Predictive Gene Signature in MAGE-A3 Antigen-Specific Cancer Immunotherapy
Both F.U.-M. and J.L. contributed equally to this work.
To detect a pretreatment gene expression signature (GS) predictive of response to MAGE-A3 immunotherapeutic in patients with metastatic melanoma and to investigate its applicability in a different cancer setting (adjuvant therapy of resected early-stage non–small-cell lung cancer [NSCLC]).
Patients were participants in two phase II studies of the recombinant MAGE-A3 antigen combined with an immunostimulant (AS15 or AS02B). mRNA from melanoma biopsies was analyzed by microarray analysis and quantitative polymerase chain reaction. These results were used to identify and cross-validate the GS, which was then applied to the NSCLC data.
In the patients with melanoma, 84 genes were identified whose expression was potentially associated with clinical benefit. This effect was strongest when the immunostimulant AS15 was included in the immunotherapy (hazard ratio [HR] for overall survival, 0.37; 95% CI, 0.13 to 1.05; P = .06) and was less strong with the other immunostimulant AS02B (HR, 0.84; 95% CI, 0.36 to 1.97; P = .70). The same GS was then used to predict the outcome for patients with resected NSCLC treated with MAGE-A3 plus AS02B; actively treated GS-positive patients showed a favorable disease-free interval compared with placebo-treated GS-positive patients (HR, 0.42; 95% CI, 0.17 to 1.03; P = .06), whereas among GS-negative patients, no such difference was found (HR, 1.17; 95% CI, 0.59 to 2.31; P = .65). The genes identified were mainly immune related, involving interferon gamma pathways and specific chemokines, suggesting that their pretreatment expression influences the tumor's immune microenvironment and the patient's clinical response.
Although recent developments of new therapies have improved the survival of patients with cancer, there are still many for whom anticancer treatment brings little or no benefit but only adverse effects. Consequently, attention has been focused on finding biomarkers to identify patients most likely to respond to a specific cancer therapy.
Gene expression profiling of some tumors, such as lymphoma, non–small-cell lung cancer (NSCLC), and breast cancer, is a powerful technique for identifying prognostic gene signatures (GSs).1–6 More recently, predictive GSs have been reported for cetuximab treatment in colorectal cancer7 and for chemotherapy in gastric cancer.8 GSs, both prognostic and predictive, have also been identified in NSCLC and breast cancer.9,10 However, results from comprehensive validation trials and prospective studies, required to generalize the use of these GSs in clinical settings, are still scarce. Accordingly, large-scale clinical trials are ongoing to validate the profiles believed to be associated with various prognoses in breast cancer,11,12 melanoma (DERMA [Adjuvant Immunotherapy with MAGE-A3 in Melanoma] trial, ClinicalTrials.gov identifier: NCT00796445), and NSCLC (MAGRIT [MAGE-A3 as Adjuvant Non–Small-Cell Lung Cancer Immunotherapy] trial, ClinicalTrials.gov identifier: NCT00480025).
Among the new targeted approaches being developed to treat cancer, immunotherapy is gaining importance. One such approach, active immunotherapy, aims to harness the host immune system to attack specific antigens presented on cancer cells and thus induce immune-mediated elimination of the malignant cells. Immunization against antigenic tumor epitopes is achieved by injection of recombinant tumor antigen protein.13 The antigen MAGE-A3 belongs to the group of tumor-specific antigens that are not expressed in normal tissues except testes and placenta14–17 but are present on the surface of various tumors including melanoma, NSCLC, and bladder and hepatocellular cancer. Both humoral and cellular immune responses against tumor antigens are induced if the recombinant MAGE-A3 protein is administered together with an immunostimulant.18–20
We have described the clinical activity of immunotherapy in a randomized phase II trial of recombinant MAGE-A3 protein combined with immunostimulant AS02B or AS15, administered to patients with early metastatic melanoma (ClinicalTrials.gov identifier: NCT0086866),21 and in another double-blind, placebo-controlled trial in which the same antigen, combined with AS02B, was given as adjuvant therapy to patients with completely resected stage IB to II NSCLC (ClinicalTrials.gov identifier: NCT00290355).22 In both studies, tumor biopsies obtained before immunotherapy were analyzed for gene expression. The aim was to investigate whether a GS could be identified as a predictor of response in patients with metastatic melanoma to the MAGE-A3 immunotherapeutic and whether this GS could be applied in another cancer setting (adjuvant) of another tumor (NSCLC).
The studies were performed with approval of all relevant review boards. This article follows the Reporting Recommendations for Tumor Marker Prognostic Studies guidelines.23 Tumor biopsy tissue, obtained just before MAGE-A3 immunotherapy from patients with appropriate consent for gene expression profiling, was used. mRNA was extracted, purified, amplified, and assayed on Affymetrix HG-U133.Plus 2.0 (Affymetrix, Santa Clara, CA) microarray gene chips by standard methods. Quantitative real-time polymerase chain reaction (qRT-PCR) for gene expression determination in both melanoma and NSCLC biopsies was done by TaqMan (Applied Biosystems, Foster City, CA) low-density arrays.
The metastatic melanoma microarray data set,21 based on 56 patients with evaluable clinical outcome who had received six or more doses of the immunotherapeutic, was used to identify genes that discriminate between patients with clinical benefit (CB) and with no CB and to develop a classification algorithm. Briefly, CB included objective responders (complete and partial) according to RECIST 1.024 and patients showing stable disease (> 4 months) or mixed response with unequivocal tumor shrinkage. For CONSORT diagram and complete definitions, see Appendix Figure A1 (online only) and Data Supplement.
For gene expression analysis, absent microarray probe sets (PSs) were discarded; retained PSs were filtered nonspecifically by interquartile range and then ranked by signal-to-noise score,25 and the top 100 were selected as the most discriminant. Hierarchical clustering analysis used the gplots R package (http://cran.r-project.org/web/packages/gplots/index.html). The samples in the melanoma data set were then used to build a classifier by supervised principal component–discriminant analysis.
To estimate the predictive performance of the classifier in the melanoma data set, the procedure for gene selection and classification was tested by cross-validation. For this, samples were omitted from the above process and then classified based on the classifier built without them. This was done by leave-one-out cross-validation (LOOCV; omitting one sample at a time) and by random selection of a training set comprising 90% of the samples (performed 500 times).
The genes found by microarray analysis to be discriminant in metastatic melanoma were corroborated by qRT-PCR. The latter data were used to build a qRT-PCR–based classifier, which was then applied to the qRT-PCR–based expression data from 157 patients in the adjuvant NSCLC study.22 Detailed descriptions of procedures, data analysis, and patients included in the gene profile analysis are provided in the Data Supplement.
The metastatic melanoma study was an open phase II trial comparing the recombinant MAGE-A3 protein combined with immunostimulants (AS02B or AS15, the latter showing better clinical activity) in 75 patients with nonresectable MAGE-A3–positive stage III or IV M1a metastatic melanoma.21 A pretreatment GS-based classifier was built using all evaluable melanoma samples (from 56 patients; Appendix Fig A1). There were 22 patients with CB (14 in the AS15 arm and eight in the AS02B arm) and 34 patients with no CB (13 in the AS15 arm and 21 in the AS02B arm). PSs were ranked according to their ability to discriminate between CB and no CB; the top 100 were found to correspond to 84 genes (Data Supplement). Hierarchical clustering was then used to visualize the expression pattern of the top 100 PSs. This identified two main clusters (Fig 1), one enriched in responders and containing all of the objective (ie, complete or partial) responders. In addition, a binary classifier based on the first principal component of the top 100 PSs was developed (Data Supplement). This component captured 56% of the total variance.
The GS-based classification was tested by LOOCV including feature selection. The prediction of CB showed a sensitivity (proportion of patients with CB correctly classified as GS positive) of 0.77 (95% CI, 0.60 to 0.95) and a specificity (proportion of patients with no CB correctly classified as GS negative) of 0.56 (95% CI, 0.39 to 0.73) using a cutoff of 0.43 (Fig 2). Sensitivity and specificity were higher when only the AS15 arm was considered (0.79; 95% CI, 0.57 to 1.00; and 0.69; 95% CI, 0.44 to 0.94, respectively). Importantly, all of the patients with objective (ie, complete or partial) response were correctly classified (clinical relevance was further confirmed with overall survival [OS] data).
The stability of the gene selection was evaluated in the LOOCV. For this purpose, the number of times that each PS was selected within the top 100 signal-to-noise–ranked PSs in the 56 LOOCV loops was recorded (Data Supplement). Sixty-eight PSs were found in at least 50 of the LOOCV loops, implying stability of the gene selection.
Finally, the classifier performance was repeatedly tested using a randomly chosen 90% of the samples as the training set. The PS selection and classifier performed, on average, similarly to the LOOCV, implying that the method is robust (Data Supplement).
Sixty-one of the 84 genes identified by microarray were corroborated by qRT-PCR (Data Supplement). These were selected on the sole basis of availability and quality of qRT-PCR assays. The expression pattern found by qRT-PCR was similar to that found by microarray; all of the genes tested showed the same direction of overexpression (CB v no CB), and most showed similar magnitude of their ratio of expression level between the two groups (Fig 3; Data Supplement). A supervised principal component–discriminant analysis classifier built with these qRT-PCR data showed results similar to those based on microarray data from the same samples (Data Supplement).
The effect of the GS on survival benefit was investigated using the GS status obtained by LOOCV. Consistent with the previous findings, OS was notably greater in the population of patients whose tumor presented the GS; median OS was 16.2 months (95% CI, 9.0 to 20.0 months) in the GS-negative population and 29.0 months (95% CI, 20.5 to 40.2 months) in the GS-positive population. However, the separation by immunostimulant showed a more than three-fold difference (OS: 16.2 months; 95% CI, 4.5 months to not reached [NR] for GS-negative patients and 53.7 months; 95% CI, 29.0 months to NR for GS-positive patients among the AS15-treated patients; Fig 4A). A smaller difference was seen in the AS02B arm (OS: 15.3 months; 95% CI, 4.2 months to NR for GS-negative patients and 21.1 months; 95% CI, 14.5 to 25.5 months for GS-positive patients), although the GS-negative curves from both arms were superimposed.
These differences were not attributable to differences in baseline characteristics; the GS was evenly distributed between the two study arms, and the treatment groups and other important potential prognostic factors were well balanced across profiles, except that the GS-positive group contained more women and more patients with stage IV disease (Data Supplement).
Most of the classifier genes have a higher expression level in patients benefiting from the treatment. Analysis of biologic functions revealed over-representation of immune-related genes in the GS (Table 1; Data Supplement), indicating a strong Th1 adaptive immunity, and interferon gamma (IFNG) pathway (CD8, granzymes, CD3D, IRF1) in the tumors of CB patients. In addition, overexpression of genes coding for specific chemokines (CCL5, CXCL2, CXCL9, and CXCL10), known to be involved in T-cell homing to the tumor, was identified. Analysis of biologic networks revealed that the GS is driven mainly by IFNG with STAT1 and IRF1 as central transcription factors, which are part of the signature (Data Supplement). Analysis of upstream direct regulators of genes in the signature reveals that STAT1 regulates B2M, CCL5, CD86, CXCL10, CXCL9, FAM26F, GBP1, GBP5, IRF1, and JAK2, whereas IRF1 regulates B2M, CCL5, CXCL10, IRF1, JAK2, PSMB10, RARRES3, and STAT1. High expression of these molecules correlated with CB from the MAGE-A3 immunotherapeutic.
|Biologic Function||P||No. of Genes|
|Antigen presentation||< .001 to .00506||27|
|Cell-to-cell signaling and interaction||< .001 to .0076||28|
|Cellular development||< .001 to .00675||27|
|Cell death||< .001 to .0058||28|
|Cellular movement||< .001 to .0076||19|
|Cell-mediated immune response||< .001 to .0076||32|
|Humoral immune response||< .001 to .0076||29|
|Hematologic system development and function||< .001 to .0076||32|
|Tissue morphology||< .001 to .0076||23|
|Immune cell trafficking||< .001 to .0076||23|
Abbreviation: GS, gene expression signature.
The predictive GS and classifier defined in patients with melanoma was tested by applying it to the population of an independent, randomized (2:1), double-blind, placebo-controlled clinical trial of MAGE-A3 plus AS02B as adjuvant treatment of 182 patients with completely resected MAGE-A3–positive NSCLC (stage IB/II). The results suggested superior clinical activity of the immunotherapeutic compared with placebo.22
Expression of the 61 genes measured by qRT-PCR in the melanoma samples was assayed by the same method, and the melanoma-defined classifier was applied to the resected NSCLC samples taken before treatment (157 of the 182 patients in the study). Sixty-one of 157 patients were classified as GS positive, and the rest were classified as GS negative. As expected, most of the genes in the GS defined in the melanoma data set were differentially expressed between GS-positive and GS-negative NSCLC samples (Data Supplement). Unsupervised hierarchical clustering showed two clusters, one containing only GS-negative patients and the other enriched in GS-positive patients as determined by the qRT-PCR melanoma classifier (Data Supplement). Immune-related genes were overexpressed in samples classified as GS positive. A pronounced association between the GS and clinical benefit is seen in the comparison of disease-free interval (DFI; the primary end point of the study) for the treatment groups (active and placebo) stratified by GS (positive v negative; Fig 4B). A lower risk of cancer recurrence is seen for patients who both received active immunization and presented the GS, compared with patients who either received placebo or did not present the GS. The baseline characteristics for GS and treatment groups were equally distributed in the two subpopulations (Data Supplement), suggesting that there was no obvious bias as a result of baseline influences.
In the original study analysis,22 which did not account for the GS, no treatment benefit in terms of OS could be discerned (whereas difference was observed in terms of DFI and disease-free survival). In contrast, when the GS is also taken into consideration, the immunotherapeutic also shows a treatment benefit as measured by OS (hazard ratio, 0.63; 95% CI, 0.22 to 1.78; P = .38 in GS-positive patients v 1.09; 95% CI, 0.56 to 2.11; P = .81 in GS-negative patients; Fig 4C).
The differences observed between the clinical responses of the GS-positive and GS-negative patients and those who received the antigen with different immunostimulants show that, within the context investigated here, the GS may have predictive value. This is supported by the clustering of CB observed in the patients with melanoma, with a higher proportion of patients showing complete or partial responses with tumor shrinkage (not occurring spontaneously in this clinical setting); the confirmation of the robustness of the GS by LOOCV and reduced training sets; the substantial clinical difference manifested by the strong association between the GS and improved OS (especially for the more active immunostimulant); the finding of a similar classifier when qRT-PCR was used instead of microarray analysis; and the success of the GS as a predictor in patients with NSCLC, a population different in many respects (tumor location, stage of disease, and therapeutic setting) from the patients used to derive the GS.
Among the patients with NSCLC who were GS-negative, the MAGE-A3–treated and placebo-treated patients did not show a difference in DFI, whereas among GS-positive patients, the active treatment was associated with clear clinical activity. These observations emphasize the potential importance of the GS as a predictive factor in melanoma and NSCLC.
It is noteworthy that in the NSCLC study population, no treatment effect on OS was observable until the GS had been taken into account. This suggests that it is possible, in a clinical study, to mask the beneficial effect of a cancer treatment by not choosing a biologically appropriate study population (in this case, patients with a specific tumor microenvironment).
To our knowledge, this is the first publication of evidence that clinical response to a cancer immunotherapeutic may be associated with an immune biomarker signature in two different settings (here, metastatic and adjuvant) and in two tumor types (here, melanoma and NSCLC). The apparent association of clinical outcome with GS in two different cancer types suggests that the same biologic mechanism might be involved in the response to the MAGE-A3 immunotherapeutic in both settings and diseases.
Our data suggest the existence of a determinative interplay between the patient's immune system and the cancer cells. Thus, a specific tumor microenvironment may favor the presence of immune effector cells in the tumor of responder patients. Among the genes differentially expressed (overexpressed in patients with CB) were MHC class I and II, T-cell markers such as CD3D and CD8 (known to be regulated by IFNG) and downstream targets of STAT1 and IRF1. Although IFNG was not detected in any of the samples by microarray gene expression profiling, it was found to be differentially expressed between CB and no CB samples when determined by qRT-PCR. Additionally, the signature contains genes involved in antigen processing and also a chemokine signature including CCL5, CXCL2, CXCL9, and CXCL10.
The presence of chemokines in the GS may be linked to the trafficking of T cells at the tumor site; tumors that express chemokines could favor the infiltration of immune effector cells, as suggested by gene expression profiling in metastatic melanoma,26 where the presence of lymphocytes in tumors correlated with the expression of a subset of six chemokines, three of which were identified here. The presence of chemokines is also correlated with infiltration of high densities of T-cell subpopulations in colorectal cancer27 and in NSCLC28 and with dendritic cells in NSCLC.29 It is likely that the predictive GS described provides a more accurate quantification of the immune infiltration associated with tumor-infiltrating lymphocytes and their effector potential. Interestingly, some of the genes we identified have been proposed to be associated with clinical response to ipilimumab.30 Therefore, the chemokine signature might also favor immune infiltration induced by active immunization. Recent evidence suggests that immune elements within the tumor may also predict response to chemotherapy,31 as described in breast cancer.32
The GS identified here seems to have little prognostic value in these study populations (compare, for example, the results for placebo-treated GS-positive and GS-negative patients with NSCLC; Fig 4B). However, this cannot be generalized, because we examined only the MAGE-A3–positive population within restricted stages. The GS was present in approximately 50% of the MAGE-A3–positive tumors included in these two trials (57% in melanoma and 39% in NSCLC) and is not necessarily associated with the expression of MAGE-A3 by these tumors (data not shown). A reliable statement about the possible prognostic value of the GS would require a wider base of data, especially for OS (Fig 4C); however, even if such an effect was observed to be weaker, the biomarkers discovered would still be useful as predictive biomarkers. The prognostic value of immune infiltration of tumors remains unclear, although recent publications suggest that infiltrated tumors have led to better outcome in several tumor types,33 and there is increasing evidence that immune infiltrates of primary tumors and metastases can be both prognostic34 and predictive,35 whereas a pattern of tumor classification according to immune infiltration has been proposed.36 We hypothesize that the GS described provides a more comprehensive characterization of the tumor microenvironment by including not only T-cell markers such as CD3G and CD8, used among others to characterize intratumoral immune cells, but also Th1 cell markers and their effector potential, T-cell activation markers (ICOS, CD86), chemokines (CXCL9, CXCL10, CCL5, CXCL2), genes expressed in natural killer cells (KLRD1, KLRB1), and HLA molecules.
Overall, we regard the results presented here (albeit obtained in studies not powered to detect differences in subset analyses) as providing support for the increasing body of data implicating the tumor's immunologic microenvironment in a patient's response to cancer therapy, which in our studies was active immunotherapy. Although the mechanistic aspects of this remain to be clarified, the empirical findings seem sufficiently robust to incorporate the GS identified here into larger scale clinical trials, to test prospectively the working hypothesis that the GS is correlated with clinical response to active immunotherapy. Validation is in progress in two phase III trials (coprimary end point; see Introduction), and further studies will be required to assess whether the GS is also relevant for other immunotherapies or even therapy other than immunotherapy. The confirmation of such a hypothesis could have a substantial effect on future strategies for optimizing the choice of therapy for individual patients.
Supported by GlaxoSmithKline Vaccines.
Presented, in part, at the 21st Joint European Organisation for Research and Treatment of Cancer–National Cancer Institute–American Association for Cancer Research Symposium, November 15-19, 2009, Boston, MA; 24th Annual Meeting of the International Society for Biological Therapy of Cancer, October 28-31, 2009, Washington, DC; and Asian Oncology Summit, April 9-11, 2010, Bali, Indonesia.
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
Clinical trial information: NCT00290355, NCT00086866.
Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.
Employment or Leadership Position: Fernando Ulloa-Montoya, GlaxoSmithKline Vaccines (C); Jamila Louahed, GlaxoSmithKline Vaccines (C); Benjamin Dizier, GlaxoSmithKline Vaccines (C); Olivier Gruselle, GlaxoSmithKline Vaccines (C); Bart Spiessens, GlaxoSmithKline Vaccines (C); Frédéric F. Lehmann, GlaxoSmithKline Vaccines (C); Vincent G. Brichard, GlaxoSmithKline Vaccines (C) Consultant or Advisory Role: Johan Vansteenkiste, GlaxoSmithKline Vaccines (C) Stock Ownership: Fernando Ulloa-Montoya, GlaxoSmithKline Vaccines; Jamila Louahed, GlaxoSmithKline Vaccines; Bart Spiessens, GlaxoSmithKline Vaccines; Frédéric F. Lehmann, GlaxoSmithKline Vaccines; Vincent G. Brichard, GlaxoSmithKline Vaccines Honoraria: None Research Funding: None Expert Testimony: None Other Remuneration: None
Conception and design: Fernando Ulloa-Montoya, Jamila Louahed, Frédéric F. Lehmann, Stefan Suciu, Wim H.J. Kruit, Alexander M.M. Eggermont, Johan Vansteenkiste, Vincent G. Brichard
Provision of study materials or patients: Johan Vansteenkiste
Collection and assembly of data: Benjamin Dizier, Olivier Gruselle, Johan Vansteenkiste
Data analysis and interpretation: Fernando Ulloa-Montoya, Jamila Louahed, Benjamin Dizier, Olivier Gruselle, Bart Spiessens, Frédéric F. Lehmann, Stefan Suciu, Wim H.J. Kruit, Johan Vansteenkiste, Vincent G. Brichard
Manuscript writing: All authors
Final approval of manuscript: All authors
|1.||AA Alizadeh, MB Eisen, RE Davis , etal: Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling Nature 403: 503– 511,2000 Crossref, Medline, Google Scholar|
|2.||SS Dave, G Wright, B Tan , etal: Prediction of survival in follicular lymphoma based on molecular features of tumor-infiltrating immune cells N Engl J Med 351: 2159– 2169,2004 Crossref, Medline, Google Scholar|
|3.||K Shedden, JM Taylor , etal: Gene expression-based survival prediction in lung adenocarcinoma: A multi-site, blinded validation study Nat Med 14: 822– 827,2008 Director's Challenge Consortium for the Molecular Classification of Lung Adenocarcinoma, Crossref, Medline, Google Scholar|
|4.||JR Kratz, J He, SK Van Den Eeden , etal: A practical molecular assay to predict survival in resected non-squamous, non-small-cell lung cancer: Development and international validation studies Lancet 379: 823– 832,2012 Crossref, Medline, Google Scholar|
|5.||LJ van't Veer, H Dai, MJ van de Vijver , etal: Gene expression profiling predicts clinical outcome of breast cancer Nature 415: 530– 536,2002 Crossref, Medline, Google Scholar|
|6.||S Paik, S Shak, G Tang , etal: A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer N Engl J Med 351: 2817– 2826,2004 Crossref, Medline, Google Scholar|
|7.||JB Baker, D Dutta, D Watson , etal: Tumour gene expression predicts response to cetuximab in patients with KRAS wild-type metastatic colorectal cancer Br J Cancer 104: 488– 495,2011 Crossref, Medline, Google Scholar|
|8.||HK Kim, IJ Choi, CG Kim , etal: Three-gene predictor of clinical outcome for gastric cancer patients treated with chemotherapy Pharmacogenomics J 12: 119– 127,2012 Crossref, Medline, Google Scholar|
|9.||KS Albain, WE Barlow, S Shak , etal: Prognostic and predictive value of the 21-gene recurrence score assay in postmenopausal women with node-positive, oestrogen-receptor-positive breast cancer on chemotherapy: A retrospective analysis of a randomised trial Lancet Oncol 11: 55– 65,2010 Crossref, Medline, Google Scholar|
|10.||CQ Zhu, K Ding, D Strumpf , etal: Prognostic and predictive gene signature for adjuvant chemotherapy in resected non-small-cell lung cancer J Clin Oncol 28: 4417– 4424,2010 Link, Google Scholar|
|11.||F Cardoso, L Van't Veer, E Rutgers , etal: Clinical application of the 70-gene profile: The MINDACT trial J Clin Oncol 26: 729– 735,2008 Link, Google Scholar|
|12.||JA Sparano, S Paik: Development of the 21-gene assay and its application in clinical practice and clinical trials J Clin Oncol 26: 721– 728,2008 Link, Google Scholar|
|13.||PG Coulie, T Connerotte: Human tumor-specific T lymphocytes: Does function matter more than number? Curr Opin Immunol 17: 320– 325,2005 Crossref, Medline, Google Scholar|
|14.||P van der Bruggen, C Traversari, P Chomez , etal: A gene encoding an antigen recognized by cytolytic T lymphocytes on a human melanoma Science 254: 1643– 1647,1991 Crossref, Medline, Google Scholar|
|15.||E De Plaen, K Arden, C Traversari , etal: Structure, chromosomal localization, and expression of 12 genes of the MAGE family Immunogenetics 40: 360– 369,1994 Crossref, Medline, Google Scholar|
|16.||P Boël, C Wildmann, ML Sensi , etal: BAGE: A new gene encoding an antigen recognized on human melanomas by cytolytic T lymphocytes Immunity 2: 167– 175,1995 Crossref, Medline, Google Scholar|
|17.||BJ Van den Eynde, P van der Bruggen: T cell defined tumor antigens Curr Opin Immunol 9: 684– 693,1997 Crossref, Medline, Google Scholar|
|18.||M Marchand, CJ Punt, S Aamdal , etal: Immunisation of metastatic cancer patients with MAGE-3 protein combined with adjuvant SBAS-2: A clinical report Eur J Cancer 39: 70– 77,2003 Crossref, Medline, Google Scholar|
|19.||D Atanackovic, NK Altorki, E Stockert , etal: Vaccine-induced CD4+ T cell responses to MAGE-3 protein in lung cancer patients J Immunol 172: 3289– 3296,2004 Crossref, Medline, Google Scholar|
|20.||WH Kruit, HH van Ojik, VG Brichard , etal: Phase 1/2 study of subcutaneous and intradermal immunization with a recombinant MAGE-3 protein in patients with detectable metastatic melanoma Int J Cancer 117: 596– 604,2005 Crossref, Medline, Google Scholar|
|21.||WHJ Kruit, S Suciu, B Dreno , etal: Selection of immunostimulant AS15 for active immunization with MAGE-A3 protein: Results of a randomized phase II study of the European Organisation for Research and Treatment of Cancer Melanoma Group in metastatic melanoma J Clin Oncol 31: 2413– 2420,2013 Link, Google Scholar|
|22.||J Vansteenkiste, M Zielinski, A Linder , etal: Adjuvant MAGE-A3 immunotherapy in resected non–small-cell lung cancer: Results of a phase III randomized study J Clin Oncol manuscript submitted for publication Google Scholar|
|23.||LM McShane, DG Altman, W Sauerbrei , etal: Reporting recommendations for tumor MARKer prognostic studies (REMARK) Br J Cancer 93: 387– 391,2005 Crossref, Medline, Google Scholar|
|24.||TR Golub, DK Slonim, P Tamayo , etal: Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring Science 286: 531– 537,1999 Crossref, Medline, Google Scholar|
|25.||P Therasse, EA Eisenhauer, J Verweij: RECIST revisited: A review of validation studies on tumour assessment Eur J Cancer 42: 1031– 1039,2006 Crossref, Medline, Google Scholar|
|26.||H Harlin, Y Meng, AC Peterson , etal: Chemokine expression in melanoma metastases associated with CD8+ T-cell recruitment Cancer Res 69: 3077– 3085,2009 Crossref, Medline, Google Scholar|
|27.||B Mlecnik, M Tosolini, P Charoentong , etal: Biomolecular network reconstruction identifies T-cell homing factors associated with survival in colorectal cancer Gastroenterology 138: 1429– 1440,2010 Crossref, Medline, Google Scholar|
|28.||CJ Moran, DA Arenberg, CC Huang: RANTES expression is a predictor of survival in stage I lung adenocarcinoma Clin Cancer Res 8: 3803– 3812,2002 Medline, Google Scholar|
|29.||MC Dieu-Nosjean, M Antoine, C Danel , etal: Long-term survival for patients with non-small-cell lung cancer with intratumoral lymphoid structures J Clin Oncol 26: 4410– 4417,2008 Link, Google Scholar|
|30.||RR Ji, SD Chasalow, L Wang , etal: An immune-active tumor microenvironment favors clinical response to ipilimumab Cancer Immunol Immunother 61: 1019– 1031,2012 Crossref, Medline, Google Scholar|
|31.||L Zitvogel, O Kepp, G Kroemer: Immune parameters affecting the efficacy of chemotherapeutic regimens Nat Rev Clin Oncol 8: 151– 160,2011 Crossref, Medline, Google Scholar|
|32.||C Denkert, S Loibl, A Noske , etal: Tumor-associated lymphocytes as an independent predictor of response to neoadjuvant chemotherapy in breast cancer J Clin Oncol 28: 105– 113,2010 Link, Google Scholar|
|33.||F Pagès, J Galon, MC Dieu-Nosjean , etal: Immune infiltration in human tumors: A prognostic factor that should not be ignored Oncogene 29: 1093– 1102,2010 Crossref, Medline, Google Scholar|
|34.||L Martinet, I Garrido, T Filleron , etal: Human solid tumors contain high endothelial venules: Association with T- and B-lymphocyte infiltration and favorable prognosis in breast cancer Cancer Res 71: 5678– 5687,2011 Crossref, Medline, Google Scholar|
|35.||WH Fridman, J Galon, F Pagès , etal: Prognostic and predictive impact of intra- and peritumoral immune infiltrates Cancer Res 71: 5601– 5605,2011 Crossref, Medline, Google Scholar|
|36.||ML Ascierto, V De Giorgi, Q Liu , etal: An immunologic portrait of cancer J Transl Med 9: 146,2011 Crossref, Medline, Google Scholar|
We thank Muriel Debois and Marc Gillet who assisted with data collection and data analysis. We thank Prof Thierry Boon and Prof Michel Symann for the critical review of the manuscript. We thank Dr Patrick Therasse for the review of the clinical information for determination of clinical benefit in the melanoma data set. We are grateful to all of the patients, whose participation made the study possible, and all participating investigators.
Helene Servais from GlaxoSmithKline Vaccines developed the manuscript according to the recommendations, documentation, and outline provided by the lead authors. All authors critically reviewed the proposed drafts of the manuscript, and their comments were taken into account and incorporated by the corresponding author. All authors approved the content of the final version of the manuscript before it was submitted. Mrs Véronique Duquenne (CROMSOURCE c/o GlaxoSmithKline Vaccines) provided editorial assistance.
GlaxoSmithKline Vaccines was the funding source and was involved in all stages of the study/project conduct and analysis. The data were stored and analyzed by employees of the sponsor for the study in patients with non–small-cell lung cancer and by employees of the European Organisation for Research and Treatment of Cancer, funded by an educational grant from GlaxoSmithKline Vaccines for the study, in patients with melanoma. GlaxoSmithKline Vaccines also bore all costs associated with the development and the publishing of the present publication.