Association of endogenous and exogenous factors with early-onset colorectal cancer: germline mutations, epigenetic modifications, diet and lifestyle habits

The incidence of early-onset colorectal cancer (eoCRC), defined as colorectal cancer occurring in young adults under the age of 50, is increasing globally. However, knowledge of the etiological factors and characteristics of CRC in young adults is far from complete. Estimating the impact of pathogenic variants (PVs) in eoCRC predisposition remains an active field of research. Therefore the 1st aim was to evaluate the associations of germline PVs and family history of CRC with eoCRC. A total of 105 eoCRCs were enrolled (mean age at diagnosis of 41.2 ± 6.7 years; 48.6% females, 51.4% males) for genetic testing through next-generation sequencing and multiplex ligation-dependent probe amplification. 20% of eoCRC carried a germline PV of genes known to be associated with CRC, of which 12.4% of mismatch repair genes, 2.9% of BRCA1-2, 3.8% of MUTYH, 0.9% of ATM, 0.9% of SDHAF2. One patient exhibited mosaicism of PVs in MSH2/MUTYH genes. 71.4% of eoCRCs didn’t have a family history of CRC; 19% of eoCRCs reported having a first degree relative (FDR) with CRC and 12.4% had a second degree relative (SDR) with CRC; three patients had both a FDR and SDR with CRC and were all Lynch patients. When comparing mutated-eoCRCs with non-mutated eoCRCs, no statistically significant differences were found in terms of age at diagnosis or sex, location of CRC, presence of FDR with CRC. Mutated eoCRC differed significantly from non-mutated eoCRCs in terms of SDRs with CRC (p < 0.001). In conclusion, 20% of eoCRCs are caused by germline pathogenic variants (PV) and a negative family history does not exclude hereditary cancer syndromes. Therefore, thorough family history should be routinely collected for all individuals with eoCRC and all patients with eoCRC should be offered multi-gene panel germline genetic testing. Identification of LS individuals as well as carriers of other relevant germline PVs (e.g. BRCA2, MUTYH) will allow at-risk individuals to take personalised preventive measures for both proband and relatives. Questionable eoCRCs' exogenous risk factors are represented by diet and lifestyle, even though a still scant literature is available to date. Moreover, most studies are small, heterogeneous, focused exclusively on peculiar dietary and drinking habits of single countries without analyzing cooking, processing, and storage techniques. Therefore, our 2nd aims were to (i) develop a unique and shared semi-quantitative food frequency questionnaire (SQFFQ) able to accurately describe dietary and drinking habits of eoCRCs and healthy controls of different countries at global level that will be involved in the future DEMETRA study (international case-control study evaluating the association of dietary, lifestyle and anthropometric factors with eoCRC of countries with different eoCRC incidence); (ii) validate the SQFFQ, making data obtained from different dietary questionnaires comparable. We designed an ad-hoc, shared, online SQFFQ to investigate the usual consumption of 329 foods, grouped into 61 food groups, over the past year. In addition to frequency of consumption, the tool investigates the portions habitually consumed using validated photographs, household measures, and standard units, as well as types of seasoning and methods of cooking. A special software was then developed to analyze responses and link them to food composition tables in order to provide a nutritional breakdown of individual and collective diets. The SQFFQ was then validated for repeatability by administering it twice, 3 weeks apart, to a sample of 30 young adults under 50 years (Internal Validation). To evaluate repeatability, the measurement error for each food group was estimated as the percentage change between the estimates of food consumption for the same individual. Afterwards, the agreement between the two measurements for each food group was measured with Cohen's kappa coefficient. Agreement levels, represented by the calculation of Cohen's kappa coefficient, were as follows: 12 food groups showed fair/sufficient agreement (Cohen's kappa 20-40), 23 foods exhibited good/moderate agreement (Cohen's kappa >40-60), and 22 food groups demonstrated high substantial agreement (Cohen's kappa >60). Therefore, the ad hoc designed SQFFQ provides a reasonably repeatable measure of dietary intake and can be used to assess the dietary and drinking habits of volunteers in this age group. Indeed, most staple foods in the Italian diet, including pasta, fruit, vegetables, legumes, eggs, meat, coffee, and tea, are well estimated by the SQFFQ. However, as yet described in literature, challenges persist in estimating the consumption of foods assumed sporadically (such as snacks) or in small quantities such as spices. Definitive conclusions will be drawn after the completion of the ongoing External validation, involving 100 volunteers from the same age group. In this phase, the SQFFQ will be validated against the gold standard, represented by a 4 days-food diary followed by a dietary recall. Once validation is complete, the international, multicenter, case-control study will start to evaluate the associations of diet, lifestyle and anthropometric factors with eoCRC comparing patients from countries with different incidence of eoCRC. Epigenetic changes are crucial in the pathogenesis of CRC, representing the missing link between CRC, specific gene expression patterns and the absence of genetic alterations. One of the most important epigenetic modifications involved in carcinogenesis is represented by an altered expression of microRNAs (miRNAs). Distinct miRNAs could be differentially expressed in patients with recurrent vs. non-recurrent eoCRC and used as simple predictive biomarkers to select the optimal post-treatment surveillance regimen in managing patients with stage I-III eoCRC. Therefore, the 3rd aim was to identify candidate miRNAs that were differentially expressed in eoCRC patients with and without recurrence and define which patients could benefit most from more aggressive surveillance. We employed a five-layer approach for the development of a simple and clinically feasible test that may be translated into clinical practice. The first phase of the study (Discovery) consisted in the systematic interrogation and profiling of miRNA expression levels in 20 formalin-fixed paraffin-embedded (FFPE) samples of stage II-III eoCRCs that did (n 10) or did not (n 10) develop recurrence in five years following curative-intent surgery. In phase two (Assay development), we performed several bioinformatic analyses to identify the best candidate miRNAs that were differentially expressed in eoCRCs with and without recurrence and provided the highest discriminatory power between the two groups. At the end of phase two, we selected 10 best performing miRNAs and quantified their expression via qPCR. This third phase utilized 88 FFPE from a larger, independent training cohort of stage I-III eoCRC who received curative-intent surgery (24 recurrent and 63 non-recurrent eoCRCs). Because there is no unique and universally accepted normalized miRNA, we employed several bioinformatic approaches to rigorously establish the ideal candidate based on intra- and inter-group expression stability. In the Assay Training phase, we optimized and trained an advanced machine learning algorithm (XGBoost) to predict the development of eoCRC recurrence based on RT-qPCR data and performed several interrogations to the model to understand its functioning. The optimized XGBoost-based 9-miRNAs risk-assessment model demonstrated a high accuracy in predicting recurrence in stage I-III eoCRCs in the training cohort with an AUC value of 0.90 (95% CI 83-95%) with a Youden index of 64.9% (CI 95%, 55%-82%), an accuracy of 81.8% (77-93%), sensitivity 84.0% (65-96%), specificity 81.0% (72-98%). We then evaluated the survival characteristics of the XGB model and observed that our 9-miRNA model can discriminate effectively between recurrent and non-recurrent cases up to 20 years after surgical resection: patients predicted to be at a high risk of recurrence by the XGB model had a statistically significant cumulative hazard of disease recurrence than those classified as low-risk (p<0.001). Finally, we performed an independent validation of our assay in a distinct and ethnically different validation cohort of 69 FFPE of stage I-III eoCRCs who received curative-intent surgery for eoCRC (9 recurrent and 60 non-recurrent). The XGBoost-based risk-assessment model, incorporating 9-miRNAs, exhibited good accuracy in predicting recurrence among stage I-III eoCRCs in the validation cohort, achieving an AUC value of 0.77 (95% CI 67.0-87.0%), with a sensitivity of 100% (88-100%), specificity of 62.9% (55-82%), accuracy 66.7% (59.0-83.0%), and Youden index of 62.9% (54-73%). To truly gauge the effects that this assay would have in a real-world scenario, we performed several decision-curve analyses. We observed a net benefit of the surveillance based on our 9-miRNA signature compared to the clinical-based surveillance especially in stage I and II high risk eoCRC, representing the subgroups of patients that could benefit more from more aggressive post-treatment follow-up strategies. Therefore, if confirmed in prospective trials, this 9-miRNA signature could establish the fundamentals of personalized medicine.