Individualized immune-related gene signature predicts immune status and oncologic outcomes in clear cell renal cell carcinoma patients

ARTICLE IN PRESS

Urologic Oncology: Seminars and Original Investigations 000 (2019) 1−8

Laboratory-Kidney cancer

Individualized immune-related gene signature predicts immune status and oncologic outcomes in clear cell renal cell carcinoma patients Ying Xiong, M.D.#,*, Li Liu, Ph.D.#, Qi Bai, M.D.#, Yu Xia, M.D., Yang Qu, M.D., Jiajun Wang, M.D., Jianming Guo, Ph.D.* Department of Urology, Zhongshan Hospital, Fudan University, Shanghai, China Received 4 March 2019; received in revised form 1 September 2019; accepted 22 September 2019

Abstract Objective: To develop an individualized immune-related gene signature that predicts oncologic outcomes and immune status of clear cell renal cell carcinoma (ccRCC). Materials and Methods: The immune-related gene pair (IGP) index was constructed and validated based on pairwise comparison in 634 ccRCC patients. Association with overall survival (OS), progression-free interval (PFI), and disease-specific survival (DSS) was evaluated by Kaplan-Meier analysis, univariate and multivariate cox regression survival analysis. Prognostic values of different risk models were compared using Harrell’s C-index. Results: The IGP index of 17 gene pairs was an adverse independent risk factor in multivariate analyses for OS (hazard ratio [HR], 1.718; P = 0.001), PFI (HR, 1.550; P = 0.006), and DSS (HR, 2.201; P = 0.001) in ccRCC patients. It showed comparable prognostic accuracy with ccA/ccB signature (C-index for OS, 0.657 vs. 0.640; P = 0.686) and better intratumor homogeneity. Immunosuppressive immune cell, markers, and pathways were all enriched in high immune-risk tumors. The integrated immune-clinical prognostic score outperformed ccA/ccB signature and University of California Integrated Staging System risk model in terms of C-index for estimation of OS, PFI, and DSS (P < 0.001). Conclusions: The proposed IGP index is a robust and promising biomarker for estimating oncologic outcomes in ccRCC. High immunerisk tumors are immunosuppressive. Ó 2019 Elsevier Inc. All rights reserved.

Keywords: Clear cell renal cell carcinoma; Prognosis; Immune microenvironment; Molecular biomarkers; Predictive models

This study was funded by grants from National Natural Science Foundation of China (81472376, 81702496, 81702497, 81702805, and 81772696), Shanghai Municipal Commission of Health and Family Planning (20174Y0042), and Zhongshan Hospital Science Foundation (2016ZSQN30, 2017ZSQN18, and 2017ZSYQ26). All these study sponsors have no roles in the study design, in the collection, analysis, and in the interpretation of data. Author contributions: Y. Xiong, L. Liu, Q. Bai for acquisition, analysis and interpretation of data, drafting of the manuscript; Y. Xia Y. Qu, and J. Wang for technical and material support; J. Guo for study concept and design, analysis and interpretation of data, drafting of the manuscript, obtained funding and study supervision. All authors read and approved the final manuscript. # These authors contributed equally to this work. *Corresponding authors. Tel.: 86-21-5479847; fax: 86-21-64735503. E-mail addresses: [email protected] (Y. Xiong), [email protected] (J. Guo). https://doi.org/10.1016/j.urolonc.2019.09.014 1078-1439/Ó 2019 Elsevier Inc. All rights reserved.

1. Introduction Renal cell carcinoma (RCC) afflicts about 271,000 patients estimated worldwide every year, accounting for 2% to 3% adult malignancies [1,2]. Clear cell renal cell carcinoma (ccRCC) is the major histological subtype, representing 70% to 80% RCC [3]. Around one third of ccRCC patients would still develop recurrences or metastases after curative surgeries [4]. Cancers develop in complex tissue microenvironment. Interactions between cancer cells and immune microenvironment play fundamental roles in tumor initiation, progression, and metastasis. A critical step in tumor progression is immune evasion and immunosuppression [5]. In particular, ccRCC has been shown to be highly

ARTICLE IN PRESS 2

Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

immune infiltrated and featured an increased immune signature [6]. It was one of the first cancers to respond to immunotherapy and among the most responsive [7]. However, no current prognostic molecular characteristics comprehensively described the tumor microenvironment. The current integrated models like University of California Integrated Staging System (UISS) have reached a ceiling for lack of genetic information in ccRCC [8]. Available public gene expression data sets make it possible to identify reliable genetic ccRCC biomarkers. Traditional methods based on gene expression level require adequate normalization, which is hard to achieve due to technical biases in different platforms and biological heterogeneity among data sets [9]. On the contrary, relative ranking of gene expression provides the opportunity to avoid data preprocessing, such as normalization and scaling. Methods based on this can be carried out on single tumor samples and get robust results in prognosis evaluation of cancer [10]. In this study, we developed and validated a personalized immune-related gene signature based on pairwise comparison, without need for normalization. The gene signature could reflect tumor immune status and oncologic outcomes of ccRCC patients. 2. Materials and methods 2.1. Patients and data collection We retrospectively analyzed the gene expression profiles of ccRCC tumor samples and normal tissues in 6 public ccRCC cohorts. Two datasets in Gene Expression Omnibus (GSE40435 [11] and GSE53757 [12]) were used to find out differentially expressed genes in tumor tissues compared with normal tissues. We took the Sato et al. cohort of 101 Japanese ccRCC patients as the discovery set for the construction of immune-related gene pair (IGP) index [13]. The microarray gene expression dataset of the Sato et al. cohort was downloaded from ArrayExpress (http://www. ebi.ac.uk/arrayexpress/experiments/E-MTAB-1980/). The prognostic value of IGP index was validated in a RNA-seq data for TCGA kidney clear cell carcinoma (KIRC) cohort, which was downloaded from the UCSC Xena (https://xenab rowser.net/heatmap/). The KIRC cohort served as the validation set to validate the prognostic value of IGP index. Genes with over 80% zero mRNA level were excluded from further analysis. Intratumor heterogeneity of IGP risk groups was assessed in 10 tumor samples with multiregion sequencing from 2 Gene Expression Omnibus datasets (GSE31610 [14] and GSE53000 [15]). Preparation of tumor samples and measurement of RNA can be found in corresponding research papers in details [11−15]. Clinical information of the TCGA KIRC validation cohort was downloaded from the TCGA Pan-cancer Clinical Data Resource, which provides high-quality clinical data [16]. Overall survival (OS) event was defined as death from any cause. Progression-free interval (PFI) was defined as the length of time during and after the treatment of a disease,

such as cancer, that a patient lives with the disease but it does not get worse. Disease-specific survival (DSS) rate represents the percentage of people in a study or treatment group who have not died from a specific disease in a defined period of time. The time period usually begins at the time of diagnosis or at the start of treatment and ends at the time of death. Patients who died from causes other than the disease being studied are not counted in this measurement [16]. Data were downloaded from February 12 to April 23, 2018. 2.2. Construction of IGP index and immune-clinical prognostic score We constructed an immune-related gene signature based on immune-related genes downloaded from the ImmPort database [17]. Only genes measured across all 6 cohorts were selected. Scores were assigned to each IGP by pairwise comparison [18]. For example, if CD247 expression was more than CXCL5 expression, the gene pair CD247CXCL5 was scored 1; otherwise CD247-CXCL5 was scored 0. This gene-pair-based method depended entirely on gene expression profile of an individualized tumor sample with no need for normalization. The constructed IGP index was not affected by batch effect or measurement platforms. IGPs with constant values (over 80% samples were assigned the same score in discovery set) were removed from further construction of IGP index because such IGPs did not provide enough discriminative information. The constant values may be attributed to biologically preferential transcription. IGPs significantly associated with OS in discovery set determined by univariate cox regression model (P < 0.01, jcoefficientj>1.5) were candidates to construct the IGP index. We applied least absolute shrinkage and selection operator cox regression model to calculate the appropriate coefficient for each prognostic IGP and estimate the likelihood deviance (glmnet, R software). Cutoff point was calculated based on the receiver operating characteristic curve with Manhattan distance method [19]. We constructed a prognostic nomogram based on results from multivariate analyses in validation set. The composite immune-clinical prognostic (ICP) score was consisted of tumor stage, histological grade, and IGP index. The prognostic value of IGP index and ICP score was evaluated by Harrell’s concordance index (C-index; Hmisc, R software). C-index and 95% confidence interval [CI] were calculated from 1,000 bootstrap samples to reduce overfitting. The performance of IGP index and ICP score was compared with UISS risk model [20] and ccA/ccB signatures classified with ClearCode34 [21,22] in terms of C-index. 2.3. Immune cell infiltration and functional annotation We evaluated infiltrations of CD8+ T cells, natural killer cells, dendritic cells, CD4 memory T cells, regulatory T cells, macrophages and neutrophils in validation set with CIBERSORT, a computational approach for inferring leukocyte

ARTICLE IN PRESS Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

representation in bulk tumor transcriptomes [23]. Metagene values for CD8+ T cells [24], natural killer cells [25], dendritic cells [25], CD4 helper T cells [25], regulatory T cells [25], macrophages [25], neutrophils [25], and immunosuppression [26] were computed to summarize an immune gene signature as the mean value each sample of genes in the category. Additional comparison of gene expression profile of high immune-risk tumors and low immune-risk tumors was performed by Gene set enrichment analysis [27] in order to gain biological understanding of IGP index. 2.4. Statistical analysis Survival curves were obtained with Kaplan-Meier analysis and differences between high and low immune-risk groups were calculated with log-rank test. Univariate and multivariate analyses were carried out to evaluate the prognostic value of IGP index. We performed 1,000 resample bootstrap validations in multivariate analyses to reduce overfitting bias. Statistical significance for comparisons between 2 groups was assessed by the Student t test, MannWhitney U test, Pearson’s chi-square test or Cochran-Mantel-Haenszel x2 test. The correlation between 2 parameters was determined with Pearson correlation coefficient and linear regression analysis. All statistical tests were 2-sided and statistical significance was defined as P < 0.05 unless specified. Data analyses were performed using SPSS Statistics 21.0 and R software. 3. Results 3.1. Construction of IGP index A total of 634 ccRCC patients (101 patients in discovery set and 533 patients in validation set) were included in construction and validation of IGP index. Baseline characteristics of patients in 2 cohorts were shown in Table S1. Among the 1,811 immune-related genes, 1,185 genes were measured across all 6 cohorts. We identified 610 differentially expressed immune-related genes in tumor samples compared with normal tissues in both GSE40435 and GSE53757 cohorts and 185745 IGPs were constructed. Eighty-five percent of IGPs with unchanged score were excluded from further analysis. We performed univariate cox regression analyses with the remaining 28,125 varied IGPs in discovery set and 153 IGPs showed strong prognostic value (P < 0.01, jcoefficientj>1.5; Fig. 1A). IGP index was constructed with 17 IGPs using least absolute shrinkage and selection operator Cox regression models. Coefficients for these seventeen IGPs were chosen when log lambda = 0.0413 (Fig. 1B). The partial likelihood deviance for each lambda was exhibited (Fig. 1C). Seventeen selected gene pairs and their coefficients were listed in Table S2. The distribution of IGP index in discovery and validation set did not show statistical significance (Fig. 1E and Table S1). The cutoff point for IGP index calculated by

3

receiver operating characteristic curve was 0.443, stratifying patients into high immune-risk and low immune-risk groups. In discovery set, patients with high immune risk (IGP score > 0.443) had significant worse OS than patients with low immune risk (IGP score < 0.443; P < 0.001; Fig. 1E). In univariate analysis, IGP index [P < 0.001, hazard ratio [HR] = 22.095(8.340−58.540)] and IGP risk group [P < 0.001, HR = 24.978(5.822−107.165)] was associated with OS in discovery set (Table S3). After adjustment for tumor stage and histological grade, the association remained significant [IGP index: P = 0.001, HR = 16.081(5.725−45.167); IGP risk groups: P = 0.001, HR = 15.728(3.533−70.026)] (Tables 1 and S4). 3.2. IGP index and patient survival in validation set To evaluate the robustness of the prognostic value of IGP index, we tested IGP index in an independent validation set. We observed high immune-risk group was associated with worse OS (P < 0.001), PFI (P < 0.001), and DSS (P < 0.001) vs. low immune-risk group via Kaplan-Meier analysis (Fig. 2A−C). Univariate cox regression analyses revealed that stage at diagnosis (P < 0.001), histological grade (P < 0.001), and IGP risk group (P < 0.001) were adverse risk factors for all three oncologic endpoints (Table S3). Only 25% patients had Eastern Cooperative Oncology Group performance status (ECOG PS) information, which may partly explain why ECOG PS was not associated with patient survival. The association between IGP index and oncologic outcomes remained significant in multivariate analyses that excluded tumor stage and histological grade as confounding variables after bootstrap validation: in the analysis of OS, the HR for death associated with IGP index was 1.718 (95% CI, 1.307−2.257; P = 0.001); in the analysis of PFI, the HR for progression was 1.550 (95% CI, 1.154−2.082); in the analysis of DSS, the HR for death was 2.201 (95% CI, 1.571−3.084; Table 1). High immune-risk group was an independent risk factor in analysis of OS (P = 0.004), PFI (P = 0.010), DSS (P = 0.002) as well (Table S4). A benefit of adjuvant sunitinib over placebo was observed in high-risk ccRCC patients [28]. Localized ccRCC patients had worse OS (P < 0.001), PFI (P < 0.001), DSS (P < 0.001) in high immune-risk group (Fig. S1A−S1C). We further assessed the association between IGP risk groups and oncologic outcomes among high-risk localized ccRCC patients. Advanced stage (stages III and IV) localized high immune-risk tumors were associated with a lower rate of 5-year OS than low immune-risk tumors [P = 0.001; HR = 2.591; 95% CI, 1.465−4.584] (Fig. S1D). High histological grade (grades III and IV) localized high immune-risk ccRCC patients had worse OS than low immune-risk tumors [P = 0.034; HR = 1.682; 95% CI, 1.039−2.723] (Fig. S1D). We found similar associations in analysis of PFI and DSS as well (Fig. S1E and F). IGP index might help select high-risk localized

ARTICLE IN PRESS 4

Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

Fig. 1. Construction of immune-related gene pair index. (A) Flow chart of immune-related gene pair selection. (B) The least absolute shrinkage and selection operator (LASSO) coefficient profiles of the selected immune-related gene pairs. (C) Partial likelihood deviance for LASSO coefficient profiles. (D) KaplanMeier analysis of overall survival (OS) in discovery set. (E) Immune-related gene pair index (IGP index) distribution in discovery and validation set.

ccRCC patients who might benefit from adjuvant tyrosine kinase inhibitor treatment.

Table 1 Multivariate cox proportional hazards analysis of ccRCC patients Variables OS, Discovery set Stage at diagnosis, per increase in stage Histological grade, per increase in grade IGP index OS, Validation set Stage at diagnosis, per increase in stage Histological grade, per increase in grade IGP index PFI, Validation set Stage at diagnosis, per increase in stage Histological grade, per increase in grade IGP index DSS, validation set Stage at diagnosis, per increase in stage Histological grade, per increase in grade IGP index

Hazard ratio

95% CI

P value*

1.634

1.059−2.524

0.027

0.932

0.475−1.829

0.837

16.081

5.725−45.167

0.001

1.655

1.430−1.915

0.001

1.316

1.040−1.666

0.035

1.718

1.307−2.257

0.001

2.380

2.012−2.814

0.001

1.517

1.183−1.946

0.003

1.550

1.154−2.082

0.006

2.660

2.127−3.327

0.001

1.293

0.968−1.727

0.094

2.201

1.571−3.084

0.001

P values < 0.05 were in bold. DSS = disease-specific survival; OS = overall survival; PFI = progression-free interval. * P value <0.05 was regarded as statistically significant, bootstrapping with 1,000 resamples was performed.

3.3. Assessment of prognostic value and intratumor heterogeneity of IGP index We compared prognostic value of IGP index and IGP risk groups with ccA/ccB signature, the best current binary classification ccRCC biomarkers [21,29]. For OS, both IGP index (C-index, 0.657) and IGP risk group (C-index, 0.618) achieved comparable C-index with ccA/ccB signature (C-index, 0.640; IGP index vs. ccA/ccB, P = 0.686; IGP risk group vs. ccA/ccB, P = 0.174; Fig. 2D). Similarly, IGP index and IGP risk group were no worse than ccA/ccB signature in terms of C-index for PFI (IGP index vs. ccA/ccB, P = 0.889; IGP risk group vs. ccA/ccB, P = 0.834) and DSS (IGP index vs. ccA/ccB, P = 0.596; IGP risk group vs. ccA/ccB, P = 0.222; Fig. 2E and F). Besides, multivariate analyses (Table S5) indicated that IGP index was able to improve the prognostic accuracy of ccA/ccB signature and UISS risk model. Primary ccRCC featured strong Intratumor molecular heterogeneity, making it difficult to accurately profiling ccRCC tumors [15]. Previous evaluation of the ccA/ccB signature across multiple tumor regions from each of 10 ccRCCs demonstrated heterogeneous expression patterns 8 of 10 cases [29]. We assessed intratumor heterogeneity with GSE31610 [4] and GSE53000 [5], in which 10 tumor samples were multiregion sequenced. Six of 10 samples

ARTICLE IN PRESS Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

5

Fig. 2. Immune-related gene pair index and patient survival in validation set. (A) Kaplan-Meier analysis of OS in validation set. (B) Kaplan-Meier analysis of progression-free interval (PFI) in validation set. (C) Kaplan-Meier analysis of disease-specific survival (DSS) in validation set. (D) C-index comparison for OS in validation set. (E) C-index comparison for PFI in validation set. (F) C-index comparison for DSS in validation set. (G) Intratumor heterogeneity of IGP index in ten tumors with multiregion sequencing. n.s. P > 0.05; error bars: 95% confidential interval (CI).

demonstrated consistent IGP risk groups (Fig. 2G). IGP risk group showed better homogeneity than ccA/ccB signature in ccRCC tumors. 3.4. Different features of immune microenvironment in high vs. low immune-risk tumors To explore whether IGP index imparted distinct features on the tumor immune microenvironment, we first analyzed the immune cell infiltration in different IGP risk groups. High immune-risk tumors were infiltrated with more regulatory T cells (P < 0.001), macrophages (P < 0.001), neutrophils (P = 0.009), and less natural killer cells (P < 0.001; Fig. 3A). In order to validate the distinct immune cell composition in IGP risk groups, we measured metagene values of these immune cells. In accordance with immune cell fraction, elevated expression of regulatory T cells (P < 0.001), macrophages (P < 0.001), neutrophils (P = 0.023), and reduced expression of natural killer cells (P = 0.001) metagene values were observed in high immune-risk tumors (Fig. 3B). We evaluated correlation between IGP index and gene clusters referring to immune suppression [30] and found that most of the markers of immune suppression (FOXP3, CTLA4, CASP3, IL10, IL10RB, TGFB1, TGFBR1, CD80, CD86, CSF2, IL1R1, and CD4) are positively correlated with IGP index (P < 0.001; Fig. 3D). High immune-risk tumors displayed heightened immunosuppression metagene values (P < 0.001), while the well-established T cell activation marker GZMB/CD8A [31] were down-regulated with statistical significance (P = 0.001; Fig. 3E and F). Additional gene expression profile

comparison between high vs. low immune-risk tumors was performed by Gene set enrichment analysis with C5 GO gene sets. Negative regulation of lymphocyte mediated immunity (P = 0.004, Q = 0.008) and negative regulation of adaptive immune response pathways (P = 0.026, Q = 0.050) were significantly up-regulated in high immune-risk tumors, further confirmed an immunosuppressive microenvironment in high immune-risk tumors (Fig. 3G and H). 3.5. ICP score combining IGP index and clinical data In multivariate analyses for 2 of the 3 endpoints in validation set, stage, grade, and IGP index were independent adverse prognostic factors, suggesting the complementary value among them. We constructed a nomogram based on multivariate analysis of OS in validation set and derived an ICP score as (33.2 £ stage + 18.2 £ grade + 35.7 £ IGP index + 5.7; Fig. 4A). The calibration plots of the nomogram were shown for 5-year OS, PFI, DSS predictions and displayed good consistency between predicted and actual observation of the 3 oncologic outcomes (Fig. 4B). Significant improvement of predictive accuracy of OS was observed in ICP score (C-index, 0.755) relative to UISS risk model (C-index, 0.720; P < 0.001; Fig. 4C). For PFI and DSS, ICP score outperformed UISS risk models as well (PFI, P < 0.001; DSS, P < 0.001; Fig. 4D and E). ICP score achieved a higher accuracy of survival estimation compared with gene signature (IGP index, IGP risk group, and ccA/ ccB signature) alone (Fig. 4C−E). The ICP score could not be compared to other prognostic risk models due to lack of essential parameters.

ARTICLE IN PRESS 6

Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

Fig. 3. Different features of immune microenvironment in high vs. low immune-risk tumors. (A) Immune cell fractions of high vs. low immune-risk tumors. (B) Immune cell metagene values in high vs. low immune-risk tumors. (C) Correlation between immune cell fractions, immune cell metagene values, and IGP index in validation set. (D) Correlation between immune suppression markers and IGP index. (E) Metagene value of immunosuppression in high vs. low immune-risk tumors. (F) GZMB/CD8A ratio in high vs. low immune-risk tumors. (G) Enrichment of “negative regulation of lymphocyte mediated immunity” gene set in high immune-risk tumors. (H) Enrichment of “negative regulation of adaptive immune response pathways” gene set in high immunerisk tumors. n.s. P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001.

4. Discussion In this study, we developed an immune-related gene signature based on 17 IGPs for ccRCC and validated it in an independent cohort. IGP index high tumors were associated with worse oncological outcomes. The association remained significant in high-risk localized tumors. IGP index was an independent adverse risk factor for OS, PFI, and DSS. The prognostic gene signature showed comparable predictive accuracy and better Intratumor homogeneity than ccA/ccB signature. It could add prognostic value to stage, grade, ccA/ccB signature, and UISS risk model. High immune-risk tumors demonstrated a highly immunosuppressive tumor microenvironment. The integrated ICP score outperformed ccA/ccB signature and the UISS risk model in analyses of OS, PFI, and DSS. The ccA/ccB gene signature was currently considered to be the best gene expression signature [29]. The IGP index was promising because it showed comparable prognostic value, less intratumor heterogeneity and ability to improve ccA/ccB signature significantly. Meanwhile, the immunerelated gene signature was developed using methods

specially designed for data across different technical platforms with microarray or RNA-Seq data despite inherent bias [10]. The IGP index derived from relative ranking of gene mRNA level, thus eliminating need for data preprocessing, making it ready to be translated into clinical utility as a single-sample estimate of survival of ccRCC. Nowadays, the effects of adjuvant targeted therapy in localized ccRCC are unclear [28]. Compared with low immune-risk tumors, advanced stage and grade tumors with high immune risk may benefit from more stringent surveillance and might be good candidates for adjuvant tyrosine kinase inhibitors or immunotherapy. We observed increased regulatory T cells, tumor-associated macrophages and neutrophils infiltration and decreased natural killer cell infiltration in high immune-risk tumors. Regulatory T cells, tumor-associated macrophages, and neutrophils have been associated with protumor functions and poor prognosis, whereas natural killer cells play a key role in antitumor immunity [32,33]. In accordance with this, immunosuppression gene signature and pathways up-regulated in high immune-risk tumors while cytotoxic signature down-regulated. IGP index accurately reflected the interaction between tumor cells and

ARTICLE IN PRESS Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

7

Fig. 4. Immune-clinical prognostic score combining immune-related gene pair index and clinical data. (A) A prognostic nomogram consisted of stage, grade, and IGP index. (B) Calibration plots of the constructed nomogram to predict 5-year OS, PFI, and DSS. (C) Comparison of predictive accuracy between ICP score and other risk models for OS in terms of C-index. (D) Comparison of predictive accuracy between ICP score and other risk models for OS in terms of C-index. (E) Comparison of predictive accuracy between ICP score and other risk models for PFI in term s of C-index. (F) Comparison of predictive accuracy between ICP score and other risk models for DSS in terms of C-index. *** P < 0.001; error bars: 95% CI.

immune microenvironment in ccRCC, which represented a strong relationship that influenced tumor progression and patient survival [5]. Prognostic biomarkers related to the tumor microenvironment are promising in improving patient management in the era of immunotherapy [34]. However, some major limitations remained. First, the results of our study should be further validated in prospective ccRCC cohort. Further, the size of the discovery set was relatively small. Besides, due to intrinsic limitations of online datasets, some important prognostic information was incomplete such as ECOG PS. Only 134 of 533 (25.1%) patients had information of ECOG PS. 5. Conclusions In conclusion, we constructed a promising prognostic biomarker based on 17 IGPs in ccRCC. The IGP index was not only an independent adverse risk factor for OS, PFI, and DSS, but also an indicator of tumor immune microenvironment for ccRCC patients. Limitations included its retrospective design in nature, single validation cohort, and unavoidable Intratumor heterogeneity. Multicentered prospective study is needed to further validate the prognostic value and explore its clinical utility. Conflict of interest The authors declare no conflict of interest.

Supplementary materials Supplementary material associated with this article can be found in the online version at https://doi.org/10.1016/j. urolonc.2019.09.014. References [1] Ferlay J, Shin HR, Bray F, Forman D, Mathers C, Parkin DM. Estimates of worldwide burden of cancer in 2008: GLOBOCAN 2008. Int J Cancer 2010;127:2893–917. [2] Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Global cancer statistics, 2012. CA Cancer J Clin 2015;65:87–108. [3] Cohen HT, McGovern FJ. Renal-cell carcinoma. N Engl J Med 2005;353:2477–90. [4] Gupta K, Miller JD, Li JZ, Russell MW, Charbonneau C. Epidemiologic and socioeconomic burden of metastatic renal cell carcinoma (mRCC): a literature review. Cancer Treat Rev 2008;34: 193–205. [5] Quail DF, Joyce JA. Microenvironmental regulation of tumor progression and metastasis. Nat Med 2013;19:1423–37. [6] Senbabaoglu Y, Gejman RS, Winer AG, Liu M, Van Allen EM, de Velasco G, et al. Tumor immune microenvironment characterization in clear cell renal cell carcinoma identifies prognostic and immunotherapeutically relevant messenger RNA signatures. Genome Biol 2016;17:231. [7] Escudier B. Emerging immunotherapies for renal cell carcinoma. Ann Oncol 2012;23(Suppl 8):viii35–40. [8] Volpe A, Patard JJ. Prognostic factors in renal cell carcinoma. World J Urol 2010;28:319–27. [9] Leek JT, Scharpf RB, Bravo HC, Simcha D, Langmead B, Johnson WE, et al. Tackling the widespread and critical impact of batch effects in high-throughput data. Nat Rev Genet 2010;11:733–9.

ARTICLE IN PRESS 8

Y. Xiong et al. / Urologic Oncology: Seminars and Original Investigations 00 (2019) 1−8

[10] Heinaniemi M, Nykter M, Kramer R, Wienecke-Baldacchino A, Sinkkonen L, Zhou JX, et al. Gene-pair expression signatures reveal lineage control. Nat Methods 2013;10:577–83. [11] Wozniak MB, Le Calvez-Kelm F, Abedi-Ardekani B, Byrnes G, Durand G, Carreira C, et al. Integrative genome-wide gene expression profiling of clear cell renal cell carcinoma in Czech Republic and in the United States. PLoS One 2013;8:e57886. [12] von Roemeling CA, Radisky DC, Marlow LA, Cooper SJ, Grebe SK, Anastasiadis PZ, et al. Neuronal pentraxin 2 supports clear cell renal cell carcinoma by activating the AMPA-selective glutamate receptor4. Cancer Res 2014;74:4796–810. [13] Sato Y, Yoshizato T, Shiraishi Y, Maekawa S, Okuno Y, Kamura T, et al. Integrated molecular analysis of clear-cell renal cell carcinoma. Nat Genet 2013;45:860–7. [14] Gerlinger M, Rowan AJ, Horswell S, Math M, Larkin J, Endesfelder D, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med 2012;366:883–92. [15] Gerlinger M, Horswell S, Larkin J, Rowan AJ, Salm MP, Varela I, et al. Genomic architecture and evolution of clear cell renal cell carcinomas defined by multiregion sequencing. Nat Genet 2014;46:225–33. [16] Liu J, Lichtenberg T, Hoadley KA, Poisson LM, Lazar AJ, Cherniack AD, et al. An integrated TCGA pan-cancer clinical data resource to drive high-quality survival outcome analytics. Cell 2018;173:400-16 e11. [17] Bhattacharya S, Andorf S, Gomes L, Dunn P, Schaefer H, Pontius J, et al. ImmPort: disseminating data to the public for the future of immunology. Immunol Res 2014;58:234–9. [18] Li B, Cui Y, Diehn M, Li R. Development and validation of an individualized immune prognostic signature in early-stage nonsquamous non-small cell lung cancer. JAMA Oncol 2017;3:1529–37. [19] Budczies J, Klauschen F, Sinn BV, Gyorffy B, Schmitt WD, DarbEsfahani S, et al. Cutoff Finder: a comprehensive and straightforward Web application enabling rapid biomarker cutoff optimization. PLoS One 2012;7:e51862. [20] Zisman A, Pantuck AJ, Dorey F, Said JW, Shvarts O, Quintana D, et al. Improved prognostication of renal cell carcinoma using an integrated staging system. J Clin Oncol 2001;19:1649–57. [21] Brooks SA, Brannon AR, Parker JS, Fisher JC, Sen O, Kattan MW, et al. ClearCode34: a prognostic risk predictor for localized clear cell renal cell carcinoma. Eur Urol 2014;66:77–84. [22] Brannon AR, Reddy A, Seiler M, Arreola A, Moore DT, Pruthi RS, et al. Molecular stratification of clear cell renal cell carcinoma by

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32] [33] [34]

consensus clustering reveals distinct subtypes and survival patterns. Genes Cancer 2010;1:152–63. Gentles AJ, Newman AM, Liu CL, Bratman SV, Feng W, Kim D, et al. The prognostic landscape of genes and infiltrating immune cells across human cancers. Nat Med 2015;21:938–45. Iglesia MD, Vincent BG, Parker JS, Hoadley KA, Carey LA, Perou CM, et al. Prognostic B-cell signatures using mRNA-seq in patients with subtype-specific breast and ovarian cancer. Clin Cancer Res 2014;20:3818–29. Bindea G, Mlecnik B, Tosolini M, Kirilovsky A, Waldner M, Obenauf AC, et al. Spatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer. Immunity 2013;39:782–95. Kardos J, Chai S, Mose LE, Selitsky SR, Krishnan B, Saito R, et al. Claudin-low bladder tumors are immune infiltrated and actively immune suppressed. JCI Insight 2016;1:e85902. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA 2005;102:15545–50. Motzer RJ, Ravaud A, Patard JJ, Pandha HS, George DJ, Patel A, et al. Adjuvant sunitinib for high-risk renal cell carcinoma after nephrectomy: subgroup analyses and updated overall survival results. Eur Urol 2018;73:62–8. Gulati S, Martinez P, Joshi T, Birkbak NJ, Santos CR, Rowan AJ, et al. Systematic evaluation of the prognostic impact and intratumour heterogeneity of clear cell renal cell carcinoma biomarkers. Eur Urol 2014;66:936–48. Dutta N, Gupta A, Mazumder DN, Banerjee S. Down-regulation of locus-specific human lymphocyte antigen class I expression in Epstein-Barr virus-associated gastric cancer: implication for viralinduced immune evasion. Cancer 2006;106:1685–93. Pan D, Kobayashi A, Jiang P, Ferrari de Andrade L, Tay RE, Luoma AM, et al. A major chromatin regulator determines resistance of tumor cells to T cell-mediated killing. Science 2018;359:770–5. Nishikawa H. [Regulatory T cells in cancer immunotherapy]. Rinsho Ketsueki 2014;55:2183–9. Noy R, Pollard JW. Tumor-associated macrophages: from mechanisms to therapy. Immunity 2014;41:49–61. Vargas AJ, Harris CC. Biomarker development in the precision medicine era: lung cancer as a case study. Nat Rev Cancer 2016;16:525– 37.