Identification and validation of immune-related biomarkers based on machine learning in patients with prostate cancer

doi:10.21203/rs.3.rs-1503908/v1

Download PDF

Research

Identification and validation of immune-related biomarkers based on machine learning in patients with prostate cancer

https://doi.org/10.21203/rs.3.rs-1503908/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this older preprint version

Read the latest preprint version →

Objectives This study aims to explore potential diagnostic biomarkers associated with prostate cancer (PCa) and analyze the pathological role of immune cell infiltration in the disease.

Methods Six previously published datasets containing gene expression data (GSE8511, GSE14206, GSE46602, GSE55945, GSE69223 and GSE71016) were downloaded from the Gene Expression Omnibus (GEO) database. Take GSE8511, GSE14206, GSE46602 and GSE55945 as the train group (127 PCa cases and 52 normal controls) and GSE69223 and GSE71016 as the test group (63 PCa cases and 62 normal controls). Differentially expressed genes (DEGs) were screened between 127 PCa cases and 52 normal controls. Candidate biomarkers of PCa were then identified via the use of a least absolute shrinkage and selector operation (LASSO) regression model and support vector machine recursive feature elimination (SVM-RFE) analyses. We performed difference analysis of these genes in the test group, with P < 0.05 were regarded statistically significant. The area under the receiver operating characteristic curve (AUC) value was obtained and used to evaluate discriminatory ability in the train group, with hub genes being identified as those with an AUC > 85%. The expression level and diagnostic value of the biomarkers in PCa were further validated in the GSE69223 and GSE71016 dataset. Lastly, CIBERSORT algorithm and ESTIMATE method were employed to assess the relative immune cell infiltrations of each sample.

Results AOX1, APOC1, ARMCX1, FLRT3, GSTM2 and HPN were identified as potential PCa diagnostic biomarkers, and these results were validated using the GSE69223 and GSE71016 datasets. Immune cell infiltration analysis revealed that AOX1, APOC1, ARMCX1, FLRT3, GSTM2 and HPN were correlated with Mast cells resting, Macrophages M0, Macrophages M2, Neutrophils, NK cells activated, T cells CD4 memory resting, T cells gamma delta, T cells follicular helper, T cells CD4 naive and B cells memory.

Conclusion We adopted a comprehensive strategy to screen biomarkers related to PCa and explore the critical role of immune cell infiltration in PCa. AOX1, APOC1, ARMCX1, FLRT3, GSTM2 and HPN can be used as diagnostic markers of PCa, providing new insights for future studies on the incidence and the molecular mechanisms of PCa.

biomarker

prostate cancer

immune infiltration

CIBERSORT

diagnostic

Prostate cancer (PCa) is a major global health issue. It causes 1.6 million cases and 366000 deaths worldwide each year. It is the second most common cancer and the fifth leading cause of cancer death among men in the world^[1]. Although there are some curable treatments such as radical prostatectomy (RP), PCa still has a high recurrence rate due to its biological characteristics, distant micrometastases and focal residuals. Salvage therapy in the early stage of recurrence can reduce the rate of distant metastases, prolong survival, and even cure the tumor. Therefore, early diagnosis of PCa can greatly reduce the mortality and improve the prognosis of patients^[2].

PCa has traditionally been diagnosed by digital rectal examination (DRE), prostate specific antigen (PSA) blood test and prostate biopsy. However, PSA lacks specificity, leading to over-diagnosis and over-treatment of PCa. There is an increasing clinical need for the identification of novel prognostic, predictive and treatment response biomarkers that can be used to provide a precision medicine approach to PCa management^[3]. For example, some studies have shown that the deletion of phosphatase and tensin homolog (PTEN) is related to the poor prognosis of PCa patients^[4]. Its loss in biopsy samples predicted an increased risk of castration-resistant prostate cancer (CRPC), metastasis and PCa specific mortality^{[5, 6]}. To prospectively analyze the potential molecular mechanisms and genomic consequences of the co-loss of BRCA2 and RB1 in prostate cancer. We can find that the loss of BRCA2 leads to the castration-resistant phenotype in human prostate cancer cell lines (LNCaP and lapc4)^[7].Therefore, it is possible to elucidate mechanisms underlying PCa development and explore novel potential diagnostic strategies for this disease. However, so far, there is no study that combined least absolute shrinkage and selector operation (LASSO) regression model and support vector machine recursive feature elimination (SVM-RFE) analyses to identify biomarkers of PCa.

In recent years, immunology researches have shown that immune cell infiltration plays an important role in the occurrence and development of PCa. For example, Flammiger et al. detected regulatory T cells (Tregs) in prostate cancer specimens by forkhead box P3 (FOXP3) immunohistochemistry found that the increased Tregs in prostate cancer specimens were associated with advanced and deteriorating prognosis^[8]. Eastham et al. performed a comparative study of normal patients and PCa patients and found that the level of transforming growth factor beta (TGF-β) in PCa patients increased, promoting both migration and invasion by PCa cells^[9]. Tissue microarray analysis confirmed decreased levels of FOXA1 protein and increased TGF-β signaling pathway in CRPC as compared with primary tumors suppress CRPC progression^[10]. In addition, some studies have shown that tyrosine hydroxylase 2 (Th2) and central memory T cell (TCM) are associated with PCa recurrence after RP and they are independent protective factors^[11]. However, so far, few studies have applied CIBERSORT algorithm to explore immune cell infiltration in PCa and candidate biomarkers of PCa.

Herein, we downloaded six publically accessible PCa-related datasets from the the Gene Expression Omnibus (GEO) database. We merged the four datasets into a metadata cohort and take it as the train group. And we merged the other two datasets into a metadata cohort and take it as the treat group. In the train group, differentially expressed genes (DEGs) were screened between 127 PCa cases and 52 normal controls. The Machine-learning algorithms was used to screen and identify diagnostic biomarkers for PCa. These candidate genes were identified and validated in the treat group. In addition, we also use CIBERSORT algorithm to study the correlation between biomarkers and PCa-infiltrating immune cells, so as to better understand the molecular immune mechanism involved in PCa as well as provide a practical and theoretical basis for subsequent research.

Gene expression data acquisition and processing

We searched all the data about PCa in the geo database (https:// www. ncbi. nlm. nih. gov/ geo/). Ultimately, raw data from the GSE8511, GSE14206, GSE46602, GSE55945, GSE69223 and GSE71016 datasets were downloaded from the GEO database. These datasets were divided into the train group (GSE8511, GSE14206, GSE46602 and GSE55945; 127 PCa cases and 52 normal controls) and the test group (GSE69223 and GSE71016; 63 PCa cases and 62 normal controls). The datasets were merged into a metadata cohort in the train group or the test group, and the combat function of the “SVA” package was used to preprocess and remove batch effects^[12]. (Table 1).

Identification of DEGs in PCa

To identify the DEGs between 127 PCa cases and 52 normal controls in the train group, the R (http://www.bioconductor.org/) package “limma” was applied in the standard comparison mode^[13]. The threshold points of DEGs were set as |log fold change (FC)| > 1, adjusted false discovery rate P < 0.05. In the train group, we compare 127 PCa cases with 52 normal controls. The volcano map was plotted to identify DEGs with “ggplot2” package in R software. Then, R package “pheatmap” was used to draw the heat map of DEGs.

Functional enrichment analysis

Gene module-related functions were identified through gene ontology (GO), Kyoto encyclopedia of genes and genomes (KEGG), and disease ontology (DO) functional enrichment analyses conducted using the R “clusterProfiler” package^[14], with P < 0.05 as the enrichment threshold. The gene set enrichment analysis (GSEA) of the train group was achieved by applying signal pathway differences. The GSEA analysis of gene expression matrix was conducted by ‘clusterProfiler’ and ‘enrichplot’ packages, and ‘c2.cp.kegg.v7.4.symbols.gmt’ was selected as the reference gene^[15]. We performed KEGG GSEA analysis on the PCa and normal cases of train group respectively. P < 0.05 were considered significant enrichment.

Screening characteristic related biomarkers via machine learning

To identify significant prognostic variables, two machine-learning algorithms were used to analyze PCa-related biomarkers. The LASSO algorithm was carried out using the “glmnet” package in R to identify the genes significantly associated with the discrimination of PCa and normal cases^[16]. Also, as a surveillant machine learning method to support vectors, the support vector machine (SVM) finds the best variables by deleting the feature vectors generated by the SVM. To avoid overfitting, a recursive feature elimination (RFE) algorithm was employed to select the optimal genes from the meta-data cohort. Therefore, to identify the set of genes with the highest discriminative power, SVM-RFE was applied to select the appropriate features. The SVM-RFE classifier from R package “e1071” was adopted for classification analysis of the selected biomarkers in the diagnosis of PCa^[17]^. Finally, we use Venn diagram to find the overlapping genes of the two algorithms, and these genes will be further verified in the test group.

Diagnostic value of the biomarkers in PCa

To test the predictive value of the identified biomarkers, we used R package "ggpubr" to analyze the differences of these genes in the test group, with P < 0.05 were regarded statistically significant^[18]. Then, We use R package "proc" to generate an ROC curve in the train group (127 PCa cases and 52 normal controls)^[19], with hub genes being identified as those with the area under the receiver operating characteristic curve (AUC) > 85%. The AUC value was utilized to determine the diagnostic effectiveness in discriminating PCa from control samples and further validated in the test group (63 PCa cases and 62 normal controls).

Assessment of immune cell infiltration

The CIBERSORT algorithm was utilized to filter 22 kinds of immune cell matrix. The putative abundance of immune cells was estimated using a reference set with 22 types of immune cell subtypes with 1,000 permutations^[20]. According to p < 0.05, the immune cell infiltration matrix was obtained. The ‘corrplot’ package was deployed to draw the correlation heatmap and boxplot to visualize the correlation of 22 kinds of immune cell infiltration^[21]. Violin plots were drawn using the “vioplot” package in R to visualize the differences in immune cell infiltration between the PCa and control samples^[22].

Analysis of correlations between identified genes and immune cell infiltration

The association of the identified gene biomarkers expression levels with the levels of infiltrating immune cells were explored through Spearman’s rank correlation analyses in R software^[23]. The resulting associations were conducted and visualized using the chart technique with “ggpubr” package^[24].

Statistical Analysis

All statistical analyses were conducted using the R software version 4.1.2 and perl version 5.32.1. P < 0.05 was regarded as statistically significant.

Identification of DEGs in PCa

The DEGs of 127 PCa cases and 52 normal controls from the GEO databases (GSE8511, GSE14206, GSE46602 and GSE55945) were identified by R package “limma”. There were 37 DEGs, including 17 upregulated and 20 downregulated genes. |log fold change (FC)| > 0 indicates that the gene is up-regulated in the train group, and |log fold change (FC)| < 0 indicates that the gene is down- regulated, which can be showed in the volcano plot (Figure 1A). The heat map showed the expression levels of 37 DEGs (Figure 1B).

Functional enrichment analysis

To assess the potential biological functions associated with the 37 DEGs. GO, KEGG, DO and GSEA analyses were conducted with the R “clusterProfiler” package. The GO results demonstrated that the majority of these genes were associated with basement membrane organization, cornification and positive regulation of secretion by cell (Figure 2A). KEGG enrichment analyses revealed the genes to be associated with the drug metabolism-cytochrome P450, nicotinate and nicotinamide metabolism and retinol metabolism (Figure 2B). The DO results demonstrated that diseases enriched by DEGs were mainly associated with chronic myeloproliferative disease, epidermolysis bullosa, integumentary system disease, vesiculobullous skin disease, peripheral primitive neuroectodermal tumor and prostate cancer (Figure 2C). The GSEA results in PCa group demonstrated that the enriched pathways mainly involved bladder cancer, cell cycle, purine metabolism, ribosome and toll like receptor signaling pathway (Figure 2D). The GSEA results in control group demonstrated that the enriched pathways mainly involved drug metabolism cytochrome P450, focal adhesion, glutathione metabolism, metabolism of xenobiotics by cytochrome P4 and vascular smooth muscle contraction (Figure 2E).

Screening diagnostic feature biomarkers for PCa

Two different algorithms were used to screen potential diagnostic biomarkers. The LASSO logistic regression algorithm was used to identify twenty-one genes from robust DEGs as potential PCa-related biomarkers (Figure 3A). Twenty-eight genes among the DEGs were determined using the SVM-RFE algorithm (Figure 3B). Finally, the gene markers obtained by the two algorithms were overlapped via Venn diagram, and the sixteen related biomarkers, including AOX1, HPN, GSTM2, APOC1, ARMCX1, FLRT3, MSMB, KRT15, GDF15, DLX1, CD177, NTNG2, CPLX3, ACSM1, ERG and CD38, were obtained (Figure 3C).

Identification and validation of diagnostic feature biomarkers for PCa

In order to produce more accurate and reliable results, GSE69223 dataset and GSE71016 dataset are used to assess the expression levels of these six genes. The expression levels of ACSM1, APOC1, DLX1, GDF15 and HPN in PCa tissues were evidently higher than those in the normal tissue (Figure 4A–E; all P < 0.05). Conversely, the opposite results were observed for AOX1, ARMCX1, CD177, FLRT3, GSTM2, KRT15, and NTNG2 (Figure 4F–L; all P < 0.05). Besides, there was no significant difference in the levels of CD38, CPLX3, ERG and MSMB between PCa tissues and the normal tissue (Supplement Figure S1A-S1D; all P > 0.05). We drew ROC curves of the above twelve meaningful genes in the train group, with hub genes being identified as those with an AUC > 85%. Then, we indentified six PCa-related diagnostic genes, with an AUC of 0.921 (95% CI 0.878−0.956) in AOX1, AUC of 0.853 (95% CI 0.782−0.919) in APOC1, AUC of 0.883 (95% CI 0.834−0.928) in ARMCX1, AUC of 0.854 (95% CI 0.796−0.904) in FLRT3, AUC of 0.877 (95% CI 0.823−0.923) in GSTM2 and AUC of 0.871 (95%CI 0.817−0.921) in HPN. Then, a powerful discrimination ability was confirmed in the GSE69223 and GSE71016 datasets, with an AUC of 0.726 (95% CI 0.641−0.810) in AOX1, AUC of 0.701 (95% CI 0.607−0.792) in APOC1, AUC of 0.722 (95% CI 0.625−0.808) in ARMCX1, AUC of 0.719 (95% CI 0.625−0.804) in FLRT3, AUC of 0.696 (95% CI 0.600−0.785) in GSTM2 and AUC of 0.789 (95%CI 0.707−0.863) in HPN (Figure 5A–F).

Assessment of immune cell infiltration

The CIBERSORT algorithm was next utilized to visualize infiltration of 22 different types of immune cells in the train group (Figure 6A). And the CIBERSORT algorithm showcased the infiltration of 22 kinds of immune cells, with heatmaps revealing strong positive correlations between T cells CD4 memory resting and plasma cells (r = 0.54) and strong negative correlations between T cells CD4 memory resting and macrophages M1 (r = − 0.51) (Figure 6B). We explored the composition of immune cells in PCa vs. normal control tissues. The results indicated that T cells CD8 in PCa was remarkably higher compared with the normal controls (P = 0.032), while mast cells resting was lower than the normal controls (P = 0.005; Figure 6C).

Correlation analysis between PCa-related biomarkers and immune infiltrating cells

Correlation analysis showed that AOX1 was positively correlated with mast cells resting (R = 0.4, P = 0.036) and negatively correlated with macrophages M0 (R = - 0.46, P = 0.013; Figure 7A). APOC1 was positively correlated with macrophages M0 (R = 0.63, P < 0.001), neutrophils (R = 0.4, P = 0.035), and macrophages M2(R = 0.4, P = 0.036) and negatively correlated with mast cells resting (R = - 0.59, P = 0.001; Figure 7B). ARMCX1 was positively correlated with NK cells activated (R = 0.38, P = 0.045) and mast cells resting (R = 0.38, P = 0.049; Figure 7C). FLRT3 was positively correlated with T cells CD4 memory resting (R = 0.41, P = 0.032) and negatively correlated with T cells gamma delta (R = - 0.43, P = 0.023; Figure 7D). GSTM2 was positively correlated with T cells follicular helper (R = 0.43, P = 0.022) and negatively correlated with T cells CD4 naive (R = - 0.41, P = 0.029; Figure 7E). HPN was positively correlated with B cells memory (R = 0.47, P = 0.011) and macrophages M0 (R = 0.41, P = 0.032) and negatively correlated with mast cells resting (R = - 0.41, P = 0.031; Figure 7F).

Nowadays, PCa remains one of the main causes of cancer-related death in men. In view of the recent success of immunotherapy in many hematological and solid malignancies, this territory is being explored widely in PCa^[25]. More and more researchers recognize that immune cell infiltration is related to the diagnosis of cancer and other diseases^[26]. Therefore, immunotherapy may become a feasible tool against PCa. The CIBERSORT algorithm has been successfully applied to identify tumor-infiltrating immune cells and its correlation with prognosis of gastric cancer, colorectal cancer, breast cancer and osteosarcoma^{[20, 27-29]}. However, its importance in prostate cancer has not yet to be elucidated. In this study, we aimed to identify candidate diagnostic biomarkers of PCa and to study the value of immune cell infiltration in PCa.

As far as we know, this is the first retrospective study to screen the diagnostic biomarkers of PCa by combining LASSO and RVM-RFE algorithm, and analyze the immune cell infiltration of PCa by CIBERSORT algorithm. We downloaded six datasets from the GEO database. Four are merged as the train group and the other two are merged as the test group. In the train group, we identified 37 DEGs, including 17 upregulated and 20 downregulated genes. The results of GSEA in PCa cases demonstrated that the enriched pathways mainly involved bladder cancer, cell cycle, purine metabolism, ribosome and toll like receptor signaling pathway.

Based on the two algorithms, we screened sixteen genes from robust DEGs as potential PCa-related biomarkers. Then, we analyzed the differences of these sixteen genes in the test group, and we found that twelve genes were statistically significant (P < 0.05). Finally, we drew the ROC curves of these twelve genes, and we identified the final six PCa-related diagnostic biomarkers (AOX1, APOC1, ARMCX1, FLRT3, GSTM2 and HPN). In order to verify the predictive value of these six diagnostic biomarkers, we calculated their ROC curves in the test group. Aldehyde oxidase produces hydrogen peroxide, which can catalyze the formation of superoxide under some conditions. Aldehyde oxidase is a candidate for amyotrophic lateral sclerosis. Aldehyde oxidase 1 (AOX1) is a flavin- containing monooxygenase, which is involved in the detoxification of a variety of aldehydes and nitrogen-containing heterocyclic compounds. Its most common substrates are sulfur-containing and nitrogen-containing alien organisms^[30]. Li et al. utilized a cox proportional hazard model to test the association between a total of 6126633 single nucleotide polymorphisms (SNPs) and PCa specific survival time, and used expression quantitative trait locus analysis to test the association between positive candidate SNPs and gene expression. The SNP rs73055188 at the AOX1 locus is associated with the specific survival time of PCa, and the expression level of AOX1 is associated with the biochemical recurrence of PCa^[31]. As an important lipoprotein, apolipoprotein is involved in maintaining lipoprotein structure, regulating the interaction between lipoprotein and cell receptor, and regulating the activity of enzymes related to intravascular lipoprotein metabolism. Apolipoprotein C-I (APOC1) located on chromosome 19, is the smallest apolipoprotein (only 6.6 kDa). It is a component of triglyceride-rich lipoproteins and high-density lipoproteins and it is very important for the metabolism of plasma lipoprotein^[32]. Su et al. use RNA interference (RNAi) to knockdown the expression of APOC1 in PCa cell lines. Then it was found that APOC1 mediated the survival, cell cycle distribution and apoptosis of PCa cells by regulating surviving/Rb /p21/caspase-3 signaling pathway^[33]. Armadillo repeat containing X-linked 1 (ARMCX1), a mammalian specific gene, encodes mitochondrial localization protein. It is a key regulator of mitochondrial transport in damaged axons and may affect neuronal survival and axon regeneration^[34]. At present, there is a lack of research between ARMCX1 and PCa. Fibronectin leucine rich transmembrane protein 3 (FLRT3) is an axon guidance-related factor related to the regulation of neuronal cell growth and morphogenesis^[35]. Balabhadrapatruni et al. used Methods of gene knockdown to show that prolyl 4-hydroxylase subunit alpha 1 (P4HA1) expression is necessary for the proliferation and invasion of PCa cells and can be used as a feasible therapeutic target. They also verified the results of FLRT3 in benign prostate, PCa and metastatic PCa tissues by western blot analysis, and found a negative correlation between P4HA1 and FLRT3. It was finally confirmed that FLRT3 promoted the migration and invasion of P4HA1-mediated PCa cells^[36]. As a member of the superfamily of multifunctional enzymes, Glutathione S-transferases play an important role in the defense of a variety of toxic and carcinogenic substances by catalyzing the binding of glutathione (GSH) conjugation to electrophilic compounds. Glutathione S-transferase mu 2 (GSTM2) is an important phase II antioxidant enzyme in eukaryotes, which is involved in the production of prostaglandin E2 (PGE2)^[37]. Several studies on GSTM2 have been carried out in PCa. Albawardi et al. found that GSTM2 was amplificated in PCa based on SNP microarrays^[38]. In addition, three other studies have shown that GSTM2 is highly methylated in PCa^[39-41]. Hepsin (HPN) is a type II transmembrane, but not secreted, protein^[42]. Ma et al. found that the protein expression level of HPN in PCa with a Gleason score > 6 was significantly higher than those with a Gleason score = 6 through single cell RNA sequencing, which determined that HPN was a potential biomarker for PCa diagnosis and stratification^[43]. The above evidences show that except armcx1, other genes play an important role in the development of PCa.

We further evaluated the immune cell infiltration of PCa using CIBERSORT to explore the role of immune cell infiltration in PCa. An increased infiltration of T cells CD8 and a decreased infiltration of mast cells resting are considered to be associated to the occurrence and progression of PCa. Correlation analysis of PCa-related biomarkers and immune infiltrating cells showed that AOX1, APOC1, ARMCX1, GSTM2 and HPN were significantly correlated with mass cells resting. And HPN, AOX1 and APOC1 were significantly correlated with macrophages M0. Florent et al. performed immunohistochemistry on the tumors from 51 patients with node-positive PCa. It was found that the high density of CD8 + T cells in the tumors was associated with the high risk of clinical progression in patients with node-positive PCa^[44]. Mast cells play an important role in inflammation, hypersensitivity and fibrotic diseases. Importantly, mast cells also play a decisive role in tumor growth. CIBERSORT algorithm was used to analyze 52 normal prostate tissues and 497 primary tumors of PCa patients from TCGA. It was found that the proportion of static mast cells in PCa was significantly different from that in normal tissues. A higher proportion of resting mast cells is associated with poor prognosis. Radiotherapy and targeted molecular therapy can affect immune infiltration of resting mast cells^[45]. Somaiyeh et al. found that etoposide-treated cancer cells M0 macrophases, THP-1 cells incubated with the supernatant of P human prostate cancer cell line (PC3) cells treated with toll like receptor 4 (TLR4) agonists had obvious protective effect on etoposide-induced PCa cells apoptosis by flow cytometry and enzyme-linked immunosorbent assay (ELISA)^[46]. A large amount of evidences mentioned above and our current findings show that several types of invasive immune cells play an important role in PCa and should become the focus of future research.

Our study has several limitations that should be taken into account. The main limitation is the small sample size of the published datasets. Our findings need to be validated in other datasets and clinical trials to determine whether AOX1, APOC1, armcx1, FLRT3, gstm2 and HPN can be used as biomarkers of PCa. CIBERSORT algorithm was based on limited retrospective gene data. Although some results of previous research are consistent with our results, the analysis of immune cell infiltration in PCa is still limited, and our conclusion should be verified by a prospective study with larger sample size.

In summary, we found that AOX1, APOC1, ARMCX1, FLRT3, GSTM2 and HPN were PCa-related biomarkers. T cells CD8、mast cells resting、Macrophages M0、NK cells activated、Neutrophils、Macrophages M2、B cells memory、T cells CD4 memory resting、T cells gamma delta、T cells follicular helper and T cells CD4 naive may be associated with the occurrence and development of PCa. Further exploring the interaction between immune cell infiltration will help to improve the immunomodulatory treatment of PCA patients. The further exploration of the interaction between immune cell infiltration and PCa will help to improve the immunomodulatory treatment of patients with PCa.

PCa: prostate cancer; DEGs: Differentially expressed genes; LASSO: least absolute shrinkage and selector operation; SVM-RFE: support vector machine recursive feature elimination; AUC :The area under the receiver operating characteristic curve; RP: radical prostatectomy; DRE: digital rectal examination; PSA: prostate specific antigen; PTEN: phosphatase and tensin homolog; CRPC: castration-resistant prostate cancer; Tregs: regulatory T cells; FOXP3: forkhead box P3; TGF-β: ransforming growth factor beta; Th2: hydroxylase 2; TCM: central memory T cell; GEO: Gene Expression Omnibus; GO: gene ontology; KEGG: Kyoto encyclopedia of genes and genomes; DO : disease ontology; GSEA: gene set enrichment analysis; SVM: support vector machine; RFE: recursive feature elimination; AOX1: Aldehyde oxidase 1; SNPs: single nucleotide polymorphisms; APOC1: Apolipoprotein C-I; RNAi: RNA interference; ARMCX1: Armadillo repeat containing X-linked 1; FLRT3: Fibronectin leucine rich transmembrane protein 3; P4HA1: prolyl 4-hydroxylase subunit alpha 1; GSH: glutathione; GSTM2: Glutathione S-transferase mu 2; PGE2: prostaglandin E2; HPN: Hepsin; PC3: P human prostate cancer cell line; TLR4: toll like receptor 4; ELISA: enzyme-linked immunosorbent assay.

Author contributions

KSY and RQZ performed study concept and design. KSY, PTD, APZ and HQL performed development of methodology and writing, review and revision of the paper. KSY, JJY and QZ provided acquisition, analysis and interpretation of data, and statistical analysis. HL provided financial support. All authors read and approved the final paper.

Availability of data and materials

All data used in this work can be acquired from the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/).

Funding

There is no funding to report.

Acknowledgments

We acknowledge GEO database for providing their platforms and contributors for uploading their meaningful datasets.

Consent for Publication

All authors have read and approved the content, and they agree to submit it for consideration for publication in the journal.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

GEO database belong to public database. The patients involved in the database have obtained ethical approval. Users can download relevant data for free for research and publish relevant articles. Our study is based on open source data, so there are no ethics approval and consent to participate.

1. Tsujino T, Komura K, Inamoto T, Azuma H. CRISPR Screen Contributes to Novel Target Discovery in Prostate Cancer. International journal of molecular sciences 2021; 22(23).

2. Carlsson SV, Vickers AJ. Screening for Prostate Cancer. The Medical clinics of North America 2020; 104(6):1051-1062.

3. Cimadamore A, Mazzucchelli R, Lopez-Beltran A, Massari F, Santoni M, Scarpelli M, et al. Prostate Cancer in 2021: Novelties in Prognostic and Therapeutic Biomarker Evaluation. Cancers 2021; 13(14).

4. Fisher KW, Montironi R, Lopez Beltran A, Moch H, Wang L, Scarpelli M, et al. Molecular foundations for personalized therapy in prostate cancer. Current drug targets 2015; 16(2):103-114.

5. Lotan TL, Carvalho FL, Peskoe SB, Hicks JL, Good J, Fedor H, et al. PTEN loss is associated with upgrading of prostate cancer from biopsy to radical prostatectomy. Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc 2015; 28(1):128-137.

6. Picanço-Albuquerque CG, Morais CL, Carvalho FL, Peskoe SB, Hicks JL, Ludkovski O, et al. In prostate cancer needle biopsies, detections of PTEN loss by fluorescence in situ hybridization (FISH) and by immunohistochemistry (IHC) are concordant and show consistent association with upgrading. Virchows Archiv : an international journal of pathology 2016; 468(5):607-617.

7. Chakraborty G, Armenia J, Mazzu YZ, Nandakumar S, Stopsack KH, Atiq MO, et al. Significance of BRCA2 and RB1 Co-loss in Aggressive Prostate Cancer Progression. Clinical cancer research : an official journal of the American Association for Cancer Research 2020; 26(8):2047-2064.

8. Flammiger A, Weisbach L, Huland H, Tennstedt P, Simon R, Minner S, et al. High tissue density of FOXP3+ T cells is associated with clinical outcome in prostate cancer. European journal of cancer (Oxford, England : 1990) 2013; 49(6):1273-1279.

9. Bansal D, Reimers MA, Knoche EM, Pachynski RK. Immunotherapy and Immunotherapy Combinations in Metastatic Castration-Resistant Prostate Cancer. Cancers 2021; 13(2).

10. Song B, Park SH, Zhao JC, Fong KW, Li S, Lee Y, et al. Targeting FOXA1-mediated repression of TGF-β signaling suppresses castration-resistant prostate cancer progression. The Journal of clinical investigation 2019; 129(2):569-582.

11. Rui X, Shao S, Wang L, Leng J. Identification of recurrence marker associated with immune infiltration in prostate cancer with radical resection and build prognostic nomogram. BMC cancer 2019; 19(1):1179.

12. Liu B, Xie Y, Wu Z. Identification of Candidate Genes and Pathways in Nonsegmental Vitiligo Using Integrated Bioinformatics Methods. Dermatology (Basel, Switzerland) 2021; 237(3):464-472.

13. Yuan K, Zeng R, Deng P, Zhang A, Liu H, Wang N, et al. Identification and Verification of Immune-Related Genes Prognostic Signature Based on ssGSEA for Adrenocortical Carcinoma (ACC). International journal of general medicine 2022; 15:1471-1483.

14. Ren C, Li M, Du W, Lü J, Zheng Y, Xu H, et al. Comprehensive Bioinformatics Analysis Reveals Hub Genes and Inflammation State of Rheumatoid Arthritis. BioMed research international 2020; 2020:6943103.

15. Zito A, Lualdi M, Granata P, Cocciadiferro D, Novelli A, Alberio T, et al. Gene Set Enrichment Analysis of Interaction Networks Weighted by Node Centrality. Frontiers in genetics 2021; 12:577623.

16. Zhang M, Zhu K, Pu H, Wang Z, Zhao H, Zhang J, et al. An Immune-Related Signature Predicts Survival in Patients With Lung Adenocarcinoma. Frontiers in oncology 2019; 9:1314.

17. Jiang H, Gu J, Du J, Qi X, Qian C, Fei B. A 21‑gene Support Vector Machine classifier and a 10‑gene risk score system constructed for patients with gastric cancer. Molecular medicine reports 2020; 21(1):347-359.

18. Cheng Q, Chen X, Wu H, Du Y. Three hematologic/immune system-specific expressed genes are considered as the potential biomarkers for the diagnosis of early rheumatoid arthritis through bioinformatics analysis. Journal of translational medicine 2021; 19(1):18.

19. Jiao Y, Li Y, Jiang P, Han W, Liu Y. PGM5: a novel diagnostic and prognostic biomarker for liver cancer. PeerJ 2019; 7:e7070.

20. Zhang C, Zheng JH, Lin ZH, Lv HY, Ye ZM, Chen YP, et al. Profiles of immune cell infiltration and immune-related genes in the tumor microenvironment of osteosarcoma. Aging 2020; 12(4):3486-3501.

21. Zhang H, Liu R, Sun L, Guo W, Ji X, Hu X. Comprehensive Analysis of Gene Expression Changes and Validation in Hepatocellular Carcinoma. OncoTargets and therapy 2021; 14:1021-1031.

22. Gao J, Shi L, Gu J, Zhang D, Wang W, Zhu X, et al. Difference of immune cell infiltration between stable and unstable carotid artery atherosclerosis. Journal of cellular and molecular medicine 2021; 25(23):10973-10979.

23. Ma C, Luo H, Cao J, Gao C, Fa X, Wang G. Independent prognostic implications of RRM2 in lung adenocarcinoma. Journal of Cancer 2020; 11(23):7009-7022.

24. Fang Y, Huang S, Han L, Wang S, Xiong B. Comprehensive Analysis of Peritoneal Metastasis Sequencing Data to Identify LINC00924 as a Prognostic Biomarker in Gastric Cancer. Cancer management and research 2021; 13:5599-5611.

25. Tsaur I, Brandt MP, Juengel E, Manceau C, Ploussard G. Immunotherapy in prostate cancer: new horizon of hurdles and hopes. World journal of urology 2021; 39(5):1387-1403.

26. Liu Z, Li H, Pan S. Discovery and Validation of Key Biomarkers Based on Immune Infiltrates in Alzheimer's Disease. Frontiers in genetics 2021; 12:658323.

27. Zeng D, Zhou R, Yu Y, Luo Y, Zhang J, Sun H, et al. Gene expression profiles for a prognostic immunoscore in gastric cancer. The British journal of surgery 2018; 105(10):1338-1348.

28. Xiong Y, Wang K, Zhou H, Peng L, You W, Fu Z. Profiles of immune infiltration in colorectal cancer and their clinical significant: A gene expression-based study. Cancer medicine 2018; 7(9):4496-4508.

29. Ali HR, Chlon L, Pharoah PD, Markowetz F, Caldas C. Patterns of Immune Infiltration in Breast Cancer and Their Clinical Implications: A Gene-Expression-Based Retrospective Study. PLoS medicine 2016; 13(12):e1002194.

30. Zhang W, Chai W, Zhu Z, Li X. Aldehyde oxidase 1 promoted the occurrence and development of colorectal cancer by up-regulation of expression of CD133. International immunopharmacology 2020; 85:106618.

31. Li W, Middha M, Bicak M, Sjoberg DD, Vertosick E, Dahlin A, et al. Genome-wide Scan Identifies Role for AOX1 in Prostate Cancer Survival. European urology 2018; 74(6):710-719.

32. Shi X, Wang J, Dai S, Qin L, Zhou J, Chen Y. Apolipoprotein C1 (APOC1): A Novel Diagnostic and Prognostic Biomarker for Cervical Cancer. OncoTargets and therapy 2020; 13:12881-12891.

33. Su WP, Sun LN, Yang SL, Zhao H, Zeng TY, Wu WZ, et al. Apolipoprotein C1 promotes prostate cancer cell proliferation in vitro. Journal of biochemical and molecular toxicology 2018; 32(7):e22158.

34. Cartoni R, Norsworthy MW, Bei F, Wang C, Li S, Zhang Y, et al. The Mammalian-Specific Protein Armcx1 Regulates Mitochondrial Transport during Axon Regeneration. Neuron 2016; 92(6):1294-1307.

35. Jauhiainen S, Laakkonen JP, Ketola K, Toivanen PI, Nieminen T, Ninchoji T, et al. Axon Guidance-Related Factor FLRT3 Regulates VEGF-Signaling and Endothelial Cell Function. Frontiers in physiology 2019; 10:224.

36. Chakravarthi BV, Pathi SS, Goswami MT, Cieślik M, Zheng H, Nallasivam S, et al. The miR-124-prolyl hydroxylase P4HA1-MMP1 axis plays a critical role in prostate cancer progression. Oncotarget 2014; 5(16):6654-6669.

37. Lv Y, Jin Y, Zhou Y, Jin J, Ma Z, Ren Z. Deep sequencing of transcriptome profiling of GSTM2 knock-down in swine testis cells. Scientific reports 2016; 6:38254.

38. Albawardi A, Livingstone J, Almarzooqi S, Palanisamy N, Houlahan KE, Awwad AAA, et al. Copy Number Profiles of Prostate Cancer in Men of Middle Eastern Ancestry. Cancers 2021; 13(10).

39. Patel PG, Wessel T, Kawashima A, Okello JBA, Jamaspishvili T, Guérard KP, et al. A three-gene DNA methylation biomarker accurately classifies early stage prostate cancer. The Prostate 2019; 79(14):1705-1714.

40. Angulo JC, Lopez JI, Dorado JF, Sanchez-Chapado M, Colas B, Ropero S. A DNA Hypermethylation Profile Independently Predicts Biochemical Recurrence Following Radical Prostatectomy. Urologia internationalis 2016; 97(1):16-25.

41. Ashour N, Angulo JC, Andrés G, Alelú R, González-Corpas A, Toledo MV, et al. A DNA hypermethylation profile reveals new potential biomarkers for prostate cancer diagnosis and prognosis. The Prostate 2014; 74(12):1171-1182.

42. Li S, Wang L, Sun S, Wu Q. Hepsin: a multifunctional transmembrane serine protease in pathobiology. The FEBS journal 2021; 288(18):5252-5264.

43. Ma X, Guo J, Liu K, Chen L, Liu D, Dong S, et al. Identification of a distinct luminal subgroup diagnosing and stratifying early stage prostate cancer by tissue-based single-cell RNA sequencing. Molecular cancer 2020; 19(1):147.

44. Petitprez F, Fossati N, Vano Y, Freschi M, Becht E, Lucianò R, et al. PD-L1 Expression and CD8(+) T-cell Infiltrate are Associated with Clinical Progression in Patients with Node-positive Prostate Cancer. European urology focus 2019; 5(2):192-196.

45. Zhang E, Dai F, Mao Y, He W, Liu F, Ma W, et al. Differences of the immune cell landscape between normal and tumor tissue in human prostate. Clinical & translational oncology : official publication of the Federation of Spanish Oncology Societies and of the National Cancer Institute of Mexico 2020; 22(3):344-350.

46. Malekghasemi S, Majidi J, Baradaran B, Aghebati-Maleki L. Prostate cancer cells modulate the differentiation of THP-1 cells in response to etoposide and TLR agonists treatments. Cell biology international 2020; 44(10):2031-2041.

Table 1 Information for selected GEO datasets

GEO accession	country	Platform	Samples		Category
GEO accession	country	Platform	PCa	Normal	Category
GSE8511	USA	GPL1708	24	16	Train group
GSE14206	Italy	GPL887	53	14	Train group
GSE46602	Denmark	GPL570	36	14	Train group
GSE55945	USA	GPL570	13	8	Train group
GSE69223	Germany	GPL570	15	15	Test group
GSE71016	USA	GPL16699	48	47	Test group

SupplementFigureS1A.tif
Supplement Figure S1A-S1D: Validation of the expression levels of PCa-related diagnostic biomarkers in the test group (all P > 0.05). (S1A) CD38; (S1B) CPLX3; (S1C) ERG; (S1D) MSMB.
SupplementFigureS1B.tif
SupplementFigureS1C.tif
SupplementFigureS1D.tif

Download PDF

Version 1

posted

You are reading this older preprint version

Read the latest preprint version →

Identification and validation of immune-related biomarkers based on machine learning in patients with prostate cancer

Status:

Version 1

Abstract

Figures

Introduction

Materials And Methods

Results

Discussion

Conclusions

Abbreviations

Declarations

References

Tables

Supplementary Files

Status:

Version 1