Search tips
Search criteria 


Logo of oncoimmunLink to Publisher's site
Oncoimmunology. 2017; 6(1): e1260214.
Published online 2016 December 7. doi:  10.1080/2162402X.2016.1260214
PMCID: PMC5283648

Analysis of the prognostic role of an immune checkpoint score in resected non-small cell lung cancer patients


Tumors develop mechanisms to recruit tolerogenic immune cells and to induce the expression of molecules that act as immune checkpoints. This regulation of the immune microenvironment favors immune tolerance to the neoplastic cells. In this study, we have investigated the prognostic role of immune-checkpoint expression markers in a cohort of resectable non-small cell lung cancer (NSCLC) patients. RNA was isolated from fresh-frozen lung specimens (tumor and normal lung) (n = 178). RTqPCR was performed to analyze the relative expression of 20 immune-related genes that were normalized by the use of endogenous genes selected by GeNorm algorithm. Patients with higher expression levels of IL23A and LGALS2 presented better outcomes. In the clustering expression patterns, we observed that patients with higher expression of immunoregulatory genes had better survival rates. Additionally, these data were used to develop a gene expression score. Since CTLA4 and PD1 were associated with prognosis based on Cox regression analysis (Z-score > 1.5), a multivariate model including these two genes was created. Absolute regression coefficients from this analysis were used in order to calculate the immune-checkpoint score: (PD1×0.116) + (CTLA4×0.059) for each case. Kaplan–Meier survival analysis showed that patients with high immune-checkpoint score have longer overall survival (OS) [NR vs. 40.4 mo, p = 0.008] and longer relapse-free survival (RFS) [82.6 vs. 23 mo, p = 0.009]. Multivariate analysis in the entire cohort indicated that the immune-checkpoint score was an independent biomarker of prognosis for OS [HR: 0.308; 95% CI, 0.156–0.609; p = 0.001] and RFS [HR: 0.527; 95% CI, 0.298–0.933; p = 0.028] in early-stage NSCLC patients. In conclusion, this score provides relevant prognostic information for a better characterization of early stage NSCLS patients with strikingly different outcomes and who may be candidates for immune-based therapies.

KEYWORDS: Biomarker, checkpoint score, NSCLC, prognostic


Lung cancer is the most common cause of tumor-related death in the world. Approximately 85% are classified as non-small cell lung cancer (NSCLC) and at the time of diagnosis the majority of patients present locally advanced or metastatic disease.1 Improvements in the understanding of the mechanisms for lung cancer initiation, maintenance, and progression have led to the discovery of a variety of molecularly defined subsets of patients characterized by specific sets of driver mutations.2 Lung cancers are among the most mutated types of tumors,3 therefore, generating new antigens which play a key part in tumor immunity 4 and improved responses to immune-based therapies in NSCLC and other lung tumors.5-7

Tumor-infiltrating immune cells have been identified in many types of cancers,8 and although some of these cells are potentially capable of eliminating neoplastic cells, ultimately they cannot prevent tumor development and progression.9 In NSCLC, it has been shown that the presence, localization, and proportion of helper and specially cytotoxic infiltrating lymphocytes are associated with a favorable prognosis.10-13 But tumors acquire mechanisms to regulate their immune microenvironment such as the release of a series of factors to subvert normal reaction mechanisms as the modulation of co-stimulatory pathways, also known as immune checkpoints,14 and the induction and attraction of suppressor cells such as myeloid-derived suppressor cells,15,16 tumor-associated macrophages,17 and regulatory T cells.18,19 The clinical implications of these immunoregulatory elements in tumors are still controversial,20 and their functional or causal relationship between immunosuppressive pathways and immune cells in the tumor microenvironment has not yet been clearly defined in NSCLC.21-24 Therefore, the study of a great variety of immune-related markers, especially those implicated in immunoregulatory processes, could provide valuable prognostic information that could help in many applications in future NSCLC clinical practice. Additionally, and regarding laboratory assessments to determine these markers, it is of crucial importance to develop robust methodologies that are not biased by observer subjectivity, are reproducible and allow accurate and affordable large scale determinations. In this study with a cohort of surgically-resected NSCLC patients we have investigated the prognostic value of the quantification of gene expression levels of a large array of immunoregulatory molecules (mRNA by RTqPCR). In this paper, we demonstrate for the first time the prognostic relevance of an immune checkpoint score (ICS) based on the relative expression of two immune checkpoint molecules in resected NSCLC patients.


Patients characteristics

The most relevant demographic and clinicopathological characteristics including age, gender, stage of disease, and histology of the 178 NSCLC patients entered in the study are shown in Table 1. The median patient age was 65 y [range: 26–85], 86.5% were males, 47.2% had SCCs, and 59% of the patients were diagnosed at stage I of the disease. Moreover, 80 (45%) relapsed and 76 (42.7%) died during the follow-up. The median follow-up was of 81.23 mo [range: 1–113].

Table 1.
Clinicopathological characteristics of the patients included in the study.

Immune-related gene expression patterns and their association with survival

We measured the expression of 20 immune-related genes in primary lung tumor and paired noncancerous tissues (adjacent healthy lung tissue) using RTqPCR. Using this criteria, we found that FOXP3 (3.87X) and CD25 (2.66X) were overexpressed, whereas CD1C (0.42X), CD127 (0.40X), PD1 (0.38X), and CCL2 (0.25X) were downregulated in the tumor compared with normal-paired tissue. Unsupervised hierarchical clustering analysis was used to group patients based on the similarity of their expression patterns. Patients were classified into a cluster tree with two major subgroups according to the expression of genes related to conventional and regulatory T cells (CD4+, CD8+, CD127, FOXP3, CD25, CTLA4, PD1, and PDL1) and genes involved in different immunoregulatory processes (IL10, IL23A, CCL2, NRP1, LGALS1, LGALS2, CD1C, and CD209). Patients in Cluster I had lower expression levels of most of the genes analyzed, whereas Cluster II comprised patients with higher gene expression levels (Fig. 1).

Figure 1.
Hierarchical cluster based on selected gene expression. Patients in the original cohort were clustered into a hierarchical tree based on the expression of immune related genes. The clustering separated the patients into two distinct groups. Red indicates ...

Kaplan–Meier survival analysis using the two major clusters showed that patients in Cluster II had longer relapse-free survival (RFS) (81.2 vs. 26.2 mo, p = 0.027) and overall survival (OS) (not reached (NR) vs. 46.6 mo, p = 0.040) than patients in Cluster I (Fig. 2). We also analyzed their prognostic value according to histology, and observed that ADC patients classified in Cluster II had a significantly better RFS (81.2 vs. 17.8 mo, p = 0.005) and OS (NR vs. 42.9 mo, p = 0.034) than patients in Cluster I. Although it is of interest that the unsupervised cluster analysis of immune-related genes was able to identify a group of patients with a better prognosis, hierarchical clustering can only be applied retrospectively and cannot be used to predict a patient's future outcome. Therefore, we next investigated the prognostic value of genes analyzed individually or as small groups.

Figure 2.
Kaplan–Meier plots for OS and RFS according to the clustering classification of patients. (A) RFS and (B) OS. Solid line represents patients classified in Cluster I, whereas dashed line represents patients in Cluster II. p-values were calculated ...

Individual immune-related genes associated with survival

Kaplan–Meier analysis indicated that patients with high levels of IL23A presented better RFS (81.2 vs. 23.4 months, p = 0.003; Fig. S1A and B) and better OS (NR vs. 43.4 mo, p = 0.001). Another gene that correlated with better RFS (NR vs. 26.2 mo, p = 0.002) and OS (NR vs. 46.6, p = 0.007) was LGALS2, which encodes galectin-2 (Fig. S1C and D).

Survival analysis was also performed according to the patient histology. Kaplan–Meier test performed with ADC patients (including ADCs and adenosquamous histology) showed the same association between high IL23A and LGALS2 and better prognosis that were found in the entire cohort. Furthermore, the group of patients with high CTLA4 expression levels had better RFS (81.2 vs. 18.2 mo, p = 0.012, Fig. S2A and B) and OS (NR vs. 37 mo, p = 0.003). We also found that high levels of IL10 correlated with a higher RFS (49.3 vs. 18.8 mo, p = 0.029, Fig. S2C and D) and OS (81.2 vs. 37 mo, p = 0.030).

Immune checkpoint score (ICS) is a prognostic biomarker for RFS and OS in NSCLC

We also intended to create a gene expression score based on a multi-gene signature, which can provide more accurate predictions than a model using single genes.25,26 Univariate Cox regression analysis was performed considering OS as a dependent variable. Genes were ordered on the basis of their prognostic power (univariate Z-score, Fig. S3) and according to this ranking, the expression of two genes, PD1 and CTLA4 (both considered immune checkpoint molecules), were found to be associated with survival (Z-score >2), and therefore, were selected to construct a risk signature. We constructed a model based on the relative contribution of these two genes in the multivariate analysis (considering the absolute regression coefficients, see Table 2), and the resulting score was named ICS, with the following equation: (PD1×0.116) + (CTLA4×0.058).

Table 2.
Results from the multivariate Cox regression model for OS.

Kaplan–Meier analysis showed that patients with a high ICS (> median) had longer RFS (82.6 vs. 23 mo, p = 0.009; Fig. 3) and OS (NR vs. 40.4 mo, p = 0.008). We also performed a stratified analysis by TNM staging and histology. We found that for ADC patients, the association between high ICS and prognosis was stronger than for the entire cohort of patients (RFS: NR vs. 16.2 mo, p < 0.001 and OS: NR vs. 34.4 mo, p = 0.002). To evaluate the potential use of the ICS as an independent prognostic biomarker, a multivariate analysis was performed, where significant analytical and clinicopathological variables from the univariate analysis were entered in the study. The variables included were: lymph node involvement, KRAS status, cluster classification, FOXP3, CD4+, CD127, IL23, LGALS2 and the ICS. Results obtained from this multivariate analysis indicated that KRAS status and the ICS were independent biomarkers for both OS and RFS, and in the later, CD127 expression was also identified as an independent biomarker (see Table 3).

Figure 3.
Kaplan–Meier plots for OS and RFS according to the immune checkpoint expression score (ICS). The score was divided as low and high according to its median. Solid line represents patients with low levels of expression, whereas dashed line represents ...
Table 3.
Multivariate Cox regression model results, including all the significant variables.


Although most of the patients in early stage NSCLC (stages I–IIIA) are treated surgically with curative intent, the associated survival is less than optimal, with a 5-y survival rate ranging from 50% for stage IA to 15% for stage IIIA.27 Currently, there are still gaps in the approach used for selecting patient's adjuvant therapies based on the surgical or TNM stage alone. So, a great challenge in the management of patients with resected NSCLC is to develop new biomarkers that could help in identifying subjects at the greatest risk of recurrence and their potential response to specific treatments. Over the last decade, the field of tumor immunology has changed, and it is now accepted that the immune system plays a pivotal role in cancer. We are in the immunotherapy era not only for the efficacy of the new therapeutic armamentarium based in the blockage of immune checkpoint inhibitors but also for the relevance of the immune-derived prognostic and predictive biomarkers related to immunoregulatory processes. Thus, the studies of immune-related markers, especially those implicated in immunoregulatory processes, like our paper, could provide valuable prognostic information in resected NSCLCs that could help in future clinical practice.

Infiltrating immune cells of the acquired and innate immune response are organized in the lung tissue20,28 and their scoring of the type, density, and localization has demonstrated to be a prognostic factor in cancer, even as useful as the pathological characteristics.9,10 Additionally, in the tumor, cancer cells and other components of the microenvironment release chemokines and chemokine receptors29,30 that on one side regulate the migration of immune cells31,32 and on the other side act as molecules that promote the proliferation and migration of the neoplastic cells.33,34 A favorable environment for the growth of tumor cells implies the specific attraction of cells with known immunosuppressive properties as regulatory T cells (Tregs),35-37 and myeloid-derived suppressor cells (MDSCs).38-41 Moreover, cancer cells can produce cytokines with immunosuppressive functions as IL-1042 and TGF-β43 together with enzymes such as indoleamine 2,3-dioxygenase (IDO)44 that can impair metabolites needed for immune cells to thrive in the environment. But of special interest is the capability of the tumoral cells to express certain immune checkpoint receptors that promote tolerance of the immune system to the tumoral cells and lead the escape from immune surveillance. The maximum exponents of the checkpoint receptor are the cytotoxic T-lymphocyte-associated protein-4 (CTLA4),45 programmed cell death protein-1 (PD1), and its ligand PDL146,47. In our work, we have analyzed the expression levels by RTqPCR of a large variety of cellular and soluble immuneregulation-related biomarkers associated with components of the physiologic or pathophysiologic immune response.

In this study, gene expression immune-related biomarkers were assessed in fresh-frozen tumor and normal lung tissue samples from resected-NSCLC patients. An introductory analysis based on unsupervised clustering of the immune-related genes indicated that the group of patients with the highest expression levels of immune-related genes had better outcomes than the other group, although most of these genes were involved in immunoregulatory processes. This is not the first study to observe that patients with higher expression of genes related to immunoregulation have better survival rates.48 In fact, in breast cancer, a molecular signature obtained from microarray data analysis, which was associated with relapse-free patients, had a higher representation of genes involved in B-cell development and antigen presentation, but also of genes involved in T-cell apoptosis, CTLA4 signaling, or activation of IL23R, which are all pathways involved in the negative regulation of effector T cells.49 In a recent study, the expression of immunosuppressive factors, such as PD1, PDL1, CTLA4, and FOXP3 measured in 481 breast tumors, was highly significant predictors of therapy response and improved outcome.50 Our observations are consistent with the present idea that there are two different phenotypes regarding the tumoral infiltration of immune cells, in the sense that the most immune components present in the tumoral microenvironment, even if they are immunoregulatory, the better prognosis of the disease.51 By contrast those tumors that are not able to recruit immune cells had worst prognosis.48,52

Since hierarchical clustering can only be applied retrospectively and cannot be used to predict a patient's future outcome, we investigated the prognostic value of genes analyzed individually. This individual survival analysis revealed associations between markers like IL23A and LGALS2 with better outcomes. IL23 is considered the master switch in several T-cell-mediated inflammatory disorders, but the antitumor activity of IL23 is controversial. On the one hand, it has been shown that pro-inflammatory cytokines, including IL17A, IL6, and IL23 can impair CD8+ T-cell-mediated immune surveillance and promote tumor neovascularization.53 But on the other hand, other groups have reported that IL23 exerts antitumor activity by stimulating T cells and natural killer (NK) cells.54 Its prognostic value was studied in ovarian cancer, and an improved OS was observed in patients with high p19 mRNA expression (expressed by the IL23A gene).55 In lung cancer, a recent study in NSCLC tumor samples and cell lines reported that gemcitabine, a chemotherapy drug indicated for first-line treatment of NSCLC, induced IL23A expression and that it was found to induce NSCLC cell line proliferation. However, they failed to correlate IL23A expression with NSCLC patient prognosis.56 For LGALS2, in contrast to galectin-1 and galectin-3, relatively few studies have examined the expression of galectin-2 in animals and human tumors. Galectins are members of a highly conserved family of β-galactoside-binding lectins, which have a broad variety of functions including immune function regulation. The most extensively studied galectin function is their regulation of apoptosis. Furthermore, galectin-1 functions as a soluble mediator used by tumor cells to evade the immune response.57 Similar to our findings that lower LGALS2 expression is associated with a worse outcome, in gastric cancer, it has been reported that decreased galectin-2 expression is associated with LN involvement and advanced clinical stage.58

In order to get a deeper insight into the immune regulatory process, a score composed with the expression of immune checkpoint related genes (ICS) was constructed following a mathematical model.25,26 Expression scores are well established methods for separating patients into prognostic group and our score includes the expression levels and regression coefficients from the Cox analysis of CTLA4 and PD1, as follows, ICS = (PD1×0.116) + (CTLA4×0.058). When dichotomizing the cohort attending to the median level of the ICS, the Kaplan–Meier analysis revealed that a high ICS (ICS above the median) was associated with longer OS and RFS. The ICS in the Cox regression model, including all the significant variable, was demonstrated to be an independent prognostic biomarker, along with KRAS status. The ICS encompasses the expression of two genes, CTLA4 and PD1, which have become of great interest in the last few years. CTLA4 overexpression is more common in ADC and appears to be an independent prognostic factor in NSCLC.45 This is because researchers have demonstrated the importance of how the immune checkpoint blockade leads to robust antitumor effects in patients with metastatic melanoma, NSCLC, and other tumor types. In fact, targeted therapy with blocking antibodies to this immune checkpoint is one of the most promising therapeutics in many different tumors. So, regarding the prognostic implication of CTLA4 expression in NSCLC, there are contradictory and opposite results in the literature. It has been described that high expression of CTLA4, but not PD1 predicts worse survival in NSCLC46 and other malignancies like nasopharyngeal59 or esophageal carcinoma.60 Contrarily, other authors have found a reduced death rate in radically resected NSCLC overexpressing CTLA4.45

In our cohort, we have found that for ADC patients, the association between high ICS and prognosis was stronger than for the entire group of patients. This results are in concordance with recent reports that indicate that in squamous NSCLC treated with nivolumab, an anti-PD1 monoclonal antibody, the improvement in OS was independent of PDL1 expression,61 whereas in contrast for nonsquamous NSCLC this benefit was only observed in those patients with PDL1 expression.62 In a similar way, PD1 is also an immune checkpoint receptor with immunosuppressive properties. However, in contrast to CTLA4, PD1 is activated during the effector stages of T-cell activation, interaction with its ligand (PDL1) occurs primarily in peripheral tissues instead of lymph nodes, and importantly, it can be expressed in tumor tissue as well as in immune cells.47,63 The expression of immune checkpoints, PD1 and PDL1, in infiltrating immune cells by IHC was correlated with better responses to immune checkpoint blockade treatment, suggesting that the presence of these biomarkers might indicate that these tumors have already been recognized by the immune system, and therefore they are key predictors of clinical treatment responses.52,64 Additionally, in one publication the expression of PD1 and PDL1 in a cohort of 125 NSCLC has been assessed in tissues in order to evaluate if they were differently expressed according to the presence or absence of EGFR mutations, ALK translocation, or KRAS mutations. Although they observed that the sensitivity to treatment was higher and the OS was longer in patients treated with EGFR TKIs when PDL1 expression was higher, no differences were observed for PD147. As for PDL1, its expression in two large NSCLC patient cohorts has been analyzed, observing that high expression of PDL1 protein or mRNA was associated with a better outcome.24 Furthermore, tumor PDL1 expression was associated with improved OS in NSCLC with adjuvant therapy.65 Also, in NSCLC patients treated with EGFR inhibitors the response rate, time to progression and survival was higher in PDL1-positive vs. PDL1-negative patients, but with no difference in PD1-positive vs. PD1-negative groups.47 However, we failed to obtain this correlation in our data when the prognostic value of the markers was analyzed individually. This discrepancy could be explained by methodological differences as we performed quantitative PCR whereas Velcheti et al. used IHC and in situ hybridization. But the IHC determinations of PD1 and PDL1 at the protein levels in NSCLC are heterogeneous even with interassay discordances. These differences are caused by the antibodies used, with a range in specificity and affinity66 and tests that harmonize immune checkpoint determinations in NSCLC and other cancers are indeed needed.67 For this reason we selected for our study the use of RTqPCR, the gold standard method for gene expression quantification with high sensitivity and specificity and clinically applicable for detecting patient subgroups with specific prognostic characteristics. Other advantages of this technology are that it requires a low RNA input, it is less time consuming than other methods, and it is robust and flexible. This is of great importance because current clinicopathological staging methods have limited success in predicting patient survival and great outcome uncertainty for same-stage NSCLC cancers remains and today, we still cannot predict which patients will be cured and which ones will suffer recurrence or death after surgical resection. Very recently, an in silico study using mRNA data from The Cancer Genome Atlas (TCGA) from 11 tumor types has demonstrated that heterogeneous immune infiltrates are present and in general are linked to improved prognosis.68 If these predictive markers are a reflection of a pre-existing immune recognition, and taking the theory that predictive markers are also likely to be of prognostic value into account,69 our results suggest that immune checkpoint marker expression may also be of future value as a new prognostic NSCLC biomarker. Thus, the ICS may reflect a favorable immune context, in which the immune system recognizes the tumor. We propose that these results may also have some therapeutic value for managing NSCLC via emerging targeted immunotherapies, especially immune checkpoint blockade-based therapies, and so further studies to asses both of these uses should be conducted in order to better understand these processes.

Taken together, our results indicate the existence of two possible immune-scenarios in NSCLCs. In the first, the tumor is recognized by the immune system and a T-cell response is activated, which in turn activates also immunoregulatory pathways. In this case, patients had better outcomes. In the second scenario, which is associated with worse outcomes, the immune system does not recognize the tumor and there is no immune response activation; therefore, immunoregulatory pathway activation is not required. These results provide new insight into the tumor immunity field in NSCLC, and could be useful in the future development of prognostic and therapeutic tools.

Patients and methods

Patients and tissue samples

This retrospective study included 178 patients with resected NSCLC from the General University Hospital of Valencia who underwent surgery between 2004 and 2013 and who fit the eligibility criteria: resected, non-pretreated stage I–IIIA patients (according to the American Joint Committee on Cancer Staging manual) with a histological diagnosis of NSCLC. The study was conducted in accordance with the Declaration of Helsinki, and the institutional ethical review board approved the protocol. The most relevant demographic and clinicopathological characteristics of the cohort are shown in Table 1. Patient tumor and adjacent normal lung specimens were obtained at the time of surgery and were preserved in RNAlater (Applied Biosystems, USA) to avoid degradation of RNA. The samples were frozen at −80 °C until the analysis. REMARK recommendations on the studies of prognostic tumor markers found in tissues, blood, and other body fluids were followed.70 RFS was estimated as the time from surgery to recurrence or death from the disease, whereas OS was defined as the time from diagnosis to the date of death or last follow up. Additionally, K-RAS mutation status was assessed for the whole cohort using the theraScreen® KRAS Pyro® kit (Qiagen). This kit is used for quantitative detection of mutations in codons 12, 13, and 61 of the human KRAS gene by pyrosequencing.

Quantitative real-time PCR of immune-related genes

RNA from frozen tissue samples was extracted using standard TRIZOL (Invitrogen) methods. RNA was reverse-transcribed using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, USA) following the manufacturer's instructions with 1.0 µg of total RNA using random hexanucleotides. The thermal cycling conditions were as follows: 10 min at 25 °C, 120 min at 37 °C, and 5 s at 85 °C. RTqPCR was performed using Universal Master Mix and TaqMan Gene Expression Assay (Applied Biosystems, USA), to analyze the relative expression of 20 immune-related genes: Chemokine ligand 2 (CCL2; assay ID Hs00234140_m1), CCL22 (Hs99999075_m1), CD1C (Hs00233509_m1), CD127 (Hs00233682_m1), CD209 (Hs00253550_m1), CD25 (Hs00166229_m1), CD4+ (Hs00181217_m1), CD8+ (Hs00233520_m1), C-type lectin domain family 4, member C (CLEC4; Hs01092462_m1), cytotoxic T-lymphocyte-associated protein 4 (CTLA4; Hs01011591), forkhead box P3 (FOXP3; Hs00203958_m1); indolamine-1 (IDO1; Hs00984148_m1) interleukin 10 (IL10; Hs00961622_m1), IL23A (Hs00413259_m1), lectin galactoside-binding soluble 1 (LGALS1; Hs00355202_m1), LGALS2 (Hs00197810_m1), neuropilin 1 (NRP1; Hs00826125_m1), programmed cell death 1 (PD1; Hs01550088_m1), PDL1 (Hs01125301_m1), and transforming growth factor β 1 (TGFB1; Hs00171257_m1). Using GeNorm software, actin β (ACTB; Hs01060665_g1), glucuronidase β (GUSB; Hs01558067_m1), and cyclin-dependent kinase inhibitor (CDKN1B; Hs00153277_m1) were use as endogenous controls. The thermal cycling parameters were as follows: 2 min at 50 °C and 10 min at 95 °C, followed by 40 cycles of 15 s at 95 °C and 1 min at 60 °C. For efficiency calculations of each gene, we used the random-primed qPCR Human Reference cDNA (Clontech, USA). Relative gene expression levels were expressed as the ratio of target gene expression to reference gene (GUSB) expression by using the Pfaffl formula.71 It was considered a gene to be overexpressed when the median of the relative gene expression of the pathological area referred to the adjacent healthy tissue was higher than 2 and underexpressed when it was less than 0.5. Gene expression levels were dichotomized as “high” and “low” according to the median of each case.

Statistical analysis

Non-supervised hierarchical analysis was performed with Cluster software (version 3.0) and visualized with Tree View software version 1.0.6 which can be found at All analyses were performed on normalized and log2-transformed dataset values. Uncentered correlation was used as the similarity metric and average linkage was used as the clustering method. Continuous variables were compared by non-parametric Mann–Whitney U and Kruskall–Wallis tests. A Spearman rank test was used to test for correlations between continuous variables, and the association between dichotomized variables was evaluated by the Chi-square test. Survival analysis was performed using a univariate Kaplan–Meier (log-rank) test method with clinicopathological variables, dichotomized gene expression marker levels, and immune cell infiltration levels. Finally, to assess the independent value of the tested biomarkers, a Cox proportional hazard model for multivariate analyses was used. All significant variables from the univariate analyses were entered into the multivariate analyses in a forward stepwise Cox regression analysis. Furthermore, we also calculated gene expression scores based on multi-gene signatures using a method previously reported.25,26 Univariate Cox regression analysis on the training cohort was used to select genes associated with mortality (Z-score > 2) which were afterwards included in a multivariate risk model. All genes were included for these purposes, and expression values for all analyses are continuous variables. For multivariate Cox regression models, missing values for genes were replaced with the average value. A probability of 95% (p < 0.05) was considered statistically significant for all analyses. The statistical analyses were done using the Statistical Package for the Social Sciences (SPSS, Chicago, IL, USA) version 15.0.

Supplementary Material


Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.


This work was supported by the Red Temática de Investigación Cooperativa en Cáncer (RD12/0036/0025) and the Fondo de Investigación Sanitaria-Fondo Europeo de Desarrollo Regional (PI09/01147, PI09/01149 and PI12/02838).


1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2016. CA Cancer J Clin 2016; 66:7-30; PMID:26742998; [PubMed] [Cross Ref]
2. Morgensztern D, Campo MJ, Dahlberg SE, Doebele RC, Garon E, Gerber DE, Goldberg SB, Hammerman PS, Heist RS, Hensing T et al. Molecularly targeted therapies in non-small-cell lung cancer annual update 2014. J Thorac Oncol 2015; 10:S1-63; PMID:25535693; [PMC free article] [PubMed] [Cross Ref]
3. Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SAJR, Behjati S, Biankin A V, Bignell GR, Bolli N, Borg A, Børresen-Dale A-L et al. Signatures of mutational processes in human cancer. Nature 2013; 500:415-21; PMID:23945592; [PMC free article] [PubMed] [Cross Ref]
4. Schreiber RD, Old LJ, Smyth MJ. Cancer immunoediting: integrating immunity's roles in cancer suppression and promotion. Science 2011; 331:1565-70; PMID:21436444; [PubMed] [Cross Ref]
5. Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, Makarov V, Havel JJ, Lee W, Yuan J, Wong P, Ho TS et al. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer. Science 2015; 348:124-8; PMID:25765070; [PMC free article] [PubMed] [Cross Ref]
6. Snyder A, Makarov V, Merghoub T, Yuan J, Zaretsky JM, Desrichard A, Walsh LA, Postow MA, Wong P, Ho TS et al. Genetic basis for clinical response to CTLA-4 blockade in melanoma. N Engl J Med 2014; 2189-99; PMID:25409260; [PMC free article] [PubMed] [Cross Ref]
7. Gubin MM, Artyomov MN, Mardis ER, Schreiber RD. Tumor neoantigens: building a framework for personalized cancer immunotherapy. J Clin Invest 2015; 125:3413-21; PMID:26258412; [PMC free article] [PubMed] [Cross Ref]
8. Bremnes RM, Sirera R, Al-saad S, Camps C, Busund L. The role of tumor-infiltrating immune cells and chronic inflammation at the tumor site on cancer development, progression, and prognosis. J Thorac Oncol 2011; 6:1-10; PMID:21178711; [PubMed] [Cross Ref]
9. Fridman W-H, Dieu-Nosjean M-C, Pagès F, Cremer I, Damotte D, Sautès-Fridman C, Galon J. The immune microenvironment of human tumors: general significance and clinical impact. Cancer Microenviron 2013; 6:117-22; PMID:23108700; [PMC free article] [PubMed] [Cross Ref]
10. Donnem T, Hald SM, Paulsen E-E, Richardsen E, Al-Saad S, Kilvaer TK, Brustugun OT, Helland A, Lund-Iversen M, Poehl M et al. Stromal CD8+ T-cell density—a promising supplement to TNM staging in non-small cell lung cancer. Clin Cancer Res 2015; 21:2635-43; PMID:25680376; [PubMed] [Cross Ref]
11. Mattsson JSM, Bergman B, Grinberg M, Edlund K, Marincevic M, Jirström K, Pontén F, Hengstler JG, Rahnenführer J, Karlsson MG et al. Prognostic impact of COX-2 in non-small cell lung cancer: a comprehensive compartment-specific evaluation of tumor and stromal cell expression. Cancer Lett 2015; 356:837-45; PMID:25449785; [PubMed] [Cross Ref]
12. Schalper KA, Brown J, Carvajal-Hausdorf D, McLaughlin J, Velcheti V, Syrigos KN, Herbst RS, Rimm DL. Objective measurement and clinical significance of TILs in non-small cell lung cancer. J Natl Cancer Inst 2015; 107:dju435; PMID:25650315; [PMC free article] [PubMed] [Cross Ref]
13. Goc J, Germain C, Vo-Bourgais TKD, Lupo A, Klein C, Knockaert S, de Chaisemartin L, Ouakrim H, Becht E, Alifano M et al. Dendritic cells in tumor-associated tertiary lymphoid structures signal a Th1 cytotoxic immune contexture and license the positive prognostic value of infiltrating CD8+ T cells. Cancer Res 2014; 74:705-15; PMID:24366885; [PubMed] [Cross Ref]
14. Joyce JA, Fearon DT. T cell exclusion, immune privilege, and the tumor microenvironment. Science 2015; 348:74-80; PMID:25838376; [PubMed] [Cross Ref]
15. Vetsika E-K, Koinis F, Gioulbasani M, Aggouraki D, Koutoulaki A, Skalidaki E, Mavroudis D, Georgoulias V, Kotsakis A. A circulating subpopulation of monocytic myeloid-derived suppressor cells as an independent prognostic/predictive factor in untreated non-small lung cancer patients. J Immunol Res 2014; 2014:659294; PMID:25436215; [PMC free article] [PubMed] [Cross Ref]
16. Ma Y, Aymeric L, Locher C, Kroemer G, Zitvogel L. The dendritic cell-tumor cross-talk in cancer. Curr Opin Immunol 2011; 23:146-52; PMID:20970973; [PubMed] [Cross Ref]
17. Carus A, Ladekarl M, Hager H, Pilegaard H, Nielsen PS, Donskov F, Ferlay J, Shin H-R, Bray F, Forman D et al. Tumor-associated neutrophils and macrophages in non-small cell lung cancer: No immediate impact on patient outcome. Lung Cancer 2013; 81:130-7; PMID:23540719; [PubMed] [Cross Ref]
18. Oleinika K, Nibbs RJ, Graham GJ, Fraser AR. Suppression, subversion and escape: the role of regulatory T cells in cancer progression. Clin Exp Immunol 2013; 171:36-45; PMID:23199321; [PubMed] [Cross Ref]
19. Mittal D, Gubin MM, Schreiber RD, Smyth MJ. New insights into cancer immunoediting and its three component phases–elimination, equilibrium and escape. Curr Opin Immunol 2014; 27:16-25; PMID:24531241; [PMC free article] [PubMed] [Cross Ref]
20. Remark R, Becker C, Gomez JE, Damotte D, Dieu-Nosjean M-C, Sautès-Fridman C, Fridman W-H, Powell CA, Altorki NK, Merad M et al. The non-small cell lung cancer immune contexture. A major determinant of tumor characteristics and patient outcome. Am J Respir Crit Care Med 2015; 191(4):377-90; PMID:25369536; [PMC free article] [PubMed] [Cross Ref]
21. Suzuki K, Kadota K, Sima CS, Nitadori J, Rusch VW, Travis WD, Sadelain M, Adusumilli PS. Clinical impact of immune microenvironment in stage I lung adenocarcinoma: tumor interleukin-12 receptor β2 (IL-12Rβ2), IL-7R, and stromal FoxP3/CD3 ratio are independent predictors of recurrence. J Clin Oncol 2013; 31:490-8; PMID:23269987; [PMC free article] [PubMed] [Cross Ref]
22. Hald SM, Bremnes RM, Al-Shibli K, Al-Saad S, Andersen S, Stenvold H, Busund L-T, Donnem T. CD4/CD8 co-expression shows independent prognostic impact in resected non-small cell lung cancer patients treated with adjuvant radiotherapy. Lung Cancer 2013; 80:209-15; PMID:23384671; [PubMed] [Cross Ref]
23. Donnem T, Kilvaer TK, Andersen S, Richardsen E, Paulsen E, Hald SM, Al-Saad S, Brustugun OT, Helland A, Lund-Iversen M et al. Strategies for clinical implementation of TNM-immunoscore in resected non-small cell lung cancer. Ann Oncol 2016; 27:225-32; PMID:26578726; [PubMed] [Cross Ref]
24. Velcheti V, Schalper KA, Carvajal DE, Anagnostou VK, Syrigos KN, Sznol M, Herbst RS, Gettinger SN, Chen L, Rimm DL. Programmed death ligand-1 expression in non-small cell lung cancer. Lab Investig 2014; 94:107-16; PMID:24217091; [PubMed] [Cross Ref]
25. Lossos IS, Czerwinski DK, Alizadeh AA, Wechser MA, Tibshirani R, Botstein D, Levy R. Prediction of survival in diffuse large-B-cell lymphoma based on the expression of six genes. N Engl J Med 2004; 350:1828-37; PMID:15115829; [PubMed] [Cross Ref]
26. Schetter AJ, Nguyen GH, Bowman ED, Mathé EA, Yuen ST, Hawkes JE, Croce CM, Leung SY, Harris CC, Aaltonen L et al. Association of inflammation-related and microRNA gene expression with cancer-specific mortality of colon adenocarcinoma. Clin Cancer Res 2009; 15:5878-87; PMID:19737943; [PMC free article] [PubMed] [Cross Ref]
27. Burotto M, Thomas A, Subramaniam D, Giaccone G, Rajan A. Biomarkers in early-stage non-small-cell lung cancer: current concepts and future directions. J Thorac Oncol 2014; 9:1609-17; PMID:25185530; [PMC free article] [PubMed] [Cross Ref]
28. Fridman WH, Pagès F, Sautès-Fridman C, Galon J. The immune contexture in human tumours: impact on clinical outcome. Nat Rev Cancer 2012; 12:298-306; PMID:22419253; [PubMed] [Cross Ref]
29. Rivas-Fuentes S, Salgado-Aguayo A, Pertuz Belloso S, Gorocica Rosete P, Alvarado-Vásquez N, Aquino-Jarquin G. Role of chemokines in non-small cell lung cancer: angiogenesis and inflammation. J Cancer 2015; 6:938-52; PMID:26316890; [PMC free article] [PubMed] [Cross Ref]
30. Mukaida N, Baba T. Chemokines in tumor development and progression. Exp Cell Res 2012; 318:95-102; PMID:22036649; [PubMed] [Cross Ref]
31. Chen G, Wang Z, Liu X, Liu F. High-level CXCR4 expression correlates with brain-specific metastasis of non-small cell lung cancer. World J Surg 2011; 35:56-61; PMID:21046385; [PubMed] [Cross Ref]
32. Yao X, Qi L, Chen X, Du J, Zhang Z, Liu S, Jemal A, Siegel R, Xu J, Ward E et al. Expression of CX3CR1 associates with cellular migration, metastasis, and prognosis in human clear cell renal cell carcinoma. Urol Oncol 2014; 32:162-70; PMID:23570708; [PubMed] [Cross Ref]
33. Otsuka S, Klimowicz AC, Kopciuk K, Petrillo SK, Konno M, Hao D, Muzik H, Stolte E, Boland W, Morris D et al. CXCR4 overexpression is associated with poor outcome in females diagnosed with stage IV non-small cell lung cancer. J Thorac Oncol 2011; 6:1169-78; PMID:21623238; [PubMed] [Cross Ref]
34. Tardáguila M, Mira E, García-Cabezas MA, Feijoo AM, Quintela-Fandino M, Azcoitia I, Lira SA, Mañes S CX3CL1 promotes breast cancer via transactivation of the EGF pathway. Cancer Res 2013; 73:4461-73; [PMC free article] [PubMed] [Cross Ref]
35. Hanagiri T, Shigematsu Y, Shinohara S, Takenaka M, Oka S, Chikaishi Y, Nagata Y, Iwata T, Uramoto H, So T et al. Clinical significance of the frequency of regulatory T cells in regional lymph node lymphocytes as a prognostic factor for non-small-cell lung cancer. Lung Cancer 2013; 81:475-9; PMID:23891508; [PubMed] [Cross Ref]
36. Pircher A, Gamerith G, Amann A, Reinold S, Popper H, Gächter A, Pall G, Wöll E, Jamnig H, Gastl G et al. Neoadjuvant chemo-immunotherapy modifies CD4(+)CD25(+) regulatory T cells (Treg) in non-small cell lung cancer (NSCLC) patients. Lung Cancer 2014; 85:81-7; PMID:24780112; [PubMed] [Cross Ref]
37. Duan M-C, Han W, Jin P-W, Wei Y-P, Wei Q, Zhang L-M, Li J-C. Disturbed Th17/Treg balance in patients with non-small cell lung cancer. Inflammation 2015; 38:2156-65; PMID:26077695; [PubMed] [Cross Ref]
38. Draghiciu O, Lubbers J, Nijman HW, Daemen T. Myeloid derived suppressor cells – an overview of combat strategies to increase immunotherapy efficacy. Oncoimmunology 2015; 4:e954829; PMID:25949858; [PMC free article] [PubMed] [Cross Ref]
39. Gabrilovich DI, Nagaraj S. Myeloid-derived suppressor cells as regulators of the immune system. Nat Rev Immunol 2009; 9:162-74; PMID:19197294; [PMC free article] [PubMed] [Cross Ref]
40. Ortiz ML, Lu L, Ramachandran I, Gabrilovich DI. Myeloid-derived suppressor cells in the development of lung cancer. Cancer Immunol Res 2014; 2:50-8; PMID:24778162; [PMC free article] [PubMed] [Cross Ref]
41. Sharma S, Dubinett S, Salgia R. CD14(+)S100A9(+) myeloid-derived suppressor cells portend decreased survival in patients with advanced lung cancer. Am J Respir Crit Care Med 2012; 186:940-1; PMID:23155211; [PMC free article] [PubMed] [Cross Ref]
42. Hatanaka H, Abe Y, Kamiya T, Morino F, Nagata J, Tokunaga T, Oshika Y, Suemizu H, Kijima H, Tsuchida T et al. Clinical implications of interleukin (IL)-10 induced by non-small-cell lung cancer. Ann Oncol 2000; 11:815-9; PMID:10997808; [PubMed] [Cross Ref]
43. Kong F-F, Zhu Y-L, Yuan H-H, Wang J-Y, Zhao M, Gong X-D, Liu F, Zhang W-Y, Wang C-R, Jiang B. FOXM1 regulated by ERK pathway mediates TGF-β1-induced EMT in NSCLC. Oncol Res 2014; 22:29-37; PMID:25700356; [PubMed] [Cross Ref]
44. Karanikas V, Zamanakou M, Kerenidi T, Dahabreh J, Hevas A, Nakou M, Gourgoulianis KI, Germenis AE. Indoleamine 2,3-dioxygenase (IDO) expression in lung cancer. Cancer Biol Ther 2007; 6:1258-62; PMID:17700060; [PubMed] [Cross Ref]
45. Salvi S, Fontana V, Boccardo S, Merlo DF, Margallo E, Laurent S, Morabito A, Rijavec E, Dal Bello MG, Mora M et al. Evaluation of CTLA-4 expression and relevance as a novel prognostic factor in patients with non-small cell lung cancer. Cancer Immunol Immunother 2012; 61:1463-72; PMID:22318401; [PubMed] [Cross Ref]
46. Deng L, Gyorffy B, Na F, Chen B, Lan J, Xue J, Zhou L, Lu Y. Association of PDCD1 and CTLA-4 gene expression with clinicopathological factors and survival in non-small-cell lung cancer: results from a large and pooled microarray database. J Thorac Oncol 2015; 10:1020-6; PMID:26134222; [PubMed] [Cross Ref]
47. D'Incecco A, Andreozzi M, Ludovini V, Rossi E, Capodanno A, Landi L, Tibaldi C, Minuti G, Salvini J, Coppi E et al. PD-1 and PD-L1 expression in molecularly selected non-small-cell lung cancer patients. Br J Cancer 2015; 112:95-102; PMID:25349974; [PMC free article] [PubMed] [Cross Ref]
48. Usó M, Jantus-Lewintre E, Bremnes RM, Calabuig S, Blasco A, Pastor E, Borreda I, Molina-Pinelo S, Paz-Ares L, Guijarro R et al. Analysis of the immune microenvironment in resected non-small cell lung cancer: the prognostic value of different T lymphocyte markers. Oncotarget 2016; 7:52849-61; PMID:27463005; [PMC free article] [PubMed] [Cross Ref]
49. Ascierto ML, Kmieciak M, Idowu MO, Manjili R, Zhao Y, Grimes M, Dumur C, Wang E, Ramakrishnan V, Wang X-Y et al. A signature of immune function genes associated with recurrence-free survival in breast cancer patients. Breast Cancer Res Treat 2012; 131:871-80; PMID:21479927; [PMC free article] [PubMed] [Cross Ref]
50. Denkert C, von Minckwitz G, Brase JC, Sinn B V, Gade S, Kronenwett R, Pfitzner BM, Salat C, Loi S, Schmitt WD et al. Tumor-infiltrating lymphocytes and response to neoadjuvant chemotherapy with or without carboplatin in human epidermal growth factor receptor 2-positive and triple-negative primary breast cancers. J Clin Oncol 2015; 33:983-91; PMID:25534375; [PubMed] [Cross Ref]
51. Gajewski TF, Schreiber H, Fu Y-X. Innate and adaptive immune cells in the tumor microenvironment. Nat Immunol 2013; 14:1014-22; PMID:24048123; [PMC free article] [PubMed] [Cross Ref]
52. Tumeh PC, Harview CL, Yearley JH, Shintaku IP, Taylor EJM, Robert L, Chmielowski B, Spasic M, Henry G, Ciobanu V et al. PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 2014; 515:568-71; PMID:25428505; [PMC free article] [PubMed] [Cross Ref]
53. Langowski JL, Zhang X, Wu L, Mattson JD, Chen T, Smith K, Basham B, McClanahan T, Kastelein RA, Oft M. IL-23 promotes tumour incidence and growth. Nature 2006; 442:461-5; PMID:16688182; [PubMed] [Cross Ref]
54. Kaiga T, Sato M, Kaneda H, Iwakura Y, Takayama T, Tahara H. Systemic administration of IL-23 induces potent antitumor immunity primarily mediated through Th1-type response in association with the endogenously expressed IL-12. J Immunol 2007; 178:7571-80; PMID:17548592; [PubMed] [Cross Ref]
55. Wolf AM, Rumpold H, Reimer D, Marth C, Zeimet AG, Wolf D. High IL-12 p35 and IL-23 p19 mRNA expression is associated with superior outcome in ovarian cancer. Gynecol Oncol 2010; 118:244-50; PMID:20603238; [PubMed] [Cross Ref]
56. Baird A-M, Leonard J, Naicker KM, Kilmartin L, O'Byrne KJ, Gray SG. IL-23 is pro-proliferative, epigenetically regulated and modulated by chemotherapy in non-small cell lung cancer. Lung Cancer 2013; 79:83-90; PMID:23116756; [PubMed] [Cross Ref]
57. Liu F-T, Rabinovich GA. Galectins as modulators of tumour progression. Nat Rev Cancer 2005; 5:29-41; PMID:15630413; [PubMed] [Cross Ref]
58. Jung J-H, Kim H-J, Yeom J, Yoo C, Shin J, Yoo J, Kang CS, Lee C. Lowered expression of galectin-2 is associated with lymph node metastasis in gastric cancer. J Gastroenterol 2012; 47:37-48; PMID:22015694; [PubMed] [Cross Ref]
59. Huang P-Y, Guo S-S, Zhang Y, Lu J-B, Chen Q-Y, Tang L-Q, Zhang L, Liu L-T, Zhang L, Mai H-Q. Tumor CTLA-4 overexpression predicts poor survival in patients with nasopharyngeal carcinoma. Oncotarget 2016; 7:13060-8; PMID:26918337; [PMC free article] [PubMed] [Cross Ref]
60. Zhang X-F, Pan K, Weng D-S, Chen C-L, Wang Q-J, Zhao J-J, Pan Q-Z, Liu Q, Jiang S-S, Li Y-Q et al. Cytotoxic T lymphocyte antigen-4 expression in esophageal carcinoma: implications for prognosis. Oncotarget 2016; 7:26670–9; PMID:27050369; [PMC free article] [PubMed] [Cross Ref]
61. Brahmer J, Reckamp KL, Baas P, Crinò L, Eberhardt WEE, Poddubskaya E, Antonia S, Pluzanski A, Vokes EE, Holgado E et al. Nivolumab versus docetaxel in advanced squamous-cell non-small-cell lung cancer. N Engl J Med 2015; 373:123-35; PMID:26028407; [PMC free article] [PubMed] [Cross Ref]
62. Borghaei H, Paz-Ares L, Horn L, Spigel DR, Steins M, Ready NE, Chow LQ, Vokes EE, Felip E, Holgado E et al. Nivolumab versus docetaxel in advanced nonsquamous non-small-cell lung cancer. N Engl J Med 2015; 373:1627-39; PMID:26412456; [PMC free article] [PubMed] [Cross Ref]
63. Ott PA, Hodi FS, Robert C. CTLA-4 and PD-1/PD-L1 blockade: new immunotherapeutic modalities with durable clinical benefit in melanoma patients. Clin Cancer Res 2013; 19:5300-9; PMID:24089443; [PubMed] [Cross Ref]
64. Herbst RS, Soria J-C, Kowanetz M, Fine GD, Hamid O, Gordon MS, Sosman JA, McDermott DF, Powderly JD, Gettinger SN et al. Predictive correlates of response to the anti-PD-L1 antibody MPDL3280A in cancer patients. Nature 2014; 515:563-7; PMID:25428504; [PMC free article] [PubMed] [Cross Ref]
65. Schmidt LH, Kümmel A, Görlich D, Mohr M, Bröckling S, Mikesch JH, Grünewald I, Marra A, Schultheis AM, Wardelmann E et al. PD-1 and PD-L1 expression in NSCLC indicate a favorable prognosis in defined subgroups. PLoS One 2015; 10:1-15; PMID:26313362; [PMC free article] [PubMed] [Cross Ref]
66. McLaughlin J, Han G, Schalper KA, Carvajal-Hausdorf D, Pelekanou V, Rehman J, Velcheti V, Herbst R, LoRusso P, Rimm DL. Quantitative assessment of the heterogeneity of PD-L1 expression in non-small-cell lung cancer. JAMA Oncol 2016; 2:46-54; PMID:26562159; [PMC free article] [PubMed] [Cross Ref]
67. Kerr KM, Tsao M-S, Nicholson AG, Yatabe Y, Wistuba II, Hirsch FR, IASLC Pathology Committee J, Drake C, Wollner I et al. Programmed death-ligand 1 immunohistochemistry in lung cancer: in what state is this art? J Thorac Oncol 2015; 10:985-9; PMID:26134220; [PubMed] [Cross Ref]
68. Iglesia MD, Parker JS, Hoadley KA, Serody JS, Perou CM, Vincent BG Genomic analysis of immune cell infiltrates across 11 tumor types. J Natl Cancer Inst 2016; 108:djw144; PMID:27335052; [PMC free article] [PubMed] [Cross Ref]
69. Angell H, Galon J. From the immune contexture to the Immunoscore: the role of prognostic and predictive immune markers in cancer. Curr Opin Immunol 2013; 25:261-7; PMID:23579076; [PubMed] [Cross Ref]
70. Altman DG, McShane LM, Sauerbrei W, Taube SE. Reporting recommendations for tumor marker prognostic studies (REMARK): explanation and elaboration. BMC Med 2012; 10:51; PMID:22642691; [PMC free article] [PubMed] [Cross Ref]
71. Pfaffl MW.. A new mathematical model for relative quantification in real-time RT-PCR. Nucleic Acids Res 2001; 29:e45; PMID:11328886; [PMC free article] [PubMed] [Cross Ref]
72. Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA 1998; 95:14863-8; PMID:9843981; [PubMed] [Cross Ref]

Articles from Oncoimmunology are provided here courtesy of Taylor & Francis