IEBBA VALERIO (IT)
ZITVOGEL LAURENCE (FR)
INST NAT SANTE RECH MED (FR)
UNIV PARIS SACLAY (FR)
WO2021063948A1 | 2021-04-08 | |||
WO2022261382A1 | 2022-12-15 | |||
WO2022157207A1 | 2022-07-28 | |||
WO2022157207A1 | 2022-07-28 |
US20220016188A1 | 2022-01-20 |
BERTRAND ROUTY; EMMANUELLE LE CHATELIER; LISA DEROSA; CONNIE P M DUONG; MARYAM TIDJANI ALOU; ROMAIN DAILLÈRE; AURÉLIE FLUCKIGER; M: "Gut microbiome influences efficacy of PD-1-based immunotherapy against epithelial tumors", SCIENCE, vol. 359, no. 6371, 5 January 2018 (2018-01-05), US, pages 91 - 97, XP055554928, ISSN: 0036-8075, DOI: 10.1126/science.aan3706
BENARD, CLEMENT, GERARD BIAU, SEBASTIEN DA VEIGA, AND ERWAN SCORNET.: "SIRIUA: Stable and Interpretable RUle Set for classification", ELECTRONIC JOURNAL OF STATISTICS, vol. 15, no. 1, 2021, pages 427 - 505
BERNARDO, DAVID, BORJA SANCHEZ, HAFID O. AL-HASSI, ELIZABETH R. MANN, MARIA C. URDACI, STELLA C. KNIGHT, AND ABELARDO MARGOLLES.: "Microbiota/Host Crosstalk Biomarkers: Regulatory Response of Human Intestinal Dendritic Cells Exposed to Lactobacillus Extracellular Encrypted Peptide ", PLOS ONE, vol. 7, no. 5, 2021, pages e36262
CARBONERO, FRANCK, ANN BENEFIEL, AMIR ALIZADEH-GHAMSARI, AND H. REX GASKINS.: "Microbial pathways in colonic sulfur metabolism and links with health and disease", FRONTIERS IN PHYSIOLOGY, vol. 3, 2012
CHANGCHIH-CHUNGCHIH-JEN LIN: "LIBSVM: A library for support vector machines", ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY, vol. 2, no. 3, 2011, pages 1 - 27
"Baseline Gut Microbiota Predicts Clinical Response and Colitis in Metastatic Melanoma Patients Treated with Ipilimumab", ANNALS OF ONCOLOGY: OFFICIAL JOURNAL OF THE EUROPEAN SOCIETY FOR MEDICAL ONCOLOGY, vol. 28, no. 6, pages 1368 - 79
CHOILSEUNGMARTIN J. BLASER: "The Human Microbiome: At the Interface of Health and Disease", NATURE REVIEWS. GENETICS, vol. 13, no. 4, 2012, pages 260 - 70
HAMILTONDANIEL AMADOR-NOGUEZOPHELIA S. VENTURELLI: "Design of Synthetic Human Gut Microbiome Assembly and Butyrate Production", NATURE COMMUNICATIONS, vol. 12, no. 1, 2021, pages 3254
ENRICA TERESA TANDA, FRANCESCO SPAGNOLO: "Integrated Analysis of Concomitant Medications and Oncological Outcomes from PD-1/PD-L1 Checkpoint Inhibitors in Clinical Practice", JOURNAL FOR IMMUNOTHERAPY OF CANCER, vol. 8, no. 2, 2020, pages e001361
COSSEAUCELINEDEIRDRE ADEVINEEDIE DULLAGHANJENNIFER L. GARDYAVINASH CHIKATAMARLASHAAN GELLATLYLORRAINE L. YU ET AL.: "The Commensal Streptococcus Salivarius K12 Downregulates the Innate Immune Responses of Human Epithelial Cells and Promotes Host-Microbe Homeostasis", INFECTION AND IMMUNITY, vol. 76, no. 9, 2008, pages 4163 - 75, XP002594153, DOI: 10.1128/IAI.00188-08
DAVAR, DIWAKAR, AMIRAN K. DZUTSEV, JOHN A. MCCULLOCH, RICHARD R. RODRIGUES, JOE-MARC CHAUVIN, ROBERT M. MORRISON, RICHELLE N. DEBL: "Fecal Microbiota Transplant Overcomes Resistance to Anti-PD-1 Therapy in Melanoma Patients ", SCIENCE, vol. 371, no. 6529, 2021, pages 595 - 602, XP055938683, DOI: 10.1126/science.abf3363
DEROSA, L., M. D. HELLMANN, M. SPAZIANO, D. HALPENNY, M. FIDELLE, H. RIZVI, N. LONG: "Negative Association of Antibiotics on Clinical Activity of Immune Checkpoint Inhibitors in Patients with Advanced Renal Cell and Non-Small-Cell Lung Cancer", ANNALS OF ONCOLOGY: OFFICIAL JOURNAL OF THE EUROPEAN SOCIETY FOR MEDICAL ONCOLOGY, vol. 29, no. 6, 2018, pages 1437 - 44
DEROSALISABERTRAND ROUTYMARINE FIDELLEVALERIO LEBBALAURIE ALLAEDOARDO PASOLLINICOLA SEGATA ET AL.: "Gut Bacteria Composition Drives Primary Resistance to Cancer Immunotherapy in Renal Cell Carcinoma Patients", EUROPEAN UROLOGY, vol. 78, no. 2, 2020, pages 195 - 206, XP086222713, DOI: 10.1016/j.eururo.2020.04.044
DEROSALISABERTRAND ROUTYGUIDO KROEMERLAURENCE ZITVOGEL.: "The Intestinal Microbiota Determines the Clinical Efficacy of Immune Checkpoint Blockers Targeting PD-1/PD-L1", ONCOIMMUNOLOGY, vol. 7, no. 6, 2018, pages e1434468, XP055954268, DOI: 10.1080/2162402X.2018.1434468
SYLVIE FRIARDJULIEN MAZIERES ET AL.: "Intestinal Akkermansia muciniphila predicts overall survival in advanced non-small cell lung cancer patients treated with anti-PD-1 antibodies: Results a phase II study", JOURNAL OF CLINICAL ONCOLOGY, vol. 39, 2021, pages 9019 - 9019
DEROSALISABERTRAND ROUTYANDREW MALTEZ THOMASVALERIO LEBBAGERARD ZALCMANSYLVIE FRIARDJULIEN MAZIERES ET AL.: "Intestinal Akkermansia Muciniphila Predicts Clinical Response to PD-1 Blockade in Patients with Advanced Non-Small-Cell Lung Cancer", NATURE MEDICINE, vol. 28, no. 2, 2022, pages 315 - 24, XP037700055, DOI: 10.1038/s41591-021-01655-5
DEROSALISA ET AL.: "Intestinal Akkermansia Muciniphila Predicts Clinical Response to PD-1 Blockade in Patients with Advanced Non-Small-Cell Lung Cance", NATURE, February 2022 (2022-02-01)
DIZMANNAZLILUIS MEZAPAULO BERGEROTMARICE ALCANTARATANYA DORFFYUNG LYOUPAUL FRANKEL ET AL.: "Nivolumab plus Ipilimumab with or without Live Bacterial Supplementation in Metastatic Renal Cell Carcinoma: A Randomized Phase 1 Trial", NATURE MEDICINE, vol. 28, no. 4, 2022, pages 704 - 12, XP037801520, DOI: 10.1038/s41591-022-01694-6
DORDEVIC, DANI, SIMONA JANCIKOVA, MONIKA VITEZOVA, AND IVAN KUSHKEVYCH: "Hydrogen Sulfide Toxicity in the Gut Environment: Meta-Analysis of Sulfate-Reducing and Lactic Acid Bacteria in Inflammatory Processes", JOURNAL OF ADVANCED RESEARCH, CHEMISTRY, BIOLOGY AND CLINICAL APPLICATIONS OF THE THIRD, 2021, pages 55 - 69
FRANKELARTHUR ELAURA A. COUGHLINJIWOONG KIMTHOMAS W. FROEHLICHYANG XIEEUGENE P. FRENKELANDREW Y. KOH: "Metagenomic Shotgun Sequencing and Unbiased Metabolomic Profiling identify Specific Human Gut Microbiota and Metabolites Associated with Immune Checkpoint Therapy Efficacy in Melanoma Patients", NEOPLASIA (NEW YORK, N.Y., vol. 19, no. 10, 2017, pages 848 - 55, XP055540712, DOI: 10.1016/j.neo.2017.08.004
FRIEDMANJONATHANLOGAN M. HIGGINSJEFF GORE: "Community Structure Follows Simple Assembly Rules in Microbial Microcosms", NATURE ECOLOGY & EVOLUTION, vol. 1, no. 5, 2017, pages 1 - 7
GACESA, RA. KURILSHIKOVA. VICH VILAT. SINHAM. A. Y. KLAASSENL. A. BOLTES. ANDREU-SANCHEZ ET AL.: "Environmental Factors Shaping the Gut Microbiome in a Dutch Population", NATURE, vol. 604, no. 7907, 2022, pages 732 - 39, XP037805444, DOI: 10.1038/s41586-022-04567-7
GHARAIBEHRAAD ZCHRISTIAN JOBIN: "Microbiota and Cancer Immunotherapy: In Search of Microbial Signals", GUT, vol. 68, no. 3, 2019, pages 385 - 88
GHOSH, TARINI S., MRINMOY DAS, IAN B. JEFFERY, AND PAUL W. O'TOOLE: "Adjusting for Age Improves Identification of Gut Microbiome Alterations in Multiple Diseases ", ELIFE, vol. 9, 2020, pages e50240
ROB KNIGHT: "Current Understanding of the Human Microbiome", NATURE MEDICINE, vol. 24, no. 4, 2018, pages 392 - 400
A. PRIETO ET AL.: "Gut Microbiome Modulates Response to Anti-PD-1 Immunotherapy in Melanoma Patients", SCIENCE (NEW YORK, N.Y., vol. 359, no. 6371, 2018, pages 97 - 103, XP055554925, DOI: 10.1126/science.aan4236
GUPTA, VINOD K., MINSUK KIM, UTPAL BAKSHI, KEVIN Y. CUNNINGHAM, JOHN M. DAVIS, KONSTANTINOS N. LAZARIDIS, HEIDI NELSON, NICHOLAS C: "A Predictive Index for Health Status Using Species-Level Gut Microbiome Profiling", NATURE COMMUNICATIONS, vol. 11, no. 1, 2020, pages 4635
CHAKSHU SAHIBERNHARD J. EIGL ET AL.: "Prognostic Factors for Overall Survival in Patients with Metastatic Renal Cell Carcinoma Treated with Vascular Endothelial Growth Factor-Targeted Agents: Results from a Large, Multicenter Study", JOURNAL OF CLINICAL ONCOLOGY: OFFICIAL JOURNAL OF THE AMERICAN SOCIETY OF CLINICAL ONCOLOGY, vol. 27, no. 34, 2009, pages 5794 - 99
IMHANN, FLORIS, MARC JAN BONDER, ARNAU VICH VILA, JINGYUAN FU, ZLATAN MUJAGIC, LISA VORK, ETTJE F. TIGCHELAAR: " Proton Pump Inhibitors Affect the Gut Microbiome", GUT, vol. 65, no. 5, 2016, pages 740 - 48
JACKSON, MATTHEW A., SERENA VERDI, MARIA-EMANUELA MAXAN, CHEOL MIN SHIN, JONAS ZIERER, RUTH C. E. BOWYER, TIPHAINE MARTIN: "Common Diseases and Prescription Medications in a Population-Based Cohort Common Diseases and Prescription Medications in a Population-Based Cohort", NATURE COMMUNICATIONS, vol. 9, no. 1, 2018, pages 2655
KAWASHIMATADAOMIAKEMI KOSAKAHUIMIN YANZIJIN GUORYOSUKE UCHIYAMARYUTARO FUKUIDAISUKE KANEKO ET AL.: "Double-Stranded RNA of Intestinal Commensal but Not Pathogenic Bacteria Triggers Production of Protective interferon-β", IMMUNITY, vol. 38, no. 6, 2013, pages 1187 - 97
LEEKARLA AANDREW MALTEZ THOMASLAURA A. BOLTEJOHANNES R. BJORKLAURA KIST DE RUIJTERFEDERICA ARMANINIFRANCESCO ASNICAR ET AL.: "Cross-Cohort Gut Microbiome Associations with Immune Checkpoint Inhibitor Response in Advanced Melanoma", NATURE MEDICINE, vol. 28, no. 3, 2022, pages 535 - 44, XP037768826, DOI: 10.1038/s41591-022-01695-5
LIMINBAOHONG WANGMENGHUI ZHANGMATTIAS RANTALAINENSHENGYUE WANGHAOKUI ZHOUYAN ZHANG ET AL.: "Symbiotic Gut Microbes Modulate Human Metabolic Phenotypes", PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, vol. 105, no. 6, 2008, pages 2117 - 22, XP055787558, DOI: 10.1073/pnas.0712038105
LIMETAANGELOBOYANG JIMAX LEVINFRANCESCO GATTOJENS NIELSEN: "Meta-Analysis of the Gut Microbiota in Predicting Response to Cancer Immunotherapy in Metastatic Melanoma", JCIINSIGHT, vol. 5, no. 23, 2020
MAGERLUKAS FREGULA BURKHARDNICOLA PETTNOAH C. A. COOKEKIRSTY BROWNHENA RAMAYSEUNGIL PAIK ET AL.: "Microbiome-Derived Inosine Modulates Response to Checkpoint Inhibitor Immunotherapy", SCIENCE (NEW YORK, N.Y.), vol. 369, no. 6510, 2020, pages 1481 - 89, XP093094032, DOI: 10.1126/science.abc3421
FANGALICIA M. COLEASCHARYA K. BALAJI ET AL.: "Intestinal Microbiota Signatures of Clinical Response and Immune-Related Adverse Events in Melanoma Patients Treated with Anti-PD-1", NATURE MEDICINE, vol. 28, no. 3, 2022, pages 545 - 56, XP037768832, DOI: 10.1038/s41591-022-01698-2
MYRIAM BENLAIFAOUIALEXIS NOLIN-LAPALME ET AL.: "A Natural Polyphenol Exerts Antitumor Activity and Circumvents Anti-PD-1 Resistance through Effects on the Gut Microbiota", CANCER DISCOVERY, vol. 12, no. 4, 2022, pages 1070 - 87
SHIRLENE PAULSTEPHANIE R. HOGUEQIN YU ET AL.: "Interaction of Bacterial Genera Associated with Therapeutic Response to Immune Checkpoint PD-1 Blockade in a United States Cohort", GENOME MEDICINE, vol. 14, no. 1, 2022, pages 35
JUSTIN T. TOMETICH, AMRITA BHATTACHARJEE, TULLIA C. BRUNO, DARIO A. A. VIGNALI: "Microbiota-Specific T Follicular Helper Cells Drive Tertiary Lymphoid Structures and Anti-Tumor Immunity against Colorectal Cancer", IMMUNITY, vol. 54, no. 12, 2021, pages 2812 - 2824, XP086899604, DOI: 10.1016/j.immuni.2021.11.003
PARK, ELIZABETH M., MANOJ CHELVANAMBI, NEAL BHUTIANI, GUIDO KROEMER, LAURENCE ZITVOGEL: "Targeting the Gut and Tumor Microbiota in Cancer", NATURE MEDICINE, vol. 28, no. 4, 2022, pages 690 - 703, XP037801560, DOI: 10.1038/s41591-022-01779-2
PESCHELSTEFANIECHRISTIAN L MULLERERIKA VON MUTIUSANNE-LAURE BOULESTEIXMARTIN DEPNER: "NetCoMi: network construction and comparison for microbiome data in R", BRIEFINGS IN BIOINFORMATICS, vol. 22, no. 4, 2021, pages bbaa290
MARYAM TIDJANI ALOUCONRAD RAUBER ET AL.: "Chemotherapy-Induced Ileal Crypt Apoptosis and the Ileal Microbiome Shape Immunosurveillance and Prognosis of Proximal Colon Cancer", NATURE MEDICINE, vol. 26, no. 6, 2020, pages 919 - 31, XP037173406, DOI: 10.1038/s41591-020-0882-8
ROUTY, BERTRAND, EMMANUELLE LE CHATELIER, LISA DEROSA, CONNIE P. M. DUONG, MARYAM TIDJANI ALOU, ROMAIN DAILLERE, AURELIE FLUCKIGER: "Gut Microbiome Influences Efficacy of PD-1-Based Immunotherapy against Epithelial Tumors ", SCIENCE (NEW YORK, N.Y., vol. 359, no. 6371, 2018, pages 91 - 97, XP055527739, DOI: 10.1126/science.aan3706
SANCHEZ-GOROSTIAGAALICIADJORDJE BAJICMELISA L. OSBORNEJUAN F. POYATOSALVARO SANCHEZ: "High-Order Interactions Distort the Functional Landscape of Microbial Consortia", PLOS BIOLOGY, vol. 17, no. 12, 2019, pages e3000550
LUIS G. BERMUDEZ-HUMARAN, VASCO AZEVEDO: "Anti-nflammatory Properties of Dairy Lactobacilli ", INFLAMMATORY BOWEL DISEASES, vol. 18, no. 4, 2012, pages 657 - 66
SHAIKHFYZA YJAMES R. WHITEJOELL J. GILLSTAIKI HAKOZAKICORENTIN RICHARDBERTRAND ROUTYYUSUKE OKUMA ET AL.: "A Uniform Computational Approach Improved on Existing Pipelines to Reveal Microbiome Biomarkers of Nonresponse to Immune Checkpoint Inhibitors", CLINICAL CANCER RESEARCH, vol. 27, no. 9, 2021, pages 2571 - 83
BIXBY, SAMUEL J. LEE, ANSHUMAN PANDA: "Endogenous Retroviral Signatures Predict Immunotherapy Response in Clear Cell Renal Cell Carcinoma", THE JOURNAL OF CLINICAL INVESTIGATION, vol. 128, no. 11, 2018, pages 4804 - 20, XP055800962, DOI: 10.1172/JCI121476
SMITHMELODYANQI DAIGUIDO GHILARDIKIMBERLY V. AMELSBERGSEAN M. DEVLINRAYMONE PAJARILLOOHN B. SLINGERLAND ET AL.: "Gut Microbiome Correlates of Response and Toxicity Following Anti-CD19 CAR T Cell Therapy", NATURE MEDICINE, vol. 28, no. 4, 2022, pages 713 - 23, XP037801533, DOI: 10.1038/s41591-022-01702-9
AUDE FLECHONGUILHEM ROUBAUD ET AL.: "Primary Results of STRONG: An Open-Label, Multicenter, Phase 3b Study of Fixed-Dose Durvalumab Monotherapy in Previously Treated Patients with Urinary Tract Carcinoma", EUROPEAN JOURNAL OF, vol. 163, 2022, XP086963519, DOI: 10.1016/j.ejca.2021.12.012
MARIE VETIZOUALEXANDRIA P. COGDILLMD A. WADUD KHAN ET AL.: "Dietary Fiber and Probiotics Influence the Gut Microbiome and Melanoma Immunotherapy Response", SCIENCE (NEW YORK, N.Y., vol. 374, no. 6575, 2021, pages 1632 - 40
MONICA D. PRAKASHKULMIRA NURGALI ET AL.: "Translocation and Dissemination of Commensal Bacteria in Post-Stroke Infection", NATURE MEDICINE, vol. 22, no. 11, 2016, pages 1277 - 84, XP037104244, DOI: 10.1038/nm.4194
TENG, HUAJING, YAN WANG, XIN SUI, JIAWEN FAN, SHUAI LI, XIAO LEI, CHEN SHI: "Gut Microbiota-Mediated Nucleotide Synthesis Attenuates the Response to Neoadjuvant Chemoradiotherapy in Rectal Cancer", CANCER CELL, vol. 41, no. 1, 2023
TERRISSE, SAFAE, LISA DEROSA, VALERIO LEBBA, FRANCOIS GHIRINGHELLI, INES VAZ-LUIS, GUIDO KROEMER, MARINE FIDELLE: "intestinal Microbiota Influences Clinical Outcome and Side Effects of Early Breast Cancer Treatment", DIFFERENTIATION, 2021, pages 1 - 19
NICOLA SEGATALAURENCE ZITVOGEL: "Gut OncoMicrobiome Signatures (GOMS) as next-Generation Biomarkers for Cancer Immunotherapy", NATURE, June 2023 (2023-06-01)
TSAYJUN-CHIEH JBENJAMIN G. WUIMRAN SULAIMANKATHERINE GERSHNERROSEMARY SCHLUGERYONGHUA LITING-AN YIE ET AL.: "Lower Airway Dysbiosis Affects Lung Cancer Progression", CANCER DISCOVERY, vol. 11, no. 2, 2021, pages 293 - 307
CAROLINE FLAMENTSYLVIE RUSAKIEWICZ ET AL.: "Anticancer Immunotherapy by CTLA-4 Blockade Relies on the Gut Microbiota", SCIENCE (NEW YORK, N.Y., vol. 350, no. 6264, 2015, pages 1079 - 84
WU, GUOJUN, TING XU, NAISI ZHAO, YAN Y. LAM, XIAOYING DING, DONGQIN WEI, JIAN FAN: "Two Competing Guilds as a Core Microbiome Signature for Health Recovery.", BIORXIV, 2022
YONEKURA, SATORU, SAFAE TERRISSE, CAROLINA ALVES COSTA SILVA, ANTOINE LAFARGE, VALERIO LEBBA, GLADYS FERRERE, ANNE-GAELLE GOUBET: "Cancer Induces a Stress Ileopathy Depending on B-Adrenergic Receptors and Promoting Dysbiosis That", CANCER DISCOVERY, vol. 12, 2022, pages 1128 - 1151
GIUSEPPE PIERACCINI, RICCARDO ZECCHI: "Microbiota Engage Aryl Hydrocarbon Receptor and Balance Mucosal Reactivity via Interleukin-22", IMMUNITY, vol. 39, no. 2, 2013, pages 372 - 85, XP055116601, DOI: 10.1016/j.immuni.2013.08.003
2018: "The Microbiome in Cancer Immunotherapy: Diagnostic Tools and Therapeutic Strategies", SCIENCE (NEW YORK, N. Y., vol. 359, no. 6382, pages 1366 - 70, XP055746988, DOI: 10.1126/science.aar6918
CLAIMS 1. A method of diagnosing intestinal dysbiosis in an individual, comprising: (i) in a sample from said individual comprising intestinal microbiota, assessing the presence or absence of bacterial species of a first species interacting group (“SIG1”) consisting of N1 bacterial species comprising at least 5, preferably at least 6, more preferably at least 7 bacterial species selected from the group consisting of Dialister invisus, Enterococcus faecalis, Haemophilus parainfluenzae, Veillonella atypica, Eggerthella lenta, Erysipelatoclostridium ramosum, Enterocloster bolteae, Alloscardovia omnicolens, Bifidobacterium dentium, Campylobacter concisus, Clostridium perfringens, Enterococcus durans, Enterococcus faecium, Klebsiella pneumoniae, Lacticaseibacillus paracasei, Lacticaseibacillus rhamnosus, Lactobacillus delbrueckii, Lactobacillus gasseri, Lactobacillus vaginalis, Lactococcus lactis, Lactococcus laudensis, Ligilactobacillus salivarius, Limosilactobacillus fermentum, Limosilactobacillus oris, Megasphaera micronuciformis, Mogibacterium diversum, Scardovia wiggsiae, Streptococcus anginosus, Streptococcus gordonii, Streptococcus infantis, Streptococcus mutans, Streptococcus parasanguinis, Streptococcus salivarius, Veillonella dispar, Veillonella parvula, Veillonella rogosae, Enterocloster aldensis, Enterocloster asparagiformis, Faecalimonas umbilicata, Gordonibacter urolithinfaciens, Actinomyces graevenitzii, Anaerostipes caccae, Blautia producta, Campylobacter gracilis, Clostridium innocuum, Clostridium scindens, Clostridium symbiosum, Collinsella SGB14754, Enorma massiliensis, Enterocloster clostridioformis, Fournierella massiliensis, Granulicatella adiacens, Hungatella hathewayi, Proteus mirabilis and Streptococcus oralis; (ii) in a sample from said individual comprising intestinal microbiota, assessing the presence or absence of bacterial species of a second species interacting group (“SIG2”) consisting of N2 bacterial species comprising at least 5, 6 or 7, preferably at least 10 to 12, more preferably at least 14 bacterial species selected from the group consisting of Dorea formicigenerans, Dorea longicatena, Eubacterium ventriosum, Eubacterium rectale, Holdemania filiformis, Parasutterella excrementihominis, Anaerostipes hadrus, Blautia obeum, Blautia wexlerae, Faecalibacterium prausnitzii, Roseburia hominis, Roseburia intestinalis, Roseburia inulinivorans, Ruminococcus bicirculans, FRNormCount= N 1 (iii) calculating a FRNormCount as follows: NSIG2 , wherein N 2 NSIG 1 is the number of bacterial species of SIG1 present in the sample and NSIG 2 is the number of bacterial species of SIG2 present in the sample; and/or (iv) calculating a S score as follows: , wherein NSIG 1 is the number of bacterial species of SIG1 present in the sample and NSIG 2 is the number of bacterial species of SIG2 present in the sample; wherein if the FRNormCount is inferior to a predetermined threshold TOPO1 and/or if the S score is superior to a predetermined threshold S2, 1 is assigned to the TOPOSCORE and the individual is likely not to have intestinal dysbiosis, and if the FRNormCount is superior to a predetermined threshold TOPO2 superior to TOPO1 and/or if the S score is inferior to a predetermined threshold S1 inferior to S2, 5 is assigned to the TOPOSCORE and the individual is likely to have intestinal dysbiosis. 2. The method of claim 1, wherein if TOPO1 ≤ FRNormCount ≤ TOPO2 and/or S1 ≤ S ≤ S2 (“grey zone”), the relative abundance of bacteria of the Akkermansia genus (Akk) is measured in a fecal material sample from said individual, wherein: a) if bacteria of the Akkermansia genus are present in the sample below a predetermined threshold (“Akk superior threshold”), 2 is assigned to the TOPOSCORE and the patient is likely not to have intestinal dysbiosis; and b) if no Akkermansia is present in the sample, 3 is assigned to the TOPOSCORE and the individual is likely to have intestinal dysbiosis; c) if bacteria of the Akkermansia genus are present in the sample above the Akk superior threshold, 4 is assigned to the TOPOSCORE and the individual is likely to have intestinal dysbiosis. 3. The method of claim 1 or claim 2, wherein: (i) the bacterial species of the first species interacting group (“SIG1”) are selected from the group consisting of Veillonella atypica, Erysipelatoclostridium ramosum, Enterocloster bolteae, Enterocloster aldensis, Alloscardovia omnicolens, Bifidobacterium dentium, Campylobacter concisus, Clostridium perfringens, Lacticaseibacillus paracasei, Lactobacillus gasseri, Lactobacillus vaginalis, Ligilactobacillus salivarius, Limosilactobacillus fermentum, Limosilactobacillus oris, Megasphaera micronuciformis, Streptococcus anginosus, Streptococcus gordonii, Streptococcus mutans, Streptococcus parasanguinis, Streptococcus salivarius, Veillonella dispar, Veillonella parvula, Actinomyces graevenitzii, Anaerostipes caccae, Blautia producta, Campylobacter gracilis, Clostridium innocuum, Clostridium scindens, Clostridium symbiosum, Collinsella SGB14754, Enorma massiliensis, Enterocloster clostridioformis, Fournierella massiliensis, Granulicatella adiacens, Hungatella hathewayi, Proteus mirabilis and Streptococcus oralis; and (ii) the bacterial species of the second species interacting group (“SIG2”) are selected from the group consisting of Dorea formicigenerans, Dorea longicatena, Eubacterium ventriosum, Eubacterium rectale, Anaerostipes hadrus, Blautia wexlerae, Faecalibacterium prausnitzii, Roseburia hominis, Roseburia intestinalis, Roseburia inulinivorans, Ruminococcus bicirculans, Ruminococcus lactaris, Candidatus Cibiobacter qucibialis, Clostridiales bacterium KLE1615, Faecalibacillus intestinalis, Lachnospira eligens, Agathobaculum butyriciproducens, Anaerobutyricum hallii, Blautia massiliensis, Clostridiaceae bacterium, Clostridium sp AF34 10BH, Lachnospira pectinoschiza, Anaerotignum faecicola, Clostridiaceae bacterium OM08 6BH, Clostridiaceae unclassified SGB4769, Clostridiales unclassified SGB15145, Clostridium fessum, Clostridium sp AM22 11AC, Clostridium sp AM333, Clostridium sp AM494BH, Coprobacter fastidiosus, Coprococcus comes, Coprococcus eutactus, Eubacterium ramulus, Faecalibacterium SGB15346, Firmicutes bacterium AF16 15, Gemmiger formicilis, Lachnospira sp NSJ 43, Lachnospiraceae bacterium OM0412BH, Lachnospiraceae bacterium WCA3601 WT 6H, Lacrimispora amygdalina, Mediterraneibacter butyricigenes, Oscillibacter sp ER4, Phocaeicola massiliensis and Roseburia sp AF0212. 4. The method of claim 1 or claim 2, wherein: (i) the bacterial species of the first species interacting group (“SIG1”) are selected from the group consisting of Dialister invisus, Enterococcus faecalis, Haemophilus parainfluenzae, Veillonella atypica, Eggerthella lenta, Erysipelatoclostridium ramosum, Enterocloster bolteae, Alloscardovia omnicolens, Bifidobacterium dentium, Campylobacter concisus, Clostridium perfringens, Enterococcus durans, Enterococcus faecium, Klebsiella pneumoniae, Lacticaseibacillus paracasei, Lacticaseibacillus rhamnosus, Lactobacillus delbrueckii, Lactobacillus gasseri, Lactobacillus vaginalis, Lactococcus lactis, Lactococcus laudensis, Ligilactobacillus salivarius, Limosilactobacillus fermentum, Limosilactobacillus oris, Megasphaera micronuciformis, Mogibacterium diversum, Scardovia wiggsiae, Streptococcus anginosus, Streptococcus gordonii, Streptococcus infantis, Streptococcus mutans, Streptococcus parasanguinis, Streptococcus salivarius, Veillonella dispar, Veillonella parvula, Veillonella rogosae, Enterocloster aldensis, Enterocloster asparagiformis, Faecalimonas umbilicata and Gordonibacter urolithinfaciens; and (ii) the bacterial species of the second species interacting group (“SIG2”) are selected from the group consisting of Dorea formicigenerans, Dorea longicatena, Eubacterium ventriosum, Eubacterium rectale, Holdemania filiformis, Parasutterella excrementihominis, Anaerostipes hadrus, Blautia obeum, Blautia wexlerae, Faecalibacterium prausnitzii, Roseburia hominis, Roseburia intestinalis, Roseburia inulinivorans, Ruminococcus bicirculans, Ruminococcus lactaris, Candidatus Cibiobacter qucibialis, Clostridiales bacterium KLE1615, Faecalibacillus intestinalis, Lachnospira eligens, Lacrimispora celerecrescens, Adlercreutzia equolifaciens, Agathobaculum butyriciproducens, Anaerobutyricum hallii, Blautia faecis, Blautia massiliensis, Clostridia unclassified SGB4447, Clostridiaceae bacterium, Clostridium sp AF34 10BH, Clostridium sp AF36 4, Eubacteriaceae bacterium, Fusicatenibacter saccharivorans, Lachnospira pectinoschiza, Lachnospiraceae bacterium and Roseburia faecis. 5. The method according to any one of claims 1 to 4, wherein: (i) the bacterial species of the first species interacting group (“SIG1”) are selected from the group consisting of Veillonella atypica, Erysipelatoclostridium ramosum, Enterocloster bolteae, Enterocloster aldensis, Alloscardovia omnicolens, Bifidobacterium dentium, Campylobacter concisus, Clostridium perfringens, Lacticaseibacillus paracasei, Lactobacillus gasseri, Lactobacillus vaginalis, Ligilactobacillus salivarius, Limosilactobacillus fermentum, Limosilactobacillus oris, Megasphaera micronuciformis, Streptococcus anginosus, Streptococcus gordonii, Streptococcus mutans, Streptococcus parasanguinis, Streptococcus salivarius, Veillonella dispar and Veillonella parvula; and (ii) the bacterial species of the second species interacting group (“SIG2”) are selected from the group consisting of Dorea formicigenerans, Dorea longicatena, Eubacterium ventriosum, Eubacterium rectale, Anaerostipes hadrus, Blautia wexlerae, Faecalibacterium prausnitzii, Roseburia hominis, Roseburia intestinalis, Roseburia inulinivorans, Ruminococcus bicirculans, Ruminococcus lactaris, Candidatus Cibiobacter qucibialis, Clostridiales bacterium KLE1615, Faecalibacillus intestinalis, Lachnospira eligens, Agathobaculum butyriciproducens, Anaerobutyricum hallii, Blautia massiliensis, Clostridiaceae bacterium, Clostridium sp AF34 10BH and Lachnospira pectinoschiza. 6. The method of any one of claims 1 to 3, wherein: - SIG1 comprises at least Enterocloster bolteae, Erysipelatoclostridium ramosum, Veillonella atypica, Clostridium symbiosum and Hungatella hathewayi, and/or - SIG2 comprises at least 10, preferably at least 12, more preferably at least 15 bacterial species selected from the group consisting of Anaerostipes hadrus, Blautia wexlerae, Dorea formicigenerans, Dorea longicatena, Eubacterium rectale, Eubacterium ventriosum, Faecalibacterium prausnitzii, Roseburia hominis, Roseburia intestinalis, Roseburia inulinivorans, Ruminococcus bicirculans, Ruminococcus lactaris, Coprococcus comes, Gemmiger formicilis and Phocaeicola massiliensis. 7. The method of claim 1, claim 2 or claim 4, wherein: - SIG1 comprises at least Dialister invisus, Enterococcus faecalis, Haemophilus parainfluenzae, Veillonella atypica, Eggerthella lenta, Erysipelatoclostridium ramosum and Enterocloster bolteae, and/or - SIG2 comprises at least 10, preferably at least 12, more preferably at least 15 bacterial species selected from the group consisting of Dorea formicigenerans, Dorea longicatena, Eubacterium ventriosum, Eubacterium eligens, Eubacterium rectale, Holdemania filiformis, Parasutterella excrementihominis, Anaerostipes hadrus, Blautia obeum, Blautia wexlerae, Faecalibacterium prausnitzii, Roseburia hominis, Roseburia intestinalis, Roseburia inulinivorans, Ruminococcus bicirculans and Ruminococcus lactaris. 8. The method of any one of the preceding claims, further comprising (v) in a sample from said individual comprising intestinal microbiota, assessing the presence or absence of functional pathways specifically related to SIG1 bacteria in the metagenome, wherein said SIG1-specific pathways are selected from purine nucleobase and pyrimidine deoxynucleotide phosphorylation and degradation, guanosine nucleotide de novo biosynthesis and L histidine degradation, (vi) in a sample from said individual comprising intestinal microbiota, assessing the presence or absence of functional pathways specifically related to SIG2 bacteria in the metagenome, wherein said SIG2-specific pathways are selected from autophagy-related pathways (polyamines such as S-adenosyl- L-methionine salvage, L-ornithine, L-arginine biosynthesis, putrescine biosynthesis) and sulfur oxidation, superpathway of β-D-glucuronide and D- glucuronate degradation, superpathway of L-alanine and L-aspartate, L- asparagine biosynthesis, wherein the presence of SIG2-specific functional pathways in the metagenome in the absence of SIG1-specific functional pathways indicates that the individual is of a “SIG2” genotype, and the presence of SIG1-specific functional pathways in the metagenome in the absence of SIG2-specific functional pathways indicates that the person is of a “SIG1” genotype. 9. The method of any one of the preceding claims, wherein said individual has a cancer amenable to immune-oncology (I-O) therapy. 10. The method of claim 9, wherein said individual has a non small cell lung cancer or a clear cell kidney cancer or an urothelial cancer or a colorectal cancer or a lymphoma. 11. A method of determining if a patient having a cancer amenable to immune-oncology (I-O) therapy is likely to be a good responder to said therapy, comprising assessing, using the method according to any one of the preceding claims, whether the patient has an intestinal dysbiosis, wherein the higher the patient’s TOPOSCORE, the lower the probability that the patient responds to said I-O therapy in absence of a microbiota-centered intervention (MCI) before administration of said I-O therapy. 12. The method of claim 11, wherein if the patient has a TOPOSCORE ≤ 2, the patient is likely to respond to said I-O therapy, and if the patient has a TOPOSCORE > 2, the patient is likely not to respond to said I-O therapy in absence of a microbiota-centered intervention (MCI) before administration of said I-O therapy. 13. The method of claim 11 or claim 12, wherein said I-O therapy is a treatment with an anti-PD1 antibody, an anti-PD-L1 antibody, an anti-PD-L2 antibody, an anti-CTLA4 antibody and/or a CAR T-cell targeting a tumor antigen, alone or combined with another antineoplastic agent. 14. The method of any of claims 11 to 13, wherein the TOPOSCORE is calculated before beginning the I-O therapy, and optionally after at least partial tumor resection in the individual. 15. Use of a TOPOSCORE calculated as described in any one of claims 2 to 8, as a theranostics tool to determine if an individual needs an MCI, wherein when the TOPOSCORE ≥ 3, the individual needs an MCI. 16. The use of claim 15, wherein when TOPOSCORE>3, the MCI is to be performed by Fecal Microbial Transplantation (FMT) and if TOPOSCORE=3, the MCI is to be performed by Fecal Microbial Transplantation (FMT) and/or by administering a bacterial composition comprising bacteria of the Akkermansia genus. 17. Use of a TOPOSCORE calculated as described in any one of claims 2 to 8, as a pharmacodynamics tool to monitor the evolution of the intestinal microbiota of an individual receiving an MCI and/or a treatment possibly impacting the intestinal microbiota and/or impacted by the intestinal microbiota. 18. Use of a TOPOSCORE calculated as described in claims any one of claims 2 to 8, as a theranostic tool for avoiding administering an I-O therapy to a patient likely to have a primary resistance thereto due to intestinal dysbiosis, and for stopping such a treatment if the patient develops a secondary resistance thereto. 19. Use of a TOPOSCORE calculated as described in any one of claims 2 to 8 for assessing whether a fecal material can be used in an MCI, wherein if the TOPOSCORE ≥ 3, the fecal material cannot be used in an MCI and if the TOPOSCORE ≤ 2, preferably if FRNormCount = 0, the fecal material can be used in an MCI. 20. A kit of parts for performing the method of any one of claims 1 to 10, comprising a primer pair and/or a nucleic acid probe specific for each of the bacterial species of the recited SIG1 and SIG2, and a primer pair and/or a nucleic acid probe specific for Akkermansia muciniphila. 21. The kit of parts of claim 20, which comprises a primer pair and/or a nucleic acid probe specific for each one of at least 20, preferably at least 30, more preferably at least 40 of the following bacterial species, wherein the ratio NSIG1:NSIG2 is about 1:3: - SIG1 bacteria: Streptococcus parasanguinis, Clostridium symbiosum, Streptococcus salivarius, Hungatella hathewayi, Clostridium scindens, Clostridium innocuum, Enterocloster aldensis, Veillonella parvula, Enterocloster bolteae, Erysipelatoclostridium ramosum, Enterocloster clostridioformis, Bifidobacterium dentium, Veillonella dispar and Actinomyces graevenitzii - SIG2 bacteria: Ruminococcus bicirculans, Faecalibacterium prausnitzii, Blautia wexlerae, Roseburia intestinalis, Gemmiger formicilis, Anaerostipes hadrus, Clostridiales bacterium KLE1615, Agathobaculum butyriciproducens, Dorea longicatena, Blautia massiliensis, Eubacterium rectale, Faecalibacterium SGB15346, Clostridium sp AF3410BH, Lachnospira eligens, Lachnospiraceae bacterium WCA3 601 WT 6H, Clostridium fessum, Anaerobutyricum hallii, Candidatus Cibiobacter qucibialis, Anaerotignum faecicola, Clostridiaceae unclassified SGB4769, Roseburia hominis, Clostridiaceae bacterium, Oscillibacter sp ER4, Clostridiaceae bacterium OM08 6BH, Roseburia inulinivorans, Phocaeicola massiliensis, Lacrimispora amygdalina, Firmicutes bacterium AF16 15, Coprococcus eutactus, Eubacterium ventriosum, Clostridiales unclassified SGB15145, Faecalibacillus intestinalis, Coprococcus comes, Roseburia sp AF0212, Clostridium sp AM494BH, Mediterraneibacter butyricigenes, Dorea formicigenerans, Coprobacter fastidiosus, Ruminococcus lactaris, Lachnospira sp NSJ 43, Clostridium sp AM22 11AC, Lachnospira pectinoschiza, Lachnospiraceae bacterium OM0412BH, Clostridium sp AM333 and Eubacterium ramulus, as well as a primer pair and/or a nucleic acid probe specific for Akkermansia muciniphila. |
41 Table 1. Comprehensive data from machine learning algorithms. a) models based on SIRUS pipeline (RF plus decision trees) as described in materials and methods. b) most predictive Random Forest (RF) model as described in materials and methods. c) models based on clinically-validated scores. d) fraction of individual hits by each model on the total OS12 data available (see Table 10). e) prediction percentage corrected for the coverage. f) difference among correct and uncorrect prediction percentages.
Example 1: Limitations in predicting clinical outcome across cohorts and cancer types using single metagenomic species (MGS) LUMIERE and ONCOBIOTICS have been two prospective observational studies recruiting 393 advanced inoperable NSCLC and 69 patients with renal cell cancer (RCC) in France and Canada since 2017. These cohorts of previously ICI-naïve or previously treated patients provided stool samples at baseline before ICI initiation, with detailed clinical data and comedications (Table 2). To study the prognostic impact of the gut microbiota composition on ICI responses in NSCLC and RCC, we performed shotgun metagenomics sequencing of frozen fecal samples in a first discovery cohort (enrolling subjects with NSCLC from 2017 to 2019), partially reported by Routy et al. (Routy et al. 2018; Derosa et al.2020), and in a validation cohort (enrolling patients with NSCLC from 2019 up to 2021), partially reported by Derosa et al. (Derosa et al. 2021). A third prospective cohort of 61 ICI-naïve advanced NSCLC and 14 RCC patients was recently incremented (Table 2). Altogether, these prospective observational cohorts provide the largest assessment of the potential impact of the gut microbiome as a biomarker of response to ICI to date, allowing investigation of specific MGS, co-abundance networks and functions of clinical relevance for cancer immunotherapy across two different histotypes of cancer.
43
44 Table 2. Patients' characteristics in the two retrospective and the prospective cohorts. ° P-value between Discovery and Validation cohorts. RCC: renal cell carcinoma; NSCLC: non-small cell lung cancer; BMI: body mass index; ECOG: Eastern Cooperative Oncology Group Performance Status; IMDC: International Metastatic Database Consortium Risk Model for Metastatic Renal Cell Carcinoma.
We first analyzed whether the fecal taxonomic composition at baseline in the discovery cohort composed of 245 NSCLC patients would predict overall survival beyond 12 months (OS>12) during a first-, ≥second line (≥2L) therapy with anti-PD-1 or anti-PD-L1 antibodies (Abs). To characterize differences in microbial composition between patient groups achieving OS<12 (non-responders (NR), n=112) or OS>12 (responders (R), n=118), we monitored the variations in stool microbial alpha diversity and performed principal coordinate analyses (PCoA) of microbial beta diversity distances (Bray-Curtis). Of note, 15 patients did not reach a 12 month- minimal follow up and could not be included in this analysis. Alpha diversity (Shannon index) was significantly different in the two groups (Figure 1A, upper panel). Next, we utilized an unsupervised PCoA to explore putative differences in general microbiota composition between short- and long-term survivors, finding a clear difference between the two groups (Figure 1A, middle panel, p=0.0039). To determine the relative contribution of each microbial species abundance at baseline to the observed group separation, MGS were ordered according to their variable importance (VIP) score (Figure 1A, right panel) which relies on the supervised Partial Least Squares Discriminant Analysis (PLS-DA). Some of the significant MGS that were associated with OS<12 were already described in antibiotic- treated or poor prognosis cancer patients (such as Hungatella hathewayi, Clostridium innocuum, Streptococcus anginosus or Actinomyces graevenitzii) (Derosa et al.2020; Tsay et al.2020). Next, we turned to the validation cohort of 148 patients with NSCLC in which slightly more therapy-naïve patients were enrolled than in the discovery cohort (Table 2). However, neither the alpha diversity (Figure 1B, upper panel, p=0.8389) nor the beta-diversity (Figure 1B, middle panel, p=0.1239) could segregate OS<12 or OS>12 subsets. Of note, 40 patients did not reach a 12 month-minimal follow up and could not be included in this analysis. Considering strictly MGS identified using the MetaPhlAn4 pipeline commonly found in both cohorts (discovery and validation), we only found two MGS associated with OS>12, namely Roseburia intestinalis and Anaerostipes hadrus. The Cox regression analysis of the effect of the relative abundance of either one of these two microbial hits on OS (considering the “high”≥0 and the “low”<0 of normalized and standardized relative abundance values as cut-off) confirmed that only R. intestinalis and A. hadrus significantly predicted long term clinical benefit to PD-1 blockade, in both, the discovery (Figure 2A) and the validation cohorts (Figure 1C). Despite these apparently encouraging results, the polinomial Support Vector Machine (poly-SVM) Receiver Operating Characteristic (ROC) curves (Chang et Lin 2011) measuring the performance of the relative abundance of these MGS to classify patients into OS<12 or ≥12 indicated AUC values around 0.5, suggesting that this model was not able to accurately predict patient prognosis (Figure 1D). Moreover, when considering another histotype of cancer such as a cohort of RCC patients, the relative abundance of Anaerostipes hadrus spp failed to predict long-term survivors (Figure 1E). In fact, both MGS correspond to highly prevalent (around 70%) and relatively abundant species (Fig.1B-C), suggesting that they are necessary but not sufficient to accurately predict immunosensitivity. Hence, as already discussed (McCulloch et al.2022), despite large and homogeneous cohorts handled by the same investigators using a clinically relevant endpoint (OS at 12 months), and optimized machine learning algorithms, we failed at identifying a prototypical MGS fingerprint robustly predicting clinical benefit to PD-1 blockade. The above results were actualized in 2023 after more patients were recruited in LUMIERE and ONCOBIOTICS prospective observational studies (NCT03084471), reaching 499 advanced NSCLC in France and Canada since 2017 and 83 renal cell cancer (RCC) in France. The actualized alpha diversity (Shannon index), differences in general microbiota composition between short- and long-term survivors and relative contribution of each microbial species abundance at baseline in patient subgroups [OS<12 months vs. OS>12 months] are shown in Discovery (Figure 1F) and Validation (Figure 1G) cohorts. Example 2: Building co-abundance networks within the microbial ecosystem of patients with NSCLC Resource and niche competition, as well as metabolic cross-feeding are among the main drivers of microbial community assembly (Friedman et al.2017; Clark et al.2021; Sanchez-Gorostiaga et al.2019). Nonetheless, the degree to which these forces are reflected in the composition of the intestinal communities of long-term responders (R) or non-responders (NR) has not been investigated to date. Here we used genome-scale species modeling to assess cooperation potential in large species interacting groups across thousands of MGS in the discovery cohort, which we attempted to corroborate in the validation cohort. Only MGS with a prevalence ≥2.5% were considered when generating the nodes within the final network, while a significant Pearson correlation coefficient and its related p-value (after Benjamini-Hochberg FDR at 10%) was employed to obtain categories defining edge thickness (Li et al. 2008). A leave-one-out cross-validation procedure was employed on the discovery cohort in order to have an averaged p-value for each correlation among two definite variables. This analysis revealed seven distinct communities apostrophed “SIG” (“species interacting group”) annotated with Greek letters, clustering at distant or opposite ends in a trade-off between competition and cooperation to predict OS at 12 months (Tables 3-4). Table 3. List of bacteria within each community found with Pearson network Table 4. SIG1 and SIG2 bacteria and their association with OS. a) community as defined by Blondel algorithm (Louvain method); b) community-averaged difference of the mOS (expressed in months) among the high and low normalized and standardized microbial relative abundances; c) community-averaged 2-base logarithm of the high/low ratio of the mOS among the high and low normalized and standardized microbial relative abundances; d) count and percentage of species with higher relative abundance in NR and responder (R) following OS12; e) Species Interacting Group (SIG) definition after merging alfa-beta and gamma-delta communities; f) number of species within each SIG; g) cumulative loss or gain of mOS (expressed in months) for SIG1 and SIG2; h) cumulative loss or gain of mOS ratio for SIG1 and SIG2. Driven by this observation, we employed Cox regression survival analysis and Kaplan-Meier curves (R packages survival, survminer, Rcpp), computing the difference (ΔHigh-Low mOS) and fold-ratio (log2FRHigh/Low mOS) in median OS for the “high”(≥0) and “low”(<0) normalized/standardized values of the relative abundance for each MGS contained within each of the seven SIGs, as we performed for A. hadrus and R. intestinalis (Table 4). Hence, out of 7 co-abundance networks representing the ecological community of the discovery cohort, we found similar average and fold-ratio differences among microbial communities mostly inhabited by NR species (α+β), and among two communities mostly inhabited by R species (γ+δ). Therefore, α+β and γ+δ communities were grouped in SIG1 and SIG2 respectively (Tables 3-4). Instead of Pearson-based correlations to establish the co-abundance network, we used the semi- parametric rank-based approach to correlation estimation for INference in Graphical models (SPRING) of statistical microbial association networks from quantitative microbiome data (that can naturally deal with the excess zeros in the data). We found similar SIG compositions utilizing three alternative computed networks (SPRING, SPIEC-EASI, CCREPE models) (not shown). SIG1 and SIG2 harbored a different microbial composition in thus far that 5% and 95% of SIG1 microbial species were associated with OS>12 or <12 respectively, while 97% and 3% for SIG2 were associated with OS>12 or <12 respectively (p<0.0001) (Table 5). Table 5. Percentage distribution of microbial species OS-related within SIG1 and SIG2, discovery cohort. χ2 statistics summarizing the numbers of MGS associated with responder (OS>12 months) or non-responder (NR, OS<12 months) patients in the microbial network for the discovery cohort (n=245) using Pearson matrices generated on normalized and standardized relative abundances of MGS having a prevalence ≥2.5%. Hence, SIG1 and SIG2 were composed of 40 “harmful” and 34 “beneficial” microbial species, respectively, because SIG1- or SIG2- related MGS led to a cumulative loss or gain in median OS of more than 10 months, respectively (Table 4). Indeed, SIG1 contained members belonging to the Enterocloster genus, and Streptococcaceae, Veillonellaceae and Lactobacillaceae families that were already associated with dismal prognosis or immunoresistant patient populations (Spencer et al.2021; Lee et al.2022; McCulloch et al. 2022; Tsay et al. 2020; Yonekura et al. 2021). Conversely, SIG2 assembled Lachnospiraceae (species from the genus Blautia, Roseburia, Dorea, Eubacterium), and Oscillospiraceae family members (Faecalibacterium prausnitzii, Ruminococcus bicirculans and R. lactaris), which are associated with general health and favorable clinical responses to ICI (Gopalakrishnan et al. 2018; Messaoudene et al. 2022). Next, we reduced this whole-population-based network down to a monodimensional score by computing a SIG1/SIG2 fold-ratio of normalized microbial counts in which, for a given patient, NSIG1 is the number of prevalent species belonging to SIG1 divided by 40, while NSIG2 is the number of prevalent species belonging to SIG2 divided by 34 in the MGS available for that particular patient (i.e., FRnormCOUNT = (NSIG1/40) / (NSIG2/34)). Theoretically, this value goes from zero to infinite. A Kernel Density Estimation (KDE) of the FRnormCOUNT parameter was performed for the discovery cohort in order to estimate the boundaries that better segregate NR and R distributions (p=0.00023) (Table 6, Figure 3). Three different regions resulted from the distribution of FRnormCOUNT values, identifying a SIG2 region (0< FRnormCOUNT <0.37), mostly harboring R (63% of patients with OS>12), an intermediate “Grey zone” (0.37≤ FRnormCOUNT <1.047), in which NR and R were equally represented, and a SIG1 region (FRnormCOUNT ≥1.047), mostly harboring NR (77% of patients with OS<12). The Cox regression analysis of the impact of the FRnormCOUNT on OS highlighted that patients with a FRnormCOUNT falling within the SIG2 exhibited a significantly prolonged clinical benefit to PD-1 blockade than patients falling into SIG1 or Grey zone (Figure 4A). As explained above, the FRnormCOUNT is based on the prevalence (presence or absence) of each SIG commensal, but not on the relative abundance (rel. abund.). However, this did not apply to one Gram negative anaerobic bacterium harboring regulatory and metabolic functions, Akkermansia muciniphila SGB9226 (Akk), for which a trichotomized distribution of the relative abundance best correlated with OS, as already reported (Derosa et al. 2021). Hence, to solve the uncertainty of the “Grey zone”, we segregated patients according to the trichotomized distribution of Akk relative abundance (Derosa et al. 2021; WO 2022/157207) (Figure 4B). Patients who harbored physiological Akk levels (0<Akk≤4.799, Akk L ) represented 29% of the Grey zone and were considered R, while patients devoid of (Akk 0 ) or harboring high Akk levels (Akk≥4.8, Akk H ) and constituting 53% and 17% of the Grey zone respectively were considered NR (Table 6). Finally, with both pieces of information (FRnormCOUNT+Akk level), we built a final categorical score of “immunoresistance-related dysbiosis”, named “TOPOSCORE”, to classify NSCLC patients into two risk categories, either R (predicting OS>12 months) or NR (predicting OS<12 months) (Figure 4B, Table 6). Indeed, the Cox regression analysis of the impact of the TOPOSCORE on OS highlighted that patients with a TOPOSCORE falling within the SIG2 + Grey zone Akk L (apostrophed “SIG2+” henceforth) exhibited a significantly prolonged clinical benefit to PD-1 blockade compared with patients with a TOPOSCORE falling within the SIG1 + Grey zone Akk 0/H (apostrophed “SIG1+” henceforth) in multivariate analyses (Table 6, Figure 4C-D). Of note, the TOPOSCORE classifier represented a prognosis marker independent from the ECOG performance status, LIPI score and PD-L1 expression in the discovery NSCLC cohorts. Table 6. Distribution of retrospective NSCLC cohorts within Toposcore regions. * Percentage calculated in each category; ** Comparing SIG2 and Grey Akk L vs SIG1 and Grey Akk 0 and Akk H . Applying the same network algorithm in the validation cohort of 148 NSCLC patients, we observed a similar co-abundance network, with 30% and 70% of SIG1 bacteria that were associated with OS>12 or <12 respectively, while 74% and 26% for SIG2 were associated with OS>12 or <12 respectively (p=0.0019) (Table 7). Of note, 15 (out of 40) and 23 (out of 34) MGS were shared with the discovery set for SIG1 and SIG2 respectively (Figure 5, Table 8). Table 7. Percentage distribution of microbial species OS-related within SIG1 and SIG2, validation cohort. χ2 statistics summarizing the numbers of MGS associated with responder (OS>12 months) or non-responder (NR, OS<12 months) patients in the microbial network for the validation cohort (n=148) using Pearson matrices generated on normalized and standardized relative abundances of MGS having a prevalence ≥2.5%. Table 8. Commonalities in bacteria species between Discovery and Validation cohorts. The boundaries of the KDE for the validation cohort were also able to accurately segregate NR and R within the 148 patients (Figure 6A, p=0.0486). As shown for the discovery cohort, the Cox regression analysis of the impact of the TOPOSCORE on OS validated that the “SIG2+” category of patients exhibited a significantly prolonged clinical benefit to PD-1 blockade compared with the “SIG1+” subgroup (Table 6, Figure 6B, p=0.0153). Here again, the TOPOSCORE classifier represented a prognosis marker independent from the ECOG performance status, LIPI score and PD-L1 expression in the validation NSCLC cohort (Figure 6C). Importantly, pooling all NSCLC patients with available PD-L1 expression (n=304), we could demonstrate the added value of the TOPOSCORE essentially in PD-L1 positive NSCLC tumor-bearing hosts (Figure 6D, p=0.003). In conclusion, the TOPOSCORE identified on a per capita basis an “immunoresistance-related dysbiosis” on an individual basis in about 33% of patients, the majority (two thirds) among whom were ICI resistant, and 67% cases devoid of dysbiosis, two thirds among whom were ICI responders (Table 6). The prognostic value of the TOPOSCORE was demonstrated either in treatment-naive or previously treated patients or in patients treated with ICI monotherapy (Figure 7A-B). The TOPOSCORE thus provides an individual diagnosis tool evaluating the risk of resistance to PD-1 blockade for advanced NSCLC patients. Example 3: Prospective validation of the TOPOSCORE in other cohorts of cancers amenable to PD-1 blockade We next applied the TOPOSCORE to a new prospective cohort of NSCLC (n=61) and RCC (n=83) treated with ICI (described in Table 2), for which baseline MGS and a >6 months clinical follow-up was available. The percentage of patients falling into SIG2+ and SIG1+ for this pooled cohort was 76% and 24%, respectively (Figure 8A). The Cox regression analysis of the impact of the TOPOSCORE on OS confirmed that the “SIG2+” subset of patients harbor a prolonged survival compared with the “SIG1+” subset (Table 6, Figure 8A, p=0.0595). Of note, considering only the prospective cohort of RCC, the TOPOSCORE classifier outperformed the IMDC score (Figure 8B). The Cox regression analysis on the OS of the NSCLC overall cohort (n=393 patients) confirmed that patients with a TOPOSCORE falling within the SIG2+ Grey Akk L region exhibited a significantly (p<0.0001) prolonged clinical benefit to PD-1 blockade compared with individuals harboring a TOPOSCORE within SIG1+ Grey Akk 0/H region (Figure 8C). Combining the results from all NSCLC patients (n=382 with follow-up >12 months ), we found that the sensitivity (Se), specificity (Sp), positive predictive value (PPV), and negative predictive value (NPV) of the TOPOSCORE are 76.8%, 48.0%, 62.7% and 64.7%, respectively (Table 9). Table 9. Calculation of sensitivity, specificity, positive and negative predictive values for the toposcore in NSCLC patients. Sensitivity A/A+C = 76.8%. Specificity D/D+B = 48.0%. Positive Predictive Value A/A+B = 62.7%. Negative Predictive Value D/D+C = 64.7%. To apply the TOPOSCORE to healthy individuals (HV), we computed the metagenomes from the public databases (n=5345) and utilized the MetaphLAn4 pipeline. To analyze the differences in the taxonomic stool composition between healthy subjects and advanced NSCLC patients (segregated into 242 R and 176 NR within the whole cohort for whom we had a follow up >12 months), we performed principal coordinate analyses of microbial beta diversity distances that unveiled significant distances using Bray-Curtis between HV and cancer groups (Figure 9A). To determine the relative contribution of each microbial species abundance at baseline to the observed three group separation, MGS were ordered according to their VIP score which relied on the supervised PLS-DA (Figure 9B). Not surprisingly, most significant MGS featured in Table 4 and the TOPOSCORE. In fact, the FRnormCOUNT applied to HV highlighted that 88.5% were SIG2, 10.4% fell in the Grey zone and 1.05% were SIG1. Using the Akk trichotomic distribution, we could calculate that 7.8% of HV are SIG1+ and could be considered a priori as “immunoresistant” (Figure 9C). In conclusion, the TOPOSCORE represents a robust biomarker predicting immunosensitivity and immunoresistance to ICI across two different cancer populations on an individual basis. Example 4: Machine learning prediction algorithms We moved further with a comparison of our TOPOSCORE with two machine-learning (ML) algorithms, namely Random Forest (RF) and Stable and Interpretable RUle Set (Sirus) (« A predictive index for health status using species-level gut microbiome profiling | Nature Communications » s. d.) in order to build predictive models on NSCLC discovery cohort (n=245) (not shown). We built six different Sirus models based on all SIG1 and SIG2 microbial species, topological parameters, FRnormCOUNT and its combination with Akk, finding out poor predictive power and coverage based on individual predictions on the validation cohort (n=148) (Table 10). Only the FRnormCOUNT parameter (Sirus model 4) held a better coverage (93%) with a better discrepancy among correct and uncorrect prediction percentages (ΔPC-U, 19%). Applying the RF algorithm on several meta-variables provided the highest cross- validated predictive power for TOPOSCORE (81.6 ± 6.4 %), but its validation was not meaningful, with a coverage of 53% and a ΔPC-U of around 8% (not shown). Thus, the referenced ML algorithms gave poor results in terms of individual prediction compared to TOPOSCORE, which, in turn, holds a 100% patients’ hits and the highest ΔPC-U, with a discrepancy of 26% (Table 10). Table 10. Toposcore and alternative machine-learning algorithms predicting response or resistance to ICI. a) models based on SIRUS pipeline (RF plus decision trees) as described in materials and methods; b) most predictive Random Forest (RF) model as described in materials and methods; c) models based on clinically-validated scores; d) fraction of individual hits by each model on the total OS12 data available (see Table 1); e) prediction percentage corrected for the coverage; f) difference among correct and uncorrect prediction percentages. Example 5: Functional metabolic fingerprints associated with SIGs Next, we explored putative mechanisms whereby the taxonomic fecal composition may influence ICI response, based on organism-specific gene hits annotated according to the Kyoto Encyclopedia of Genes and Genomes Orthology in each NSCLC cohort. Following these annotations, reads from each sample were reconstructed into metabolic pathways using the MetaCyc hierarchy of pathway classifications, by means of the HUMAnN 3.0 pipeline. We retrieved 493 pathways (unclassified excluded, 381 at 20% prevalence cutoff), with 372 pathways common among discovery and validation cohorts (not shown). PCoA based on Bray Curtis Dissimilarity Index showed significant compositional differences in the functional pathways across sample types among NR (OS<12 months) versus R patients (OS>12 months), in both the discovery and the validation cohorts (Figure 10A-B). Analyzing the VIP plots derived from supervised PLS-DA of retrieved pathways, six metabolic traits were found to be common among both cohorts (not shown). The PLS-DA VIP analysis indicated that patients who had OS>12 showed pathway enrichment for the biosynthesis of deoxy-thymidine diphosphate-l-rhamnose, a naturally occurring deoxy-hexose (non- digestible carbohydrate) with strong propionigenic potential, as previously shown in RCC patients treated with a probiotic C. butyricum that boosted the efficacy of ICI (Dizman et al.2022). Analyzing the pathway composition of SIGs in common among discovery and validation cohorts (Figure 10C-E), we found that SIG1 harbored 183 pathways mostly related to dismal prognosis in other studies (such as CMP-legionaminate biosynthesis reported for CAR-T cells (Smith et al. 2022), or the superpathway of thiamine diphosphate biosynthesis, L-Histidine degradation, superpathway of L-Lysine, L- Threonine and L-Methionine biosynthesis I, or pyridoxal 5'-phosphate salvage II (plants), already reported in advanced (as opposed to localized) breast cancers (Terrisse et al. 2021), while SIG2 retained the metabolic potential of mannan degradation and sulfur oxidation which remain of unknown significance between overt proinflammatory effects and elimination of SIG1-associated lactic acid bacteria (not shown). Example 6: Development of a user-friendly PCR-based TOPOSCORE assay Lastly, to transform the TOPOSCORE into a clinically actionable diagnosis tool, we contemplated i) circumventing the costly, laborious and time- consuming method of shotgun metagenomics by means of a PCR-based assay that can be performed within 48 hours for determining bacteria prevalence and ii) to restrain the numbers of SIG1- and SIG2-associated bacteria to expedite the assay. Based on the most prevalent (Figure 11) MGS species found for the gut oncoimmunological fingerprints across various studies (Park et al. 2022) as well as probing set design feasibility and specificity, we focused our PCR-based assay on 16 SIG2- and 7 SIG1- associated bacteria in addition to Akkermansia spp. (Figure 12). First, we attempted to re-run the TOPOSCORE calculation for the discovery cohort based on the MGS-based relative abundances of these 24 (instead of the full set of 75 bacteria). The Cox regression analysis of the impact of the “restricted 24-based TOPOSCORE'' on the OS of the NSCLC cohort (n=393 patients) confirmed that patients with a TOPOSCORE falling within the SIG2+ Grey Akk L region (1< FRnormCOUNT <2.065) exhibited a significantly prolonged clinical benefit to PD-1 blockade compared with individuals harboring a TOPOSCORE within SIG1+ Grey Akk 0/H region (Figure 8C-D). Secondly, Spearman correlation indexes between MGS relative abundance (and therefore prevalence) and PCR amplification for each of the 24 bacterium-specific DNA validated the strategy of switching from MGS to PCR (Figure 12, Figure 8E). Thirdly, when utilizing the 24 bacteria-specific probe sets on the remaining fecal DNA (originally extracted for MGS) from the whole NSCLC cohort (n=313), we re-demonstrated and confirmed that OS was superior in those patients harboring a PCR-based TOPOSCORE within the SIG2+ Grey Akk L region (1< FRnormCOUNT <2.065) (Figure 8E). In conclusion, we demonstrated that the development of a quick test, easily translated into clinical routine, is feasible but requires a prospective validation at this stage, on an independent cohort. Example 7: Development of a further user-friendly PCR-based TOPOSCORE assay SIG2 species were selected after a random forest classifier within Sci-Kit learn package v1.0.1, having L1 regularization, with the following parameters: clf = RandomForestClassifier (n_estimators=est, class_weight='balanced_subsample', random_state=0, oob_score=True, n_jobs=-1, max_depth=2, bootstrap=True, criterion='gini'). The 536 microbial species retrieved by the aforementioned RF model built on the TOPOSCORE within the discovery cohort (n=245) were hierarchically ordered following their descending Variable Feature Importance (VFI), and only microbial species with a VIF value higher than 1 were selected. Table 11 Microbial species selected by RF model applied on TOPOSCORE. Among the 536 microbial species 30 were selected based on their descending order of VIF. In bold the species falling within SIG1, while in italics the species falling within SIG2. Underlined are the species belonging to the selected 24-based species PCR assay. In order to ease such an assay, the selected species within SIG2 are reduced to seven: Eubacterium_rectale, Dorea_formicigenerans, Lachnospira_eligens, Faecalibacterium_prausnitzii, Parasutterella_excrementihominis, Dorea_longicatena, Eubacterium_ventriosum. A TOPOSCORE_7_7 (seven members of SIG1 and seven members of SIG2) was calculated with: As shown in table 12 below, this TOPOSCORE is still predictive. Table 12. Coverage = 1.00 DISCUSSION Cross-cohort microbiome-trained machine learning consistently predicted outcomes of PD-(L)1 therapy despite heterogeneity between cohorts across geographical distribution but failed to reproducibly identify a gut fingerprint that robustly predicts clinical outcome on an individual basis. Three meta-analyses applying uniform computational approaches across different cancer types and therapies have not explained discrepancies among published cohorts (Gharaibeh et Jobin 2019; Limeta et al. 2020; Shaikh et al. 2021). Focusing on melanoma, two recent meta-analyses exploiting MGS data bases and using machine learning methodology did not entirely converge on the final “microbiotypes” associated with responses or resistance to immunotherapy (Lee et al.2022; McCulloch et al.2022). In the present study, we readdressed this question using a different strategy. The emerging challenge of contemporary oncology is to reconstruct ecosystem networks and detect patterns of microbial species or communities leading to user-friendly diagnosis tools predicting the individual risk of immune resistance. As already discussed, baseline microbiota composition was optimally associated with clinical outcome when considering OS at 1 year after initiation of treatment (McCulloch et al.2022; Heng et al. 2009). This reconstruction, which gave close to similar co-abundance networks using PEARSON or SPRING algorithms, was based on the assumption that SIG assembled a cooperative ecosystem of functionally related bacteria/archaea located at opposite ends and matching with clinical benefit and resistance respectively. Such opposite SIGs (SIG1 and SIG2) should represent suitable surrogates of the holo-ecosystem, considering the ratio of prevalence of each SIG member rather than a single or a couple of MGS species or genera of interest for each person. This rationale better handles the inherent heterogeneity among individuals, having in mind that the prevalence of each SIG1 member is lower (around 60% of SIG1 MGS have a prevalence <15%) than that of each SIG2 member (around 60% of SIG2 MGS have a prevalence >60%), in HV and cancer patients (Figure 11). In fact, SIG1 contained 40 species belonging to the Enterocloster genus, and Streptococcacae and Veillonellaceae families as well as the Lactobacillales order (Enterococcaceae+Lactobacillaceae families) already identified in immune resistant patients (Spencer et al.2021; Lee et al.2022; McCulloch et al.2022; Tsay et al.2020; Yonekura et al.2021). Interestingly, the Gut Microbiome Health Index (GMHI), an index proposed to predict the disease likelihood of an individual based on its fecal microbial composition (Gupta et al.2020) found similar species in the non-healthy group to the ones here retrieved in our SIG1, and none of the species falling within the healthy group were retrieved in our SIG2, mainly because NSCLC patients are already diseased. Instead, SIG2 was composed of 34 species gathering Lachnospiraceae and Oscillospiraceae family members largely reported for their strong association with favorable clinical responses (Gopalakrishnan et al.2018; Spencer et al.2021; McCulloch et al.2022; Yonekura et al.2021; Frankel et al.2017; Chaput et al.2017). Computing the SIG1/SIG2 FRnormCOUNT on the 393 NSCLC patients, we found that 70% of individuals classified within SIG1+ were NR while 63% falling into SIG2+ were R. For individuals within the intermediate category (0.37< FRnormCOUNT<1.047), the relative abundance of Akkermansia spp., Verrucomicrobiaceae family member not statistically retained in the Pearson network model, efficiently complemented the FRnormCOUNT, allowing identification of NR in 59% and 84% cases of Akk 0 and Akk H (>4.8%) respectively, while classifying R in 59% cases of Akk L individuals. The robustness of the TOPOSCORE could be extended to another cohort of advanced patients amenable to ICI (including 61 prospective cases of chemotherapy naive NSCLC and 83 RCC). Extrapolating the calculations to our discovery and validation cohorts gathering 454 NSCLC patients, the TOPOSCORE Sensibility, Specificity, PPV, and NPV were 76.3%, 48%, 62.2% and 64.6%, respectively. Of note, the alternative state-of-the-art machine learning algorithms (including the SIRUS (Bénard et al.2021)) and the Random Forest model ((« Random Forests | SpringerLink » s. d.)) did not perform as well. It is interesting to outline that 7.8% of HV fell into the SIG1+ category, suggesting that the TOPOSCORE could subserve the utility to eliminate distinct fecal transplantation donors. In contrast, the 4.5% of healthy individuals with MG profile harboring a FRnormCOUNT at zero (optimal TOPOSCORE) should be preferentially selected for FMT. The Lactobacillales order was heavily represented within SIG1+, with 19/40 spp. including Enterococcaceae, Lactobacillaceae and Streptococcacae family members. Together with Veillonellaceae representatives (V. atypica, V. dispar, V. parvula, V. rogosae), they comprise many microbial components of the oral cavity, that can transit from the supraglottic compartment down to the bronchoalveolar space or the small intestine, as a result of pH fluctuations and/or co-medications (proton pump inhibitors) or dysphagia (Lee et al.2022; Tsay et al.2020; Cortellini et al.2020; Imhann et al.2016; Jackson et al.2018). Oralization of the intestinal microbiota has been linked to failure of immunotherapy and immune-related adverse events (McCulloch et al.2022). Many oral commensals suppress epithelial cell inflammatory responses by dampening PRR through Toll-like receptor (TLR) or NOD-like receptor (NLR) expression and signaling, while others suppress inflammatory responses by inhibiting NF-kB or releasing immunosuppressive anti-inflammatory cytokines, such as IL-10 (Cosseau et al. 2008; Bernardo et al.2012; Santos Rocha et al.2012). For instance, commensal Lactobacilli spp. with tryptophanase activity generates indole derivatives that can function as aryl hydrocarbon (AhR) ligands promoting the expansion of anti-inflammatory CD4+ Foxp3+ regulatory T cells (Treg) (Zelante et al.2013). Double-stranded RNA from intestinal lactic acid bacteria induces interferon-β production by dendritic cells via TLR3 activation, thereby dampening inflammation (Kawashima et al. 2013). Veillonella spp. have the capacity to prime or expand TH17 pro-angiogenic and -oncogenic lymphocytes, that contribute to dismal prognosis and resistance to cytotoxicants in NSCLC (Tsay et al. 2020). The Enterocloster gen. (E. aldensis, E. asparagiformis, E. bolteae) represents a vancomycin-sensitive clade of immunosuppressive bacteria, dominant in the intestinal microbiota of people and patients suffering from aging and chronic inflammatory disorders including cancer patients (Limeta et al.2020; Ghosh et al.2020). By causing a beta-adrenergic receptor-dependent stress ileopathy and an Enterocloster genus- dominated dysbiosis, some malignancies (and other pathological disorders such as stroke) may increase gut permeability, favoring translocation of inflammatory mediators and bacteremia with immunosuppressive potential (Stanley et al.2016; Yonekura et al. 2021). Hence, resistance to ICI appears to be driven by the over-representation of harmful bacteria more than by the absence of favorable MGS species, when considering the relative weight of SIG1+ (or Akk H ) versus SIG2+ (or Akk 0 or Akk L ) in the performance of the TOPOSCORE. This conclusion was also reached by McCulloch et al. finding that unfavorable “microbiotypes” composed of Gram negative Proteobacteria influenced the peripheral inflammatory tonus, the neutrophil-to-lymphocyte ratio and enterocyte exfoliation, paving the way to resistance to PD-1 blockade (McCulloch et al.2022). We also showed that it was possible to obtain comparable results using a PCR rather than an MGS-based TOPOSCORE, leveraging this diagnosis test within the routine tool box. In most cases, a good correlation was obtained between the two methods for the specific microorganism of interest. One exception was Faecalibacterium prausnitzii. The PCR assay used in the calculations was developed several years ago, and as more information about the heterogeneity of this species has merged, there is a clear need for further improvement of this particular PCR assay. Implementation of additional targets within SIG1 and SIG2 for probe set designing could further improve the diagnostic potential of such a PCR-based- TOPOSCORE assay. The TOPOSCORE offers the unmet medical need of patient stratification based on “gut dysbiosis”, in order to ascribe resistance to ICI to an objective deviation from the “healthy” taxonomic composition (rather than to a cell-intrinsic molecular cue) and to guide the outcome of microbiota-centered interventions, for instance following switch from SIG1+ towards SIG2+ on an individual basis, at least in advanced lung and kidney cancer patients. More data and patient incrementation in trials are needed to design a TOPOSCORE in other malignancies (such as melanoma). Example 8: A SIG2+ gut microbiota signature at baseline is associated with a better response to CAR T cell therapy We prospectively and longitudinally collected fecal material, from patients receiving commercial CD19 CAR-T cells, at different time-points (Oncobiotics (Discovery of Microbiome-based Biomarkers for Patients With Cancer Using Metagenomic Approach); Sponsors: Gustave Roussy, Cancer Campus, Grand Paris ; Sponsor Protocol N: CSET 2017/2619, ID-RCB N: 2017-A02010-53)): at baseline before lymphodepletive chemotherapy, between 7 and 15 days after CAR-T cell infusion, and 3 months after CAR-T cell infusion. Shotgun metagenomic analyses were performed on the patient’s fecal samples, aiming at correlating the composition of the gut microbiota with response to CAR-T cells therapy. Patients characteristics are described in Table 13. As expected, most of the patients had diffuse large B cell lymphoma and received axi-cel (CD28 co-stimulatory domain). The overall response rate observed (48.7%) was concordant with the ones observed in clinical trials. Table 13: Patients characteristics Metagenomic sequencing were performed for 41 patients (data are still being collected) and analyses were obtained so far for 22 patients. Strikingly, we observed an absence of Akkermansia muciniphila in the fecal material from most of our B cell lymphoma cohort (90.9 %). The patient’s TOPOSCORE was monitored using the metagenomic- based TOPOSCORE assay described in Example 2. As shown in Figure 13, the TOPOSCORE at baseline predicts overall response rate (ORR) and progression-free survival (PFS) in this cohort (p=0.0509). Example 9: Co-abundance networks within the microbial ecosystem of NSCLC patients and novel TOPOSCORE calculation Here we used prevalence and/or relative abundances of metagenomics species (MGS) to assess their cooperative potential within large species interacting groups (SIG) and the clinical relevance of SIGs for the response to PD1 blockade in the discovery cohort. Building up intestinal communities (species interacting groups) Each MGS was categorized as either “low” or “high” based on the median of its relative abundance in the whole population of 245 subjects (≤ median) or > median respectively). For those MGS which had a majority of null abundances (i.e., median = 0), the MGS were categorized as “present” or “absent” (relative abundance > 0 or = 0 respectively). Cox Proportional Hazard (CoxPH) models were run to select MGS associated with the clinical variable “Overall Survival” with a Hazard Ratio (HR) ≤0.80 or ≥1.25 respectively. The purpose of this selection was to discard MGS with HR close to 1, which are unlikely to participate in the robustness of the signature. Among the 536 MGS identified by the shot gun MG of the discovery cohort, a total of 266 MGS was retained in the model (Table 14). The Akkermansia muciniphila MGS was not considered in this screening because its distribution was trichotomic with no linear dose-effect relationship with patient prognosis (Lisa Derosa et al.2022b). Each pair of these 266 MGS was then analyzed by a Fisher’s exact test on 2x2 contingency tables based on their absence/presence co-occurrences and scored by the by -log10(p) x sign(OR – 1) metrics, where p is the Fisher p-value and OR, the Odds Ratio of the 2x2 table. This metrics defined a score proportional to the significance of the interaction between two MGS (-log10(p)) that is negative in case of co-exclusion pattern (OR < 1) or positive in case of co-occurrence (OR > 1). Interactions with a Bonferroni-corrected p-value ≤ 0.05 were retained for analysis. A total of 180 connected MGS were then clustered with Ward’s method and Manhattan distance resulting in the identification of 7 clusters (C1 to C7) (Table 14). Two clusters (C5 and C6) contained 37 MGS mostly (95%) associated with OS<12 (HR ≥ 1.25) that were used to define the SIG1 signature. Three clusters (C1, C2, C3) contained 45 MGS all associated with OS>12 months (HR ≤ 0.80) that were used to define the SIG2 signature (Table 14). All the other clusters failed to correlate with OS. In addition, interactions within SIG1 and SIG2 MGS were 99% and 100% positive respectively (co-occurrence patterns), while edges in-between SIG1 and SIG2 MGS were 98% negative (co-exclusion patterns), thus reflecting a significant and opposite topological separation (data not shown). These results are supported by the fact that SIG1 contained members belonging to the Enterocloster genus, and Streptococcaceae, Veillonellaceae and Lactobacillaceae families that were already associated with dismal prognosis and immunoresistance (Yonekura et al.2021; Spencer et al.2021; Lee et al.2022; McCulloch et al.2022; Tsay et al.2020). Conversely, SIG2 contained Lachnospiraceae (species from the genus Blautia, Roseburia, Dorea, Eubacterium), and Oscillospiraceae family members (Faecalibacterium prausnitzii, Ruminococcus bicirculans and R. lactaris), which were found associated with general health and favorable clinical responses to ICI (Gopalakrishnan et al.2018; Messaoudene et al.2022). Table 14: List of the 266 MGS retained in the CoxPH model, of the 180 selected in each of the seven clusters and of 82 selected in SIG 1 or 2 Scoring system for each individual Next, we reduced the information of this whole-population-based network down to a unidimensional score. Each patient of the discovery cohort was then scored with a S score computed as the difference of proportions between present (relative abundance > 0) SIG2 and SIG1 MGS and scaled from 0 to 1: S = (#SIG2/45 - #SIG1/37 + 1)/2. A score of 0 indicates that all MGS of the SIG1 signature have strictly positive relative abundances and all MGS of the SIG2 signature have null relative abundances. Conversely, a score of 1 indicates that all MGS of the SIG1 and SIG2 signature have null and strictly positive relative abundances respectively. The performance of this S score as predictor of OS12 was analyzed by a Receiver Operating Characteristic (ROC) analysis. Two S scores, 0.5351 and 0.7911, were identified as local maxima of the Youden index (Specificity + Specificity – 1, Figure 14), and were used as cut-off values to define three categories: SIG1 if S ≤ 0.5351, SIG2 if S ≥ 0.7911, and “gray zone” otherwise (Figure 15A). Hence, 69% and 23% of patients falling into S scoring ≤ 0.5351 and S ≥ 0.7911 presented an OS<12 months respectively (Figure 15A, Table 15). By extension, the Cox regression analysis of the clinical impact of the S score on OS highlighted that patients with a S score ≥ 0.7911 (called “SIG2” henceforth) exhibited a significantly prolonged clinical benefit to PD-1 blockade than patients falling into the Gray zone or a S score ≤ 0.5351 (called “SIG1” henceforth) (Figure 15B). About 22.5%, 31%, and 46.5% of patients in the discovery cohort fell into SIG2, SIG1 and Gray zone category respectively. Next, we analyzed the intraindividual dynamics of the S score in 32 NSCLC patients who were sampled twice, before and within 3 months after treatment start. Interestingly, 33% and 25% of SIG2 and SIG1 joined the Gray zone respectively while half of patients classified in the Gray score shifted to SIG2 and no patients changed from Gray to SIG1 (Figure 15C). Altogether, 67%, 50% and 70% of individuals within SIG2, Gray or SIG1 remained in the same category respectively (Figure 15C). Responders at 3 months (n=12) resulted from SIG2 (n=8) and Grey (n=4) subject categories. Non responders (n=20) resulted from Grey (n=6) and SIG1 subjects (n=14) subject categories. Refining the predictive model To solve the uncertainty of the “Gray zone” which represented about half of NSCLC patients, we segregated individuals according to the trichotomized distribution of Akkermansia muciniphila (Akk) relative abundance (Figure 15D) (Lisa Derosa et al. 2022a). Normal levels of Akk (0<Akk≤4.799, Akk norm ) may be considered as a surrogate of host fitness in comparison with abnormal levels (Akk≥4.8, Akk high ) or the absence of Akk (Akk 0 )(Lisa Derosa et al. 2022a). Starting from here, Gray zone patients who harbored physiological Akk levels (about 19% of the whole cohort) were considered comparable to SIG2 subject category, while Gray zone patients devoid of (Akk 0 ) or harboring high Akk levels and constituting about 23% and 4% of the whole cohort, respectively, were considered comparable to SIG1 subject category (Table 1). Ultimately, according to S scoring and Akk level, we built a final binomial categorical score of “immunoresistance-related” dysbiosis, named “TOPOSCORE”, to classify NSCLC patients into two risk categories, either SIG2+ (comprising SIG2 [TOPOSCORE = 1] + Gray zone Akk norm [TOPOSCORE = 2]) or SIG1+ (encompassing SIG1 [TOPOSCORE = 5] + Gray zone Akk high [TOPOSCORE = 4] + Gray zone Akk 0 [TOPOSCORE = 3]) individuals (Figure 15D, Table 15). Indeed, the Cox regression analysis of the clinical impact of the TOPOSCORE on progression-free and overall survivals highlighted that patients with a TOPOSCORE falling within “SIG2+” exhibited a significantly prolonged clinical benefit to PD-1 blockade (Table 15, Figure 15E, left and right panels) compared with patients with a TOPOSCORE falling within “SIG1+”. Moreover, even after adjusting for renown risk factors (age, gender, body mass index, antibiotic use, PD-L1 expression, line of treatment and ECOG performance status), the TOPOSCORE had an independent relationship with overall survival in multivariate analyses (Table 16, HR= 0.47 (0.33-0.67), p=0.001). Finally, the intraindividual dynamics of the TOPOSCORE in the same 96 NSCLC individuals sampled twice (baseline and 3 months) described above showed the relative stability of the SIG phenotype with 67% and 74% patients remaining in their SIG2+ and SIG1+ category respectively (Figure 16C). Scoring validation in lung cancer We next applied the TOPOSCORE to a NSCLC validation cohort of 254 patients. The proportions of patients falling into SIG1, Gray Akk high , Akk 0 , Akk norm , SIG2 were approximately similar to those described in the discovery cohort with 29%, 7%, 16%, 21%, 27% respectively (Table 15). Here, 44.2% and 22% of patients falling into S scoring ≤ 0.5351 (SIG1) and S ≥ 0.7911 (SIG2) presented an OS<12 months respectively (Figure 16A, Table 15). 73 Table 15. Distribution of NSCLC and GU patients within the TOPOSCORE categorization. * Percentage calculated in each category considering patients with follow-up > 12 months ** Comparing SIG2 and Grey Akk L vs SIG1 and Grey Akk 0 and Akk H
Table 16. Multivariate analyses of the TOPOSCORE in discovery NSCLC patients - Cox proportional-hazards univariate and multivariate analyses for DISCOVERY cohort ¹Total patients included=214; missing data: 31; total events: 156
Table 17. Multivariate analyses of the TOPOSCORE in validation NSCLC patients - Cox proportional-hazards univariate and multivariate analyses for VALIDATION cohort ¹Total patients included=193; missing data: 61; total events: 72
As shown for the discovery cohort, the Cox regression analysis of the association of the TOPOSCORE with PFS and OS validated that the “SIG2+” (TOPOSCORE= 1 or 2) category of patients exhibited a significantly prolonged clinical benefit to PD-1 blockade compared with the “SIG1+” (TOPOSCORE= 3 to 5) subgroup (Table 15, Figure 16B, p =0.058 for PFS, p=0.0034 for OS). Here again, the TOPOSCORE classifier represented an independent and more robust prognosis marker than PD-L1, age and antibiotics uptake in multivariate analyses (Table 17, HR= 0.59 (0.36-0.97), p=0.041). Importantly, pooling all NSCLC patients from the discovery and validation cohorts with an available PD-L1 immunohistochemical tumor labeling (n=344), we could demonstrate the added value of the TOPOSCORE not only in PD-L1 negative tumors but also in PD-L1 positive NSCLC patients (Figure 16D, p=0.046 and p=0.032 respectively). The prognostic value of the TOPOSCORE was demonstrated not only in previously treated- patients but also in treatment-naive individuals, being in anti-PD-1 Ab monotherapy in PD-L1>50% cases or in chemo-immunotherapy (Figure 16E-F). Example 10: Prospective validation of the TOPOSCORE in other cancer cohorts amenable to PD-1 blockade We next extended the use of the lung cancer-related TOPOSCORE to a new prospective cohort pooling 83 RCC (from ONCOBIOTICS) and 133 unorthelial cancer (UC) (from IOPREDI study) treated with anti-PD(L)1 antibodies in 2 nd L therapy, for which baseline samples and >6 months-clinical follow-up were available. The percentage of patients falling into SIG1+ for RCC and UC cohorts were 35% and 57%, respectively (Figure 17A). Pooling all urinary tract malignancies, we found that the proportions of patients falling into SIG1, Gray Akk high , Akk 0 , Akk norm , SIG2 were approximately similar to those described in NSCLC with 26.4%, 1.4%, 21%, 24%, 27.3% respectively (Table 15). As found in NSCLC, 80% and 36% of RCC+UC patients within SIG1 and SIG2 had an OS<12 months, respectively (Table 15). The Cox regression analysis of the impact of the TOPOSCORE on PFS and OS confirmed that the “SIG1+” subset of patients harbors a reduced clinical benefit compared with the “SIG2+” subset (Table 15, Figure 17B, left and right panels, p=0.0039 for PFS and p<0.0001 for OS). Finally, we applied the TOPOSCORE to healthy individuals (HV) instead of cancer patients, computing the metagenomes of public databases (n=5345) and utilizing the MetaPhlAn 4.0 pipeline. To analyze the differences in the taxonomic stool composition between HV and the advanced NSCLC patients (segregated into OS> or <12 months) described above, we performed PCoA of Bray-Curtis distances on batch- corrected with MMUPHin (Ma et al. 2022) and normalized/standardized data that unveiled significant separation among HV and cancer groups. To determine the relative contribution of each MGS abundance at baseline to the observed three group separation, MGS were ordered according to their VIP score which relied on the supervised PLS-DA (Figure 18A). Not surprisingly, 9 out of the 37 significant MGS were listed in SIG1 (such as Enterocloster clostridioformis, E. bolteae, Clostridium symbiosum…) or SIG2 (Coprococcus comes, Dorea longicatena…). In fact, the S score applied to HV highlighted that 68.5% were SIG2, 27.5% fell in the Gray zone and 4% were SIG1. Using the Akk trichotomic distribution, we could calculate that about 20% of HV are SIG1+ and could be considered a priori as inappropriate donors of fecal microbial transplantation (Figure 17A, Figure 18B). Altogether, the TOPOSCORE allowed to conclude that 53%, 58%, 35% and 57% of 1 st L NSCLC, 2 nd L NSCLC, 1 st +2 nd L RCC, >2 nd L UC patients harbor a gut dysbiosis (defined by the percentage of SIG1+ individuals) in our cohorts (Figure 17A) that was associated with immuno- resistance independently of other prognosis factors (Tables 16-17). In conclusion, the TOPOSCORE represents a robust biomarker predicting immune-sensitivity (and immune-resistance thereof) to ICI across lung and genitourinary tract cancer populations and a useful tool to follow the dynamics of gut dysbiosis on an individual basis. Example 11: Challenging the TOPOSCORE of Example 9 with machine learning approaches The Sensitivity, Specificity, Positive and Negative Predictive Values of the TOPOSCORE were calculated in the discovery cohort of NSCLC as 74.1%, 56.8%, 69.8% and 61.9%, respectively, with an AUC=0.66 [95% confidence interval 0.59 - 0.73]. This performance of the TOPOSCORE was compared with that of two machine-learning algorithms. First, Random Forest (RF) applied on relative abundances of all microbial species with SIAMCAT provided an AUC of 0.651±0.012 in the discovery cohort (Figure 19A). Then, in an effort to evade the usual taxonomy-based RF modeling, we employed an innovative Metagenome Assembled Genomes (MAGs)-based RF model, which was built on relative abundances of 284 high-quality MAGs identified based on their capacity for keeping their interactions stable despite dramatic environmental perturbations. These 284 MAGs are organized as two competing guilds which support RF models for discriminating cases from controls across various disease states including insulin resistance and colon cancer (Wu et al. 2022). These 284 MAGs were used to construct a RF model for predicting personalized immunoresistance with an AUC of 0.69 (95% CI 0.62-0.76) in this discovery cohort (Figure 19B). Thus, the referenced machine-learning algorithms gave similar results in terms of individual prediction in the discovery cohort compared to the TOPOSCORE. Example 12: Functional pathways associated with SIG1 and SIG2 MGS To explore putative microbial functions underlying SIG1 and SIG2 compositions, we employed an analysis of MG pathways by means of HUMAnN 3.0 pipeline. This pipeline first annotates microbial-specific gene hits according to the Kyoto Encyclopedia of Genes and Genomes Orthology, then reconstructs microbial metabolic pathways using the MetaCyc hierarchy. We thus retrieved 664 pathways (unclassified excluded, and 441 at 20% prevalence cutoff) in the whole cohort of NSCLC 499 patients, with 11 and 57 pathways exclusively present in SIG1 and SIG2 microbial communities, respectively, and 76 shared pathways (for a total of 144 pathways) (Figure 19C, Table 18). PLS-DA ordination plot showed significant compositional differences in the functional pathways across sample types among SIG1+ and SIG2+ patients, while the VIP plot showed discriminant and significant pathways for each cohort (Figure 19D-E). SIG2 metabolic functions encompass sulfur oxidation, tRNA charging and processing, stachyose and mannan degradation, L-glutamate and L glutamine as well as L-Arginine and L-ornithine biosynthesis, L-tryptophane and dTDP-L-rhamnose_biosynthesis, as well as the pentose phosphate pathways. In contrast, SIG1 metabolic functions gather fatty acid betaoxidation, 5’deoxyadenosine and L phenylalanine degradation, purine and L-histidine degradation and guanosine and L-lysine biosynthesis. A hierarchical clustering based on the overall 144 pathway abundances related to SIG1 and SIG2 MGS, was applied to the 499 NSCLC patients, showing a clear separation of patients into two distinct groups, cluster 1 harboring 73% of SIG1+ individuals enriched in SIG1 functional pathways and cluster 2 harboring 70% of SIG2+ patients enriched in SIG2 functional pathways (χ 2 statistic with Yates correction = 88.305, p< 0.00001). Thus, even if 76 pathways are in common among SIG1 and SIG2 groups of microbial species, the overall 144 pathway distribution mirrored the seesaw balance between these interactive groups, dividing them into two distinct functional patterns related to SIG1 and SIG2 MGS functions (Fisher exact test < 0.00001). Table 18. List of pathways distinctive for SIG1 and SIG2 Example 13: Development of a new user-friendly qPCR-based TOPOSCORE assay Finally, to transform the TOPOSCORE into a clinically actionable diagnosis tool, we contemplated to circumvent the costly, laborious, and time-consuming method of shotgun metagenomics by means of a qPCR-based assay that can be performed within 48 hours for determining bacteria prevalence. Based on the most prevalent MGS species found for the gut oncomicrobial signatures across various cohorts (Park et al.2022; Thomas et al.2023) and the feasibility of designing bacteria- specific and reliable probe sets (Figure 20), we focused our qPCR-based assay on 15 SIG2- and 5 SIG1-associated bacteria in addition to Akkermansia spp. (Figure 21). First, we attempted to re-run the TOPOSCORE calculation for the whole NSCLC cohort based on the MGS-based relative abundances of these 21 microbial species (instead of the full set of 83 bacteria). In Figure 17C left panel, we show the survival curve of the 393 NSCLC patients confirming that an 83 MGS-based TOPOSCORE can discriminate between favorable versus dismal prognosis. A comparable performance was obtained using the 21 MGS-based TOPOSCORE determined by shotgun MG analyses (Figure 17C, right panel). Importantly, Spearman correlation indices between MGS relative abundance (and therefore prevalence) and PCR quantification of each of the 21 bacterium-specific DNA validated the strategy of switching from shotgun MG to qPCR (two examples are shown for two bacteria (Figure 17D), refer to Figure 21 for the other 19 bacteria). Lastly, when utilizing the 21 bacteria-specific probe sets on the remaining fecal DNA (originally extracted for shotgun MG) from the whole NSCLC cohort (n=313), we confirmed that OS was superior in those patients harboring a qPCR-based TOPOSCORE falling within the SIG2+category (Figure 17E, p=0.0015). We validated the robustness of the qPCR- based TOPOSCORE to predict OS in a prospective cohort of 96 NSCLC patients (Figure 17F). In conclusion, we demonstrated that a quick test, easily translatable into clinical routine, is feasible and reliable to predict survival during immunotherapy of lung cancer. DISCUSSION about Examples 9-13 Despite the use of microbiome-trained machine learning across different geographical cohorts, consistent prediction of PD-(L)1 therapy outcomes remains elusive. Notably, there has not been a reproducible microbiome signature to reliably predict individual clinical outcomes. This observation is consistent with three meta- analyses spanning various cancer types and therapies that failed to resolve discrepancies in existing cohorts (Gharaibeh and Jobin 2019; Limeta et al.2020; Shaikh et al. 2021). Recently, two meta-analyses focused on melanoma utilized shotgun MG databases and machine learning methodology to offer partial clarity on the “microbiotypes” associated with responses or resistance to immunotherapy (Lee et al. 2022; McCulloch et al.2022). In the present study, we pivot to an ecosystem-based strategy. Hence, our work suggests gut residence of cooperative ecosystems yielding consistent co- abundance patterns harboring opposite clinical relevance (sensitivity versus resistance to ICI) in a seesaw manner. Computing the TOPOSCORE on 715 advanced cancer patients, we found that around 50% of individuals could be classified within SIG1+ among whom about 63% had an OS<12 months. The prevalence of each SIG1 member is lower than that of each SIG2 member. Around 50% of SIG1 MGS have a prevalence <15% while about 55% of SIG2 MGS have a prevalence >50% in HV and cancer patients (Figure 20). SIG1 contained 37 species belonging to the Enterocloster genus, Streptococcaceae, Veillonellaceae and Lactobacillaceae families, already identified in immune resistant patients (Gilbert et al. 2018; L. Derosa et al.2018; Mager et al.2020; Overacre-Delgoffe et al.2021). In fact, SIG1 comprises many microbial components of the oral cavity, that can transit from the supraglottic compartment down to the bronchoalveolar space or the small intestine, as a result of pH fluctuations and/or co-medications (proton pump inhibitors) or dysphagia (Lee et al.2022; Tsay et al.2020; Cortellini et al.2020; Imhann et al.2016; Jackson et al. 2018). Oralization of the intestinal microbiota has been linked to failure of immunotherapy and immune-related adverse events (McCulloch et al.2022), (Cosseau et al.2008; Bernardo et al.2012; Santos Rocha et al.2012). Veillonella spp. have the capacity to expand TH17 pro-angiogenic and -oncogenic lymphocytes that contribute to dismal prognosis and resistance to cytotoxicants in NSCLC (Tsay et al. 2020). The Enterocloster genus (E. aldensis, E. asparagiformis, E. bolteae) represents a vancomycin-sensitive clade of immunosuppressive bacteria, dominant in the intestinal microbiota of people and patients suffering from aging and chronic inflammatory disorders including cancer (Limeta et al.2020; Ghosh et al.2020). By causing a beta- adrenergic receptor-dependent stress ileopathy and an Enterocloster genus-dominated dysbiosis, some malignancies (and other pathological disorders such as stroke) may increase gut permeability, favoring translocation of inflammatory mediators and bacteremia with immunosuppressive potential (Yonekura et al. 2021; Stanley et al. 2016). Interestingly, the Gut Microbiome Health Index (GMHI), an index proposed to predict the disease likelihood of an individual based on its fecal microbial composition (Gupta et al.2020) found species in the non-healthy group similar to the ones retrieved here in SIG1, but none of the species falling within the healthy group were retrieved in our SIG2, mainly because NSCLC patients are already diseased. Instead, SIG2 was composed of 45 species gathering Lachnospiraceae and Oscillospiraceae family members largely reported for their strong association with healthy status and favorable clinical responses (Yonekura et al. 2021; Gopalakrishnan et al. 2018; Spencer et al. 2021; McCulloch et al. 2022; Frankel et al. 2017; Chaput et al. 2017). Indeed, SIG2 functional fingerprints (stachyose degradation (Spencer et al.2021), biosynthesis of L- ornithine and L-Arginine of polyamine pathway (Danlos et al.2021; Peyraud et al.2022; Geiger et al.2016; Canale et al.2021), purine ribonucleoside degradation (Teng et al. 2023) plead in favor of metabolic patterns that could keep in check tumor cell cycle and reactivate immunosurveillance. Sensitivity, specificity, positive and negative predictive values of the TOPOSCORE were 74.1%, 56.8%, 69.8% and 61.9%, respectively, with an AUC=0.66 [95% confidence interval 0.59 - 0.73]. Of note, the alternative state-of-the-art machine learning algorithms (including SIRUS, SIAMCAT and MAGs-based RF) performed equally well (Figure 19). The robustness of the TOPOSCORE was not only illustrated in NSCLC but also in stage IV urinary tract bearing cancer patients amenable to ICI (including 83 (1L+2L) RCC, and 133 (>2L) UC. Interestingly, the fraction of SIG1+ patients was lower in RCC (34%) than in UC (57%), suggesting that platinum salt-based chemotherapy might be less efficient than tyrosine kinase inhibitors to prevent gut dysbiosis. Indeed, the longitudinal scoring of patients will be instrumental to understand the impact of each therapy on the gut homeostasis. Likewise, the TOPOSCORE can also represent a valuable tool to select donors of fecal microbial transplantation (FMT). It is interesting to outline that about 21% of HV fell into the SIG1+ category, suggesting that the TOPOSCORE could help dismiss donor candidates of FMT in favor of the 26% fraction (1399/5345) that resides in the top 10% of the TOPOSCORE (0.90). In contrast, only 6% cancer patients (43/715) scored 0.90 and could theoretically be preferentially selected for FMT. The TOPOSCORE also covers the unmet medical need of patient stratification based on “gut dysbiosis” in order to ascribe resistance to ICI to an objective deviation from the “healthy” taxonomic composition (rather than to a cell- intrinsic molecular cue), and to guide the outcome of microbiota-centered interventions. Hence, the TOPOSCORE represents an actionable diagnosis tool for the pharmacodynamics of live biotherapeutics, FMT and prebiotics. More specifically, the TOPOSCORE offers a friendly user process to quickly assess gut dysbiosis in a given individual at any time of the disease. Indeed, we showed that it was possible to obtain comparable results using a 21 bacteria-probe set-based qPCR rather than a 83 MGS- based shotgun MG TOPOSCORE, leveraging this diagnosis test within the routine tool- box. Admittedly, incrementation of additional MGS into the qPCR-based TOPOSCORE may improve its performance. The TOPOSCORE may fluctuate with patient accrual, disease selection and geography. Of note, the TOPOSCORE was computed based on 12 months-overall survival, suggesting that it may not be helpful to predict response rates at the first CT scan. Despite these limitations, our work offers a new method of dimension reduction of clinical relevance to assess gut dysbiosis in cancer patients amenable to immunotherapy. Example 14: Use of the TOPOSCORE measured as described in Example 13 in colorectal cancer The TOPOSCORE of patients enrolled in ATEZOTRIBE clinical trial was assessed using the 21 MGS-based PCR asay of Example 13. ATEZOTRIBE is a randomised clinical trial with 150 colorectal cancers in two arms, with or without anti-PDL-1 Ab (atezolizumab). As shown in Figure 22 (only MSS patients shown) the TOPOSCORE predicts overall survival only in the immunotherapy (anti-PDL-1 Ab) arm (not the chemotherapy arm). Example 15: Further examples of qPCR-based TOPOSCORE assays Ten combinations of 50 bacterial species, each differring by the one species, were used to calculate the TOPOSCORE of NSCLC patients (N=20). The 49 bacterial species common to all of these combinations are Ruminococcus bicirculans, Faecalibacterium prausnitzii, Blautia wexlerae, Roseburia intestinalis, Gemmiger formicilis, Anaerostipes hadrus, Streptococcus parasanguinis, Clostridiales bacterium KLE1615, Agathobaculum butyriciproducens, Dorea longicatena, Clostridium symbiosum, Blautia massiliensis, Eubacterium rectale, Faecalibacterium SGB15346, Clostridium sp AF34 10BH, Lachnospira eligens, Lachnospiraceae bacterium WCA3601 WT 6H, Streptococcus salivarius, Clostridium fessum, Anaerobutyricum hallii, Hungatella hathewayi, Candidatus Cibiobacter qucibialis, Anaerotignum faecicola, Clostridium scindens, Clostridium innocuum, Clostridiaceae unclassified SGB4769, Roseburia hominis, Clostridiaceae bacterium, Oscillibacter sp ER4, Clostridiaceae bacterium OM08 6BH, Roseburia inulinivorans, Phocaeicola massiliensis, Enterocloster aldensis, Veillonella parvula, Lacrimispora amygdalina, Firmicutes bacterium AF16 15, Coprococcus eutactus, Eubacterium ventriosum, Enterocloster bolteae, Clostridiales unclassified SGB15145, Faecalibacillus intestinalis, Coprococcus comes, Roseburia sp AF02 12, Erysipelatoclostridium ramosum, Clostridium sp AM49 4BH, Mediterraneibacter butyricigenes, Dorea formicigenerans, Coprobacter fastidiosus ad Enterocloster clostridioformis. In each assay, one additional bacterial species was used in addition to the above 49 species. These variable species were: J1: Ruminococcus lactaris, J2: Bifidobacterium dentium, J3: Lachnospira sp NSJ 43, J4: Clostridium sp AM2211AC, J5: Lachnospira pectinoschiza, J6: Lachnospiraceae bacterium OM0412BH, J7: Clostridium sp AM333, J8: Veillonella dispar, J9: Eubacterium ramulus and J10: Actinomyces graevenitzii. Akkermansia was also used to refine the TOPOSCORE in the grey zone. The AUC obtained with these combinations of bacterial species are shown in Figure 23. REFERENCES Bénard, Clément, Gérard Biau, Sébastien Da Veiga, and Erwan Scornet. 2021. « SIRUS: Stable and Interpretable RUle Set for classification ». Electronic Journal of Statistics 15 (1): 427-505. Bernardo, David, Borja Sánchez, Hafid O. Al-Hassi, Elizabeth R. Mann, María C. Urdaci, Stella C. Knight, and Abelardo Margolles. 2012. « Microbiota/Host Crosstalk Biomarkers: Regulatory Response of Human Intestinal Dendritic Cells Exposed to Lactobacillus Extracellular Encrypted Peptide ». PLOS ONE 7 (5): e36262. Carbonero, Franck, Ann Benefiel, Amir Alizadeh-Ghamsari, and H. Rex Gaskins. 2012. « Microbial pathways in colonic sulfur metabolism and links with health and disease ». Frontiers in Physiology 3. Chang, Chih-Chung, and Chih-Jen Lin. 2011. « LIBSVM: A library for support vector machines ». ACM Transactions on Intelligent Systems and Technology 2 (3), 27, 1- 27. Chaput, N., P. Lepage, C. Coutzac, E. Soularue, K. Le Roux, C. Monot, L. Boselli, et al.2017. « Baseline Gut Microbiota Predicts Clinical Response and Colitis in Metastatic Melanoma Patients Treated with Ipilimumab ». Annals of Oncology: Official Journal of the European Society for Medical Oncology 28 (6): 1368-79. Cho, Ilseung, and Martin J. Blaser. 2012. « The Human Microbiome: At the Interface of Health and Disease ». Nature Reviews. Genetics 13 (4): 260-70. Clark, Ryan L., Bryce M. Connors, David M. Stevenson, Susan E. Hromada, Joshua J. Hamilton, Daniel Amador-Noguez, and Ophelia S. Venturelli. 2021. « Design of Synthetic Human Gut Microbiome Assembly and Butyrate Production ». Nature Communications 12 (1): 3254. Cortellini, Alessio, Marco Tucci, Vincenzo Adamo, Luigia Stefania Stucci, Alessandro Russo, Enrica Teresa Tanda, Francesco Spagnolo, et al. 2020. « Integrated Analysis of Concomitant Medications and Oncological Outcomes from PD-1/PD-L1 Checkpoint Inhibitors in Clinical Practice ». Journal for Immunotherapy of Cancer 8 (2): e001361. Cosseau, Celine, Deirdre A. Devine, Edie Dullaghan, Jennifer L. Gardy, Avinash Chikatamarla, Shaan Gellatly, Lorraine L. Yu, et al. 2008. « The Commensal Streptococcus Salivarius K12 Downregulates the Innate Immune Responses of Human Epithelial Cells and Promotes Host-Microbe Homeostasis ». Infection and Immunity 76 (9): 4163-75. Davar, Diwakar, Amiran K. Dzutsev, John A. McCulloch, Richard R. Rodrigues, Joe-Marc Chauvin, Robert M. Morrison, Richelle N. Deblasio, et al.2021. « Fecal Microbiota Transplant Overcomes Resistance to Anti–PD-1 Therapy in Melanoma Patients ». Science 371 (6529): 595-602. Derosa, L., M. D. Hellmann, M. Spaziano, D. Halpenny, M. Fidelle, H. Rizvi, N. Long, et al. 2018. « Negative Association of Antibiotics on Clinical Activity of Immune Checkpoint Inhibitors in Patients with Advanced Renal Cell and Non-Small-Cell Lung Cancer ». Annals of Oncology: Official Journal of the European Society for Medical Oncology 29 (6): 1437-44. Derosa, Lisa, Bertrand Routy, Marine Fidelle, Valerio Iebba, Laurie Alla, Edoardo Pasolli, Nicola Segata, et al.2020. « Gut Bacteria Composition Drives Primary Resistance to Cancer Immunotherapy in Renal Cell Carcinoma Patients ». European Urology 78 (2): 195-206. Derosa, Lisa, Bertrand Routy, Guido Kroemer, and Laurence Zitvogel.2018. « The Intestinal Microbiota Determines the Clinical Efficacy of Immune Checkpoint Blockers Targeting PD-1/PD-L1 ». Oncoimmunology 7 (6): e1434468. Derosa, Lisa, Bertrand Routy, Laurence Zitvogel, Andrew M. Thomas, Gerard Zalcman, Sylvie Friard, Julien Mazieres, et al. 2021. « Intestinal Akkermansia muciniphila predicts overall survival in advanced non-small cell lung cancer patients treated with anti-PD-1 antibodies: Results a phase II study. » Journal of Clinical Oncology, Volume 39, Issue 15 suppl, 9019-9019. Derosa, Lisa, Bertrand Routy, Andrew Maltez Thomas, Valerio Iebba, Gerard Zalcman, Sylvie Friard, Julien Mazieres, et al. 2022a. “Intestinal Akkermansia Muciniphila Predicts Clinical Response to PD-1 Blockade in Patients with Advanced Non-Small- Cell Lung Cancer.” Nature Medicine 28 (2): 315–24. Derosa, Lisa et al.2022b. “Intestinal Akkermansia Muciniphila Predicts Clinical Response to PD-1 Blockade in Patients with Advanced Non-Small-Cell Lung Cancer.” Nature Medicine, February. Dizman, Nazli, Luis Meza, Paulo Bergerot, Marice Alcantara, Tanya Dorff, Yung Lyou, Paul Frankel, et al. 2022. « Nivolumab plus Ipilimumab with or without Live Bacterial Supplementation in Metastatic Renal Cell Carcinoma: A Randomized Phase 1 Trial ». Nature Medicine 28 (4): 704-12. Dordević, Dani, Simona Jančíková, Monika Vítězová, and Ivan Kushkevych. 2021. « Hydrogen Sulfide Toxicity in the Gut Environment: Meta-Analysis of Sulfate- Reducing and Lactic Acid Bacteria in Inflammatory Processes ». Journal of Advanced Research, Chemistry, Biology and Clinical Applications of the Third Gasotransmitter, Hydrogen Sulfide (H2S), 55-69. Frankel, Arthur E., Laura A. Coughlin, Jiwoong Kim, Thomas W. Froehlich, Yang Xie, Eugene P. Frenkel, and Andrew Y. Koh. 2017. « Metagenomic Shotgun Sequencing and Unbiased Metabolomic Profiling Identify Specific Human Gut Microbiota and Metabolites Associated with Immune Checkpoint Therapy Efficacy in Melanoma Patients ». Neoplasia (New York, N.Y.) 19 (10): 848-55. Friedman, Jonathan, Logan M. Higgins, and Jeff Gore.2017. « Community Structure Follows Simple Assembly Rules in Microbial Microcosms ». Nature Ecology & Evolution 1 (5): 1-7. Gacesa, R., A. Kurilshikov, A. Vich Vila, T. Sinha, M. a. Y. Klaassen, L. A. Bolte, S. Andreu- Sánchez, et al. 2022. « Environmental Factors Shaping the Gut Microbiome in a Dutch Population ». Nature 604 (7907): 732-39. Gharaibeh, Raad Z., and Christian Jobin.2019. « Microbiota and Cancer Immunotherapy: In Search of Microbial Signals ». Gut 68 (3): 385-88. Ghosh, Tarini S., Mrinmoy Das, Ian B. Jeffery, and Paul W. O’Toole.2020. « Adjusting for Age Improves Identification of Gut Microbiome Alterations in Multiple Diseases ». ELife 9, e50240. Gilbert, Jack A., Martin J. Blaser, J. Gregory Caporaso, Janet K. Jansson, Susan V. Lynch, and Rob Knight.2018. « Current Understanding of the Human Microbiome ». Nature Medicine 24 (4): 392-400. Gopalakrishnan, V., C. N. Spencer, L. Nezi, A. Reuben, M. C. Andrews, T. V. Karpinets, P. A. Prieto, et al. 2018. « Gut Microbiome Modulates Response to Anti-PD-1 Immunotherapy in Melanoma Patients ». Science (New York, N.Y.) 359 (6371): 97-103. Gupta, Vinod K., Minsuk Kim, Utpal Bakshi, Kevin Y. Cunningham, John M. Davis, Konstantinos N. Lazaridis, Heidi Nelson, Nicholas Chia, and Jaeyun Sung.2020. « A Predictive Index for Health Status Using Species-Level Gut Microbiome Profiling ». Nature Communications 11 (1): 4635. Heng, Daniel Y. C., Wanling Xie, Meredith M. Regan, Mark A. Warren, Ali Reza Golshayan, Chakshu Sahi, Bernhard J. Eigl, et al.2009. « Prognostic Factors for Overall Survival in Patients with Metastatic Renal Cell Carcinoma Treated with Vascular Endothelial Growth Factor-Targeted Agents: Results from a Large, Multicenter Study ». Journal of Clinical Oncology: Official Journal of the American Society of Clinical Oncology, 27 (34): 5794-99. Imhann, Floris, Marc Jan Bonder, Arnau Vich Vila, Jingyuan Fu, Zlatan Mujagic, Lisa Vork, Ettje F. Tigchelaar, et al.2016. « Proton Pump Inhibitors Affect the Gut Microbiome ». Gut 65 (5): 740-48. Jackson, Matthew A., Serena Verdi, Maria-Emanuela Maxan, Cheol Min Shin, Jonas Zierer, Ruth C. E. Bowyer, Tiphaine Martin, et al.2018. « Gut Microbiota Associations with Common Diseases and Prescription Medications in a Population-Based Cohort ». Nature Communications 9 (1): 2655. Kawashima, Tadaomi, Akemi Kosaka, Huimin Yan, Zijin Guo, Ryosuke Uchiyama, Ryutaro Fukui, Daisuke Kaneko, et al. 2013. « Double-Stranded RNA of Intestinal Commensal but Not Pathogenic Bacteria Triggers Production of Protective Interferon-β ». Immunity 38 (6): 1187-97. Lee, Karla A., Andrew Maltez Thomas, Laura A. Bolte, Johannes R. Björk, Laura Kist de Ruijter, Federica Armanini, Francesco Asnicar, et al. 2022. « Cross-Cohort Gut Microbiome Associations with Immune Checkpoint Inhibitor Response in Advanced Melanoma ». Nature Medicine 28 (3): 535-44. Li, Min, Baohong Wang, Menghui Zhang, Mattias Rantalainen, Shengyue Wang, Haokui Zhou, Yan Zhang, et al.2008. « Symbiotic Gut Microbes Modulate Human Metabolic Phenotypes ». Proceedings of the National Academy of Sciences of the United States of America 105 (6): 2117-22. Limeta, Angelo, Boyang Ji, Max Levin, Francesco Gatto, and Jens Nielsen.2020. « Meta- Analysis of the Gut Microbiota in Predicting Response to Cancer Immunotherapy in Metastatic Melanoma ». JCI Insight, 5 (23). Mager, Lukas F., Regula Burkhard, Nicola Pett, Noah C. A. Cooke, Kirsty Brown, Hena Ramay, Seungil Paik, et al. 2020. « Microbiome-Derived Inosine Modulates Response to Checkpoint Inhibitor Immunotherapy ». Science (New York, N.Y.) 369 (6510): 1481-89. McCulloch, John A., Diwakar Davar, Richard R. Rodrigues, Jonathan H. Badger, Jennifer R. Fang, Alicia M. Cole, Ascharya K. Balaji, et al. 2022. « Intestinal Microbiota Signatures of Clinical Response and Immune-Related Adverse Events in Melanoma Patients Treated with Anti-PD-1 ». Nature Medicine 28 (3): 545-56. Messaoudene, Meriem, Reilly Pidgeon, Corentin Richard, Mayra Ponce, Khoudia Diop, Myriam Benlaifaoui, Alexis Nolin-Lapalme, et al.2022. « A Natural Polyphenol Exerts Antitumor Activity and Circumvents Anti-PD-1 Resistance through Effects on the Gut Microbiota ». Cancer Discovery 12 (4): 1070-87. Newsome, Rachel C., Raad Z. Gharaibeh, Christine M. Pierce, Wildson Vieira da Silva, Shirlene Paul, Stephanie R. Hogue, Qin Yu, et al. 2022. « Interaction of Bacterial Genera Associated with Therapeutic Response to Immune Checkpoint PD-1 Blockade in a United States Cohort ». Genome Medicine 14 (1): 35. Overacre-Delgoffe, Abigail E., Hannah J. Bumgarner, Anthony R. Cillo, Ansen H. P. Burr, Justin T. Tometich, Amrita Bhattacharjee, Tullia C. Bruno, Dario A. A. Vignali, and Timothy W. Hand.2021. « Microbiota-Specific T Follicular Helper Cells Drive Tertiary Lymphoid Structures and Anti-Tumor Immunity against Colorectal Cancer ». Immunity 54 (12): 2812-2824.e4. Park, Elizabeth M., Manoj Chelvanambi, Neal Bhutiani, Guido Kroemer, Laurence Zitvogel, and Jennifer A. Wargo.2022. « Targeting the Gut and Tumor Microbiota in Cancer ». Nature Medicine 28 (4): 690-703. Peschel, Stefanie, Christian L Müller, Erika von Mutius, Anne-Laure Boulesteix, and Martin Depner.2021. « NetCoMi: network construction and comparison for microbiome data in R ». Briefings in Bioinformatics 22 (4): bbaa290. Roberti, Maria Paula, Satoru Yonekura, Connie P. M. Duong, Marion Picard, Gladys Ferrere, Maryam Tidjani Alou, Conrad Rauber, et al. 2020. « Chemotherapy-Induced Ileal Crypt Apoptosis and the Ileal Microbiome Shape Immunosurveillance and Prognosis of Proximal Colon Cancer ». Nature Medicine 26 (6): 919-31. Routy, Bertrand, Emmanuelle Le Chatelier, Lisa Derosa, Connie P. M. Duong, Maryam Tidjani Alou, Romain Daillère, Aurélie Fluckiger, et al. 2018. « Gut Microbiome Influences Efficacy of PD-1-Based Immunotherapy against Epithelial Tumors ». Science (New York, N.Y.) 359 (6371): 91-97. Sanchez-Gorostiaga, Alicia, Djordje Bajić, Melisa L. Osborne, Juan F. Poyatos, and Alvaro Sanchez. 2019. « High-Order Interactions Distort the Functional Landscape of Microbial Consortia ». PLoS Biology 17 (12): e3000550. Santos Rocha, Clarissa, Omar Lakhdari, Hervé M. Blottière, Sébastien Blugeon, Harry Sokol, Luis G. Bermúdez-Humarán, Vasco Azevedo, et al. 2012. « Anti-Inflammatory Properties of Dairy Lactobacilli ». Inflammatory Bowel Diseases 18 (4): 657-66. Shaikh, Fyza Y., James R. White, Joell J. Gills, Taiki Hakozaki, Corentin Richard, Bertrand Routy, Yusuke Okuma, et al.2021. « A Uniform Computational Approach Improved on Existing Pipelines to Reveal Microbiome Biomarkers of Nonresponse to Immune Checkpoint Inhibitors ». Clinical Cancer Research 27 (9): 2571-83. Smith, Christof C., Kathryn E. Beckermann, Dante S. Bortone, Aguirre A. De Cubas, Lisa M. Bixby, Samuel J. Lee, Anshuman Panda, et al. 2018. « Endogenous Retroviral Signatures Predict Immunotherapy Response in Clear Cell Renal Cell Carcinoma ». The Journal of Clinical Investigation 128 (11): 4804-20. Smith, Melody, Anqi Dai, Guido Ghilardi, Kimberly V. Amelsberg, Sean M. Devlin, Raymone Pajarillo, John B. Slingerland, et al.2022. « Gut Microbiome Correlates of Response and Toxicity Following Anti-CD19 CAR T Cell Therapy ». Nature Medicine 28 (4): 713-23. Sonpavde, Guru P., Cora N. Sternberg, Yohann Loriot, Aurelien Marabelle, Jae Lyun Lee, Aude Fléchon, Guilhem Roubaud, et al. 2022. “Primary Results of STRONG: An Open-Label, Multicenter, Phase 3b Study of Fixed-Dose Durvalumab Monotherapy in Previously Treated Patients with Urinary Tract Carcinoma.” European Journal of Cancer 163 Spencer, Christine N., Jennifer L. McQuade, Vancheswaran Gopalakrishnan, John A. McCulloch, Marie Vetizou, Alexandria P. Cogdill, Md A. Wadud Khan, et al. 2021. « Dietary Fiber and Probiotics Influence the Gut Microbiome and Melanoma Immunotherapy Response ». Science (New York, N.Y.) 374 (6575): 1632-40. Stanley, Dragana, Linda J. Mason, Kate E. Mackin, Yogitha N. Srikhanta, Dena Lyras, Monica D. Prakash, Kulmira Nurgali, et al.2016. « Translocation and Dissemination of Commensal Bacteria in Post-Stroke Infection ». Nature Medicine 22 (11): 1277-84. Teng, Huajing, Yan Wang, Xin Sui, Jiawen Fan, Shuai Li, Xiao Lei, Chen Shi, et al. 2023. “Gut Microbiota-Mediated Nucleotide Synthesis Attenuates the Response to Neoadjuvant Chemoradiotherapy in Rectal Cancer.” Cancer Cell 41 (1). Terrisse, Safae, Lisa Derosa, Valerio Iebba, François Ghiringhelli, Ines Vaz-Luis, Guido Kroemer, Marine Fidelle, et al. 2021. « Intestinal Microbiota Influences Clinical Outcome and Side Effects of Early Breast Cancer Treatment ». Cell Death & Differentiation, 1-19. Thomas, Andrew Maltez, Marine Fidelle, Bertrand Routy, Guido Kroemer, Jennifer A. Wargo, Nicola Segata, and Laurence Zitvogel. 2023. “Gut OncoMicrobiome Signatures (GOMS) as next-Generation Biomarkers for Cancer Immunotherapy.” Nature Reviews. Clinical Oncology, June. Tsay, Jun-Chieh J., Benjamin G. Wu, Imran Sulaiman, Katherine Gershner, Rosemary Schluger, Yonghua Li, Ting-An Yie, et al. 2021. « Lower Airway Dysbiosis Affects Lung Cancer Progression ». Cancer Discovery, 11(2): 293-307. Vétizou, Marie, Jonathan M. Pitt, Romain Daillère, Patricia Lepage, Nadine Waldschmitt, Caroline Flament, Sylvie Rusakiewicz, et al.2015. « Anticancer Immunotherapy by CTLA-4 Blockade Relies on the Gut Microbiota ». Science (New York, N.Y.) 350 (6264): 1079-84. Wu, Guojun, Ting Xu, Naisi Zhao, Yan Y. Lam, Xiaoying Ding, Dongqin Wei, Jian Fan, et al. 2022. “Two Competing Guilds as a Core Microbiome Signature for Health Recovery.” bioRxiv. Yonekura, Satoru, Safae Terrisse, Carolina Alves Costa Silva, Antoine Lafarge, Valerio Iebba, Gladys Ferrere, Anne-Gaelle Goubet, et al.2022. « Cancer Induces a Stress Ileopathy Depending on B-Adrenergic Receptors and Promoting Dysbiosis That Contribute to Carcinogenesis ». Cancer Discovery, Volume 12, Issue 4, 1128-1151. Zelante, Teresa, Rossana G. Iannitti, Cristina Cunha, Antonella De Luca, Gloria Giovannini, Giuseppe Pieraccini, Riccardo Zecchi, et al. 2013. « Tryptophan Catabolites from Microbiota Engage Aryl Hydrocarbon Receptor and Balance Mucosal Reactivity via Interleukin-22 ». Immunity 39 (2): 372-85. Zitvogel, Laurence, Yuting Ma, Didier Raoult, Guido Kroemer, and Thomas F. Gajewski. 2018. « The Microbiome in Cancer Immunotherapy: Diagnostic Tools and Therapeutic Strategies ». Science (New York, N.Y.) 359 (6382): 1366-70.