1 INTRODUCTION

Cancer is considered the most serious and prevalent disease in human beings around the world. The morbidity and mortality rate of cancer is the highest in the world. The 2020 cancer report estimates 19.3 million new cancer cases and 10.0 million cancer-associated deaths.1 Owing to the increasing and ageing world population, the global cancer burden is expected to be 22.2 and 28.4 million new cases in 2030 and 2040, respectively, as depicted by the current trends.1, 2 In recent years, great efforts have been made for cancer prevention, screening, early detection, standardized treatment and regular follow-up, however, the world still bears a large tumour burden due to the unclear pathogenesis of most tumours and the unavailability of potential biomarkers.3 It is crucial to extensively explore the pathogenesis of tumours and effective screening markers can be exploited as therapeutic targets.

Since the first case of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections occurred in December 2019, the number of global coronavirus disease (COVID-19) cases is still steadily increasing. As of 1 November 2021, the global COVID-19 pandemic has led to 246,929,884 confirmed cases and 5,003,404 deaths over 188 countries/regions across the world.4 Clear evidence exists that patients with co-morbidities are more susceptible to the COVID-19, and are more likely to manifest complications and mortality after infection.5, 6 During the epidemic period, due to ageing, decreased immunity, delay in diagnosis, treatment and follow-up, surgery, radiotherapy and chemotherapy and tumour-related multiple co-morbidities, cancer has been identified as an individual risk factor for COVID-19.6-9 A cohort study from China showed that the overall prevalence of COVID-19 in cancer patients was considerably elevated than the overall incidence in the general population (1% vs. 0.29%).10, 11 In addition, various studies have reported that cancer patients are more likely to have serious clinical outcomes after suffering from COVID-19.7, 10, 12-14 This might provide a clue that pays attention to the internal relationship between tumour and COVID-19, which can lead to the preventing as well as controlling of COVID-19 in cancer patients.

Transmembrane Serine Protease 2 (TMPRSS2) is a multifunctional encoding gene and is considered one of the members of the serine protease family. TMPRSS2 contains four domains, that is, protease domain, type II transmembrane domain, receptor class A domain and scavenger receptor cysteine-rich domain.15 Among them, the serine protease domain of the underlined protease cleaves, followed by secreting into the cell culture medium after being autocleavage. Thus, it participates in viruses in host cell processes.15, 16 TMPRSS2 has been reported for its contribution to the process of human influenza viruses, coronaviruses including SARS-CoV, SARS-CoV-2, Middle East respiratory syndrome coronavirus (MERS-CoV) and human coronavirus 229E (HCoV-229E) and entering host cells.17, 18 Currently, modulating the expression or activity of TMPRSS2 is considered a potential intervention against human influenza viruses and coronaviruses including COVID-19.18 On the other hand, multiple studies revealed that the expression of TMPRSS2 was found to be considerably down-regulated in tumour tissues compared to non-tumorous ones, and abnormal expression of TMPRSS2 was closely related to tumour growth, invasion, metastasis and prognosis in various cancers, especially prostate cancer.19, 20 More importantly, the Inhibition of TMPRSS2 expression can reduce prostate or head and neck cancer cell invasion and metastasis, and reduce human lung Calu-3 cells infection with SARS-CoV-2.21, 22 In addition, the TMPRSS2 knockout mouse model in the cancer study showed that TMPRSS2 inhibition is safe and effective for molecular therapy of tumours with few on-target side effects.23 Therefore, a systematic and in-depth investigation of the function of TMPRSS2 in multiple tumours and COVID-19 could pave the way for precision medicine and TMPRSS2-targeted strategies.

Herein, to investigate the potential relationship between tumours and COVID-19, and assess the expression level of TMPRSS2 and its prognostic value in different carcinomas, we systematically studied the expression of TMPRSS2 and its medical consequences in different types of carcinomas while employing multiple recognized online network databases. Furthermore, we examined the co-altered genes with TMPRSS2 for common cancer types and performed functional enrichment analysis. Therefore, the analyses may provide the potential value of TMPRSS2 expression for the survival of patients associated with cancer, and give potential direction to prevent COVID-19 pandemic for specific tumour patients.

2 MATERIALS AND METHODS 2.1 Analyses of Oncomine dataset

In this study, a public web-based microarray database Oncomine (www.oncomine.org) was employed for the analysis of the TMPRSS2 transcription levels in cancerous specimens followed by comparing the obtained results with the healthy specimens (controls).24 The thresholds were restricted in the following manner: fold-change = 1.5; p = 0.001; data type: mRNA.

2.2 Analysis of GEPIA dataset

Gene Expression Profiling Interactive Analysis (GEPIA) (http://gepia.cancer-pku.cn) offers significant interactive and customizable tasks, such as profile plotting, differential expression analysis, correlation analysis, estimating RNA sequencing expression data based on 8587 healthy and 9736 cancer samples in Genotype–Tissue Expression (GTEx) and TCGA projects.25 We used GEPIA to verify the differences in TMPRSS2 gene expression in both healthy and different types of cancer tissues. In addition, profile plotting based on cancer pathological stage or type of cancer, survival rate of patient, similar gene detection, correlation analyses and dimensionality reduction analysis can be carried out via the GEPIA dataset.

2.3 UALCAN database analysis

UALCAN (http: //ualcan. path.uab.edu/index.html) is a user-friendly integrated data-mining platform and is used for the extensive analysis of data obtained from cancer OMICS.26 It is built on PERL-CGI and can also be employed for gene expression analysis, methylation of promoter, prognosis and correlation. In this study, the UALCAN database was employed for analysing the expression pattern and promoter methylation profiling of TMPRSS2 mRNA.

2.4 cBioPortal dataset and muTarget database analysis

TCGA (cancergenome.nih.gov/) is a comprehensive database, which has both sequencing and pathological data of 30 various forms of carcinomas. For cancer genomics, cBioPortal (http://www.cbioportal.org/) is an open-access platform, which can be utilized for the multi-functional visualization of complex cancer genomics, integrative analysis and clinical profiling.27 We employed cBioPortal to evaluate the TMPRSS2 alterations frequency, mRNA expression z-scores (RNA Seq V2 RSEM) and copy number variance via subtypes of each carcinoma from the TCGA PanCanAtlas dataset. To further clarify the relationship between TMPRSS2 gene mutation and protein expression, we used the muTarget database (https://www.mutarget.com/) to analyse the effect of TMPRSS2 gene mutation on protein expression levels in different tumour types. The cut-off p-value was regarded as <0.01.28

2.5 TIMER analysis

TIMER (https://cistrome.shinyapps.io/timer/) database was employed to validate the TMPRSS2 expression level in different form of carcinomas,29 followed by estimating the Spearman's correlation analysis between the TMPRSS2 expression levels and six immune infiltrates, such as CD8+ T cells, CD4+ T cells, B cells, macrophages, dendritic cells (DCs) and neutrophils obtained from four common carcinomas. The expression scatter plots were formed between a pair of user-defined genes in a given cancer type by using a correlation module, followed by revealing the level of gene expression via log2 RSEM.

2.6 PrognoScan database analysis

An online database PrognoScan (http://dna00.bio.kyutech.ac.jp/PrognoScan/) provides a platform to assess effective tumour biomarkers and their significant therapeutic targets.30 PrognoScan was employed for exploring the correlation between the TMPRSS2 expression and survival rate of patients in different carcinomas. According to the obtained results, the threshold was found to be Cox p < 0.05.

2.7 The Kaplan–Meier plotter analysis

A web-based online database Kaplan–Meier plotter (www.kmplot.com) is used to study prognostic implications of genes in various forms of carcinoma. The underlined database comprised data, associated with the rate of survival and gene expression in 7461 samples, obtained from 21 different types of tumour.31 We employed this database for the validation of the TMPRSS2 prognostic value in different types of cancers, with an HR with 95% CI and log-rank p-value.

2.8 Protein–Protein interaction analysis

An online interface GeneMANIA (https://genemania.org/) is a user-friendly data mining platform. It can be used for the generation of genes correlated to a set of input genes, based on protein and genetic interactions, pathways, co-localization, co-expression and protein domain similarity.32 STRING (https://string-db.org/) is associated with protein–protein interactions.33 Herein, we employed both the GeneMANIA and STRING servers to explore the related genes of TMPRSS2.

2.9 Co-expressed and pathway analysis

The R2: Genomics Analysis and Visualization Platform-V-3.2.0 (https://hgserver1.amc.nl/) was employed for the integrative analysis of the positively and negatively co-expressed genes of TMPRSS2 in TCGA dataset of four different forms of carcinomas, that is, colorectal, breast, lung and ovarian, and the cut-off p-value was regarded as <0.01. The co-expressed genes in the four common tumours are obtained through the intersection of the Venn diagram. Then, we used the React me tool (https://reactome.org/) to explore pathways shared by TMPRSS2-correlated genes and subsequently categorized them according to their KEGG pathway.34

3 RESULTS 3.1 The expression level of TMPRSS2in many cancers

Databases including Oncomine and TCGA were employed for the evaluation of TMPRSS2 differential expression patterns in many kinds of cancer by UALCAN and GEPIA. The results obtained from the Oncomine database indicated that relative to normal tissues, the expression level of TMPRSS2 was reduced in many tumour tissues and cancers including breast, bladder, gastric, colorectal, lung, kidney, prostate, ovarian and sarcoma cancer. There is only one dataset study that shows an elevated expression of TMPRSS2 in breast, kidney and liver cancer, respectively, as depicted in Figure 1A. The UALCAN, TIMER and GEPIA were used to further explore the TMPRSS2 expression in various cancers. The results of the UALCAN and TIMER databases indicated that the expression of TMPRSS2 was decreased in many kinds of cancer, that is, colon adenocarcinoma (COAD), head and neck squamous cell carcinoma (HNSC), breast invasive carcinoma (BRCA), kidney renal papillary cell carcinoma (KIRP), oesophageal carcinoma (ESCA), lung adenocarcinoma (LUAD), kidney renal clear cell carcinoma (KIRC), rectum adenocarcinoma (READ), liver hepatocellular carcinoma (LIHC), sarcoma (SAEC), lung squamous cell carcinoma (LUSC), thyroid carcinoma (THCA), skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD) and thymoma (THYM). However, elevated expression of TMPRSS2 was reported in bladder urothelial carcinoma (BLCA), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), cholangiocarcinoma (CHOL), glioblastoma multiforme (GBM), kidney chromophobe (KICH), pancreatic adenocarcinoma (PAAD), prostate adenocarcinoma (PRAD) and uterine corpus endometrial carcinoma (UCEC; Figure 1B,C, Figure S1). The results of the GEPIA database are similar to the results of Oncomine, suggesting that the expression of TMPRSS2 was considerably reduced in COAD, KICH, BRCA, HNSC, KIRP, LUAD, KIRC, LUSC and LIHC tumour tissues, however, TMPRSS2 expression was considerably elevated in the UCEC, as depicted in Figure S2.

Expression pattern of TMPRSS2 mRNA in many cancers by Oncomine, UALCAN and GEPIA databases. (A) The expression pattern of TMPRSS2 mRNA in various cancers was searched in the Oncomine database. The underlined graphic was generated from the Oncomine database, revealing the number of datasets (p < 0.01) mRNA over (red) or under expression (blue) of TMPRSS2 (tumour tissue vs. corresponding normal tissue). The threshold was considered with the underlined parameters: p and fold-change were equal to 0.001 and 1.5, respectively, and the data type was mRNA. (B) The expression of TMPRSS2 mRNA in many cancers was searched in the UALCAN database. Boxes represent the median and the 25th and 75th percentiles; green and red boxes indicate normal and tumour tissues respectively. (C) The expression level of TMPRSS2 mRNA in many cancers was searched in the TIMER database. Boxes indicate the median and the 25th and 75th percentiles; blue and red boxes indicate normal and tumour tissues respectively. Blue and red dashed lines indicate the average value of normal and tumour tissues respectively. *p = 0.05，**p = 0.01, ***p = 0.001

3.2 TMPRSS2 promoter methylation in many cancers

Due to the lower expression of TMPRSS2 in a variety of tumour tissues, we evaluated the gene promoter of TMPRSS2. UALCAN database was used to verify the level of methylation of TMPRSS2 promoter in various cancers. The results of the UALCAN database showed that the methylation levels of TMPRSS2 promoter in BRCA, CESC, ESCA, HNSC, KIRC, KIRP and UCEC were considerably elevated relative to that in normal tissue, as depicted in Figure 2. In contrast, the methylation levels of TMPRSS2 promoter in COAD, PRAD, READ and testicular germ cell tumours (TGCT) were relatively decreased than those in normal tissue. TMPRSS2 promoter methylation level was negatively correlated with gene expression level, it is indicated that TMPRSS2 promoter hypermethylation in various cancers may trigger itself and elevates its level accordingly.

TMPRSS2 promoter methylation in various cancers using the UALCAN database. The methylation levels of TMPRSS2 promoter in BRCA, CESC, ESCA, HNSC, KIRC, KIRP and UCEC were considerably elevated than that in normal tissue. On the contrary, the methylation levels of TMPRSS2 promoter in COAD, PRAD, READ and TGCT were slightly reduced relative to that in normal tissue. Boxes indicate the median and the 25th and 75th percentiles; green and red boxes show normal and tumour tissues respectively. Green and red dashed lines indicate the average value of all normal and tumour tissues respectively. **p = 0.001, ***p = 0.001

3.3 TMPRSS2 genetic variation in various cancers

The gene mutations of TMPRSS2 were explored in 32 common types of the tumour by the TCGA PanCancer Atlas database, mainly including mutation and copy number mutation. As showed in Figure 3, TMPRSS2 was only altered in 371 (3%) of 10,953 queried patients and 371 (3%) of 10,967 queried samples. Total 329 mutations were present within amino acids 1 to 492aa of TMPRSS2. Among them, there are 7 missense mutation sites, 19 truncating sites, 11 inframe mutation sites and 242 fusion mutation sites respectively. Mutation sites were located in a hotspot in SRCR_2 and trypsin domains. Among the 32 datasets, the percentage of TMPRSS2 alteration frequency varied from 0% to 42.7% in various cancers. The highest alteration frequency was found in prostate cancer (42.7%), whereas other types of tumours exhibited very low mutation alteration (<10%) among all of the query cancer samples. In addition, our research found that the gene mutation of TMPRSS2 cannot cause changes in its own expression, but it can lead to the differential expression of multiple genes in melanoma and uterine cancer. The most significant differences in melanoma are NOMO1, PIGT, VKORC1, PDIA5 and DDX11, and the most significant differences in uterine cancer are CTU2, EMC8, TUBG1, CLPP and LONP1. The details are shown in Figure 4. The mutation frequency of the TMPRSS2 gene is not high in most tumour types, suggesting that the mutation of the TMPRSS2 gene itself may have little effect on the TMPRSS2 gene function.

Frequency of mutations, CNAs and expression in many cancers obtained from cBioPortal web. (A) TMPRSS2 was only altered in 371 (3%) of 10,953 queried patients in 371 (3%) of 10,967 queried samples. (B) Total 329 mutations were found within amino acids 1 to 492 of TMPRSS2. Among them, there are 7 missense mutation sites, 19 truncating sites, 11 inframe mutation sites and 242 fusion mutation sites respectively. Mutation sites were found in a hotspot in SRCR_2 and trypsin domains. (C) In 29 cancer studies, the expression of TMPRSS2 mRNA (RNA Seq V2) was generated from the cBioPortal web. The x-axis has been categorized based on cancer type and y-axis indicates the expression level of BMP5 mRNA. The expression frequency revealed fusions (violet), missense mutations (green), no mutations (blue) and truncating (deep blue). (D) Among the 32 datasets, the percentage of TMPRSS2 alteration frequency varied from 0% to 42.7% in many cancers. The highest alteration frequency was found in prostate cancer (42.7%), whereas other types of tumours all exhibited very low mutation alteration (<10%) among all of the query cancer samples. The alteration frequency revealed fusions (violet), mutations (green) and multiple

The effect of TMPRSS2 gene mutation on the expression of the five most significant genes in melanoma and uterine cancer obtained from muTarget database. Histogram of the effect of TMPRSS2 gene mutation on the expression of (A) DDX11 in melanoma, (B) NOMO1 in melanoma, (C) PDIA5 in melanoma, (D) PIGT in melanoma, (E) VKORC1 in melanoma, (F) CLPP in uterine cancer, (G) CTU2 in uterine cancer, (H) EMC8 in uterine cancer, (I) LONP1 in uterine cancer and (J) TUBG1 in uterine cancer. Boxes indicate the median and the 25th and 75th percentiles. Green and red boxes indicate wild and mutant tissues. Green and red dashed lines indicate the average value of wild and mutant tissues respectively

3.4 TMPRSS2 gene and protein partners in various cancers

GeneMANIA web was used for the prediction of the functionally similar genes including TMPRSS2, which accumulates data on co-localization, co-expression, genetic interactions, involved cascades, prediction of physical interactions and shared protein domains, as depicted in Figure 5A. The predicted functionally similar genes of TMPRSS2 were KDM3A, POU2F1, SLC37A1, C1orf116, SLC44A4, RASEF, TRPM4, ACPP, KLK4, KLK3, KLK2, PDE9A, CX3CL1, NGF, EMX2, AR, ALDOB, RBM47, KCNK5 and SLC45A3. At the same time, the protein–protein interaction network of the TMPRSS2 was predicted by the STRING database (Figure 5C). It was found that 11 proteins interacted with TMPRSS2 proteins. The predicted protein partners of TMPRSS2 were FKBP5, AR, NKX3-1, TMPRSS4, ETV1, ERG, SLC45A3, PTEN, ETV4 and FAM3B. Hence, 27 predicted genes and proteins associated with TMPRSS2 might have a role in regulating cancer development (mediated by TMPRSS2) and prognosis. The mutations and copy number alterations (CNAs) were analysed in the 27 predicted–associated genes of TMPRSS2 via the cBioPortal database. There was an alteration in queried genes of 3225 (38%) out of 10,953 queried patients. The highest alteration frequency was observed in lung squamous carcinoma (>30%), while the lowest alteration (<10%) was observed in colorectal cancer among all of the query cancer samples, as depicted in Figure 5B. Enrichment analysis revealed that the underlined genes were commonly enriched in PID AR TF and PID HNF3A pathways, transcriptional misregulation in cancer, regulation of hormone levels, signalling by nuclear receptors, regulation of growth and positive regulation of hydrolase activity, as depicted in Figure 5D.

Interaction and co-occurrence functional protein partners of TMPRSS2. (A) Expected functional partners of TMPRSS2 generated from GeneMANIA. Circles displayed depicting nodes. Predicted functional partners are indicated post observing co-localization, co-expression, genetic and physical interactions, cascades and shared protein domains (predicted). (B) Predicted functional partners of TMPRSS2 generated from STRING. Circles revealed indicating nodes. Predicted functional partners are indicating post considering physical and functional associations. (C) The alteration frequency of 27 partners gene signature using cBioPortal web. The highest alteration frequency was observed in lung squamous carcinoma (>30%). Green depicts the mutations in alteration frequency, while violet, red, deep blue and grey colours depict fusions, amplifications, deep deletions and multiple alterations respectively. (D) The enrichment analysis of 27 partners gene signature. The obtained results revealed that the underlined genes are largely enriched in PID AR TF and PID HNF3A pathways, transcriptional misregulation in cancer, regulation of hormone levels, signalling by nuclear receptors, regulation of growth and positive regulation of hydrolase activity (p < 0.05)

3.5 Prognostic analysis of TMPRSS2 in many cancers

We explored the prognostic value of TMPRSS2 mRNA expression in many types of cancers by PrognoScan and Kaplan–Meier plotter databases. The PrognoScan database results revealed that the TMPRSS2 expression level was considerably linked with the prognosis of brain, blood, colorectal, breast, ovarian, lung and soft tissue cancer. In breast cancer, GSE9893, GSE7390, GSE12276 and GSE6532 datasets revealed that the patients with up-regulated TMPRSS2 expression had significantly poor overall survival (OS), relapse-free survival (RFS) and distant metastasis-free survival (DMFS) compared to patients with lower TMPRSS2 expression (Figure 6A and Table 1). The opposing results were found in colorectal cancer datasets. The GSE17536, GSE14333 and GSE17536 datasets revealed that the patient's group with an elevated level of TMPRSS2 mRNA expression had significantly better OS and disease-free survival (DFS) than the low expression group (Figure 6B and Table 1). Analysis of GSE31210 and GSE13213 datasets of PrognoScan revealed considerably poor OS and RFS of lung cancer patients in the down-regulated TMPRSS2 mRNA expression group relative to their elevated expression counterparts (Figure 6C and Table 1). High OS and DFS ratio were shown in the low TMPRSS2 expression group of ovarian cancer relative to the elevated expression group, according to the DUKE-OC, GSE9891 and GSE26712 (205102_at) datasets, while one alteration revealed by dataset GSE26712 (211689_s_at) reversed the correlation of reduced expression with poor DFS of ovarian cancer patients, as depicted in Figure 6D and Table 1.

Association between the expression level of TMPRSS2 and prognosis in four common kinds of cancers, that is, lung, breast, ovarian and colorectal cancer by PrognoScan database. The survival curve comparing patients with an elevated (red) and reduced (blue) expression in the breast (A), colorectal (B), lung (C) and ovarian cancer (D) that were retrieved from the PrognoScan database. Survival curve analysis was carried out by threshold Cox p < 0.05. The dotted lines indicate maximum and minimum values of the survival average

TABLE 1. Prognosis analysis of TMPRSS2 expression in various cancer patients (PrognoScan database) Dataset Cancer type Endpoint Cohort Contributor Probe ID N Cox p Ln (HR) HR (95% CI) GSE31210 Lung cancer RFS NCCRI Okayama 1570433_at 204 8.80481E−07 −0.525474 0.59 (0.48–0.73) GSE31210 Lung cancer OS NCCRI Okayama 1570433_at 204 0.000115681 −0.551071 0.58 (0.44–0.76) GSE31210 Lung cancer RFS NCCRI Okayama 205102_at 204 0.000358629 −0.527135 0.59 (0.44–0.79) DUKE-OC Ovarian cancer OS Duke Bild 205102_at 133 0.00473586 1.68887 5.41 (1.68–17.48) GSE31210 Lung cancer OS NCCRI Okayama 205102_at 204 0.00581929 −0.520625 0.59 (0.41–0.86) GSE13213 Lung cancer OS Nagoya Tomida A_23_P29067 117 0.00614575 −0.3926 0.68 (0.51–0.89) GSE31210 Lung cancer RFS NCCRI Okayama 226553_at 204 0.00700726 −0.267109 0.77 (0.63–0.93) GSE9893 Breast cancer OS Montpellier, Bordeaux, Turin Chanrion 13277 155 0.0108505 0.388917 1.48 (1.09–1.99) GSE17536 Colorectal cancer OS MCC Smith 205102_at 177 0.0109941 −0.945643 0.39 (0.19–0.81) GSE7390 Breast cancer RFS Uppsala, Oxford, Stockholm, IGR, GUYT, CRH Desmedt 205102_at 198 0.0111074 0.255479 1.29 (1.06–1.57) GSE9891 Ovarian cancer OS AOCS, RBH, WH, NKI-AVL Tothill 205102_at 278 0.0123117 −1.00512 0.37 (0.17–0.80) GSE14333 Colorectal cancer DFS Melbourne Jorissen 226553_at 226 0.0131981 −0.240839 0.79 (0.65–0.95) Jacob−00182-UM Lung cancer OS UM Shedden 211689_s_at 178 0.0143181 −0.262144 0.77 (0.62–0.95) GSE30929 Soft tissue cancer DRFS MSKCC (1993–2008) Gobble 205102_at 140 0.0166028 −2.50239 0.08 (0.01–0.63) GSE6532-GPL570 Breast cancer RFS GUYT Loi 1570433_at 87 0.0189748 2.63604 13.96 (1.54–126.26) GSE6532-GPL570 Breast cancer DMFS GUYT Loi 1570433_at 87 0.0189748 2.63604 13.96 (1.54–126.26) GSE26712 Ovarian cancer DFS MSKCC Bonome 211689_s_at 185 0.0255409 0.573233 1.77 (1.07–2.93) GSE4412-GPL96 Brain cancer OS UCLA Freije 211689_s_at 74 0.0271724 0.46614 1.59 (1.05–2.41) GSE14333 Colorectal cancer DFS Melbourne Jorissen 1570433_at 226 0.0276288 −0.274568 0.76 (0.60–0.97) GSE17536 Colorectal cancer DSS MCC Smith 205102_at 177 0.0300716 −0.926539 0.40 (0.17–0.91) GSE2658 Blood cancer DSS Arkansas Zhan 211689_s_at 559 0.0317345 −0.416225 0.66 (0.45–0.96) GSE4412-GPL97 Brain cancer OS UCLA Freije 226553_at 74 0.0361025 1.22568 3.41 (1.08–10.72) Jacob−00182-MSK Lung cancer OS MSK Shedden 205102_at 104 0.0362315 −1.0361 0.35 (0.13–0.94) GSE26712 Ovarian cancer