Development and Validation of An Immune-Related Gene Pair Signature in Skin Cutaneous Melanoma

doi:10.21203/rs.3.rs-51753/v1

PDF

Research article

Development and Validation of An Immune-Related Gene Pair Signature in Skin Cutaneous Melanoma

https://doi.org/10.21203/rs.3.rs-51753/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 01 Dec, 2020

Read the published version in Clinical, Cosmetic and Investigational Dermatology →

Version 1

posted 20 Aug, 2020

You are reading this latest preprint version

Skin cutaneous melanoma (SKCM) is a common skin malignancy worldwide, and its metastasis and mortality rates are high. The molecular characteristics exhibited by tumor-immune interactions have drawn the attention from researchers. Therefore, increased knowledge and new strategies to identify effective immune-related biomarkers may improve the clinical management of SKCM by providing more accurate prognostic information. In this study, we established a prognostic immune-related gene pair (IRGP) signature for predicting the survival of SKCM patients. The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases provided gene expression profiles together with clinical information, and the samples were randomly divided into three groups including the training, testing, and validation datasets. The regression model of least absolute shrinkage and selection operator (LASSO) helped to identify a 13-IRGP signature with a significant relation to the survival of SKCM patients. The training, TCGA, and independent sets have an average value of area under the curve of 0.79, 0.76, and 0.82, respectively. In addition, this 13-IRGP signature can noticeably divide SKCM patients into high-risk group and low-risk group with significantly different prognoses. Many biological activities such as gene family were enriched among the genes in our IRGP signature. While analyzing the risk signature and clinical characteristics, there was a large difference in the risk score between T stage and tumor stage grouping. Finally, we constructed a nomogram and forest plots of the risk score and clinical features. In summary, we developed a robust 13-IRGP prognostic signature in SKCM, which can identify and provide new insights into immunological biomarkers.

Allergy & Immune Disorders

Immunology

Skin cutaneous melanoma

TCGA

immune-related gene pair

signature

bioinformatic

Skin cutaneous melanoma (SKCM) acts as a common skin malignancy around the world [1]. It was estimated that 287,723 new cases and 60,712 cancer-related deaths occurred in 2018 [2]. Thus, SKCM is a complex malignant tumor with high metastasis and mortality rates [3]. Significant progress has been made in understanding SKCM biology, genetics and treatment methods; however, the prognosis is still poor due to the high probability of invasion and metastasis. The diagnosis of melanoma may indicate more aggressive treatment, lifelong monitoring and a worse prognosis [4]. Therefore, increased knowledge and new strategies to identify effective biomarkers may improve the clinical management of SKCM by providing more accurate prognostic information.

Recently, the number of approved cancer immunotherapies has become unprecedented, including immune checkpoint inhibitor monoclonal antibodies and adoptive cell therapy with chimeric antigen receptor (CAR)-T cell therapy [5]. Cancer immunotherapy has become one of the most exciting and rapidly expanding fields. It is now considered to be one of the pillars of treatment alongside surgery, radiotherapy and chemotherapy [6]. In SKCM, most patients remain unresponsive to checkpoint inhibitors. Therefore, trials of different combinations of immunotherapies are being evaluated [7]. Surgical treatment is still the most suitable treatment for metastatic SKCM currently, but targeted molecular therapy and immunotherapy can effectively assist in the development of melanoma therapies [8]. At the same time, immunotherapy has shifted from therapy based on cytokine to the blockade of immune checkpoints of cytotoxic T-lymphocyte-associated antigen-4 (CTLA-4) and programmed cell-death protein 1 (PD-1) mediated by antibody [9]. Targeted therapy with inhibitors of BRAF and MEK is currently an important cornerstone for the treatment of metastatic SKCM [10]. Nevertheless, the molecular characteristics of tumor-immune interactions remain to be extensively studied in future studies.

With the development of genomics and transcriptomics, studies about immune-related genes (IRGs) or prognostic signatures in cancers are increasingly being reported. For example, Li et al. [11] used transcriptomic data and clinicopathological information from the TCGA database to identify an immune-related signature specific to uveal melanoma prognosis. In another study of breast cancer [12], many tumor immune-related genes (IRGs) were identified for predicting patients’ prognosis. In SKCM, a novel IRG prognostic biomarker associated with the tumor microenvironment based on the TCGA and GEO databases was established [13]. In addition, some other biological markers have been identified and discovered as prognostic biomarkers for SKCM. For example, alternative splicing events could be significant indicators for SKCM prognosis [14]. Another study found a five-miRNA (miR-25, miR-204, miR-211, miR-510, miR-513c) signature as a potential significant biomarker for SKCM prognosis based on the GEO and TCGA databases [15].

Recently, more and more studies have paid attention to immune-related gene pairs in human cancers. To be specific, a signature of 33 immune-related gene pairs that can predict clinical outcomes of hepatocellular carcinoma was established by Sun et al. [16]. In this study, they used the gene expression levels of IRGs from the TCGA database to construct IRGPs. Based on LASSO penalized Cox proportional hazards regression, the frequency of the 33-IRGP model was the highest. For colorectal cancer, Wu et al. [17] also identified a 19 IRGP signature with 36 unique genes. The above studies helped us to better understand how to identify cancer patients with a high mortality. However, there is no report about IRGPs in SKCM.

In the present study, based on the TCGA and GEO databases, we constructed and validated a 13-IRGP prognostic signature for SKCM. This study provides new insights for prognosis prediction in SKCM patients.

2.1 Sources of immune-related genes

In this study, the gene list of IRGs was downloaded from the InnateDB database (https://www.innatedb.com/) [18], which recorded the endogenous IRGs of multiple species under the support of the literature with manual correction. Here, the endogenous IRGs of humans were used, and a total of 1,039 IRGs were identified (the genes with duplicate names were removed). The list of genes is provided in Supplementary Table 1.

2.2 Data source and download

We used the TCGA GDC API to download the latest RNA-seq mRNA expression data together with clinical follow-up information of SKCM (Supplementary Table 2) from the TCGA database (https://portal.gdc.cancer.gov/) [19] until December 2018. This dataset included 472 SKCM patient samples, of which 471 were tumor tissues and one was a normal sample. Three microarray datasets (GSE54467 [20], GSE22155 [21] and GSE65904 [22]) were also downloaded from the GEO database (http://www.ncbi.nlm.nih.gov/geo) [23] for validation of this signature. In brief, we used the gene expression as MINiML files from NCBI. The GSE54467 dataset contained 79 SKCM samples with clinical features (Supplementary Table 3), the GSE22155 dataset contained 79 samples with clinical features (all stage III and IV samples, Supplementary Table 4), and the GSE65904 dataset contained 214 samples (Supplementary Table 5).

2.3 Data preprocessing

The RNA-seq data of 471 samples from the TCGA database were preprocessed in the following steps: 1) samples with no clinical information or with an overall survival (OS) of 0 days were removed; 2) data from normal samples were removed; 3) genes with an FPKM value of 0 in over half of all samples were removed; and 4) only IRGs’ expression profiles were retained. For the GEO datasets, the samples were preprocessed as follows: 1) data from normal tissues were removed and only the primary tumor data were kept; 2) the overall survival (OS) time of was converted from years or months into days; 3) the microarray probes were mapped to the human gene symbol using the bioconductor R package; and 4) only the expression profiles of IRGs were retained. After data preprocessing, the sample sizes of the GSE54467 and GSE22155 datasets were only 70 and 43, respectively. We combined the above two datasets as an independent validation set. The clinical information of the three datasets is shown in Table 1.

Table 1

The clinical information for three datasets
TCGA		Independent validation dataset	GSE65904
Event
Alive	229	35		106
Dead	213	78		100
T stage			Stage
T0	23		General		21
T1	41		In-transit		15
T2	76		Local	11
T3	89		Primary	14
T4	143		Regional	138
TX	74		X	7
N stage
N0	222
N1	73
N2	48
N3	54
NX	49
M stage
M0	401
M1	21
MX	24
Tumor stage
I	76	24
II	133	25
III	168	20
IV	20	43
X	49	1
Gender
Female	168			85
Male	278			121
PrimarySite
Extremities	188
Extremities&Trunk	4
Head&Neck	32
Trunk	162
Unknown	60
Age
0 ~ 40	60	10		20
40 ~ 50	75	15		18
50 ~ 60	102	33		40
60 ~ 70	88	31		54
70 ~ 100	121	24		74

2.4 Construction of IRGPs

The method to build the IRGPs was as described by Wu et al. [17]. In brief, we first constructed (1,039*1,038)/2 IRGPs on the basis of 1,039 IRGs. IRGP values were calculated following rules below. If the first IRG in a particular IRGP held smaller gene expression level compared with the second IRG, their IRGP score = 1; otherwise, their IRGP score = 0. Next, we filtered IRGPs by computing the IRGPs for each dataset. If the IRGP was a unique value of 0 or 1 for all samples in that dataset, we removed the IRGPs (possibly due to platform factors).

2.5 Establishment and validation of the prognostic signature based on IRGPs

The TCGA dataset was used as the training set with a total of 220,718 IRGPs with 446 samples used for subsequent modeling and analysis. The IRGPs of the other two datasets were used as the validation sets to verify the signature. First, we divided 446 SKCM samples into two sets, the training set and the testing set. For avoiding the random allocation bias that could impact the stability exhibited by subsequent signature, all samples were resampled 100 times into random groups. Here, the group sampling was based on the training: testing set with a ratio of 0.5: 0.5. We chose the most suitable training set and testing set considering two conditions: 1) the distribution of age, follow-up time, tumor stage, as well as death rate of patients in two groups were similar; and 2) the sample size of the two groups was similar after clustering by the gene expression profile. The final training set data are shown in Supplementary Table 6 and contained a total of 222 SKCM samples. The testing set data are shown in Supplementary Table 7 and contained a total of 224 samples.

The prognostic IRGPs (p-value < 0.05) in the training dataset were then selected via a univariate Cox proportional hazard regression model with a long-rank test. To further narrow the IRGP number while maintaining high accuracy, we performed LASSO regression for filtering the IRGPs in the risk model by using the glmnet R package [24]. Here, we used the most stable gene pair for constructing the prognosis signature for SKCM. By calculating the risk score of each sample, we used the median risk score to divide samples in the training set into two groups, high-risk group (Risk-H) and low-risk group (Risk-L). The receiver operating characteristic (ROC) curves were plotted by virtue of the signature performance. The Kaplan-Meier curves of these samples were plotted with the KMsurv R package.

2.6 Functional enrichment analysis

First, we performed gene family enrichment analysis based on the IRGs in our signature with Fisher’s exact test. The clusterProfiler R package [25] was then used for analyzing to the IRGPs for molecular function (MF), cellular component (CC), and biological process (BP) enrichment through studying the Gene Ontology (GO) terms, as well as the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment.

2.7 The nomogram and forest plots of the risk score and clinical features

The nomogram can visually and effectively show the results of the risk model and is convenient to use in the prediction of survival outcomes. It uses the length of the line to indicate the influence degree of different variables on the result as well as the influence of different values of variables on the result. Here, the nomogram was generated with the rms R package. Forest plots can simply and intuitively display the statistical summary results of different research factors. The general form of the forest plot is in a plane rectangular coordinate system, with an invalid line perpendicular to the X axis (usually the coordinate X = 1 or 0) as the center. Several line segments parallel to the X axis represent the magnitude of each study effect and its 95% confidence interval.

2.8 Statistical analysis

We carried out all the statistical analyses in the R language (version 3.5.1) environment. The log-rank test assisted in evaluating the difference in survival between two groups. In the analysis of the risk signature and clinical characteristics, the Kruskal-Wallis test was used to determine differences between multiple groups. The survcomp R package helped to calculate the C-index. A p-value less than 0.05 means the difference is statistically significant.

3.1 Consistency assessment of datasets on different platforms

In this study, to evaluate the differences between RNA-seq data from the TCGA database and gene expression microarray data from the GEO database, we used IRG and IRGP data to cluster the correlation between SKCM samples. The results showed that IRG data can clearly separate the data from the two platforms (Supplementary Fig. 1), while IRGP data can also separate the data from the two platforms (Supplementary Fig. 2), but the difference between two platforms was significantly reduced, and the correlation was closer than that of IRGs (Supplementary Fig. 3). This suggested that after IRGP construction from three datasets based on different platforms, the differences were eliminated.

3.2 Construction of the IRGP signature in SKCM

We extracted the IRGPs shared by the three datasets. The TCGA dataset was used as the training set with a total of 220,718 IRGPs with 446 samples used for subsequent modeling and analysis. First, a total of 446 SKCM patients from the TCGA database were included in our study, which were split into a training dataset (n = 222) and a testing dataset (n = 224) in a random manner. The detailed clinical information is shown in Table 2. Based on the univariate Cox proportional hazards regression model, a total of 1,614 significantly prognostic IRGPs were selected (Supplementary Table 8). As shown in Fig. 1, we determined the relationships between the p-values of 1,614 genes and the hazard ratios (HRs). From the results, the HRs corresponding to IRGPs with a significant p-value usually deviate from 1.

Table 2

The clinical information of the training and testing datasets
Clinical features	Training set	Testing set
Event
Alive	115	114
Dead	107	110
T stage
T0	13	10
T1	21	20
T2	40	36
T3	43	46
T4	72	71
TX	33	41
N stage
N0	110	112
N1	35	38
N2	23	25
N3	31	23
NX	23	26
M stage
M0	201	200
M1	11	10
MX	10	14
Tumor stage
I	42	34
II	60	73
III	86	82
IV	11	9
X	23	26
Gender
Female	84	84
Male	138	140
PrimarySite
Extremities	84	104
Extremities&Trunk	3	1
Head&Neck	19	13
Trunk	87	75
Unknown	29	31
Age
0 ~ 40	23	37
40 ~ 50	44	31
50 ~ 60	49	53
60 ~ 70	46	42
70 ~ 100	60	61

Next, aiming at further filtering above IRGPs for reducing the number in the risk model, we constructed a prognostic gene signature with the help of the LASSO regression model on the TCGA training dataset. We finally obtained 13 IRGPs to build this signature (Supplementary Table 9). Risk score = SIRPA_vs_IFI27*1.02 + IDO1_vs_APOBEC3A*-0.99 + ING4_vs_CNOT4*-1.49 + NLRP1_vs_EGFR*-0.67 + NFKBIE_vs_ITGB3*0.71 + IFIT5_vs_CASP6*-0.53 + RGMB_vs_ACHE*0.69 + CXCL14_vs_RSAD2*0.46 + CD36_vs_TLR8*0.55 + IFIT2_vs_FANCC*-0.25 + NFKBIE_vs_CXCL10*0.12 + TRIM45_vs_CD1D*0.12 + KCNJ8_vs_CTLA4*-0.02. Considering that the overall survival distribution of the samples was approximately 2 years (Supplementary Fig. 4), we calculated the prediction effect on all datasets exhibited by the model for 1, 5, and 7 years. The average AUC of the training set was 0.79, of the TCGA dataset was 0.756, of the independent dataset was 0.818, and of the GSE65904 independent testing dataset was 0.675 (Fig. 2A-D, left). The OS distribution of the samples in the Risk-H and Risk-L groups was also assessed (Fig. 2A-D, middle). The Risk-H and Risk-L groups were not significantly different in terms of size in the 0- and 1-year periods. After the 5th year, the Risk-H group was gradually smaller than the Risk-L group, and the change became increasingly significant as the OS was extended (Fig. 2A-D, right).

3.3 IRGP signature evaluation and validation for survival prediction

The K-M survival curves of the Risk-H and Risk-L groups were predicted via the risk model based on the 13 IRGPs in the training set, TCGA dataset and two GEO datasets (Fig. 3). As shown in Fig. 3A, we found significant survival differences between the Risk-H and Risk-L groups (with log-rank test p-value less than 0.0001). Similar results were also validated in the TCGA, independent, and GSE65904 datasets (with log-rank test p-value less than 0.01, Fig. 3B-D). Moreover, in the testing set of TCGA, Risk-H and Risk-L samples also had significant differences in prognosis (Supplementary Fig. 5).

3.4 Biological functions related to the IRGP signature

Based on the above IRGP signature, we next performed functional annotations of the IRGPs. As shown in Table 3, four gene families, namely, the repulsive guidance molecule family, Erb-b2 receptor tyrosine kinases, tetratricopeptide repeat domain containing, and integrin beta subunits, exhibited a significant enrichment (Fisher’s exact test p-value < 0.01). GO and KEGG pathway enrichment analyses were used for further confirming the biological activity of the 13 IRGPs (Fig. 4, Supplementary Table 10). The enriched biological pathways are shown in Supplementary Table 11.

Table 3

The gene family enrichment analysis of IRGPs signature
Gene family	Genes	p-value	p-adj
Repulsive guidance molecule family	RGMB	0.00430607	0.107651741
Erb-b2 receptor tyrosine kinases	EGFR	0.005379802	0.134495062
Tetratricopeptide repeat domain containing	IFIT5/IFIT2	0.007020208	0.175505211
Integrin beta subunits	ITGB3	0.010731814	0.268295359
Toll like receptors	TLR8	0.011798894	0.29497235
V-set domain containing	SIRPA/CTLA4	0.01356257	0.339064249
Apolipoprotein B mRNA editing enzyme catalytic subunits	APOBEC3A	0.013929739	0.348243483
Caspases	CASP6	0.014993507	0.374837679
Potassium voltage-gated channel subfamily J	KCNJ8	0.018178205	0.454455119
FA complementation groups	FANCC	0.024517971	0.612949276
Pyrin domain containing	NLRP1	0.027673098	0.691827462
Scavenger receptors	CD36	0.029771072	0.74427679
Blood group antigens	ACHE	0.044335676	1
Chemokine ligands	CXCL14	0.048458289	1
PHD finger proteins	ING4	0.093676864	1
Tripartite motif containing	TRIM45	0.098571875	1
RNA binding motif containing	CNOT4	0.206994163	1
Endogenous ligands	CXCL10	0.22069409	1
Ankyrin repeat domain containing	NFKBIE	0.231657695	1
CD molecules	CD1D	0.349347048	1
unknown	IFI27:RSAD2:IDO1:	NA	NA

3.5 Risk signature and clinical characteristics analysis

Furthermore, we analyzed the relationships between clinical features such as T stage, N stage, M stage, age and tumor stage and the risk score. The results are shown in Fig. 5 (Supplementary Fig. 6 for M stage) and revealed that there was a large difference in the risk score between the T stage and tumor stage groups (p-value < 0.05), indicating that our risk model was closely associated with the above clinical characteristics in SKCM. Because of the significant correlation between the clinical characteristics of T stage and tumor stage and the prognosis of SKCM, there were significant prognostic differences between the samples of the above clinical characteristic groupings (Supplementary Fig. 7). TNM, age, and different pathological stage prognostic risk models were constructed. Then, we compared these models with our 13-IRGP risk signature. The C-index results of the models are shown in Fig. 6. From the results, our 13-IRGP signature had the highest C-index, which demonstrated that the model exhibited a better performance compared with the other models. In addition, we constructed a T stage + tumor stage + risk score multifactor prognostic signature. The C-index of this model reached 0.76, indicating that the performance of the multifactor model was improved.

3.6 The nomogram and forest plots of the risk score and clinical features

Finally, we constructed a nomogram with clinical features including T stage, N stage, M stage, age, and risk score (Fig. 7). Here, two nomograms, TNM stage + age + risk score and tumor stage + age + risk score, were constructed. From the results, the characteristics of the risk score the most significantly affected SKCM patient survival prediction, demonstrating the better prediction effect of the 13-IRGP risk model. Then, we described the clinical features, including T stage, N stage, M stage, and age, and the risk score by displaying forest plots (Fig. 8). The TNM stage and tumor stage were used to construct forest plots with the risk score. From the results, the HR of the risk score was the largest in both models, approximately 1.9, with a p-value of < 0.001 (Supplementary Table 12).

In this study, a robust 13-IRGP signature was employed to estimate prognosis in SKCM, which can identify and provide new insights into immunological biomarkers.

Recently, with the development of genomics and transcriptomics, studies about IRGs or prognostic signatures in cancers, including melanoma, have been increasingly reported. In one study of melanoma, Huang et al. [13] first found that 63 IRGs were associated with the OS of patients. Based on Cox regression and LASSO analysis, they identified an IRG signature including 8 IRGs, namely PSME1, CDC42, CMTM6, HLADQB1, HLA-C, CXCR6, CD8B, and TNFSF13, which could better predict the prognosis of patient compared with existing recorded data. For uveal melanoma, an immune-related signature with two genes of MANEAL and SLC44A3, was established [11]. In addition, the researchers revealed a moderate association between the immune checkpoints that contained PD-1, CTLA-4, IDO, and TIGIT and such signature. Li et al. [26] used RNA-seq data from 527 head and neck cancer patients from the TCGA database to establish a prognostic prediction model using multivariable Cox regression analysis. Finally, they identified a 10-IRG signature including SEMA3G, GNRH1, ZAP70, PLAU, SFTPA2, CCL26, DKK1, GAST, PDGFA and STC1. Other studies on IRG signatures in human cancers include breast cancer [12], colorectal cancer [27], hepatocellular cancer [28], lung cancer [29], and gliomas [30]. Therefore, based on TCGA and microarray methods, IRG signatures for predicting the prognosis of patients have increasingly become the focus of research.

We constructed a 13-IRGP prognostic signature for SKCM in this study. There have been no reports about IRGP signatures for SKCM until now. However, in colorectal cancer [17], Wu et al. used six public cohorts, including a training cohort (n = 565) together with five independent validation cohorts (n = 572, 290, 90 177 and 68). They established a 19-IRGP signature that contained 36 unique genes with an obvious relation to patients’ survival. When combined sex, stage, and other clinical factors, the IRGP signature and clinical factor combination showed a higher prognostic accuracy than the individual IRGP signature. In another study of hepatocellular carcinoma [16], researchers also identified a signature of 33 IRGPs that can predict clinical outcomes based on TCGA and GEO data. Compared with traditional studies about gene signatures, since pairwise comparison helped to generate our IRGP, the calculation of risk score was decided by the gene expression of the same patient. The obtained prognostic signature is capable of overcoming the batch effect of various platforms on the one hand and shows no need for scaling and normalization of data on the other hand.

In our study, LASSO penalized Cox regression was adopted for constructing a 13-IRGP prognostic signature of SKCM. We also validated this signature in testing and validation datasets. The training set, the TCGA dataset, and the independent dataset held an average AUC of 0.79, 0.756, and 0.818, respectively. The significant survival differences between the Risk-H and Risk-L groups were determined by K-M survival plots. Similar results were also validated in the TCGA, independent, and GSE65904 datasets. The above results proved the accuracy and robustness of our prognostic signature in SKCM patients.

Our 13-IRGP signature consists of a total of 25 IRGs in SKCM. Here, IFIT5_vs_CASP6 and NFKBIE_vs_CXCL10 were two IRGPs with the largest coefficients. IFIT5, as one of the human IFIT gene families, can affect different biological activities of cancers, such as antiviral immune response, host innate immunity, replication, PAMP recognition, as well as double-stranded RNA signaling [31]. In bladder cancer, it was shown to be positively correlated with pathological characteristics and predicts poor prognosis in patients [32]. Additionally, this gene is capable of inducing the epithelial-mesenchymal transition (EMT) as well as promoting the migration and invasion of cells [32]. CXCL10, as a "key driver chemokine" as well as an effective target for treating autoimmune diseases [33], has been a target for novel cancer therapy in immune activation functions [34]. In melanoma, CXCL10 was reported as a key candidate gene by integrated bioinformatics analysis [35]. The above research results indicated that the IRGs in our signature may be involved in the occurrence and development of SKCM.

However, the study exhibits some limitations. First, adopting retrospective analysis, the study requires a prospective cohort for validating study results and this signature. Second, the molecular functions of the genes in the IRGP signature are still unknown. Finally, RNA-seq and microarray data were used to construct the signature in the study, thus it is necessary to verify the model with IHC or western blotting in a large sample size of clinical tissues.

To sum up, a 13-IRGP prognostic signature is developed here, which will serve as a suitable predictive method for identifying SKCM patients who may be better treated with immunotherapy.

Authors’ contributions

Ran Xie: Conception and design of the research, Acquisition of data, and Drafting the manuscript.

Suwei Dong: Conception and design of the research, Acquisition of data, and Drafting the manuscript. Acquisition of data.

Jie Jiang: Acquisition of data.

Conghui Yang: Analysis and interpretation of data.

Lanjiang Li: Statistical analysis and Revision of manuscript for important intellectual content.

Sheng Zhao: Revision of manuscript for important intellectual content.

Yunlei Li: Revision of manuscript for important intellectual content.

Chun Wang: Conception and design of the research, Acquisition of data, and Revision of manuscript for important intellectual content.

Shujuan Li: Analysis and interpretation of data.

Yanbin Xiao: Analysis and interpretation of data and Revision of manuscript for important intellectual content.

Long Chen: Analysis and interpretation of data and Revision of manuscript for important intellectual content.

Competing interests

The authors declare no conflicts of interest.

Acknowledgements

We greatly thank the Department of PET/CT Center, Yunnan Cancer Hospital, The Third Affiliated Hospital of Kunming Medical University. We also gratefully thank American Journal Experts (https://www.aje.cn/) for editing the present paper (7876-83B2-73C2-1F71-2579).

Funding

This study was supported by the Project funded by China Postdoctoral Science Foundation (No.2019M653501)，the National Natural Science Foundation of China (No. 81960496，81760495)，the Funding Project of Oriented Postdoctoral Training in Yunnan Province，the 100 Young and Middle-aged Academic and Technical Backbone Incubation Projects of Kunming Medical University，Reserve candidates for Kunming's young and middle-aged academic and technical leaders (17th)，Special funding for the cultivation of high-level health technical personnel in Yunnan Province (H-2018006)，Scientific Research Fund of Yunnan Provincial Department of Education(2019J1286).

Availability of data and materials

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Lai V, Cranwell W, Sinclair R. Epidemiology of skin cancer in the mature patient. Clin Dermatol. 2018; 36: 167-76.
Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018; 68: 394-424.
Rastrelli M, Tropea S, Rossi CR, Alaibac M. Melanoma: epidemiology, risk factors, pathogenesis, diagnosis and classification. In Vivo. 2014; 28: 1005-11.
Bsirini C, Smoller BR. Histologic mimics of malignant melanoma. Singapore Med J. 2018; 59: 602-7.
Koo SL, Wang WW, Toh HC. Cancer Immunotherapy - The Target is Precisely on The Cancer and Also Not. Ann Acad Med Singapore. 2018; 47: 381-7.
Bergman PJ. Cancer Immunotherapies. Vet Clin North Am Small Anim Pract. 2019; 49: 881-902.
Simeone E, Grimaldi AM, Festino L, Trojaniello C, Vitale MG, Vanella V, Palla M, Ascierto PA. Immunotherapy in metastatic melanoma: a novel scenario of new toxicities and their management. Melanoma Manag. 2019; 6: MMT30.
Pavri SN, Clune J, Ariyan S, Narayan D. Malignant Melanoma: Beyond the Basics. Plast Reconstr Surg. 2016; 138: 330e-40e.
Luke JJ, Flaherty KT, Ribas A, Long GV. Targeted agents and immunotherapies: optimizing outcomes in melanoma. Nat Rev Clin Oncol. 2017; 14: 463-82.
Albittar AA, Alhalabi O, Glitza Oliva IC. Immunotherapy for Melanoma. Adv Exp Med Biol. 2020; 1244: 51-68.
Li YZ, Huang Y, Deng XY, Tu CS. Identification of an immune-related signature for the prognosis of uveal melanoma. Int J Ophthalmol. 2020; 13: 458-65.
Yang H, Zhao K, Kang H, Wang M, Wu A. Exploring immune-related genes with prognostic value in microenvironment of breast cancer from TCGA database. Medicine (Baltimore). 2020; 99: e19561.
Huang R, Mao M, Lu Y, Yu Q, Liao L. A novel immune-related genes prognosis biomarker for melanoma: associated with tumor microenvironment. Aging (Albany NY). 2020; 12.
Xue D, Cheng P, Jiang J, Ren Y, Wu D, Chen W. Systemic Analysis of the Prognosis-Related RNA Alternative Splicing Signals in Melanoma. Med Sci Monit. 2020; 26: e921133.
Lu T, Chen S, Qu L, Wang Y, Chen HD, He C. Identification of a five-miRNA signature predicting survival in cutaneous melanoma cancer patients. PeerJ. 2019; 7: e7831.
Sun XY, Yu SZ, Zhang HP, Li J, Guo WZ, Zhang SJ. A signature of 33 immune-related gene pairs predicts clinical outcome in hepatocellular carcinoma. Cancer Med. 2020; 9: 2868-78.
Wu J, Zhao Y, Zhang J, Wu Q, Wang W. Development and validation of an immune-related gene pairs signature in colorectal cancer. Oncoimmunology. 2019; 8: 1596715.
Breuer K, Foroushani AK, Laird MR, Chen C, Sribnaia A, Lo R, Winsor GL, Hancock RE, Brinkman FS, Lynn DJ. InnateDB: systems biology of innate immunity and beyond--recent updates and continuing curation. Nucleic Acids Res. 2013; 41: D1228-33.
Cancer Genome Atlas Research N, Weinstein JN, Collisson EA, Mills GB, Shaw KR, Ozenberger BA, Ellrott K, Shmulevich I, Sander C, Stuart JM. The Cancer Genome Atlas Pan-Cancer analysis project. Nat Genet. 2013; 45: 1113-20.
Jayawardana K, Schramm SJ, Haydu L, Thompson JF, Scolyer RA, Mann GJ, Muller S, Yang JY. Determination of prognosis in metastatic melanoma through integration of clinico-pathologic, mutation, mRNA, microRNA, and protein information. Int J Cancer. 2015; 136: 863-74.
Jonsson G, Busch C, Knappskog S, Geisler J, Miletic H, Ringner M, Lillehaug JR, Borg A, Lonning PE. Gene expression profiling-based identification of molecular subtypes in stage IV melanomas with different clinical outcome. Clin Cancer Res. 2010; 16: 3356-67.
Cirenajwis H, Ekedahl H, Lauss M, Harbst K, Carneiro A, Enoksson J, Rosengren F, Werner-Hartman L, Torngren T, Kvist A, Fredlund E, Bendahl PO, Jirstrom K, Lundgren L, Howlin J, Borg A, Gruvberger-Saal SK, Saal LH, Nielsen K, Ringner M, Tsao H, Olsson H, Ingvar C, Staaf J, Jonsson G. Molecular stratification of metastatic melanoma using gene expression profiling: Prediction of survival outcome and benefit from molecular targeted therapy. Oncotarget. 2015; 6: 12297-309.
Edgar R, Domrachev M, Lash AE. Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 2002; 30: 207-10.
Friedman J, Hastie T, Tibshirani R. Regularization Paths for Generalized Linear Models via Coordinate Descent. J Stat Softw. 2010; 33: 1-22.
Yu G, Wang LG, Han Y, He QY. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS. 2012; 16: 284-7.
Li L, Wang XL, Lei Q, Sun CZ, Xi Y, Chen R, He YW. Comprehensive immunogenomic landscape analysis of prognosis-related genes in head and neck cancer. Sci Rep. 2020; 10: 6395.
Lin K, Huang J, Luo H, Luo C, Zhu X, Bu F, Xiao H, Xiao L, Zhu Z. Development of a prognostic index and screening of potential biomarkers based on immunogenomic landscape analysis of colorectal cancer. Aging (Albany NY). 2020; 12: 5832-57.
Chen W, Ou M, Tang D, Dai Y, Du W. Identification and Validation of Immune-Related Gene Prognostic Signature for Hepatocellular Carcinoma. J Immunol Res. 2020; 2020: 5494858.
Qu Y, Cheng B, Shao N, Jia Y, Song Q, Tan B, Wang J. Prognostic value of immune-related genes in the tumor microenvironment of lung adenocarcinoma and lung squamous cell carcinoma. Aging (Albany NY). 2020; 12: 4757-77.
Deng X, Lin D, Zhang X, Shen X, Yang Z, Yang L, Lu X, Yu L, Zhang N, Lin J. Profiles of immune-related genes and immune cell infiltration in the tumor microenvironment of diffuse lower-grade gliomas. J Cell Physiol. 2020.
Pidugu VK, Pidugu HB, Wu MM, Liu CJ, Lee TC. Emerging Functions of Human IFIT Proteins in Cancer. Front Mol Biosci. 2019; 6: 148.
Huang J, Lo UG, Wu S, Wang B, Pong RC, Lai CH, Lin H, He D, Hsieh JT, Wu K. The roles and mechanism of IFIT5 in bladder cancer epithelial-mesenchymal transition and progression. Cell Death Dis. 2019; 10: 437.
Karin N, Razon H. Chemokines beyond chemo-attraction: CXCL10 and its significant role in cancer and autoimmunity. Cytokine. 2018; 109: 24-8.
Tokunaga R, Zhang W, Naseem M, Puccini A, Berger MD, Soni S, McSkane M, Baba H, Lenz HJ. CXCL9, CXCL10, CXCL11/CXCR3 axis for immune activation - A target for novel cancer therapy. Cancer Treat Rev. 2018; 63: 40-7.
Huang L, Chen J, Zhao Y, Gu L, Shao X, Li J, Xu Y, Liu Z, Xu Q. Key candidate genes of STAT1 and CXCL10 in melanoma identified by integrated bioinformatical analysis. IUBMB Life. 2019; 71: 1634-44.

PDF

Journal Publication

published 01 Dec, 2020

Read the published version in Clinical, Cosmetic and Investigational Dermatology →

Version 1

posted 20 Aug, 2020

You are reading this latest preprint version

Development and Validation of An Immune-Related Gene Pair Signature in Skin Cutaneous Melanoma

Status:

Journal Publication

Version 1

Abstract

Figures

1 Introduction

2 Methods

2.1 Sources of immune-related genes

2.2 Data source and download

2.3 Data preprocessing

2.4 Construction of IRGPs

2.5 Establishment and validation of the prognostic signature based on IRGPs

2.6 Functional enrichment analysis

2.7 The nomogram and forest plots of the risk score and clinical features

2.8 Statistical analysis

3 Results

3.1 Consistency assessment of datasets on different platforms

3.2 Construction of the IRGP signature in SKCM

3.3 IRGP signature evaluation and validation for survival prediction

3.4 Biological functions related to the IRGP signature

3.5 Risk signature and clinical characteristics analysis

3.6 The nomogram and forest plots of the risk score and clinical features

4 Discussion

Declarations

References

Status:

Journal Publication

Version 1