Advertisement

Development and validation of a 6-gene signature for the prognosis of loco-regional control in patients with HPV-negative locally advanced HNSCC treated by postoperative radio(chemo)therapy

  • Author Footnotes
    1 Shared first authorship.
    Shivaprasad Patil
    Footnotes
    1 Shared first authorship.
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Helmholtz-Zentrum Dresden - Rossendorf, Institute of Radiooncology - OncoRay, Dresden, Germany
    Search for articles by this author
  • Author Footnotes
    1 Shared first authorship.
    Annett Linge
    Footnotes
    1 Shared first authorship.
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany
    Search for articles by this author
  • Marianne Grosser
    Affiliations
    Institute of Pathology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany
    Search for articles by this author
  • Fabian Lohaus
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany
    Search for articles by this author
  • Volker Gudziol
    Affiliations
    Department of Otorhinolaryngology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    Department of Otorhinolaryngology, Head and Neck Surgery, Municipal Hospital Dresden, Germany
    Search for articles by this author
  • Max Kemper
    Affiliations
    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany

    Department of Otorhinolaryngology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany
    Search for articles by this author
  • Alexander Nowak
    Affiliations
    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany

    Department of Oral and Maxillofacial Surgery, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany
    Search for articles by this author
  • Dominik Haim
    Affiliations
    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany

    Department of Oral and Maxillofacial Surgery, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany
    Search for articles by this author
  • Inge Tinhofer
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Berlin, Germany

    Department of Radiooncology and Radiotherapy, Charité University Medicine Berlin, Germany
    Search for articles by this author
  • Volker Budach
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Berlin, Germany

    Department of Radiooncology and Radiotherapy, Charité University Medicine Berlin, Germany
    Search for articles by this author
  • Maja Guberina
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Essen, Germany

    Department of Radiotherapy, Medical Faculty, University of Duisburg-Essen, Essen, Germany
    Search for articles by this author
  • Martin Stuschke
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Essen, Germany

    Department of Radiotherapy, Medical Faculty, University of Duisburg-Essen, Essen, Germany
    Search for articles by this author
  • Panagiotis Balermpas
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Frankfurt, Germany

    Department of Radiotherapy and Oncology, Goethe-University Frankfurt, Germany
    Search for articles by this author
  • Claus Rödel
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Frankfurt, Germany

    Department of Radiotherapy and Oncology, Goethe-University Frankfurt, Germany
    Search for articles by this author
  • Henning Schäfer
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Freiburg, Germany

    Department of Radiation Oncology, Medical Center, Medical Faculty, University of Freiburg, Germany
    Search for articles by this author
  • Anca-Ligia Grosu
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Freiburg, Germany

    Department of Radiation Oncology, Medical Center, Medical Faculty, University of Freiburg, Germany
    Search for articles by this author
  • Amir Abdollahi
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Heidelberg, Germany

    Heidelberg Institute of Radiation Oncology (HIRO), National Center for Radiation Research in Oncology (NCRO), University of Heidelberg Medical School and German Cancer Research Center (DKFZ), Germany

    Heidelberg Ion Therapy Center (HIT), Department of Radiation Oncology, University of Heidelberg Medical School, Germany

    National Center for Tumor Diseases (NCT), University of Heidelberg Medical School and German Cancer Research Center (DKFZ), Germany

    Translational Radiation Oncology, University of Heidelberg Medical School and German Cancer Research Center (DKFZ), Germany
    Search for articles by this author
  • Jürgen Debus
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Heidelberg, Germany

    Heidelberg Institute of Radiation Oncology (HIRO), National Center for Radiation Research in Oncology (NCRO), University of Heidelberg Medical School and German Cancer Research Center (DKFZ), Germany

    Heidelberg Ion Therapy Center (HIT), Department of Radiation Oncology, University of Heidelberg Medical School, Germany

    National Center for Tumor Diseases (NCT), University of Heidelberg Medical School and German Cancer Research Center (DKFZ), Germany

    Clinical Cooperation Unit Radiation Oncology, University of Heidelberg Medical School and German Cancer Research Center (DKFZ), Germany
    Search for articles by this author
  • Ute Ganswindt
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Munich, Germany

    Department of Radiotherapy and Radiation Oncology, University Hospital, Ludwig-Maximilians-Universität, Munich, Germany
    Search for articles by this author
  • Claus Belka
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Munich, Germany

    Department of Radiotherapy and Radiation Oncology, University Hospital, Ludwig-Maximilians-Universität, Munich, Germany

    Clinical Cooperation Group Personalized Radiotherapy in Head and Neck Cancer, Helmholtz Zentrum Munich, Neuherberg, Germany
    Search for articles by this author
  • Steffi Pigorsch
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Munich, Germany

    Department of Radiation Oncology, Technische Universität München, Germany
    Search for articles by this author
  • Stephanie E. Combs
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Munich, Germany

    Department of Radiation Oncology, Technische Universität München, Germany

    Department of Radiation Sciences (DRS), Institut für Innovative Radiotherapie (iRT), Helmholtz Zentrum Munich, Neuherberg, Germany
    Search for articles by this author
  • Simon Boeke
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Tübingen, Germany

    Department of Radiation Oncology, Faculty of Medicine and University Hospital Tübingen, Eberhard Karls Universität Tübingen, Germany
    Search for articles by this author
  • Daniel Zips
    Affiliations
    German Cancer Consortium (DKTK), Partner Site Tübingen, Germany

    Tumour- and Normal Tissue Bank, University Cancer Centre (UCC), University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany
    Search for articles by this author
  • Gustavo B. Baretton
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany

    Institute of Pathology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    Tumour- and Normal Tissue Bank, University Cancer Centre (UCC), University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany
    Search for articles by this author
  • Michael Baumann
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Helmholtz-Zentrum Dresden - Rossendorf, Institute of Radiooncology - OncoRay, Dresden, Germany

    Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    German Cancer Research Center (DKFZ), Heidelberg, Germany
    Search for articles by this author
  • Mechthild Krause
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Helmholtz-Zentrum Dresden - Rossendorf, Institute of Radiooncology - OncoRay, Dresden, Germany

    Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany
    Search for articles by this author
  • Steffen Löck
    Correspondence
    Corresponding author at: Modeling and Biostatistics in Radiation Oncology Group, OncoRay – National Center for Radiation Research in Oncology Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Fetscherstraße 74, 01307 Dresden, Germany.
    Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany
    Search for articles by this author
  • Affiliations
    German Cancer Research Center (DKFZ), Heidelberg, Germany, and German Cancer Consortium (DKTK), Partner Site Dresden, Germany

    OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Helmholtz-Zentrum Dresden - Rossendorf, Germany

    Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Germany

    National Center for Tumor Diseases (NCT), Partner Site Dresden, Germany, German Cancer Research Center (DKFZ), Heidelberg, Germany, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, and Helmholtz Association/Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Germany
  • Author Footnotes
    1 Shared first authorship.

      Highlights

      • A novel 6-gene signature was developed based on full-transcriptome data for LRC prognosis in patients with HPV-negative HNSCC treated with PORT-C.
      • The prognostic performance was improved by adding relevant clinical parameters, CD44 expression, and the 15-gene hypoxia classifier.
      • The developed models were validated on an independent patient cohort.
      • Successful technical validation was performed by nanoString technology.

      Abstract

      Purpose

      The aim of this study was to develop and validate a novel gene signature from full-transcriptome data using machine-learning approaches to predict loco-regional control (LRC) of patients with human papilloma virus (HPV)-negative locally advanced head and neck squamous cell carcinoma (HNSCC), who received postoperative radio(chemo)therapy (PORT-C).

      Materials and methods

      Gene expression analysis was performed using Affymetrix GeneChip Human Transcriptome Array 2.0 on a multicentre retrospective training cohort of 128 patients and an independent validation cohort of 114 patients from the German Cancer Consortium - Radiation Oncology Group (DKTK-ROG). Genes were filtered based on differential gene expression analyses and Cox regression. The identified gene signature was combined with clinical parameters and with previously identified genes related to stem cells and hypoxia. Technical validation was performed using nanoString technology.

      Results

      We identified a 6-gene signature consisting of four individual genes CAV1, GPX8, IGLV3-25, TGFBI, and one metagene combining the highly correlated genes INHBA and SERPINE1. This signature was prognostic for LRC on the training data (ci = 0.84) and in validation (ci = 0.63) with a significant patient stratification into two risk groups (p = 0.005). Combining the 6-gene signature with the clinical parameters T stage and tumour localisation as well as the cancer stem cell marker CD44 and the 15-gene hypoxia-associated signature improved the validation performance (ci = 0.69, p = 0.001).

      Conclusion

      We have developed and validated a novel prognostic 6-gene signature for LRC of HNSCC patients with HPV-negative tumours treated by PORT-C. After successful prospective validation the signature can be part of clinical trials on the individualization of radiotherapy.

      Keywords

      Patients with locally advanced head and neck squamous cell carcinoma (HNSCC) are currently treated with postoperative radio-chemotherapy (PORT-C) or primary radio-chemotherapy. While the inclusion of chemotherapy improved the outcome significantly [
      • Cooper J.S.
      • Pajak T.F.
      • Forastiere A.A.
      • Jacobs J.
      • Campbell B.H.
      • Saxman S.B.
      • et al.
      Postoperative concurrent radiotherapy and chemotherapy for high-risk squamous-cell carcinoma of the head and neck.
      ,
      • Lavaf A.
      • Genden E.M.
      • Cesaretti J.A.
      • Packer S.
      • Kao J.
      Adjuvant radiotherapy improves overall survival for patients with lymph node-positive head and neck squamous cell carcinoma.
      ], treatment response is still heterogeneous. To further individualize therapy in order to improve treatment response, additional biomarkers need to be identified and applied in personalized treatment escalation or de-escalation strategies [
      • Rades D.
      • Seibold N.D.
      • Gebhard M.P.
      • Noack F.
      • Schild S.E.
      • Thorns C.
      Prognostic factors (including HPV status) for irradiation of locally advanced squamous cell carcinoma of the head and neck (SCCHN)Prognosefaktoren (inklusive HPV-Status) bei der Bestrahlung lokal fortgeschrittener Plattenepithelkarzinome im Kopf-Hals-Bereich (SCCHN).
      ,
      • Baumann M.
      • Krause M.
      • Overgaard J.
      • Debus J.
      • Bentzen S.M.
      • Daartz J.
      • et al.
      Radiation oncology in the era of precision medicine.
      ].
      Human papilloma virus (HPV) associated HNSCC have shown a very high rate of loco-regional control (LRC) and overall survival (OS) after PORT-C [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ], suggesting that a number of patients are currently being overtreated. Hence, dose de-escalation trials are currently ongoing in patients with HPV-positive oropharyngeal carcinoma with the overall aim to reduce radiation-induced side effects while achieving similar response rates [

      De-escalation of Adjuvant Radio (Chemo) Therapy for HPV-positive Head-neck Squamous Cell Carcinomas (DELPHI). ClinicalTrials.gov [Internet], Bethesda (MD): National Library of Medicine (US), 2000 Feb 29. Identifier NCT03396718 2018:[registered 2018 Jan 11, updated 2019 Feb 18].

      ,
      • Seiwert T.Y.
      • Foster C.C.
      • Blair E.A.
      • Karrison T.G.
      • Agrawal N.
      • Melotek J.M.
      • et al.
      Optima: A phase II dose and volume de-escalation trial for human papillomavirus-positive oropharyngeal cancer.
      ,
      • Nichols A.C.
      • Lang P.
      • Prisman E.
      • Berthelet E.
      • Tran E.
      • Hamilton S.
      • et al.
      Treatment de-escalation for HPV-associated oropharyngeal squamous cell carcinoma with radiotherapy vs. trans-oral surgery (ORATOR2): Study protocol for a randomized phase II trial.
      ]. However, the treatment response of HPV-negative tumours is still heterogeneous and novel biomarkers are needed to distinguish between patients at low and high risk of loco-regional recurrence or distant metastases (DM).
      Developments in the field of omics technologies over the past decades have provided an invaluable opportunity to study the molecular mechanisms of cancer and identify prognostic or predictive biomarkers, e.g., transcriptomic data has been used to identify promising gene signatures that predict recurrence and response to therapy [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ,
      • Scott J.G.
      • Berglund A.
      • Schell M.J.
      • Mihaylov I.
      • Fulp W.J.
      • Yue B.
      • et al.
      A genome-based model for adjusting radiotherapy dose (GARD): a retrospective, cohort-based study.
      ,
      • Kim H.S.
      • Kim S.C.
      • Kim S.J.
      • Park C.H.
      • Jeung H.-C.
      • Kim Y.B.
      • et al.
      Identification of a radiosensitivity signature using integrative metaanalysis of published microarray data for NCI-60 cancer cells.
      ,
      • Chung C.H.
      • Parker J.S.
      • Ely K.
      • Carter J.
      • Yi Y.
      • Murphy B.A.
      • et al.
      Gene expression profiles identify epithelial-to-mesenchymal transition and activation of nuclear factor-κB Signaling as characteristics of a high-risk head and neck squamous cell carcinoma.
      ]. Most prognostic gene signatures for patients with locally advanced HNSCC have been developed based on data from treatment with primary radiotherapy, while for patients treated with PORT-C only a very few prognostic gene signatures are existing. Previously, we have shown that the cancer stem cell markers CD44 and SLC3A2 as well as the 15-gene [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ] and 26-gene [
      • Eustace A.
      • Mani N.
      • Span P.N.
      • Irlam J.J.
      • Taylor J.
      • Betts G.N.J.
      • et al.
      A 26-gene hypoxia signature predicts benefit from hypoxia-modifying therapy in laryngeal cancer but not bladder cancer.
      ] hypoxia-associated signatures were related to LRC after PORT-C [
      • Linge A.
      • Löck S.
      • Gudziol V.
      • Nowak A.
      • Lohaus F.
      • von Neubeck C.
      • et al.
      Low cancer stem cell marker expression and low hypoxia identify good prognosis subgroups in HPV(-) HNSCC after postoperative radiochemotherapy: A multicenter study of the DKTK-ROG.
      ]. The value of CD44 and the 15-gene hypoxia signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ] was successfully validated on an independent cohort [
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ]. Furthermore, we developed and validated a 7-gene signature using the expression of a limited set of 178 pre-defined genes evaluated by the nanoString nCounter system [
      • Schmidt S.
      • Linge A.
      • Zwanenburg A.
      • Leger S.
      • Lohaus F.
      • Krenn C.
      • et al.
      Development and validation of a gene signature for patients with head and neck carcinomas treated by postoperative radio(chemo)therapy.
      ].
      Although the number of gene signatures keeps increasing with technological advancements, few have clinical utility. The reasons for this include: failure to account for differences of signature levels between patients with and without event (loco-regional failure, overall survival etc.), missing independent validation in multiple datasets and limited improvement of clinical outcome [
      • Teutsch S.M.
      • Bradley L.A.
      • Palomaki G.E.
      • Haddow J.E.
      • Piper M.
      • Calonge N.
      • et al.
      The evaluation of genomic applications in practice and prevention (EGAPP) initiative: Methods of the EGAPP working group.
      ,
      • Michiels S.
      • Ternès N.
      • Rotolo F.
      Statistical controversies in clinical research: prognostic gene signatures are not (yet) useful in clinical practice.
      ]. The performance of gene signatures can be enhanced by their combination with clinical features [
      • Pittman J.
      • Huang E.
      • Dressman H.
      • Horng C.-F.
      • Cheng S.H.
      • Tsou M.-H.
      • et al.
      Integrated modeling of clinical and gene expression information for personalized prediction of disease outcomes.
      ] or by combining multiple gene signatures related to different biological mechanisms [
      • Zhao X.I.
      • Rødland E.A.
      • Sørlie T.
      • Naume B.
      • Langerød A.
      • Frigessi A.
      • et al.
      Combining gene signatures improves prediction of breast cancer survival.
      ].
      Therefore, in this study we (i) developed and validated a novel gene signature from full-transcriptome data using machine-learning approaches to predict LRC for patients with locally advanced HPV-negative HNSCC, who were treated with PORT-C; (ii) combined our signature with clinical parameters; and (iii) extended the signature by including the genes of the 15-gene hypoxia-associated signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ] and the cancer stem cell marker CD44, which were previously validated for association with LRC [
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ]. The analyses were based on the gene expression data of 25,246 genes evaluated by the Affymetrix GeneChip system. Technical validation of the gene expression data was performed using nanoString nCounter technology.

      Materials and methods

      Patient data

      For the purpose of this retrospective study, we used two different cohorts of patients with locally advanced HNSCC that were presented previously [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ,
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ,
      • Schmidt S.
      • Linge A.
      • Zwanenburg A.
      • Leger S.
      • Lohaus F.
      • Krenn C.
      • et al.
      Development and validation of a gene signature for patients with head and neck carcinomas treated by postoperative radio(chemo)therapy.
      ]. As the training cohort, we used the postoperative HNSCC cohort of the German Cancer Consortium - Radiation Oncology Group (DKTK-ROG), which consists of 221 patients treated with PORT-C between 2004 and 2012 in 9 different institutions [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ]. The validation cohort consisted of 152 additional patients, who were treated with PORT or PORT-C between 1999 and 2006 at the DKTK partner site Dresden [
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ]. All patients received surgery followed by postoperative radio(chemo)therapy. Patients meeting the following criteria were included: histologically proven squamous cell carcinoma, curatively intended cisplatin-based PORT-C/PORT according to standard protocols covering the former tumour region and the neck nodes. Inclusion criteria, data collection, handling and analyses of biomaterial were described previously [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ,
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ].
      It was shown that patients with HPV-DNA positive tumours have a very good prognosis [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ]. Thus, in this study, we only considered patients with HPV16-DNA negative tumours. This reduced the training cohort to the final sample size of 128 patients and the validation cohort to 114 patients. Patient characteristics are presented in Table 1 and the study design is shown in Supplementary Fig. 1.
      Table 1Patient characteristics for the training and validation cohort. IQR: Inter-quartile range, R: Residual tumour, ECE: Extracapsular extension, UICC: Union for International Cancer Control.
      Training cohort (2004–2011)Validation cohort (1999–2006)
      CharacteristicsMedian (IQR)Median (IQR)
      Follow-up (months)41.7 (17.8–59.4)35.1 (9.5–61.4)
      Age (years)56.5 (50.0–63.0)52.4 (46.8–60.7)
      Dose (Gy)64.0 (63.0–66.0)64.0 (60.0–64.0)
      Treatment time (days)44.0 (43.0–48.0)43.0 (41.0–44.0)
      Number of pts%Number of pts%
      Gender
       Male/female99/2977.3/22.799/1586.8/13.2
      Smoking status
       Never/current or former/missing7/83/385.5/64.8/29.78/81/257.0/71.1/21.9
      Alcohol status
       Never/current or former/missing13/64/5110.2/50.0/39.89/81/247.9/71.1/21.0
      Tumour localization
       Oral cavity/oropharynx/hypopharynx/ larynx48/59/21/037.5/46.1/16.4/070/25/12/761.5/21.9/10.5/6.1
      Tumour grade
       G1/G2/G34/82/423.1/64.1/32.83/62/492.6/54.4/43.0
      R status
       0/1/missing73/55/057.0/43.0/082/23/971.9/20.2/7.9
      ECE status
       0/1/missing61/67/047.7/52.3/076/37/166.6/32.5/0.9
      UICC stage (2010)
       1/2/3/40/4/23/1010/3/1/18.0/78.92/3/29/801.8/2.6/25.4/70.2
      T stage
       1/2/3/424/49/34/2118.8/38.3/26.5/16.427/51/20/1623.7/44.7/17.5/14.1
      N stage
       0/1/2/314/20/80/1410.9/15.7/62.5/10.913/26/73/211.4/22.8/64.0/1.8
      Chemotherapy
       Yes/no128/0100/028/8624.6/75.4
      Loco-regional failure2721.13228.1
      Distant metastases3124.22723.7
      Death from all causes5341.47061.4
      Figure thumbnail gr1
      Fig. 1Scheme for identifying the 6-gene signature. Univariable Cox regression and differential gene expression analysis were performed over 100 bootstraps of the training cohort. Genes were filtered based on the fold-change value and their false discovery rate (FDR) adjusted p-values. Six different feature selection methods were applied afterwards. Feature selection was performed over 1000 bootstraps on the training cohort. Finally, the occurrence of the genes was used to define the 6-gene signature.
      Disease status and first site of relapse were evaluated by the respective treating institutions. An experienced radiation oncologist (F.L) reviewed the radiotherapy treatment plan and radiological images of the recurrence for every loco-regional failure. Formalin-fixed and paraffin-embedded (FFPE) blocks of the resected tumour specimens were centrally collected at the DKTK partner site Dresden. Extraction of genomic DNA and total RNA were performed as described previously [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ,
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ]. Ethical approval for multicentre retrospective analyses of clinical and biological data was obtained from the Ethics Committees of all DKTK partner sites.

      DNA extraction and PCR array-based analyses of HPV status

      DNA extraction and PCR-array based analyses of HPV status was performed as described previously [
      • Lohaus F.
      • Linge A.
      • Tinhofer I.
      • Budach V.
      • Gkika E.
      • Stuschke M.
      • et al.
      HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
      ]. Briefly, the QIAamp DNA FFPE tissue kit (Qiagen GmbH, Hilden, DE) was used to extract genomic DNA. The LCD-Array HPV 3.5 kit (CHIPRON GmbH, Berlin, DE) was used to assess HPV DNA positivity and to determine the HPV DNA genotype.

      Microarray analysis

      Whole transcriptome analysis was performed using the Affymetrix GeneChip Human Transcriptome 2.0 Array (Thermo Fisher Scientific Inc., Waltham, MA, USA). Probe-level intensity files were used to perform quality control in Transcriptome Analysis Console (TAC) 4.0 (Applied Biosystems, Waltham, Massachusetts, USA) as per manufacturer’s instructions. Data normalization was performed using the Signal Space Transformation in conjunction with the Robust Multiarray Average method (SST-RMA). Since the training and validation cohort data was collected during different time intervals, batch normalization was performed using the ComBat method [
      • Johnson W.E.
      • Li C.
      • Rabinovic A.
      Adjusting batch effects in microarray expression data using empirical Bayes methods.
      ] to adjust for batch effects between the two cohorts. Coding genes were filtered for further analysis.

      nanoString analysis

      For technical validation of the developed microarray-based gene-signature, gene expression analyses was performed using nanoString elements technology (nanoString Technologies, Seattle, WA, USA) as previously described [
      • Linge A.
      • Löck S.
      • Gudziol V.
      • Nowak A.
      • Lohaus F.
      • von Neubeck C.
      • et al.
      Low cancer stem cell marker expression and low hypoxia identify good prognosis subgroups in HPV(-) HNSCC after postoperative radiochemotherapy: A multicenter study of the DKTK-ROG.
      ,
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ]. Briefly, total RNA along with reporter and capture probes specific to the genes of interest were mixed and incubated at 62 °C for 22 hours. Then the samples were kept at 4 °C for a maximum of 18 hours and were subjected to the nCounter system. The raw counts were logarithmised and then normalized by subtracting the mean of the log-transformed counts of the seven reference genes ACTR3, B2M, GNB2L1, NDFIP1, POLR2A, RPL11, and RPL37A. Two samples of the training cohort were omitted from the analysis due to insufficient tumour material or too low RNA yield, i.e. 126 and 114 patients were available for training and validation, respectively.

      Clinical endpoints and general statistical analysis

      Loco-regional control (LRC) was the primary endpoint of this study. Overall survival (OS) and freedom from distant metastases (DM) were the secondary endpoints. The endpoints were calculated from the first day of radiotherapy to the date of event or censoring. The Kaplan-Meier method was used to estimate survival curves, which were compared using log-rank tests. Mann-Whitney-U tests for continuous variables and chi-squared (χ2) tests for categorical variables were used to evaluate differences between the training and validation cohort. Cox regression was used to test the prognostic value of the genes and clinical features for the endpoints. The validity of the proportional hazards assumption of the Cox models was assessed through the chi-squared test for the fit of the Schoenfeld residuals. The described statistical tests were performed using Python (Python Software Foundation. Python Language Reference, version 3.7) and SPSS 25 software (IBM Corporation, Armonk, NY, USA). For all analyses, two-sided tests were performed and p-values < 0.05 were considered as statistically significant.

      Statistical framework to identify gene signatures and perform model predictions

      To identify the gene signature, the following steps were carried out: (i) The gene expression data of the training cohort was z-transformed to mean 0 and standard deviation of 1. Means and standard deviations from the training cohort were used to transform the gene expression data of the validation cohort. (ii) Differential gene expression (DGE) analysis was performed on the training cohort, between patients with LRC and patients without LRC until 24 months after treatment that had a minimum follow-up of 24 months. We ended up with 108 patients that fit this criterion. DGE analysis was performed over 100 bootstraps of the training cohort using the Limma R package [
      • Ritchie M.E.
      • Phipson B.
      • Wu D.i.
      • Hu Y.
      • Law C.W.
      • Shi W.
      • et al.
      Limma powers differential expression analyses for RNA-sequencing and microarray studies.
      ]. (iii) Univariable Cox regression analysis was performed on the training cohort based on 100 bootstraps to filter genes with high prognostic value for LRC. (iv) Genes that in at least 40% of all bootstraps fulfilled the following criteria were selected: (a) a fold-change (FC) of ≥ 1.5 and false discovery rate (FDR) adjusted p-values of ≤ 0.1 in the DGE analysis and in univariable Cox regression, or (b) a FC of ≥ 1.5 and genes were part of the 838 cancer-related genes from the COSMIC database [
      • Tate J.G.
      • Bamford S.
      • Jubb H.C.
      • Sondka Z.
      • Beare D.M.
      • Bindal N.
      • et al.
      COSMIC: The Catalogue Of Somatic Mutations In Cancer.
      ].
      To develop a reduced gene signature for the prognosis of LRC, we applied additional feature selection algorithms to the selected genes on the training cohort. Six feature selection methods (minimum redundancy maximum relevance (MRMR), mutual information matrix (MIM), Pearson correlation, Spearman correlation, elastic-net Cox regression, univariable Cox regression) were applied on 1000 bootstraps of the training cohort. For every feature selection algorithm, the five genes that most often occurred in the bootstraps were considered. The genes that were present in ≥50% of the feature selection methods were selected for further analysis. If two genes had the same number of occurrences, both of them were selected. To increase robustness of the signature, genes that were highly correlated in the training cohort (Spearman correlation coefficient r ≥ 0.8) were combined to create a new metagene, using the median expression of the individual genes. A schematic representation of the procedure is illustrated in Fig. 1. Finally, the selected genes were included in a multivariable Cox model for LRC to develop a risk score for each patient of the training cohort. The risk score was defined by: ∑ coefficient of the gene in the multivariable Cox model (βi) × expression value of the gene (xi). For patient stratification into groups at low and high risk of loco-regional recurrence, an optimal risk score was calculated using the maximally selected rank statistics (maxstat) R package [

      Hothorn T. maxstat: Maximally Selected Rank Statistics. R Packag Version 07-25 Https//CRANR-ProjectOrg/Package=maxstat 2017.

      ]. The median value of 1000 bootstraps on the training cohort was used, leading to a binary gene classifier.
      To further improve the prognostic value of the developed gene signature, additional multivariable Cox regression models were developed including clinical parameters and previously identified biomarkers. The developed gene classifier was combined with (i) those clinical parameters that were significantly associated with LRC in univariable Cox regression, (ii) the cancer stem-cell marker CD44 and the 15-gene hypoxia-associated signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ] that have previously been validated [
      • Linge A.
      • Löck S.
      • Gudziol V.
      • Nowak A.
      • Lohaus F.
      • von Neubeck C.
      • et al.
      Low cancer stem cell marker expression and low hypoxia identify good prognosis subgroups in HPV(-) HNSCC after postoperative radiochemotherapy: A multicenter study of the DKTK-ROG.
      ,
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ], and (iii) the combination of (i) and (ii). For these analyses, a CD44 classifier was built on the training cohort using an optimal expression value, determined as described above. For the 15-gene hypoxia-associated signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ], k-means clustering was performed on the 15 genes on the training cohort and the cluster centres were transferred to perform classification on the validation cohort. Risk scores and optimal stratification cut-offs were determined as outlined for the developed gene signature.
      For independent validation, coefficients of Cox regression and the optimal risk scores of the training cohort were applied to the validation cohort unchanged. To evaluate the prognostic performance of the developed models we calculated the concordance index (ci). The 95% confidence interval (CI) of the ci was estimated from 1000 bootstrap samples of the validation cohort and validation was declared successful if the 95% CI did not contain 0.5. Differences in discriminatory performance between the different models were tested based on 1000 bootstraps for both cohorts. A heuristic p-value was determined by assessing the fraction of bootstrap runs for which the baseline model (clinical only or genes only) had a higher ci than the combined model (genes + clinical). For technical validation of the developed gene signature, the same analyses were repeated based on the gene expression data obtained by nanoString technology.

      Results

      Patient data, treatment parameters, and tumour characteristics of both cohorts are summarized in Table 1. PORT-C was the standard treatment for all patients in the training cohort, while patients in the validation cohort received PORT (n = 86) or PORT-C (n = 28). The training cohort mainly included patients with oropharyngeal (46.1%) and oral cavity carcinomas (37.5%). In the validation cohort, 21.9% of patients were diagnosed with oropharyngeal and 61.5% with oral cavity carcinomas. Patients in the training cohort had somewhat higher LRC and OS than patients in the validation cohort, while DM was similar (Supplementary Fig. 2).
      Figure thumbnail gr2
      Fig. 2Patient stratification by the 6-gene signature combined with the clinical parameters tumour localization and T stage (A, B) and by the extended gene signature (6-gene signature, CD44 and the 15-gene hypoxia-associated signature) combined with the same clinical parameters (C, D) for loco-regional tumour control (LRC). p-values originate from log-rank tests. Heatmap of the 6-gene signature (E) and the extended gene signature (F) as well as tumour localization (oral cavity, dark; others, light), T stage status (1–2, light; 3–4, dark), risk group (low risk, light; high risk, dark) and LRC during follow-up (yes, light; no, dark) for the training and the validation cohort.
      Based on differential gene-expression analysis, Cox regression, and the COSMIC database [
      • Tate J.G.
      • Bamford S.
      • Jubb H.C.
      • Sondka Z.
      • Beare D.M.
      • Bindal N.
      • et al.
      COSMIC: The Catalogue Of Somatic Mutations In Cancer.
      ], 34 genes related to LRC were selected (Supplementary Table 1). Additional feature selection (Supplementary Table 2) further reduced this number to the finally selected six genes CAV1, GPX8, IGLV3-25, INHBA, SERPINE1, and TGFBI. Due to their high correlation (r = 0.80), the median of the expression values of INHBA and SERPINE1 was used to form a new metagene. From multivariable Cox regression, an individual risk score (RS) was calculated for every patient: RS=(0.140×CAV1+0.527×GPX8-0.425×IGLV3_25+0.363×metagene_INHBA_SERPINE1+0.235×TGBFI). An optimal risk score of 0.90 was used as the cut-off for stratifying patients into groups at high and low risk of loco-regional recurrence. The resulting 6-gene signature was significantly associated to LRC in training [ci = 0.84, 95% confidence interval (0.76–0.92)], and in validation [ci = 0.63 (0.53–0.73), Table 2] and showed a significant patient stratification (training: p < 0.001, validation: p = 0.005, Supplementary Fig. 3).
      Table 2Multivariable Cox regression of loco-regional tumour control for the 6-gene signature, the extended gene signature, clinical parameters and their combinations. HR: hazard ratio, CI: confidence interval.
      ParameterCoefficient (ß)HR (95 % CI)p-valueci Training (95 % CI)ci Validation (95 % CI)
      6-gene signature0.84 (0.76–0.92)0.63 (0.53–0.73)
      CAV10.141.15 (0.67–1.96)0.61
      GPX80.531.69 (1.10–2.61)0.017
      IGLV3-25−0.430.65 (0.38–1.12)0.12
      Metagene INHBA + SERPINE10.361.44 (0.75–2.76)0.28
      TGFBI0.231.26 (0.68–2.37)0.46
      Extended gene signature0.82 (0.74–0.90)0.67 (0.57–0.77)
      6-gene classifier (≥0.90 vs < 0.90)2.178.76 (3.70–20.71)<0.001
      CD44 classifier (≥-0.12 vs < 0.12)1.113.04 (1.00–9.26)0.050
      Hypoxia classifier (high vs low hypoxia)0.531.69 (0.76–3.76)0.20
      Clinical parameters0.69 (0.59–0.79)0.62 (0.52–0.72)
      Tumour localization (oral cavity vs others)0.772.17 (1.01–4.65)0.047
      T stage (3,4 vs 1,2)0.852.34 (1.08–5.09)0.032
      6-gene signature and clinical parameters0.82 (0.74–0.90)0.66 (0.56–0.76)
      6-gene classifier (≥0.90 vs < 0.90)2.279.68 (3.99–23.47)<0.001
      Tumour localization (oral cavity vs others)0.111.12 (0.50–2.49)0.78
      T stage (3,4 vs 1,2)0.671.95 (0.88–4.31)0.10
      Extended gene signature and clinical parameters0.83 (0.77–0.89)0.69 (0.59–0.79)
      6-gene classifier (≥0.90 vs < 0.90)2.067.88 (3.15–19.69)<0.001
      CD44 classifier (≥-0.12 vs < 0.12)1.223.40 (1.09–10.54)0.035
      Hypoxia classifier (high vs low hypoxia)0.391.47 (0.63–3.47)0.38
      Tumour localization (oral cavity vs others)0.181.19 (0.52–2.76)0.70
      T stage (3,4 vs 1,2)0.691.99 (0.86–4.60)0.11
      Figure thumbnail gr3
      Fig. 3Patient stratification by the 6-gene signature (A-D) and the extended gene signature (E-H) combined with the clinical parameters tumour localization and T stage for overall survival (left) and for freedom from distant metastases (right). p-values originate from log-rank tests.
      On the training cohort, tumour localization (oral cavity vs others) and T stage were significantly associated to LRC in univariable Cox regression (Supplementary Table 3). Age was not considered, since the result that lower age was related to more loco-regional recurrences seemed questionable. A multivariable model including the two clinical parameters tumour localization and T stage showed a validation performance similar to the 6-gene signature [training: ci = 0.69 (0.59–0.79), validation: ci = 0.62 (0.52–0.72); Table 2]. Combining the two significant clinical parameters and the 6-gene signature resulted in an improved performance [training: ci = 0.82 (0.74–0.90), validation: ci = 0.66 (0.56–0.76), Table 2, p = 0.081]. The optimal risk score for the combined model (2.27) was used for stratifying patients into groups at high and low risk of loco-regional recurrence. This stratification led to significant differences in LRC for the training (p < 0.001) and for the validation cohort (p = 0.005), see Fig. 2 A, B and E.
      The 6-gene classifier was extended by the CD44 and the 15-gene hypoxia-associated classifier. Multivariable Cox regression revealed a ci of 0.82 (0.74–0.90) on the training cohort and of 0.67 (0.57–0.77) in validation for the prognosis of LRC (Table 2) with a significant patient stratification (Supplementary Fig. 3). The multivariable model additionally including the clinical parameters T stage and localization resulted in the highest validation performance [training: ci = 0.83 (0.77–0.89), validation: ci = 0.69 (0.59–0.79), Table 2]. This performance was somewhat higher than that of the clinical model alone (p = 0.075, statistical trend). Based on the optimal risk score of this model (3.28), significant differences in LRC were observed for the training and validation cohort (p = 0.001), see Fig. 2 C, D, and F. Means and standard deviations of the expression values are given in Supplementary Table 4 for both cohorts. Spearman correlations between the considered genes and clinical parameters are shown in Supplementary Figure 4. While higher correlations up to 0.8 were observed between the genes of the 6-gene signature, correlations to the clinical parameters tumour localisation and T stage were moderate.
      The 6-gene signature and the extended gene signature determined for LRC, along with the clinical features tumour localization (oral cavity vs others) and T stage were trained and validated for OS and DM (Supplementary Tables 5 and 6, respectively). For OS, the gene classifier only moderately added to the performance of the clinical features (Supplementary Table 5). For DM, the extended gene classifier combined with the clinical parameters showed the best validation performance with a substantial contribution from the gene classifier (Supplementary Table 6). Corresponding Kaplan-Meier curves are presented in Fig. 3. We repeated the presented analysis of OS and DM using the clinical features significantly associated to the respective endpoint (Supplementary Table 3) instead of tumour localization and T stage. Similar results were observed (Supplementary Tables 7 and 8).
      We technically validated our results using nanoString nCounter technology. The Spearman correlation between the gene expression data obtained by nanoString analyses and Affymetrix analyses was evaluated for the 22 genes included in the extended gene signature. The median correlation was 0.78 (range −0.15 to 0.91, Supplementary Figure 5). Except for CD44 all genes showed a correlation > 0.50. Repeating the same analysis workflow as presented for the Affymetrix data led to very similar results with almost identical Kaplan-Meier estimates of LRC (Supplementary Table 9 and Supplementary Figure 6). Means and standard deviations of the gene expressions are given in Supplementary Table 10. The proportional hazards assumption was fulfilled for all presented models for LRC except for the 6-gene signature combined with clinical parameters based on the Affymetrix platform.

      Discussion

      The overall aim of our study was to develop and validate a novel gene signature from full-transcriptome data using machine-learning approaches to predict LRC of patients with HPV-negative locally advanced HNSCC who received PORT-C. We have identified and successfully validated a 6-gene signature containing the genes CAV1, GPX8, IGLV3-25, INHBA, SERPINE1, and TGFBI. We showed that the prognostic power could be further increased by combining the 6-gene signature with relevant clinical parameters and by extending it with the additional validated biomarkers CD44 and the 15-gene hypoxia signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ].
      Five of the genes of the 6-gene signature have known associations to cancer: CAV1 (Caveolin-1) is a major structural protein which is involved in multiple cancer-associated processes, including tumour growth, cell migration, cell death and angiogenesis [
      • Goetz J.G.
      • Lajoie P.
      • Wiseman S.M.
      • Nabi I.R.
      Caveolin-1 in tumor progression: The good, the bad and the ugly.
      ]. A loss-of-function study using si-CAV1 in HNSCC cell lines revealed that CAV1 mediates cell migration and invasion. The study also showed that miR133a overexpression downregulated CAV1, suggesting this miRNA could be used as a therapeutic target [
      • Nohata N.
      • Hanazawa T.
      • Kikkawa N.
      • Mutallip M.
      • Fujimura L.
      • Yoshino H.
      • et al.
      Caveolin-1 mediates tumor cell migration and invasion and its regulation by miR-133a in head and neck squamous cell carcinoma.
      ]. GPX8 (Glutathione peroxidase 8) is a redox enzyme that resides in endoplasmic reticulum. In breast cancer, GPX8 is essential in regulating tumour aggressiveness and is associated with poor prognosis [
      • Khatib A.
      • Solaimuthu B.
      • Ben Yosef M.
      • Abu Rmaileh A.
      • Tanna M.
      • Oren G.
      • et al.
      The glutathione peroxidase 8 (GPX8)/IL-6/STAT3 axis is essential in maintaining an aggressive breast cancer phenotype.
      ]. IGLV3-25 (Immunoglobulin lambda variable 3–25) encodes variable domain of immunoglobulin light chains that participate in antigen recognition. IGLV3-21 (Immunoglobulin lambda variable 3–21) defines a new poor prognosis subgroup in chronic lymphocytic leukaemia patients [
      • Stamatopoulos B.
      • Smith T.
      • Crompot E.
      • Pieters K.
      • Clifford R.
      • Mraz M.
      • et al.
      The light chain IgLV3-21 defines a new poor prognostic subgroup in chronic lymphocytic leukemia: Results of a multicenter study.
      ]. But the role of variable domain of immunoglobulin light chains is not well defined in HNSCC. In this study, we suggest a prognostic ability of IGLV3-25 in HNSCC. The INHBA (Inhibin, beta A) gene encodes an individual form the TGF-ß superfamily of proteins. INHBA is overexpressed in oral squamous cell carcinoma (OSCC) and is significantly correlated with regional lymph node metastases. Patients with high expression of INHBA show shortened survival. miR-143 and miR-145 regulate the expression of INHBA, making it a therapeutic target [
      • Bufalino A.
      • Cervigne N.K.
      • de Oliveira C.E.
      • Fonseca F.P.
      • Rodrigues P.C.
      • Macedo C.C.S.
      • et al.
      Low miR-143/miR-145 cluster levels induce activin a overexpression in oral squamous cell carcinomas, which contributes to poor prognosis.
      ]. SERPINE1 (Serpin Family E Member 1) encodes a member of the serine protease inhibitor superfamily. Overexpression of SERPINE1 in patients with HNSCC enhances tumour cell migration and invasion. It also plays a role in the development of metastases and poor prognosis [
      • Pavón M.A.
      • Arroyo-Solera I.
      • Céspedes M.V.
      • Casanova I.
      • León X.
      • Mangues R.
      uPA/uPAR and SERPINE1 in head and neck cancer: Role in tumor resistance, metastasis, prognosis and therapy.
      ]. Expression of SERPINE1 is associated with radiation resistance [
      • Lee Y.-C.
      • Yu C.-C.
      • Lan C.
      • Lee C.-H.
      • Lee H.-T.
      • Kuo Y.-L.
      • et al.
      Plasminogen activator inhibitor-1 as regulator of tumor-initiating cell properties in head and neck cancers.
      ], activation of hypoxia-related factors [
      • Sun Z.-J.
      • Yu G.-T.
      • Huang C.-F.
      • Bu L.-L.
      • Liu J.-F.
      • Ma S.-R.
      • et al.
      Hypoxia induces TFE3 expression in head and neck squamous cell carcinoma.
      ], and increased cisplatin resistance of head and neck tumour cells [
      • Pavón M.A.
      • Arroyo-Solera I.
      • Téllez-Gabriel M.
      • León X.
      • Virós D.
      • López M.
      • et al.
      Enhanced cell migration and apoptosis resistance may underlie the association between high SERPINE1 expression and poor outcome in head and neck carcinoma patients.
      ]. TGFBI (Transforming Growth Factor Beta Induced) gene encodes a protein that becomes a part of extracellular matrix and helps in cell adhesion and cell migration. TGFBI overexpression promotes OSCC and is associated with poor prognosis [
      • Wang B.-J.
      • Chi K.-P.
      • Shen R.-L.
      • Zheng S.-W.
      • Guo Y.
      • Li J.-F.
      • et al.
      TGFBI promotes tumor growth and is associated with poor prognosis in oral squamous cell carcinoma.
      ].
      The prognostic value of the IGLV3-25 gene has not been reported so far. In our study, overexpression of IGLV3-25 was associated with better prognosis, which needs further confirmation. The IGLV3-25 gene participates in the recognition phase of humoral immunity triggering the clonal expansion and differentiation of B lymphocytes into immunoglobulins-secreting plasma cells [
      • Lefranc M.P.
      Immunoglobulin and T cell receptor genes: IMGT® and the birth and rise of immunoinformatics.
      ]. A recent study suggests that cancer can trigger acquired humoral immunity as a response against the developing tumour [

      Joseph CG, Darrah E, Shah AA, Skora AD, Livia A, Wigley FM, et al. Association of the Autoimmune Disease Scleroderma with an Immunologic Response to Cancer 2014;343:152–7. doi:10.1126/science.1246886.Association.

      ]. This may be related to our finding that overexpression of the IGLV3-25 gene was associated to better prognosis.
      Gene Ontology revealed that the six identified genes were enriched in biological processes like apoptosis, cell adhesion, and extracellular organization. In addition, results of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis revealed that the genes were enriched in p53 signalling pathway, TGF-beta signalling pathway, focal adhesion etc. Our result that a higher expression of CAV1, GPX8, INHBA, SERPINE1, and TGFBI is associated to poor prognosis is in line with the literature [
      • Goetz J.G.
      • Lajoie P.
      • Wiseman S.M.
      • Nabi I.R.
      Caveolin-1 in tumor progression: The good, the bad and the ugly.
      ,
      • Nohata N.
      • Hanazawa T.
      • Kikkawa N.
      • Mutallip M.
      • Fujimura L.
      • Yoshino H.
      • et al.
      Caveolin-1 mediates tumor cell migration and invasion and its regulation by miR-133a in head and neck squamous cell carcinoma.
      ,
      • Bufalino A.
      • Cervigne N.K.
      • de Oliveira C.E.
      • Fonseca F.P.
      • Rodrigues P.C.
      • Macedo C.C.S.
      • et al.
      Low miR-143/miR-145 cluster levels induce activin a overexpression in oral squamous cell carcinomas, which contributes to poor prognosis.
      ,
      • Pavón M.A.
      • Arroyo-Solera I.
      • Céspedes M.V.
      • Casanova I.
      • León X.
      • Mangues R.
      uPA/uPAR and SERPINE1 in head and neck cancer: Role in tumor resistance, metastasis, prognosis and therapy.
      ,
      • Wang B.-J.
      • Chi K.-P.
      • Shen R.-L.
      • Zheng S.-W.
      • Guo Y.
      • Li J.-F.
      • et al.
      TGFBI promotes tumor growth and is associated with poor prognosis in oral squamous cell carcinoma.
      ].
      Most of the presented models performed better on the training cohort than in validation. Given the models were developed on the training cohort, the fact that they performed somewhat less well in validation is to be expected. Differences between the characteristics of the two cohorts might contribute to this as well. The validation cohort had a higher percentage of oral cavity tumours, only 24.6% of the patients had received concurrent chemotherapy, and overall survival was somewhat lower than for the training cohort (Supplementary Fig. 2). Differences in gene expressions and clinical parameters may have caused a shift in risk scores, causing imbalanced risk groups for the validation cohort. Equally, in our validation cohort, the older age of the samples meant that there was a higher technical rate of failure of RNA analysis. Furthermore, for the analyses of LRC and DM, the competing risk of death was not explicitly accounted for but considered as censoring, which may overestimate the true probability of the corresponding events [
      • Dutz A.
      • Löck S.
      Competing risks in survival data analysis.
      ]. Additional analyses based on the cumulative incidence function and Gray’s test led to similar results with significant patient stratifications or statistical trends. Exemplary results for the validation cohort are shown in Supplementary Figure 7 for the model combining the 6-gene signature with the two clinical parameters and for the extended gene signature with the two clinical parameters, based on Affymetrix and nanoString data.
      Previously we developed a 7-gene signature [
      • Schmidt S.
      • Linge A.
      • Zwanenburg A.
      • Leger S.
      • Lohaus F.
      • Krenn C.
      • et al.
      Development and validation of a gene signature for patients with head and neck carcinomas treated by postoperative radio(chemo)therapy.
      ] based on the same training cohort using nanoString nCounter data of 178 selected cancer-related genes. This signature included the genes INHBA and SERPINE1 that were also identified for the 6-gene signature developed in the present study using Affymetrix GeneChip data of 25,246 genes, showing the robustness of these two genes. The other five genes ACTN1, CD24, HILPDA, P4HA2, and TCF3 were not selected for the 6-gene signature. Differences between the two signatures may arise from the different sequencing platforms used. In addition, the differing number of patients with available gene expression data in both the studies might be a contributing factor. Also, in the present study, genes may be selected that were not present in the limited nanoString dataset and thereby modify the signature.
      The cancer stem-cell marker CD44 and the 15-gene hypoxia-associated signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ] have been validated for LRC in patients treated with locally advanced HPV16 DNA-negative HNSCC treated by PORT-C[
      • Linge A.
      • Löck S.
      • Krenn C.
      • Appold S.
      • Lohaus F.
      • Nowak A.
      • et al.
      Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
      ]. The 15-gene hypoxia-associated signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ] has proven to be a useful predictive biomarker for the selection of patients with HNSCC that benefit from hypoxic modification of primary radiotherapy with nimorazole [
      • Toustrup K.
      • Sørensen B.S.
      • Lassen P.
      • Wiuf C.
      • Alsner J.
      • Overgaard J.
      Gene expression classifier predicts for hypoxic modification of radiotherapy with nimorazole in squamous cell carcinomas of the head and neck.
      ]. Here we extended the 6-gene signature with these two biomarkers in combination with important clinical parameters, leading to improved prognostic performance for LRC, OS, and DM. Since, the three biomarkers are related to different biological processes, combining them may help to better uncover the underlying mechanisms associated with patient prognosis. The 6-gene signature and the extended gene signature determined for LRC, combined with clinical features were successfully validated for the secondary endpoints OS and DM adding to the robustness of the signature. Still, the extended gene signature combined with clinical parameters included five features for predicting only 27 loco-regional failures. This led to overfitting on the training cohort (ci = 0.83), while the model showed a more realistic performance in validation (ci = 0.69). Compared to the model containing only clinical features (ci = 0.62), this extended model performed somewhat better in validation (p = 0.075). While these results show that the studied biomarkers can explain some of the heterogeneity in treatment response, the small improvement also highlights the fact that it is difficult to improve on basic clinical prognostication.
      The developed 6-gene signature and the corresponding models were technically validated using nanoString technology showing very similar prognostic results and significant patient stratifications. Concerning the correlations between both measurements, our data are in line with a recent study that showed a high median correlation between expression measurements by nanoString and Affymetrix [
      • Schmidt S.
      • Linge A.
      • Grosser M.
      • Lohaus F.
      • Gudziol V.
      • Nowak A.
      • et al.
      Comparison of GeneChip, nCounter, and Real-Time PCR–Based Gene expressions predicting locoregional tumor control after primary and postoperative radiochemotherapy in head and neck squamous cell carcinoma.
      ]. The CD44 gene was the only gene that had a low correlation between the two methods, also when considering the individual CD44 probe expressions from Affymetrix. In nanoString analysis, the expression of CD44 was low compared to the other genes and close to the negative controls. This might have induced a large amount of noise, leading to a lower correlation with the Affymetrix results [
      • Schmidt S.
      • Linge A.
      • Grosser M.
      • Lohaus F.
      • Gudziol V.
      • Nowak A.
      • et al.
      Comparison of GeneChip, nCounter, and Real-Time PCR–Based Gene expressions predicting locoregional tumor control after primary and postoperative radiochemotherapy in head and neck squamous cell carcinoma.
      ]. In addition, the pair of CD44 nanoString probes does very likely not cover all known CD44 variant isoforms and thus may not be as representative for the overall CD44 gene expression [
      • Hudson D.L.
      • Speight P.M.
      • Watt F.M.
      Altered expression of CD44 isoforms in squamous-cell carcinomas and cell lines derived from them.
      ]. This may result in the lower comparability of nanoString results with Affymetrix data, which includes several probe pairs in parallel and thus covering more CD44 variant isoforms, or other methods such as immunohistochemistry.
      In the future, a multiplatform omics analysis may further improve the signature by combining data from different platforms, e.g., by combining the developed 6-gene signature with other biomarkers developed on the same cohort at the other DKTK-ROG partner sites [
      • Balermpas P.
      • Rödel F.
      • Rödel C.
      • Krause M.
      • Linge A.
      • Lohaus F.
      • et al.
      CD8+ tumour-infiltrating lymphocytes in relation to HPV status and clinical outcome in patients with head and neck cancer after postoperative chemoradiotherapy: A multicentre study of the German cancer consortium radiation oncology group (DKTK-ROG).
      ,
      • Balermpas P.
      • Rödel F.
      • Krause M.
      • Linge A.
      • Lohaus F.
      • Baumann M.
      • et al.
      The PD-1/PD-L1 axis and human papilloma virus in patients with head and neck cancer after adjuvant chemoradiotherapy: A multicentre study of the German Cancer Consortium Radiation Oncology Group (DKTK-ROG).
      ,
      • De-Colle C.
      • Mönnich D.
      • Welz S.
      • Boeke S.
      • Sipos B.
      • Fend F.
      • et al.
      SDF-1/CXCR4 expression in head and neck cancer and outcome after postoperative radiochemotherapy.
      ,
      • Tawk B.
      • Debus J.
      • Schwager C.
      • Linge A.
      • Ganswindt U.
      • Tinhofer I.
      • et al.
      Discovery of a reliable and robust methylome classifier of HPV driven head and neck cancer with favorable response to chemoradiation: A multicenter study of the German Cancer Consortium Radiation Oncology Group (DKTK-ROG).
      ,
      • Stangl S.
      • Tontcheva N.
      • Sievert W.
      • Shevtsov M.
      • Niu M.
      • Schmid T.E.
      • et al.
      Heat shock protein 70 and tumor-infiltrating NK cells as prognostic indicators for patients with squamous cell carcinoma of the head and neck after radiochemotherapy: A multicentre retrospective study of the German Cancer Consortium Radiation Oncology Gro.
      ,
      • Eder T.
      • Hess A.K.
      • Konschak R.
      • Stromberger C.
      • Jöhrens K.
      • Fleischer V.
      • et al.
      Interference of tumour mutational burden with outcome of patients with head and neck cancer treated with definitive chemoradiation: a multicentre retrospective study of the German Cancer Consortium Radiation Oncology Group.
      ,
      • Guberina M.
      • Sak A.
      • Pöttgen C.
      • Tinhofer-Keilholz I.
      • Budach V.
      • Balermpas P.
      • et al.
      ERCC2 gene single-nucleotide polymorphism as a prognostic factor for locally advanced head and neck carcinomas after definitive cisplatin-based radiochemotherapy.
      ], similar as recently published [
      • Löck S.
      • Linge A.
      • Lohaus F.
      • Ebert N.
      • Gudziol V.
      • Nowak A.
      • et al.
      Biomarker signatures for primary radiochemotherapy of locally advanced HNSCC – hypothesis generation on a multicentre cohort of the DKTK-ROG.
      ]. Validation will be performed on the currently recruiting prospective HNprädBio trial of the DKTK-ROG (NCT02059668) [

      Observational Study on Biomarkers in Head and Neck Cancer (HNprädBio). ClinicalTrials.gov [Internet], Bethesda (MD): National Library of Medicine (US), 2000 Feb 29. Identifier NCT02059668 2014:[registered 2014 Feb 11, updated 2019 Feb 18].

      ]. If successful, this could help to develop a clinically actionable framework integrating biological differences into radiotherapy dose planning, i.e. to perform a randomised controlled trial on dose escalation for patients classified as high risk to improve LRC and OS or on dose de-escalation for patients at low risk to reduce treatment-related side effects.
      In conclusion, we identified a novel 6-gene signature prognostic for LRC in patients with locally advanced HNSCC treated with PORT-C. Cox models were trained on a multicentre patient cohort and were independently validated. The prognostic performance was increased by extending the 6-gene signature with the validated biomarkers CD44 and the 15-gene hypoxia signature [
      • Toustrup K.
      • Sørensen B.S.
      • Nordsmark M.
      • Busk M.
      • Wiuf C.
      • Alsner J.
      • et al.
      Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
      ], and by its combination with the clinical parameters T stage and tumour localisation. Prospective validation of the signature is planned in the currently recruiting HNprädBio trial of the DKTK-ROG (NCT02059668) [

      Observational Study on Biomarkers in Head and Neck Cancer (HNprädBio). ClinicalTrials.gov [Internet], Bethesda (MD): National Library of Medicine (US), 2000 Feb 29. Identifier NCT02059668 2014:[registered 2014 Feb 11, updated 2019 Feb 18].

      ] before potential application in an interventional trial.

      Declaration of Competing Interest

      Michael Baumann, CEO and Scientific Chair of the German Cancer Research Center (DKFZ, Heidelberg) is responsible for collaborations with a large number of companies and institutions worldwide. In this capacity, he has signed contracts for research funding and/or collaborations, including commercial transfers, with industry and academia on behalf of his institute(s) and staff. He is a member of several supervisory boards, advisory boards and boards of trustees. Michael Baumann confirms that there is no conflict of interest for this paper. Dr. Baumann confirms that, to the best of his knowledge, none of the above funding sources were involved in the preparation of this paper.
      In the past 5 years, Dr. Mechthild Krause received funding for her research projects by IBA (2016), Merck KGaA (2014–2018 for preclinical study; 2018–2020 for clinical study), Medipan GmbH (2014–2018). Dr. Mechthild Krause and Dr. Annett Linge are involved in an ongoing publicly funded (German Federal Ministry of Education and Research) project with the companies Medipan (2019–2022), Attomol GmbH (2019–2022), GA Generic Assays GmbH (2019–2022), Gesellschaft für medizinische und wissenschaftliche genetische Analysen (2019–2022), Lipotype GmbH (2019–2022) and PolyAn GmbH (2019–2022). Dr. Krause and Dr. Linge confirm that, to the best of their knowledge, none of the above-mentioned funding sources were involved in the preparation of this paper.
      The Department of Radiation Oncology Tübingen receives within the frame of research agreements financial and technical support as well as sponsoring for travels and scientific symposia from Elekta AB (Stockholm, Sweden), TheraPanacea (Paris, France), Philips GmbH (Best, The Netherlands); Dr. Sennewald Medizintechnik GmbH (München, Germany), PTW Freiburg (Germany).

      Acknowledgements

      FFPE specimens were kindly provided by the local tissue banks and pathologists including the Normal and Tumor Tissue Bank (TNTB) of the National Center for Tumor Diseases (NCT/UCC) Dresden.

      Appendix A. Supplementary data

      The following are the Supplementary data to this article:

      References

        • Cooper J.S.
        • Pajak T.F.
        • Forastiere A.A.
        • Jacobs J.
        • Campbell B.H.
        • Saxman S.B.
        • et al.
        Postoperative concurrent radiotherapy and chemotherapy for high-risk squamous-cell carcinoma of the head and neck.
        N Engl J Med. 2004; 350: 1937-1944
        • Lavaf A.
        • Genden E.M.
        • Cesaretti J.A.
        • Packer S.
        • Kao J.
        Adjuvant radiotherapy improves overall survival for patients with lymph node-positive head and neck squamous cell carcinoma.
        Cancer. 2008; 112: 535-543https://doi.org/10.1002/cncr.23206
        • Rades D.
        • Seibold N.D.
        • Gebhard M.P.
        • Noack F.
        • Schild S.E.
        • Thorns C.
        Prognostic factors (including HPV status) for irradiation of locally advanced squamous cell carcinoma of the head and neck (SCCHN)Prognosefaktoren (inklusive HPV-Status) bei der Bestrahlung lokal fortgeschrittener Plattenepithelkarzinome im Kopf-Hals-Bereich (SCCHN).
        Strahlentherapie Und Onkol. 2011; 187: 626-632
        • Baumann M.
        • Krause M.
        • Overgaard J.
        • Debus J.
        • Bentzen S.M.
        • Daartz J.
        • et al.
        Radiation oncology in the era of precision medicine.
        Nat Rev Cancer. 2016; 16: 234-249
        • Lohaus F.
        • Linge A.
        • Tinhofer I.
        • Budach V.
        • Gkika E.
        • Stuschke M.
        • et al.
        HPV16 DNA status is a strong prognosticator of loco-regional control after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: Results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group.
        Radiother Oncol. 2014; 113: 317-323https://doi.org/10.1016/j.radonc.2014.11.011
      1. De-escalation of Adjuvant Radio (Chemo) Therapy for HPV-positive Head-neck Squamous Cell Carcinomas (DELPHI). ClinicalTrials.gov [Internet], Bethesda (MD): National Library of Medicine (US), 2000 Feb 29. Identifier NCT03396718 2018:[registered 2018 Jan 11, updated 2019 Feb 18].

        • Seiwert T.Y.
        • Foster C.C.
        • Blair E.A.
        • Karrison T.G.
        • Agrawal N.
        • Melotek J.M.
        • et al.
        Optima: A phase II dose and volume de-escalation trial for human papillomavirus-positive oropharyngeal cancer.
        Ann Oncol. 2019; 30: 297-302
        • Nichols A.C.
        • Lang P.
        • Prisman E.
        • Berthelet E.
        • Tran E.
        • Hamilton S.
        • et al.
        Treatment de-escalation for HPV-associated oropharyngeal squamous cell carcinoma with radiotherapy vs. trans-oral surgery (ORATOR2): Study protocol for a randomized phase II trial.
        BMC Cancer. 2020; 20https://doi.org/10.1186/s12885-020-6607-z
        • Toustrup K.
        • Sørensen B.S.
        • Nordsmark M.
        • Busk M.
        • Wiuf C.
        • Alsner J.
        • et al.
        Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer.
        Cancer Res. 2011; 71: 5923-5931
        • Scott J.G.
        • Berglund A.
        • Schell M.J.
        • Mihaylov I.
        • Fulp W.J.
        • Yue B.
        • et al.
        A genome-based model for adjusting radiotherapy dose (GARD): a retrospective, cohort-based study.
        LancetOncol. 2017; 18: 202-211
        • Kim H.S.
        • Kim S.C.
        • Kim S.J.
        • Park C.H.
        • Jeung H.-C.
        • Kim Y.B.
        • et al.
        Identification of a radiosensitivity signature using integrative metaanalysis of published microarray data for NCI-60 cancer cells.
        BMC Gen. 2012; 13
        • Chung C.H.
        • Parker J.S.
        • Ely K.
        • Carter J.
        • Yi Y.
        • Murphy B.A.
        • et al.
        Gene expression profiles identify epithelial-to-mesenchymal transition and activation of nuclear factor-κB Signaling as characteristics of a high-risk head and neck squamous cell carcinoma.
        Cancer Res. 2006; 66: 8210-8218
        • Eustace A.
        • Mani N.
        • Span P.N.
        • Irlam J.J.
        • Taylor J.
        • Betts G.N.J.
        • et al.
        A 26-gene hypoxia signature predicts benefit from hypoxia-modifying therapy in laryngeal cancer but not bladder cancer.
        Clin Cancer Res. 2013; 19: 4879-4888
        • Linge A.
        • Löck S.
        • Gudziol V.
        • Nowak A.
        • Lohaus F.
        • von Neubeck C.
        • et al.
        Low cancer stem cell marker expression and low hypoxia identify good prognosis subgroups in HPV(-) HNSCC after postoperative radiochemotherapy: A multicenter study of the DKTK-ROG.
        Clin Cancer Res. 2016; 22: 2639-2649
        • Linge A.
        • Löck S.
        • Krenn C.
        • Appold S.
        • Lohaus F.
        • Nowak A.
        • et al.
        Independent validation of the prognostic value of cancer stem cell marker expression and hypoxia-induced gene expression for patients with locally advanced HNSCC after postoperative radiotherapy.
        Clin Transl Radiat Oncol. 2016; 1: 19-26
        • Schmidt S.
        • Linge A.
        • Zwanenburg A.
        • Leger S.
        • Lohaus F.
        • Krenn C.
        • et al.
        Development and validation of a gene signature for patients with head and neck carcinomas treated by postoperative radio(chemo)therapy.
        Clin Cancer Res. 2018; 24: 1364-1374
        • Teutsch S.M.
        • Bradley L.A.
        • Palomaki G.E.
        • Haddow J.E.
        • Piper M.
        • Calonge N.
        • et al.
        The evaluation of genomic applications in practice and prevention (EGAPP) initiative: Methods of the EGAPP working group.
        Genet Med. 2009; 11: 3-14
        • Michiels S.
        • Ternès N.
        • Rotolo F.
        Statistical controversies in clinical research: prognostic gene signatures are not (yet) useful in clinical practice.
        Ann Oncol Off J Eur Soc Med Oncol. 2016; 27: 2160-2167https://doi.org/10.1093/annonc/mdw307
        • Pittman J.
        • Huang E.
        • Dressman H.
        • Horng C.-F.
        • Cheng S.H.
        • Tsou M.-H.
        • et al.
        Integrated modeling of clinical and gene expression information for personalized prediction of disease outcomes.
        Proc Natl Acad Sci USA. 2004; 101: 8431-8436
        • Zhao X.I.
        • Rødland E.A.
        • Sørlie T.
        • Naume B.
        • Langerød A.
        • Frigessi A.
        • et al.
        Combining gene signatures improves prediction of breast cancer survival.
        PLoS One. 2011; 6: e17845
        • Johnson W.E.
        • Li C.
        • Rabinovic A.
        Adjusting batch effects in microarray expression data using empirical Bayes methods.
        Biostatistics. 2007; 8: 118-127https://doi.org/10.1093/biostatistics/kxj037
        • Ritchie M.E.
        • Phipson B.
        • Wu D.i.
        • Hu Y.
        • Law C.W.
        • Shi W.
        • et al.
        Limma powers differential expression analyses for RNA-sequencing and microarray studies.
        Nucleic Acids Res. 2015; 43: e47
        • Tate J.G.
        • Bamford S.
        • Jubb H.C.
        • Sondka Z.
        • Beare D.M.
        • Bindal N.
        • et al.
        COSMIC: The Catalogue Of Somatic Mutations In Cancer.
        Nucleic Acids Res. 2019; 47: D941-D947
      2. Hothorn T. maxstat: Maximally Selected Rank Statistics. R Packag Version 07-25 Https//CRANR-ProjectOrg/Package=maxstat 2017.

        • Goetz J.G.
        • Lajoie P.
        • Wiseman S.M.
        • Nabi I.R.
        Caveolin-1 in tumor progression: The good, the bad and the ugly.
        Cancer Metastasis Rev. 2008; 27: 715-735https://doi.org/10.1007/s10555-008-9160-9
        • Nohata N.
        • Hanazawa T.
        • Kikkawa N.
        • Mutallip M.
        • Fujimura L.
        • Yoshino H.
        • et al.
        Caveolin-1 mediates tumor cell migration and invasion and its regulation by miR-133a in head and neck squamous cell carcinoma.
        Int J Oncol. 2011; https://doi.org/10.3892/ijo-00000840
        • Khatib A.
        • Solaimuthu B.
        • Ben Yosef M.
        • Abu Rmaileh A.
        • Tanna M.
        • Oren G.
        • et al.
        The glutathione peroxidase 8 (GPX8)/IL-6/STAT3 axis is essential in maintaining an aggressive breast cancer phenotype.
        Proc Natl Acad Sci U S A. 2020; 117: 21420-21431
        • Stamatopoulos B.
        • Smith T.
        • Crompot E.
        • Pieters K.
        • Clifford R.
        • Mraz M.
        • et al.
        The light chain IgLV3-21 defines a new poor prognostic subgroup in chronic lymphocytic leukemia: Results of a multicenter study.
        Clin Cancer Res. 2018; 24: 5048-5057
        • Bufalino A.
        • Cervigne N.K.
        • de Oliveira C.E.
        • Fonseca F.P.
        • Rodrigues P.C.
        • Macedo C.C.S.
        • et al.
        Low miR-143/miR-145 cluster levels induce activin a overexpression in oral squamous cell carcinomas, which contributes to poor prognosis.
        PLoS One. 2015; 10: e0136599
        • Pavón M.A.
        • Arroyo-Solera I.
        • Céspedes M.V.
        • Casanova I.
        • León X.
        • Mangues R.
        uPA/uPAR and SERPINE1 in head and neck cancer: Role in tumor resistance, metastasis, prognosis and therapy.
        Oncotarget. 2016; 7: 57351-57366https://doi.org/10.18632/oncotarget.10344
        • Lee Y.-C.
        • Yu C.-C.
        • Lan C.
        • Lee C.-H.
        • Lee H.-T.
        • Kuo Y.-L.
        • et al.
        Plasminogen activator inhibitor-1 as regulator of tumor-initiating cell properties in head and neck cancers.
        Head Neck. 2016; 38: E895-E904
        • Sun Z.-J.
        • Yu G.-T.
        • Huang C.-F.
        • Bu L.-L.
        • Liu J.-F.
        • Ma S.-R.
        • et al.
        Hypoxia induces TFE3 expression in head and neck squamous cell carcinoma.
        Oncotarget. 2016; 7: 11651-11663
        • Pavón M.A.
        • Arroyo-Solera I.
        • Téllez-Gabriel M.
        • León X.
        • Virós D.
        • López M.
        • et al.
        Enhanced cell migration and apoptosis resistance may underlie the association between high SERPINE1 expression and poor outcome in head and neck carcinoma patients.
        Oncotarget. 2015; 6: 29016-29033
        • Wang B.-J.
        • Chi K.-P.
        • Shen R.-L.
        • Zheng S.-W.
        • Guo Y.
        • Li J.-F.
        • et al.
        TGFBI promotes tumor growth and is associated with poor prognosis in oral squamous cell carcinoma.
        J Cancer. 2019; 10: 4902-4912
        • Lefranc M.P.
        Immunoglobulin and T cell receptor genes: IMGT® and the birth and rise of immunoinformatics.
        Front Immunol. 2014; 5: 1-22https://doi.org/10.3389/fimmu.2014.00022
      3. Joseph CG, Darrah E, Shah AA, Skora AD, Livia A, Wigley FM, et al. Association of the Autoimmune Disease Scleroderma with an Immunologic Response to Cancer 2014;343:152–7. doi:10.1126/science.1246886.Association.

        • Dutz A.
        • Löck S.
        Competing risks in survival data analysis.
        Radiother Oncol. 2019; 130: 185-189https://doi.org/10.1016/j.radonc.2018.09.007
        • Toustrup K.
        • Sørensen B.S.
        • Lassen P.
        • Wiuf C.
        • Alsner J.
        • Overgaard J.
        Gene expression classifier predicts for hypoxic modification of radiotherapy with nimorazole in squamous cell carcinomas of the head and neck.
        Radiother Oncol. 2012; 102: 122-129https://doi.org/10.1016/j.radonc.2011.09.010
        • Schmidt S.
        • Linge A.
        • Grosser M.
        • Lohaus F.
        • Gudziol V.
        • Nowak A.
        • et al.
        Comparison of GeneChip, nCounter, and Real-Time PCR–Based Gene expressions predicting locoregional tumor control after primary and postoperative radiochemotherapy in head and neck squamous cell carcinoma.
        J Mol Diagnostics. 2020; 22: 801-810
        • Hudson D.L.
        • Speight P.M.
        • Watt F.M.
        Altered expression of CD44 isoforms in squamous-cell carcinomas and cell lines derived from them.
        Int J Cancer. 1996; 66: 457-463https://doi.org/10.1002/(SICI)1097-0215(19960516)66:4<457::AID-IJC8>3.0.CO;2-V
        • Balermpas P.
        • Rödel F.
        • Rödel C.
        • Krause M.
        • Linge A.
        • Lohaus F.
        • et al.
        CD8+ tumour-infiltrating lymphocytes in relation to HPV status and clinical outcome in patients with head and neck cancer after postoperative chemoradiotherapy: A multicentre study of the German cancer consortium radiation oncology group (DKTK-ROG).
        Int J Cancer. 2016; 138: 171-181
        • Balermpas P.
        • Rödel F.
        • Krause M.
        • Linge A.
        • Lohaus F.
        • Baumann M.
        • et al.
        The PD-1/PD-L1 axis and human papilloma virus in patients with head and neck cancer after adjuvant chemoradiotherapy: A multicentre study of the German Cancer Consortium Radiation Oncology Group (DKTK-ROG).
        Int J Cancer. 2017; 141: 594-603
        • De-Colle C.
        • Mönnich D.
        • Welz S.
        • Boeke S.
        • Sipos B.
        • Fend F.
        • et al.
        SDF-1/CXCR4 expression in head and neck cancer and outcome after postoperative radiochemotherapy.
        Clin Transl Radiat Oncol. 2017; 5: 28-36
        • Tawk B.
        • Debus J.
        • Schwager C.
        • Linge A.
        • Ganswindt U.
        • Tinhofer I.
        • et al.
        Discovery of a reliable and robust methylome classifier of HPV driven head and neck cancer with favorable response to chemoradiation: A multicenter study of the German Cancer Consortium Radiation Oncology Group (DKTK-ROG).
        J Clin Oncol. 2018; 36
        • Stangl S.
        • Tontcheva N.
        • Sievert W.
        • Shevtsov M.
        • Niu M.
        • Schmid T.E.
        • et al.
        Heat shock protein 70 and tumor-infiltrating NK cells as prognostic indicators for patients with squamous cell carcinoma of the head and neck after radiochemotherapy: A multicentre retrospective study of the German Cancer Consortium Radiation Oncology Gro.
        Int J Cancer. 2018; 142: 1911-1925
        • Eder T.
        • Hess A.K.
        • Konschak R.
        • Stromberger C.
        • Jöhrens K.
        • Fleischer V.
        • et al.
        Interference of tumour mutational burden with outcome of patients with head and neck cancer treated with definitive chemoradiation: a multicentre retrospective study of the German Cancer Consortium Radiation Oncology Group.
        Eur J Cancer. 2019; 116: 67-76https://doi.org/10.1016/j.ejca.2019.04.015
        • Guberina M.
        • Sak A.
        • Pöttgen C.
        • Tinhofer-Keilholz I.
        • Budach V.
        • Balermpas P.
        • et al.
        ERCC2 gene single-nucleotide polymorphism as a prognostic factor for locally advanced head and neck carcinomas after definitive cisplatin-based radiochemotherapy.
        Pharmacogenomics J. 2021; 21: 37-46
        • Löck S.
        • Linge A.
        • Lohaus F.
        • Ebert N.
        • Gudziol V.
        • Nowak A.
        • et al.
        Biomarker signatures for primary radiochemotherapy of locally advanced HNSCC – hypothesis generation on a multicentre cohort of the DKTK-ROG.
        Radiother Oncol. 2022; 169: 8-14
      4. Observational Study on Biomarkers in Head and Neck Cancer (HNprädBio). ClinicalTrials.gov [Internet], Bethesda (MD): National Library of Medicine (US), 2000 Feb 29. Identifier NCT02059668 2014:[registered 2014 Feb 11, updated 2019 Feb 18].