International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study

Articles

International validation of the consensus Immunoscore for the classification of colon cancer: a prognostic and accuracy study Franck Pagès, Bernhard Mlecnik, Florence Marliot, Gabriela Bindea, Fang-Shu Ou, Carlo Bifulco, Alessandro Lugli, Inti Zlobec, Tilman T Rau, Martin D Berger, Iris D Nagtegaal, Elisa Vink-Börger, Arndt Hartmann, Carol Geppert, Julie Kolwelter, Susanne Merkel, Robert Grützmann, Marc Van den Eynde, Anne Jouret-Mourin, Alex Kartheuser, Daniel Léonard, Christophe Remue, Julia Y Wang, Prashant Bavi, Michael H A Roehrl, Pamela S Ohashi, Linh T Nguyen, SeongJun Han, Heather L MacGregor, Sara Hafezi-Bakhtiari, Bradly G Wouters, Giuseppe V Masucci, Emilia K Andersson, Eva Zavadova, Michal Vocka, Jan Spacek, Lubos Petruzelka, Bohuslav Konopasek, Pavel Dundr, Helena Skalova, Kristyna Nemejcova, Gerardo Botti, Fabiana Tatangelo, Paolo Delrio, Gennaro Ciliberto, Michele Maio, Luigi Laghi, Fabio Grizzi, Tessa Fredriksen, Bénédicte Buttard, Mihaela Angelova, Angela Vasaturo, Pauline Maby, Sarah E Church, Helen K Angell, Lucie Lafontaine, Daniela Bruni, Carine El Sissy, Nacilla Haicheur, Amos Kirilovsky, Anne Berger, Christine Lagorce, Jeffrey P Meyers, Christopher Paustian, Zipei Feng, Carmen Ballesteros-Merino, Jeroen Dijkstra, Carlijn van de Water, Shannon van Lent-van Vliet, Nikki Knijn, Ana-Maria Mușină, Dragos-Viorel Scripcariu, Boryana Popivanova, Mingli Xu, Tomonobu Fujita, Shoichi Hazama, Nobuaki Suzuki, Hiroaki Nagano, Kiyotaka Okuno, Toshihiko Torigoe, Noriyuki Sato, Tomohisa Furuhata, Ichiro Takemasa, Kyogo Itoh, Prabhu S Patel, Hemangini H Vora, Birva Shah, Jayendrakumar B Patel, Kruti N Rajvik, Shashank J Pandya, Shilin N Shukla, Yili Wang, Guanjun Zhang, Yutaka Kawakami, Francesco M Marincola, Paolo A Ascierto, Daniel J Sargent*, Bernard A Fox, Jérôme Galon

Summary Lancet 2018; 391: 2128–39 Published Online May 10, 2018 http://dx.doi.org/10.1016/ S0140-6736(18)30789-X See Comment page 2084 INSERM, Laboratory of Integrative Cancer Immunology, Paris, France (Prof F Pagès MD, B Mlecnik PhD, F Marliot MSc, G Bindea PhD, T Fredriksen MSc, B Buttard MSc, M Angelova PhD, A Vasaturo PhD, P Maby PhD, S E Church PhD, H K Angell PhD, L Lafontaine BSc, D Bruni PhD, C El Sissy MD, A Kirilovsky PhD, Prof A Berger PhD, C Lagorce PhD, J Galon PhD); Université Paris Descartes, Sorbonne Paris Cité, Paris, France (Prof F Pagès, B Mlecnik, F Marliot, G Bindea, T Fredriksen, B Buttard, M Angelova, A Vasaturo, P Maby, S E Church, H K Angell, L Lafontaine, D Bruni, C El Sissy A Kirilovsky, Prof A Berger, C Lagorce, J Galon); Centre de Recherche des Cordeliers, Université Pierre et Marie Curie, Sorbonne Universités, Paris, France (Prof F Pagès, B Mlecnik, F Marliot, G Bindea, T Fredriksen, B Buttard, M Angelova, A Vasaturo, P Maby, S E Church, H K Angell, L Lafontaine, D Bruni, C El Sissy, A Kirilovsky, Prof A Berger, C Lagorce, J Galon); Immunomonitoring Platform, Laboratory of Immunology, AP-HP, Assistance Publique-Hopitaux de Paris, Georges Pompidou

2128

Background The estimation of risk of recurrence for patients with colon carcinoma must be improved. A robust immune score quantification is needed to introduce immune parameters into cancer classification. The aim of the study was to assess the prognostic value of total tumour-infiltrating T-cell counts and cytotoxic tumour-infiltrating T-cells counts with the consensus Immunoscore assay in patients with stage I–III colon cancer. Methods An international consortium of 14 centres in 13 countries, led by the Society for Immunotherapy of Cancer, assessed the Immunoscore assay in patients with TNM stage I–III colon cancer. Patients were randomly assigned to a training set, an internal validation set, or an external validation set. Paraffin sections of the colon tumour and invasive margin from each patient were processed by immunohistochemistry, and the densities of CD3+ and cytotoxic CD8+ T cells in the tumour and in the invasive margin were quantified by digital pathology. An Immunoscore for each patient was derived from the mean of four density percentiles. The primary endpoint was to evaluate the prognostic value of the Immunoscore for time to recurrence, defined as time from surgery to disease recurrence. Stratified multivariable Cox models were used to assess the associations between Immunoscore and outcomes, adjusting for potential confounders. Harrell’s C-statistics was used to assess model performance. Findings Tissue samples from 3539 patients were processed, and samples from 2681 patients were included in the analyses after quality controls (700 patients in the training set, 636 patients in the internal validation set, and 1345 patients in the external validation set). The Immunoscore assay showed a high level of reproducibility between observers and centres (r=0·97 for colon tumour; r=0·97 for invasive margin; p<0·0001). In the training set, patients with a high Immunoscore had the lowest risk of recurrence at 5 years (14 [8%] patients with a high Immunoscore vs 65 (19%) patients with an intermediate Immunoscore vs 51 (32%) patients with a low Immunoscore; hazard ratio [HR] for high vs low Immunoscore 0·20, 95% CI 0·10–0·38; p<0·0001). The findings were confirmed in the two validation sets (n=1981). In the stratified Cox multivariable analysis, the Immunoscore association with time to recurrence was independent of patient age, sex, T stage, N stage, microsatellite instability, and existing prognostic factors (p<0·0001). Of 1434 patients with stage II cancer, the difference in risk of recurrence at 5 years was significant (HR for high vs low Immunoscore 0·33, 95% CI 0·21–0·52; p<0·0001), including in Cox multivariable analysis (p<0·0001). Immunoscore had the highest relative contribution to the risk of all clinical parameters, including the American Joint Committee on Cancer and Union for International Cancer Control TNM classification system. Interpretation The Immunoscore provides a reliable estimate of the risk of recurrence in patients with colon cancer. These results support the implementation of the consensus Immunoscore as a new component of a TNM-Immune classification of cancer. Funding French National Institute of Health and Medical Research, the LabEx Immuno-oncology, the Transcan ERAnet Immunoscore European project, Association pour la Recherche contre le Cancer, CARPEM, AP-HP, www.thelancet.com Vol 391 May 26, 2018

Articles

Institut National du Cancer, Italian Association for Cancer Research, national grants and the Society for Immunotherapy of Cancer. Copyright © 2018 Elsevier Ltd. All rights reserved.

Introduction Prognosis of patients with resectable colorectal cancer relies on histopathological criteria of tumour invasion according to the American Joint Committee on Cancer (AJCC) and Union for International Cancer Control (UICC) TNM classification system and on features of tumour-cell differentiation.1–3 This anatomy-based system provides useful but incomplete prognostic infor­ mation.4 New ways to classify cancer that focuse on tumour cells, molecular pathways, mutation status, and tumour gene expression-based stratification4,5 only have a moderate prediction accuracy and limited clinical usefulness. In-situ immune cell infiltrate in tumours is associated with a favourable prognostic effect.4,6–19 In colorectal cancer, we have shown that time to recurrence and overall survival strongly correlate with the strength of the in-situ adaptive immune reaction8,13,17,20,21 in the colon tumour and the invasive margin.22 We have previously characterised the subtypes of immune cells that infiltrate tumours23 and proposed that the intra-tumoral immune

context­ure (ie, type, functional orientation, density, and location of immune cells) of solid tumours could be a dominant determinant of clinical outcome.7,8 We showed the usefulness of a derived immune score termed Immunoscore to predict clinical outcome in patients with early20 and advanced24,25 stage colorectal cancer. The consensus Immunoscore is a scoring system to summarise the density of CD3+ and CD8+ T-cell effectors within the tumour and its invasive margin. A revolutionary change is taking place in oncology by evaluating, and activating with immunotherapy, the preexisting intra-tumoral immunity of patients.26,27 The preexisting intratumoral immune profile can now be assessed by digital pathology. An international consort­ ium was initiated with the support of the Society for Immunotherapy of Cancer (SITC)28 to validate the consensus Immunoscore in clinical practice for patients with stage I–III colon cancer. Here we report the final results of an analysis of colonic tumours to assess the relation between Immunoscore and prognosis and

Research in context Evidence before this study We searched MEDLINE via PubMed for evidence available by Jan 1, 2018, for an association between immune cell infiltrates of tumours and prognosis in various cancer types. 250 published articles implicating cytotoxic T cells, memory T cells, and T-helper cell subpopulations in prognosis of cancer patients were retrieved (28 different cancer types were analysed). Tumour infiltration by cytotoxic CD8+ T cells was associated with a good prognosis in 98% of the studies. By contrast, other T-helper subpopulations had a good or a bad prognostic effect depending on the method used and on the cancer types. Colorectal cancer was the most studied cancer type. The quantification of immune infiltrates had a prognostic importance superior to that of the American Joint Committee on Cancer (AJCC) and Union for International Cancer Control (UICC) TNM classification system. Multiple ways to classify cancer, colorectal cancer in particular, have been proposed. These rely on tumour cell characteristics, such as morphology, molecular pathways, mutation status, tumour cell origin, and tumour gene expression. Despite the importance of the intratumour immune reaction to predict patients at high risk and low risk of recurrence and death, no immune marker is used in the classification of colorectal cancer at present. Added value of this study After the first World Immunotherapy Council meeting on Feb 21–24, 2012, bringing together 17 cancer international organisations, a decision was made, together with the Society

www.thelancet.com Vol 391 May 26, 2018

for Immunotherapy of Cancer, to initiate an international Immunoscore consortium to assess the prognostic value of total T-lymphocyte and cytotoxic T-lymphocyte tumour infiltrate. The international Immunoscore consortium of expert pathologists and immunologists identified a strategy for an international multicentre study to validate the consensus Immunoscore in patients with stage I–III colon cancer. The purpose of the study was to agree on how to harmonise the Immunoscore assay, to demonstrate the feasibility and reproducibility of the Immunoscore, to demonstrate the utility of the Immunoscore to predict patients with stage II colon cancer at high risk of recurrence, and to validate the prognostic value of the Immunoscore in routine clinical settings. Implications of all the available evidence Compared with existing prognostic tools, the consensus Immunoscore can be used to better define the prognosis of cancer patients, better identify patients at high risk of tumour recurrence (including patients with stage II colon cancer), and stratify patients who could benefit from adjuvant therapies. The consensus Immunoscore study represents the largest international consortium validating a standardised immune parameter to stratify patients with cancer. The Immunoscore had a prognostic value superior to that of existing tumour parameters. This initiative and these results support the implementation of the consensus Immunoscore as a new component for a TNM-Immune classification of cancer.

European Hospital, Paris, France (Prof F Pagès, F Marliot, C El Sissy, N Haicheur MSc, A Kirilovsky); Inovarion, Paris, France (B Mlecnik); Cancer Center Statistics, Department of Health Sciences Research, Mayo Clinic, Rochester, MN, USA (F-S Ou PhD, J P Meyers BA, D J Sargent PhD); Department of Pathology, Providence Portland Medical Center, Portland, OR, USA (Prof C Bifulco MD); Institute of Pathology, University of Bern, Bern, Switzerland (Prof A Lugli MD, Prof I Zlobec PhD, Prof T T Rau MD); Department of Medical Oncology, University Hospital of Bern, Bern, Switzerland (M D Berger MD); Pathology Department, Radboud University, Nijmegen, Netherlands (Prof I D Nagtegaal PhD, E Vink-Börger PhD, J Dijkstra BSc, C van de Water BASc, S van Lent-van Vliet BSc, N Knijn MD); Department of Pathology, University Erlangen-Nürnberg, Erlangen, Germany (Prof A Hartmann MD, C Geppert MD, J Kolwelter MD); Department of Surgery, University Erlangen-Nürnberg, Erlangen, Germany (Prof S Merkel MD, Prof R Grützmann MD); Institut Roi Albert II, Department of Medical Oncology Cliniques Universitaires St-Luc, Brussels, Belgium (M Van den Eynde PhD); Institut de Recherche Clinique et Experimentale (Pole MIRO), Université Catholique de Louvain, Brussels, Belgium (M Van den Eynde); Department of Pathology, Cliniques Universitaires St-Luc, Brussels, Belgium (Prof A Jouret-Mourin PhD); Institut de Recherche Clinique et Experimentale (Pole GAEN), Université Catholique de Louvain, Brussels, Belgium (Prof A Jouret-Mourin); Institut Roi Albert II, Department of Digestive Surgery, Cliniques Universitaires St-Luc Université Catholique de Louvain, Brussels, Belgium (Prof A Kartheuser MD, D Léonard PhD, C Remue MD); Curandis Laboratories, Boston, MA, USA (Prof J Y Wang PhD); Department of Pathology and Laboratory Medicine, University Health Network, Toronto, ON, Canada

2129

Articles

2130

to compare the Immunoscore with existing prognostic risk parameters, including microsatellite instability (MSI) status.

Methods

Study design and patients An international consortium of 14 expert centres in 13 countries in North America, Europe, and Asia

(appendix) was initiated to evaluate the standard­ ised Immunoscore assay in primary tumours from patients with stage I–III colon cancer (appendix). After stratification by centre, T stage, N stage, and relapse, patients from Switzerland, Germany, France, the USA, the Czech Republic, and Canada for whom data were available before Jan 1, 2016 were randomly assigned (1:1) to a training set and an internal validation set. An

A 3539 patients with stage I–III colon cancer assessed for Immunoscore

858 excluded* 357 biomarker quality control 577 clinical data quality control

2681 randomised

700 in training set

B

636 in internal validation set

978 in external validation set (1345 including Asian patients)

C

Training set

Internal validation set

Patients without event (%)

100 80 60 Immunoscore Low Intermediate High

40 20

0 Number at risk Low 155 Intermediate 357 High 188

D

HR for high vs low Immunoscore 0·29 (95% CI 0·15–0·57); p<0·00040

HR for high vs low Immunoscore 0·20 (95% CI 0·10–0·38); p<0·0001

0

1

2

3

4

5

6

7

8

0

1

2

3

4

5

6

7

8

127 294 156

95 262 144

86 232 135

74 213 120

66 175 108

51 141 86

39 113 70

32 91 53

162 304 170

130 268 149

114 243 139

101 223 117

84 199 104

69 171 81

56 135 64

45 117 47

38 100 41

E

Training set and internal validation set

External validation set

100 Patients without event (%)

(Prof J Y Wang, Prof M H A Roehrl PhD, S Hafezi-Bakhtiari MD); Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, ON, Canada (Prof J Y Wang, P Bavi MD, Prof M H A Roehrl); Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, NY, USA (Prof Ml H A Roehrl); University Health Network, Toronto, ON, Canada (Prof P S Ohashi PhD, L T Nguyen PhD, S Han MSc, H L MacGregor BSc, Prof B G Wouters PhD); Department of Oncology-Pathology, Karolinska Institutet, Karolinska University, Stockholm, Sweden (Prof G V Masucci PhD, E K Andersson PhD); Department of Oncology, First Faculty of Medicine, Charles University and General University Hospital in Prague, Prague, Czech Republic (Prof E Zavadova PhD, M Vocka MD, J Spacek MD, Prof L Petruzelka PhD, B Konopasek PhD); Institute of Pathology, First Faculty of Medicine, Charles University, Prague, Czech Republic (P Dundr PhD, H Skalova MD, K Nemejcova PhD); General University Hospital in Prague, Prague, Czech Republic (P Dundr, H Skalova, K Nemejcova); Department of Pathology (Prof G Botti PhD, Prof F Tatangelo PhD) and Colorectal Surgery Department (Paolo Delrio PhD), Istituto Nazionale per lo Studio e la Cura dei Tumori, “Fondazione G.Pascale” Naples, Italy; IRCCS Istituto Nazionale Tumori “Regina Elena”, Rome, Italy (Prof G Ciliberto PhD); Center for Immuno-Oncology, University Hospital of Siena, Istituto Toscano Tumori, Siena, Italy (M Maio PhD); Molecular Gastroenterology and Department of Gastroenterology, Humanitas Clinical and Research Center, Rozzano, Milan, Italy (Prof L Laghi MD, Prof F Grizzi PhD); Humanitas University, Rozzano, Milan, Italy (Prof F Grizzi); NanoString Technologies, Seattle, WA, USA (S E Church); Translational Science, Oncology, IMED Biotech Unit, AstraZeneca, Cambridge, UK (H K Angell);

80 60 40 20 HR for high vs low Immunoscore 0·24 (95% CI 0·15–0·37); p<0·0001 0

0

Number at risk Low 317 Intermediate 661 High 358

1 257 562 305

2 3 4 5 6 Time to recurrence (years from surgery) 209 505 283

187 455 252

158 412 224

135 346 189

107 276 150

HR for high vs low Immunoscore 0·36 (95% CI 0·25–0·53); p<0·0001

7

8

0

1

84 230 117

70 191 94

239 445 294

166 332 227

2 3 4 5 6 Time to recurrence (years from surgery) 128 259 190

100 223 160

77 189 133

68 170 111

55 139 88

7

8

51 102 64

36 77 52

Figure 1: Study design and prognostic value of the Immunoscore (A) The Immunoscore international consortium study design (further details are provided in the appendix). (B−E) Kaplan-Meier curves for time to recurrence according to the Immunoscore based on three categories in the training set, the internal validation set, the training set plus internal validation set cohorts (5 year survival: 71% [95% CI 66−77] for patients with low Immunoscore versus 84% [81−87] for patients with intermediate Immunoscore versus 92% [89−95] for patients with high Immunoscore), and the external validation set (5 year percentage survival: 57% [95% CI 50−65] for patients with low Immnoscore versus 73% [68−78] for patients with intermediate Immunoscore versus 82% [77−87] for patients with high Immunoscore). *76 patients were excluded because of both bad quality biomarker data and improper clinical data.

www.thelancet.com Vol 391 May 26, 2018

Articles

independent external validation set with data received after Jan 1, 2016, was also established. Samples were excluded if counting data were missing or the staining intensity was too low, (biomarker quality control) or if the patient did not meet the inclusion criteria (clinical quality control; eg, rectal cancer, metastatic cancer, received neoadjuvant chemotherapy, Tis/Tx tumour, or missing mortality or recurrence status; appendix). Stage II colorectal cancer was considered high-risk if positive for the biomarkers for venous emboli, lymphatic invasion, or perineural invasion (VELIPI+) or T4 stage II, whereas low-risk colorectal cancer was negative for VELIPI markers (VELIPI–) and T1–3 stage II.29 Ethical, legal, and social implications were approved by an ethical review board at each centre.

Procedures Expert pathologists and immunologists from the con­ sortium28 agreed to use standardised operating pro­ cedures and dedicated image-analysis software to quantify the density of CD3+ and CD8+ T-cell effectors in the tumour and its invasive margin and to thereby determine the consensus Immunoscore.4,8,13,20 Pathologists from all participating centres selected one tumour block from each patient that contained the colon tumour and invasive margin. Two tissue paraffin sections of 4 µm were processed for immunohisto­ chemistry according to the following protocol, as recommended by the reference centre (Immuno­monitor­ ing Platform, Hôpital Européen Georges Pompidou AP-HP, INSERM, Paris, France): antigen retrieval with a Tris base buffer (pH=8) for 60 min, quenching of endogenous peroxidase activity, incubation with antibody against CD8 (C8/144B, 3 µg/ml; Dako, Glostrup, Denmark) for 32 min at 37°C and with antibody against CD3 (2GV6, 0·4 µg/ml; Ventana, Tuscon, AZ, USA) for 20 min at 37°C; revelation with the Ultraview Universal DAB IHC Detection Kit (Ventana, Tuscon, AZ, USA), and counter­staining with Mayer’s haematoxylin. Digital images of the stained tissue sections were obtained at 20× magnification and 0·45 µm/pixel resolution. The densities of CD3+ and CD8+ T cells in colon tumour and invasive margin regions were determined using a specially developed Immunoscore module that was integrated into the image-analysis system of a Developer XD digital pathology software (Definiens, Munich, Germany). The mean and distribution of the staining intensities were monitored to obtain an internal quality control of each slide (appendix). For each case, CD3+ and CD8+ cell densities in colon tumour and invasive margin regions were compared with that obtained in the training and internal validation sets and converted into percentiles. The mean of four percentiles (two markers, two regions) was calculated and converted into an Immunoscore. In a three-category Immunoscore analysis, a 0–25% density was scored www.thelancet.com Vol 391 May 26, 2018

as low, a density between 25% and 70% was scored as intermediate, and a density between 70% and 100% density was scored as high (appendix). In a two-category Immunoscore analysis, a 0–25% density was scored as low, and a density between 25% and 100% was scored as intermediate plus high. The biomarker reference centre optimised the immun­o­­staining protocols and sent them to the centres with a batch of control slides to stain (n=3–10). A 20% maximum deviation for the staining intensity (as compared with adjacent slides stained by the reference centre) was tolerated. To achieve a good level of consistency between centres, the reference centre wrote and distributed a user’s manual to each participating centre that described the Immuno­score module of the software and illustrated (with 38 examples) how to select tumour regions. The Immunoscore module had been made available to the consortium so that each centre could analyse its own patient cohort. The reference centre validated each case at the end of the study to ensure the quality and intensity of the staining. Samples were excluded from the analysis if counts were missing from a tumour region or if staining intensity was low (ie, ≤152 arbitrary units [AU]; appendix). MSI status was determined for patients from whom enough sample material was available. Genomic DNA was extracted from paired formalin-fixed paraffinembedded tumour and normal colonic tissue, and MSI status was assessed with the molecular new Bethesda panel and by immunohistochemistry (using MLH1, MSH2, MSH6, and PMS2 antibodies). Patients with deficient mismatch repair and proficient mismatch repair were denoted microsatellite unstable (MSI) and microsatellite stable (MSS), respectively. Detailed protocols are available in the appendix.

Outcomes The primary endpoint was to evaluate the prognostic value of Immunoscore for time to recurrence, defined as time from surgery to disease recurrence. Additional outcomes of interest were disease-free survival (time from surgery to first observation of disease recurrence or death due to any cause) and overall survival (time from surgery to death due to any cause; appendix).

Statistical analysis Statisticians external to the consortium (DJS, F-SO, JPM) did the statistical analyses using SAS version 9.3. Demographics and disease characteristics were com­ pared descriptively across the training set, internal validation set, and external validation set and compared by Kruskal-Wallis and χ² tests when applicable. Bivariable associ­ation between Immunoscore and timeto-event out­comes was evaluated by the stratified logrank test and stratified Cox proportional hazards model. Stratified multivariable Cox models were used to assess the associations between Immunoscore and outcomes

Digestive Surgery Department, AP-HP, Assistance Publique-Hopitaux de Paris, Georges Pompidou European Hospital, Paris, France (Prof A Berger, C Lagorce); University of Medicine and Pharmacy “Grigore T. Popa” Iaşi, Department of Surgical Oncology, Regional Institute of Oncology, Iaşi, Roumania (A-M Mușină PhD, D-V Scripcariu PhD); Division of Cellular Signaling, Institute for Advanced Medical Research, Keio University School of Medicine, Tokyo, Japan (B Popivanova PhD, M Xu PhD, T Fujita PhD, Prof Y Kawakami PhD); Department of Translational Research and Developmental Therapeutics against Cancer, Yamaguchi University School of Medicine, Yamaguchi, Japan (Prof S Hazama PhD); Department of Gastroenterological, Breast and Endocrine Surgery, Yamaguchi University Graduate School of Medicine, Yamaguchi, Japan (Prof N Suzuki PhD, Prof H Nagano PhD); Department of Surgery, Kindai University, School of Medicine, Osaka-sayama, Japan (Prof K Okuno PhD); Department of Pathology (Prof T Torigoe PhD, Prof N Sato PhD) and Department of Surgery, Surgical Oncology, and Science (Prof T Furuhata PhD, Prof I Takemasa PhD), Sapporo Medical University School of Medicine, Sapporo, Japan; Department of Immunology and Immunotherapy, Kurume University School of Medicine, Kurume, Japan (Prof K Itoh PhD); The Gujarat Cancer & Research Institute, Asarwa, Ahmedabad, India (Prof P S Patel PhD, H H Vora PhD, B Shah MD, J B Patel PhD, K N Rajvik PhD, S J Pandya MCh, Prof S N Shukla MD); Institute for Cancer Research of School of Basic Medical Science, Department of Pathology of the First Affiliated Hospital, Health Science Center of Xi’an Jiaotong University, Xian, China (Prof Y Wang PhD, Prof G Zhang MD); Research Branch, Sidra Medical and Research Centre, Doha, Qatar (F M Marincola MD); Melanoma, Cancer Immunotherapy and Innovative Therapies Unit,

2131

Articles

*Dr Sargent died in September, 2016 Correspondence to: Dr Jérôme Galon, INSERM UMRS1138, Laboratory of Integrative Cancer Immunology. Cordeliers Research Center, 75006 Paris, France [email protected] or Prof Franck Pagès, Immunomonitoring Platform, Laboratory of Immunology, Hôpital Européen Georges Pompidou, 75015 Paris, France [email protected] See Online for appendix

C 100 80 60 40 20 0

50 μm

Staining intensity intervals 0

D

E r=0·98

1000 800 600 400 200 0

0

200 400 600 800 1000 USA (cells per mm2)

CD3 and CD8 in colon tumour (whole slide)

A B C r=0·98 D E F G H A B C D E F G H

Pathologists from different centres Pearson’s r 0 0·5 0·8 1·0

104 103 102 101 100 Mean density of CD3+ lymphocytes in colon tumour (whole slide)

F

Immunoscore (CD3, CD8)

High, intermediate, low Concordant (%)

G

Infiltrating immune cells, H&E

High, intermediate, low Discordant (%)

10 20 25 30 40 50

Figure 2: Determination of the immune cell densities by image analysis software with a dedicated Immunoscore module (A) The colon tissue is divided into tiles, with tumour tissue highlighted in red and the invasive margin highlighted in brown. (B) Representative immunohistochemistry of CD3+ cells infiltrating a colonic tumour (left, in brown), and a histogram of the staining intensities of positive cells detected by the software in a case with an adequate immunostaining intensity, leading to a valid counting (right; mean intensity >152 arbitrary units; black triangle). (C) Histogram of the mean density of CD3+ T cells in the colon tumour for all patients of the cohort (training set plus internal validation set plus external validation set, 3539 patients). (D) Example of a 2 × 2 correlation of the mean densities for CD3+ cells in the tumour between pathologists from the USA and the Netherlands (r=0·98). (E) Correlation matrix illustrating the reproducibility of the CD3+ and CD8+ cell counting in a set of control slides with colon tumour sections by eight pathologists (tagged A–H) from five centres. The mean of all 2 × 2 correlations between the eight pathologists doing digital pathology is r=0·97. (F–G) H&E stain evaluation for immune-cell infiltration was done by 11 independent evaluators on 268 representative cases that were randomly selected (n=135 in cohort 1; n=133 in cohort 2). Each evaluator investigated all cases using the same set of nine referent slides (three for each Immunoscore category). Pie charts show the degree of agreement between (F) Immunoscore and (G) H&E stain evaluation for tumour-infiltrating T cells by independent evaluations. Cases with more than 50% discordance between evaluators were considered ambiguous. H&E=haematoxylin and eosin.

with adjustment for potential confounders. Model performance was assessed by Harrell’s C-statistics. Participating centres were used as the stratification factors, and the variables adjusted in the multivariable models were age, sex, T stage, N stage, and MSI status. The predictive accuracy of the Immunoscore was evaluated by the integrated area under the ROC curve (iAUC) with 1000 × bootstrap resampling. The performance of risk prediction models was compared using the likelihood ratio p value. The relative importance of each parameter to survival risk was assessed using the χ² from Harrell’s rms R package (version 3.2.3).

Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study after the statistical analysis by the external statistical group and had final responsibility for the decision to submit for publication.

Results Tumour samples from 3539 patients with AJCC/UICC TNM stage I–III colon cancer were collected between 2132

>370

152 Quality control cutoff Mean intensity

Colon Invasive margin tumour (red) (brown)

Cells per mm2

B 5 mm

% of cells

A

Netherlands (cells per mm2)

Istituto Nazionale per lo Studio e la Cura dei Tumori, Fondazione “G. Pascale”, Napoli, Italy (Prof P A Ascierto MD); Laboratory of Molecular and Tumor Immunology, Earle A. Chiles Research Institute, Robert W Franz Cancer Center, Providence Portland Medical Center, Portland, OR, USA (B A Fox PhD, C Paustian PhD, Z Feng MD, C Ballesteros-Merino PhD); and Department of Molecular Microbiology andImmunology, Oregon Health and Science University, Portland, OR, USA (B A Fox)

2013 and 2015, and Immunoscore data were retrieved between 2013 and 2016. 357 patients were excluded after biomarker quality control, and 577 patients were excluded after clinical data quality control. Samples from 2681 patients were included in the analyses (700 patients in the training set, 636 patients in the internal validation set, and 1345 patients in the external validation set [including Asian patients]; figure 1A). The demographic and clinical characteristics of the patients were well balanced between the training set and internal validation set (appendix). 1380 (52%) patients were men, and the median age was 69 years (IQR 60–77). 451 (17%) patients had stage I colorectal cancer, 1434 (54%) patients had stage II colorectal cancer, and 763 (29%) patients had stage III colo­ rectal cancer. The mean number of lymph nodes examined was 18·8 per patient. 474 (18%) patients had a relapse, and 930 (35%) patients died. The median follow-up time for all patients was 96 months (95% CI 93–100). The median follow-up time was 111 months (95% CI 105–116) in the training set, 113 months (107–119) in the internal validation set, and 83 months (78–88) in the external validation set. The median survival time from surgery to death from any cause was 158 months (95% CI 143–171). www.thelancet.com Vol 391 May 26, 2018

Articles

www.thelancet.com Vol 391 May 26, 2018

A Low Immunoscore Intermediate Immunoscore High Immunoscore

Disease-free survival (%)

100 80 60 40

HR for high vs low 0·31 (95% CI 0·23–0·41); p<0·0001 HR for intermediate vs low 0·57 (95% CI 0·47–0·69); p<0·0001 HR for high vs intermediate 0·56 (95% CI 0·42–0·73); p<0·0001 c-index=0·61 (95% CI 0·55–0·67)

20 0

0 Number at risk Low Immunoscore 702 Intermediate Immunoscore 1271 High Immunoscore 708

1

2

3

4

5

6

7

8

566 1069 594

468 930 535

411 839 469

358 763 417

321 675 359

259 526 273

194 406 208

150 324 172

B MSI, low Immunoscore MSI, high Immunoscore

100 Disease-free survival (%)

A significant positive correlation was found be­ tween the densities of CD3+ and CD8+ cells in each tumour region and survival in the training set. 155 (22%) patients had a low Immunoscore, 357 (51%) patients had an inter­mediate Immunoscore, and 188 (27%) patients had a high Immunoscore. Patients with high Immunoscore had the lowest risk of recurrence. Recurrence and survival at 3 years and 5 years are shown in the appendix. Recurrence at 3 years was seen in ten (5%) patients with high Immuno­score, in 58 (17%) mediate Immunoscore, and in patients with an inter­ 42 (26%) patients with a low Immunoscore (unadjusted hazard ratio [HR] for high vs low Immunoscore 0·20, 95% CI 0·10-0·38; p<0·0001; figure 1B; appendix). Similar results were found with an adjusted model (HR for high vs low Immuno­ score 0·25, 95% CI 0·13–0·48; p<0·0001), disease-free survival, and overall survival (appendix). Patients with a high Immunoscore had the longest survival. Overall survival at 3 years was recorded for 164 (87%) patients with a high Immuno­ score, 305 (85%) patients with an intermediate Immuno­ score, and 117 (76%) patients with a low Immunoscore. Overall survival at 5 years was recorded for 155 (82%) patients with a high Immunoscore, 274 (77%) patients with an intermediate Immunoscore, and 94 (62%) patients with a low Immunoscore (unadjusted HR for high vs low Immunoscore 0·53, 95% CI 0·38–0·75; p=0·0004). In the training set, a significant difference in time to recurrence was also found between patients with high and intermediate Immunoscore (unadjusted HR 0·38, 95% CI 0·21−0·70; p=0·0011), overall survival (p=0·035), and disease-free survival (p=0·0050). Re­ grouping the Immunoscore into two categories also allowed the identification of patients with distinct clinical outcome for time to recurrence, disease-free survival, and overall survival (appendix). These results were confirmed in the internal and external validation set. Patients with the highest est risk of recurrence. Immunoscore had the low­ Recurrence at 3 years in the internal validation set was recorded for ten (6%) patients with a high Immunoscore, 31 (10%) patients with an inter­mediate Immuno­score, and 33 (20%) patients with a low Immunoscore (unadjusted HR for high vs low Immunoscore 0·29, 95% CI 0·15–0·57; p=0·00040. Recurrence at 3 years in the external validation set was recorded for 42 (14%) patients with a high Immunoscore, 104 (24%) patients with an intermediate Immunoscore, and 85 (36%) patients with a low Immunoscore (un­adjusted HR for high vs low Immunoscore 0·36, 95% CI 0·24–0·53; p<0·0001; figure 1E; appendix). Similar results were obtained with adjusted models (appendix). A beneficial effect of a high Immunoscore was also confirmed for disease-free survival and overall survival in the internal and external validation sets (all p<0·05; appendix). When stratified into two Immunoscore categories, patients with high Immuno­score also had a significantly prolonged time

MSS, low Immunoscore MSS, high Immunoscore

80 60 40 20 0

MSI-high: HR for high vs low Immunoscore 0·56 (95% CI 0·34–0·90); p=0·0150 MSS: HR for high vs low 0·56 (95% CI 0·46–0·68); p<0·0001 0

Number at risk MSI, low Immunoscore 49 MSI, high Immunoscore 255 MSS, low Immunoscore 353 MSS, high Immunoscore 922

1

2

39 206 273 791

31 178 225 690

3 4 5 Time from surgery (years) 23 164 200 621

21 146 174 565

19 130 155 502

6

7

8

13 101 127 379

11 71 94 286

9 58 72 234

Figure 3: Kaplan-Meier estimates of disease-free survival (A) Kaplan-Meier curves for disease-free survival according to the Immunoscore in all cohorts of patients with AJCC/UICC TNM stage I–III colon cancers. (B) Disease-free survival according to the Immunoscore and MSI status in 1579 patients with known MMR status. The Immunoscore is based on two categories (low Immunoscore vs intermediate plus high Immunoscore). 5 year survival: 56% (95% CI 43−73) for patients with low Immunoscore and MSI (green), 75% (69−80) for patients with high Immunoscore and MSI (red), 53% (48−59) for patients with low Immunoscore and MSS (black), and 72% (69−74) for patients with high Immunoscore and MSS (blue). MSI=microsatellite instability. MMR=mismatch repair. MSS=microsatellite stable.

to recurrence, overall survival, and disease-free survival in the training, internal validation, and external vali­ dation sets (all p<0·05; appendix). The two independent cohorts (internal and external validation sets) thus confirmed the findings from the first cohort (training set). We investigated the analytical performance of Immuno­ score. Software was used to determine the mean staining intensities of each slide, which allowed the reliability of each staining to be judged and avoided underestimation of the total cell count (figure 2A–B; appendix). This measurement ensured a high level (>80%) of homogeneity between centres for intensity of staining in a set of control slides. Overall, the mean density in the colon tumour and invasive margin was 693 cells per mm² and 1169 cells per mm², respectively, for CD3+ T cells and 241 cells per mm² and 435 cells per mm², respectively, for CD8+ cells, with a large variation between patients 2133

Articles

(figure 2C; appendix). Selected images of tissue stained for CD3 and CD8 (n=36) from five centres in Belgium, Canada, China, France, and the USA (representative of the whole cohort) with Immunoscores ranging from 2·5th to 90th percentiles were reanalysed by eight pathologists from different centres, and a strong interobserver re­producibility was found for the determi­ nation of mean cell densities in each tumour region (r=0·97 for tumour; r=0·97 for invasive margin; p<0·0001; figure 2D–E; appendix). We detected a 2·1% variation in the mean percentile of CD3+ and CD8+ T-cell den­sities between observers (appendix). This shows the reproducibility of Immunoscore within the ranges of immune cell densities that are found in colon tumours. The reproducibility of the results and the prognostic performance of the Immunoscore were compared with

Unadjusted stratified Cox model Immunoscore, three-category (CD3/CD8 colon tumour and invasive margin)

that of a visual assessment of the density of tumourinfiltrating T cells in tumour tissue stained with haematoxylin and eosin. To this end, 268 representative cases from the cohort were assessed by 11 observers. Only 4% of cases were concordant between all observers, and a total absence of concordance was evident in 45% of the cases (figure 2G; appendix). By contrast, an 88% con­cordance between observers was evident for the Immunoscore (figure 2F; appendix). A 48% discordance was found between the Immunoscore and the density of tumour-infiltrating T cells, as determined by visual assessment (appendix). Furthermore, patient survival was predicted more accurately with the Immunoscore than by visual assessment of tumour-infiltrating T cell (appendix). Overall, the consensus Immunoscore quantification was standardised, reproducible, robust, and quantitative.

Time to recurrence (314/1562)*

Disease-free survival (590/1562)*

Overall survival (491/1562)*

HR (95% CI)

HR (95% CI)

HR (95% CI)

·· ··

p value ·· <0·0001†

C index (95% CI) 0·62 (0·56–0·68) ··

·· ··

p value ··

C index (95% CI) 0·58 (0·54–0·63)

<0·0001†

··

··

p value ··

··

C index (95% CI) 0·58 (0·53–0·63)

<0·0001†

··

Intermediate vs low

0·447 (0·349–0·572)

<0·0001‡

0·588 (0·487–0·710)

<0·0001‡

··

0·617 (0·503–0·757)

<0·0001‡

··

High vs low

0·239 (0·168–0·341)

<0·0001‡

0·429 (0·338–0·545)

<0·0001‡

··

0·496 (0·384–0·640)

<0·0001‡

··

Multivariable stratified Cox model Immunoscore, three-category (CD3/CD8 colon tumour and invasive margin)

·· ··

··

0·74 (0·67–0·80)

<0·0001†

 ··

··

··

0·66 (0·61–0·71)

<0·0001†

··

··

··

··

0·64 (0·58–0·69)

<0·0001†

··

Intermediate vs low

0·488 (0·381–0·626)

<0·0001‡

··

0·622 (0·515–0·753)

<0·0001‡

··

0·654 (0·532–0·805)

<0·0001‡

··

High vs low

0·328 (0·229–0·472)

<0·0001‡

··

0·511 (0·401–0·652)

<0·0001‡

··

0·558 (0·429–0·726)

<0·0001‡

··

Sex Women vs men

·· 0·811 (0·646–1·017)

·· ··

·· 0·867 (0·735–1·022)

·· ··

0·1355†

··

0·872 (0·728–1·044)

··

0·1355‡

··

<0·0001†

··

<0·0001†

··

0·0059†

··

1·611 (0·551–4·710)

0·3839‡

··

1·491 (0·778–2·860)

0·2289‡

··

1·636 (0·800–3·345)

0·1774‡

··

T3 vs T1

2·707 (0·994–7·371)

0·0514‡

··

2·084 (1·133–3·833)

0·0182‡

··

2·088 (1·065–4·096)

0·0322‡

··

T4 vs T1

4·991 (1·803–13·818)

0·0020‡

··

2·850 (1·518–5·351)

0·0011‡

··

2·657 (1·322–5·339)

0·0061‡

··

N stage

<0·0001†

··

<0·0001†

··

<0·0001†

··

N1 vs N0

1·943 (1·477–2·555)

<0·0001‡

··

1·563 (1·274–1·918)

<0·0001‡

··

1·327 (1·056–1·667)

0·0150‡

··

N2 vs N0

3·118 (2·315–4·200)

<0·0001‡

··

2·272 (1·793–2·879)

<0·0001‡

··

1·992 (1·530–2·594)

<0·0001‡

··

MSI status (derived) Deficient MMR vs proficient MMR Age

··

 ··

0·0894† 0·0894‡

T2 vs T1

T stage (grouped T4 version)

 ··

0·0696† 0·0696‡

 ··

··

··

··

0·0064†

··

0·6677†

··

0·6107†

··

0·608 (0·425–0·870)

0·0064‡

··

0·953 (0·767–1·185)

0·6677‡

··

1·063 (0·841–1·343)

 ··

0·6107‡

··

0·999 (0·995–1·003)

0·5695†

··

1·001 (1·000–1·002)

0·2023†

··

1·002 (1·000–1·003)

0·0143†

··

(Table 1 continues on next page)

2134

www.thelancet.com Vol 391 May 26, 2018

Articles

Time to recurrence (314/1562)*

Disease-free survival (590/1562)*

Overall survival (491/1562)*

HR (95% CI)

HR (95% CI)

HR (95% CI)

p value

C index (95% CI)

p value

C index (95% CI)

p value

C index (95% CI)

(Continued from previous page) Multivariable stratified Cox model with AJCC/UICC TNM stage

··

Immunoscore, three-category (CD3/CD8 colon tumour and invasive margin)

··

··

0·72 (0·65–0·78)

<0·0001†

··

·· ··

··

0·67 (0·62–0·72)

<0·0001†

··

·· ··

··

0·68 (0·63–0·74)

<0·0001†

··

Intermediate vs low

0·489 (0·382–0·626)

<0·0001‡

··

0·623 (0·516–0·752)

<0·0001‡

··

0·668 (0·544–0·821)

0·0001‡

··

High vs low

0·327 (0·228–0·469)

<0·0001‡

··

0·502 (0·394–0·639)

<0·0001‡

··

0·560 (0·431–0·728)

<0·0001‡

··

0·776 (0·619–0·974)

0·0286‡

··

0·809 (0·686– 0·954)

0·0118‡

··

0·800 (0·668–0·959)

Sex Women vs men

0·0286†

0·0118†

··

·· ··

<0·0001†

··

<0·0001†

··

<0·0001†

··

II vs I

2·210 (1·336–3·660)

0·0020‡

··

1·672 (1·236–2·262)

0·0008‡

··

1·661 (1·197–2·305)

0·0024‡

··

III vs I

5·391 (3·292–8·826)

<0·0001‡

··

3·166 (2·336–4·29)

<0·0001‡

··

2·655 (1·905–3·702)

<0·0001‡

··

AJCC/UICC TNM stage

··

MSI status (derived) Deficient MMR vs proficient MMR Age

··

··

··

 ··

0·0160† 0·0160‡

0·0118†

··

0·5907†

··

0·7999†

··

0·631 (0·441–0·903)

0·0118‡

··

0·942 (0·758–1·171)

0·5907‡

··

1·031 (0·816–1·302)

··

0·7999‡

··

1·002 (0·993–1·012)

0·6138†

··

1·032 (1·025–1·040)

<0·0001†

··

1·045 (1·036–1·054)

<0·0001†

··

MSI=microsatellite instability. MMR=mismatch repair. AJCC/UICC=American Joint Committee on Cancer/Union for International Cancer Control. *Numbers in parentheses are number of events/total number of patients. †Stratified type-3 Wald p value. ‡Stratified covariate Wald p value. The multivariate analysis was stratified by centre.

Table 1: Multivariate analysis of three-category Immunoscore AJCC/UICC TNM stage I–III adjusted for MSI status

In patients with stage II tumours (n=1434), a high Immunoscore was associated with the lowest risk of recurrence and the highest disease-free survival and overall survival (p<0·05; appendix). Recurrence at 3 years was recorded for 23 (6%) patients with high Immunoscore, 73 (11%) patients with intermediate Immunoscore, and 77 (20%) patients with low Immuno­ score. Recurrence at 5 years was seen in 31 (8%) patients with high Immunoscore, 95 (14%) patients with intermediate Immunoscore, and 91 (23%) patients with low Immuno­ score (unadjusted HR for high vs low Immuno­ score 0·33, 95% CI 0·21–0·52; p<0·0001; C index=0·60; appendix). The beneficial effect of a high Immunoscore persisted in the presence of signs of tumour emboli (ie, in patients with venous emboli, lymphatic invasion, or perineural invasion; appendix; data not shown). Similar results were found for the adjusted model for Immunoscore, age, sex, T stage, N stage, and MSI status and when stratified by city centres. Similar results were also found for diseasefree survival and overall survival (appendix) and when regrouping the Immunoscore into two categories (appendix). The Immunoscore thus significantly predicted survival in patients with stage II colon cancer. The results of the bivariable analysis, in which all patients at all stages were pooled (n=2681), showed that the Immunoscore was a strong predictor for time to www.thelancet.com Vol 391 May 26, 2018

recurrence, overall survival, and disease-free survival (all p<0·0001; appendix). Disease-free survival at 5 years was recorded for 428 (75%) patients with a high Immuno­ score, 688 (70%) patients with an intermediate Immunoscore, and 271 (57%) patients with a low Immuno­score (figure 3A; appendix). The data showed that Immuno­ score is valid in each continent (unadjusted HR for high vs low Immuno­score for time to recurrence in North America 0·13, 95% CI 0·05–0·30; p<0·0001; Asia 0·19, 0·04–0·83; p=0·0271; and Europe 0·31, 0·22–0·43; p<0·0001). 1579 patient samples were included in the analysis of MSI status. When stratified into three Immunoscore categories, MSI tumours were associated with a high Immunoscore in 138 (45%) of 304 cases, whereas a high Immunoscore was observed in 273 (21%) of 1275 patients with MSS tumours. When stratified into two Immuno­ score categories, patients with high Immunoscore had prolonged disease-free survival (figure 3B; appendix), time to recurrence, and overall survival, irrespective of their microsatellite status (unadjusted HR for high vs low Immunoscore in MSI-high 0·56, 95% CI 0·34–0·90; p=0·0150; and unadjusted HR for high vs low Immunoscore in MSS 0·56, 0·46–0·68; p<0·0001; figure 3B; appendix). A similar profile was found for highly infiltrated tumours with a high Immuno­ score (in the three-category Immunoscore analysis; 2135

Articles

A Mucinous (colloid) Sidedness Sex T stage N stage MSI Venous emboli Lymphatic invasion Perineural invasion VELIPI Differentiation Immunoscore (high, intermediate, low) All clinical parameters Clinical plus Immunoscore (high, intermediate, low)

*

Predictive accuracy (iAUC)

0·65

0·60

0·55

0·50

B Relative variable contribution

N stage

Sidedness Mucinous (colloid) MSI Sex

Immunoscore (high, intermediate, low)

T stage N stage VELIPI

Differentiation

Clinical parameters

VELIPI Differentiation T stage

Clinical parameters plus Immunoscore

C Relative variable contribution

AJCC/UICC TNM stage

Differentiation

VELIPI

Clinical parameters

Immunoscore (high, intermediate, low)

AJCC/UICC TNM stage

Sidedness Mucinous (colloid) MSI Sex

VELIPI Differentiation

Clinical parameters plus Immunoscore

Figure 4: Performance of Immunoscore compared with clinico-pathological parameters including AJCC/UICC TNM staging (A) Clinical performance of tumour-related and immune-related risk parameters. The predictive accuracy for overall survival based on the iAUC with 1000 × bootstrap resampling for each parameter is shown in a box plot. Median values of 1000 × bootstrap resampling are shown with thick lines and the point estimates (based on the initial data) are shown by diamonds. (B) Relative importance of each risk parameter to survival risk using the χ² proportion test for clinical parameters (including T stage and N stage; left) and for clinical parameters plus Immunoscore (right). (C). Relative importance of each risk parameter to survival risk using the χ² proportion test for clinical parameters (including UICC/AJCC TNM classification; left) and for clinical parameters plus Immunoscore (right). iAUC=integrated area under the ROC curve. MSI=microsatellite instability. AJCC/UICC=American Joint Committee on Cancer/Union for International Cancer Control. *Significant log likelihood ratio p<0·0001 when adding the three-category Immunoscore to the clinical model.

appendix). Additionally, patients with weakly infiltrated MSI tumours did not have a survival advantage as compared with patients with MSS tumours. Similar results were found when analysing Immunoscore and 2136

MSI in patients with stage II tumours (appendix). Immunoscore was also significant (all p<0·0001) for time to recurrence, overall survival, and disease-free survival within the subgroup of patients with stage II cancer, MSS, and who were not receiving chemotherapy. In multivariable stratified Cox model combining MSI status with the Immunoscore, MSI remained a significant factor for time to recurrence but was not a significant factor for disease-free survival and overall survival and was dependent on Immunoscore (table 1; appendix). The beneficial effect of the MSI status was therefore mainly related to its capacity to induce a strong immunity (ie, a high Immunoscore). Immunoscore was a significant predictor of diseasefree survival in an unadjusted stratified Cox model (table 1; C index=0·58). Similar results were found for time to recurrence and overall survival (table 1; appendix). In each independent dataset and in the entire cohort, the results of the Cox multivariable analysis for time to recurrence adjusted for Immunoscore, age, sex, T stage, N stage, and MSI status, and when stratified by centre, revealed a significant prognostic value of Immunoscore in the three-category (table 1) and the two-category (appendix) analysis. Immunoscore was a significant predictor for time to recurrence, disease-free survival, and overall survival, with C indexes of 0·74 (p<0·0001), 0·66 (p<0·0001), and 0·64 (p<0·0001), respectively (appendix). Furthermore, without dichotomisation, when considered as a linear variable, the Immunoscore remained a highly sig­ nificant factor in a multivariate analysis for time to recurrence, disease-free survival, and overall survival (all p<0·0001). The ability of Immunoscore to predict overall survival was superior to that of existing tumour risk parameters such as grade of differentiation, MSI status, muc­inous colloid type, sidedness, venous emboli, lymphatic invasion, perineural invasion, and VELIPI (figure 4A). Immunoscore had similar iAUC values as T stage and N stage. Furthermore, adding the Immunoscore to a model that combines all clinical variables significantly improved the prediction for overall survival (likelihood ratio p<0·0001). We analysed the relative contribution of each parameter to predict survival and observed that the most important variables were N stage, T stage, differentiation, VELIPI, sex, and MSI status. Immunoscore was stronger than all these clinical parameters (figure 4B). Furthermore, the Immunoscore was predictive not only for patients with T1–3 and N0–1 tumours, but for patients with T4 and N2 tumours. Overall survival HR was 0·57 (95% CI 0·41–0·80; Wald p=0·0011) in the T4 subgroup and 0·67 (0·45–0·98; p=0·0417) in the N2 subgroup (appendix). Finally, a multivariate Cox analysis for overall survival, stratified by centre, for all available clinical parameters revealed that all Immunoscore categories were significant (HR for high vs low Immunoscore 0·47, 95% CI www.thelancet.com Vol 391 May 26, 2018

Articles

0·33–0·65; p<0·0001). T stage and N stage were significant risk parameters, whereas all other existing tumour risk para­ meters were no longer significant (table 2). Thus, of all clinical parameters, the relative contribution to the risk showed that Immunoscore (47%) was better than AJCC/UICC TNM staging (28%), grade of differentiation (15%), VELIPI (8%), sex (<3%), mucinous, and MSI status (figure 4C; appendix). The results of the multivariable analysis for overall survival using AJCC/UICC TNM staging showed that TNM staging and Immunoscore remained significant tumour risk parameters.

Discussion Since the early 1900s, immune infiltration of cancers has been suspected to be a positive factor for patient outcome;30 however, these insights did not have a major influence on cancer classification or clinical decision making. The recent demonstration of the prognostic importance of the immune contexture,7 together with new image assessment software to enumerate cells in tumours, has led the SITC to promote an international validation of the derived Immunoscore assay.28 This study complied with the STARD reporting guidelines for diagnostic accuracy studies (appendix). Each centre of the consortium in North America, Europe, and Asia analysed its own cohort under the supervision of the reference centre in France. A strict validation of the staining quality and intensity ensured staining homogeneity between centres and repro­ ducibility of the cell counting. By using multiple quality controls, we showed that Immunoscore quantification is highly reproducible, objective, and robust, whereas the density of tumour-infiltrating lymphocytes is highly subjective and less reproducible. Discrepancies between the score of tumour-infiltrating lymphocyte density and Immunoscore were seen in 48% of the cases. Those differences also reflect the fact that haematoxylin and eosin staining of tumour-infiltrating lymphocytes gives a crude and subjective semi-quantitative evaluation of undefined cell populations with possible opposite functions (eg, CD4+ T cells with Th1 orientation vs Th2 orientation vs immune cells with regulatory functions [Treg cells], natural killer [NK] cells, NK T cells, B-cell subsets, innate lymphoid cells, or cytotoxic CD8 T cells). By contrast, Immunoscore is a quantification of specific T-cell subsets in specific tumour regions. We previously showed that of all immune cells involved in the in-situ immune reaction, CD3+ and CD8+ cells provided the optimal combination for prognostic purpose. The accuracy of prediction of survival times for the different patient groups was greater with a combined analysis of colon tumour plus invasive margin regions than with a single-region analysis.8 The markers, regions of interest, procedure, measurements, and strategy to quantify Immunoscore were decided during the consensus meeting in 2012. CD3 and CD8 were also chosen as www.thelancet.com Vol 391 May 26, 2018

HR (95% CI)

Wald p value

0·90 (0·72–1·12)

0·3400

T2 vs T1

1·49 (0·62–3·57)

0·3686

T3 vs T1

1·91 (0·84–4·38)

0·1238

T4 vs T1

2·36 (1·01–5·55)

0·0484

Female vs male T stage

N stage N1 vs N0

1·16 (0·89–1·52)

0·2770

N2 vs N0

1·58 (1·15–2·17)

0·0052

1·20 (0·94–1·54)

0·1488

Moderate vs well

0·91 (0·66–1·24)

0·5403

Poor-undifferentiated vs well

1·37 (0·9–2·08)

0·1421

Mucinous (colloid) type (yes vs no)

1·02 (0·78–1·33)

0·8741

Sidedness distal vs proximal

0·96 (0·76–1·21)

0·7362

VELIPI (yes vs no) Differentiation

Immunoscore Intermediate vs low

0·67 (0·52–0·86)

0·0014

High vs low

0·47 (0·33–0·65)

<0·0001

0·93 (0·68–1·27)

0·6356

MSI status (MSI vs MSS)

The significance of the Cox multivariate regression model was evaluated with the Wald p value. None of the parameters violated the Cox proportional hazards assumption (all p value >0·05). Events per total number of patients was 341 deaths per 1107 total patients. C index 0·64 (95% CI 0·59–0·69). HR=hazard ratio. MSI=microsatellite instability. MSS=microsatellite stable. VELIPI=venous emboli, lymphatic invasion, perineural invasion.

Table 2: Cox multivariate regression analysis of overall survival stratified by centre, combining Immunoscore with T stage, N stage, sex, VELIPI, histological grade, mucinous-colloid type, sidedness, and MSI status

markers because of the quality of the staining and the stability of these antigens. The statistical analyses were done by an independent and external group. Using these rigorous practices, we showed that the Immunoscore assay segregated patients into sub­ categories with significant (p<0·0001) differences for time to recurrence, disease-free survival, and overall survival. The immune infiltrate varied widely between patients, with 21% of patients with MSS tumours having a high Immunoscore. This reflects that the quality and the density of the immune infiltrate is affected by many factors such as the genetic background of the patient, the pre-existing tumour microenvironment, the tumour genetic features, and the gut microbiota, each of which is variable between individuals and influences each other. Multivariable analyses showed that this scoring system adds substantial power to discriminate patients with varied survival characteristics in addition to that provided by established prognostic variables, even for tumours with MSI features, as previously reported.12 Beyond the results obtained in localised cancers, the relevance of the Immunoscore could extend to metastatic disease because the Immunoscore identifies tumours that are likely to metastasise31 and predicts the prognosis of patients with brain24 and liver or lung25 metastases from primary colorectal cancer. 2137

Articles

A limitation of the study could be the heterogeneity of the study population of patients, who came from 13 countries and received standard-of-care treatment in real-life clinical practice. In all cohorts, patients who received preoperative treatment were systemically excluded to avoid possible bias on the immune infiltrate level within the primary resected tumours. Such broad evaluation in non-randomised studies was inherently the design of the study, and it allowed us to detect the robustness of the Immunoscore assay across multiple ethnicities, lifestyles, and types of patient care. Most of the failures in our series were due to deviation from the predefined, standardised operating procedure, which will be corrected in the future. It will be now important to validate the standardised Immunoscore assay and determine its prognostic value and prediction of chemotherapy response in randomised clinical trials with patients who have stage II or stage III colorectal cancer and are treated with adjuvant chemotherapy (eg, N0147 trial, International Duration Evaluation of Adjuvant therapy [IDEA] collaboration) or neo-adjuvant therapy (eg, PRODIGE 22 trial, FOxTROT trial). Whether patients with stage III tumours and a high Immunoscore could be eligible for a reduced chemo­therapy regimen should also be tested.29 These lines of investigation could provide a new categorisation of patients that would modify therapeutic strategy in colonic cancers. The reproducibility and robustness of the consensus Immunoscore as a strong prognostic factor in associ­ ation with the TNM classification system favours its implementation as a new component in the classification of cancer, designated TNM-Immune. This would re­ present the first standardised immune-based assay for the classification of cancer.4,32,33 Contributors JG and FP wrote the draft report and designed the study. JG initiated and coordinated the study. JG, FMM, PAA, BAF, and CB represented the International Immunoscore steering committee. DJS designed all statistical analyses. DJS, F-SO, and JPM did the statistical analysis. All authors analysed Immunoscore in their respective cohorts. All authors discussed early version of the report and provided comments and suggestions for change. All authors have approved the final report for publication. Declaration of interests JG, FP and, BM have patents associated with the immune prognostic biomarkers. JG is co-founder of HalioDx biotechnology company. All other authors declare no competing interests. Immunoscore a registered trademark owned by the National Institute of Health and Medical Research. Acknowledgments In memoriam of Daniel J Sargent. The work was supported by the SITC, by grants from National Institute of Health and Medical Research (INSERM), the LabEx Immuno-oncology, the Transcan ERAnet European project, Association pour la Recherche contre le Cancer, CARPEM, AP-HP, INCA translationnel, Japan-AMED (P-DIRECT) and MEXT (Grants-in-aid for Scientific Research-S), and Ministry of Health of the Czech Republic (AZV CR 15-28188A) Progres Q25-LF1, and with support from PathForce and Definiens. We thank the following organisations for supporting the Immunoscore project after the World Immunotherapy Cancer meeting in 2012: European Academy of Tumor Immunology; La Fondazione Melanoma Onlus;

2138

National Cancer Institute; Biotherapy Development Association; Canadian Cancer Immunotherapy Consortium; Cancer Immunotherapy Consortium; Cancer Research Institute; Association for Cancer Immunotherapy (Cinvasive marginT); Committee for Tumor Immunology and Bio-therapy; European Society for Cancer Immunology and Immunotherapy; Italian Network for Tumor Biotherapy; Japanese Association of Cancer Immunology; Nordic Center for Development of Antitumour Vaccines; Progress in Vaccination Against Cancer; Adoptive engineered T cell Targeting to Activate Cancer Killing (ATTACK); Tumor Vaccine and Cell Therapy Working Group; and Institut National du Cancer, France. References 1 Locker GY, Hamilton S, Harris J, et al. ASCO 2006 update of recommendations for the use of tumor markers in gastrointestinal cancer. J Clin Oncol 2006; 24: 5313–27. 2 Sobin L, Wittekind C. TNM classification of malignant tumors. New York, NY: Wiley-Liss, 2002. 3 Weitz J, Koch M, Debus J, Hohler T, Galle PR, Buchler MW. Colorectal cancer. Lancet 2005; 365: 153–65. 4 Galon J, Mlecnik B, Bindea G, et al. Towards the introduction of the ‘Immunoscore’ in the classification of malignant tumours. J Pathol 2014; 232: 199–209. 5 Guinney J, Dienstmann R, Wang X, et al. The consensus molecular subtypes of colorectal cancer. Nat Med 2015; 21: 1350–56. 6 Fridman WH, Pages F, Sautes-Fridman C, Galon J. The immune contexture in human tumours: impact on clinical outcome. Nat Rev Cancer 2012; 12: 298–306. 7 Galon J, Angell HK, Bedognetti D, Marincola FM. The continuum of cancer immunosurveillance: prognostic, predictive, and mechanistic signatures. Immunity 2013; 39: 11–26. 8 Galon J, Costes A, Sanchez-Cabo F, et al. Type, density, and location of immune cells within human colorectal tumors predict clinical outcome. Science 2006; 313: 1960–64. 9 Koelzer VH, Dawson H, Andersson E, et al. Active immunosurveillance in the tumor microenvironment of colorectal cancer is associated with low frequency tumor budding and improved outcome. Transl Res 2015; 166: 207–17. 10 Laghi L, Bianchi P, Miranda E, et al. CD3+ cells at the invasive margin of deeply invading (pT3–T4) colorectal cancer and risk of post-surgical metastasis: a longitudinal study. Lancet Oncol 2009; 10: 877–84. 11 Lee WS, Park S, Lee WY, Yun SH, Chun HK. Clinical impact of tumor-infiltrating lymphocytes for survival in stage II colon cancer. Cancer 2010; 116: 5188–99. 12 Mlecnik B, Bindea G, Angell HK, et al. Integrative analyses of colorectal cancer show Immunoscore is a stronger predictor of patient survival than microsatellite instability. Immunity 2016; 44: 698–711. 13 Mlecnik B, Tosolini M, Kirilovsky A, et al. Histopathologic-based prognostic factors of colorectal cancers are associated with the state of the local immune reaction. J Clin Oncol 2011; 29: 610–18. 14 Nosho K, Baba Y, Tanaka N, et al. Tumour-infiltrating T-cell subsets, molecular changes in colorectal cancer and prognosis: cohort study and literature review. J Pathol 2010; 222: 350–66. 15 Ogino S, Galon J, Fuchs CS, Dranoff G. Cancer immunology— analysis of host and tumor factors for personalized medicine. Nat Rev Clin Oncol 2011; 8: 711–19. 16 Ogino S, Nosho K, Irahara N, et al. Lymphocytic reaction to colorectal cancer is associated with longer survival, independent of lymph node count, microsatellite instability, and CpG island methylator phenotype. Clin Cancer Res 2009; 15: 6412–20. 17 Pages F, Berger A, Camus M, et al. Effector memory T cells, early metastasis, and survival in colorectal cancer. N Engl J Med 2005; 353: 2654–66. 18 Sinicrope FA, Rego RL, Ansell SM, Knutson KL, Foster NR, Sargent DJ. Intraepithelial effector (CD3+)/regulatory (FoxP3+) T-cell ratio predicts a clinical outcome of human colon carcinoma. Gastroenterology 2009; 137: 1270–79. 19 Emile JF, Julie C, Le Malicot K, et al. Prospective validation of a lymphocyte infiltration prognostic test in stage III colon cancer patients treated with adjuvant FOLFOX. Eur J Cancer 2017; 82: 16–24.

www.thelancet.com Vol 391 May 26, 2018

Articles

20 Pages F, Kirilovsky A, Mlecnik B, et al. In situ cytotoxic and memory T cells predict outcome in patients with early-stage colorectal cancer. J Clin Oncol 2009; 27: 5944–51. 21 Mlecnik B, Bindea G, Angell HK, et al. Functional network pipeline reveals genetic determinants associated with in situ lymphocyte proliferation and survival of cancer patients. Sci Transl Med 2014; 6: 228ra37. 22 Broussard EK, Disis ML. TNM staging in colorectal cancer: T is for T cell and M is for memory. J Clin Oncol 2011; 29: 601–03. 23 Bindea G, Mlecnik B, Tosolini M, et al. Spatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer. Immunity 2013; 39: 782–95. 24 Berghoff AS, Fuchs E, Ricken G, et al. Density of tumorinfiltrating lymphocytes correlates with extent of brain edema and overall survival time in patients with brain metastases. Oncoimmunology 2016; 5: e1. 25 Mlecnik B, Van den Eynde M, Bindea G, et al. Comprehensive intrametastatic immune quantification and major impact of immunoscore on survival. J Natl Cancer Inst 2018; 110: 97–108. 26 Concha-Benavente F, Ferris RL. Oncogenic growth factor signaling mediating tumor escape from cellular immunity. Curr Opin Immunol 2017; 45: 52–59.

www.thelancet.com Vol 391 May 26, 2018

27 Whiteside TL, Demaria S, Rodriguez-Ruiz ME, Zarour HM, Melero I. Emerging opportunities and challenges in cancer immunotherapy. Clin Cancer Res 2016; 22: 1845–55. 28 Galon J, Pages F, Marincola FM, et al. The immune score as a new possible approach for the classification of cancer. J Transl Med 2012; 10: 1. 29 Grothey A, Sobrero AF, Shields AF, et al. Duration of adjuvant chemotherapy for stage III colon cancer. N Engl J Med 2018; 378: 1177–88. 30 MacCarty W, Mahle A. Relation of differentiation and lymphocytic infiltration to postoperative longevity in gastric carcinoma. J Lab Clin Med 1921; 6: 473. 31 Mlecnik B, Bindea G, Kirilovsky A, et al. The tumor microenvironment and Immunoscore are critical determinants of dissemination to distant metastasis. Sci Transl Med 2016; 8: 327ra26. 32 Benson AB, 3rd, Hamilton SR. Path toward prognostication and prediction: an evolving matrix. J Clin Oncol 2011; 29: 4599–601. 33 Emens LA. It’s TIME for a biomarker-driven approach to cancer immunotherapy. J Immunother Cancer 2016; 4: 43.

2139