Microbiome-based Diagnostic Tool for the Screening of Colorectal Cancer (GUILTI)

Overview

Colorectal cancer (CRC) is one of the most common cancer and cause of cancer death worldwide. Population-based screening programs for average risk populations have proven effective in reducing both incidence and mortality of CRC through early detection of cancer. The fecal immunochemical testing (FIT), has still a suboptimal diagnostic yield, with both missed adenomas and, mainly, unnecessary colonoscopies.The identification of novel, non-invasive biomarkers is currently one of the research areas driving most expenditure forces in the field of CRC.A large body of evidence shows that alterations of the gut microbiome and the enrichment of specific taxa(e.g. Fusobacterium nucleatum, Parvimonas micra, and others) are involved in the pathogenesis of CRC. Moreover, recent studies, have discovered common microbial signatures able to reproducibly discriminate between patients with CRC and healthy controls.The goal of this observational study to develop a gut microbiome based diagnostic tool for the identification of CRC and advanced colorectal adenomas in patients enrolled in the national colorectal cancer (CRC) screening program (50-69 year-old) and among who refer to all centers involved in this study for screening colonoscopy with positivity of FIT, of both sex. The primary endpoint of the study is to develop a gut microbiome-based diagnostic tool for the identification of CRC and advanced colorectal adenomas in patients involved in the national CRC screening program, using both statistical and machine learning approaches. The secondary endpoints are: * The association of clinical and colonoscopy outcomes with FIT results; * The characterization of gut microbiome from an ecological, taxonomic, phylogenetic and functional point of view; * The association between microbiome signatures with clinical and colonoscopy outcomes, through statistical and machine-learning algorithms. At baseline, enrolled patients will provide a fecal sample within 2 weeks from enrollment and demographic, clinical characteristics and laboratory data will be recorded. Enrolled patients will be scheduled for colonoscopy, as for clinical practice, within 4 weeks from the positive FIT and histology of resected lesions will be assessed by experienced pathologists according to the WHO classification and the Vienna criteria. Clinical, endoscopic and microbial data will be combined through statistical and machine learning algorithms to identify specific microbial biomarkers associated with CRC and develop a new diagnostic tool, based on a scoring system. This tool will be validated, and its diagnostic performances will be compared with traditional screening methods.

Colorectal cancer (CRC) is the 3rd most common cancer and the 2nd most common cause of cancer death worldwide, with nearly 2 million new cases and one million deaths in 2020. In the last decades, a population-based screening programs for average-risk populations have established as an effective strategy to reduce both incidence and mortality of CRC through early detection of cancer. The fecal immunochemical testing (FIT), the reference diagnostic tool in most countries, has still a suboptimal diagnostic yield, with both missed adenomas and, mainly, unnecessary colonoscopies. For these reasons, in the last years the identification of novel, non-invasive biomarkers is absolutely advocatein the field of CRC. A large body of evidence shows that alterations of the gut microbiome and the enrichment of specific taxa (e.g. Fusobacterium nucleatum, Parvimonas micra, and others) are involved in the pathogenesis of CRC. Moreover, recent studies, including metagenomic meta-analyses from our group, have discovered common microbial signatures able to reproducibly discriminate between patients with CRC and healthy controls. Based on this evidence, international guidelines have recently advocated the exploitation of microbiome-based biomarkers for the screening of CRC in clinical practice, but such studies are not yet available to date. The primary objective is to develop a gut microbiome-based diagnostic tool for the identification of CRC and advanced colorectal adenomas. The secondary objectives are: * To associate FIT with clinical and colonoscopy outcomes * To characterize gut microbiome of enrolled patients * To associate microbiome signatures with clinical and colonoscopy outcomes. This is an observational prospective multicenter study, in which patients will be selected among those enrolled in the national colorectal cancer (CRC) screening program and among who refer to all centers involved in this study for screening colonoscopy. Patients with all inclusion criteria and none of the exclusion criteria (detailed in the specific section of this website) will be considered for this study.N=1202 patients will be enrolled, based on sample size calculation and including a validation group. N=911 patients will be needed, and we will add 91 patients to cover a 10% potential drop-out and 200 patients as a validation group (20% of the study cohort). Patients enrolled in this validation cohort will have the same exclusion and inclusion criteria of other patients and will undergo the same study procedures. At baseline all enrolled patients will provide a fecal sample (collected using a buffer for genome preservation) within 2 weeks from enrollment and will be stored at -80°C at each clinical center and assigned de-identified IDs. Moreover demographic, clinical characteristics and laboratory data will be recorded. For all enrolled patients clinicians will record the following data: * Familiar history of CRC; * Comorbidities; * Drug intake; * Gastrointestinal (GI) and alarm symptoms (iron-deficiency anemia, hematochezia, unexplained weight loss, abdominal pain, sudden change in bowel habits); * Dietary habits; * Smoking and alcohol consumption; * BMI (body mass index); * Information about previous FIT and/or colonoscopies and/or virtual colonoscopies. All enrolled patients will undergo a colonoscopy within 4 weeks from the positive FIT, after colonoscopy endoscopic characteristics (size, site, shape based on Paris class.) and histopathological characteristics of detected lesions will be collected. Advanced colorectal adenomas will be defined as adenomas larger than or equal to 10 mm, and/or with villous components higher than or equal to 25%, and/or high-grade dysplasia.Moreover, patients diagnosed with CRC, will also undergo a total-body CT scan to stage the disease and will be referred to an oncological pathway in clinical practice, and the following data will be collected: presence of lymph node metastasis and of organ metastasis, TNM (Tumor-Node-Metastasis) class..The enrollment phase will last 48 months.At the end of the enrollment phase, the investigators will perform the microbiome analysis, and the investigators will combined clinical, endoscopic and microbial data through statistical and machine learning algorithms to identify specific microbial biomarkers associated with CRC and advanced colorectal adenomas, to develop a new diagnostic tool, based on a scoring system. Study Outcomes are detailed in the specific section of this website. Microbiome analysis will be performed with shotgun sequencing techniques. Moreover to obtain a panel of microbial species mostly associated with CRC and advanced colorectal adenomas, the investigators will exploit machine learning applications, for instance, using the widely adopted Random Forest algorithms, as well as meta-analytical approaches integrating our large cohort with previously published metagenomic cohorts. A defined panel of microbial species mostly associated with CRC and advanced colorectal adenomas will provide a comprehensive risk score based on microbiome features, with two specific goals, to improve the accuracy testing of the FIT diagnostic tool and to be easily interpretable by clinicians.All the collected clinical data will be statistically combined to rank microbial features from the most to least associated. These rankings will be exploited to evaluate a microbiome profile from a new individual, and to report how many of the defined species in the previous panel are found, and hence linked with CRC and advanced colorectal adenomas. Associations between identified species and makers of interest will be estimated through robust statistical analysis methods, such as partial correlation or linear modeling. Then the associations will be ranked and compared across markers for the same species, giving a priority score for each species. Particular care will be dedicated to identifying potential confounding markers known to be associated with differences in microbiome composition, for instance, age, that instead should be used as covariates in the statistical modeling. In addition to statistical modeling, associations between species and markers will also be evaluated via ML approaches, particularly those providing feature importance scores for the input features (species, in our case), that can be used to rank the species across the different markers (e.g. Random Forest, LASSO, etc). Continuous variables will be reported as mean ± standard deviation or as median and interquartile range (IQR), and categorical variables were summarized as frequency and percentage. Comparisons of variables will be made by t-test (or Mann-Whitney/Kruskal-Wallis), Chi-square test, or Fisher\'s exact test, as appropriate. A p-value \<0.05 will be considered to indicatestatistical significance. Logistic regression models will be performed to identify the presence of variables independently associated with outcomes. Variables considered in the models were selected through stepwise model selection and guided by clinical relevance. All statistical analyses were performed using SPSS v. 28.0 for Macintosh (SPSS Inc., Chicago, USA). The investigators will employ a Machine Learning (ML) framework based on the scikit-learn Python package using a cross-validation approach with 100 bootstrap iterations and an 80/20 random splitting into training and testing folds. Classification and regression ML algorithms will be trained on both microbiome species-level taxonomic relative abundances and functional potential profiles. Species-level taxonomic abundances will be estimated by MetaPhlAn 4 (using the latest databases available, currently named Oct22) and normalized using the arcsin-sqrt transformation for compositional data. Functional potential profiles will be estimated using the latest HUMAnN 4 version and the investigators will consider both abundance estimations of single microbial gene families and of metabolic pathways. Partial correlations between species and markers will be computed using the Spearman index, to avoid biases due to outlier values and corrected for covariates, such as sex, age, and body-mass index, known to have an impact on microbiome composition. Partial correlation values will be corrected using the Benjamini-Hochberg procedure for the false-discovery rate.