- Record: found
- Abstract: found
- Article: found
A costimulatory molecule-related signature in regard to evaluation of prognosis and immune features for clear cell renal cell carcinoma
research-article
Author(s):
Xiaoliang Hua
1
,
2
,
3 ,
Shengdong Ge
1
,
2
,
3 ,
Jiong Zhang
1
,
2
,
3 ,
Haibing Xiao
1
,
2
,
3 ,
Sheng Tai
1
,
2
,
3 ,
Cheng Yang
1
,
2
,
3
,
,
Li Zhang
1
,
2
,
3
,
,
Chaozhao Liang
1
,
2
,
3
,
4
,
Publication date (Electronic):
18 September 2021
Journal:
Cell Death Discovery
Publisher:
Nature Publishing Group UK
Read this article at
There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.
Abstract
Costimulatory molecules have been proven to enhance antitumor immune responses, but their roles in clear cell renal cell carcinoma (ccRCC) remain unexplored. In this study, we aimed to explore the gene expression profiles of costimulatory molecule genes in ccRCC and construct a prognostic signature to improve treatment decision-making and clinical outcomes. We performed the first comprehensive analysis of costimulatory molecules in patients with ccRCC and identified 13 costimulatory molecule genes with prognostic values and diagnostic values. Consensus clustering analysis based on these 13 costimulatory molecular genes showed different distribution patterns and prognostic differences for the two clusters identified. Then, a costimulatory molecule-related signature was constructed based on these 13 costimulatory molecular genes, and validated in an external dataset, showing good performance for predicting a patient’s prognosis. The signature was an independent risk factor for ccRCC patients and was significantly correlated with patients’ clinical factors, which could be used as a complement for clinical factors. In addition, the signature was associated with the tumor immune microenvironment and the response to immunotherapy. Patients identified as high-risk based on our signature exhibited a high mutation frequency, a high level of immune cell infiltration, and an immunosuppressive microenvironment. High-risk patients tended to have high cytolytic activity scores and immunophenoscore of CTLA4 and PD1/PD-L1/PD-L2 blocker than low-risk patients, suggesting these patients may be more suitable for immunotherapy. Therefore, our signature could provide clinicians with prognosis predictions and help guide treatment for ccRCC patients.
Related collections
Most cited references46
- Record: found
- Abstract: found
- Article: not found
Cancer statistics, 2020
Rebecca Siegel, Kimberly D Miller, Ahmedin Jemal (2020)
Each year, the American Cancer Society estimates the numbers of new cancer cases and deaths that will occur in the United States and compiles the most recent data on population-based cancer occurrence. Incidence data (through 2016) were collected by the Surveillance, Epidemiology, and End Results Program; the National Program of Cancer Registries; and the North American Association of Central Cancer Registries. Mortality data (through 2017) were collected by the National Center for Health Statistics. In 2020, 1,806,590 new cancer cases and 606,520 cancer deaths are projected to occur in the United States. The cancer death rate rose until 1991, then fell continuously through 2017, resulting in an overall decline of 29% that translates into an estimated 2.9 million fewer cancer deaths than would have occurred if peak rates had persisted. This progress is driven by long-term declines in death rates for the 4 leading cancers (lung, colorectal, breast, prostate); however, over the past decade (2008-2017), reductions slowed for female breast and colorectal cancers, and halted for prostate cancer. In contrast, declines accelerated for lung cancer, from 3% annually during 2008 through 2013 to 5% during 2013 through 2017 in men and from 2% to almost 4% in women, spurring the largest ever single-year drop in overall cancer mortality of 2.2% from 2016 to 2017. Yet lung cancer still caused more deaths in 2017 than breast, prostate, colorectal, and brain cancers combined. Recent mortality declines were also dramatic for melanoma of the skin in the wake of US Food and Drug Administration approval of new therapies for metastatic disease, escalating to 7% annually during 2013 through 2017 from 1% during 2006 through 2010 in men and women aged 50 to 64 years and from 2% to 3% in those aged 20 to 49 years; annual declines of 5% to 6% in individuals aged 65 years and older are particularly striking because rates in this age group were increasing prior to 2013. It is also notable that long-term rapid increases in liver cancer mortality have attenuated in women and stabilized in men. In summary, slowing momentum for some cancers amenable to early detection is juxtaposed with notable gains for other common cancers.
- Record: found
- Abstract: found
- Article: not found
Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources.
Da Huang, Brad T. Sherman, Richard A. Lempicki (2009)
DAVID bioinformatics resources consists of an integrated biological knowledgebase and analytic tools aimed at systematically extracting biological meaning from large gene/protein lists. This protocol explains how to use DAVID, a high-throughput and integrated data-mining environment, to analyze gene lists derived from high-throughput genomic experiments. The procedure first requires uploading a gene list containing any number of common gene identifiers followed by analysis using one or more text and pathway-mining tools such as gene functional classification, functional annotation chart or clustering and functional annotation table. By following this protocol, investigators are able to gain an in-depth understanding of the biological themes in lists of genes that are enriched in genome-scale studies.
- Record: found
- Abstract: found
- Article: found
Inferring tumour purity and stromal and immune cell admixture from expression data
Malignant solid tumour tissues consist of not only tumour cells but also tumour-associated normal epithelial and stromal cells, immune cells and vascular cells. Stromal cells are thought to have important roles in tumour growth, disease progression1 2 and drug resistance3. Infiltrating immune cells act in a context-dependent manner, and whereas antitumor effects of infiltrating T-lymphocytes have been observed in ovarian cancer4 5 6, associations with tumour growth, invasion and metastasis were described in colorectal cancer7 8. The comprehensive understanding of tumour-associated normal cells in tumour tissues may provide important insights into tumour biology and aid in the development of robust prognostic and predictive models. Gene expression profiling of cancer has resulted in the identification of molecular subtypes and the development of models for prediction prognosis and has enriched our knowledge of the molecular pathways of tumorigenesis9 10 11 12 13. Increasing evidence suggests that the infiltration of tumour-associated normal cells influences the analysis of clinical tumour samples by genomic approaches, such as gene expression profiles or copy number data, and biological interpretation of the results requires considerable attention to sample heterogeneity14 15 16. Several methods have been proposed to estimate the fraction of tumour cells in clinical tumour samples by using DNA copy number array data14 15 or by using next-generation sequencing data17. DNA copy number-based estimation of tumour purity is rapidly gaining traction in predicting the purity of tumour samples; however, such methods are limited to samples with available copy number profiles. Previous studies have attempted to deconvolve gene expression data into gene expression profiles from their constituent cellular fractions, whereas others have focused on deconvolution of microarray data obtained from normal tissue into cell-type-specific profiles, by calculating enrichment scores18 19 20 21 22. These methods take advantage of the differences in transcriptome properties of distinct cell types. Here we present a new algorithm that takes advantage of the unique properties of the transcriptional profiles of cancer samples to infer tumour cellularity as well as the different infiltrating normal cells, called ESTIMATE (Estimation of STromal and Immune cells in MAlignant Tumour tissues using Expression data). We focus on stromal and immune cells that form the major non-tumour constituents of tumour samples and identify specific signatures related to the infiltration of stromal and immune cells in tumour tissues1. By performing single-sample gene set-enrichment analysis (ssGSEA)13 23, we calculate stromal and immune scores to predict the level of infiltrating stromal and immune cells and these form the basis for the ESTIMATE score to infer tumour purity in tumour tissue. Finally, we describe the biological characteristics of stromal and immune scores in The Cancer Genome Atlas (TCGA) data sets24 25 26 27 28 29. Results Estimation of infiltrating cells and tumour purity An overview of ESTIMATE algorithm is shown in Fig. 1. We devised two gene signatures: (1) a ‘stromal signature’ that was designed to capture the presence of stroma in tumour tissue, and (2) an ‘immune signature’ that aimed to represent the infiltration of immune cells in tumour tissue (Supplementary Data 1). To generate these signatures, we performed the following steps (Fig. 1). Genes associated with the quantity of infiltrating immune cells in tumour tissue were identified using leukocyte methylation scores, which were previously shown to correlate with the presence of leukocytes in ovarian carcinomas15. Gene expression profiles of normal hematopoietic samples were compared with those of other normal cell types. The overlap between the two gene sets constituted the immune signature. Stromal-related genes were selected among non-hematopoiesis genes by comparison of the tumour cell fraction and matched stromal cell fraction after laser-capture microdissection in breast, colorectal and ovarian cancer data sets30 31 32. Genes with high variability in cancer cell lines and genes highly expressed in glioma stem-like cells were filtered to make up the stromal signature. We used single-sample ssGSEA13 23 of these two signatures to generate scores that reflect the presence of each cell type in tumour samples and combined represent a measurement of tumour purity. In order to evaluate the reliability of the stromal and the immune signatures, we obtained three ovarian carcinoma tumour samples and performed microbead-based cell sorting to separate tumour and non-tumour cell fractions. The epithelial, tumour cell-containing, cell fraction was enriched using an EpCAM antibody. Transcriptional profiles were obtained from the bulk tumour samples as well as the EpCAM-positive and EpCAM-negative cell fractions. Although tumour cells may not necessarily express EpCAM and some normal epithelial cells may express EpCAM33, a significant reduction in stromal signature scores (paired t-test, P=0.0042) and a declining trend in immune signature scores (paired t-test, P=0.072) were observed in all three EpCAM-positive cell fractions compared with the EpCAM-negative cell fractions, suggesting that these signatures are associated with the amount of non-epithelial cells in tumour samples (Fig. 2a). In the three data sets used in the process of gene selection, there was a significant reduction in the stromal and immune scores in the tumour cell fraction (Fig. 2b; Supplementary Fig. S1). Similarly, the microdissected stroma-enriched fraction in the three independent public data sets, which were not used in construction of the gene signature was significantly decreased (ovarian cancer (GSE29156), P=2.5 × 10−5; breast cancer (GSE10797), P=1.9 × 10−7; lung cancer (GSE33363), P=5.7 × 10−7 by paired t-test; Fig. 2c). Although immune scores in the tumour cell-enriched fraction were lower than those in bulk tumour- or stroma-enriched fraction (ovarian cancer, P=0.0030; breast cancer P=3.2 × 10−7; lung cancer P=0.0044 by paired t-test; Fig. 2d), one tumour-enriched sample retained a high immune score (Fig. 2b), suggesting that immune cells were retained in the microdissected tumour cell-enriched fraction. This observation may reflect the challenges in microdissecting tumour and immune cells that intermix in many tumours. It could also be related to differences between infiltrating immune cells and immune cells surrounding the tumour4 5 6. To evaluate the association of the stromal and immune scores with tumour purity, we compared ESTIMATE scores with predictions of tumour purity based on the ABSOLUTE method15. ABSOLUTE establishes the fraction of tumour cells in a tumour sample based on somatic DNA copy number alterations and has been shown to provide highly accurate prediction of tumour purity. Immune and stromal signature scores of TCGA Agilent array-based expression profiles of ovarian cancer (n=417; 28 samples used to define the immune signature were not included in this analysis) showed a significant correlation of both stromal and immune scores with ABSOLUTE tumour purity predictions (Pearson’s correlation coefficient or r, −0.65 and −0.60; distance r, 0.65 and 0.58) (Fig. 3a,b). Importantly, ESTIMATE scores showed an increased correlation with tumour purity compared with stromal-only and immune-only scores (Pearson’s r, −0.69; distance r, 0.69) (Fig. 3c). There was a positive correlation between stromal and immune scores (Pearson’s r, 0.62; distance r, 0.58), and samples with low tumour purity showed high stromal and immune scores (Fig. 3d). Specific samples were associated with high stromal but not high immune scores, and vice versa, suggesting variable infiltrating patterns (Supplementary Data 2). To illustrate the broad utility of the ESTIMATE algorithm, we applied this model to 10 TCGA tumour types for which both DNA copy number and gene expression data sets were available, profiled on four different platforms (Table 1)24 26 27 28 29. These 10 tumour types were among the first cancers to be characterized by TCGA and were included in TCGA’s Pan-Cancer effort. To confirm the accuracy of the ESTIMATE algorithm, receiver operating characteristic (ROC) curve analysis34 using ABSOLUTE-based tumour purity was performed. Tumour samples were divided into high- and low-purity groups based on several cutoff values of ABSOLUTE-based tumour purity (0.9, 0.8, 0.7 and 0.6), and the area under the ROC curve (AUC) for each cutoff was measured. For example, a cutoff of 0.7 for tumour purity resulted in Agilent-based ESTIMATE score AUC of 0.89 in the TCGA ovarian cancer data set used as the training set (Fig. 3f). Next, we applied the ROC analysis to other data sets by using the same procedure. Similar AUC values were observed across different expression platforms as well as different tumour types (Fig. 4a; Supplementary Figs S2–S6). Immune cells not only infiltrate the tumour cell region but have also been demonstrated to associate with stromal cells, in a cancer-type-specific manner4. The correlation between stromal and immune scores varied across cancer types, ranging from high (GBM, Pearson’s r=0.8) to modest (KIRC, Pearson’s r=0.38; Fig. 3d; Supplementary Fig. S7). This suggests that the stromal and immune signatures do not measure the same phenotype and reflects the variable association between immune cells and tumour stroma across cancers. Pathology-based estimates of the percentage of tumour cells, stromal cells and infiltrating lymphocytes, evaluated from hematoxylin-eosin-stained slides, were less correlated with ESTIMATE, stromal and immune scores (Fig. 5). Prediction of tumour purity using ESTIMATE In order to facilitate tumour purity prediction using ESTIMATE signatures, we transformed the scoring system to a [0,1] range. First, a regression curve for ESTIMATE score and tumour purity based on ABSOLUTE in the TCGA data set was established. By applying the nonlinear least squares method to the modified TCGA Affymetrix data (n=995) (Supplementary Fig. S8a), ESTIMATE-based tumour purity prediction model was developed. There was a high correlation between ESTIMATE-based and DNA copy number-based tumour purity (Pearson’s r=0.74) (Supplementary Fig. S8b). Validating the capacity of ESTIMATE to predict tumour purity was performed using an independent data set (n=195) composed of seven publicly available data sets including both Affymetrix microarray expression data and matched SNP array copy number data (Supplementary Table S1). Moreover, ESTIMATE-based tumour purities were highly correlated with the ABSOLUTE-based tumour purities in the independent validation set (Pearson’s r=0.87) (Fig. 4b; Supplementary Fig. S8c). When four cutoff values (ABSOLUTE-based tumour purity of 0.9, 0.8, 0.7 and 0.6) were applied, the average and standard deviation of the accuracy per cutoff was 0.87±0.050 (Supplementary Table S2). ESTIMATE provided tumour purity predictions in individual samples with a 95% confidence interval of the validity of the prediction (Fig. 4c). To show the specificity of the tumour purity prediction, we used copy number and expression data from 27 cancer cell lines samples (GSE34211). The root-mean-square error of ESTIMATE and ABSOLUTE were 0.006 and 0.051, respectively, indicating consistent absence of immune and stromal signals (Supplementary Fig. S9). Next, we calculated ESTIMATE scores using the expression profiles from 10 normal ovarian epithelium samples (GSE18520). The ESTIMATE-predicted tumour purity was 0.68±0.12 (Supplementary Table S3), suggesting that normal ovarian epithelium may have some stromal or immune cell components. In addition, to clarify whether alteration of gene expression levels related to cell adhesion, migration or wound-healing processes that occur within tumour cells would affect our stromal, immune and ESTIMATE scores, we used public microarray data (GSE17708) from 26 lung adenocarcinoma cell lines treated or untreated by transforming growth factor beta 1. Although our stromal scores slightly increased, the estimated tumour purity was unaffected (Supplementary Fig. S10). We investigated the correlation of the stromal, immune and ESTIMATE scores with methylation-based estimates of the fraction of leukocytes in tumour tissues15. A high correlation between our immune score and leukocyte methylation score was observed across all tumour types (Pearson’s r=0.75±0.091) (Supplementary Fig. S11). Interestingly, stromal scores were not strongly correlated with leukocyte methylation score (Pearson’s r=0.51±0.089). These findings showed that our immune scores were specifically associated with the presence of leukocytes across different tumour types. Patterns of stromal and immune cell scores across different tumour types Using both TCGA and non-TCGA data sets from 10 different tumour types (Supplementary Table S1), we examined the distribution of stromal and immune score per tumour type (Fig. 6; Supplementary Fig. S12, Supplementary Table S4). As reported previously, lung adenocarcinomas showed lower purity compared with other tumour types15. The relatively high levels of stroma found in clear cell renal cell carcinoma and breast carcinoma may be associated with the high levels of adipocyte content that is characteristic of both tumour types35 36. In high-grade serous ovarian carcinoma, high stromal or immune scores reflect the presence of mesenchymal or immunoreactive gene expression subtypes that have been reported previously30 37. Clear cell renal cell carcinomas are considered to be immunogenic tumours, and this characteristic is captured by the relatively high levels of immune signature expression38. Immunogenicity is not known as a property of lung squamous cell carcinoma; however, this disease is characterized by a high percentage (>95%) of patients with a history of smoking, which has been linked to lung inflammation39 40. Lung squamous cell carcinomas showed relatively high immune cell scores and have recently been associated with susceptibility to immunomodulatory therapeutics such as ipilimumab40. Further investigation is needed to show that the presence of infiltrating immune cells is a biomarker for immunotherapy response. The similarity in the distribution of stromal and immune scores between lung squamous cell carcinoma and head and neck squamous cell carcinoma suggests that these tumours may harbour a similar genomic profile but also share comparable tumour cellularities28. The impact of tumour purity on somatic mutations To examine the impact of tumour purity on the ability to detect genetic alterations, we assigned samples with ESTIMATE scores in the top 25% to a low-purity subgroup, and samples with the bottom 25% ESTIMATE scores to a high-purity subgroup, per tumour type. We observed a reduced number of mutations per megabase in low-purity head and neck squamous cell carcinomas and clear cell renal cell carcinomas, (unpaired t-test with Benjamini–Hochberg FDR correction, adjusted P=0.055 and 0.055) but not in other tumour types, suggesting that the sequencing coverage used for TCGA samples is sufficient to comprehensively detect somatic sequence variants (Supplementary Fig. S13). Next, we evaluated the mutation spectrum of high- and low-purity subgroups by measuring the relative contribution of the two types of transition base substitution (A>G/G>A and T>C/C>T) and the four classes of transversion base substitutions (C>A/A>C, C>G/G>C, T>A/A>T and T>G/G>T). Two of the ten TCGA data sets (head and neck squamous cell carcinoma, lung squamous cell carcinoma) showed a significantly decreased fraction of T>A substitutions in the low-purity group compared with the high-purity group (unpaired t-test with Benjamini–Hochberg FDR correction, adjusted P=0.015 and 0.015, respectively) (Supplementary Table S5). The ratio of transitions and transversions was significantly associated with purity level in head and neck squamous cell carcinoma (adjusted P=0.018). Discussion We have developed a new algorithm to infer the level of infiltrating stromal and immune cells in tumour tissues and tumour purity using gene expression data. The predictive ability of this method has been validated in large and independent data sets. Genomic, transcriptomic and proteomic analyses using clinical tumour tissue are affected by the fraction of tumour cells present, and methods for evaluation of the non-tumour portions of tumour samples could provide an important context to genomic data analysis15. ESTIMATE scores were significantly correlated with the tumour purity of clinical cancer samples as well as cancer cell line samples and provide an accessible and straightforward approach to obtain a measure of the amount of tumour cells in a biological sample. The ESTIMATE algorithm may be further optimized by including signature of endothelial cells and tumour-type-specific normal epithelial cells. Tumour purity of clinical tumour samples is routinely determined by pathologists through visual evaluation of hematoxylin- and eosin-stained slides. In this study, histological estimates of the percentage of tumour cells, stromal cells and infiltrating lymphocytes did not correlate well with ESTIMATE, stromal and immune scores, consistent with the weak correlation between DNA copy number-based tumour purity and histological tumour purity15. This discrepancy between genomic- or transcriptomic-based and pathology-based estimates might be affected by the sensitivity of histopathological examination to interobserver bias and variability in accuracy15 41 or the difference in tissue sections42 in the same sample between nucleic acid extraction and histological evaluation. The contribution of immune cells to ovarian carcinoma is well recognized5 6, and we chose to use the TCGA ovarian carcinoma samples as the basis for development of the immune signature, as four types of principal information were available: tumour tissue for cell-sorting experiments, estimates of the amount of desmoplasia, immunohistochemistry-based counts of the number of leukocytes and methylation leukocyte scores. Importantly, the performance of ESTIMATE in both TCGA and non-TCGA ovarian carcinoma data sets was not distinctively better compared with other tumour types, and we thus believe that the method used to develop the signature is not biased towards ovarian cancers. The fibroblast/mesenchymal nature of stromal cells separates their gene expression profile from that of the epithelial tumour cells, thus providing a rationale to seek a signature that is characteristic of stromal cells in general, despite the notion that stromal cells may be tumour-type-specific. As expression data sets from three cancer types (ovary, breast and colon) were used to compare tumour cell fractions and matched stromal cell fractions after laser-capture microdissection, we suggest that some of the diversity in tumour-associated stroma among various cancer types was captured. Importantly, the ESTIMATE accuracy among ovarian, breast and colon cancer TCGA samples was not notably better than that of other tumour types, suggesting that the stromal signature can be broadly applied. The dependency of ESTIMATE on infiltrating stromal and immune cells resulted in some limitations, such as the inability to accurately infer tumour cellularity of hematopoietic or stromal tumours (for example, leukaemia, sarcoma and gastrointestinal stromal tumours) because of the high and tumour-intrinsic expression of stromal- or immune-related genes. Owing to the lack of data, we were unable to evaluate ESTIMATE in the context of tumour types such as prostate or pancreas cancer that may present with atypical patterns of tumour-associated cells—that is, increased fractions of normal epithelial cells. Additional methods may be needed to predict cancer cell fractions for such malignancies. The diverse pattern of the presence of stroma and immune cells across tumour types further emphasizes the different context-dependent ways in which tumour-associated normal cells function and more broadly illustrates the impact of the tumour microenvironment on tumorigenesis and homeostasis. Epithelial-to-mesenchymal transitions in tumour cells have been frequently described43. It is possibility that some overlap exists between the stromal expression signature and a mesenchymal tumour cell phenotype. However, the strong correlation with tumour purity may suggest that epithelial-to-mesenchymal transition is often confused with the increased presence of tumour-associated stroma. Low tumour purities may reduce the sensitivity of somatic mutation detection44. We did not observe an association of tumour purity with mutation rates except in head and neck squamous cell carcinomas and clear cell renal cell carcinoma, suggesting that the impact of tumour purity to identify somatic mutations is less compared with other factors such as depth or coverage or the mutation detection algorithm applied. We noted differences in mutational profile and spectrum between high and low stromal/immune subgroups in several tumour types. The consistent reduction in T>A substitutions in some low-purity cases suggests that the tumour microenvironment can have an impact on mutational processes or alternatively that the types of mutations in the tumour can alter stromal and immune infiltrations. Our ESTIMATE method for the assessment of stromal and immune cells in tumour tissues may provide an additional avenue to increase our understanding of molecular phenotype. Our results show that the levels of stromal and immune cells in tumour tissue can be associated with clinical characteristics. Further refinement of the lineage characteristics of infiltrating cells, such as distinguishing between various types of leukocytes, may reveal a more consistent pattern of clinical associations than what we have currently described. Novel therapeutics such as ipilumimab and nivolumab alters T-lymphocyte checkpoint control and may be particularly effective in tumours with intrinsically high levels of infiltrating leukocytes. Whether ESTIMATE immune scores could serve as a biomarker for immunotherapy response is a topic for further investigation. The ESTIMATE method can be applied for assessment of the presence of stromal cells and the infiltration of immune cells in tumour samples using gene expression data. The method is publicly available through the SourceForge software repository ( https://sourceforge.net/projects/estimateproject/). The application of ESTIMATE to publicly available microarray expression data sets, as well as new microarray or RNA-seq-based transcriptome profiles, may help in elucidating the facilitating roles of the microenvironment to neoplastic cell and provide new insights into context in which genomic alterations occur. Methods Data preparation TCGA level 3 gene expression levels were obtained from the TCGA Data Portal45 in March 2013. In this study, we used 10 tumour types from four platforms: Affymetrix HT-HG-U133A (one-colour type—that is, one RNA sample is labelled with a fluorophore and hybridized to a microarray), Agilent G4502A (two-colour type—that is, one sample and one reference are labelled with different fluorophores and hybridized together on a same microarray), RNAseq (quantified as Reads Per Kilobase per Million mapped reads)46 and RNAseqV2 (quantified through RNA-seq by Expectation Maximization)47 (Table 1). The tumour types selected for our study were among the first tumour types analysed through TCGA and were selected as cancer types studied in TCGA’s Pan-Cancer project. In addition, we used 31 data sets of microarray expression or SNP array copy numbers from Gene Expression Omnibus48 and ArrayExpress49, glioblastoma expression data set from the Repository of Molecular Brain Neoplasia Data50, cancer cell line expression data set from Cancer Cell Line Encyclopedia (CCLE)51 and a glioma stem-like cell expression data set from researchers at MD Anderson Cancer Center (Supplementary Table S1). Microbead-based cell sorting First, the tissue of a fresh frozen ovarian cancer sample was diced into 1-mm pieces. The tissue was further enzymatically dissociated with 0.8 mg/ml HBSS Liberase Research Grade (#05-401-119-001; Roche) and incubated at 37 °C for 1 h, followed by mechanical dissociation using an 18-G needle. To isolate single cells, the resulting cell suspension was filtered using a 40-μm filter. Lastly, the remaining cells were separated into an epithelial tumour cell fraction and a non-epithelial tumour-associated stromal fraction. For cell sorting, we used antibody-coated microbeads that recognize the epithelial cell surface marker EpCAM (#130-061-101; Miltenyi Biotec), which results in an EpCAM-positive tumour cell fraction and an EpCAM-negative tumour-associated stromal cell fraction. To test the efficiency of our procedure we performed gene expression profiling on three bulk tumours, three EpCAM-positive fractions and three EpCAM-negative fractions after cell sorting using Illumina BeadChip Human HT-12 v4 according to the manufacturer’s instructions. This study was approved by the institutional ethics review board at The University of Texas MD Anderson Cancer Centre (Lab 07-0108). All patients provided written informed consent for the collection of samples and subsequent analysis. Microarray data processing Probes from Affymetrix HG-U133A, HG-U133Plus2.0 and HT_HG-U133A GeneChip platforms were mapped to a transcript database and combined in one probe set per gene, as described previously52. Expression levels from these Affymetrix data sets were individually established using RMA and quantile normalization53. Raw data from Affymetrix Human 133 × 3 P array were processed using the Bioconductor rma package with the default setting. On the Agilent G4112F platform, data normalization was carried out in GeneSpring GX 11.5 (Agilent Technologies) by setting the raw signal threshold to 1.0 and using 75th percentile normalization54. Quantile normalization was performed for Illumina Human HT-12 v4 microarray data using the Bioconductor preprocessCore package. On Affymetrix Human 133 × 3P array, Agilent G4112F and Illumina Human HT-12 v4 probes measuring the same gene were averaged to obtain one expression value per gene and sample. Gene selection A flowchart of gene selection in this study is shown in Supplementary Fig. S1. To analyse expression data measured from six different platforms, we extracted 10,412 common genes. In the gene selection process, we used the significance analysis of microarray55 method to detect differentially expressed genes (more than twofold and q A, C>G, C>T, T>A, T>C and T>G)66 67. We then calculated the relative contribution of each of the six classes of base substitutions and compared them between the two subgroups. Next, we extracted the respective high and low stromal/immune score subgroups based on the 75th and 25th percentiles of each score per tumour type and combined each subgroup’s expression data and mutation data. Statistical analysis We conduced all computations with R 2.13.2 (ref. 68) and used standard statistical tests as appropriate. Where appropriate, P-values were corrected for multiple testing using the Benjamini–Hochberg false discovery rate method69. Author contributions K.Y. and R.G.W.V. conceived and designed the present study. M.S. performed the experiments. K.Y., H.K. and R.G.W.V. analysed the data. K.Y., R.V., H.K., W.T.-G. and R.G.W.V. developed and coded the ESTIMATE algorithm. E.M., V.T., H.S., P.W.L., D.A.L., S.L.C., G.G., K.S.-H., G.B.M. and TCGA contributed data/materials/analysis tools. K.Y. and R.G.W.V. wrote the manuscript. All authors read and approved the final manuscript. Additional information How to cite this article: Yoshihara, K. et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat. Commun. 4:2612 doi: 10.1038/ncomms3612 (2013). Supplementary Material Supplementary Figures, Tables and References Supplementary Figures S1-S13, Supplementary Tables S1-S5 and Supplementary References Supplementary Data 1 A gene list of stromal and immune signatures. Supplementary Data 2 A list of stromal, immune, and ESTIMATE scores in TCGA data sets (RNAseqV2). Supplementary Note 1 List of The Cancer Genome Atlas Research Network contributors
Author and article information
Contributors
Cheng Yang:
ORCID: http://orcid.org/0000-0002-6663-3017
chengyang_md@163.com
Li Zhang:
ORCID: http://orcid.org/0000-0003-2599-952X
lzhang@ahmu.edu.cn
Chaozhao Liang:
ORCID: http://orcid.org/0000-0003-2317-1323
liang_chaozhao@ahmu.edu.cn
Journal
Journal ID (nlm-ta): Cell Death Discov
Journal ID (iso-abbrev): Cell Death Discov
Title:
Cell Death Discovery
Publisher:
Nature Publishing Group UK
(London
)
ISSN
(Electronic):
2058-7716
Publication date
(Electronic):
18
September
2021
Publication date PMC-release: 18
September
2021
Publication date Collection: 2021
Volume: 7
Electronic Location Identifier: 252
Affiliations
[1
]GRID grid.412679.f, ISNI 0000 0004 1771 3402, Department of Urology, , The First Affiliated Hospital of Anhui Medical University, ; Hefei, China
[2
]GRID grid.186775.a, ISNI 0000 0000 9490 772X, Anhui Province Key Laboratory of Genitourinary Diseases, , Anhui Medical University, ; Hefei, China
Author information
Xiaoliang Hua
http://orcid.org/0000-0001-5095-8872
Shengdong Ge
http://orcid.org/0000-0001-8187-9894
Cheng Yang
http://orcid.org/0000-0002-6663-3017
Li Zhang
http://orcid.org/0000-0003-2599-952X
Chaozhao Liang
http://orcid.org/0000-0003-2317-1323
Article
Publisher ID: 646DOI: 10.1038/s41420-021-00646-2
PMC ID: 8449780
PubMed ID: 34537809
SO-VID: af5bdedb-d1bb-41f8-95c4-19ce5b3dda17
Copyright © © The Author(s) 2021
License:
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
History
Date
received
: 8
May
2021
Date
revision received
: 17
August
2021
Date
accepted
: 7
September
2021
Categories
Subject:
Article
Custom metadata
issue-copyright-statement © The Author(s) 2021
Data availability: