Whole Exome and Transcriptome Analyses Integrated With

Published OnlineFirst May 2, 2018; DOI: 10.1158/2326-6066.CIR-17-0453

Research Article Cancer Immunology Research Whole Exome and Transcriptome Analyses Integrated with Microenvironmental Immune Signatures of Lung Squamous Cell Carcinoma Jeong-Sun Seo1,2,3,4, Ji Won Lee2,3, Ahreum Kim2,3, Jong-Yeon Shin2,4, Yoo Jin Jung5, Sae Bom Lee5, Yoon Ho Kim5, Samina Park6, Hyun Joo Lee6, In-Kyu Park6, Chang-Hyun Kang6, Ji-Young Yun2,4, Jihye Kim2,4, and Young Tae Kim2,5,6

Abstract

The immune microenvironment in lung squamous cell carci- immune competent. We analyzed infiltrating stromal and noma (LUSC) is not well understood, with interactions between immune cells to further characterize the tumor microenviron- the host immune system and the tumor, as well as the molecular ment. Elevated expression of macrophage 2 signature genes in the pathogenesis of LUSC, awaiting better characterization. To date, immune competent subtype confirmed that tumor-associated no molecularly targeted agents have been developed for LUSC macrophages (TAM) linked inflammation and mutation-driven treatment. Identification of predictive and prognostic biomarkers cancer. A negative correlation was evident between the immune for LUSC could help optimize therapy decisions. We sequenced score and the amount of somatic copy-number variation (SCNV) whole exomes and RNA from 101 tumors and matched noncancer of immune genes (r ¼0.58). The SCNVs showed a potential control Korean samples. We used the information to predict detrimental effect on immunity in the immune-deficient subtype. subtype-specific interactions within the LUSC microenvironment Knowledge of the genomic alterations in the tumor microenvi- and to connect genomic alterations with immune signatures. ronment could be used to guide design of immunotherapy Hierarchical clustering based on gene expression and mutational options that are appropriate for patients with certain cancer profiling revealed subtypes that were either immune defective or subtypes. Cancer Immunol Res; 1–12. 2018 AACR.

Introduction immune system (7–9). Most cells in tumor stroma have some capacity to suppress a tumor, although this capacity changes as the Lung cancer is the second leading cause of death in Korea. The cancer progresses; invasion and metastasis can follow (10–13). most common type of primary lung cancer, lung adenocarcino- Immune and stromal characteristics have emerged as prognos- ma, has been characterized at the molecular level (1, 2). Lung tic and predictive factors that could be used to guide a person- squamous cell carcinoma, which accounts for 30% of all lung alized approach in cancer immunotherapy (14, 15). Analyses of cancers (3), is not well characterized due to poor understanding of genomic alterations, especially somatic mutations, have been the cancer's genomic evolution (4) and the antitumor activity of used to predict response to immunotherapy (16, 17). Here, we immune cells (5, 6). Genomic alterations in the tumor charac- used genomic and transcriptomic analysis to deﬁne molecular terize various stages of cancer progression. Immune defenses, on subtypes of tumors with immune responses. We show that the other hand, are governed by tumor stroma, including base- genomic alterations affect the tumor microenvironment and ment membrane, extracellular matrix, vasculature, and cells of the tumor development in a subtype-speciﬁc manner. The data show how genomic alterations and tumor microenvironment impact 1Precision Medicine Center, Seoul National University Bundang Hospital, cancer proliferation and invasion, and how predicted roles of Seongnamsi, Korea. 2Genomic Medicine Institute (GMI), Medical Research immune cells and their interactions with cancer cells in lung 3 Center, Seoul National University, Seoul, Republic of Korea. Department of squamous cell carcinoma (LUSC) might affect cancer therapy Biomedical Sciences, Seoul National University College of Medicine, Seoul, and patient survival. Republic of Korea. 4Macrogen Inc., Seoul, Republic of Korea. 5Seoul National University Cancer Research Institute, Seoul, Republic of Korea. 6Department of Thoracic and Cardiovascular Surgery, Seoul National University Hospital, Materials and Methods Seoul, Republic of Korea. RNA and whole-exome sequencing Note: Supplementary data for this article are available at Cancer Immunology All protocols of this study were approved by the Institutional Research Online (http://cancerimmunolres.aacrjournals.org/). Review Board of Seoul National University Hospital (IRB J.-S. Seo, J.W. Lee, A. Kim, and J.-Y. Shin contributed equally to this article. No:1312-117-545). Corresponding Authors: Jeong-Sun Seo, Precision Medicine Center, Seoul One hundred and one cases of lung squamous cell cancer National University Bundang Hospital, Seongnamsi 13605, Korea. Phone: 82- samples, taken between 2011 and 2013, were included. Of these 31-600-3001; E-mail: [email protected]; and Young Tae Kim, Department of 101 patients we excluded two patients, a patient treated with one Thoracic and Cardiovascular Surgery, Seoul National University Hospital, Seoul cycle of weekly docetaxel 65 mg and cisplatin 48 mg regimen 03080, Republic of Korea. Phone: 82-22-072-3161; Email: [email protected] preoperatively, and another patient who died of massive pulmonary doi: 10.1158/2326-6066.CIR-17-0453 embolism at 16 days after operation, from subsequent survival 2018 American Association for Cancer Research. analysis. All the tumor and matched adjacent noncancer

www.aacrjournals.org OF1

Seo et al.

Table 1. Clinical data summary ment software. The preprocessing pipeline on the GTAK website Korean (n ¼ 101) was followed (19). The raw read counts were generated using Patient characteristics Number of patients HTSeq-count for each annotated gene. Age at diagnosis, years Median 70 Range 35–83 Unsupervised subtype clustering Sex With the Ensembl gene set, the number of raw reads aligned to Male 95 each gene was computed by HT-seq count and was normalized by Female 6 the Variance Stabilizing Data (VSD) method with use of the R Smoking status package DEseq2. The variance for each gene was calculated, and Never-smoker 12 the top 1,000 genes by variance were selected for PCA analysis Former smoker 61 Current smoker 28 (20). PCA analysis using the 1,000 most variable genes was Median follow-up, months 45 conducted with all tumor and noncancer control samples. Sam- Tumor stage ples were clustered based on principal components into three I51groups noncancer control with 95% confidence interval by hier- II 29 archical clustering methods as implemented in the R package rgl III 21 (21). When analyzing RNA sequencing data, batch effects should IV 0 T stage be considered if experimental conditions and library preparation T1 27 varied. All of our samples were processed in the same batches, thus T2 58 additional batch-effect corrections were not necessary (22). T3 15 T4 1 Differentially expressed gene analysis N stage Differentially expressed genes of tumor subtypes compared with N0 62 N1 24 noncancer control expression in noncancer control cells were fi P < N2 15 determined by the signi cance criteria (adjusted 0.05, |Log2 Recurrancy 31 (fold change)| 1, and base mean 100) as implemented in the R Total LN packages DESeq2 and edgeR. The adjusted P value for multiple Median 30 testing was calculated by using the Benjamini–Hochberg correction – Range 5 66 from the computed P value (23). The centered VSD values of the differentially expressed gene list were applied to the array hierarchical clustering algorism (Cluster 3.0) with uncentered correlation control tissue specimens were grossly dissected immediate after and average linkage (24). The gene expression pattern was visual- surgery and preserved in liquid nitrogen. Data on clinical ized with use of JAVA treeview. The hierarchical tree by arrays was features such as smoking history, pathologic TNM stage, tumor generated by the clustering process and two types of gene sets in size, and degree of differentiations were collected (Table 1; differentially expressed genes (subtype A-UP and B-DOWN, sub- Supplementary Table S1). For RNA-seq, we extracted RNA from type A-DOWN and B-UP) were selected and enriched for Gene tissue using RNAiso Plus (Takara Bio Inc.), followed by puri- Ontology (GO) gene sets by Gene Set Enrichment Analysis (GSEA) fication using RNeasy MinElute (Qiagen). RNA was assessed for in order to determine genes enriched in ranked gene lists. quality and was quantified using an RNA 6000 Nano LabChip on a 2100 Bioanalyzer (Agilent Technologies). The RNA-seq Fragments per kilobase million (FPKM) calculation and libraries were prepared as previously described (18). normalization For whole-exome sequencing, genomic DNA was extracted and Raw reads (HTseqcounts) were normalized using FPKM as 3 mg from each sample was sheared and used for the construction implemented in the R package edgeR, and the FPKM values were of a paired-end sequencing library as described in the protocol transformed to log2 values and adjusted to the median-centered provided by Illumina. Enrichment of exonic sequences was then gene expression values by subtracting the row-wise median from performed for each library using the SureSelect Human All Exon the expression values in each row (Cluster 3.0). The centered and 50Mb Kit (Agilent Technologies) following the manufacturer's log2-transformed VSD and FPKM expression were used to illus- instructions. trate gene expression pattern in a heat map. Libraries for RNA and whole-exome sequencing were sequenced with Illumina TruSeq SBS Kit v3 on a HiSeq 2000 GSEA and network analysis sequencer (Illumina Inc.) to obtain 100-bp paired-end reads. The GO analysis of the gene expression data was performed using image analysis and base calling were performed using the Illu- GSEA (v2.24) desktop tools (permutation ¼ 1,000) and visual- mina pipeline (v1.8) with default settings. ized by the Enrichment Map tool in Cytoscape [P 0.05, false discovery rate (FDR) q value 0.1, and similarity 0.5; ref. 25]. RNA-seq analysis To characterize the LUSC transcriptome profile in cancer and Tumor microenvironment analysis noncancer control cells, we performed RNA-seq for 101 LUSC and The fractions of stromal and immune cells in tumor samples matched noncancer control samples. Total RNA extracted from were estimated by Estimation of STromal and Immune cells in lung specimens and depleted of ribosomal RNA was sequenced at MAlignant Tumours using Expression data (ESTIMATE) scores, the desired depth (100) on RNA-Seq (Illumina HiSeq). The with predictions of tumor purity based on the absolute method reads were aligned to the human genome (version GRCh37) with previously reported (26). The abundance of six infiltrating þ þ the Spliced Transcripts Alignment to a Reference (STAR) align- immune cell types (B cells, CD4 T cells, CD8 T cells,

OF2 Cancer Immunol Res; 2018 Cancer Immunology Research

Genomic Landscape and Microenvironmental Immune Signature

X neutrophils, macrophages, and dendritic cells) in the two sub- Cell-cycle-related SCNV ¼ Copy-number change at copy-number types of LUSC was estimated with the Tumor IMmune Estimation Resource (TIMER) algorithm (27). region of genes used in cell-cycle score

Immune score, stromal score, tumor purity, and cell-cycle score Immune score, stromal score, and tumor purity were made using X Immune-related SCNV ¼ Copy-number change at copy-number ESTIMATE. The gene set for cell-cycle score was used to calculate the cell-cycle score from the one in Davoli and colleagues (28). region of genes used in immune score

Statistical analyses Prediction of neoantigens Quantitative data are presented as mean standard deviation. We predicted neoantigens using the pVAC-Seq pipeline (36). We used R-3.2.3 to perform the statistical analyses. The normality of We used nonsynonymous mutations to follow the pVAC-seq the variables was tested by the Shapiro–Wilk normality test (29). For pipeline. Amino acid changes and transcript sequences were two groups, significance (P value) for normally distributed variables annotated by variant effect predictor. Epitopes predicted by was determined by an unpaired Student t test, and nonnormally HLAminer and were filtered by RNA expression (FPKM>1) and distributed variables were analyzed by the Mann–Whitney U test. coverage (tumor coverage >10 and noncancer control coverage For more than two groups, Kruskal–Wallis and one-way ANOVA >5). tests were used for the nonparametric and parametric methods, respectively (30). Statistically significant differences were tested at P Availability of data and material values < 0.05. R value was computed by Pearson and distance Whole-exome and RNA sequencing data are available under correlation. Survival graph was plotted by the Kaplan–Meier meth- the NCBI Sequence Read Archive (SRA) study accession no. od, and comparison between the subtypes was analyzed with the SRP114315. log-rank test, and Cox proportional hazards model was also used for survival analysis (SPSS for Windows, version 22; SPSS Inc.). Results Identification of LUSC subtypes Somatic mutation detection using whole-exome sequencing In this study, 101 LUSC and matched noncancer control fi To nd somatic mutations, we performed whole-exome samples were used to discover significant differential gene expres- sequencing (100 ). Reads were aligned to the NCBI human sion, and principal component analysis (PCA) on entire samples reference genome (hg19) using BWA (31). Picard was applied was performed to identify distinctive clusters based on gene to mark duplicates and we used Genome Analysis Toolkit (GATK variability between LUSC and noncancer control samples. PCA Indel Realigner) to improve alignment accuracy. Somatic single- with the top 1,000 most variable genes and unsupervised hier- nucleotide variants (SNV) from 101 LUSC samples with matched archical clustering with the k-means algorithm were applied to noncancer control samples were called using MuTect (32). GATK's RNA-seq data and distinguished noncancer control from tumor Haplotype Caller was also used for indel detection. All variants samples. In contrast, 19 tumor samples overlapped and were were annotated with information from several databases using more closely associated with the noncancer control group with ANNOVAR (33). 95% confidence interval ellipsoids (Fig. 1A; Supplementary Fig. S1). Additional PCA with 101 tumor samples distinguished the 19 Somatic copy-number variant detection tumor samples (subtype B) from 82 other major tumor samples To detect somatic copy-number variants from whole-exome (subtype A) at the 95% confidence interval. Thus, PCA revealed sequencing data, EXCAVATOR analysis was applied (34). GISTIC two molecular subtypes, A and B. These two subtypes were also fi analysis was used to identify recurrent ampli cation and deletion distinguishable in The Cancer Genome Atlas (TCGA) LUSC peaks (35). cohort (n ¼ 431) by PCA and unsupervised hierarchical clustering at 95% confidence interval (Supplementary Fig. S2). Calculation of SCNV levels To identify the pattern of genomic alterations in LUSC, we fi We obtained ampli ed and deleted copy-number variants sequenced the exome of independent noncancer control samples using EXCAVATOR and GISTIC analyses. Arm and focal SCNV (n ¼ 101) and tumor samples (n ¼ 101). The number of total levels of each patient were respectively calculated by summing the mutations in subtype B was less than that in subtype A (P < 0.001 copy-number changes at each copy-number event. Arm, chromo- by Mann–Whitney U test; Fig. 1B; Supplementary Fig. S3A). some, and focal CNV level were normalized to the mean and Mutations in genes encoding TP53, NAV3, CDH10, KMT2D, standard deviation among the samples (28). X NFE2L2, CTNNA3, KEAP1, NTRK3, RB1, NOTCH1, PTEN, FGFR2 Total SCNV ¼ Copy-number change at total and EGFR were also identified in the current study (Fig. 1C). Mutational signature combinations in subtype A were significant- copy-number region ly different from those in subtype B (P < 0.01 by Mann–Whitney X U test) showing more irregular patterns [Fig. 1D; Supplementary ¼ Arm-level SCNV Copy-number change at arm Fig. S3D, (32)]. We compared the burden of somatic copy-num- > copy-number region ber variation (SCNV; ref. 28). Arm-level SCNVs (length 98% of a chromosome arm), focal-level SCNVs (length < 98% of a X chromosome arm), and total SCNVs (all of a chromosome), Focal-level SCNV ¼ Copy-number change at focal which comprise the burden of SCNV, were mostly found in copy-number region subtype A (Fig. 1E and F). To elucidate the effect of genomic alteration on the tumor microenvironment, we also investigated

www.aacrjournals.org Cancer Immunol Res; 2018 OF3

Seo et al.

Figure 1. Identification of the subtypes of LUSC. A, Principal component analysis (PCA) of the top 1,000 most variable genes among the subtype A, B, and noncancer control were performed across all samples. 3D PCA scores plots of top 1,000 of most variable genes was drawn as meshes containing cancer and noncancer control points (left) and subtype A and B points (right) based on K-means clustering (k-means ¼ 2) on the first three PCs with 95% confidence interval ellipsoids. B, Ratio of somatic synonymous and nonsynonymous mutations in mutations per megabase and subtypes were classified. C, Name of significantly mutated genes (left), distribution of mutations across 101 LUSC, and frequency of significantly mutated genes (right) were plotted. D, The mutational signature revealed by somatic mutations in whole-exome sequencing. E, Arm-level CNV, focal-level CNV, and total level CNV of individual sample displayed in each column. F, Significant, focally amplified (red), and deleted (blue) regions are plotted across chromosomal locations.

OF4 Cancer Immunol Res; 2018 Cancer Immunology Research

Genomic Landscape and Microenvironmental Immune Signature

major histocompatibility complex (MHC) and tumor-specific 10 6 by unpaired Student t test; Supplementary Fig. S4). antigen (TSA) level, and found that subtype A carried more Cytolytic activity is a metric of immune-mediated cell destruction, deletions of MHC regions than did subtype B (Supplementary and immune infiltration predicts more cytolytic activity in sub- Fig. S3C). TSA level was also higher in subtype A than in subtype B type B of LUSC (37). Similarly, TCGA LUSC cohort results (P < 0.001 by Mann–Whitney U test; Supplementary Fig. S3D). identified different microenvironmental factors in the two sub- Our transcriptome analysis thus defined two LUSC subtypes, A types. Subtype B (n ¼ 328) in the TCGA LUSC cohort was and B, with different patterns of somatic mutations. consistent with the description of our LUSC subtype B (n ¼ 19; higher immune and stromal scores, lower tumor purity, and Identification of differentially expressed genes in subtypes higher CYT score; Supplementary Fig. S5). The TCGA cohort We analyzed differential expression of 4,807 genes in subtype A contained more subtype B than subtype A samples; the Korean and 1,471 genes in subtype B compared with noncancer control and TCGA cohorts likely had different subtype compositions. expression, and matched the significance criteria applied in this Immune cells are important in tumors. Stromal cells influence study (DESeq2: false discovery rate (FDR) < 0.1, P < 0.05). We tumor proliferation and inflammation. In order to evaluate derived a heat map depicting 5,387 differentially expressed genes tumor-associated stroma, the effects of activated and normal in subtype A and B by excluding the overlapped genes (Supple- stroma on two tumor subtypes were described by 50 stromal mentary Table S2). Differentially expressed genes were signifi- signature genes as previously reported (38). The heat map of cantly enriched in several pathways through GO term enrichment LUSC and noncancer control cells using active and normal stroma analysis. Genes for which expression was upregulated were main- genes indicated three groups: activated stromal-rich samples ly involved with the cell-cycle and DNA replication (n ¼ 1,826) in (subtype A), normal and activated stromal-rich samples (subtype subtype A and with immune and defense pathways (n ¼ 1,876) in B), and normal stromal-rich samples (noncancer control). Mean subtype B (Fig. 2A). normalized gene expression of the 25 exemplar activated and Further functional gene enrichment analysis using GSEA normal stromal genes was higher in subtype B than in subtype A (v2.2.3; ref. 25) proved that these differentially expressed genes (activated stroma P ¼ 0.0247; normal stroma P < 0.01 by Mann– were enriched in pathways related to cytoskeleton, mitosis, cell Whitney U test: Fig. 3A; Supplementary Figs. S6A and S6B). The cycle, and chromatin modification in subtype A and pathways activated stromal signature genes were associated with carcinoma- related to defense response, immune response, cytokine, and associated fibroblasts (CAF) and were overexpressed in subtype B metabolic and biosynthetic process in subtype B. Using network as compared with subtype A and noncancer control. Likewise, the analysis with Cytoscape (P 0.05, FDR q-value 0.1, and normal stromal genes, which reflect the composition of immune similarity 0.5), we found that the increase in expression and cells, were more highly expressed in subtype B than in subtype A. activation of pathway-related genes in subtype B was linked to Overexpression of growth factors and chemokines related to regulation of immune cell differentiation and apoptotic process- CAFs, such as HGF, FGF7, CXCL12, MMP2, IL6, CCL2, NFkB, es, which would increase immune cell abundance. Genes char- mediated tumor promotion and aggressive invasion in subtype B acterizing subtype A, on the other hand, were linked to cell-cycle (Supplementary Fig. S6C; ref. 39). and microtubule pathways that would enhance cancer cell pro- To analyze tumor microenvironment integration in subtypes liferation (Fig. 2B). Consistent with GSEA, network analysis A and B, we assessed the correlation of stromal and immune showed that cell cycle and immune system gene sets were differ- scores as well as tumor purity. There was a positive correlation entially correlated in each subtype. Both immune response and between stromal and immune scores (subtype A, Pearson r, cell-cycle pathways were differently upregulated in the subtypes, 0.79; distance r, 0.78; subtype B Pearson r,0.46;distancer, showing an interaction between immune response and somatic 0.54); samples with high tumor purity showed low stromal and mutations. immune scores (Fig. 3B). Subtypes A and B differed in their association with stromal and immune scores, suggesting vari- The immune landscape of the microenvironment in LUSC ation in microenvironments and in interactions with stromal To identify the roles of immunity and distribution of infiltrat- and immune cells. ing immune cells in tumor and noncancer control samples, we We estimated (with use of the TIMER algorithm) the abun- computed stromal and immune scores along with tumor purity dance of tumor-infiltrating immune cells in six cell types (B cells, þ þ based on the ESTIMATE method (26). The stromal and immune CD4 T cells, CD8 T cells, neutrophils, macrophages, and scores in subtype B compared with subtype A were significantly dendritic cells) to predict immune cell profiling in subtypes A increased (stromal core P ¼ 9.10 10 25; immune score P ¼ 7.51 and B. Noncancer control A and noncancer control B subtypes had 10 19 by unpaired Student t test; Fig. 3A; Supplementary Fig. no significant differences. Dendritic cells were the most abundant S4). The high immune scores suggested that recruitment of cells among both tumor and noncancer control cells. Although immune cells was more enhanced in subtype B than in A, which macrophages were not the most abundant cells in tumors and implicates not only immune cells surrounding the tumor but also noncancer control samples, macrophages seemed to be the most immune cells infiltrating the tumor (26). This observation may influential in a subtype-specific manner (P ¼ 6.27 10 9 by reflect the fact that the different portion of infiltrating immune Mann–Whitney U test; Figs. 3C and D). The proportion of and stromal cells was intermixed in a dissecting tumor, and this macrophages in subtype B was significantly higher than in subtype can be useful for predicting the degree of tumor purity and tumor A, supporting the previous observations that macrophages are microenvironment (tumor purity P ¼ 9.80 10 20 by unpaired present in large numbers in the tumor microenvironment and Student t test; Supplementary Fig. S4). As subtype B had more promote tumor progression and metastasis (40). tumor-infiltrating immune cells than did subtype A, cytolytic Intratumoral immune coordination was assessed by analyzing activity of subtype B must have been higher. Indeed, the cytolytic the correlation between selected immune cell markers. Pairwise activity score was higher in subtype B than in subtype A (P ¼ 2.07 comparisons of immune cell abundance levels were done by

www.aacrjournals.org Cancer Immunol Res; 2018 OF5

Seo et al.

Figure 2. Molecular subtypes at transcriptomic expression level. A, Heat map depicted 5,387 differentially expressed genes in subtypes A and B, and two distinct clustered genes in differentially expressed genes were selected. Top 10 GO gene sets in either clusters were determined based on the rank of enrichment log10(P value) of pathway and the matched significance criteria (P < 0.05 and FDR q value < 0.1) B, Network visualization based on gene enrichment analysis. Nodes represent subtype A (red) and B (blue) networks, and green edges represent genetic overlap between networks using GO enrichment analysis (permutation ¼ 1,000). The size of each node reflects the number of genes included in the network. Genes in significant networks were annotated and grouped with simplified GO terms. Networks meeting the cutoff conditions detailed at the bottom of the figure (right) were visualized with the Enrichment Map plugin for Cytoscape (P 0.05, FDR q value 0.1, and similarity 0.5).

OF6 Cancer Immunol Res; 2018 Cancer Immunology Research

Genomic Landscape and Microenvironmental Immune Signature

Figure 3. The immune landscape of the microenvironment in LUSC. A, The heat map depicted the expression of stromal genes and the tumor microenvironmental factor [cytolytic activity (CYT), purity, immune, stromal], which divided into three groups, describing activated stroma-rich samples (subtype A), normal and activated stroma-rich samples (subtype B), and normal stroma-rich samples (noncancer control) gene expression. B, Scatterplots between stromal and immune scores with tumor purity gradient were shown, and correlation coefficient was indicated by each of the subtypes. The color grading corresponds to the tumor purity, indexed as shown on the color bar at the bottom right of the figure. C, The abundance of infiltrating B cells, CD4þ T cells, CD8þ T cells, neutrophil, macrophages, and dendritic cells in two subtypes was estimated, and each P value was indicated by each of the subtypes (Mann–Whitney U test and Kruskal–Wallis test). Box represents the median (thick line) and the quartiles (line). D, The median-centered and log2-transformed expression level (log2fpkmþ1) of M1 and M2 signature genes in subtypes and noncancer control was box plotted with corresponding Mann–Whitney U and Kruskal–Wallis test results. Box represents the median (thick line) and the quartiles (line).

www.aacrjournals.org Cancer Immunol Res; 2018 OF7

Seo et al.

Figure 4. Impacts of genomic alterations on tumor microenvironments in LUSC. A, The level of total CNVs in cell-cycle–related gene sets is displayed across classes (left), and the level of total CNVs in immune-related gene sets is displayed across classes (right). B, The association between genomic alteration load (the number of mutations, the level of CNV) and cell-cycle score is shown (left) and the association between genomic alteration load and immune score is shown (right). C, The correlations between genomic alterations (CNVs, mutations) and immune-related properties were displayed. D, The cell-cycle scores are shown in each sample, and the high score densities and low score densities of cell-cycle score are plotted on the x and y axes. E, The immune scores are shown in each sample, and the high score densities and low score densities of immune score are plotted on the x and y axes. F, The correlation between CNV level (arm, focal) and cell-cycle score or immune score is displayed across subtype.

measuring Pearson correlation coefficients (r). The relationships expression of macrophage 2 signature genes in subtype B, a implied by these correlations were visualized as r values (Sup- previously validated gene set (43), suggests that tumor-associated plementary Fig. S7). Correlation between immune cells was macrophages (TAM) promote tumor growth in subtype B cancer greater in tumor cells than in noncancer control cells, and the (Fig. 3D; Supplementary Figs. S8A and S8B). We examined the degree of correlation was higher in subtype B than in subtype A. reproducibility of the immune parameters by subtype in both our Dendritic cells were more correlated with other immune cells in LUSC and TCGA LUSC cohorts. All Immune scores, activated subtype B than in subtype A, and the correlations of dendritic cells stromal gene expression, and macrophage 2 activity patterns by þ with neutrophil and neutrophil with CD8 T cells were increased subtype were highly reproducible across both cohorts. LUSC and in subtype B. Dendritic cells and neutrophils seemed to have TCGA LUSC cohort subtype B showed higher immune infiltration þ immediate connection with CD8 T cells, confirming previous (Supplementary Fig. S9). The immune microenvironment in findings that the cross-talk between neutrophils and recruited subtype B was converted to escape mode during the immunoedit- dendritic cells accelerates antigen presentation to T cells and ing process of LUSC in order to help proliferation of resistant generates an antigen-specific immune response (41, 42). Elevated clones in an immunocompetent host. This complexity of subtype-

OF8 Cancer Immunol Res; 2018 Cancer Immunology Research

Genomic Landscape and Microenvironmental Immune Signature

Figure 5. SCNV in LUSC immune pathway. The diagrams show the genes with SCNV in the four immune-related pathways and the percentage of samples with SCNV in immune-related pathway across classes; copy-number deletion (blue), copy-number amplification (pink), the percentage of samples with SCNV in subtype A (red box), the percentage of samples with SCNV in subtype B (blue box) specific microenvironments should inform therapeutic strategies subtype A. Focal-level SCNVs contributed more to immune scores for treatment of LUSC. than arm-level SCNVs (Fig. 4D and E). In an association test, focal- level SCNVs were more highly correlated with immune score (Fig. Impacts of somatic copy-number variants on tumor 4F). These findings indicate that focal-level SCNVs are more microenvironment influential on the tumor microenvironment and might be better To identify the effect of genomic alterations on tumor micro- targets for immunotherapy. Focal-level SCNVs had a stronger environments in LUSC, we investigated the association between correlation with LUSC immunity and may make it easier to somatic genomic alterations and the immune score. The SCNVs determine the target genes for drug interventions than arm-level were significantly higher in subtype A (P < 0.001 by Mann– SCNVs, because arm-level SCNVs are large genomic defects that Whitney test; Fig. 4A) and were negatively correlated with the may affect multiple targets (44, 45). To assess the role of focal- immune score (Pearson coefficient: 0.58) and with other level SCNV in the tumor microenvironment, we investigated immune-related properties such as the stromal score (stromal focal-level SCNVs in immune-related pathways via KEGG enrich- þ cell infiltration), CYT score (immune cytolytic activity), CD4 ment analysis. Subtype A had a high prevalence of SCNVs on T-cell infiltration (Pearson coefficient: 0.42), macrophage infil- immune-related pathways (B-cell receptor signaling pathway, tration (Pearson coefficient: 0.39), and dendritic cell infiltration chemokine signaling pathway, T-cell receptor signaling pathway, (Pearson coefficient: 0.41; Fig. 4B and C). These results sug- and Toll-like receptor signaling pathway of KEGG; Fig. 5). Genes gested that the negative correlation between immune score and harboring SCNV deletions, which lead to loss of function, were SCNVs in subtype A may influence the immune response of enriched in subtype A only for immune-related pathways of

www.aacrjournals.org Cancer Immunol Res; 2018 OF9

Seo et al.

Figure 6. The expression pattern of immune checkpoints in LUSC. A, The expression of immune checkpoint genes was analyzed between subtypes in our LUSC samples

(n ¼ 101) and TCGA LUSC cohort (n ¼ 431), respectively. The comparison of median-centered and log2-transformed expression (log2fpkm) of immune checkpoint genes was performed between subtypes, and P value was indicated by Mann–Whitney U or unpaired Student t test based on the normality. A two-color

scale was used, with blue indicating low expression values and red representing highly expressed genes. B, The median-centered and log2-transformed expression level (log2fpkm) of immune checkpoint genes in both cohorts was box plotted with corresponding Kruskal–Wallis test results. Box represents the median (thick line) and the quartiles (line).

immune system processes, immune responses, and response to dation gene set with previously studied subtypes (Supplemen- cytokine (Supplementary Fig. S10). These data suggested that tary Fig. S11; ref. 46). The subtype A expression pattern resembled focal-level SCNVs may drive the low immune response in that of the classical subtype, in which TP63, AKR1C3, FOXE1, and tumormicroenvironmentsandwouldbegoodtargetstocon- TXN genes were over-expressed. Subtype B appeared to over- trol immunity. express basal related genes (S100A7, MMP13, and SERPINB3) By analyzing the tumor microenvironment with RNA sequenc- and secretory related genes (ARHGDIB and TNFRSF14). Except ing and whole-exome sequencing, we deﬁned two immune- for the secretory-related genes, the other genes were all over- related subtypes: subtype A (immune defective) and subtype B expressed in subtype A relative to subtype B. The expression (immune competent). The log2 VSD expression pattern between pattern of previous validated genes differed between subtypes. tumors and noncancer control tissue was consistent over the vali- Thus, our clustering method with gene selection for subtyping

OF10 Cancer Immunol Res; 2018 Cancer Immunology Research

Genomic Landscape and Microenvironmental Immune Signature

achieved similar results to those of the previous study. Our immune regions than did subtype B. However, these results did not explain subtyping approach categorized the LUSC into immune-defective the role of the tumor microenvironment. Further study is needed and -competent subtypes based on the pattern of infiltrating to validate the results. In particular TSA content was predicted but immune, stromal cells, immune cell composition. Our data also not validated in the clinical environment. suggest that genomic alterations, especially SCNVs, decrease Our genetic results thus provide evidence for an immune immune-related activities and immune cells in cell proliferation– response to cancer in humans and indicate a mechanism of related subtypes of LUSC, and that activated stroma and TAM lead tumor-intrinsic resistance to cytolytic activity in tumors with a to the evolution of cancer cells in the immune-related subtype. high burden of somatic mutations. Analysis of genomic altera- Although the subtypes differed in immune activation, mutation tions and their impact on the tumor microenvironment in a burden, and SCNV, the subtypes had no significant differences in subtype-specific manner might identify patients who could ben- clinical features or overall survival in both cohorts (our LUSC efit from cancer immunotherapies that boost the immune system. cohort P value ¼ 0.223; TCGA LUSC cohort P value ¼ 0.54; log- rank test; Supplementary Figs. S12 and S13). It was difficult to Disclosure of Potential Conflicts of Interest evaluate to what extent clinical features were affected by low No potential conflicts of interest were disclosed. immune activities due to the high SCNV in subtype A and the increased activity of TAMs and activated stroma in subtype B. Authors' Contributions We found that all immune checkpoint genes were more Conception and design: J.-S. Seo, A. Kim, J.-Y. Shin, Y.T. Kim highly expressed in subtype B than subtype A in both our Development of methodology: A. Kim, J.-Y. Shin, J.-Y. Yun, J. Kim Acquisition of data (provided animals, acquired and managed patients, LUSC samples and the TCGA LUSC cohort (Fig. 6A), and two provided facilities, etc.): J.-Y. Shin, Y.H. Kim, S. Park, H.J. Lee, I.-K. Park, independent cohorts had similar expression patterns for C.-H. Kang, J.-Y. Yun, J. Kim, Y.T. Kim immune checkpoints (Fig. 6B). Both PD-1 and its ligand PD-L1 Analysis and interpretation of data (e.g., statistical analysis, biostatistics, as well as other immune checkpoint genes are highly expressed computational analysis): J.W. Lee, A. Kim, J.-Y. Shin in subtype B samples. PD-1 and PD-L1 are involved in immune Writing, review, and/or revision of the manuscript: J.W. Lee, A. Kim, J.-Y. Shin, tolerance by preventing stimulation of an immune response Y.T. Kim Administrative, technical, or material support (i.e., reporting or organizing and inducing tumor immune escape (47) and might be valu- fi data, constructing databases): J.-S. Seo, J.W. Lee, A. Kim, Y.H. Kim, Y.T. Kim able targets for new drugs in subtype B. However, dif culty Study supervision: J.-S. Seo, C.-H. Kang, Y.T. Kim with inducibility of PD-L1 protein expression and with accurate assays complicates their use. More reliable biomarkers are Acknowledgments needed (48). This work has been supported by Macrogen, Inc. (grant no. MGR17-01) and the National Research Foundation of Korea (NRF) grant funded by the Discussion Korea government (MSIP; No. NRF-2014R1A2A2A05003665).

Although the study of PD1 expression for immunotherapy has The costs of publication of this article were defrayed in part by the produced benefit in many clinical cases, some patients do not payment of page charges. This article must therefore be hereby marked respond to checkpoint blockade (49). For such nonresponders, advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate tumor-specific antigens with appropriate MHC-binding affinity this fact. might be more useful for immunotherapy (36, 50, 51). We analyzed the MHC region and TSA levels for both subtypes and Received August 19, 2017; revised December 18, 2017; accepted April 25, found that subtype A had more TSA and more deletions of MHC 2018; published first May 2, 2018.

References 1. Park JY, Jang SH. Epidemiology of lung cancer in Korea: recent trends. 10. Li H, Fan X, Houghton J. Tumor microenvironment: the role of the tumor Tuberc Respir Dis (Seoul) 2016;79:58–69. stroma in cancer. J Cell Biochem 2007;101:805–15. 2. Lee CT. Epidemiology of lung cancer in Korea. Cancer Res Treat 2002;34: 11. Vannucci L. Stroma as an active player in the development of the tumor 3–5. microenvironment. Cancer Microenviron 2015;8:159–66. 3. Kim Y, Hammerman PS, Kim J, Yoon JA, Lee Y, Sun JM, et al. Integrative and 12. Ramamonjisoa N, Ackerstaff E. Characterization of the tumor microenvi- comparative genomic analysis of lung squamous cell carcinomas in East ronment and tumor-stroma interaction by non-invasive preclinical imag- Asian patients. J Clin Oncol 2014;32:121–8. ing. Front Oncol 2017;7:3. 4. Yates LR, Campbell PJ. Evolution of the cancer genome. Nat Rev Genet 13. Siniard RC, Harada S. Immunogenomics: using genomics to personalize 2012;13:795–806. cancer immunotherapy. Virchows Arch 2017;471:209–19. 5. Gajewski TF, Schreiber H, Fu YX. Innate and adaptive immune cells in the 14. Becht E, de Reynies A, Giraldo NA, Pilati C, Buttard B, Lacroix L, et al. tumor microenvironment. Nat Immunol 2013;14:1014–22. Immune and stromal classiﬁcation of colorectal cancer is associated with 6. Yang Y. Cancer immunotherapy: harnessing the immune system to battle molecular subtypes and relevant for precision immunotherapy. Clin cancer. J Clin Invest 2015;125:3335–7. Cancer Res 2016;22:4057–66. 7. Green S, Dawe DE, Banerji S. Immune signatures of non-small cell lung 15. Wang RF. A special issue on cancer immunotherapy. Cell Res 2017;27:1–2. cancer. J Thorac Oncol 2017;12:913–5. 16. Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, Makarov V, Havel JJ, 8. Schoenhals JE, Seyedin SN, Anderson C, Brooks ED, Li YR, Younes AI, et al. et al. Cancer immunology. Mutational landscape determines sensitivity Uncovering the immune tumor microenvironment in non-small cell lung to PD-1 blockade in non-small cell lung cancer. Science 2015;348: cancer to understand response rates to checkpoint blockade and radiation. 124–8. Transl Lung Cancer Res 2017;6:148–58. 17. Palmieri G, Colombino M, Cossu A, Marchetti A, Botti G, Ascierto PA. 9. Safonov A, Jiang T, Bianchini G, Gyorffy B, Karn T, Hatzis C, et al. Immune Genetic instability and increased mutational load: which diagnostic tool gene expression is associated with genomic aberrations in breast cancer. best direct patients with cancer to immunotherapy? J Transl Med Cancer Res 2017;77:3317–24. 2017;15:17.

www.aacrjournals.org Cancer Immunol Res; 2018 OF11

Seo et al.

18. Ju YS, Kim JI, Kim S, Hong D, Park H, Shin JY, et al. Extensive genomic and focal somatic copy-number alteration in human cancers. Genome Biol transcriptional diversity identified through massively parallel DNA and 2011;12:R41. RNA sequencing of eighteen Korean individuals. Nat Genet 2011;43: 36. Hundal J, Carreno BM, Petti AA, Linette GP, Griffith OL, Mardis ER, et al. 745–52. pVAC-Seq: A genome-guided in silico approach to identifying tumor 19. Soundararajan R, Stearns TM, Griswold AL, Mehta A, Czachor A, Fukumoto neoantigens. Genome Med 2016;8:11. J, et al. Detection of canonical A-to-G editing events at 30 UTRs and 37.RooneyMS,ShuklaSA,WuCJ,GetzG,HacohenN.Molecularand microRNA target sites in human lungs using next-generation sequencing. genetic properties of tumors associated with local immune cytolytic Oncotarget 2015;6:35726–36. activity. Cell 2015;160:48–61. 20. Bailey P, Chang DK, Nones K, Johns AL, Patch AM, Gingras MC, et al. 38. Moffitt RA, Marayati R, Flate EL, Volmar KE, Loeza SG, Hoadley KA, et al. Genomic analyses identify molecular subtypes of pancreatic cancer. Nature Virtual microdissection identifies distinct tumor- and stroma-specific sub- 2016;531:47–52. types of pancreatic ductal adenocarcinoma. Nat Genet 2015;47:1168–78. 21. Inamura K, Fujiwara T, Hoshida Y, Isagawa T, Jones MH, Virtanen C, et al. 39. Banat GA, Tretyn A, Pullamsetti SS, Wilhelm J, Weigert A, Olesch C, et al. Two subclasses of lung squamous cell carcinoma with different gene Immune and inflammatory cell composition of human lung cancer expression profiles and prognosis identified by hierarchical clustering and stroma. PLoS One 2015;10:e0139073. non-negative matrix factorization. Oncogene 2005;24:7105–13. 40. Pollard JW. Tumour-educated macrophages promote tumour progression 22. Liu Q, Markatou M. Evaluation of methods in removing batch effects on and metastasis. Nat Rev Cancer 2004;4:71–8. RNA-seq data. Infect Dis Transl Med 2016;2:3–9. 41. Schuster S, Hurrell B, Tacchini-Cottier F. Crosstalk between neutrophils 23. Love MI, Huber W, Anders S. Moderated estimation of fold change and and dendritic cells: a context-dependent process. J Leukoc Biol 2013;94: dispersion for RNA-seq data with DESeq2. Genome Biol 2014;15:550. 671–5. 24. de Hoon MJ, Imoto S, Nolan J, Miyano S. Open source clustering software. 42. Tate MD, Brooks AG, Reading PC, Mintern JD. Neutrophils sustain effective Bioinformatics 2004;20:1453–4. CD8(þ) T-cell responses in the respiratory tract following influenza 25. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, infection. Immunol Cell Biol 2012;90:197–205. et al. Gene set enrichment analysis: a knowledge-based approach for 43. Shaykhiev R, Krause A, Salit J, Strulovici-Barel Y, Harvey BG, O'Connor TP, interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A et al. Smoking-dependent reprogramming of alveolar macrophage polar- 2005;102:15545–50. ization: implication for pathogenesis of chronic obstructive pulmonary 26. Yoshihara K, Shahmoradgoli M, Martinez E, Vegesna R, Kim H, Torres- disease. J Immunol 2009;183:2867–83. Garcia W, et al. Inferring tumour purity and stromal and immune cell 44. Beroukhim R, Mermel CH, Porter D, Wei G, Raychaudhuri S, Donovan J, admixture from expression data. Nat Commun 2013;4:2612. et al. The landscape of somatic copy-number alteration across human 27. Li B, Severson E, Pignon JC, Zhao H, Li T, Novak J, et al. Comprehensive cancers. Nature 2010;463:899–905. analyses of tumor immunity: implications for cancer immunotherapy. 45. Weir BA, Woo MS, Getz G, Perner S, Ding L, Beroukhim R, et al. Charac- Genome Biol 2016;17:174. terizing the cancer genome in lung adenocarcinoma. Nature 2007;450: 28. Davoli T, Uno H, Wooten EC, Elledge SJ. Tumor aneuploidy correlates with 893–8. markers of immune evasion and with reduced response to immunother- 46. Wilkerson MD, Yin X, Hoadley KA, Liu Y, Hayward MC, Cabanski CR, et al. apy. Science 2017;355:. Lung squamous cell carcinoma mRNA expression subtypes are reproduc- 29. Ghasemi A, Zahediasl S. Normality tests for statistical analysis: a guide for ible, clinically important, and correspond to normal cell types. Clin Cancer non-statisticians. Int J Endocrinol Metab 2012;10:486–9. Res 2010;16:4864–75. 30. Hazra A, Gogtay N. Biostatistics series module 3: comparing groups: 47. He J, Hu Y, Hu M, Li B. Development of PD-1/PD-L1 pathway in tumor numerical variables. Indian J Dermatol 2016;61:251–60. immune microenvironment and treatment for non-small cell lung cancer. 31. Li H, Durbin R. Fast and accurate long-read alignment with Burrows- Sci Rep 2015;5:13110. Wheeler transform. Bioinformatics 2010;26:589–95. 48. Tsiatas M, Mountzios G, Curigliano G. Future perspectives in cancer 32. do Valle IF, Giampieri E, Simonetti G, Padella A, Manfrini M, Ferrari A, et al. immunotherapy. Ann Transl Med 2016;4:273. Optimized pipeline of MuTect and GATK tools to improve the detection of 49. Roh W, Chen PL, Reuben A, Spencer CN, Prieto PA, Miller JP, et al. somatic single nucleotide polymorphisms in whole-exome sequencing Integrated molecular analysis of tumor biopsies on sequential CTLA-4 data. BMC Bioinformatics 2016;17:341. and PD-1 blockade reveals markers of response and resistance. Sci Transl 33. Wang K, Li M, Hakonarson H. ANNOVAR: functional annotation of genetic Med 2017;9:eaah3560. variants from high-throughput sequencing data. Nucleic Acids Res 2010; 50. Matsushita H, Vesely MD, Koboldt DC, Rickert CG, Uppaluri R, Magrini VJ, 38:e164. et al. Cancer exome analysis reveals a T-cell-dependent mechanism of 34. Xu D, Olman V, Wang L, Xu Y. EXCAVATOR: a computer program for cancer immunoediting. Nature 2012;482:400–4. efficiently mining gene expression data. Nucleic Acids Res 2003;31:5582–9. 51. Gubin MM, Zhang X, Schuster H, Caron E, Ward JP, Noguchi T, et al. 35. Mermel CH, Schumacher SE, Hill B, Meyerson ML, Beroukhim R, Getz G. Checkpoint blockade cancer immunotherapy targets tumour-specific GISTIC2.0 facilitates sensitive and confident localization of the targets of mutant antigens. Nature 2014;515:577–81.

OF12 Cancer Immunol Res; 2018 Cancer Immunology Research

Whole Exome and Transcriptome Analyses Integrated with Microenvironmental Immune Signatures of Lung Squamous Cell Carcinoma

Jeong-Sun Seo, Ji Won Lee, Ahreum Kim, et al.

Cancer Immunol Res Published OnlineFirst May 2, 2018.

Updated version Access the most recent version of this article at: doi:10.1158/2326-6066.CIR-17-0453

Supplementary Access the most recent supplemental material at: Material http://cancerimmunolres.aacrjournals.org/content/suppl/2018/05/02/2326-6066.CIR-17-0453 .DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerimmunolres.aacrjournals.org/content/early/2018/06/06/2326-6066.CIR-17-0453. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.