Products and uses thereof for predicting the sensitivity of a subject to cancer immunotherapy

Abstract

The present invention relates to the field of oncology. It more particularly relates to a method for evaluating the prognosis or the response to a therapeutic treatment of a subject having a squamous cell carcinoma by assessing the multinucleated giant cells status of a SCC tumor of the subject. This method may be advantageously computer-implemented for ease of use in the clinic.

Description

[0001] PRODUCTS AND USES THEREOF FOR PREDICTING THE SENSITIVITY OF A

[0002] SUBJECT TO CANCER IMMUNOTHERAPY

[0003] FIEED OF THE INVENTION

[0004] The present invention relates to the field of oncology. It more particularly relates to a method for evaluating the prognosis or the response to a therapeutic treatment of a subject having a squamous cell carcinoma (SCC) by assessing the multinucleated giant cells (MGC) status of a SCC tumor of the subject. This method may be advantageously computer-implemented for ease of use in the clinic.

[0005] BACKGROUND OF THE INVENTION

[0006] Among non-communicable diseases, cancer is anticipated to become the leading cause of death in the next coming years (Sung H. et al). In 2020, 19.3 millions of cancers had been diagnosed worldwide resulting in nearly 10 million deaths (Sung H. et al). Squamous Cell Carcinoma (SCC) is one of the most frequent carcinomas and arises in several organs including lung, head and neck, esophagus, skin, and uterine cervix: among these, head and neck SCC (HNSCC) is one of the most common, with 890,000 new cases and 450,000 deaths globally in 2020. Its incidence is rising, expected to escalate by 20% by 2030, representing 1.08 million of new cases per year (Sung H. et al). Current treatments for HNSCC involve the combination of surgery, chemotherapy and radiotherapy; alongside, immunotherapy is now approved for recurrent, metastatic, and unresectable carcinomas (Ferris R. L. et al). However, over 50% of patients experience recurrent or metastatic disease within three years of diagnosis (Ferris R. L. et al) and frequently suffer from numerous complications or disabilities even after recovery (Johnson D. E. et al). Gold standard patient stratification for HNSCC is the TNM staging from the last edition of the American Join Committee on Cancer (AJCC - 8th edition) where patients with equivalent staging receive similar treatment although they often hold heterogenous outcomes. Such suboptimal stratification results in inadequate therapy leading to either undertreatment compromising patient recovery or overtreatment which increases the risk of adverse events and unnecessary clinical expenses. Therefore, there is an unmet clinical need for novel individual prognostic biomarkers to guide therapeutic decisions.

[0007] Improving treatment requires a more complete understanding of cancer biology. Recent work has highlighted the long under-appreciated role of the tumor micro-environment (TME) in shaping all stages of the disease, from its genesis through its development and growth, as well as the response of patients to treatment, the onset of metastasis and likelihood of relapse (Hanahan D. et Weinberg R. A.). The TME comprises blood vessels, extracellular matrix, secreted proteins, and non-malignant cells including cancer-associated fibroblasts and immune cells. Among them, tumor-associated macrophages (TAM) are particularly abundant and exert a major influence on tumor biology (Pittet M. J. et al). Recent studies employing single-cell RNA sequencing (scRNAseq) have described several TAM transcriptional programs that are conserved across tumors (Mulder K. et al ', of particular interest, macrophages bearing the Triggering Receptor Expressed on Myeloid Cells-2 (TREM2) have been shown to exert immunosuppressive and pro-tumoral functions in various cancers (Colonna M; Molgora M. et al).

[0008] Alongside conventional TAMs, some tumors, including a proportion of HNSCC, also contain macrophages that exhibit a large cytoplasm and multiple nuclei, and are termed multinucleated giant cells (MGC) (Burkhardt A et Gebbers J.O.). In SCC, myeloid cells from the mononuclear phagocyte system (macrophages and monocytes) fused together to form MGC that perform a “foreign body reaction” targeting extracellular keratin produced by the carcinomatous cells (Burkhardt A et Gebbers J.O.). Accordingly, their presence is well-documented in keratinizing SCC (Patil S. et al), yet the skilled person knows little about their biology or effects in these tumors. While some studies suggest a positive association between the presence of MGC and patient survival in esophageal and oral SCC in patients naive to treatment (Wang H. et al, de Meideros V.A. et al),- others on the contrary show a potentially detrimental effect of these cells and of macrophages in oral SCC (Pandiar D. et al. ; Hsieh C.-Y. et al.).

[0009] Conventional investigation of these cells by flow cytometry and scRNAseq has been hampered by their extraordinary size, which renders them unable to pass through standard sample filters; therefore, the clinical impact of MGC in SCC remains largely unexplored.

[0010] Herein, inventors used samples from several large cohorts of patients with SCC, in particular HNSCC, to ask about the association between MGC and survival following resection of their tumors. Inventors investigated the interaction between MGC, cancer treatments (in particular chemotherapy and immunotherapy) and patient outcomes, and now herein describe an invention enabling their findings to be exploited in the clinic and that can be readily adapted to SCC arising in different tissues. Lastly, inventors used a spatial transcriptomic approach on tumor microscopic slides and herein reveal the unique scRNA signature of MGC within keratin-rich tumor niches, and their relationship to TREM2 -expressing TAM. SUMMARY OF THE INVENTION

[0011] Inventors deeply explored the clinical impact of multinucleated giant cells (MGC) in Squamous Cell Carcinoma (SCC), and herein identify both MGC and TREM2Hlghmacrophages as independent valuable prognostic biomarkers allowing optimal stratification of cancer patients for the first time. These new biomarkers, as well as correlated biomarkers, also identified in the context of the present invention, such as the herein below described proteins expressed or secreted by MGC, are advantageously usable to guide therapeutic decision and limit the risk of adverse events resulting from inappropriate therapeutic options.

[0012] These predictive biomarkers are in particular able to secure identification of SCC patients prone to respond or, on the contrary, resist to a proposed therapy, for example a chemotherapy or immunotherapy.

[0013] A first object of the invention relates to a method for evaluating the prognosis or the response to a therapeutic treatment of a subject having a SCC, wherein the method comprises a step of assessing the MGC status of a SCC tumor of the subject.

[0014] In a particular aspect, the multinucleated giant cells (MGC) status of a SCC tumor is assessed in the context of said method by a):

[0015] - determining a “MGC or TREM2Hlghmacrophages-tumor” ratio defined as i) the number of MGC, or of TREM2Hlghmacrophages, per mm2of SCC tumor area, or ii) the number of MGC, or of TREM2Hlghmacrophages, per the number of cancer cells, in a SCC tumor sample,

[0016] - detecting or measuring the expression by MGC, and optionally by distinct macrophages, of the CHIT1 and / or TREM2 gene(s), in particular of the ( HITT FBP1 and / or TREM2 gene(s), in a SCC tumor sample, and / or

[0017] - detecting or measuring the secretion by MGC, and optionally by distinct macrophages, of CHIT1, FBP1 and / or TREM2 protein(s), for example of CHIT1 and / or TREM2 protein(s), in a blood sample and / or in a SCC tumor sample of the subject; and b) determining that the tumor has a MGCHlgh, MGC1111, or MGCLowstatus, a MGCHlghstatus of the tumor being correlated to a favorable prognosis or positive (i.e., good or partial) response to the treatment, a MGCLowstatus of the tumor being correlated to a bad prognosis or resistance to the treatment, and a MGCIntstatus being correlated to an intermediate prognosis or resistance (i.e., to a poor or intermediate response) to the treatment.

[0018] This method may be used in addition for monitoring SCC tumor evolution (progression or regression) in a subject; for monitoring the response of a subject to a treatment of SCC; for selecting the appropriate treatment of SCC for a subject in need thereof; for selecting subjects capable of responding to a treatment of SCC; or for selecting subjects for enrolment in a clinical trial for the treatment of SCC.

[0019] The method may be used in addition for monitoring SCC tumor evolution in a subject; for monitoring the response of a subject to a treatment of SCC; for selecting the appropriate treatment of SCC for a subject in need thereof; for selecting subjects capable of responding to a treatment of SCC; or for selecting subjects for enrolment in a clinical trial for the treatment of SCC.

[0020] The method may be a partially or fully computer-implemented method. This method preferably uses a classifier as herein described by inventors, trained to assess the MGC status of a tumor.

[0021] A computer-implemented method of training a classifier for assessing the MGC status of a SCC tumor is in particular herein described. In this method, at least one classifier is trained with images wherein MGC, or both MGC and tumor cells, are annotated, in order for the classifier to provide a “MGC -tumor” ratio of the number of MGC per the number of cancer cells or per mm2of tumor area, thereby assessing the MGC status and allowing to evaluate the prognosis or the response to a therapeutic treatment of a subject having a SCC. In a particular aspect, two distinct classifiers are trained, a first classifier being trained for detecting (for example counting) MGC in a SCC tumor sample, and a second classifier being trained for detecting (for example counting) tumor cells in a SCC tumor sample or for measuring the SCC tumor area.

[0022] Also herein described is a computer-implemented method of training a classifier for assessing the MGC status of a SCC tumor, wherein the method comprises: a) providing a training set of tumor images each tumor being obtained from a subject suffering from a SCC, or preprocessed information obtained from said training set, as input to a classifier, said training set comprising i) images of MGCHlghtumors, obtained from subjects suffering from a SCC known as having a MGCHlghstatus, and ii) images of MGCLowtumors, obtained from subjects suffering from a SCC known as having a MGCLowstatus; b) generating an output of the classifier for each image, said output classifying the tumor image input as having a MGCHlghor MGCLowstatus; and c) evaluating the classifier’s performance for distinguishing between MGCHlghand MGCLowstatus by comparing, for each image, the output of the classifier to the known actual status of the tumor; wherein the classifier is considered as a trained accurate (i.e., efficient) classifier to determine the MGC status of a tumor, if it exhibits an Area Under the ROC Curve (ROC AUC) for each tumor above 0.65. A particular computer-implemented method of training such a classifier for assessing or determining the MGC status of a SCC tumor, comprises the following steps: a)

[0023] - of providing a training set comprising images of healthy tissue and / or images of SCC tumor, or preprocessed information obtained from said training set, as input to a first classifier module, and / or

[0024] - of providing a training set comprising images of healthy tissue and / or images of SCC tumor, or preprocessed information obtained from said training set, as input to a second classifier module; b) of generating

[0025] - an output of the first classifier for each image, said output consisting in an image wherein cells identified by the first classifier as tumor cells are annotated or labelled; and / or

[0026] - an output of the second classifier for each image, said output consisting in an image wherein cells identified by the second classifier as MGC, or as TREM2HIGH macrophages, are annotated or labelled; and c) of evaluating

[0027] - the performance of the first classifier for detecting tumor cells by comparing, for each image, the output of the classifier to the known actual identity of the cells, and / or

[0028] - the performance of the second classifier for detecting MGC, or TREM2HIGH macrophages, by comparing, for each image, the output of the classifier to the known actual identity of the cells; wherein the first trained classifier is considered as efficient for identifying tumor cells if it exhibits a mean Average Precision (mAP) on tumor images from a testing set above 0.2, and wherein the second trained classifier is considered as efficient for identifying MGC, or TREM2HIGHmacrophages, if it exhibits a mean Average Precision (mAP) on tumor images from a testing set above 0.2.

[0029] Also herein described are new methods and predictive tools for evaluating or determining the prognostic of a SCC patient. The methods and tools of the invention may be used in addition for identifying patients who can respond and benefit from a SCC treatment, for selecting an appropriate treatment of SCC for a particular patient or subject in need thereof, for monitoring the effects of a treatment on a SCC patient or in other words for monitoring the response of a subject to a treatment of SCC, for monitoring SCC tumor evolution in a subject; or for selecting or disqualifying a subject having a SCC for inclusion (enrolment) in a clinical trial (the clinical trial being for evaluating a therapy directed against a SCC), for example. Herein described in particular is a computing system comprising:

[0030] - a memory storing at least one instruction of a classifier trained preferably according to a method of the invention, and

[0031] - a processor accessing to the memory for reading said instruction(s) and executing the method according to the invention.

[0032] Also described is a kit comprising i) a memory storing at least one instruction of a classifier trained according to a method of the invention on a support, and ii) at least one detection means that specifically recognizes a MGC or TREM2Hlghmacrophage, or that specifically recognizes a CHIT1, FBP1 or TREM2 gene or protein, as well as uses thereof in particular for analyzing the MGC status of a SCC.

[0033] DETAILED DESCRIPTION OF THE INVENTION

[0034] Squamous cell carcinoma (SCC) is one of the most frequent carcinomas and arises in several organs including for example head and neck (HN or H&N), lung, esophagus, skin, and uterus cervix (CE).

[0035] Patients with head and neck squamous cell carcinomas (HNSCC) often have poor outcomes as current risk-management and treatment strategies are suboptimal; yet novel prognostic biomarkers are generally difficult to integrate into clinical practice.

[0036] Here, inventors report the presence of multinucleated giant cells (MGC) - a type of macrophages -, or the presence of TREM2HIGHmacrophages, in tumors from patients with SCC, in particular HNSCC, and reveal the correlation between their presence and a favorable prognosis in treatment-naive and preoperative-chemotherapy-, or neoadjuvant immunotherapy-, treated patients.

[0037] Importantly, they observed a MGC density (i.e., MGC number per mm2of SCC tumor area), or a TREM2HIGHmacrophages density, increase following preoperative chemotherapy and neoadjuvant immunotherapy and discovered a critical role of these cells in anti-tumoral response.

[0038] To facilitate clinical translation of MGC density as a prognostic (bio)marker, inventors developed a model, in particular a deep-learning model, to automate its quantification on routinely stained pathological (whole slide) images (WSI).

[0039] Inventors also defined an MGC-specific RNA signature using spatial transcriptomics and discovered a close resemblance to TREM2-expressing mononuclear tumor-associated macrophages (also herein identified as “TREM2 macrophages” or “TREM2Hlgh macrophages”), which co-localize with MGC in keratin-rich tumor niches, and were also associated by inventors with a good response to SCC treatment.

[0040] Inventors now herein provide for the first time a method for evaluating the prognosis or the response to a therapeutic treatment of a subject having a cancer, in particular a SCC, more particularly a HNSCC. This method comprises a step of assessing the MGC status of a SCC tumor of the subject.

[0041] In the context of the present invention, the evaluation of the prognosis of a subject having a SCC refer to the prediction of the likely (probable) course of the SCC, or in other words, to the prospect of recovery or death as anticipated from the usual course of the disease. Thus, the term “course” refers both to the progression of the disease, i.e. to detectable changes (either positive - when an improvement is observed - or negative - when a complication is observed - to the subject), and to the outcome of the disease (i.e., high chance of healing from the disease, or low chance of recovery from the disease / high risk of death).

[0042] The prognosis is closely related to the response of the subject to a therapeutic treatment. Indeed, a favourable response (sensitivity or responsiveness) of the patient to the treatment is correlated with a positive (good) prognosis, a partially favourable response of the patient to the treatment is correlated with a positive but incomplete response to the treatment, and an unfavourable or negative response of the patient to the treatment (i.e., resistance) is correlated with a negative (bad) prognosis. For example, if a subject’s SCC is an aggressive type or has already metastasized to other areas, an oncologist may give the subject a bad prognosis. Thus, three categories of prognosis may be distinguished, herein distributed between “poor” (negative), “partial” (positive but incomplete) and “good” (positive).

[0043] By “sensitivity” or “responsiveness” is intended herein the likelihood that a patient will respond to a cancer treatment as herein described.

[0044] By “resistant” or “resistance” is intended herein the likelihood that a patient will not respond to such a cancer treatment.

[0045] The prognosis may be more precisely measured for example in terms of overall survival (OS), defined as the time from treatment to death, regardless of disease recurrence. The prognosis may also be more precisely measured for example in terms of progression free interval (PFI) defined as the appearance of a new tumor event whether it was a progression of disease, local recurrence, distant metastasis, new primary tumors all sites, or death of the subject suffering of a SCC without new tumor event, including cases with a new tumor event whose type is not available (N / A) according to Eberle et al. , or defined by Institut Gustave Roussy as events for progression of SCC, appearance of a second SCC, death by SCC, or death of unknown cause with SCC. In the context of the present invention, the “subject” (or “patient”) is a mammal. In a particular embodiment, the mammal is a human being, whatever its age or sex. The patient typically has a tumor. Unless otherwise specified in the present disclosure, the “tumor” is a cancerous or malignant tumor, more specifically a squamous cell carcinoma (SCC).

[0046] In a particular aspect, the subject is a subject who has not been previously exposed to a treatment of cancer or, in other words, who did not receive any pre-operative therapy (a “treatment-naive patient”).

[0047] In another particular and preferred aspect, the subject is a subject who has received a treatment, for example a chemotherapy, in particular an induction chemotherapy (ICT), or an immunotherapy, in particular a neoadjuvant immunotherapy.

[0048] In a particular aspect, the subject who has received a treatment is a subject suffering of a HNSCC, in particular of a SCC of the oral cavity, for example a tongue SCC, a floor of the mouth SCC, a gum SCC, a cheek mucosa SCC or any other known SCC of the oral cavity.

[0049] In another particular aspect, the subject suffering of a SCC of the oral cavity is a subject suffering of a floor of the mouth SCC, a gum SCC and / or a cheek mucosa SCC, in particular a subject who does not suffer from a tongue SCC.

[0050] In another particular aspect, the subject who has received a treatment is a subject suffering of a esophageal SCC.

[0051] In another particular aspect, the subject who has received a treatment is not a subject suffering of a esophageal SCC.

[0052] The term “treatment” refers to any act intended to ameliorate the health status of patients such as therapy, prevention, prophylaxis and retardation of the disease or of the symptoms of the disease. It designates both a curative treatment and / or a prophylactic treatment of the disease. A curative treatment is defined as a treatment resulting in cure or a treatment alleviating, improving and / or eliminating, reducing and / or stabilizing a disease or the symptoms of a disease or the suffering that it causes directly or indirectly. A prophylactic treatment comprises both a treatment resulting in the prevention of a disease and a treatment reducing and / or delaying the progression and / or the incidence of a disease or the risk of its occurrence or recurrence. In certain aspects, such a term refers to the improvement or eradication of a disease, a disorder, an infection or symptoms associated with it. In other aspects, this term refers to minimizing the spread or the worsening of the disease. Treatments described in the context of the present invention do not necessarily imply 100% or complete treatment. Rather, there are varying degrees of treatment recognized by one of ordinary skill in the art as having a potential benefit or therapeutic effect. Preferably, the term “treatment” refers to the application or administration of a composition including one or more active agents to a subject who has a disorder / disease. In a particular aspect, the treatment is cancer treatment, more particularly a SCC treatment. Preferably, the anti -cancer treatment is selected from the group consisting of resection surgery, chemotherapy, radiotherapy or immunotherapy. Preferably, the therapeutic compound is a chemotherapeutic or immunotherapeutic compound.

[0053] The term “induction chemotherapy” (ICT) refers to the first-line treatment of cancer with a chemotherapeutic drug, this induction therapy being performed before any therapeutic surgical or resection step (surgery) and before any radiotherapy (radiation therapy). The term “induction chemotherapy” is herein considered as equivalent to the term “neoadjuvant chemotherapy”.

[0054] Examples of chemotherapeutic compounds usable in the context of the invention may be, without limitation, alkylating agents, antimetabolites, plant alkaloids, topoisomerase inhibitors, or antitumor antibiotics. For example, if the SCC is a HNSCC, a platinum (cisplatin, carboplatin or oxaliplatin, ), a taxane (docetaxel or paclitaxel), methotrexate, 5-FU, and / or capecitabin may be used in the context of an induction chemotherapy (cf. David G Pfister et al., JNCCN, 2020). Docetaxel (or paclitaxel), cisplatin and / or 5-FU may be used in a particular aspect, preferably in combination. A targeted-therapy involving for example cetuximab may also be used for treating HNSCC in the context of ICT.

[0055] The term “neoadjuvant” in relation with an “immunotherapy” refers to the administration of an immunotherapeutic agent before any therapeutic surgical or resection step (surgery) and before any radiotherapy (radiation therapy). It aims to reduce the size or extent of the cancer before using radical treatment intervention, thus both making procedures easier and more likely to succeed and reducing the consequences of a more extensive treatment technique, which would be required if the tumor were not reduced in size or extent. Not everyone is suitable for neoadjuvant therapy because it can be extremely toxic. Some patients react so severely that further treatments, especially surgery, are precluded.

[0056] Examples of immunotherapeutic compounds may be, without limitation, an antibody, in particular a monoclonal antibody, a cytokine, or interferon.

[0057] For example, if the SCC is a HNSCC, an immunotherapeutic agent such as pembrolizumab, nivolumab or a combination thereof may be used in the context of a neoadjuvant immunotherapy (cf. David G Pfister et al., JNCCN, 2020).

[0058] The term “monoclonal antibody based-therapy” refers to the use of a monoclonal antibody exerting direct antitumor effects which result in the death of tumor cells, such as for example cetuximab (an anti-EGFR) and afatinib which may be used in the treatment of HNSCC. In a particular aspect of the present invention, the method of the invention is performed after at least partial, for example total, (surgical) resection of the cancerous tumor and / or metastases thereof, in the subject. The method can however also be performed on a biological sample of the subject before any therapeutic surgical step, for example on a tumor biopsy.

[0059] In a preferred embodiment, the method is performed on a biopsy from the subject.

[0060] In a particular aspect, the method is performed on a biological sample of the subject following at least partial resection of the cancerous tumor and / or metastases thereof.

[0061] Carcinoma refers to a cancer that arise from an epithelial tissue. Such epithelial tissue can be found in the skin or in the lining of internal organs. The two main types of carcinoma are squamous cell carcinoma (SCC) and adenocarcinoma (ADK). Preferably the carcinoma is not an adenocarcinoma. In particular, the carcinoma is not a lung adenocarcinoma or a breast adenocarcinoma.

[0062] “Squamous cell carcinoma” (or “SCC”) arise from epithelial squamous cell and can be found in several areas of the body: skin, head and neck, esophagus, lung, uterine cervix, vagina, anus. SCC can also be found in additional locations but with a low frequency.

[0063] In a particular aspect, the SCC is selected from head and neck squamous cell carcinoma (HNSCC), lung carcinoma, esophagus carcinoma, skin carcinoma and uterus cervix carcinoma (CESCC), for example from head and neck squamous cell carcinoma (HNSCC), lung carcinoma, skin carcinoma and uterus cervix carcinoma, in particular from HNSCC and uterus cervix carcinoma. In a particular aspect, the SCC is not a esophagus carcinoma.

[0064] Preferably the SCC is a HNSCC, for example a SCC of the oral cavity (for example of the gum, cheek mucosa, tongue, or floor of the mouth), a SCC of the sinus(es), a SCC of the nasal cavity(ies), a SCC of the pharynx (in particular of the nasopharynx or of the hypopharynx), a SCC of the larynx, even more preferably a SCC of the tongue, of the floor of the mouth, or of the larynx.

[0065] In a particularly preferred aspect, the SCC is a SCC of the floor of the mouth.

[0066] In a particularly preferred aspect, the SCC is a SCC of the larynx.

[0067] In a particular aspect, the HNSCC is not a SCC of the tongue.

[0068] In another particular aspect, the SCC is not a oesophageal SCC.

[0069] In a preferred aspect, the SCC is a CESCC (also herein identified as “CESC”).

[0070] In the context of the present invention, the particular SCC treatment, may be selected from the group consisting of resection surgery, chemotherapy, radiotherapy or immunotherapy, preferably from a combination of surgery and chemotherapy, surgery and immunotherapy, or surgery and radiation. The therapeutic treatment for a particular SCC which may be evaluated by the method of the invention is typically a treatment recommended by a health authority recognized by health professionals [cf. guidelines recommendations from the NCCN (National Comprehensive Cancer Network), from the ASCO (American Society of Clinical Oncology) and from the ESMO (European society of medical oncology) for example] .

[0071] If the SCC is a HNSCC, the treatment is for example selected from radiotherapy, chemotherapy, immunotherapy, monoclonal antibody based-therapy and a combination of several of said therapies (cf. David G Pfister et al., JNCCN, 2020). In a particular aspect, the treatment is a chemotherapy involving a drug for example a taxane (like docetaxel or paclitaxel), a platinum (like cisplatin, carboplatin or oxaliplatin), methotrexate, 5-FU, and / or capecitabin; or an immunotherapy or monoclonal antibody based-therapy (involving for example cetuximab or an immune checkpoint inhibitor such as pembrolizumab or nivolumab). Additionally, surgical resection followed by adjuvant radiation therapy or chemotherapy plus radiation therapy is a classical treatment for HNSCC, especially in the case of oral cavity cancer (cf. Daniel E. Johnson, et al., “Head and neck squamous cell carcinoma”; Nature Reviews Disease Primers, 6, Article number 92 (2020)).

[0072] If the SCC is a uterus cervix carcinoma, the treatment is selected from a combination of radiation therapy and chemotherapy, the chemotherapy involving for example cisplatin, carboplatin, capecitabine, gemcitabine, topotecan, fluorouracil, pemetrexed, vinorelbine, irinotecan and / or paclitaxel; surgery, radiation therapy and chemotherapy either given separately or in combination; immunotherapy involving for example pembrolizumab and / or nivolumab (preferably for recurrent or metastatic cervical cancer); and monoclonal antibody based-therapy involving for example bevacizumab, tisotumab vetodin-tftv, and / or cemiplimab (cf. Nadeem R. Abu-Rustum, et al., JNCCN, 2023).

[0073] Multinucleated giant cells (“MGC”) are particular tumor associated macrophages (TAM) resulting from the fusion of myeloid cells from the mononuclear phagocyte system (macrophages and monocytes), that exhibit a large cytoplasm and multiple nuclei.

[0074] The term “MGC status” used in relation with a SCC tumor of a subject refers to the classification of the SCC tumor as MGCHlgh, MGC1'11. or MGCLow, said classification allowing evaluation of the prognosis or response to a therapeutic treatment of the subject.

[0075] The multinucleated giant cells (MGC) status of a SCC tumor may be assessed, in the context of the herein above described method of the invention, by a): - determining a “MGC or TREM2Hlghmacrophages-tumor” ratio defined as i) the number of MGC, or of TREM2Hlghmacrophages, per mm2of SCC tumor area, or ii) the number of MGC, or of TREM2Hlghmacrophages, per the number of cancer cells, in a tumor sample, typically in a SCC tumor sample,

[0076] - detecting or measuring the expression, preferably by MGC, and optionally by distinct macrophages, of the CHIT1 and / or TREM2 gene(s), in particular of the CHIT1 , FBP1 and / or TREM2 gene(s), in a tumor sample, typically in a SCC tumor sample, and / or

[0077] - detecting or measuring the secretion, preferably by MGC, and optionally by distinct macrophages, of CHIT1, FBP 1 and / or TREM2 protein(s), for example of CHIT1 and / or TREM2 protein(s), in a blood sample and / or in a SCC tumor sample of the subject; and b) determining that the tumor has a MGCHlgh, MGCInt, or MGCLowstatus, a MGCHlghstatus of the tumor being correlated to a favorable prognosis or positive response to the treatment, a MGCLowstatus of the tumor being correlated to a bad prognosis or resistance to the treatment, and a MGCIntstatus being correlated to an intermediate prognosis or resistance to the treatment (or in other words to a “poor” or “intermediate” response).

[0078] In a particular aspect, step b) includes determining that the tumor has a MGCIntstatus, the MGClmstatus being correlated to (positive but) partial or poor response of the subject to the treatment.

[0079] The term “sample”, as used in the context of the present invention, refers to a biological sample obtained from a subject suffering of a cancer, typically a SCC. Implementation of the methods of the invention involve obtaining such a (biological) sample from a subject.

[0080] The sample can be a fluid or liquid sample, or a solid sample. It is preferably a fresh sample. It can also be a frozen sample or a fixed sample (for example a formalin-fixed paraffin-embedded sample).

[0081] The solid sample is typically a tumor sample. The tumor sample may be obtained from the subject during (surgical) partial or total resection of a tumor or during biopsy. The biopsy is generally removed from a solid tumor or from tissues or organs suspected to comprise tumor cells. Tissue biopsies are typically utilized when a known tumor’s location is suspected or confirmed and available for extraction. The tumor sample can be for example a tumor biopsy, a whole tumor piece, a tumor bed sample, a metastasis sample, a metastatic lymph node cells sample, or a combination thereof.

[0082] The fluid or liquid sample can be a blood, plasma, lymphatic fluid, or pleural sample.

[0083] In a particular aspect, the multinucleated giant cells (MGC) status of a SCC is assessed by: a) determining, a “MGC or TREM2Hlghmacrophages-tumor” ratio defined as i) the number of MGC, or of TREM2Hlghmacrophages, per mm2of SCC tumor area, or ii) the number of MGC, or of TREM2Hlghmacrophages, per the number of cancer cells, in a SCC tumor sample, and b) determining that the tumor has a MGCHlgh, MGCIntor MGCLowstatus, a MGCHlghstatus of the tumor being correlated to a favorable prognosis or positive response to the treatment, a MGCLowstatus of the tumor being correlated to a bad prognosis or resistance to the treatment and a MGClmstatus being correlated to an intermediate prognosis or resistance to the treatment as explained herein above. In a particular aspect, step b) includes determining that the tumor has a MGClmstatus as indicated herein above, correlated to (positive but) partial or poor response to the treatment.

[0084] TREM2Hlghmacrophages designate (mononucleated or multinucleated giant) macrophages bearing the Triggering Receptor Expressed on Myeloid Cells-2 (TREM2).

[0085] The term “TREM2hlgh” refers to the relative level of expression of the TREM2 gene between macrophages of interest and reference macrophages, the TREM2 level being identified as “high” if the level of expression is higher in the macrophages of interest than in the reference macrophages. In the context of the invention, the macrophage of interest may be for example a mononuclear macrophage or a MGC. For example, if the macrophage of interest is a mononuclear macrophage, the reference macrophage is typically a macrophage from a different population of macrophages in the same tissue (for example a population of macrophages from a healthy part of the same tissue or a macrophage from the same diseased tissue which express TREM 2 at a lower level). The same applies to a MGC as a macrophage of interest.

[0086] The evaluation of the TREM2 level of expression may be easily determined or assessed by the skilled person via RNA sequencing (“RNA-seq”). Indeed, this technology allows the examination of the quantity and sequences of RNA in a sample using next-generation sequencing (NGS). It enables the investigation and discovery of the transcriptome, which is the total cellular content of RNAs, including mRNA, rRNA, and tRNA. The RNA-seq workflow involves several steps, such as RNA extraction, reverse transcription into cDNA, adapted ligation, amplification, and sequencing. It is widely used in cancer research to measure gene expression, identify gene fusions, mutations, and changes in gene expression over time or in different groups or treatments. RNA-seq is a highly sensitive and accurate ( / efficient) tool for measuring expression across the transcriptome, providing visibility into previously undetected changes occurring in disease, for example under different environmental conditions. Note that RNA-seq data can come from various sources and techniques such as bulk-RNA-sequencing, single-cell RNA sequencing or spatial transcriptomics techniques. In this context, differentially expressed genes (DEGs), such as T1U-.M2. are genes that show significant changes in expression levels between different experimental conditions, such as disease versus healthy tissue or treated versus untreated samples. RNA-seq technology is commonly used to identify DEGs by quantifying the abundance of RNA transcripts, allowing researchers to understand how gene expression is regulated under different biological conditions. In the experimental part of the present description, DEG analyzes were performed using the Seurat v4 package. DEGs obtained from the “RNA” matrix of the Seurat object were calculated on normalized values with a log fold change threshold of 0.25 and a min.pct threshold of 0.25. The Wilcoxon-rank sum test was used. However, the skilled person will appreciate that other conditions could be selected since other methods are available in the art to analyze RNA sequencing data [cf. Corchete, L.A., Rojas, E.A., Alonso-Lopez, D. et al. Systematic comparison and assessment of RNA-seq procedures for gene expression quantitative analysis. Sci Rep 10, 19737 (2020); Li, H., Zhou, J., Li, Z. et al. A comprehensive benchmarking with practical guidelines for cellular deconvolution of spatial transcriptomics. Nat Commun 14, 1548 (2023) and Li, B., Zhang, W., Guo, C. et al. Benchmarking spatial and single-cell transcriptomics integration methods for transcript distribution prediction and cell type deconvolution. Nat Methods 19, 662-670 (2022)] .

[0087] Inventors observed, and herein reveal for the first time, that TREM2 -expressing mononuclear macrophages and MGC cluster together in keratin-rich carcinoma niches, TREM2 -expressing mononuclear macrophages share a similar transcriptional program with MGC and are similarly associated with good response to a SCC treatment such as immunotherapy, in particular neoadjuvant immunotherapy, and chemotherapy, in particular induction (or neoadjuvant) chemotherapy. Similarly to MGC, TREM2 -expressing macrophages are a biomarker of good prognosis in SCC, in particular in head and neck squamous cell carcinoma (HNSCC).

[0088] In order to determine the number of MGC, or of TREM2Hlghmacrophages, per mm2of SCC tumor area or per the number of cancer cells, the skilled person can count the cells seen under a microscope (classically used method), possibly with the help of a QuPath software for whole slide image analysis.

[0089] A manual count can be performed directly on a volume of sample deposited on a laboratory slide. The count may also be performed on the basis of one or more images of the tumor.

[0090] The assessment of the MGC status may for example comprise the steps of: a) determining the ratio of the number of MGC and / or TREM2Hlghmacrophages, per mm2of tumor area or per number of cancer cells, the number(s) of MGC, of TREM2Hlghmacrophages, and / or of cancer cells or the tumor area being obtained from (one or several) image(s) of the SCC tumor; b) comparing the ratio of step a) to a reference ratio for a SCC tumor of the same tissue origin, and c) determining the MGC status of the tumor, a tumor being considered as having a MGCHlghstatus if the ratio is equal to or above (>) the reference ratio, and as having a MGCLowstatus if the ratio is below (<) the reference ratio.

[0091] In a particular aspect where the reference ratio is a range of values, the MGC status may be considered as a MGClmstatus if the ratio is equal to or above (>) the lower value of the range and below (<) the upper value of the range.

[0092] The expressions “reference value” or “reference expression level” used in the present description may refer to the concentration of a biomarker in a control sample derived from one or more subjects (reference subject or population) having a cancer, in particular a SCC. The reference value or level is typically the median value obtained from the reference population (for example the median concentration of the CHIT1 or TREM2 protein in the reference population). The reference value typically varies in a range of values defined for a given population suffering of a SCC of identified tissue origin. Thus, said value may vary depending on the tissue origin of the SCC. The reference value can further be a “reference ratio” involving two distinct biomarkers (such as the number of MGC, or TREM2Hlghmacrophages, per mm2of tumor area or per number of cancer cells), or a “reference percentage (%) or proportion” of one several biomarkers obtained from control sample(s), the control sample(s) being typically biopsy(ies) or sample(s) of resected tumor(s).

[0093] As indicated, the reference ratio may vary depending on the tissue origin of the cancer, typically of the SCC, and may be defined as a single value or as a range of values.

[0094] If step b) of any one of the herein above described methods includes determining that the tumor has a MGClmstatus, the reference ratio is typically defined in relation to a range of values, below which the status will be considered as MGCLow, above which the status will be considered as MGCHlghand, within which, it will be considered as MGCInt.

[0095] For example:

[0096] - if the SCC is a HNSCC of the oral cavity, the MGC status may be determined as being MGCHlghif the ratio of the number of MGC per mm2of tumor area is equal to or above (>) 1, as being MGCLowif said ratio is below (<) 0,2, or as being MGCIntin between, i.e., if equal to or above (>) 0,2 and below (<) 1; and (or) - if the SCC is a HNSCC of the oral cavity, the MGC status may be determined as being MGCHlghif the ratio of the number of MGC per the number of cancer cells is below (<) 1 and equal to or above (>) 0,0006, or as being MGChtin between.

[0097] Inventors further herein reveal the unique gene signature of MGC identified from single-cell RNA-sequencing (scRNA). The MGC signature includes at least CHIT1 , TREM2, and FBP1. The relevant signature as determined by scRNA includes in addition the following genes: MARCO, DCSTAMP, TYROBP, CHI3L1, MM I >9. CTSS, CTSZ, CTSD, CTSB, CD68, APOE, SPP1 and OSCAR. This signature of any of the three biomarkers (CHIT1, TREM2 or FBP1), or two or three thereof, or of more biomarkers among the 10 biomarkers, may be used by the skilled person to identify and count MGC in the context of a method of the invention.

[0098] In another aspect, the multinucleated giant cells (MGC) status of a SCC tumor is assessed, in the context of the herein above described method of the invention, by detecting the presence or by measuring the expression of additional herein revealed biomarkers, in particular by: a)

[0099] - detecting or measuring, in particular measuring, the expression, preferably by MGC, and optionally by distinct macrophages, of at least one gene, optionally several genes, selected from CHIT1, TREM2, FBP1, MARCO, DCSTAMP, TYROBP, CHI3L1, MMP9, CTSS, CTSZ, CTSD, CTSB, CD68, APOE, SPP1 and OSCAR, in particular of the CHIT1, TREM2 and / or FBP1 gene(s), more particularly of the CHIT1 and / or TREM2 gene(s), in a SCC tumor sample, and / or

[0100] - detecting or measuring, in particular measuring, the secretion, preferably by MGC, and optionally by distinct macrophages, of at least one protein, possibly several proteins, selected from CHIT1, TREM2, FBP1, MARCO, DCSTAMP, TYROBP, CHI3L1, MMP9, CTSS, CTSZ, CTSD, CTSB, CD68, APOE, SPP1 and OSCAR, in particular of the CHIT1, TREM2 and / or FBP1 protein(s), more particularly of the CHIT1 and / or TREM2 protein(s), in a blood sample and / or in a SCC tumor sample of the subject, and b) determining that the tumor has a MGCHlgh, MGClm. or MGCLowstatus, a MGCHlghstatus of the tumor being correlated to a favorable prognosis or positive response to the treatment, a MGCLowstatus of the tumor being correlated to a bad prognosis or resistance to the treatment and a MGClmstatus being correlated to an intermediate prognosis or resistance to the treatment. Optionally, step b) may include determining that the tumor has a MGClmstatus as indicated herein above, correlated to (positive but) partial response to the treatment.

[0101] In a particular aspect, the expression of the CHIT1 gene or the secretion of the CHIT1 protein is not detected or measured in the context of the method of the invention. Similarly to the level of expression of TREM2, the level of expression of any one of the herein identified genes (CHIT1, TREM2, FBP1, MARCO, DCSTAMP, TYROBP, CHI3L1, MMP9, CTSS, CTSZ, CTSD, CTSB, CD68, APOE, SPP1 and OSCAR) may be easily determined or assessed, preferably measured, by the skilled person via RNA sequencing (“RNA-seq”). These genes are DEGs (differentially expressed genes) that show significant changes in expression levels between different experimental conditions, such as disease versus healthy tissue or treated versus untreated samples, and RNA-seq technology is commonly used to identify DEGs by quantifying the abundance of RNA transcripts. In the context of the present invention, the measure is thus a “relative” measurement compared to another reference condition which may for example correspond to the measure obtained from a healthy tissue, or from distinct macrophages in the same tissue.

[0102] The “distinct macrophages” mentioned herein above designates macrophages different from MGC also present in the tumor, in particular TREM2Hlghmononuclear macrophages, MGC being also herein identified as TREM2Hlghpolynuclear macrophages.

[0103] The expression of the herein described biomarkers may be measured for example as a concentration, as a percentage (%), as a ratio, or as a proportion of one or several of said biomarkers relative to one, several or each of the other biomarkers.

[0104] The expression of a particular gene or protein may be compared to a reference expression of the corresponding gene or protein, or to a reference percentage (%), ratio or proportion of the corresponding gene or protein with respect to other genes or proteins, for a SCC tumor of the same tissue origin, the MGC status of the tumor being determined in comparison to said reference expression, %, ratio or proportion, as corresponding to a MGCHlghstatus if the expression is equal to or above (>) the reference expression, %, ratio or proportion, as corresponding to a MGCLowstatus if the expression is below (<) the reference expression, %, ratio or proportion, and, if the reference expression, %, ratio or proportion varies in a range of values, as corresponding to a MGCIntstatus if within said range.

[0105] Once again, the reference value, expression, percentage (%), ratio or proportion may vary depending on the tissue origin of the SCC.

[0106] The skilled person knows how to detect or measure the expression of a nucleic acid, for example of a desoxyribonucleic acid (DNA) such as a gene, or of a ribonucleic acid (RNA). The skilled person can use any well-known techniques such as for example bulk-RNA-sequencing, single- cell RNA sequencing or a spatial transcriptomics technique, on a biological sample as herein described, preferably a tumor sample.

[0107] In order to detect or measure the expression of an amino acid sequence, for example a peptide or protein, the skilled person can use any well-known detection or dosing method of a protein, such as for example immunohistochemistry (“IHC” - which uses an antibody directed against an antigen of the protein of interest), immunofluorescence, western blot, flow cytometry, a proteomic technique, ELISA, a radioimmunoassay (RIA), on a biological sample as herein described, preferably a tumor (SCC) or blood sample, even more preferably a tumor sample (typically in the form of a tissue or suspension of cells).

[0108] For example, any one of the herein described proteins (for example CHIT1, TREM2 or FBP1) may be detected by IHC using commercially available antibodies capable of recognizing and binding the protein of interest. Another test classically used in immunology to detect the presence of a protein in the cytoplasm or on the surface of a cell is flow cytometry. Here again, commercially available antibodies capable of binding a protein of interest may be used by the skilled person.

[0109] In some aspects of the invention, identification of any of the herein above biomarker of interest (cell, nucleic acid or protein) involves using at least one binding agent.

[0110] Furthermore, it is contemplated that a binding agent may be specific or not to the considered biomarker. Alternatively, different conformations may serve as the basis for binding agents capable of distinguishing between similar biomarkers.

[0111] The binding agent may be a polypeptide (for example when the biomarker is a protein or a cell). The polypeptide is, in particular embodiments, an antibody. In further embodiments, the antibody is a monoclonal or polyclonal antibody. The antibody can be bi-specific, recognizing two different epitopes. The antibody, in some embodiments, immunologically binds to more than one epitope from the same biomarker. In some embodiments of the invention, the binding agent is an aptamer.

[0112] For example, when CHIT1 (SEQ ID NO: 1, 2, 3 or 4) is to be detected, the antibody to be used can be for example a rabbit anti-CHITl antibody (for example the rabbit polyclonal antibody to CHIT1 from Biorbyt, catalog number orb377995 - https: / / www.biorbyt.com / chitl-antibody- orb377995.html), when TREM2 (SEQ ID NO: 6, 7 or 8) is to be detected, the antibody to be used can be for example a rabbit anti-TREM2 antibody (for example the rabbit monoclonal antibody to TREM2 from Cell Signaling, clone D814C, reference 91068 https: / / www.cellsignal.com / products / primary-antibodies / trem2-d8i4c-rabbit-mab / 91068), and when FBP1 (SEQ ID NO: 10) is to be detected, the antibody to be used can be for example a rabbit anti-FBP 1 antibody (for example the rabbit polyclonal antibody to FBP 1 from Invitrogen, reference PA5-76734 - https: / / www.thermofisher.com / antibody / product / FBPl-Antibody- Polyclonal / PA5-76734).

[0113] The binding agent may otherwise be a nucleic acid (typically when the biomarker is a gene or a RNA whose expression is to be measured).

[0114] For example, when the expression of CHIT1 (SEQ ID NO: 5) is to be detected, a nucleic acid probe capable of detecting the expression of the mRNA transcript of the CHIT1 gene may be used, when the expression of TREM2 (SEQ ID NO: 9) is to be detected, a nucleic acid probe capable of detecting the expression of the mRNA transcript of the TREM2 gene may be used, and when the expression of FBP1 (SEQ ID NO: 11) is to be detected, a nucleic acid probe capable of detecting the expression of the mRNA transcript of the FBP1 gene may be used. Nucleic acids capable of recognizing the herein described genes of interest are available in the art or may be easily synthesized by the skilled person.

[0115] In some embodiments of the invention, the binding agent is labeled. In further embodiments, the label is radioactive, fluorescent, chemiluminescent, an enzyme, or a ligand. It is also specifically contemplated that a binding agent is unlabeled, but may be used in conjunction with a detection agent that is labeled. A detection agent is a compound that allows for the detection or isolation of itself so as to allow detection of another compound that binds, directly or indirectly. An indirect binding refers to binding among compounds that do not bind each other directly but associate or are in a complex with each other because they bind the same compounds or compounds that bind each other.

[0116] Other embodiments of the invention involve a second binding agent in addition to a first binding agent. The second binding agent may be any of the entities discussed above with respect to the first binding agent, such as an antibody. It is contemplated that a second antibody may bind to the same of different epitopes as the first antibody. It is also contemplated that the second antibody may bind the first antibody or another epitope than the one recognized by the first antibody.

[0117] As discussed earlier, binding agents may be labeled or unlabeled. Any polypeptide binding agent used in methods of the invention may be recognized using at least one detection agent. A detection agent may be an antibody that binds to a polypeptide binding agent, such as an antibody. The detection agent antibody, in some embodiments, binds to the Fc-region of a binding agent antibody. In further aspects, the detection agent is biotinylated and possibly incubated, in additional aspects, with a second detection agent comprising streptavidin and a label. It is contemplated that the label may be radioactive, fluorescent, chemiluminescent, an enzyme, or a ligand. In some cases, the label is an enzyme, such as horseradish peroxidase.

[0118] The ELISA assay for example is a sandwich assay. In a sandwich assay, more than one antibody will be employed. Typically, ELISA method can be used, wherein the wells of a microtiter plate are coated with a set of antibodies which recognize the protein of interest. A sample containing or suspected of containing the protein of interest is then added to the coated wells. After a period of incubation sufficient to allow the formation of antibody-antigen complexes, the plate(s) can be washed to remove unbound moieties and a detectably labelled secondary binding molecule added. The secondary binding molecule is allowed to react with any captured sample marker protein, the plate washed and the presence of the secondary binding molecule detected using methods well known in the art.

[0119] In the methods herein described, any classical method, well-known by the skilled person, of determining the presence or measuring the expression level of a biomarker of interest, such as typically IHC, a flow cytometry technology such as Fluorescence-activated cell sorting (FACS), ELISA, radioimmunoassay (RIA) and mass spectrometry for example can be used.

[0120] FACS can be used for distinguishing and separating into two or more containers specific cells from a heterogeneous mixture of biological cells, based upon the specific light scattering and fluorescent characteristics of each cell. RIA can otherwise be used in certain aspects of the invention easily identifiable by the skilled person. This technology uses radiolabeled molecules in a stepwise formation of immune complexes.

[0121] A RIA is a very sensitive in vitro assay technique used to measure concentrations of substances, usually measuring antigen or protein concentrations (for example, the levels of MGC secretion product levels in blood) by use of antibodies.

[0122] Any of the herein above described methods of the invention may be used in addition for making treatment decision as soon as possible for a particular patient, for example for selecting the appropriate treatment of SCC for a subject in need thereof or for selecting subjects capable of responding to a treatment of SCC, thereby limiting toxicities associated to inappropriate therapy; for monitoring the response of a subject to a treatment of SCC; for monitoring SCC tumor evolution (i.e., progression or regression) or stabilization in a subject; or for selecting subjects for enrolment in a clinical trial for the treatment of a SCC.

[0123] In a particular aspect, the method of the invention is used for monitoring SCC tumor evolution in a subject; for monitoring the response of a subject to a treatment of SCC; for selecting the appropriate treatment of SCC for a subject in need thereof; for selecting a subject capable of responding to a treatment of SCC; or for selecting or disqualifying a subject having a SCC for enrolment in a clinical trial for the treatment of SCC.

[0124] Indeed, if the subject is identified, using a method according to the present invention, as resistant to a particular treatment of cancer, this means that said treatment will not be efficient in the subject and will in addition possibly generate unwanted deleterious side effects in the subject. In such circumstances, the method advantageously further comprises an additional step of selecting a distinct or complementary cancer treatment, for example a distinct chemotherapy or immunotherapy, typically involving a “compensatory molecule” to be used alone or in combination with the originally preselected therapeutic drug(s) or with (a) distinct therapeutic drug(s), as the appropriate therapeutic treatment of cancer for the subject. On the contrary, if the subject is identified, using a method according to the present invention, as sensitive to the proposed particular treatment of cancer, this means that said treatment is an appropriate option for the subject.

[0125] Also herein described is a method of selecting or disqualifying a subject having a SCC for inclusion in a clinical trial, the clinical trial being for evaluating a SCC treatment, which method comprises a step of predicting or assessing the sensitivity of a subject having a SCC using a method according to the present invention comprising a step of assessing the multinucleated giant cells (MGC) status of a SCC tumor of the subject.

[0126] Although the herein described methods can in principle be performed in vivo, ex vivo and in vitro, it is particularly envisaged that they are performed in vitro.

[0127] In a preferred embodiment, the herein described method including a step of assessing the MGC status of a SCC tumor, in particular of a HNSCC tumor, more particularly a oral SCC (for example a SCC of the tongue or a SCC of the floor of the mouth), a pharynx SCC or a larynx SCC, of a subject (for example for evaluating the prognosis or the response to a therapeutic treatment of a subject having a SCC) is a partially or fully computer-implemented method.

[0128] The term “computer-implemented method” refers to a method which involves a programmable apparatus / device, in particular a computer, computer network, or readable medium carrying a computer program, in which at least one step of the method is performed by using at least one computer program. A computer-implemented method may further comprise at least one step that is not performed by using a computer program.

[0129] Inventors have developed a model (also herein identified by the term “classifier”) to automatically determine the MGC status of a SCC tumor of a subject, in particular of a HNSCC, more particularly a oral SCC (for example a SCC of the tongue or a SCC of the floor of the mouth), of a pharynx SCC or of a larynx SCC, thereby providing a ready-to-use tool for clinicians.

[0130] Inventors designed in particular a deep learning model which is described in the experimental part in as much technical details as possible. This model can automatically and reproducibly detect MGC on pathological whole slide image (WSI) stained by standard coloration.

[0131] Herein described in particular is a partially or fully computer-implemented method of the invention involving a step of determining the ratio of the number of MGC per mm2of tumor area or per number of cancer cells wherein the number(s) of MGC and / or of cancer cells, or the tumor area, is(are) obtained from image(s) of the SCC tumor. This computer implemented method preferably uses a classifier trained to assess the “MGC status” of a tumor, as herein below described.

[0132] In a particular computer-implemented method of training a classifier for (accurately, i.e. efficiently) assessing the MGC status of a SCC tumor, at least one classifier is trained with (a set of) images wherein MGC, or both MGC and tumor cells, are annotated, in order for the classifier to provide a “MGC-tumor” ratio of the number of MGC per the number of cancer cells or per mm2of tumor area, thereby assessing the MGC status and allowing in particular to evaluate the prognosis or the response to a therapeutic treatment of a subject having a SCC.

[0133] In a particular aspect, at least two distinct classifiers are trained, a first classifier being trained for detecting (and preferably counting) MGC in a SCC tumor sample, and a second classifier being trained for detecting (and preferably counting) tumor cells in a SCC tumor sample and / or for measuring the SCC tumor area.

[0134] As used herein, the term “classifier” refers to an algorithm that implements (i.e., predict) classification, i.e. that can determine a likelihood score or a probability that an object, here a cell, classifies within a group of objects (e.g., a MGC or a tumor cell) as opposed to one or several other groups of objects (e.g., for the MGC, a distinct type of macrophage such as a TREM2-expressing or TREM2Hlghmononuclear macrophage, and for the tumor cell, a healthy cell of the same origin), and that maps said input object to a category (e.g. regarding MGC, a MGCHlgh, MGCmtor MGCLowstatus, and regarding tumor cell, a “diseased” status or “healthy” status).

[0135] The term “classifier” may refer to one or multiple classifiers. For example, multiple classifiers may be trained, which may process data in parallel and / or as a pipeline. For example, output of one type of classifier (e.g., from intermediate layers of a neural network) may be fed as input into another type of classifier.

[0136] The classifier is preferably a detection model that predicts a set of information in relation with the detected object, in particular that predicts both i) the classification of the detected object (as herein above explained) and ii) the coordinates of the detected object. Such a detection model preferably contains a classification module and a “contour” generation module (cf. “bounding boxes” described in the experimental part).

[0137] Examples of classifiers that can be used in the context of the present invention include for example, but are not limited to, neural networks of various architectures (e.g., artificial, deep, convolutional, fully connected) and supervised machine learning classifiers such as Support Vector Machine (SVM) classifier, random forest classifier, decision tree classifier, K-nearest neighbor classifier (KNN), logistic regression classifier, nearest neighbor classifier, Gaussian mixture model (GMM), nearest centroid classifier and linear regression classifier. It is not an exhaustive list and the skilled person in the art will be able to identify similar algorithms that can be equally used, although they are not specifically mentioned here. Details and rules of functioning of the mentioned algorithms have already been widely described in the literature. An important contribution is the (training) set of input data provided to the classifier (i.e., a set of images of SCC tumor of known tissue origin comprising MGC and / or TREM2Hlghmacrophages in addition to tumor cells, or a set of images of SCC tumor of known surface area, or preprocessed information obtained from a set of such images). Based on this input data, it is possible to create a suitable model using any appropriate supervised or unsupervised machine learning techniques. The selection of appropriate algorithms is therefore of secondary nature and can be carried out in many different ways and in various combinations obvious to those skilled in the art. The FCOS convolutional neural network was for example used by inventors to detect MGC as well as the nuclei of tumor cells, and a Pixel-classifier based on a Random Forest Model was used to detect TREM2 expressing mononuclear macrophages and MGC. Preferably, the classifier is selected from random forest (RF) classifier, Support Vector Machine (SVM) classifier, decision tree classifier, K-nearest neighbor classifier (KNN), logistic regression classifier, nearest neighbor classifier, Gaussian mixture model (GMM) classifier, nearest centroid classifier, linear regression classifier, a neural network such as an artificial, deep, convolutional or fully connected neural network, and a transformer model (deep learning architecture). More preferably, the classifier is selected from Support Vector Machine (SVM) classifier, random forest (RF) classifier, neural network, in particular convolutional neural network (CNN), and transformer model. Even more preferably, the classifier is a convolutional neural network (CNN) or a transformer model.

[0138] A classifier utilizes some training data to understand how given input objects belong to a category / class or another. The classifier may be provided with a training set of annotated images of biological samples, preferably tumor samples, from subjects suffering of a SCC, preferably from subjects suffering of a SCC of known tissue origin, the origin being any origin as herein above identified. In a particular aspect, the images used to prepare the training set are obtained from a SCC cancerous tumor, the cancer being for example selected from head and neck squamous cell carcinoma (HNSCC), lung carcinoma, esophagus carcinoma, skin carcinoma and uterus cervix carcinoma. In a particular embodiment, the images used to prepare the training set are obtained from a HNSCC cancerous tumor, more particularly an oral SCC (for example a SCC of the tongue, a SCC of the floor of the mouth, a gum SCC or a cheek mucosa SCC), a pharynx SCC, a larynx SCC or a mixture thereof. The use of images from different SCC sources is preferred to prepare the training set in order to avoid obtaining a classifier that is too specific of a very particular SCC and thus incapable of detecting a distinct one (for example incapable of detecting a SCC of the floor of the mouth when considering the gum SCC as the “very particular SCC”).

[0139] In a preferred embodiment, in the training data set of images provided as an input to the classifier to be trained, each image is (manually) annotated, i.e., validated, by a skilled human person (typically a pathologist) by means of a known methodological independent approach such as a human eye-based method (“manual” or “pathologist-only approach” methodology), possibly with the help of a tool such as a software for annotating images (Da Qian, et al. “DigestPath: A benchmark dataset with challenge review for the pathological detection and segmentation of digestive-system.” Medical Image Analysis 80 (2022): 102485. Litjens, Geert, et al. “1399 H&E-stained sentinel lymph node sections of breast cancer patients: the CAMELY ON dataset.” GigaScience 7.6 (2018): giy065.), for example the open-source QuPath software (Bankhead, Peter, et al. “QuPath: Open source software for digital pathology image analysis.” Scientific reports 7.1 (2017): 7-7), the open-source Automated Slide Analysis Platform (ASAP) software (htps: / / github.com / computationalpathologygroup / ASAP). or the proprietary NDP.View2 software (https : / / www .hamamatsu . com / j p / en / product / life -science -and-medical- systems / digital-slide-scanner / U 12388-01.html) . This is to ensure for example that a particular biomarker is actually the expected biomarker, for example a particular cell identified as a MGC is actually a MGC and not a different macrophage or a completely different cell, and / or a cell identified as a tumor cell is actually a tumor cell and not a healthy cell.

[0140] Indeed, in order to distinguish SCC tumor cells of a particular origin from SCC tumor cells of a different origin, a particular classifier must be trained with annotated images (whose cells are correctly identified) obtained from SCC tumors of the desired origins so that said classifier learns to distinguish said cells.

[0141] Similarly, if the classifier is to be trained for measuring the SCC tumor area, images from the training data set are prepared (i.e., annotated) so that the classifier learns to correctly distinguish a tumor area from a healthy area.

[0142] Alternatively, the classifier may be provided with preprocessed information obtained from such training sets of images.

[0143] The accuracy ( / performance) of the classifier may be assessed using any method known by the skilled person. In particular, the classifier’s performance may be assessed by calculating for each of the biomarker of interest (for example MGC, TREM2Hlghmacrophage or tumor cell), a threshold or cut-off value. In a particular aspect, this value may be defined as 0.5. Another cutoff value may be easily obtained by the skilled person who knows how to select an appropriate performance measure, for example best accuracy or Fl-score, and then a relevant value, among a range of cutoffs preferably between 0 and 1, for a (particular) method of the invention trained for example for a particular SCC or for a particular population of subjects suffering from a SCC, in particular a population who is treatment-naive or a population who has received a particular neoadjuvant SCC therapeutic treatment.

[0144] The skilled person knows how to use any additional distinct performance measure selected for example from accuracy, precision, recall, mean Average Precision (mAP), area under the ROC curve, or Fl -score, and can easily determine an appropriate cut-off or threshold for the selected performance measure.

[0145] Also herein described is a computer-implemented method of training a classifier for (accurately) assessing or determining the MGC status of a SCC tumor. A particular method comprises the following steps of: a) providing a training set of tumor images, each tumor being obtained from a subject suffering from a SCC, or preprocessed information obtained from said training set, as input to a classifier, said training set comprising i) images of MGCHlghtumors, obtained from subjects suffering from a SCC known as having a MGCHlghstatus, ii) images of MGCLowtumors, obtained from subjects suffering from a SCC known as having a MGCLowstatus, and optionally iii) images of MGCInttumors, obtained from subjects suffering from a SCC known as having a MGCIntstatus; b) generating an output of the classifier for each image, said output classifying the tumor image input as having a MGCHlghor MGCLowstatus, or as having a MGCHlgh, MGCIntor MGCLowstatus; and c) evaluating the classifier’s performance for distinguishing between MGCHlghand MGCLowstatus, or between MGCHlgh, GClnand MGCLowstatus, by comparing, for each image, the output of the classifier to the known actual status of the tumor; wherein the classifier is considered as a trained performing classifier to determine the MGC status of a tumor, if it exhibits an Area Under the ROC Curve (ROC AUC) on tumor images from a testing set above 0.5, preferably above 0.6, in particular above 0.65.

[0146] Each tumor image from the testing set of images has been obtained from a subject suffering from a SCC. As further explained herein below, the testing set (similarly to the training set) comprises i) images of MGCHlghtumors, obtained from subjects suffering from a SCC known as having a MGCHlghstatus, ii) images of MGCLowtumors, obtained from subjects suffering from a SCC known as having a MGCLowstatus, and optionally iii) images of MGC1"1tumors, obtained from subjects suffering from a SCC known as having a MGClmstatus.

[0147] A particular system (also herein identified as a “detection classifier” or “detection model”) usable in the context of the invention herein described comprises several, for example at least two, distinct modules, a first classifier module for use for detecting cancer cells, a second classifier module for use for detecting MGC (or TREM2HIGHmacrophages). The system then calculates a ratio for the studied SCC sample, said ratio consisting in the number of MGC (or TREM2HIGHmacrophages) detected by the second classifier module per the number of cancer cells detected by the first classifier module.

[0148] For example, if 100 cancer cells are detected by the first module and 10 MGC are detected by the second module, the ratio is 10 / 100 = 0.1. Said ratio is then to be compared to a reference ratio (reference threshold) in order to assess the MGC status of the tested SCC sample in a method according to the invention for evaluating the prognosis or the response to a therapeutic treatment. A particular computer-implemented method of training such a (detection-based) classifier for (accurately / efficiently) assessing or determining the MGC status of a SCC tumor, comprises the following steps: a)

[0149] - of providing a training set comprising images of healthy tissue and / or images of SCC tumor, preferably images of healthy tissue and images of SCC tissue, or preprocessed information obtained from said training set, as input to a first classifier module (for use for detecting cancer cells), and / or

[0150] - of providing a training set comprising images of healthy tissue and / or images of SCC tumor, or preprocessed information obtained from said training set, as input to a second classifier module (for use for detecting MGC and / or TREM2HIGHmacrophages); b) of generating

[0151] - an output of the first classifier for each image, said output consisting in an image wherein cells identified by the first classifier as tumor cells are annotated or labelled, i.e., are made visible (or in other words are made distinguishable from healthy cells in particular (and preferably also from other structures that are not cells), for example surrounded by a bounding box, and identified as tumor cells; and / or

[0152] - an output of the second classifier for each image, said output consisting in an image wherein cells identified by the second classifier as MGC (or as TREM2HIGHmacrophages) are annotated or labelled, i.e., are made visible (or in other words are made distinguishable from cells of a different nature and preferably also from other structures that are not cells), for example surrounded by a bounding box, and identified as a MGC (or TREM2HIGHmacrophage); and c) of evaluating

[0153] - the performance of the first classifier for identifying ( / detecting) tumor cells by comparing, for each image, the output of the classifier to the known actual identity of the cells (established by a pathologist), and / or

[0154] - the performance of the second classifier for identifying ( / detecting) MGC (or TREM2HIGHmacrophages) by comparing, for each image, the output of the classifier to the known actual identity of the cells (established by a pathologist); wherein the first trained classifier is considered as efficient for identifying tumor cells if it exhibits for example a mean Average Precision (mAP) on tumor images from a testing set above 0, preferably above 0.2, in particular above 0.5, and wherein the second trained classifier is considered as efficient for identifying MGC (or TREM2HIGHmacrophages) if it exhibits for example a mean Average Precision (mAP) on tumor images from a testing set above 0, preferably above 0.2, in particular above 0.5.

[0155] The testing set (similarly to the training set) used in relation with the first classifier comprises i) images of tumors or tumor cells obtained from subjects suffering from a SCC, optionally ii) images of healthy tissue or healthy cells obtained from subjects suffering from a SCC and / or from healthy subjects, and optionally iii) images of other structures that are not cells obtained from subjects suffering from a SCC and / or from healthy subjects (the nature of said cells and structures being known in order to allow the test, i.e., the evaluation of performance).

[0156] The testing set (similarly to the training set) used in relation with the second classifier comprises i) images of MGC (and / or TREM2Hlghmacrophages), obtained from subjects suffering from a SCC, and ii) images of cells of a different nature, obtained from subjects suffering from a SCC and / or from healthy subjects [the MGC cells (and / or TREM2Hlghmacrophages) being identified as such in order to allow the test, i.e., the evaluation of performance].

[0157] This method may further comprise the following additional steps: d) of generating an output of the detection classifier for each SCC tumor image, said output classifying the SCC tumor image input as having a MGCHlgh, MGClm. or MGCLowstatus; and e) of evaluating the detection classifier’s performance for distinguishing between MGCHlghand MGCLowstatus, or between MGCHlgh, MGClmand MGCLowstatus, by comparing, for each SCC tumor image, the output of the classifier to the known actual status of the tumor; wherein the classifier is considered as a trained efficient classifier to determine the MGC status of a tumor, for example if it exhibits an Area Under the ROC Curve (ROC AUC) on tumor images from a testing set above 0.65.

[0158] Again, each tumor image from the testing set of tumor images has been obtained from a subject suffering from a SCC and the testing set (similarly to the training set) comprises i) images of MGCHlghtumors, obtained from subjects suffering from a SCC known as having a MGCHlghstatus, ii) images of MGCLowtumors, obtained from subjects suffering from a SCC known as having a MGCLowstatus, and optionally iii) images of MGC1"1tumors, obtained from subjects suffering from a SCC known as having a MGCIntstatus.

[0159] Thus, the evaluation of the performance of any of the herein above described trained classifiers is preferably checked with a testing set (also herein identified as “test set”) comprising annotated images (i.e., images comprising (annotated) tumor cells, MGC, TREM2HIGHmacrophages, or any biomarkers the identity of which has been validated by a skilled person, i.e., by a pathologist) distinct from the images of the set used to train the classifier (the “training set”), the detection (encompassing identification and classification) of the cells and biomarkers, or determination of the MGC status, derived from each image of the test set being obtained and processed using the same method as that used to obtain and process the biomarkers and MGC status of each image with the training set.

[0160] The performance may be further evaluated by confronting the MGC status determined for a particular SCC sample by a method of the invention to new data (for example images from a test set) or to the actual health data of the subject from whom the considered SCC sample came. Such health data provide valuable information. For example, they indicate the relapse of the patient or his survival in the short, medium or long term, or allow the production of Kaplan- Meier curves for qualitative analysis and / or quantitative measurements via log-rank tests.

[0161] In a particular aspect, any one of the herein above described method of the invention comprising a step of assessing the MGC status of a SCC tumor of a subject uses a classifier trained to assess the MGC status of a tumor with a training method according to the invention as herein above described.

[0162] Any partially or fully computer-implemented method herein described may contain additional step(s) selected from:

[0163] - determining a “MGC or TREM2Hlghmacrophages-tumor” ratio defined as i) the number of MGC, or of TREM2Hlghmacrophages, per mm2of SCC tumor area, or ii) the number of MGC, or of TREM2Hlghmacrophages, per the number of cancer cells, in a SCC tumor sample,

[0164] - detecting or measuring the expression of the CHIT1 , FBP1 and / or TREM2 gene(s) in a SCC tumor sample, and / or

[0165] - detecting or measuring the secretion of a CHIT1, FBP1 and / or TREM2 protein(s) in a blood sample or in a SCC tumor sample of the subject, in order to confirm the status assessed with the computer-implemented method.

[0166] In a particular embodiment, the invention concerns an in vitro method of determining the MGC status of a SCC tumor, wherein the method comprises the steps of:

[0167] (i) providing image(s) from the SCC tumor of a subject, or preprocessed information obtained from said image(s), as an input to a classifier trained to distinguish between a MGCHlghand MGCLowstatus, optionally between a MGCHlgh, a MGCmtand a MGCLowstatus, and

[0168] (ii) using the classifier to identify the MGC status of the SCC tumor of the subject as a MGCHlghstatus if the ratio of the number of MGC per mm2of tumor area or per the number of cancer cells is equal to or above a reference ratio, as a MGCLowstatus if the ratio of the number of MGC per mm2of tumor area or per the number of cancer cells is below a reference ratio, and as a GClnstatus if the ratio of the number of MGC per mm2of tumor area or per the number of cancer cells is in between, as an output of the classifier.

[0169] The method of determining the MGC status of a subject’s SCC tumor according to the invention may be performed once or several time during a subject’s lifetime. Thus, it is possible to monitor the evolution of the MGC status.

[0170] Also herein described are new predictive tools to assess SCC patient’s prognostic, to predict response to SCC treatment or identify patients who can respond and benefit from a SCC treatment, to monitor the effects of a SCC treatment, or to select an appropriate SCC treatment, for example.

[0171] Herein described in particular is a computing system comprising:

[0172] - a memory storing at least one instruction of a classifier trained preferably according to a method of the invention, or a system (“detection classifier”) according to the invention, and

[0173] - a processor accessing to the memory for reading said instruction(s), or to the system, and executing the method according to the invention.

[0174] A particular kit of the invention comprises i) a memory storing at least one instruction of a classifier trained according to a method of the invention on a support, ii) one or several detection means, for example a detection means that specifically recognizes a MGC or TREM2Hlghmacrophage, or that specifically recognizes a CHIT1, FBP1 or TREM2 gene or protein, possibly in suitable containers means, and, optionally, iii) a leaflet providing one or several reference cell number or density, reference ratio, reference percentage (%), reference proportion, and / or reference concentration.

[0175] The detection means (element ii)) of the kit is for example an agent that allows visualization of one of the herein described biomarkers such as for example a MGC, a TREM2Hlghmacrophage or a tumor cells. Such an agent is typically a reagent standardly used for colorimetric or enzymatic assays.

[0176] The detection means may otherwise be an agent capable of specifically recognizing (i.e., linking or binding) one of the herein described biomarkers such as for example a gene selected from CHIT1 , TREM2 and FBP1, a protein encoded by one or several of said genes, or a peptide or protein expressed by a biomarker cell as herein identified such as for example a MGC, TREM2Hlghmacrophage or tumor cell. The MGC detection means may be for example a TREM2 antibody as herein described (capable of detecting the TREM2 protein expressed by MGC).

[0177] The tumor cell detection means may be selected for example a cytokeratin such as for example a CK5 or CK6 antibody.

[0178] The detection means allowing the detection of a biomarker gene is typically a nucleic acid probe specific to the considered biomarker gene.

[0179] The detection means allowing the detection of a biomarker protein is typically an antibody, in particular an antibody as herein described.

[0180] In a particular aspect, the kit comprises at least one additional agent specifically recognizing ( / capable of detecting and binding) one of the following genes: MARCO, DCSTAMP, TYROBP, CHI3L1,MMP9, CTSS, CTSZ, CTSD, CTSB, CD68,APOE, SPP1 and OSCAR or its expression product.

[0181] In further aspects, the binding agent is labelled or a detection agent is included in the kit. It is contemplated that the kit may include one, at least one or several, biomarker binding agents attached to a non-reacting solid support, such as a tissue culture dish or a plate with multiple wells. It is further contemplated that such a kit includes one or several detectable agents. The kit may also comprise a positive control or several positive controls that can be used to determine whether a particular detection means is capable of specifically recognizing its corresponding biomarker.

[0182] In some aspects, the invention concerns kits for carrying out a method of the invention comprising, in suitable container means: (a) agent(s) that specifically recognizes all or part of a given biomarker; and, (b) at least one positive control, for example at least two positive controls, that can be used to determine whether the agent is capable of specifically recognizing all or part of said given biomarker. The kit may also include other reagents that allow visualization or other detection of anyone of the herein described biomarkers, such as reagents for colorimetric or enzymatic assays.

[0183] Also herein described are uses of a herein described kit, in particular for analyzing the MGC status of a SCC. The kit may be used to implement any methods herein described by inventors, for example for evaluating the prognosis of a subject having a SCC or for evaluating, predicting, assessing or monitoring the response (sensitivity or resistance) of a subject having a SCC to a particular therapy. The invention will be further illustrated by the following figures and examples. However, these examples and figures should not be interpreted in any way as limiting the scope of the present invention.

[0184] FIGURES

[0185] Figure 1. Multinucleated Giant Cells density is a biomarker of prognosis in patients with HNSCC.

[0186] A: therapeutic sequence of the HNSCC TCGA and GR treatment-naive cohorts. B: a nonkeratinizing HNSCC (left panel) and a keratinizing HNSCC (right panel), stained by H&E. Whole slide images (WSI) from TCGA (scale bar 250 pm). Histograms showing the number of KLowand KHlghpatients. C: keratinizing SCC of the oral cavity from a patient in the GR cohort, with MGC-rich granuloma surrounding keratin debris. Left panel: low magnification of carcinoma section (scale bar 2 mm). Middle panel: high magnification of a granuloma containing MGC and keratin (scale bar 50 pm). Right panel: MGC are highlighted in yellow and keratin in red. D: histograms showing the proportion of MGChlghpatients (TCGA and IGR cohorts) among no, low, moderate, and high carcinoma keratinization groups. E: flow chart of TCGA cohort of treatment-naive patients. Among the 527 patients with HNSCC, 110 were included. F: OS curve of TCGA patients stratified according to MGC density in their tumors. The vertical tick mark on the curves means that a patient was censored at this time. G: PFI curve of TCGA patients stratified according to MGC density in their tumors. H: flow chart of the GR cohort of treatment-naive patients. Among the 419 patients with HNSCC, 284 were included. I: OS curve of GR patients stratified according to MGC density in their tumors. J: PFI curve of GR patients stratified according to MGC density in their tumors. K: flow chart of the total GR and TCGA cohorts (n = 394). L: OS curve of all patients stratified according to MGC density in their tumors. M: PFI curve of all patients stratified according to MGC density in their tumors.

[0187] Figure 2. High MGC density after preoperative chemotherapy is associated with improved outcomes in patients with HNSCC.

[0188] A: therapeutic sequence of the induction chemotherapy (ICT) treated cohort of patients with HNSCC treated at GR. B: flow chart of the GR cohort of patients treated with ICT. C: keratinizing oral SCC from a patient treated at GR whose tumor responded well to ICT, showing areas of keratin surrounded by numerous MGC. Left panel: low magnification of carcinoma section (scale bar 2 mm). Middle panel: high magnification of a granuloma containing MGC and keratin (scale bar 50 pm). Right panel: MGC are highlighted in yellow and keratin in red. D: histogram showing the % of patients with no residual tumor, and whose surgical resected tumors were MGCHlghand MGCInt / Low, in both treatment-naive and ICT patients. E: OS curve of patients treated by ICT at GR, stratified by their tumor content and MGC density on surgical resection. The vertical tick mark on the curves means that a patient was censored at this time. F : histogram showing the proportion of MGCHlghand MGCLow / Intpatients according to the pathological response status of their tumors (poor vs. good / partial). G: left panel: low magnification of an oral SCC with a high tumor content (poor response to ICT) and no MGC (scale bar 2 mm). Right panel: high magnification on viable carcinomatous cells (scale bar 50 pm). H: left panel: low magnification of an oral SCC with a low tumor content (partial response to ICT) and numerous MGC (scale bar 2 mm). Right panel: high magnification of a granuloma with keratin and MGC surrounded by fibrosis (scale bar 50 pm).

[0189] Figure 3. Automatic detection of MGC and tumor cells on H&E / HES whole slide image by deep learning.

[0190] A: overview of the methodology. Manual annotations of MGC and tumor cells were used to train two cell-detection models to identify and count MGC and carcinomatous cells in SCC. An automatic score was computed as the ratio of MGC to tumor cells. The score was applied first to TCGA and GR oral cavity cohorts and second to TCGA CESC (also herein identified as CESCC) and larynx cohorts. B: keratinizing oral SCC with granulomas, from the TCGA cohort. Left panel: low magnification of the carcinoma (scale bar 2 mm). Middle panel: high magnification of the carcinoma where automatically detected carcinomatous cells are highlighted in red (scale bar 50 pm). Right panel: high magnification of the carcinoma where automatically detected MGC are highlighted in yellow (scale bar 50 pm). C: correlation between the number of MGC per patient quantified manually by a pathologist and automatically detected by the model. Patients were from the oral cavity TCGA cohort (n = 110 patients). D: Correlation between the number of MGC per patient quantified manually by a pathologist and automatically detected by the model. Patients are from the oral cavity GR cohort (n = 231 patients). E: OS curve of oral cavity TCGA and GR patients (n = 341 patients) stratified according to MGC density by the cell-detection model. The vertical tick mark on the curves means that a patient was censored at this time. F : PFI curve of oral cavity TCGA and GR patients (n = 341 patients) stratified according to MGC density by the cell-detection model. G: OS curve of uterine cervix SCC TCGA patients (n = 189 patients) stratified according to MGC density by the cell-detection model. H: OS curve of larynx TCGA patients (n = 100 patients) stratified according to MGC density by the cell-detection model.

[0191] Figure 4. Spatial transcriptomics reveal a unique MGC signature.

[0192] A: low magnification of a representative HES section of a tumor selected for spatial transcriptomic analysis. Inset: high magnification of an MGC. B: same HES section with the overlay of the spots analyzed by Visium technology. One spot is covering a single MGC. C: low magnification of a representative HES of an MGCHlghtumor from a patient in the GR cohort. D: overlay of the seven cell populations analyzed by unsupervised clustering. E: same HES section showing pathologist annotations of the tumor area (blue) and the MGC (yellow). F : low magnification of a representative HES of an MGCLowtumor from a patient in the GR cohort. G: overlay of the seven cell populations analyzed by unsupervised clustering. H: same HES section showing pathologist annotations of the tumor area (blue). I: projection of the Visium spots onto a UMAP space; cells from nine different patients. J: histograms showing the number of spots capturing the different cell types in MGCHlghand MGCLowcarcinomas. K: Volcano plot showing the MGC RNA signature extracted from the DEG analysis between supervised MGC spots and all the other non-MGC spots. L: Volcano plot showing the DEG between MGCHlghand MGCLowtumors from patients in TCGA cohort. M: Venn diagram between supervised MGC signature (GR) and bulk RNA analysis of MGCHlghtumors (TCGA), highlighting 12 common genes.

[0193] Figure 5. TREM2-expressing mononuclear macrophages and MGC cluster together in keratin-rich carcinoma niches.

[0194] A: representative low magnification image of a keratin-centered granuloma in an MGCHlghoral cavity SCC, stained by DAPI (blue), CHIT1 (yellow), CD163 (red), CD68 (cyan) and TREM2 (green). The merged image is shown in the upper left panel (scale bar 50 pm). B: representative high magnification image of a small granuloma. The arrow indicates mononuclear macrophages (scale bar 20 pm). C: histogram showing the ratio of TREM2Hlghmononuclear macrophages over the tumor surface between MGCHlghand MGCLowtumors (n = 9 patients). D: correlation between the ratio of TREM2Hlghmononuclear macrophages over the tumor surface and the ratio of TREM2HlghMGC over the tumor surface (n = 9 patients). E: histograms showing the distance of TREM2HlghMGC from keratin. F: histograms showing the distance of TREM2Hlghmononuclear macrophages from keratin.

[0195] Figure 6. TREM2-expressing mononuclear macrophages share a similar transcriptional program with MGC and are associated with good response to neoadjuvant immunotherapy.

[0196] A: integration of four scRNAseq HNSCC datasets into a UMAP space, showing ten distinct clusters of monocytes and macrophages (n = 74 patients and 23,635 cells). B: mapping of the Visium MGC spots on the HNSCC UMAP space. C: histogram showing the repartition of MGC spots into the different macrophage clusters; all MGC spots are in the TREM2Hlghmacrophage cluster. D: therapeutic sequence of the HNSCC cohort treated by neo-adjuvant immunotherapy (Luoma A.M. et al). E: histogram showing the proportion of macrophage subsets in tumors that responded poorly or partially / well after neo-adjuvant immunotherapy. F : histogram showing the ratio of macrophage subsets in tumors that responded well / partially over those that responded poorly. G: histogram showing the proportion of TREM2Hlghmacrophages between tumors that responded well / partially and poorly. H: before-after plot showing the increase in the proportion of TREM2HlghTAM in tumors that responded well / partially versus poorly to neo-adjuvant immunotherapy (ICB = Immune Checkpoint Blockade). The before category corresponds to biopsy and the after category corresponds to surgical resection after neo-adjuvant immunotherapy.

[0197] Figure 7 (Extended Data related to Figure 1).

[0198] A: correlation between the degree of keratinization (%) and the number of MGC / mm2per patient, in TCGA and GR cohorts (n = 394 patients). B: histograms showing the proportion of MGC according to the level of keratinization of the carcinoma (TCGA and GR cohorts). C: flow chart of the treatment-naive oral cavity SCC cohort (TCGA and GR) where patients are stratified according to their degree of keratinization (KeratinHlghand KeratinLow) and their MGC / mm2content. D: OS curve of GR and TCGA patients with oral cavity SCC stratified into KHlghMGCHlgh, KLowMGCLow / Intand KHlghMGCLow / Int. The vertical tick mark on the curves means that a patient was censored at this time. E: PFI curve of GR and TCGA patients with oral cavity SCC stratified into KHlghMGCHlgh, KLowMGCLow / Intand KHlghMGCLow / Int. F : correlation between MGC / mm2quantified by two pathologists on 30 WSI (20 from GR and 10 from TCGA cohort). Figure 8 (Extended Data related to Figure 3).

[0199] A: flow chart of the digital slides cohort of patients with oral cavity SCC from TCGA and GR. Among the 394 patients with oral cavity SCC (1414 slides), 341 were included (949 slides). B: correlation between MGC / mm2quantified by a pathologist and by the cell detection model in the TCGA oral cavity cohort (n = 110 patients). C: correlation between MGC / mm2quantified by a pathologist and by the cell detection model in the digital GR oral cavity cohort (n = 231 patients). D: box plot showing the time required to quantify the absolute number of MGC per digital slide in the TCGA CESC cohort, by a pathologist (red) and by an Al augmented pathologist (green). The speed reduction factor is shown. E: box plot showing the same approach in the TCGA laryngeal SCC cohort. F. Correlation between the number of MGC quantified by a pathologist and an Al augmented pathologist, in TCGA CESC. G: Similar correlation in TCGA larynx. H: PFI curve of TCGA CESC patients stratified in MGCHlghand MGCLow(n = 189 patients). The vertical tick mark on the curves means that a patient was censored at this time. I: PFI curve of TCGA laryngeal SCC patients stratified in MGCHlghand MGCLow(n = 100 patients).

[0200] Figure 9 (Extended Data related to Figure 4).

[0201] A: low magnification of a representative HES section of an SCC selected for spatial transcriptomic analysis. First Inset: high magnification of an MGC. Second inset: overlay of one selected spot that is covering a single MGC. B: mapping of the supervised Visium MGC spots (red) on the Visium UMAP space. The histogram is showing the repartition of selected MGC spots into the different clusters; the majority of the supervised MGC spots are in the unsupervised MGC cluster. C: OS curve of TCGA oral cavity patients stratified in CHIT1- TREM2Hlghand CHITl-TREM2Low. The vertical tick mark on the curves means that a patient was censored at this time. D: Visium spots with high CK5 RNA content (upper image) and immunohistochemistry staining with CK5 antibody on a serial slide (lower image). E: Visium spots with high CHIT1 RNA content (upper image) and immunohistochemistry staining with CHIT1 antibody on a serial slide (lower image). F: Visium spots with high CD68 RNA content (upper image) and immunohistochemistry staining with CD68 antibody on a serial slide (lower image).

[0202] Figure 10 (Extended Data related to Figure 5).

[0203] A: representative low magnification image of a granuloma made of multiple MGC in oral cavity SCC from GR cohort, stained by DAPI (white), TREM2 (red), CD68 (cyan) and CD44 (green). Individual stainings are shown in the small left panels. Merged image is shown in the center large panel. Scale bar 50 pm. B: histogram showing the ratio of TREM2HlghMGC over the tumor surface between MGCHlghand MGCLowtumors.

[0204] Throughout this application, various references describe the state of the art to which this invention pertains. The disclosures of these references are hereby incorporated by reference into the present disclosure.

[0205] Other characteristics and advantages of the invention are given in the following experimental section (with reference to figures 1 to 10), which should be regarded as illustrative and not limiting the scope of the present application.

[0206] EXPERIMENTAL PART

[0207] EXAMPLE 1 - Trem2-expressing multinucleated giant macrophages are a biomarker of good prognosis in head and neck squamous cell carcinoma (HNSCC).

[0208] MATERIAL AND METHODS

[0209] Patients

[0210] The Cancer Genome Atlas (TCGA) oral cavity cohorts. Data from 527 HNSCC patients were retrieved from the TCGA website. Exclusion criteria were patients with only frozen sections, low quality slides, no diagnostic slides, absence of surgical resection, only biopsies, HPV (Human Papilloma Virus) positive tumors or tumor localization elsewhere than oral cavity: this led to exclusion of 417 patients, leaving data from 110 patients eligible for further analyses (Table 1).

[0211] Tablet: Initial patient and tumor characteristics according to cohorts

[0212] GR TCGA TCGA+GR Induction

[0213] N=288 N=110 N=398 chemothe rapy

[0214] N=52

[0215] Sex

[0216] Male 204 (71%) 77 (70%) 281 (71%) 51 (98%)

[0217] Female 84 (29%) 33 (30%) 117 (29%) 1 (2%)

[0218] Age (years)

[0219] Median [range] 57 [19-89] 60 [19-87] 58 [19-89] 53 [20-69]

[0220] (IQR) (50-66) (50-66) (50-66) (44-63)

[0221] < 65 years 204 (71%) 78 (71%) 282 (71%) 45 (87%)

[0222] > 65 years 84 (29%) 32 (29%) 116 (29%) 7 (13%)

[0223] Tobacco consumption

[0224] Never 46 (16%) 53 (48%) 99 (25%) 14 (27%)

[0225] < 20 PY 46 (16%) 8 (7%) 54 (14%) 4 (8%)

[0226] > 20 PY 152 (53%) 49 (45%) 201 (51%) 34 (65%)

[0227] Smoker but unknown PY 44 (15%) 0 (0%) 44 (11%) 0 (%) Alcohol consumption

[0228] No 108 (37%) 34 (31%) 142 (36%) 27 (52%)

[0229] Yes 152 (53%) 72 (65%) 224 (56%) 25 (48%)

[0230] Unknown 28 (10%) 4 (4%) 32 (8%) 0 (0%)

[0231] Location

[0232] Floor 111 (39%) 0 111 (28%) 36 (69%)

[0233] Tongue 177 (61 %) 110 (100%) 287 (72%) 16 (31%)

[0234] Clinical stage

[0235] T1 54 (19%) 0 (0%)

[0236] T2 58 (20%) 3 (6%)

[0237] T3 61 (21%) 4 (8%)

[0238] T4 115 (40%) 37 (71%)

[0239] TX 0 8 (15%)

[0240] NO 195 (68%) 12 (23%)

[0241] N1 35 (12%) 2 (4%)

[0242] N2a 8 (3%) 3 (6%)

[0243] N2b 31 (11%) 13 (25%)

[0244] N2c 14 (5%) 9 (17%)

[0245] N3 3 (1%) 6 (12%)

[0246] NX 2 (1%) 7 (13%)

[0247] Stage

[0248] I 54 (19%) 6 (5%) 60 (15%) 0 (0%)

[0249] II 58 (20%) 30 (27%) 88 (22%) 2 (4%)

[0250] III 48 (17%) 29 (26%) 77 (19%) 2 (4%)

[0251] IV 128 (44%) 39 (35%) 167 (42%) 40 (77%)

[0252] Unknown 0 6 (5%) 6 (2%) 8 (15%)

[0253] Pathological stage (V7)

[0254] PTI 58 (20%) pT2 88 (31%) pT3 29 (10%) pT4a 112 (39%)

[0255] Missing 1 (<1%) pNO 144 (50%) pNl 48 (17%) pN2a 3 (1%) pN2b 52 (18%) pN2c 37 (13%) pN3 4 (1%)

[0256] Missing 0 pStage (V7)

[0257] I 45 (16%) 6 (6%) 51 (13%)

[0258] II 49 (17%) 21 (21%) 70 (18%)

[0259] III 44 (15%) 26 (26%) 70 (18%)

[0260] IV 149 (52%) 48 (43%) 197 (49%)

[0261] Unknown 1 (<1%) 9 (5%) 10 (3%)

[0262] Pathological stage (V8) pTl 45 (16%) pT2 54 (19%) pT3 75 (26%) pT4a 112 (39%)

[0263] Missing 2 (1%) pNO 142 (50%) pNl 38 (17%) pN2a 11 (1%) pN2b 20 (18%) pN2c 4 (13%) pN3b 73 (1%)

[0264] Missing 0 pStage (V8)

[0265] I 29 (10%)

[0266] II 40 (14%)

[0267] III 59 (20%)

[0268] IV 158 (55%)

[0269] Unknown 2 (1%) Vascular embol

[0270] No 214 (74%)

[0271] Yes 67 (23%)

[0272] Not evaluated 7 (2%)

[0273] Perineural invasion

[0274] No 134 (47%)

[0275] Yes 147 (51%)

[0276] Not evaluated 7 (2%)

[0277] MGC

[0278] Low (0 - < 0.2) 181 (63%) 81 (74%) 262 (66%) 16 (31%)

[0279] Intermediate (> 0.2 - < 1) 65 (23%) 18 (16%) 83 (21%) 6 (12%)

[0280] High (> l) 42 (15%) 11 (10%) 53 (13%) 21 (40%)

[0281] Fibrosis 9 (17%)

[0282] Keratin

[0283] Low (< 10%) 70 (24%) 32 (29%) 102 (26%)

[0284] High (> 10%) 218 (76%) 78 (71%) 296 (74%)

[0285] MGC low / int - Keratin 70 (24%) 32 (29%) 102 (26%) low

[0286] MGC low / int - Keratin 176 (61%) 67 (61%) 243 (61%) high

[0287] MGC high - Keratin low 0 (0%) 0 (0%) 0 (0%)

[0288] MGC high - Keratin high 42 (15%) 11 (10%) 53 (13%)

[0289] Resection

[0290] R0 225 (78%)

[0291] R1 62 (22%)

[0292] Missing 1 (<1%)

[0293] Nodal extra capsular extension pNO (V7) 144 (50%)

[0294] No 65 (23%)

[0295] Yes 78 (27%)

[0296] Missing 1 (<1%) Adjuvant treatment

[0297] None 134 (47%) 2 (4%)

[0298] RT 62 (22%) 13 (25%)

[0299] RT-CT 89 (31%) 37 (71%)

[0300] Curie 3 (1%) 0

[0301] These patients were considered “treatment-naive” because they did not receive pre-operative therapy. Inventors had access to the bulk RNAseq data for 108 of these patients.

[0302] TCGA larynx (HNSCC) and uterine cervix squamous cell carcinoma (CESC, also herein identified as CESCC) cohorts.

[0303] Data from 307 patients with uterine cervix carcinoma and from 527 patients with HNSCC were retrieved from the TCGA.

[0304] Exclusion criteria for uterine cervix were patients without squamous cell carcinoma, frozen sections, low quality slides and no diagnostic slides: this led to the exclusion of 118 patients, leaving data from 189 patients eligible for further analyses.

[0305] Exclusion criteria for larynx were frozen sections, low quality slides, no diagnostic slides and tumor localization elsewhere than larynx: this led to exclusion of 427 patients, leaving data from 100 patients eligible for further analyses.

[0306] GR (Gustave Roussy) cohorts. Inventors had access to samples from two previously published cohorts of patients with HNSCC, treated at GR (Marhic, A. et al. ; Moya-Plana, A. et al.). These patients did not receive pre-operative therapy and were thus considered as “treatment-naive”. Exclusion criteria were patients with not enough representative pathological diagnostic slides, low quality slides, or the presence of neoadjuvant therapy: out of 419 patients, 284 were included (cf. Table 1). Inventors further had access to a retrospective cohort of 52 patients with HNSCC treated by induction chemotherapy at GR. All Formalin-Fixed Paraffin-Embedded (FFPE) blocks and slides came from the pathology laboratory of GR. All patients gave an informed consent. For each patient, the study protocol was approved by the institutional committee of GR. All experiments were in accordance with the Declaration of Helsinki. The REporting recommendations for tumor MARKer prognostic studies (REMARK) were followed. Pathologist annotations and quantifications

[0307] A pathologist quantified the Multinucleated Giant cells (MGC) per mm2of tumor for each hematoxylin and eosin (H&E) / hematoxylin, eosin and saffron (HES) slide from each patient from TCGA (n = 110 slides) and GR cohorts (n = 1,304 slides). A second pathologist quantified the MGC per mm2of tumor for a subset of the patients (TCGA n = 10 slides; GR n = 20 slides): the counts from the two pathologists showed almost perfect agreement (r2= 0.9606, p-value < 0.0001 and weighted kappa = 0.876) (Fig. 7F).

[0308] MGC -related stratification was based on two thresholds for manual quantification: MGCHlgh(> 1 MGC / mm2), MGCInt(> 0.2 MGC / mm2and < 1 MGC / mm2) and MGCLow(< 0.2 MGC / mm2). These thresholds were determined with the maximally selected rank statistics.

[0309] The annotations were carried out with QuPath version 0.4.3, an open-source software for digital pathology images analysis (Bankhead, P. et al.).

[0310] Stratification of patients treated by induction chemotherapy, into good, partial, and poor responders, was made by a pathologist using the original grading system published by Braun et al. in 1989.

[0311] Survival curves

[0312] Overall survival (OS) and progression free interval (PFI) were the two clinical endpoints used for the analysis. Log-rank test and log-rank test for trend were performed. Multivariable analyses were performed using Cox model and p-value of Wald test was presented.

[0313] TCGA cohorts. Clinical data were retrieved from the GDC Data Portal (https: / / portal.gdc.cancer.gov / ). OS and PFI were retrieved from an integrated TCGA pancancer clinical data resource (Liu J et al.), in which the authors recommended using OS and PFI as clinical endpoints. PFI was defined as follows: “1 for patient having a new tumor event whether it was a progression of disease, local recurrence, distant metastasis, new primary tumors all sites, or died with the cancer without new tumor event, including cases with a new tumor event whose type is N / A. 0 for censored otherwise.” (Liu J et al.).

[0314] GR cohorts. Clinical data of treatment naive patients were retrieved from the two ancillary annotated cohorts (Marhic, A. etal. ; Moya-Plana, A. etal.). PFI was defined as follows: “events for progression or second HNSCC or death by HNSCC or death of unknown cause with HNSCC”. GR clinical data of induction chemotherapy patients were retrieved from the GR patient’s database. Digitization of HES pathological slides

[0315] Inventors retrieved the 1,304 physical microscopic slides from the two cohorts of treatment- naive patients from GR (Marhic, A. et al.-, Moya-Plana, A. et al.). Multiple slides per patient were available (ranging 1-17). While most of the slides had a standard microscopic size (75 x 25 mm), several of them were larger (75 x 50 mm) and did not physically fit into the digital slide scanner. Therefore, all large microscope slides were removed from the analysis and 44 patients with large slides only were excluded from the digital analysis. Quality control was performed by a pathologist to exclude slides with common artifacts such as tissue folds, blurring or faint staining, resulting in the exclusion of 9 additional patients. Overall, 839 slides were analyzed, representing tumors from 231 patients (of the original 284 included patients). Slides were digitized at 20X resolution, with an Olympus VS 120.

[0316] Slides from the TCGA cohort were already digitized and one slide per patient was quantified (n = 110 patients and n = 110 slides).

[0317] Training of deep learning models

[0318] Training data for MGC. To train the MGC detection model, 16 oral cavity whole slide images (WSI) from the TCGA cohort and 12 oral cavity WSI from the GR cohort were retrieved. A pathologist annotated a total of 5,000 MGC by drawing bounding boxes using QuPath software version 0.4.3. Rather than exhaustively annotating all MGC in a small number of slides, inventors opted to maximize the variability by non-exhaustively annotating some of the MGC on many slides. The annotation instructions were to draw bounding boxes around each MGC, and then to draw an englobing bounding box that ensured that all MGC were exhaustively annotated in the latter. As such, all pixels outside the englobing boxes were not used to train the detection model. Inventors randomly annotated additional negative patches, i.e., patches that contained no MGC, by drawing an englobing bounding box. All englobing bounding boxes were extracted with both tissue image and annotations using an in-house software at 20x, resulting in fields of view (FOV) with varying size and with exhaustively annotated MGC. All FOV of 20 WSI were randomly assigned to the training set, and the remainder FOV from 8 WSI to the validation set. All the training and validation slides from the GR cohort, that were used to train the MGC detection model, were discarded from the rest of the study.

[0319] Training data for tumor cells. The same protocol was applied to obtain ground-truth annotations for training the tumor detection model . In total, 814,606 tumor cells were collected from 15 WSI of TCGA tongue and 52 WSI of the GR cohort. Additionally, a pathologist exhaustively annotated the viable tumor cells in 107 WSI from the GR cohort, ensuring that there were not any tumor cells outside the annotated regions. Patches outside the annotated regions were added to the training data with an associated empty ground-truth since these patches do not contain tumor cells. The final training data contained 13,624 256x256 pixels patches with at least one tumor cell, and 106,000 patches without tumor cells. All FOV of 47 WSI were randomly assigned to the training set, and FOV from the remainder 20 WSI to the validation set. Here again, all the training and validation GR slides for the tumor detection model were discarded from the rest of the study.

[0320] Model architecture and hyper-parameters . Inventors trained the Fully Convolutional One-Stage (FCOS) (Tian Z. et al.) detection architecture for both tumor cells and MGC automatic annotation. Contrary to region proposal models, detections of FCOS are made at the pixel level: each pixel notably produces four values that are reconstructed as a predicted bounding box, and an additional probability value for each class - 1 in our case for both tumor and MGC. During training, each proposed bounding box is first matched with one ground-truth bounding box, or no bounding box. Then, a regression loss and a classification loss are computed based on the predicted box coordinates and its associated probability. Inventors used the ResNet50 architecture (He K. etal.) pre-trained on ImageNet (Deng J. et al.) as the backbone model. Both classification and regression heads comprised four ReLU-activated convolutional layers of kernel size 3 and stride 1, with batch normalization. Contrary to the original implementation of FCOS, image rescaling was discarded to input patches with the same magnification as the model. Data augmentation consisted of rotations, flips, shifts, and color jitter. For each input patch, the same data augmentation was performed to the associated ground-truth bounding boxes using the Albumentations (Buslaev A. et al.) python library version 1.2.1. The detection model parameters were stochastically updated using the Adam optimizer (Kingma D. P. et al.) from errors computed by the focal loss (Lin T.-Y. et al.) cost function with a learning rate of 10-4, a regularization of 10-4 and a batch size of four patches. Training was conducted for up to 2000 epochs on a NVIDIA A40 GPU. Inventors selected the model snapshot with the maximum mean average precision at intersection-over-union of 0.5 (mAP@50) on the validation set for both tumor and MGC tasks for the remainder of their study. All deep learning implementation was done in python 3.8, pytorch (Paszke A. et al. ) version 1.13 and torchvision version 0.14.0.

[0321] Cell detection model performances . Inventors assessed the performances of the tumor cell detection model. Briefly, ground truth tumor cells (annotated by a pathologist) were compared to automatically detected tumor cells and the average precision was computed using the MMdetection package version 3.1.0 (Chen K. et al.). To assess the performance of the MGC detection model, they compared the number of automatically detected MGC to the number of MGC that were quantified by a pathologist (Fig. 3C and 3D). Artificial Intelligence (Al)-based biomarker inference

[0322] Once both tumor detection and MGC detection models were trained, the biomarker was computed for each WSI as follows. Patches of width 2048 pixels were extracted from a WSI in a sliding window fashion with an overlap of 256 pixels for both sides. Inventors used a non-0 overlap to remove border artifacts (such as false positives) arising from a lack of context at the borders. Each patch was forwarded into both the tumor and MGC models, producing two lists of predicting bounding box coordinates and their predicted probabilities. Bounding boxes predicted with a probability below 0.4 were discarded and inventors applied non-maximum suppression with a threshold of 0.6 intersection-over-union to remove highly overlapping predictions. Inventors finally computed the proposed MGC biomarker as the ratio of the number of predicted MGC to the number of predicted tumor cells for each patient.

[0323] Spatial transcriptomic

[0324] Library preparation. The Visium Spatial Gene Expression Slide, Visium formalin-fixed paraffin-embedded (FFPE) Reagent kit, and Visium human transcriptome probe kit (10X Genomics) were used to generate sequencing libraries. RNA quality was accessed for all samples using the DV200 method. For library constructions, 5pm cryosections of 6.5 x 6.5 mm, from each FFPE sample were placed into the 4 capture areas of Spatial Transcriptomics slides (10X Genomics). Samples were then deparaffinized, stained, imagined, and decross-linked. Probe hybridization and ligation to RNA were then performed, followed by the single-stranded ligation product release and extension. Libraries were then prepared following the manufacturer's instructions. All the libraries were sequenced with a 10-base index read (dual index), a 28-base Readl containing cell identifying barcodes and unique molecular identifiers (UMIs), and a 90-base Read2 containing transcript sequences on an Illumina NovaSeq 6000.

[0325] Visium data analysis. All sections were processed individually using the Seurat package, with spots containing less than 500 UMI and 500 RNA counts were discarded from further analyses. Afterwards, all slides were merged into one object used for differential expressed genes (DEG) analysis, and integration was performed as described in the scRNAseq section. The Loupe software (10X Genomics®) was used to explore the data.

[0326] Identification of MGC signature. To identify the DEG between the MGC selected spots and non-MGC spots, inventors used the R package Seurat, and the volcano plot was plotted using the Enhanced Volcano R package. Only the genes with a log2 fold change (FC) over 0.25 and an adjusted p-value less than 0.05 were considered significant. TCGA bulk RNA sequencing data

[0327] Preprocessing of data. Bulk RNAseq data of primary tumors from 108 out of the 110 patients with tongue squamous cell carcinoma (SCC) were used for survival analysis. Data were retrieved using the TCGABiolinks R package, then normalized using the EDAseq package based on gene length; genes whose expression was zero in more than 25% of the samples were removed.

[0328] Differentially expressed genes (DEG). The 108 patients with available histological slides were split into three groups based on the presence of MGC as annotated by a pathologist. DEGs were calculated between groups using the R package DESeq2, and the volcano plot was plotted using the Enhanced Volcano Rpackage. Only genes with a log2 fold change over 0.25 and an adjusted p-value less than 0.05 were considered significant.

[0329] Gene set enrichment analysis. The R package clusterprofiler was used to analyze the DEGs between different groups for GO analyses. For Visium GO analysis, only genes with log2FC superior to 1.5 were considered.

[0330] Ven diagram analysis. The analysis was made in the interactivenn website (Heberle H. et al.). The top DEG whose adjusted p-value was lower than 0.05 and log2 fold change higher than 0.25 were extracted from the TCGA bulk RNAseq data and from the Visium MGC signature.

[0331] MoMac verse

[0332] The MoMac -Verse RDS object was retrieved from CellXGene website (htps : / / macro verse . ustaveroussy . fr / 2021 _MoMac_VERSE / ) (Mulder K. et al.). Then, multimodal reference mapping was used to map the MGC Visium single spots onto the MoMac - Verse UMAP. The normalization method was logarithmic, reference reduction was Principal Component Analysis (PCA) and 50 dimensions were used for FindTransferAnchors function (Seurat v4).

[0333] Multiplex immunofluorescence

[0334] Nine FFPE blocs of HNSCC (4 MGCHlghpatients and 5 MGCLowpatients) were stained by multiplex immunofluorescence. The experimental and translational pathology platform of GR performed the stainings using a preset routine protocol. Briefly, FFPE blocs were cut into 3 pm thick sections. Multiplex staining was performed on a Bond RX (Leica Biosystems). Slides were baked in high pH for 20 min at 100°C for epitope retrieval. Then, multiple rounds of staining were performed, each round included endogenous peroxidase blocking for 10 min, non-specific sites blocking for 5 min, primary antibody incubation, secondary HRP-labeled antibody incubation for 10 min and OPAL reactive fluorophore for 10 min that covalently label the primary epitope, and heat denaturation of antibodies. The sequence of the antibodies with the associated Opal dyes is as follow: 1) rabbit anti-TREM2 (clone D814C, Cell Signaling, reference 91068), 1 / 400 dilution, Ih, 37°C, with OPAL 520, 1 / 100 dilution; 2) rabbit anti- CHIT1 (polyclonal, Biorbyt, reference orb377995), 1 / 50 dilution, 30min, ambient temperature, with OPAL 570, 1 / 100 dilution; 3) mouse anti-CD68 (clone KPI, DAKO, reference M0814), 1 / 1000 dilution, Ih, ambient temperature, with OPAL480; 4) mouse anti-CD163 (clone 10D6, Diagnostic BioSystems, reference Mob 460-05), 1 / 600 dilution, 15min ambient temperature, with OPAL690, 1 / 100 dilution. Nuclei were visualized by a final incubation with the Spectral DAPI (Akoya Biosciences) for 5min. Slides were mounted with mounting medium for fluorescence. Finally, images were acquired on the Polaris 2 (Akoya Biosciences).

[0335] Immunofluorescence image analysis

[0336] A pathologist manually annotated the tumor, the keratin and the technical artifact regions on the nine immunofluorescence images, using QuPath vO.4.3 (Bankhead P. et al.). In addition, a small proportion of TREM2-expressing mononuclear macrophages and MGC were manually annotated by a pathologist. These annotations were used to train a pixel-classifier based on a random forest model. The classifier model was applied to predict the presence of TREM2- expressing MGC and mononuclear macrophages, in the tumor regions of the nine immunofluorescence images. Measurements of the areas and distances of these regions to the closest keratin-rich regions were exported and analyzed with a homemade R script47 to produce graphics of surface ratios and distances to keratin regions.

[0337] Single-cell RNA sequencing data analysis

[0338] Dataset availability. Previously published scRNAseq data have been deposited at Gene Expression Omnibus (GEO: GSE139324, GSE164690, GSE200996) and are publicly available (Cillo A. et al. Kurten C.H.L. et al., Song H. et al.,' Luoma A.M. et al ).

[0339] Preprocessing of data. Cells that expressed fewer than 500 counts or 200 UMI, had more than 7.5% mitochondrial reads, or had more than 6,000 unique molecular identifiers (UMIs), were filtered out. The RNA matrix was sc-transformed (normalization and variance stabilization), and a PCA was performed (Becht, E. et al ), from which the first 30 significant Principal Components were selected for UMAP analysis. Following the identification of major immune populations (T cells, MNPs, Mast Cells, B cells, NK Cells, etc.) using canonical markers, each population was extracted, and the preprocessing was carried out again. Integration of multiple datasets. Different scRNAseq experiments were integrated using the Seurat R package based on dataset origin using reciprocal PCA (rpca) based on 30 first PCA dimensions, with k.anchor = 20.

[0340] Annotations of immune populations . Clusters were annotated using the MoMac verse, and the FindTransferAnchors function of the Seurat R package using the 30 first PCA dimensions.

[0341] Differentially expressed genes (DEG). DEG analyses were performed using the Seurat v4 package. DEGs obtained from the “RNA” matrix of the Seurat object were calculated on normalized values with a log fold change threshold of 0.25 and a min.pct threshold of 0.25. The Wilcoxon-rank sum test was used.

[0342] Data availability

[0343] Digital slides. WSI were generated from publicly available cohorts (TCGA) and from institutional cohorts (GR). The digital slides from the TCGA can be retrieved from their website. The digital slides generated from Gustave Roussy cohorts cannot be made publicly available due to general data protection regulations and institutional guidelines.

[0344] Single-cell RNA sequencing data (scRNAseq). Previously published scRNAseq data have been deposited at Gene Expression Omnibus (GEO: GSE139324, GSE164690, GSE200996) and are publicly available (Cillo A. et al.,' Kurten C.H.L. et al., Song H. et al.,' Luoma A.M. et al ). Bulk RNA sequencing data. Data from the TCGA can be retrieved from their website.

[0345] Code availability

[0346] The source code is not publicly available due to proprietary and confidentiality constraints. Availability may be made upon reasonable request. All methods and software packages used in the context of inventors’ experiments have been documented and explained in ways accessible to the skilled person and to broader scientific audience.

[0347] RESULTS

[0348] MGC are a biomarker of good prognosis in patients with HNSCC.

[0349] Inventors first characterized the presence and abundance of MGC in tumors from patients with HNSCC who underwent primary surgery (Fig. 1A). They made use of cohorts from The Cancer Genome Atlas (TCGA) and Gustave Roussy (GR), consisting of 394 oral cavity patients, and observed that 292 bore keratinizing tumors, while the remaining 102 did not (Fig. IB). Within keratinizing tumors, they observed MGC that were aggregated in large granulomatous clusters and in close contact with extracellular keratin (Fig. 1C). Interestingly, none of the non- keratinizing tumors contained MGC, while 33% of keratinizing (“K”) tumors contained high (see Methods) densities of these cells (Fig. ID). Although the abundance of MGC was significantly associated with the keratinization level of the tumors, the presence of keratin did not guarantee the presence of MGC as 37% of patients with highly keratinizing SCC were devoid of MGC infiltration (Fig 7A, B).

[0350] Inventors then asked whether the density of MGC in these tumors was related to outcomes for patients. Using microscopic slides, routinely stained by hematoxylin, eosin (H&E) + / - saffron (HES), from a TCGA cohort of 110 patients with HNSCC (Fig. IE; Supplementary Table 1), inventors stratified patients into three groups (MGCHlgh, MGClmand MGCLow) according to the MGC density in their tumor. Inventors saw that higher densities of MGC were associated with longer overall survival (OS) (Fig. IF) and progression-free interval (PFI) (Fig. 1G). This observation was validated in two additional retrospective cohorts of patients with HNSCC treated at GR, with the same outcome (n = 284 patients) (Fig. 1H, I, J; Supplementary Table 1). Moreover, combining TCGA and GR cohorts (n = 110 + 284 = 394 patients) also confirmed that higher densities of MGC were associated with longer OS and PFI in a dose-dependent manner (Fig. IK, L, M; Supplementary Table 1).

[0351] Supplementary Table 1: Initial patient and tumor characteristics according to cohort

[0352] GR TCGA+GR Induction

[0353] N=284 N=394 chemotherapy

[0354] N=52

[0355] Sex

[0356] Male 202 (71%) 77 (70%) 279 (71%) 51 (98%)

[0357] Female 82 (29%) 33 (30%) 115 (29%) 1 (2%)

[0358] Age (years)

[0359] Median [range] 57 [19-89] 60 [19-87] 58 [19-89] 53 [20-69]

[0360] (IQR) (50-67) (50-66) (50-66) (44-63)

[0361] < 65 years 200 (70%) 78 (71%) 278 (71%) 45 (87%)

[0362] > 65 years 84 (30%) 32 (29%) 116 (29%) 7 (13%) Tobacco consumption

[0363] Never 44(15%) 53 (48%) 97 (25%) 14 (27%)

[0364] <20 PY 49 (17%) 8 (7%) 57 (14%) 4 (8%)

[0365] >20 PY 154 (54%) 49 (45%) 203 (52%) 34 (65%)

[0366] Smoker but unknown 37(13%) 0(0%) 37(9%) 0 (%)

[0367] PY

[0368] Alcohol consumption

[0369] No 103 (36%) 34(31%) 137 (35%) 27 (52%)

[0370] Yes 161 (57%) 72 (65%) 233 (59%) 25 (48%)

[0371] Unknown 20 (7%) 4 (4%) 24 (6%) 0 (0%)

[0372] Location

[0373] Floor 109 (38%) 0 109 (28%) 36 (69%)

[0374] Tongue 175 (62%) 110(100%) 285 (72%) 16 (31%)

[0375] Pathological stage (V7)

[0376] PTI 95 (20%) pT2 116 (31%) pT3 54 (10%) pT4a 13 (39%) pT4b 1 (<1%) Missing 5 (2%) pNO 143 (50%) pNl 47 (17%) pN2a 6 (1%) pN2b 52(18%) pN2c 34(13%) pN3 2 (1%) Missing 0 pStage (V7)

[0377] I 65(16%) 8(7%) 73(19%)

[0378] II 54(17%) 22(20%) 76(19%)

[0379] III 57(15%) 27 (25%) 84(21%)

[0380] IVa 100 (52%) 48 (44%) 147 (37%) IVb 3 (1%) 3 (1%)

[0381] Unknown 5 (2%) 5 (5%) 11 (3%)

[0382] Pathological stage (V8)

[0383] PTI 45 (16%) pT2 66 (23%) pT3 154 (54%) pT4a 13 (5%) pT4b 1 (<1%) Missing 5 (2%) pNO 143 (50%) pNl 34 (12%) pN2a 13 (5%) pN2b 22 (8%) pN2c 4 (1%) pN3a 1 (<1%) pN3b 66 (23%)

[0384] Missing 1 (<1%) p Stage (V8)

[0385] I 32 (11%)

[0386] II 50 (18%)

[0387] III 82 (29%)

[0388] IVa 50 (18%)

[0389] IVb 68 (24%)

[0390] Unknown 2 (1%)

[0391] Vascular embol

[0392] No 213 (75%)

[0393] Yes 64 (23%)

[0394] Not evaluated 7 (2%)

[0395] Perineural invasion

[0396] No 133 (47%)

[0397] Yes 144 (51%)

[0398] Not evaluated 7 (2%) MGC

[0399] Low (0 - < 0.2) 179 (63%) 81 (74%) 260 (66%) 16 (31%)

[0400] Intermediate (> 0.2 - 65 (23%) 18 (16%) 83 (21%) 6 (12%))

[0401] High (> l) 40 (14%) 11 (10%) 51 (13%) 21 (40%)

[0402] Fibrosis 9 (17%)

[0403] Keratin

[0404] Low (< 10%) 70 (25%) 32 (29%) 102 (26%)

[0405] High (> 10%) 24 (75%) 78 (71%) 292 (74%)

[0406] MGC low / int - 70 (25%) 32 (29%) 102 (26%)

[0407] Keratin low

[0408] MGC low / int - 174 (61%) 67 (61%) 241 (61%)

[0409] Keratin high

[0410] MGC high - Keratin 0 (0%) 0 (0%) 0 (0%) low

[0411] MGC high - Keratin 40 (14%) 11 (10%) 51 (13%) high

[0412] Resection

[0413] R0 220 (78%)

[0414] R1 63 (22%)

[0415] Missing 1 (<1%)

[0416] Nodal extra capsular extension pNO (V7) 143 (50%)

[0417] No 61 (21%)

[0418] Yes 80 (28%)

[0419] Adjuvant treatment

[0420] None 134 (47%) 2 (4%)

[0421] RT 70 (25%) 13 (25%)

[0422] RT-CT 77 (27%) 37 (71%)

[0423] Curie 3 (1%) 0 Table 2: Initial characteristics according to MGC level in the 398 patients of the GR + TCGA cohort (without the patients of the induction cohort)

[0424] MGC low MGC intermediate MGC high p-value

[0425] N=262 N=83 N=53

[0426] Sex 0.75

[0427] Male 188 (72%) 56 (67%) 37 (70%)

[0428] Female 74 (28%) 27 (33%) 16 (30%)

[0429] Age (years)

[0430] Median [range] 58 [23-89] 59 [19-85] 58 [24-74]

[0431] (IQR) (50-66) (50-67) (50-62)

[0432] <65 years 182(69%) 55 (66%) 45 (85%) 0.046

[0433] >65 years 80(31%) 28 (34%) 8(15%)

[0434] Tobacco consumption 0.15

[0435] Never 63 (24%) 23 (28%) 13 (25%)

[0436] < 20 PY 39(15%) 12(14%) 3(6%)

[0437] > 20 PY 137 (52%) 38 (46%) 26(49%)

[0438] Smoker but unknown 23(9%) 10(12%) 11 (21%)

[0439] PY

[0440] Alcohol consumption 0.71

[0441] No 91 (38%) 33 (43%) 18(38%)

[0442] Yes 151 (62%) 44 (57%) 29(62%)

[0443] Unknown 20 6 6

[0444] Location 0.68

[0445] Floor 76 (29%) 20 (24%) 15(28%)

[0446] Tongue 186(71%) 63 (76%) 38 (72%)

[0447] Clinical stage

[0448] T1 30(17%) 13(20%) 11 (26%) 0.21

[0449] T2 41 (23%) 14 (22%) 3 (7%)

[0450] T3 35 (19%) 13(20%) 13(31%)

[0451] T4 75 (41%) 25 (38%) 15(36%)

[0452] Unknown 81 18 11 NO 120 (67%) 44 (68%)

[0453] N1 22 (12%) 8 (12%) 5 (12%)

[0454] N2a-b 28 (16%) 8 (12%) 3 (7%)

[0455] N2c-3 10 (6%) 5 (8%) 2 (5%)

[0456] Unknown 82 18 12

[0457] I 34 (13%) 15 (18%) 11 (21%) 0.028

[0458] II 62 (24%) 22 (27%) 4 (8%)

[0459] III 48 (19%) 12 (15%) 17 (32%)

[0460] IV 113 (44%) 33 (40%) 21 (40%)

[0461] Unknown 5 1 0

[0462] Pathological stage (V7) pTl 36 (20%) 11 (17%) pT2 55 (30%) 23 (36%) 10 (24%) pT3 18 (10%) 5 (8%) 6 (14%) pT4a 72 (40%) 25 (39%) 15 (36%)

[0463] Unknown 81 18 11 pNO 86 (48%) 29 (45%) 29 (69%) 0.20 pNl 29 (16%) 15 (23%) 4 (10%) pN2a-b 37 (20%) 13 (20%) 5 (12%) pN2c-3 29 (16%) 8 (12%) 4 (10%) Unknown 81 18 11

[0464] I 31 (12%) 8 (10%) 12 (23%) 0.33

[0465] II 49 (19%) 14 (17%) 7 (13%)

[0466] III 42 (17%) 18 (22%) 10 (19%)

[0467] IV 131 (52%) 42 (51%) 24 (45%)

[0468] Unknown 9 1 0

[0469] Pathological stage (V8)

[0470] PTl 29 (16%) 9 (14%) 7 (17%) 0.99 pT2 34 (19%) 13 (20%) 7 (17%) pT3 45 (25%) 17 (27%) 13 (31%) pT4a 72 (40%) 25 (39%) 15 (36%)

[0471] Unknown 82 19 11 pNO 85 (47%) 29 (45%) 28 (67%) 0.32 pNl 24(13%) 9(14%) 5(12%) pN2a-b 19 (10%) 9 (14%) 3 (7%) pN2c-3 53 (29%) 18 (28%) 6 (14%)

[0472] Unknown 81 18 11

[0473] I 19(11%) 4(6%) 6(14%) 0.73

[0474] II 27(15%) 7(11%) 6(14%)

[0475] III 34 (19%) 15 (23%) 10 (24%)

[0476] IV 100(56%) 38 (59%) 20(48%)

[0477] Unknown 82 19 11

[0478] Resection 0.20

[0479] R0 136 (75%) 53 (83%) 36 (86%)

[0480] R1 45 (25%) 11 (17%) 6(14%)

[0481] Unknown 81 19 11

[0482] Nodal ECE 0.084 pNO (V7) 86 (48%) 29 (45%) 29 (69%)

[0483] No 44 (24%) 14(22%) 7(17%)

[0484] Yes 50 (28%) 22 (34%) 6(14%)

[0485] Unknown 82 18 11

[0486] Vascular embol 0.35

[0487] No 129 (73%) 51 (81%) 34 (81%)

[0488] Yes 75 (27%) 12(19%) 8(19%)

[0489] Not evaluated 86 20 11

[0490] Perineural invasion 0.056

[0491] No 75 (43%) 33 (52%) 26 (62%)

[0492] Yes 101 (57%) 30 (48%) 16 (38%)

[0493] Not evaluated 86 20 11

[0494] Demographic and disease characteristics of the patients were well balanced between the three groups (Supplementary Table 2). Supplementary Table 2: Initial characteristics according to MGC level in the 394 patients of the GR and TCGA cohorts (without the patients of the induction cohort)

[0495] MGC low MGC intermediate MGC high p-value

[0496] N=260 N=83 N=51

[0497] Sex 0.74

[0498] Male 187 (72%) 56 (67%) 36 (71%)

[0499] Female 73 (28%) 27 (33%) 15 (29%)

[0500] Age (years)

[0501] Median [range] 58 [23-89] 59 [19-85] 58 [24-74]

[0502] (IQR) (50-67) (50-67) (50-62)

[0503] < 65 years 180 (69%) 55 (66%) 43 (84%) 0.061

[0504] > 65 years 80 (31%) 28 (34%) 8 (16%)

[0505] Tobacco consumption 0.19

[0506] Never 62 (24%) 22 (27%) 13 (25%)

[0507] < 20 PY 41 (16%) 13 (16%) 3 (6%)

[0508] > 20 PY 138 (53%) 39 (47%) 26 (51%)

[0509] Smoker but unknown PY 19 (7%) 9 (11%) 9 (18%)

[0510] Alcohol consumption 0.68

[0511] No 87 (36%) 33 (41%) 17 (36%)

[0512] Yes 156 (64%) 47 (59%) 30 (64%)

[0513] Unknown 17 3 4

[0514] Location 0.85

[0515] Floor 74 (28%) 21 (25%) 14 (27%)

[0516] Tongue 186 (72%) 62 (75%) 37 (73%)

[0517] Pathological stage (V7) pTl 56 (32%) 23 (37%) 16 (40%) pT2 75 (43%) 28 (44%) 13 (32%) pT3 34 (19%) 10 (16%) 10 (25%) pT4a-b 11 (6%) 2 (3%) 1 (2%)

[0518] Unknown 84 20 11 pNO 86 (48%) 29 (45%) 28 (70%) pNl 28 (16%) 15 (23%) 4 (10%) pN2a-b 40 (22%) 14 (22%) 4 (10%) pN2c-3 25 (14%) 7 (11%) 4 (10%)

[0519] Unknown 81 18 11

[0520] I 44 (17%) 13 (16%) 16 (31%) 0.17

[0521] II 51 (20%) 15 (19%) 10 (20%)

[0522] III 51 (20%) 21 (26%) 12 (24%)

[0523] Iva-b 108 (43%) 31 (40%) 13 (25%)

[0524] Unknown 6 2 0

[0525] Pathological stage (V8)

[0526] PTI 29 (16%) 9 (14%) 7 (18%) pT2 41 (23%) 15 (24%) 10 (25%) pT3 95 (54%) 37 (59%) 22 (55%) pT4a-b 11 (6%) 2 (3%) 1 (2%)

[0527] Unknown 84 20 11 pNO 86 (48%) 29 (45%) 28 (70%) pNl 21 (12%) 9 (14%) 4 (10%) pN2a-b 21 (12%) 11 (17%) 3 (8%) pN2c-3b 50 (28%) 16 (25%) 5 (12%)

[0528] Unknown 82 18 11

[0529] I 20 (11%) 5 (8%) 7 (18%) 0.24

[0530] II 33 (19%) 8 (12%) 9 (22%)

[0531] III 45 (25%) 22 (34%) 15 (38%)

[0532] IVa 33 (19%) 13 (20%) 4 (10%)

[0533] IVb 47 (26%) 16 (25%) 5 (12%)

[0534] Unknown 82 19 11

[0535] Resection 0.18

[0536] RO 133 (74%) 53 (83%) 34 (85%)

[0537] R1 46 (26%) 11 (17%) 6 (15%)

[0538] Unknown 81 19 11

[0539] Nodal ECE 0.078 pNO (V7) 86 (48%) 29 (45%) 28 (70%) No 40 (22%) 14 (22%) 7 (18%)

[0540] Yes 53 (30%) 22 (34%) 5 (12%)

[0541] Unknown 81 18 11

[0542] Vascular embol 0.21

[0543] No 128 (74%) 51 (81%) 34 (85%)

[0544] Yes 46 (26%) 12 (19%) 6 (15%)

[0545] Not evaluated 86 20 11

[0546] Perineural invasion 0.063

[0547] No 75 (43%) 33 (52%) 25 (62%)

[0548] Yes 99 (57%) 30 (48%) 15 (38%)

[0549] Not evaluated 86 20 11

[0550] Table 3: Hazard ratio (HR) of death estimated in Cox model

[0551] Univariate analysis (398 patients, 229 deaths)

[0552] HR of death 95%CI

[0553] MGC low 1

[0554] MGC mt 0.68 0.49-0.95

[0555] MGC high 0.53 0.34-0.83

[0556] Multivariate analysis [adjusted for sex, age (< 65 years / > 65 years), p-stage (V7), location (tongue / floor), tobacco and alcohol consumption] based on 388 patients (222 deaths)

[0557] HR of death 95%CI

[0558] MGC low 1

[0559] MGC mt 0.65 0.46-0.91

[0560] MGC high 0.50 0.32-0.80

[0561] The MGC level was significantly associated with survival (Wald test p-value = 0.0017). In multivariate analyses adjusted for sex, age, p-stage, location, tobacco, and alcohol consumption, the density of MGC in the tumor was significantly associated with OS (p-value = 0.0016) and PFI (p-value = 0.0005) (cf. Supplementary Tables 3 and 4). Table 3: Hazard ratio (HR) of death estimated in Cox model.

[0562] Univariate analysis (394 patients, 225 deaths)

[0563] HR of death 95%CI

[0564] MGC low 1

[0565] MGC mt 0.69 0.49-0.96

[0566] MGC high 0.51 0.32-0.81

[0567] Multivariate analysis (adjusted for sex, age (< 65 years / > 65 years), p-stage (V7), location (tongue / floor), tobacco and alcohol consumption) based on 386 patients (219 deaths)

[0568] HR of death 95%CI

[0569] MGC low 1

[0570] MGC mt 0.63 0.45-0.89

[0571] MGC high 0.50 0.31 -0.80

[0572] The MGC level was significantly associated with survival (Wald test p-value = 0.0016).

[0573] Table 4: Hazard ratio (HR) of events estimated in Cox model

[0574] Univariate analysis (398 patients, 148 events)

[0575] HR of event 95%CI

[0576] MGC low 1

[0577] MGC mt 0.60 0.39-0.93

[0578] MGC high 0.34 0.18-0.65

[0579] Multivariate analysis [adjusted for sex, age (< 65 years / > 65 years), p-stage (V7), location (tongue / floor), tobacco and alcohol consumption] based on 388 patients (139 events)

[0580] HR of event 95%CI

[0581] MGC low 1

[0582] MGC mt 0.54 0.34-0.85

[0583] MGC high 0.34 0.18-0.66

[0584] The MGC level was significantly associated with survival (Wald test p-value = 0.0004). Supplementary Table 4: Hazard ratio (HR) of events estimated in Cox model.

[0585] Univariate analysis (394 patients, 151 events)

[0586] HR of event 95%CI

[0587] MGC low 1

[0588] MGC mt 0.66 0.44-1.00

[0589] MGC high 0.31 0.16-0.61

[0590] Multivariate analysis (adjusted for sex, age (< 65 years / > 65 years), p-stage (V7), location (tongue / floor), tobacco and alcohol consumption) based on 386 patients (146 events)

[0591] HR of event 95%CI

[0592] MGC low 1

[0593] MGC mt 0.61 0.40-0.94

[0594] MGC high 0.30 0.15-0.60

[0595] The MGC level was significantly associated with survival (Wald test p-value = 0.0005).

[0596] Importantly, high keratinization in the context of low / intermediate densities of MGC was not associated with a better OS and PFI (Fig 7C, 7D and 7E; Supplementary Table 5), showing that it is not high keratinization per se that correlates with outcome.

[0597] Table 5: Initial characteristics according to MGC-keratin categories in the 398 patients of the GR + TCGA cohort (without the patients of the induction cohort)

[0598] MGC low / int - MGC low / int - MGC high - p-value

[0599] Keratin low Keratin high Keratin high

[0600] N=102 N=243 N=53

[0601] Sex 0.04

[0602] Male 82 (80%) 162 (67%) 37 (70%)

[0603] Female 20 (20%) 81 (33%) 16 (30%)

[0604] Age (years)

[0605] Median [range] 58 [26-89] 59 [19-87] 58 [24-74]

[0606] (IQR) (51-65) (49-67) (50-62)

[0607] < 65 years 74 (73%) 163 (67%) 45 (85%) 0.032

[0608] > 65 years 28 (27%) 80 (33%) 8 (15%)

[0609] Tobacco consumption 0.07 Never 22 (22%) 64 (26%) 13 (25%)

[0610] < 20 PY 20 (20%) 31 (13%) 3 (6%)

[0611] > 20 PY 52 (51%) 123 (51%) 26 (49%)

[0612] Smoker but 8 (8%) 25 (10%) 11 (21%) unknown PY

[0613] Alcohol consumption 0.61

[0614] No 33 (35%) 91 (41%) 18 (38%)

[0615] Yes 62 (65%) 133 (59%) 29 (62%)

[0616] Unknown 7 19 6

[0617] Location 0.13

[0618] Floor 36 (35%) 60 (25%) 15 (28%)

[0619] Tongue 66 (65%) 183 (75%) 38 (72%)

[0620] Clinical stage

[0621] T1 16 (23%) 27 (15%)

[0622] T2 15 (21%) 40 (23%) 3 (7%)

[0623] T3 12 (17%) 36 (20%) 13 (31%)

[0624] T4 27 (39%) 73 (41%) 15 (36%)

[0625] Unknown 32 67 11

[0626] NO 46 (66%) 118 (67%) 31 (76%) 0.85

[0627] N1 9 (13%) 21 (12%) 5 (12%)

[0628] N2a-b 12 (17%) 24 (14%) 3 (7%)

[0629] N2c-3 3 (4%) 12 (7%) 2 (5%)

[0630] Unknown 32 68 12

[0631] I 18 (18%) 31 (13%) 11 (21%) 0.026

[0632] II 25 (25%) 59 (25%) 4 (8%)

[0633] III 14 (14%) 46 (19%) 17 (32%)

[0634] IV 44 (44%) 102 (43%) 21 (40%)

[0635] Pathological stage

[0636] PTI 15 (21%) 32 (18%) 11 (26%) 0.39 pT2 27 (39%) 51 (29%) 10 (24%) pT3 4 (6%) 19 (11%) 6 (14%) pT4a 24 (34%) 73 (42%) 15 (36%)

[0637] Unknown 32 68 11 pNO 30 (43%) 85 (48%) 29 (69%) 0.28 pNl 13 (19%) 31 (18%) 4 (10%) pN2a-b 16 (23%) 34 (19%) 5 (12%) pN2c-3 11 (16%) 26 (15%) 4 (10%)

[0638] Unknown 32 67 11

[0639] I 11 (14%) 28 (12%) 12 (23%) 0.26

[0640] II 23 (18%) 40 (17%) 7 (13%)

[0641] III 15 (17%) 45 (19%) 10 (19%)

[0642] IV 49 (51%) 124 (52%) 24 (45%)

[0643] Unknown 4 6 0

[0644] Pathological stage (V8) pTl 16 (23%) 22 (13%) 7 (17%) 0.48 pT2 11 (16%) 36 (21%) 7 (17%) pT3 19 (27%) 43 (25%) 13 (31%) pT4a 24 (34%) 73 (42%) 15 (36%)

[0645] Unknown 32 69 11 pNO 29 (41%) 85 (48%) 28 (67%) 0.27 pNl 11 (16%) 22 (13%) 5 (12%) pN2a-b 8 (11%) 20 (11%) 3 (7%) pN2c-3 22 (31%) 49 (28%) 6 (14%)

[0646] Unknown 32 67 11

[0647] I 8 (11%) 15 (9%) 6 (14%) 0.85

[0648] II 9 (13%) 25 (14%) 6 (14%)

[0649] III 16 (23%) 33 (19%) 10 (24%)

[0650] IV 37 (53%) 101 (58%) 20 (48%)

[0651] Unknown 32 69 11

[0652] Resection 0.27

[0653] R0 57 (81%) 132 (75%) 36 (86%)

[0654] R1 13 (19%) 43 (25%) 6 (14%)

[0655] Unknown 32 68 11 Nodal ECE 0.10 pNO (V7) 30 (43%) 85 (48%) 29 (69%)

[0656] No 17 (25%) 41 (23%) 7 (17%)

[0657] Yes 22 (32%) 50 (28%) 6 (14%)

[0658] Unknown 33 67 11

[0659] Vascular embol 0.016

[0660] No 43 (63%) 137 (80%) 34 (81%)

[0661] Yes 25 (37%) 34 (20%) 8 (19%)

[0662] Not 34 64 11 evaluated

[0663] Perineural invasion 0.11

[0664] No 33 (49%) 75 (44%) 26 (62%)

[0665] Yes 35 (51%) 96 (56%) 16 (38%)

[0666] Not 34 72 11 evaluated

[0667] Supplementary Table 5: Initial characteristics according to MGC-keratin categories in the

[0668] 394 patients of the GR and TCGA cohorts (without the patients of the induction cohort)

[0669] MGC low / int - MGC low / int - MGC high - p-value

[0670] Keratin low Keratin high Keratin high

[0671] N=102 N=241 N=51

[0672] Sex 0.041

[0673] Male 82 (80%) 161 (67%) 36 (71%)

[0674] Female 20 (20%) 80 (33%) 15 (29%)

[0675] Age (years)

[0676] Median [range] 58 [26-89] 59 [19-87] 58 [24-74]

[0677] (IQR) (51-65) (50-67) (50-62)

[0678] < 65 years 74 (73%) 161 (67%) 43 (84%) 0.039

[0679] > 65 years 28 (27%) 80 (33%) 8 (16%)

[0680] Tobacco consumption 0.12

[0681] Never 22 (22%) 62 (26%) 13 (25%)

[0682] < 20 PY 20 (20%) 34 (14%) 3 (6%) > 20 PY 54 (53%) 123 (51%) 26 (51%)

[0683] Smoker but unknown 6 (6%) 22 (9%) 9 (18%)

[0684] PY

[0685] Alcohol consumption 0.40

[0686] No 31 (32%) 89 (40%) 17 (36%)

[0687] Yes 67 (68%) 136 (60%) 30 (64%)

[0688] Unknown 4 16 4

[0689] Location 0.12

[0690] Floor 36 (35%) 59 (24%) 14 (27%)

[0691] Tongue 66 (65%) 182 (76%) 37 (73%)

[0692] Pathological stage (V7) pTl 25 (36%) 54 (32%) 16 (40%) pT2 37 (54%) 66 (39%) 13 (32%) pT3 5 (7%) 39 (23%) 10 (25%) pT4a-b 2 (3%) 11 (6%) 1 (2%)

[0693] Unknown 33 71 11 pNO 30 (43%) 85 (49%) 28 (70%) pNl 13 (19%) 30 (17%) 4 (10%) pN2a-b 18 (26%) 36 (21%) 4 (10%) pN2c-3 9 (13%) 23 (13%) 4 (10%)

[0694] Unknown 32 67 11

[0695] I 18 (14%) 39 (17%) 16 (31%) 0.12

[0696] II 23 (18%) 43 (18%) 10 (20%)

[0697] III 17 (17%) 55 (23%) 12 (24%)

[0698] IV 42 (51%) 98 (42%) 13 (25%) Unknown 2 6 0

[0699] Pathological stage (V8) pTl 16 (23%) 22 (13%) 7 (18%) pT2 15 (22%) 41 (24%) 10 (25%) pT3 36 (52%) 96 (56%) 22 (55%) pT4a-b 2 (3%) 11 (6%) 1 (2%)

[0700] Unknown 33 71 11 pNO 30 (43%) 85 (49%) 28 (70%) pNl 10(14%) 20(11%) 4(10%) pN2a-b 9(13%) 23(13%) 3(8%) pN2c-3 20 (29%) 46(26%) 5(12%)

[0701] Unknown 33 67 11

[0702] I 9(13%) 16(9%) 7(18%) 0.34

[0703] II 12(17%) 29(17%) 9(22%)

[0704] III 18(26%) 49(28%) 15(38%)

[0705] IVa 11 (16%) 35 (20%) 4(10%)

[0706] IVb 20 (29%) 43 (25%) 5 (12%)

[0707] Unknown 32 69 11

[0708] Resection 0.35

[0709] R0 56 (80%) 130 (75%) 34 (85%)

[0710] R1 14(20%) 43 (25%) 6(15%)

[0711] Unknown 32 68 11

[0712] Nodal ECE 0.069 pNO (V7) 30 (43%) 85 (49%) 28 (70%)

[0713] No 17(24%) 37(21%) 7(18%)

[0714] Yes 23 (33%) 52 (30%) 5 (12%)

[0715] Unknown 32 67 11

[0716] Vascular embol 0.0073

[0717] No 43 (63%) 136 (80%) 34 (85%)

[0718] Yes 25 (37%) 33 (20%) 6(15%)

[0719] Not evaluated 34 72 11

[0720] Perineural invasion 0.12

[0721] No 33 (49%) 75 (44%) 25 (62%)

[0722] Yes 35(51%) 94 (56%) 15(38%)

[0723] Not evaluated 34 72 11

[0724] Taken together, these data show that patients with HNSCC bearing keratinizing tumors are more likely to have positive outcomes in the presence of high densities of MGC. Induction-chemotherapy induces MGC formation.

[0725] Next, Inventors extended investigation to include a cohort of patients with HNSCC (n = 52 patients) treated by three cycles of induction-chemotherapy (ICT) followed by surgery and adjuvant therapy (Fig. 2A). Patients were classified into three groups: no residual tumor, and two groups with residual tumors having either high (MGCHlgh) or low / intermediate (MGCLow / Int) densities of MGC (Fig. 2B). Interestingly, high densities of MGC in surgical resections was more common after ICT compared to treatment-naive TCGA and GR patients (Fig. 2C, D), suggesting that ICT might induce the formation of MGC. As observed in treatment-naive patients, high densities of MGC were also associated with longer OS, comparable to patients with no residual tumor (Fig. 2E). In addition, the degree of pathological response after ICT was similarly correlated with MGC density: good / partial responders were more likely to have high densities of MGC in their tumors compared to the poor responders (Fig. 2F, G, H).

[0726] In summary, MGC density in tumors is associated with good outcome in patients with HNSCC either in the presence or absence of ICT prior to surgical resection.

[0727] Automatic detection of MGC and tumor cells on H&E / HES WSI by deep learning.

[0728] Effective stratification of patients with cancer is a mainstay of good clinical management, yet few novel biomarkers prove amenable to quick and easy measurement in the hospital setting: for example, here, it would take a trained pathologist up to 30 minutes per slide to assess the density of MGC. Therefore, inventors next aimed to establish a high-throughput automatic method to accurately (i.e., efficiently) detect and enumerate MGC on whole-slide images (WSI) of patients with SCC. Inventors trained two models on routinely H&E / HES stained WSI: one that detects tumor cells and one that detects MGC (Fig. 3A, B). Over 5,000 manually annotated MGC from TCGA and GR cohorts were used to train the MGC detection model and more than 800,000 manually annotated carcinomatous cells were used to train the tumor cells detection model. A proxy of the biomarker was computed as the total number of detected MGC over the total number of detected tumor cells, across all available slides for each patient.

[0729] After training, inventors validated the performance of both models on 110 slides from TCGA cohort and 839 slides from GR cohort that were different from those used fortraining. Inventors found that the models performed comparably to manual quantification by a pathologist, both for MGC in TCGA (r2 = 0.8792; p < 0.0001) and GR (r2 = 0.7811; p < 0.0001) cohorts (Fig. 3C, D), as well as for tumor cells in TCGA (average precision = 0.688) and GR (average precision = 0.679) cohorts. Importantly, on the 949 slides (representing 341 patients) that passed quality control (Fig 8A), the automatic MGC density correlated with the manual MGC density in TCGA (r2 = 0.7833; p < 0.0001) and GR (r2 = 0.4036; p < 0.0001) (Fig 8B, C) cohorts. Accordingly, patients were efficiently stratified into MGCHlgh, MGClnand MGCLowgroups and the automatic MGC biomarker was again associated with longer OS and PFI (Fig. 3E, F).

[0730] Inventors then asked whether the automatic detection model could be used to measure MGC density in SCC samples from other organs, namely uterine cervix and larynx. Initially, the MGC predictions gave false positive MGC detections in both cervical and laryngeal SCC cohorts from TCGA; these were due to tissue-specific features of the larynx and cervix that were absent in the oral cavity and so not included in the training data. Inventors overcame this issue by calibrating the MGC detection model to increase its sensitivity. Doing so, inventors excluded the risk of false negatives, while leading to abundant (true and false) positives that were then manually corrected by a pathologist: although slower than full automation, compared to a pathologist-only approach, this method was approximatively 10 times faster (Fig 8 D, E) and reached similar levels of accuracy / performance (Fig 8 F, G). Using this approach, inventors then categorized patients into high and low MGC / tumor ratios and found that increased proportions of MGC again correlated with longer OS in both uterine cervix squamous cell carcinoma (CESC, also herein identified as CESCC) and SCC of the larynx (Fig. 3G, H), and with longer PFI in CESC (Fig 8 H, I).

[0731] Altogether, their automatic cell detection approach was rapid and efficient, allowing easy assessment of the MGC-to-tumor cell ratio that acted as a clear biomarker of prognosis in patients with oral cavity SCC, CESC, and SCC of the larynx.

[0732] Spatial transcriptomics identifies MGC as a specific population of macrophages.

[0733] Having identified the relationship between MGC density and prognosis of patients with SCC, inventors next wanted to understand the biological features of these cells and so their likely role within the tumor microenvironment (TME). Inventors exploited Visium spatial transcriptomic technology (10X Genomics) to analyze the single giant cell transcriptomes of MGC - on formalin-fixed paraffin-embedded (FFPE) HNSCC tumor sections from GR cohort (Fig. 4A, B). Such technology allows the assessment of transcriptomes on spots, larger than single cells, directly on histological slides. Importantly, because the diameter of a Visium spot is 55pm, this technology is considered having a pseudo-single cell resolution. Here, they took advantage of this limit, as one spot corresponds roughly to the size of an individual MGC, allowing the definition of their specific transcriptomic signature.

[0734] Inventors analyzed slides from nine patients: six with MGCHlghtumors, and three that were MGCLow(Fig. 4C-H). Unsupervised clustering revealed seven distinct clusters of cells that matched with manual morphology-based annotations of the slides. When projected onto a UMAP space, MGC formed a clear and distinct cluster (Fig. 41). Of note, MGC were the only cell type with significantly different abundance between MGCHlghand MGCLowtumor samples (Fig. 4J). To narrow MGC signature, spots only covering MGC were manually selected (Fig 9A) and were shown to overlap with the unsupervised analysis (Fig 9B). Differentially expressed genes (DEG) between annotated MGC spots and all the non-MGC spots included CHIT1, FBP1, MMP9, SPP1, APOE, CHI3L1, CTSS, CTSB, CTSD, CTSZ, TYROBP, CD68, DC-STAMP, MARCO and TREM2: this confirmed the macrophage nature of MGC (Fig. 4K). Gene Ontology (GO) analysis of this supervised signature showed upregulation of pathways involved in bone resorption, tissue remodeling and extracellular matrix disassembly, as well as macrophage activation, phagocytosis, antigen presentation and ROS production.

[0735] To validate this signature, inventors investigated the transcriptomic differences between the MGCHlghand MGCLowtumors from patients in TCGA oral cavity SCC cohort. They retrieved the bulk RNA sequencing data (bulk RNAseq) for 108 patients (out of the initial 110 patients, two did not have available RNA-seq data) for which they had quantified MGC density on WSI. DEG analysis between the two groups showed a clear difference between the bulk signatures of tumors that were MGCHlghand MGCLow(Fig. 4L). Among genes that were more highly expressed in the MGCHlghtumors were macrophage-related genes (SPP1, MARCO, DC- STAMP, OSCAR) and MGC-specific genes (CHIT1, FBP1), which confirmed their spatial transcriptomic data (Fig. 4M). GO analysis of the TCGA RNA bulk signature also highlighted pathways common to the Visium-defmed MGC signature, such as bone resorption, tissue remodeling and macrophage differentiation. Inventors also investigated the potential influence of CHIT1 and TREM2 on patient survival in the TCGA oral cavity cohort. Remarkably, CHIT1 and TREM2 were associated with a good OS in the TCGA oral cavity SCC cohort (Fig 9C).

[0736] Finally, inventors performed immunohistochemistry for representative markers to validate the RNA signature at the protein level. CK5 protein and RNA expression was restricted to carcinomatous cells, while CHIT1 and CD68 protein and RNA expression patterns overlayed the MGC area (Fig 9 D, E, F).

[0737] Taken together these results revealed the core macrophage signature of MGC and suggested a potential anti-tumoral function through expression of genes and proteins associated with phagocytosis, antigen presentation, keratin resorption and tissue remodeling. TREM2-expressing mononuclear tumor associated macrophage (TAM) and MGC cluster together in keratin-enriched niches.

[0738] Inventors then asked about the relative localization of MGC and TAM in oral cavity SCC. Inventors designed a 5 -color multiplex imaging panel containing DAPI and antibodies recognizing CHIT1, CD163, CD68 and TREM2. They selected CD68 and CD163 because of their pan-macrophage nature and CHIT1 and TREM2 to study their expression in MGC and TAM. While confirming that MGC express CHIT1 and TREM2 proteins (Fig. 5A, B, Fig. 10A), inventors observed that mononuclear TAM surrounding MGC were also expressing CHIT1 and TREM2 in these tumors (Fig. 5B). They confirmed that MGC were more present in MGCHlghpatients (Fig. 10B). Interestingly, TREM2 -expressing mononuclear TAM were significantly more abundant in MGCHlghcompared to MGCLowtumors (Fig. 5C, D). These two populations of macrophages were preferentially clustering in keratin-rich niches (Fig. 5E, F), leading to speculate that mononuclear TAM might be the precursors of MGC.

[0739] TREM2-expressing mononuclear tumor associated macrophage (TAM) share a similar transcriptional program with MGC and are associated with a good response to neoadjuvant immunotherapy.

[0740] Having established the common localization and abundance features of MGC and TREM2- expressing mononuclear TAM, inventors next compared their transcriptomes. They used their spatial transcriptomics MGC signature together with scRNA-seq data from four publicly available datasets on HNSCC tumors: two from patients without pre-operative therapy (Cillo A. R. et al.; Kurten C. H. L. et al). one from patient treated by induction chemotherapy (Song H. et al), and another from patients treated by neoadjuvant immunotherapy (Luoma A. M. et al Schoenfield J. D. etal) (n = 74 patients and 23,635 monocytes / macrophages). After integrating the datasets, they projected them onto a UMAP space to create an HNSCC universe (HNSCC- verse) and identified ten different clusters of monocytes and macrophages, annotated using the MoMac-verse (Mulder K. et al as a reference (Fig. 6A). Strikingly, when projected onto the HNSCC-verse, the supervised MGC spots mapped entirely onto the TREM2Hlghmacrophage cluster (Fig. 6B, C), indicative of a high level of transcriptomic similarity.

[0741] In addition, scRNAseq data revealed that TREM2Hlghmononuclear macrophages were more abundant in patients whose tumors showed good or partial responses after neo-adjuvant immunotherapy (Fig. 6D). Of note, poor responders had low level of TREM2Hlghmacrophages (Fig. 6E) with a 3.5-fold ratio of good and partial responders over poor responders (Fig. 6F, G). Using the available before and after treatment scRNAseq data from this cohort, they saw that the number of TREM2Hlghmacrophages dramatically increased after immunotherapy (Immune Checkpoint Blockade) in good and partial responders, whereas it did not in poor responders (Fig. 6H). Because they previously showed that MGC were more abundant in patients with tumors that responded well to induction chemotherapy (ICT) (Fig. 2F), these results highlighted a correlation between TREM2HlghTAM, MGC abundance and positive response to neoadjuvant immunotherapy in patients with HNSCC.

[0742] DISCUSSION

[0743] Herein, inventors observed that the density of MGC in HNSCC patients associates in a dosedependent manner with a longer OS and PFI. They first observed this phenomenon in an exploratory cohort from TCGA (n = 110), and then confirmed it in an independent GR validation cohort (n = 284). As in treatment-naive patients (n = 394), the association was also observed in preoperative chemotherapy-treated (n = 52) patients. Given the urgent need for improved methods for patient stratification in HNSCC, they developed a deep learning approach to automatically quantify the MGC-to-tumor-cell ratio on standard H&E / HES WSI. This automatic detection was as effective as the pathologist, but much quicker. Inventors then investigated the underlying biology of MGC using spatial transcriptomics, and found that they closely resemble TREM2HlghTAM, which co-localize with MGC in keratin-enriched tumor niches, correlate with MGC abundance, and associate with response to neo-adjuvant immunotherapy.

[0744] The accumulation of MGC is well described in response to foreign body (Camicer-Lombarte et al.). Indeed, following the introduction of a foreign body, such as surgical sutures or bone protheses, macrophages and monocytes fused together to form foreign-body-giant-cells (FBGC) (Camicer-Lombarte et al.). Usually, it results in fibrotic encapsulation of the foreign body, leading in extreme cases to its rejection from the organism. In HNSCC, the carcinoma synthesizes the keratin that is seen as a foreign body by the immune system (Patil S. et al.). However, it is not the presence of keratin per se, but rather of keratin associated MGC that is associated with longer survival of patients. The open question is why only a fraction of patients with keratinizing SCC have high densities of MGC in their tumors.

[0745] The presence of MGC in SCC after preoperative therapy has been long known (Safaii H. et al. ), and is part of the histological scoring for pathological response on surgical resection following anti-tumor therapy (Braun O. M. et al.). The original grading in HNSCC published in 1989 by Braun et al., described 4 grades based on the percentage of residual vital tumor cells, the presence of fibrotic connective tissue, keratin pearls and giant cell infiltration (Braun O. M. et a / .). Here, inventors demonstrated the prognosis significance of MGC, as well as TREM2hlghmacrophage, in SCC patients, in particular HNSCC patients, treated by ICT before surgery or neoadjuvant immunotherapy, as well as in treatment-naive patients. More particularly, they demonstrate that MGC is an independent factor directly associated with an improved survival of these patients. Surprisingly, the same was demonstrated for TREM2hlghmacrophage. Moreover, they extended their findings to define the first evidence-based threshold for the relationship between MGC density and patient risk, and so demonstrated the potential of this approach for informing therapeutic decisions.

[0746] Inventors took advantage of the newly developed Visium spatial transcriptomics approach for FPPE sections (10X Genomics) which allowed to analyze the whole transcriptomes of individual MGC. This revealed a specific “single giant cell RNA signature” of MGC in HNSCC, which inventors then compared with the transcriptomes of other well-known macrophage subsets, using the MoMac verse - a recently published single-cell atlas of monocytes and macrophages from different cancers (Mulder K. et al.). Interestingly, they found strong similarities between TREM2Hlghmacrophages and MGC; moreover, the abundance of TREM2Hlghmononuclear TAM correlated with that of MGC across different datasets of HNSCC. However, TREM2Hlghmacrophages are known to be pro-tumoral in several cancers (Molgora M. et al., 2023), including colorectal (Molgora M. et al., 2020), breast (Molgora M. et al., 2020; Timperi E. et al ), and lung adenocarcinomas (Maynard A. et al ). Moreover, targeting TREM2HlghTAM enhances the efficacy of immunotherapy in a model of ovarian adenocarcinoma (Binnewies M. et al. ), and TREM2 -expressing macrophages are more abundant in non-responding melanomas patients treated by immunotherapy (Xiong D et al.). Interestingly, neither adenocarcinomas nor melanomas produce keratin: in contrast, this study described a good prognosis for TREM2HlghTAM in SCC, in particular HNSCC, - a carcinoma known to produce extracellular keratin.

[0747] Given the growing recognition of the importance of spatial factors within the TME, inventors also investigated the localization of TREM2Hlghmononuclear TAM and MGC in HNSCC tumors by multiplex immunofluorescence. Additionally, they performed a multiplex fluorescent analysis to assess the spatial localization of TREM2Hlghmononuclear and multinucleated TAM in the tumor. Of note, a recent spatial analysis of the tumoral niche in breast and colorectal adenocarcinoma revealed that TREM2HlghTAM were found in the necrotic areas of the tumor (Matusiak M. A. et al.). Here, inventors showed that TREM2HlghTAM and MGC clustered together within tumor niches containing extracellular keratin. Without wishing to be bound by a particular theory, inventors believe that the keratin produced by the SCC creates specific intra- tumoral niches that lead to the initial accumulation of TREM2HlghTAM that subsequently fuse into anti-tumoral MGC. These observations highlight the pivotal roles played by TREM2- expressing TAM in cancer and urge caution for current clinical trials of anti-TREM2 therapies in keratinizing SCC where such an approach could hamper patients’ prognoses.

[0748] Inventors also herein showed the favorable impact of CHIT1 on OS and most importantly identified MGC as a major source of CHIT1 production by single giant cell analysis.

[0749] The recent and progressive digitization of pathology laboratories opens the way for artificial intelligence algorithms that will lead to unprecedented improvements in diagnostic and prognostic applications (Berbis M. A. etal.). Manual quantification of MGC is time-consuming for overwhelmed pathologists (van der Laak J et al.) and impossible to apply routinely: thus, inventors designed an advantageous deep learning model to automatically quantify the MGC- to-tumor cell ratio. This tool is easily integrated into the pathological workflow, where it identifies and quantifies MGC and tumor cells on routine diagnostic histopathological slides stained for example by H&E / HES, without the need for additional costly techniques such as immunohistochemistry or genetic testing. This approach is faster, yet equally as efficient as enumeration by pathologists, and similarly stratified patients with HNSCC. This approach can be used advantageously to evaluate the prognosis, or response to treatment, of a subject suffering of a SCC, and identify patients in need of therapeutic escalation / de-escalation according to predicted prognosis.

[0750] Lastly, because SCC are widely spread across the body, inventors also screened for the presence of MGC in laryngeal SCC and cervical squamous cell carcinoma (CESC). They confirmed that high MGC density is an indicator of favorable prognosis also in SCC of larynx and uterine cervix.

[0751] To conclude, inventors showed that TREM2 -expressing MGC and TAM are major contributors to good prognosis in treatment-naive and pre-operatively treated SCC, and their monitoring will be of tremendous value to improve the clinical management of patients suffering from SCC, in particular HNSCC.

[0752] REFERENCES

[0753] Bankhead, P., Loughrey, M. B., Fernandez, J. A., Dombrowski, Y., McArt, D. G., Dunne, P.

[0754] D., McQuaid, S., Gray, R. T., Murray, L. J., Coleman, H. G., et al. QuPath: Open source software for digital pathology image analysis. Sci. Rep. 7, 1-7 (2017).

[0755] Becht, E., Mclnnes, L., Healy, J., Dutertre, C. A., Kwok, I. W. H., Ng, L. G., Ginhoux, F. & Newell, E. W. Dimensionality reduction for visualizing single-cell data using UMAP. Nat. Biotechnol. 37, 38-47 (2019).

[0756] Berbis, M. A., McClintock, D. S., Bychkov, A., Van der Laak, J., Pantanowitz, L., Lennerz, J. K., Cheng, J. Y., Delahunt, B., Egevad, L., Eloy, C., et al. Computational pathology in 2030: a Delphi study forecasting the role of Al in pathology within the next decade. eBioMedicine 88, 104427 (2023).

[0757] Bill, R., Wirapati, P., Messemaker, M., Roh, W., Zitti, B., Duval, F., Park, J. C., Saal, T. M., Hoelzl, J., Tarussio, D., et al. CXCL9:SPP1 macrophage polarity identifies a network of cellular programs that control human cancers. Science (80-. ). 381, 515-524 (2023).

[0758] Binnewies, M., Pollack, J. L., Rudolph, J., Dash, S., Abushawish, M., Lee, T., Jahchan, N. S., Canaday, P., Lu, E., Nomg, M., et al. Targeting TREM2 on tumor-associated macrophages enhances immunotherapy. Cell Rep. 37, 109844 (2021).

[0759] Braun, O. M., Neumeister, B., Popp, W., Scherrer, R., Dobrowsky, E., Dobrowsky, W., Rausch,

[0760] E. M., Strassl, H., Krisch, K. & Holzner, J. H. Histologic tumor regression grades in squamous cell carcinoma of the head and neck after preoperative radiochemotherapy. Cancer 63, 1097-1100 (1989).

[0761] Burkhardt A. B., & Gebbers J.-O. Giant Cell Stromal Reaction in Squamous Cell Carcinomata. Virchows Arch. A Path. Anat. Histol. 280, 263-280 (1977).

[0762] Buslaev, A., Iglovikov, V. I., Khvedchenya, E., Parinov, A., Druzhinin, M. & Kalinin, A. A. Albumentations: Fast and flexible image augmentations. Information 11, 1-20 (2020).

[0763] Camicer-Lombarte, A., Chen, S. T., Malliaras, G. G. & Barone, D. G. Foreign Body Reaction to Implanted Biomaterials and Its Impact in Nerve Neuroprosthetics. Front. Bioeng. Biotechnol. 9, 1-22 (2021)

[0764] Cillo, A. R., Kurten, C. H. L., Tabib, T., Qi, Z., Onkar, S., Wang, T., Liu, A., Duvvuri, U., Kim, S., Soose, R. J., et al. Immune Landscape of Viral- and Carcinogen-Driven Head and Neck Cancer. Immunity 52, 183-199. e9 (2020).

[0765] Chen, K., Wang, J., Pang, J., Cao, Y., Xiong, Y., Li, X., Sun, S., Feng, W., Liu, Z., Xu, J., et al. MMDetection: Open MMLab Detection Toolbox and Benchmark. (2019).

[0766] Colonna, M. The biology of TREM receptors. Nat. Rev. Immunol. (2023) doi: 10.1038 / s41577- 023-00837-1.

[0767] Corchete, L.A., Rojas, E.A., Alonso-Lopez, D. et al. Systematic comparison and assessment of RNA-seq procedures for gene expression quantitative analysis. Sci Rep 10, 19737 (2020) de Medeiros, V. A., de Pontes Santos, H. B., de Brito Monteiro, B. V., da Paz, A. R., Alves, P.

[0768] M. & Nonaka, C. F. W. Absence of multinucleated giant cell reaction as an indicator of tumor progression in oral tongue squamous cell carcinoma. Eur. Arch. Oto-Rhino-Laryngology 279, 3123-3130 (2022).

[0769] Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K. & Fei-Fei, L. ImageNet: A Large-Scale Hierarchical Image Database. IEEE 248-255 (2009) doi: 10.1109 / CVPR.2009.5206848.

[0770] Ferris, R. L., Blumenschein, G., Fayette, J., Guigay, J., Colevas, A. D., Licitra, L., Harrington, K., Kasper, S., Vokes, E. E., Even, C., et al. Nivolumab for Recurrent Squamous-Cell Carcinoma of the Head and Neck. N. Engl. J. Med. 375, 1856-1867 (2016).

[0771] Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: The next generation. Cell 144, 646-674 (2011).

[0772] He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learning for Image Recognition Kaiming. IEEE (2016) doi : 10.1109 / CVPR.2016.90.

[0773] Heberle, H., Meirelles, V. G., da Silva, F. R., Telles, G. P. & Minghim, R. InteractiVenn: A web-based tool for the analysis of sets through Venn diagrams. BMC Bioinformatics 16, 1- 7 (2015).

[0774] Johnson, D. E., Burtness, B., Leemans, C. R., Lui, V. W. Y ., Bauman, J. E. & Grandis, J. R. Head and neck squamous cell carcinoma. Nat. Rev. Dis. Prim. 6, (2020).

[0775] Kingma, D. P. & Ba, J. L. Adam: A method for stochastic optimization. 3rd Int. Conf. Learn. Represent. ICLR 2015 - Conf. Track Proc. 1-15 (2015).

[0776] Kurten, C. H. L., Kulkami, A., Cillo, A. R., Santos, P. M., Roble, A. K., Onkar, S., Reeder, C., Lang, S., Chen, X., Duvvuri, U., et al. Investigating immune and non-immune cell interactions in head and neck tumors by single-cell RNA sequencing. Nat. Commun. 12, 1- 16 (2021).

[0777] Li, B., Zhang, W., Guo, C. et al. Benchmarking spatial and single-cell transcriptomics integration methods for transcript distribution prediction and cell type deconvolution. Nat Methods 19, 662-670 (2022). https: / / doi.org / 10.1038 / s41592-022-01480-9

[0778] Li, H., Zhou, J., Li, Z. et al. A comprehensive benchmarking with practical guidelines for cellular deconvolution of spatial transcriptomics. Nat Commun 14, 1548 (2023)

[0779] Lin, T.-Y., Goyal, P., Girshick, R., He, K. & Dollar, P. Focal Loss for Dense Object Detection. IEEE (2017) doi: 10.1109 / ICCV.2017.324.

[0780] Liu, J., Lichtenberg, T., Hoadley, K. A., Poisson, L. M., Lazar, A. J., Chemiack, A. D., Kovatich, A. J., Benz, C. C., Levine, D. A., Lee, A. V., et al. An Integrated TCGA PanCancer Clinical Data Resource to Drive High-Quality Survival Outcome Analytics. Cell 173, 400-416.el l (2018).

[0781] Luoma, A. M., Suo, S., Wang, Y., Gunasti, L., Porter, C. B. M., Nabilsi, N., Tadros, J., Ferretti, A. P., Liao, S., Gurer, C., et al. Tissue-resident memory and circulating T cells are early responders to pre-surgical cancer immunotherapy. Cell 185, 2918-2935. e29 (2022).

[0782] Marhic, A., Guerlain, J., Benmoussa, N., Breuskin, L, Honart, J. F., Janot, F., Moya-Plana, A., Temam, S. & Gorphe, P. Replacement of lip-split mandibulotomy by pull-through approach for T3-4 oral carcinomas. Int. J. Oral Maxillofac. Surg. (2021) doi:10.1016 / j.ijom.2020.10.011.

[0783] Matusiak, M., Hickey, J. W., Luca, B., Lu, G., Kidzihski, L., Shu, S., Colburg, D. R. C., Phillips, D. J., Brubaker, S. W., Charville, G. W., et al. A spatial map of human macrophage niches links tissue location with function. bioRxiv 2022.08.18.504434 (2022).

[0784] Maynard, A., McCoach, C. E., Rotow, J. K., Harris, L., Haderk, F., Kerr, D. L., Yu, E. A., Schenk, E. L., Tan, W., Zee, A., et al. Therapy-Induced Evolution of Human Lung Cancer Revealed by Single-Cell RNA Sequencing. Cell 182, 1232-1251.e22 (2020).

[0785] Molgora, M., Esaulova, E., Vermi, W., Hou, J., Chen, Y., Luo, J., Brioschi, S., Bugatti, M., Omodei, A. S., Ricci, B., et al. TREM2 Modulation Remodels the Tumor Myeloid Landscape Enhancing Anti-PD-1 Immunotherapy. Cell 182, 886-900. el7 (2020).

[0786] Molgora, M., Liu, Y. A., Colonna, M. & Celia, M. TREM2: A new player in the tumor microenvironment. Semin. Immunol. 67, 101739 (2023).

[0787] Moya-Plana, A., Auperin, A., Guerlain, J., Gorphe, P., Casiraghi, O., Mamelle, G., Melkane, A., Lumbroso, J., Janot, F. & Temam, S. Sentinel node biopsy in early oral squamous cell carcinomas: Long-term follow-up and nodal failure analysis. Oral Oncol. 82, 187-194 (2018).

[0788] Mulder, K., Patel, A. A., Kong, W. T., Piot, C., Halitzki, E., Dunsmore, G., Khalilnezhad, S., Irac, S. E., Dubuisson, A., Chevrier, M., et al. Cross-tissue single-cell landscape of human monocytes and macrophages in health and disease. Immunity 54, 1883-1900. e5 (2021).

[0789] Nadeem R. Abu-Rustum, Catheryn M. Yashar MD, et al., NCCN Guidelines Insights: Cervical cancer, Version 1.2024, JNCCN, V21, (Dec. 2023).

[0790] Pandiar, D., Ramani, P., Krishnan, R. P. & Monica, K. Multifaceted multinucleated giant cells in oral squamous cell carcinoma. Oral Oncol. 121, (2021).

[0791] Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury Google, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems. NeurlPS 8026- 8037 (2019).

[0792] Patil, S., Rao, R. S. & Ganavi, B. S. A foreigner in squamous cell carcinoma! J. Int. oral Heal. JIOH 5, 147-14750 (2013).

[0793] Pittet, M. J., Michielin, O. & Migliorini, D. Clinical relevance of tumour-associated macrophages. Nat. Rev. Clin. Oncol. (2022) doi: 10.1038 / s41571-022-00620-6.

[0794] Pfister, Spencer, Adelstein, Adkins, Anzai et al. Head and Neck Cancers, version 2.2020, NCCN Clinical Practice Guidelines in Oncology. JNatl Compr Cane Neto. 2020, Jul;18(7):873-898.

[0795] Safaii, H. & Azar, H. A. Keratin Granulomas in Irradiated Squamous Cell Carcinoma of Various Sites. Cancer Res. (1966).

[0796] Schoenfeld, J. D., Hanna, G. J., Jo, V. Y., Rawal, B., Chen, Y., Catalano, P. S., Lako, A., Ciantra, Z., Weirather, J. L., Criscitiello, S., et al. Neoadjuvant Nivolumab or Nivolumab Plus Ipilimumab in Untreated Oral Cavity Squamous Cell Carcinoma A Phase 2 Open-Label Randomized Clinical Trial. Jama Oncol. 6, 1563-1570 (2020).

[0797] Song, H., Lou, C., Ma, J., Gong, Q., Tian, Z., You, Y., Ren, G., Guo, W., Wang, Y., He, K., et al. Single-Cell Transcriptome Analysis Reveals Changes of Tumor Immune Microenvironment in Oral Squamous Cell Carcinoma After Chemotherapy. Front. Cell Dev. Biol. 10, 1-14 (2022).

[0798] Sung, H., Ferlay, J., Siegel, R. L., Laversanne, M., Soerjomataram, L, Jemal, A. & Bray, F. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA. Cancer J. Clin. 71, 209-249 (2021).

[0799] Tian, Z., Shen, C., Chen, H. & He, T. FCOS: Fully convolutional one-stage object detection. Proc. IEEE Int. Conf. Comput. Vis. 2019-Octob, 9626-9635 (2019).

[0800] Timperi, E., Gueguen, P., Molgora, M., Magagna, L, Kieffer, Y., Lopez-Lastra, S., Sirven, P., Baudrin, L. G., Baulande, S., Nicolas, A., et al. Lipid-Associated Macrophages Are Induced by Cancer-Associated Fibroblasts and Mediate Immune Suppression in Breast Cancer. Cancer Res. 82, 3291-3306 (2022). van der Laak, J., Litjens, G. & Ciompi, F. Deep learning in histopathology: the path to the clinic. Nat. Med. 27, 775-784 (2021).

[0801] Wang, H., Zhou, J., Li, J., Geng, Y., Meng, P., Ma, C., Zhu, Z., Zhang, W., Hong, L., Quan, Y., et al. A study of multinucleated giant cells in esophageal cancer. Clin. Immunol. 222, 108600 (2021).

[0802] Xiong, D., Wang, Y. & You, M. A gene expression signature of TREM2hi macrophages and y5 T cells predicts immunotherapy response. Nat. Commun. 11, 1-12 (2020).

Claims

CLAIMS1. A method for evaluating the prognosis or the response to a therapeutic treatment of a subject having a squamous cell carcinoma (SCC), wherein the method comprises a step of assessing the multinucleated giant cells (MGC) status of a SCC tumor of the subject.

2. The method according to claim 1, wherein the multinucleated giant cells (MGC) status of a SCC tumor is assessed by a)- determining a “MGC or TREM2Hlghmacrophages-tumor” ratio defined as i) the number of MGC, or of TREM2Hlghmacrophages, per mm2of SCC tumor area, or ii) the number of MGC, or of TREM2Hlghmacrophages, per the number of cancer cells, in a SCC tumor sample,- detecting or measuring the expression of the CHIT1 , FBP1 and / or TREM2 gene(s) in a SCC tumor sample, and / or- detecting or measuring the secretion of a CHIT1, FBP1 and / or TREM2 protein(s) in a blood sample or in a SCC tumor sample of the subject; and b) determining that the tumor has a MGCHlgh, MGClm. or MGCLowstatus, a MGCHlghstatus of the tumor being correlated to a favorable prognosis or positive response to the treatment, a MGCLowstatus of the tumor being correlated to a bad prognosis or resistance to the treatment, and a MGClmstatus being correlated to an intermediate prognosis or resistance to the treatment.

3. The method according to claim 2, wherein the assessment of the MGC status comprises the steps of: a) determining the ratio of the number of MGC per mm2of tumor area or per number of cancer cells, the number(s) of MGC and / or of cancer cells or the tumor area being obtained from image(s) of the SCC tumor; b) comparing the ratio of step a) to a reference ratio for a SCC tumor of the same tissue origin, and c) determining the MGC status of the tumor, a tumor being considered as having a MGCHlghstatus if the ratio is equal to or above (>) the reference ratio, and as having a MGCLowstatus if the ratio is below (<) the reference ratio.

4. The method according to claim 3, wherein the method is a partially or fully computer- implemented method.

5. The method according to anyone of claims 1 to 4, wherein the SCC is selected from head and neck squamous cell carcinoma (HNSCC), lung carcinoma, esophagus carcinoma, skin carcinoma and uterus cervix carcinoma, and is preferably HNSCC in particular a SCC of the oral cavity or of the larynx.

6. The method according to anyone of claims 1 to 5, wherein:- if the SCC is a HNSCC of the oral cavity, the MGC status is determined as being MGCHlghif the ratio of the number of MGC per mm2of tumor area is equal to or above (>) 1, as being MGCLowif said ratio is below (<) 0,2, or as being MGCmtif in between (>0,2 and <1); or- if the SCC is a HNSCC of the oral cavity, the MGC status is determined as being MGCHlghif the ratio of the number of MGC per the number of cancer cells is below (<) 1 and equal to or above (>) 0,0006.

7. The method of any one of claims 1 to 6, wherein the method is used in addition for monitoring SCC tumor evolution in a subject; for monitoring the response of a subject to a treatment of SCC; for selecting the appropriate treatment of SCC for a subject in need thereof; for selecting a subject capable of responding to a treatment of SCC; or for selecting or disqualifying a subject having a SCC for enrolment in a clinical trial for the treatment of SCC.

8. A computer-implemented method of training a classifier for assessing the multinucleated giant cells (MGC) status of a squamous cell carcinoma (SCC) tumor, wherein the classifier is trained with images wherein MGC, or both MGC and tumor cells, are annotated, in order for the classifier to provide a “MGC-tumor” ratio of the number of MGC per the number of cancer cells or per mm2of tumor area, thereby assessing the MGC status and allowing to evaluate the prognosis or the response to a therapeutic treatment of a subject having a SCC.

9. The computer-implemented method of claim 8, wherein two distinct classifiers are trained, a first classifier being trained for detecting MGC in a SCC tumor sample, and a second classifier being trained for detecting tumor cells in a SCC tumor sample or for measuring the SCC tumor area.

10. A computer-implemented method of training a classifier for assessing the multinucleated giant cells (MGC) status of a squamous cell carcinoma (SCC) tumor, wherein the method comprises:a) providing a training set of tumor images each tumor being obtained from a subject suffering from a SCC, or preprocessed information obtained from said training set, as input to a classifier, said training set comprising i) images of MGCHlghtumors, obtained from subjects suffering from a SCC known as having a MGCHlghstatus, and ii) images of MGCLowtumors, obtained from subjects suffering from a SCC known as having a MGCLowstatus; b) generating an output of the classifier for each image, said output classifying the tumor image input as having a MGCHlghor MGCLowstatus; and c) evaluating the classifier’s performance for distinguishing between MGCHlghand MGCLowstatus by comparing, for each image, the output of the classifier to the known actual status of the tumor; wherein the classifier is considered as a trained efficient classifier to determine the MGC status of a tumor, if it exhibits an Area Under the ROC Curve (ROC AUC) for each tumor above 0.65.

11. A particular computer-implemented method of training such a classifier for assessing or determining the MGC status of a SCC tumor, comprises the following steps: a)- of providing a training set comprising images of healthy tissue and / or images of SCC tumor, or preprocessed information obtained from said training set, as input to a first classifier module, and / or- of providing a training set comprising images of healthy tissue and / or images of SCC tumor, or preprocessed information obtained from said training set, as input to a second classifier module; b) of generating- an output of the first classifier for each image, said output consisting in an image wherein cells identified by the first classifier as tumor cells are annotated or labelled; and / or- an output of the second classifier for each image, said output consisting in an image wherein cells identified by the second classifier as MGC, or as TREM2HIGH macrophages, are annotated or labelled; and c) of evaluating- the performance of the first classifier for detecting tumor cells by comparing, for each image, the output of the classifier to the known actual identity of the cells, and / or- the performance of the second classifier for detecting MGC, or TREM2HIGH macrophages, by comparing, for each image, the output of the classifier to the known actual identity of the cells;wherein the first trained classifier is considered as efficient for identifying tumor cells if it exhibits a mean Average Precision (mAP) on tumor images from a testing set above 0.2, and wherein the second trained classifier is considered as efficient for identifying MGC, or TREM2HIGHmacrophages, if it exhibits a mean Average Precision (mAP) on tumor images from a testing set above 0.2.

12. The method of claim 10 or 11, wherein the images used to prepare the training set are obtained from a cancerous tumor, the cancer being selected from head and neck squamous cell carcinoma (HNSCC), lung carcinoma, esophagus carcinoma, skin carcinoma and uterus cervix carcinoma, in particular HNSCC.

13. The method of any one of claims 1 to 12, wherein the method uses a classifier trained to assess the MGC status of a tumor with the method according to any one of claims 8 to 12.

14. The method of anyone of claims 8 to 13, wherein the classifier is selected from random forest (RF) classifier, Support Vector Machine (SVM) classifier, decision tree classifier, K- nearest neighbor classifier (KNN), logistic regression classifier, nearest neighbor classifier, Gaussian mixture model (GMM) classifier, nearest centroid classifier, linear regression classifier, a neural network such as an artificial, deep, convolutional or fully connected neural network, and a transformer model.

15. A computing system comprising:- a memory storing at least one instruction of a classifier trained according to the method of anyone of claims 8 to 11, and- a processor accessing to the memory for reading said instruction(s) and executing the method according to any one of claims 1 to 7.

16. A kit characterized in that it comprises a memory storing at least one instruction of a classifier trained according to the method of anyone of claims 8 to 12 on a support, and at least one detection means that specifically recognizes a MGC or TREM2Hlghmacrophage, or that specifically recognizes a CHIT1, FBP1 or TREM2 gene or protein.

17. Use of the kit of claim 16 for analyzing the MGC status of a SCC.

Images