NEW YORK – A team led by investigators at Fudan University has outlined a proteomics and machine-learning strategy to find and validate apparent blood-based biomarkers for colorectal cancer (CRC) using data on proteins found in extracellular vesicles in the blood.
Their diagnostic model, dubbed ColonTrack, demonstrated "superior diagnostic accuracy and sensitivity for CRC, particularly in early-stage disease detection," senior and co-corresponding author Hao-Jie Lu, with Fudan University, and colleagues wrote in Cell Reports Medicine on Wednesday, suggesting the model "could become a valuable tool for improving cancer diagnosis and management."
For their retrospective study, the researchers used mass spectrometry and data-independent acquisition mass spec-based proteomics to profile thousands of proteins in extracellular vesicles isolated from tissue and matched blood plasma samples collected from 40 individuals with CRC and 20 unaffected control individuals. They focused in on 21 candidate biomarkers for CRC that were subsequently assessed in samples from 20 more CRC cases and 20 unaffected control individuals, using parallel reaction monitoring mass spec (PRM-MS) and machine learning.
With targeted PRM and machine learning, the team focused in on half a dozen potential markers for CRC: CTTN, HNRNPK, PSMC6, MYH9, NAP1L1, and EIF3B. That set was whittled down further through enzyme-linked immunosorbent assay (ELISA)-based testing on samples from 519 individuals with CRC, 243 unaffected controls, and 264 cases of benign conditions such as inflammatory bowel disease or advanced colorectal adenoma.
From there, a set of three markers was used to validate the ColonTrack machine-learning model for identifying early-stage CRC in hundreds more individuals with or without CRC or a benign colorectal condition who participated in the external or benchmarking cohorts for ColonTrack.
The integrated proteomic analysis of plasma and tissue EVs led to a three-protein panel — CTTN, HNRNPK, and PSMC6 — and the ColonTrack model based on this panel exhibited "outstanding diagnostic performance for CRC," the authors reported.
When they compared ColonTrack's performance to the CRC marker mSeptin-9 in 126 individuals with stage I to III CRC, 68 individuals with advanced colorectal adenoma cases, and 68 unaffected controls, the researchers found that it was particularly informative when it came to finding stage II or III cases. While the use of mSeptin-9 made it possible to pick up just over 96 percent of CRC cases, for example, the detection rate was 94.3 percent for ColonTrack.
That jumped to 98.2 percent detection for stage II CRC and 95.7 percent detection for stage III cancer, compared to 78.6 percent and 86.9 percent detection rates, respectively, with the mSeptin-9 marker.
In a broader patient set that spanned 431 CRC cases from several cohorts considered in the study, ColonTrack had similar or improved performance for detecting CRC compared to mSeptin-9 for stage I, II, or III CRC. On the other hand, ColonTrack was edged out by mSeptin-9 when it came to detecting stage IV CRC cases. There, mSeptin-9-based approaches picked up all of the stage IV cases, while ColonTrack detected nearly 82 percent.
"The observed differences in the performance of ColonTrack and mSeptin-9 across various stages of CRC emphasize the complementary roles these biomarkers could play in clinical practice," the authors reported. "While ColonTrack excels in early CRC detection, mSeptin-9's high sensitivity for advanced-stage CRC further underscores the importance of combining multiple diagnostic tools to achieve more comprehensive and reliable CRC screening."
The authors noted that there is a need to dig into the proposed plasma markers further to determine whether they are specific to CRC tumors or reflect indirect changes related to the disease, and to explore their potential performance in CRC patients from other human populations.
"Further research into the biological mechanisms underlying these biomarkers is necessary to better understand their role and to guide clinical management and the development of potential drug targets," the authors wrote, noting that the work "would benefit from validation in larger, prospective cohorts to assess the diagnostic model's applicability and robustness across different patient populations."