Serum biomarkers in colorectal cancer : a study of a large number of serum markers leading up to prognostic modelling

Background and aims Colorectal cancer (CRC) is one of the most common cancers globally. Different biomarkers are expressed differently depending on the nature of the tumor. The detection of biomarkers is crucial to monitor survival. No superior biomarker has been identified for patient prognostics in CRC, and even carcinoembryonic antigen (CEA) falls short in determining outcomes. Multiplex assays have been developed to detect biomarkers with a higher accuracy than previously, allowing the use of artificial intelligence (AI) models to evaluate the prognosis. Biomarkers interleukin 8 (IL-8), carbohydrate antigen 125 (CA125), carbohydrate antigen 242 (CA242), and carbohydrate antigen 19-9 (CA19- 9) seem to play a role in CRC prognostics, whereas trypsin inhibitor (TATI) and tumor-associated trypsin-2 (TAT-2) are prognostic in other gastrointestinal malignancies. This study aimed to evaluate and validate the prognostic value of biomarkers in CRC and to build prognostic machine-learned survival models. Materials and Methods Preoperative serum samples were gathered from 658 CRC surgery patients at Helsinki University Hospital in 19982005. Proximity extension assays (PEAs) were done on 148 serum samples, from which 92 different immunological and oncological markers were detected. Furthermore, 48 cytokines and chemokines were analyzed from 328 serum samples and machine learning was used to find tentative prognostic markers for prognostic models. CA125, CA242, CA19-9, and CEA were analyzed and validated on 322 CRC patients by enzyme-linked immunosorbent assay (ELISA). TATI, TAT- 2, and CRP were defined for 494 CRC patients with time-resolved immunofluorometric assays (TR-IFMA). Results In survival analyses for 92 biomarkers analyzed using PEA, Kallikrein 13 (KLK13) and Mucin-16 (MUC-16) emerged as independent prognostic markers. Patients with high MUC-16 levels exhibited a poorer prognosis than patients with low MUC-16 levels, whereas patients with low KLK13 levels exhibited a poorer prognosis than patients with high levels. From the 48 analyzed biomarkers, only IL-8 emerged as a significant prognostic factor for CRC. The multivariate model was developed based on all 48 biomarkers using a random survival forest analysis, which yielded two tentative candidate CRC prognostic markers: IL-2Ra and IL-8. The findings for MUC-16 (the same molecule as CA125) and CA125 correlated strongly. Further validation of CA125 confirmed its significance and its independent prognostic value as a biomarker for CRC patients. Conclusions In CRC, serum CA125 (MUC-16) emerged as a strong significant and independent negative prognostic factor. The multivariate machine-learned model served as a promising calculator, and such tools such will most likely enjoy further clinical use.