KAIST Team Maps Amino Acid Network to Predict Colorectal Cancer Recurrence

A single preoperative blood sample may help improve the prediction of recurrence or metastasis in patients with colorectal cancer. A joint research team from KAIST, Gangnam Severance Hospital, and Asan Medical Center has shown that, as colorectal cancer advances, the network of relationships among circulating amino acids—a kind of metabolic map—undergoes systematic remodeling. Building on this finding, the researchers developed an analytical method that showed higher predictive performance than a carcinoembryonic antigen (CEA)-only model and models based solely on individual amino acid levels.

Using this new framework, the team showed that the circulating amino acid network undergoes stage-dependent remodeling that reflects systemic metabolic reprogramming, then used these network-derived features to build a new strategy for predicting recurrence or metastasis.

Cancer cells require large amounts of nutrients to grow and proliferate, and amino acids serve as building blocks for proteins as well as fuel for energy production and DNA synthesis. Colorectal cancer is marked by pronounced changes in amino acid metabolism that extend beyond tumor tissue into the bloodstream, making blood amino acids a metabolic biomarker of interest. Previous research had focused mainly on individual amino acid concentrations rather than how amino acids are interconnected.

Using fluorine-19 nuclear magnetic resonance (¹⁹F NMR) spectroscopy, the team simultaneously quantified 18 circulating amino acids in a small serum sample and analyzed both their relative abundance and their relationships as a network. As colorectal cancer progressed, the proportion of branched-chain amino acids such as valine and leucine decreased, while glycine and serine, which cancer cells need to proliferate rapidly, increased. Glycine stood out: despite being actively used by proliferating cancer cells, its relative blood abundance rose rather than fell, and a glycine-centered interaction pattern emerged as further evidence of systemic remodeling. 

The team applied these pairwise amino acid interaction features to machine-learning models identifying patients with recurrence or metastasis. In nested cross-validation, the correlation-based model outperformed a CEA-only model, and a combined model incorporating CEA, individual amino acid levels, and interaction-derived features achieved the highest overall performance, also outperforming a model based on individual amino acid levels alone.

"We hope this will lead to new precision medicine technologies that can predict recurrence risk more accurately using a blood sample alone and help establish personalized treatment strategies," said Ji Min Lee, senior author of the study published in Advanced Science.