Lipid remodeling in serum and tumor margins tracks colorectal cancer progression

The Blood and the Tumor Keep Different Fat Ledgers: Study Maps How Colorectal Cancer Reshapes the Body’s Lipid Chemistry

Colorectal cancer does not merely rewrite the genome of the cells it invades — it quietly reshuffles the fat chemistry of the entire body, and it does so in strikingly different ways depending on where you look. Writing in the Journal of Translational Medicine, researchers from Wroclaw Medical University and collaborating Polish clinical centers report that lipid molecules drifting through the bloodstream of colorectal cancer patients tell a very different story from the lipids embedded inside the tumor itself and in the healthy-looking surgical margin that surrounds it. By simultaneously profiling preoperative serum and paired tumor–margin tissue from 122 patients spanning every stage of the disease, the team uncovered fat-molecule signals that rise and fall in step with advancing tumor stage, spread to lymph nodes, and even invasion into blood vessels and nerves. The findings stop short of a ready-made diagnostic test, but they sketch one of the most detailed maps to date of how the body’s lipid economy is remodeled as colorectal cancer progresses.

Lipids are far more than biological padding. They form the membranes that enclose every cell, serve as dense energy reserves, and act as potent signaling molecules that switch cellular programs on and off. Tumors exploit all three functions: dividing cells must manufacture new membranes at a furious pace, fast-growing tissue burns fatty acids as fuel, and cancer cells deploy lipids such as prostaglandins and ceramides to dampen immune attacks and survive stress. This is why metabolic reprogramming, the rewiring of how cells acquire and use nutrients, is now considered a hallmark of cancer. Yet most lipidomic investigations examine a single biological compartment at a time, typically blood or tumor tissue, and track individual molecular species one by one. What such work has rarely delivered is a systematic, stage-by-stage comparison of the circulating and local lipid landscapes within the same patients, and that is precisely the gap this cross-sectional study of colorectal cancer stages I through IV set out to address.

The cohort comprised 122 patients with histologically confirmed colorectal cancer ranging from stage I to stage IV, with blood drawn before any surgical intervention and serum separated for analysis. During surgery, researchers collected two tissue samples from each patient: the tumor itself and an adjacent, tumor-free margin of the same tissue. This paired design is analytically powerful because it lets each patient serve as their own reference: the difference between tumor and margin captures changes specific to the malignant tissue, while the margin itself characterizes the surrounding peritumoral field. The study was approved by the Ethics Committee of Wroclaw Medical University, conducted in accordance with the Declaration of Helsinki, and included written informed consent from every participant. Crucially, the work is framed as cross-sectional — a snapshot of patients at different disease stages rather than a follow-up of the same individuals over time — which shapes how its conclusions can be used.

To read the lipid code hidden in these samples, the team used untargeted lipidomics based on ultra-high-performance liquid chromatography coupled to quadrupole time-of-flight mass spectrometry. In this workflow, each sample is pushed through a column packed with microscopic stationary-phase particles; as thousands of lipid molecules traverse it, they separate according to subtle differences in chemical affinity. The emerging stream is ionized by electrospray and fed into the mass spectrometer, which measures the exact mass-to-charge ratios of molecules and their fragments, generating thousands of discrete features per run. Because the approach is untargeted, it makes no assumptions about which lipids will appear, capturing the full chemical landscape rather than a predetermined panel. The analysis ran in both negative and positive ionization modes, a detail that matters because each mode favors a different chemical family. Negative ionization efficiently captures acidic phospholipids and related species, while positive ionization is better suited to neutral fats such as triglycerides; running both in parallel yields a far more complete census of the lipidome than either mode alone.

Where the study departs from convention is in how it interprets that mountain of data. Rather than chasing thousands of individual lipid species, each vulnerable to noise and to ambiguity in structural identification, the researchers aggregated feature-level signals using the MS-DIAL lipid ontology framework, which assigns detected molecules to chemically coherent classes and subclasses. The result is a set of ontology-level signals, each summarizing the collective behavior of an entire lipid domain rather than a single molecule. These signals were then tested against clinical staging with models matched to the type of outcome: ordinal proportional-odds models for the ordered TNM, T and N categories, and logistic regression for the binary outcomes of angioinvasion and neuroinvasion. In an ordinal proportional-odds model, an odds ratio above one means that a higher lipid signal raises the odds of landing in a more advanced disease category, while a ratio below one signals the opposite. Every model was computed both unadjusted and after statistical adjustment for age and sex, the two demographic factors most likely to confound lipid measurements.

The circulating-blood results arrived in two flavors. In negative ionization mode the associations were predominantly inverse: stronger signals of particular lipid ontologies tended to accompany less advanced disease. The clearest example was MMPE, a methylated derivative of the membrane phospholipid phosphatidylethanolamine. Patients with higher serum MMPE signals had 27 percent lower odds of falling into a more advanced TNM category (adjusted odds ratio 0.73, 95 percent confidence interval 0.55 to 0.98; nominal p = 0.034) and 28 percent lower odds of a higher N stage, which reflects lymph-node involvement (adjusted OR 0.72, 95 percent CI 0.53 to 0.99; nominal p = 0.042). In other words, the more abundant this modified phospholipid family was in the blood, the less likely the cancer had climbed the staging ladder, an intriguing hint that some circulating membrane lipids track, or even oppose, aggressive disease.

Positive ionization mode told the opposite story. Two triglyceride-related domains rose with deeper tumor growth into the bowel wall: an esterified triglyceride class labeled TG-EST was associated with higher T stage (adjusted OR 1.45, 95 percent CI 1.06 to 1.96; nominal p = 0.018), and ether-linked triglycerides, a structurally unusual fat family in which an ether bond replaces one ester linkage, showed an even stronger association (adjusted OR 2.26, 95 percent CI 1.08 to 4.74; nominal p = 0.031). A further ontology class abbreviated BRSE was linked both to more advanced overall TNM stage (adjusted OR 1.45, 95 percent CI 1.05 to 2.00; nominal p = 0.024) and to higher N stage (adjusted OR 1.51, 95 percent CI 1.08 to 2.11; nominal p = 0.016). The contrast is conceptually important: phospholipid-like signals in negative mode pointed downward with severity, while storage-fat signals in positive mode pointed upward, a two-directional pattern that studies confined to a single chemical family would miss.

The tissue results may be the most biologically provocative of all. Here the team modeled both the adjacent-margin background and the within-patient tumor-versus-margin difference, asking which lipid domains distinguish malignant tissue from its own surroundings. Two signals stood out. A higher tumor-versus-margin bile acid signal was associated with more advanced TNM stage (adjusted OR 1.70, 95 percent CI 1.05 to 2.75; nominal p = 0.031) and with roughly a doubling of the odds of angioinvasion, the invasion of blood vessels that gives tumors their principal route to seed distant metastases (adjusted OR 1.99, 95 percent CI 1.13 to 3.52; nominal p = 0.018). The finding resonates with a growing body of work on bile acids, cholesterol-derived molecules modified by gut microbes, as signaling agents that can influence inflammation and epithelial proliferation in the intestine. Separately, a tumor-versus-margin signal of a specialized ceramide subclass, labeled Cer_EODS, tracked higher T stage (adjusted OR 1.65, 95 percent CI 1.14 to 2.40; nominal p = 0.008) and higher N stage (adjusted OR 1.71, 95 percent CI 1.16 to 2.51; nominal p = 0.006). Ceramides sit at the center of sphingolipid signaling, governing programmed cell death and membrane architecture, which makes their local remodeling a plausible accomplice of tumor progression.

The authors are careful about how far these numbers can be pushed. All p values are nominal, with no correction for the multiple statistical tests performed across dozens of lipid ontologies, so some associations would be expected to appear by chance alone. Even where confidence intervals exclude the null value, they are often wide, and thresholds of statistical significance must be read with corresponding caution. The cross-sectional design captures each patient at a single moment, so it cannot establish whether lipid changes drive progression or merely accompany it. Age and sex were accounted for, but other confounders such as diet, medication and the gut microbiome were not. Most importantly, the conclusions are framed as hypothesis-generating: the identified lipid domains are candidate signals for targeted, species-resolved validation in independent cohorts, not confirmatory biomarkers. That candor is what separates this work from premature claims of a blood test for cancer staging. What the study does establish, with unusual clarity, is methodological: that aggregating untargeted lipidomics data at the ontology level can extract stage-related, clinically coherent signals from both blood and tissue within a single analytical framework.

The research is embedded in a larger European effort. It was funded by the European Union through the NextGenerationEU program as part of Poland’s National Recovery and Resilience Plan, with support from the Medical Research Agency, within a project developing a colorectal cancer risk model for patients with inflammatory bowel disease based on omic profiles of serum and tissue. That context matters because people with inflammatory bowel disease face elevated colorectal cancer risk, and stratifying them accurately remains an unsolved clinical problem. If the serum and tissue lipid domains identified here survive validation in longitudinal, species-resolved studies, they could eventually complement colonoscopy and standard staging by flagging biologically aggressive disease or monitoring progression noninvasively. For now, the paper’s central message is one of structured complexity: colorectal cancer progression is accompanied not by a single uniform lipid signature but by compartment-specific remodeling, in which the blood, the tumor, and the margin around it each keep their own molecular accounts. Disentangling those accounts, the authors suggest, may prove a decisive step toward translating the lipidome into clinical practice.