig. 9A). These results suggested that immune microenvironment subtype Ⅱ may be an immune subtype, which further explains the close relationship between keloid and immune response. In addition, the immune score of each immune microenvironment subtype was compared. The immune score of patients with subtype Ⅱ was higher than that of patients with subtype Ⅰ (Fig. 9B).

External validation of differential expression of characteristic genes.

Firstly, we used GSE44270, GSE7890, GSE145725 and GSE83286 to analyze the expression pattern of characteristic genes in keloid patients. Compared with the control group, the expression of PTPRC, STAT3 and IL1R1 in keloid samples decreased significantly, while MAPK1 increased significantly (Fig. 9C-F). Then we explored the expression trend of 4 characteristic genes in different tumors through pan-cancer analysis. The results showed that the 4 characteristic genes were down-regulated in most tumors (Fig. S4A-D). Moreover, the prognosis showed that the 4 characteristic genes were closely related to the overall survival of at least 8 tumor patients (Fig. S4E-H). Finally, we analyzed the expression of 4 characteristic immune microenvironment-related genes in human primary keloid fibroblasts by RT-PCR. The results were consistent with the datasets analysis, the expression levels of PTPRC, STAT3 and IL1R1 were significantly decreased in keloid fibroblasts, while the expression of MAPK1 gene was significantly increased (Fig. 9G).

It has been reported that keloid was easy to recur after surgery and difficult to cure, and will seriously affect the quality of life of patients due to aesthetic problems and functional symptoms (Naik ,2022; Hawash et al.,2021). Therefore, it is very important to establish an effective risk prognosis model and explore the factors closely related to the progression of keloid for the treatment of keloid. The data of this study suggest that the immune microenvironment was closely related to the development and occurrence of keloids. In addition, we used PPI, WGCNA, univariate analysis, LASSO regression analysis (Lars algorithm), SVM recursive feature elimination (SVM-REF) algorithm and multivariate logistic regression analysis to identify 4 characteristic genes related to immune microenvironment associated with keloid pathological biomarkers. and built a risk model that can stably predict keloid progression. Then 5 algorithms, namely, logistic regression (gradient descent method), naive Bayesian classification, support vector machine (SVM) classification, bp neural network classification and K-Nearest Neighbor (KNN) classification, were used to verify and evaluate the classification effect of the 4 characteristic genes, and the nomograph wa established to visual logical regression model. Finally, based on the 4 characteristic genes, we found 2 different immune microenvironment subtypes in keloid patients. Each subtype showed different biological functions and pathways, immune cell infiltration and immune characteristics. These results provide a new perspective for studying the relationship between keloid immune microenvironment and the diagnosis, classification and treatment of keloid patients.

It is known that the key cells of keloid formation are keloid fibroblasts, and their immune microenvironment is mainly composed of keloid fibroblasts and infiltrating immune cells (Macarak et al., 2021; Shan and Wang, 2021). Some researchers have reported that keloid fibroblasts can promote or inhibit keloid development through immune processes such as stress induced by immune cells in the microenvironment. In the analysis of keloid tissues, it was found that the number of macrophages, T cells, mast cells and Langerhans cells in keloid was significantly increased, and these immune cells could secrete cytokines to stimulate collagen production and promote scar process (Wilgus et al., 2020; Wang et al., 2022; Lv et al., 2020). Some researchers have also found that macrophages, Treg cells, CD⁸⁺ T cells, dendritic cells and natural killer cells play an important role in the pathological process of keloid when analyzing keloid immune infiltration (Niessen et al., 2004). However, the regulatory role between immune cells and keloid fibroblasts in keloid microenvironment has not been fully explored, and the role of immune cells in keloid pathology was still unclear. In this study, based on two machine algorithms (Glmnet algorithm and Lars algorithm), we carried out LASSO regression on 28 immune cells to screen out 8 immune cells with the strongest correlation (Effector memory CD8 T cell, Macrophage, Monocyte, Effector memory CD4 T cell, Type 17 T helper cell, Activated B cell, Natural killer cell and Immature differentiation cell) with keloid. Subsequently, 19 differentially expressed genes strongly related to the above 8 immune cells were identified, and PPI indicated that these 19 genes showed strong synergistic effect. Functional enrichment analysis showed that these differentially expressed genes were mainly involved in phosphatase activity and immune response related pathways, which also indicated that the effect of immune cells on keloid might be a synergistic process of immune genes and immune cells.

With the development of high-throughput sequencing technology and the establishment of disease databases, data mining to extract potential disease information from a large number of data has become an important analytical tool in clinical diagnosis and treatment. Machine learning, statistics and bioinformatics are common methods of data mining, and machine learning algorithms and statistical algorithms have been widely used in clinical practice, which can accurately predict the clinical status of patients and new biomarkers (Ichimasa et al., 2018; Kang et al., 2021). Li et al., established a mortality prediction model for patients with heart failure in intensive care unit based on machine learning and mining MIMIC-III database (Li et al., 2021). Some researchers also established a nomogram model of ependymoma prognosis using LASSO-Based machine algorithm based on the SEER database (Jia et al., 2021). In this study, we have identified the key characteristic factors MAPK1, PTPRC, STAT3 and IL1R1 using statistical methods such as univariate analysis and multivariate logistic regression, and machine learning algorithms such as SVM-REF and LASSO regression (Lars algorithm), and built a stable risk prediction model. Then the performance of the 4 feature genes was evaluated by five algorithms, namely, logical regression (gradient descent method), naive Bayesian classification, support vector machine (SVM) classification, bp neural network classification and K-Nearest Neighbor (KNN) classification, and the reliability of the classification algorithm was measured by ROC, accuracy, accuracy and F1. Combined with GSEA results, these 4 genes were associated with immune microenvironment and have good diagnostic ability in predicting keloid risk. Mitogen-activated protein kinase 1 (MAPK1) is a member of MAPK family, which participates in proliferation, development and differentiation, and belongs to MAPK signal pathway molecule (Zhang et al., 2022). It is reported that in head and neck tumors, abnormal pathway based on MAPK1 mutation can affect the immune microenvironment and ErbB3 signal to regulate the survival time of patients (Ngan et al., 2020). In breast cancer, the disturbance of mutation driven MAPK pathway is related to the negative regulation of immune response (Hendrickx et al., 2017). However, MAPK1 has not been studied in keloid and its immune microenvironment. Protein tyrosine phosphatase receptor type C (PTPRC) is an important regulator of T/B-cell antigen receptor-mediated activation, and plays a vital role in the innate immune system. Its abnormality can lead to immune deficiency, autoimmune or malignant tumor (Al Barashdi et al., 2021). Studies have shown that PTPRC can be used as a diagnostic tool for acute lymphoblastic leukemia to distinguish immature lymphoblastic cells from plasma cell tumors. In addition, PTPRC controls the signal pathway of the immune system, so it has been studied as a therapeutic target for many immune diseases including autoimmunity and organ transplantation (Rheinländer et al., 2018). Signal transducer and activator of transcription (STAT) protein is a family of cytoplasmic transcription factors that participate in cell proliferation, differentiation, angiogenesis and other biological processes (Zou et al., 2020). More and more evidences show that overactivation of STAT3 mediates tumor-induced immunosuppression at multiple levels. In addition, it is also over-activated in immune cells in TME and CAFs, inhibits the production of pro-inflammatory mediators such as IFN-γ and antigen presentation, and inhibits the tumor killing activity of effector cells (Zou et al., 2020; Piperi et al.,2019). In keloid, STAT3 can promote inflammatory cell death and fibrosis of keloid fibroblasts by inducing autophagy dysfunction and inhibiting the inflammatory process (Lee et al., 2022). Interleukin-1 receptor type 1 (IL1R1) belongs to the interleukin-1 receptor family and is an important mediator of many cytokines induced immune and inflammatory responses, playing an important role in various cells and organs (Tian et al., 2022). However, there are no reports about the effect of IL-1R1 on immune response and keloid. Based on the current reports of 4 characteristic genes, we can find that these 4 genes are basically involved in immune response or immune microenvironment, and the abnormal expression of 4 genes may be a predictive factor for the occurrence of keloid. In addition, the external validation dataset, the constructed nomogram, calibration curve and ROC all showed the reliable and stable diagnostic ability of these 4 characteristic genes.

In this study, we comprehensively explored 28 kinds of immune cells and immune-related genes in the keloid immune microenvironment, and determined 8 kinds of key immune infiltrating cells and 4 key immune characteristic genes (MAPK1, PTPRC, STAT3 and IL1R1) that can accurately predict the occurrence and development of keloid based on statistics and machine learning algorithms. Targeting these immune cells and genes may provide theoretical basis for the treatment of keloid. Compared with previous studies on keloid immunity, the immune cell spectrum we studied is more comprehensive and the evaluation method is more comprehensive. And added the content of immune microenvironment and immune characteristic gene analysis, which have not been reported in keloid before. In addition, the sample size of keloid research in the past is relatively small. In this study, we combine the data samples from previous studies into a large dataset for analysis, evaluate the immune characteristics of keloid patients, and use these datasets to independently verify the characteristic genes, which makes our results reliable. In addition, the sample size of previous studies on keloid was relatively small. In this study, data samples from previous studies were combined into a large dataset to analysis and evaluate the immune characteristics of patients with keloid, and these datasets were also used for independent verification of characteristic genes, thus making our results reliable. At the same time, based on the combined big dataset, we divided keloid patients into two different subtypes according to the relevant characteristic genes of immune microenvironment. Functional enrichment and KEGG analysis showed that subtype Ⅰ was mainly involved in cell division, chromosome structure and phosphorylation. However, subtype Ⅱ mainly involves immune response and other related functions, and subtype Ⅰ showed higher immune cell infiltration and immune score. These results suggest that epigenetic and immune microenvironment may be the key factors for the pathological heterogeneity of keloid. These data are also described for the first time in keloid, which can provide accurate data reference for the prevention and treatment of keloid. However, there are still some shortcomings in this study. Firstly, all the analyses in this study are based on open database datasets, and an additional number of clinical samples are needed for experimental evaluation and verification. Secondly, at present, the amount of keloid samples collected from the database is still relatively small, and more samples need to be collected for verification. Finally, more clinical information about keloid needs to be collected for evaluation.

In this study, a series of statistical methods, bioinformatics methods and machine learning algorithms were used to illustrate the correlation between the pathological process of keloid and the immune microenvironment. In addition, we identified 4 differentially expressed characteristic genes related to immune microenvironment. These 4 feature genes, whether they are classifiers based on five machine learning algorithms, or risk prediction models and nomographs, can stably and reliably predict the occurrence and development of keloid. In addition, based on 4 characteristic genes and immune microenvironment, we have carried out molecular classification of keloid patients. In conclusion, our research data comprehensively explored the immune microenvironment mode of keloid patients, which could be helpful to provide references and strategies for the follow-up pathological mechanism, early prevention and treatment of keloid.

ssGSEA: single sample gene set enrichment analysis; LASSO: Least absolute shrinkage and selection operator; WGCNA: weighted gene co-expression network analysis; RT-PCR: quantitative real-time polymerase chain reaction; PPI: protein-protein interaction; ROC: receiver operator characteristic curve; AUC: Area Under Curve; SVM-REF: support vector machine recursive feature elimination; SVM: support vector machine; KNN: K-Nearest Neighbor; GEO: Gene Expression Omnibus; GO: Gene Ontology; KEGG: Kyoto Encyclopedia of Genes and Genomes; FC: fold change; FDR: False Discovery Rate; BP: biological process; MF: molecular function; CC: cellular component; MCODE: Molecular Complex Detection; CDF: cumulative distribution function; MSigDB: Molecular Signatures Database; DEGs: differential expression genes; GSVA: gene set variation analysis.

Author contributions

Yongyan Pei: Plan writing, project management, draft writing and funding acquisition; Yikai Wu: Statistical analysis and data management; Mengqi Zhang: Bioinformatics analysis and data management; Xuemin Su: Cell culture and RT-qPCR, data analysis; Jiaji Zhao: Machine learning analysis, review and funding acquisition; Hua Cao: Funding acquisition, writing review and editing.

Conflict of interests

The authors declare that they have no competing interests.

Consent for publication

Not applicable.

Ethics approval and consent to participate

Not applicable.

Funding

This work was supported by Scientific research project of Guangdong Bureau of Traditional Chinese Medicine (20221215). The Project of Young Creative talents in Colleges and Universities of Guangdong Province Department of Education (2021KQNCX037). Guangdong Medical Research Foundation(B2023365). Scientific and Technological innovation team project of Guangdong Medical Products Administration(2022TDZ22).

Data Availability Statement

All datasets used for analysis (GSE44270, GSE7890 and GSE92566, GSE145725 and GSE83286) were derived from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) database, a publicly available dataset.

Acknowledgements

We thank all GEO data builders and data contributors, as well as the team that built the Sangerbox tool, and thank to Shanghai Ordovician Biotechnology Co., LTD. for providing the platform of biological information analysis.

Al Barashdi MA, Ali A, McMullin MF, Mills K (2021) Protein tyrosine phosphatase receptor type C (PTPRC or CD45). J Clin Pathol 74:548–552. https://doi.org/10.1136/jclinpath-2020-206927
Bhattacharya S, Dunn P, Thomas CG et al (2018) ImmPort, toward repurposing of open access immunological assay data for translational and clinical research. Sci Data 5:180015. https://doi.org/10.1038/sdata.2018.15
Breuer K, Foroushani AK, Laird MR et al (2013) InnateDB: systems biology of innate immunity and beyond–recent updates and continuing curation. Nucleic Acids Res 41:D1228–D1233. https://doi.org/10.1093/nar/gks1147
Charoentong P, Finotello F, Angelova M et al (2017) Pan-cancer Immunogenomic Analyses Reveal Genotype-Immunophenotype Relationships and Predictors of Response to Checkpoint Blockade. Cell Rep 18:248–262. https://doi.org/10.1016/j.celrep.2016.12.019
Chen Y, Jin Q, Fu X et al (2019) Connection between T regulatory cell enrichment and collagen deposition in keloid. Exp Cell Res 383:111549. https://doi.org/10.1016/j.yexcr.2019.111549
Ekstein SF, Wyles SP, Moran SL, Meves A (2021) Keloids: a review of therapeutic management. Int J Dermatol 60:661–671. https://doi.org/10.1111/ijd.15159
Elsaie ML (2021) Update on management of keloid and hypertrophic scars: A systemic review. J Cosmet Dermatol 20:2729–2738. https://doi.org/10.1111/jocd.14310
Feng C, Shan M, Xia Y et al (2022) Single-cell RNA sequencing reveals distinct immunology profiles in human keloid. Front Immunol 13:940645. https://doi.org/10.3389/fimmu.2022.940645
Hawash AA, Ingrasci G, Nouri K, Yosipovitch G (2021) Pruritus in Keloid Scars: Mechanisms and Treatments. Acta Derm Venereol 101:adv00582. https://doi.org/10.2340/00015555-3923
Hendrickx W, Simeone I, Anjum S et al (2017) Identification of genetic determinants of breast cancer immune phenotypes by integrative genome-scale analysis. Oncoimmunology. 6: e1253654. https://doi.org/.1080/2162402X.2016.1253654
Ichimasa K, Kudo S-E, Mori Y et al (2018) Artificial intelligence may help in predicting the need for additional surgery after endoscopic resection of T1 colorectal cancer. Endoscopy 50:230–240. https://doi.org/10.1055/s-0043-122385
Jia Z, Yan Y, Wang J et al (2021) Development and validation of prognostic nomogram in ependymoma: A retrospective analysis of the SEER database. Cancer Med 10:6140–6148. https://doi.org/10.1002/cam4.4151
Kang J, Choi YJ, Kim I-K et al (2021) LASSO-Based Machine Learning Algorithm for Prediction of Lymph Node Metastasis in T1 Colorectal Cancer. Cancer Res Treat 53:773–783. https://doi.org/10.4143/crt.2020.974
Lee S-Y, Lee AR, Choi JW et al (2022) IL-17 Induces Autophagy Dysfunction to Promote Inflammatory Cell Death and Fibrosis in Keloid Fibroblasts via the STAT3 and HIF-1α Dependent Signaling Pathways. Front Immunol 13:888719. https://doi.org/10.3389/fimmu.2022.888719
Li F, Xin H, Zhang J et al (2021) Prediction model of in-hospital mortality in intensive care unit patients with heart failure: machine learning-based, retrospective analysis of the MIMIC-III database. BMJ Open 11:e044779. https://doi.org/10.1136/bmjopen-2020-044779
Liu J, Huang Y, Gong Y et al (2022) CTHRC1 + fibroblasts are stimulated by macrophage-secreted SPP1 to induce excessive collagen deposition in keloids. Clin Transl Med 12:e1115. https://doi.org/10.1002/ctm2.1115
Liu S, Yang H, Song J et al (2022) Keloid: Genetic susceptibility and contributions of genetics and epigenetics to its pathogenesis. Exp Dermatol 31:1665–1675. https://doi.org/10.1111/exd.14671
Lv W, Ren Y, Hou K et al (2020) Epigenetic modification mechanisms involved in keloid: current status and prospect. Clin Epigenetics 12:183. https://doi.org/10.1186/s13148-020-00981-8
Macarak EJ, Wermuth PJ, Rosenbloom J, Uitto J (2021) Keloid disorder: Fibroblast differentiation and gene expression profile in fibrotic skin diseases. Exp Dermatol 30:132–145. https://doi.org/10.1111/exd.14243
Naik PP (2022) Novel targets and therapies for keloid. Clin Exp Dermatol 47:507–515. https://doi.org/10.1111/ced.14920
Ngan H-L, Liu Y, Fong AY et al (2020) MAPK pathway mutations in head and neck cancer affect immune microenvironments and ErbB3 signaling. Life Sci Alliance 3:e201900545. https://doi.org/10.26508/lsa.201900545
Niessen FB, Schalkwijk J, Vos H, Timens W (2004) Hypertrophic scar formation is associated with an increased number of epidermal Langerhans cells. J Pathol 202:121–129. https://doi.org/10.1002/path.1502
Piperi C, Papavassiliou KA, Papavassiliou AG (2019) Pivotal Role of STAT3 in Shaping Glioblastoma Immune Microenvironment. Cells 8:1398. https://doi.org/10.3390/cells8111398
Rheinländer A, Schraven B, Bommhardt U (2018) CD45 in human physiology and clinical medicine. Immunol Lett 196:22–32. https://doi.org/10.1016/j.imlet.2018.01.009
Ru B, Wong CN, Tong Y et al (2019) TISIDB: an integrated repository portal for tumor-immune system interactions. Bioinformatics 35:4200–4202. https://doi.org/10.1093/bioinformatics/btz210
Shan M, Liu H, Hao Y et al (2022) The Role of CD28 and CD8 + T Cells in Keloid Development. Int J Mol Sci 23:8862. https://doi.org/10.3390/ijms23168862
Shan M, Wang Y (2022) Viewing keloids within the immune microenvironment. Am J Transl Res 14:718–727
Shen W, Song Z, Zhong X (2017) Sangerbox: A comprehensive, interaction-friendly clinical bioinformatics analysis platform. iMeta 1:e36. https://doi.org/10.1002/imt2.36
Tian Y, Sun Y, Wu J et al (2020) Impact of IL1R1 polymorphisms on the risk of head and neck cancer in Chinese Han population. Gene 757:144927. https://doi.org/10.1016/j.gene.2020.144927
Wang X, Liang B, Li J et al (2022) Identification and characterization of four immune-related signatures in keloid. Front Immunol 13:942446. https://doi.org/10.3389/fimmu.2022.942446
Wilgus TA, Ud-Din S, Bayat A (2020) A Review of the Evidence for and against a Role for Mast Cells in Cutaneous Scarring and Fibrosis. Int J Mol Sci 21:9673. https://doi.org/10.3390/ijms21249673
Xie T, Pei Y, Shan P et al (2022) Identification of miRNA-mRNA Pairs in the Alzheimer’s Disease Expression Profile and Explore the Effect of miR-26a-5p/PTGS2 on Amyloid-β Induced Neurotoxicity in Alzheimer’s Disease Cell Model. Front Aging Neurosci 14:909222. https://doi.org/10.3389/fnagi.2022.909222
Yu G, Wang L-G, Han Y, He Q-Y (2012) clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 16:284–287. https://doi.org/10.1089/omi.2011.0118
Zhang Y, Zhou J, Liu S, Ji Z (2022) MAPK1 Is Regulated by LOC102188416/miR-143-3p Axis in Dairy Goat Mammary Epithelial Cells. Genes (Basel) 13:1013. https://doi.org/10.3390/genes13061013
Zou S, Tong Q, Liu B et al (2020) Targeting STAT3 in Cancer Immunotherapy. Mol Cancer 19:145. https://doi.org/10.1186/s12943-020-01258-7

No competing interests reported.

supplementarymaterials.docx

Download PDF

Journal Publication

published 12 Dec, 2023

Read the published version in Biochemical Genetics →

Editorial decision: Major revision
06 Oct, 2023
Reviews received at journal
05 Sep, 2023
Reviewers agreed at journal
31 Aug, 2023
Reviewers invited by journal
31 Aug, 2023
Submission checks completed at journal
05 Aug, 2023
Editor assigned by journal
05 Aug, 2023
First submitted to journal
04 Aug, 2023

You are reading this latest preprint version

Screening of keloid immune microenvironment-related genes and their effects on keloid proliferation and migration

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Materials and methods

Data acquisition

Analysis of immune cell infiltration

Screening of key characteristic immune cells

Weighted correlation network analysis (WGCNA)

Identification of immune microenvironment-related differentially expressed genes

Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG)

Establishment of risk prediction model

Validation of characteristic genes in risk prediction model

Establishment and evaluation of nomogram

Consistent cluster recognition of different immune microenvironment subtypes

Gene set enrichment analysis (GSEA) and Gene set enrichment analysis (GSVA)

Cell culture

Real-time RT-PCR

Statistical analysis

Results

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1