Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Mar 9.
Published before final editing as: Gut. 2025 Dec 24:gutjnl-2025-335618. doi: 10.1136/gutjnl-2025-335618

Metabolomic pattern of ultra-processed food intake and its association with colorectal cancer risk

Mengxi Du 1,2,3, Xinyu Wang 1, Dong Hang 4, Fenglei Wang 2, Yujia Lu 1, Kai Wang 1,3, Alaina M Bever 1,5, Ana Nogal 1, Danielle E Haslam 2,6, Shuji Ogino 1,7,8,9,10, Jeffrey A Meyerhardt 11, Liming Liang 1,12, Qi Sun 1,2,6, Curtis Huttenhower 9,10,12,13, Andrew T Chan 3,10,13, Frank B Hu 1,2,6,10, Mingyang Song 1,2,3,10
PMCID: PMC12967296  NIHMSID: NIHMS2144676  PMID: 41443985

Abstract

Background

High ultra-processed food (UPF) intake has been linked to colorectal cancer (CRC), but underlying mechanisms remain unclear.

Objective

To evaluate a metabolomic pattern of UPF intake and its association with CRC risk.

Design

Integrating food frequency questionnaire data and high-throughput metabolomic profiling in 1,740 participants (mean age at blood draw: 59.9 years; >95% non-Hispanic whites) from nested case-control studies within the Nurses’ Health Study and Health Professionals Follow-up Study, we derived and validated a UPF-related metabolomic pattern as a weighted sum of metabolites selected via elastic net regression with 10-fold cross-validation. We evaluated prospective associations of this pattern and individual metabolites with CRC risk using multivariable conditional logistic regression in 686 pairs of incident CRC cases and matched controls.

Results

Among 222 metabolites, we constructed a UPF metabolomic pattern comprising 50 metabolites, primarily lipids and amino acids, with 22 positively and 28 inversely associated with total UPF intake (pattern vs. intake: Spearman rho = 0.35). The pattern was associated with higher CRC risk (highest vs. lowest quintile: odds ratio [95% confidence interval]: 1.71 [1.15–2.43], p-trend = 0.002). Correlations of individual metabolites with UPF intake were moderately aligned with their associations with CRC risk (rho = 0.50). N2, N2-dimethylguanosine, a marker of meat/poultry intake, was positively associated with CRC risk (1.96 [1.27–3.03]), while 21-deoxycortisol, related to cortisol biosynthesis, was inversely associated (0.59 [0.41–0.85]).

Conclusion

We developed a UPF metabolomic pattern. The pattern and several metabolites were associated with CRC risk, providing biological insights into potential pathways underlying the UPF-CRC relationship.

Keywords: metabolomics, ultra-processed foods, colorectal cancer, prospective cohorts

INTRODUCTION

Colorectal cancer (CRC) remains one of the most prevalent cancer types and the second leading cause of cancer deaths in the US.1 It is intricately tied to suboptimal diets.2 Ultra-processed foods (UPFs), which make up nearly 60% of Americans’ daily calorie intake,3 have emerged as a potential risk factor for CRC.4,5 High UPF intake has been associated with a higher risk of CRC, CRC precursors, and clinical outcomes after CRC diagnosis.6–8 A recent systematic review and meta-analysis of four large prospective cohorts reported an 11% increased risk of CRC associated with high UPF intake.5 While the strength and consistency of evidence vary, these findings raise important concerns regarding the potential role of high UPF intake in CRC development. It is also recognized that not all UPFs confer equal risk9 – some fiber-rich, plant-based UPFs, such as whole-grain bread or cold breakfast cereals, may have favorable effects. Given the widespread intake of UPFs and the potential for heterogeneous health effects across subtypes, elucidating the underlying biological pathways linking UPF intake to CRC risk is critical to refine dietary guidance and CRC prevention strategies.

High UPF intake may increase CRC risk by introducing carcinogens, such as N-nitroso compounds and acrylamides in processed meat.10–13 Randomized controlled trials and observational studies have associated high UPF intake with weight gain, possibly leading to obesity, an established risk factor for CRC.14–16 More recently, emerging evidence suggests that food additives, including emulsifiers, artificial sweeteners, coloring agents, and nanoparticles, can affect CRC risk and other gastrointestinal diseases by altering the gut microbial composition and gut permeability and thereby triggering inflammation.17–21 Of note, each of these mechanisms is closely related to changes in circulating metabolites.22–24 Therefore, a more in-depth understanding of the metabolic responses to UPF intake can help decipher the biological pathways that underlie the UPF-CRC relationship.

In the current study, we integrated high-throughput metabolomic profiling with epidemiologic data collected from two large US cohorts – the Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS) – to develop and validate metabolomic patterns associated with intake of total UPFs and UPF subgroups and to prospectively evaluate their associations with CRC risk in a nested case-control study.

MATERIALS AND METHODS

Study Design and Participants

The study design and participant selection are summarized in Supplementary Figure 1. Details of the NHS and HPFS cohorts have been documented elsewhere.25,26 Briefly, the NHS enrolled 121,700 female nurses, aged 30–55 years old, in 1976, and the HPFS enrolled 51,529 male health professionals, aged 40–75 years old, in 1986. Cohort participants completed self-administered questionnaires regarding medical conditions, family history, and lifestyle every two years and reported dietary intakes via validated FFQs every four years.27,28 A subset of participants, including 32,826 NHS participants, provided blood samples between 1989 and 1990, and 18,225 HPFS participants provided blood samples between 1993 and 1995. Participants who provided a blood specimen were generally similar to those who did not in terms of demographics and lifestyle, although they were more health-conscious (e.g., more likely to be non-smokers, having a normal BMI, etc.) (Supplementary Table 1).29 These differences should not influence the internal validity of comparisons between cases and controls who provided blood specimens.

To derive the metabolomic pattern of UPFs in the training set, we included participants of the NHS and HPFS who had available metabolomic profiling data from prior nested case-control studies of ovarian cancer and Parkinson’s disease (n=1,740). We selected these case-control studies to retain a wide range of metabolites available in the cohorts while minimizing variations introduced by other disease case-control pairs that may potentially bias the association between the pattern and CRC risk. We then validated the pattern and evaluated its association with CRC risk in a nested case-control study of 686 pairs of CRC cases and matched controls (testing set).

Ethics Statement

The study protocol was approved by the institutional review boards (IRB) of Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, and those participating cancer registries as required (IRB Protocol Number P10372).

Assessment of Ultra-processed Food Intake

In NHS and HPFS, participants reported foods and drinks that they typically consumed over the past year using validated FFQs with ~130 food items.27,28 The FFQs captured the frequency of food consumption in nine categories, ranging from “never/less than 1 per month” to “more than 6 per day” with standard portion sizes. Based on the reported frequency and portion size, we estimated daily UPF consumption using the Nova classification system.30,31

Nova classifies foods into four categories based on the extent of industrial processing: 1) unprocessed or minimally processed foods, 2) processed culinary ingredients, 3) processed foods, and 4) UPFs. UPFs are defined as formulations of ingredients, mostly for exclusive industrial use, typically produced by various industrial techniques and processes.30 The methods of categorizing UPFs using the FFQ data from NHS and HPFS have been described previously.31 UPFs were further classified into eight mutually exclusive subgroups, including grain products (e.g., bread and breakfast cereals), fats/sauces, sweets (e.g., packaged sweet snacks or desserts), beverages (e.g., sugar or artificially sweetened beverages), meat products (e.g., animal protein-based ready-to-eat foods), dairy foods (e.g., flavored yogurt/dairy-based desserts), packaged savory snacks, and ready-to-eat/ready-to-heat mixed dishes.6,8 To minimize measurement errors, we estimated UPF intakes (servings per day) from the FFQ administered at the time closest to blood collection (average intake based on 1986 and 1990 FFQs for NHS participants;1990 and 1994 FFQs for HPFS participants). We removed liquor when estimating UPF consumption due to its known associations with metabolic dysfunction and an established relationship with CRC.

Measurement of Metabolomics

Blood specimens were collected by participants at home with the assistance of a local medical professional and returned on ice packs by overnight courier, and at least 95% of samples arrived within 26 hours of blood collection in both cohorts. Upon arrival, samples were centrifuged, and plasma, white blood cells, and red blood cells were aliquoted into cryotubes and archived in liquid nitrogen freezers at −130 degrees Celsius.

Plasma metabolomic profiling data were generated using the high-throughput liquid chromatography-mass spectrometry (LC-MS) method at the Broad Institute of MIT and Harvard as previously documented.28,29 In brief, high-resolution, accurate mass profiling data were acquired using LC-MS systems comprised of Nexera X2 UHPLC systems (Shimadzu Corp., Marlborough, MA) coupled to a Q Exactive or Exactive Plus Orbitrap Mass Spectrometer (Thermo Fisher Scientific, Waltham, MA). Hydrophilic interaction liquid chromatography with positive ion mode mass spectrometry detection was used to separate polar metabolites, and C18 chromatography with negative ion mode detection and C8 chromatography with positive ion mode detection were used to profile lipids and metabolites with intermediate polarity. Raw data were processed using TraceFinder 3.3 (Thermo Fisher Scientific, Waltham, MA, USA) and Progenesis QI (Nonlinear Dynamics, Newcastle upon Tyne, UK). The annotated metabolite identities were confirmed using authentic reference standards or reference samples. Pooled reference plasma samples (QCs) were interspersed every 20 participant samples to monitor analytical drift and maintain inter-assay reproducibility. Metabolite levels were reported as measured LC-MS peak areas, which are proportional to metabolite concentration.

In this study, we excluded unknown metabolites to facilitate biological interpretation. Among a total of 636 known metabolites, we excluded those with a mean coefficient of variation >25% among blinded QC samples or an intraclass correlation coefficient <0.40 to ensure within-person reproducibility over 1–2 years (n=27).32 After these exclusions, metabolite peak areas were then log-transformed and converted to z-scores with a mean of 0 and a standard deviation of 1 within each metabolomics sub-study. For metabolites missing less than 10%, imputation was performed using half of the minimum measured value. Metabolites with remaining missing values were further removed (n=388). A total of 228 and 296 annotated metabolites were retained among participants of the training and testing set, respectively. Only metabolites (n=222) that are captured in both sets were included for model training and testing.

Ascertainment of Colorectal Cancer Cases and Matched Controls

In the NHS and HPFS, CRC diagnoses were self-reported on biennial questionnaires and confirmed, with permission, via review of medical records and pathologic reports. A study physician, blinded to exposure information, reviewed records to confirm CRC diagnosis and to extract information on anatomic location, stage, and histologic type of cancer.

To construct the nested case-control study, we identified incident CRC cases diagnosed after blood draws, ranging from one month to 25 years. Risk set sampling was used to randomly select one control for each CRC case – matched on cohort, age, and date and fasting status at blood collection – from participants with archived blood samples who were alive and free of CRC at the time of case diagnosis. A total of 686 pairs of CRC cases and controls were included in the analysis.

Statistical Analysis

We used a two-phase approach for data analysis (Supplementary Figure 2). In Phase I, we identified the metabolomic pattern of UPF intake among 1,740 participants from the training set. We applied elastic net regression with 10-fold cross-validation to select the UPF-related metabolites for the pattern, which was calculated as the sum of the metabolite levels weighted by the regression coefficients. To avoid overfitting, the metabolomic pattern in the training set was obtained using the leave-one-out cross-validation (LOOCV) approach. We then validated the pattern by evaluating the Spearman correlation between the metabolomic pattern score (i.e., predicted UPF intake) and the self-reported UPF intake among the 1,740 participants in the training set and 686 pairs of CRC cases and controls. We used the same approach to derive and validate the metabolomic patterns of individual UPF subgroups. To assess the potential influence of inclusion of future cases or preclinical disease on metabolite selection, we conducted sensitivity analyses among controls only and among participants who did not develop type 2 diabetes, cancer, or cardiovascular disease within two years after blood draw. We also evaluated whether a less conservative Nova classification may affect the findings by using the secondary Nova categorization for UPF assessment, in which ambiguous items (popcorn, pancake or waffles, potato or corn chips, red meat as a sandwich or mixed dish, cream, and tomato sauce) were classified as UPFs.31

In Phase II, we evaluated the associations with CRC risk of UPF intake and metabolomic pattern scores, as well as individual metabolites constituting the metabolomic patterns, among 686 pairs of CRC cases and matched controls. Multivariable conditional logistic regression was used to estimate ORs and 95% CIs, adjusting for a wide range of demographic and lifestyle factors, including age, sex, race, fasting status, total calorie intake, alcohol intake, physical activity, smoking status, pack-years, body mass index (BMI), menopausal status and menopausal hormone therapy for women, aspirin use, and intakes of fruits and vegetables to account for potential confounding (Supplementary Methods). Of note, fasting status was included as a matching factor and additionally adjusted for in multivariable models to account for residual variation in fasting duration.

Considering dietary quality can be a potential mediator, we additionally adjusted for it in separate sensitivity analyses. Given that adiposity can influence systemic metabolism and circulating metabolites, we considered BMI to act primarily as a confounder rather than a mediator. Additionally, considering UPFs are highly correlated with sociodemographic, lifestyle factors, and cancer characteristics, we further conducted stratification and correlation analyses to assess whether the CRC association varied by these factors. The Benjamini-Hochberg false discovery rate (FDR) was used to account for multiple comparisons. All statistical analyses were performed in SAS (version 9.4) and R version 4.2.0. Statistical significance was considered at the α = 0.05 level.

Patient and Public Involvement Statement

No patients were directly involved throughout the research process, including formulation of the research questions or the outcome measures, study design, recruitment, the conduct of the study, and dissemination of the results.

RESULTS

Characteristics of the Study Participants

In the training set, participants were grouped into quintiles according to their intake of total UPFs (Supplementary Table 2). Each quintile included over 340 participants. The mean age at blood draw was 59.9 years, most participants were non-Hispanic whites, and over 60% of the blood samples were collected after at least 8 hours of fasting. Individuals with higher UPF intake had higher total calorie intake, more pack-years of smoking, lower diet quality (AHEI-2010), lower intake of fruits and vegetables, less physical activity, and were more likely to be overweight or obese.

For CRC case-control pairs, the mean age at blood draw was 61.5 years (Table 1 and Supplementary Table 3). Compared to controls, CRC cases had slightly more pack-years of smoking, higher overweight or obesity, and lower use of aspirin.

Table 1.

Characteristics of Study Participants at the Time of Blood Collection among Colorectal Cancer Cases and Matched Controls in the Nurses’ Health Study and Health Professionals Follow-up Study1,2

Cases (n=686) Controls (n=686)
Age at blood draw, years, mean (SD) 61.5 (8.20) 61.5 (8.10)
Women, % 60.0 60.3
Non-Hispanic White, % 96.7 98.6
Fasting ≥ 8 hours, % 63.4 63.3
Total calorie intake, kcal, mean (SD) 1906 (534) 1952 (560)
AHEI-2010, mean (SD) 51.5 (10.0) 51.9 (10.1)
Alcohol intake, g/day, mean (SD) 8.5 (12.3) 8.2 (11.2)
Physical activity, MET-hours/week, mean (SD) 22.0 (20.8) 22.9 (21.6)
Smoking, pack-years, mean (SD) 14.3 (19.1) 12.3 (17.1)
Smoking status
 Never, % 41.9 42.4
 Past, % 45.8 47.1
 Current, % 12.3 10.5
BMI status
 Underweight (<18.5 kg/m2), % 0.50 0.90
 Normal (18.5 to 24.9 kg/m2), % 45.4 56.2
 Overweight (25 to 29.9 kg/m2), % 40.5 31.8
 Obese (>30 kg/m2), % 13.5 11.1
Use of aspirin, % 38.3 45.1
Fruit intake, serving/day 2.5 (1.3) 2.6 (1.4)
Vegetable intake, serving/day 3.4 (1.5) 3.4 (1.5)
Open in a new tab

Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, Body Mass Index; MET, Metabolic Equivalent of Task

1.

Values are means (SD) for continuous variables; and percentages for categorical variables and are standardized to the age distribution of the study population. The values of polytomous variables may not sum to 100% due to rounding.

2.

Cases and controls were matched on cohort, age, and date and fasting status at blood collection.

Characteristics of the Metabolomic Pattern of Total UPFs

Among 222 metabolites, the elastic net regression model selected a combination of 50 metabolites (22 positively and 28 inversely) associated with UPF intake, while robust to the effects of collinearity among metabolites (Figure 1 and Supplementary Table 4). These metabolites belong to 5 annotated classes and 1 unannotated class, including 1) carboxylic acids and derivatives, 2) glycerophospholipids, 3) fatty acyls, glycerolipids, and sphingolipids, 4) steroids and steroid derivatives, and 5) others.

Figure 1. Characteristics of UPF-related Metabolomic Pattern.

Figure 1.

Open in a new tab

The mean pattern score significantly increased from quintile 1 (Q1) to Q5 (p-value <0.0001).

The metabolomic pattern was statistically significantly correlated with observed UPF intake (Spearman correlation coefficient, rho = 0.35, p = 2.2 × 10−16 in the training set; rho = 0.27, p = 2.2 × 10−16 in the testing set). The correlations between individual metabolites and total UPF intake range from −0.16 to 0.15. The magnitude and direction were largely consistent with the weights generated from the elastic net regression. Model performance and sensitivity analysis results are presented in Supplementary Results and Supplementary Figures 3 and 4, which support the robustness of metabolite selection.

Association between UPF-related Metabolomic Pattern and CRC Risk

After adjusting for potential confounders, the metabolomic pattern of UPF intake was associated with a higher risk of CRC (highest vs. lowest quintile: multivariable OR [95%CI] = 1.66 [1.14 to 2.43], p-trend = 0.002) (Table 2). Additional adjustment for overall diet quality using AHEI-2010 did not materially change the association (Supplementary Table 5). In contrast, further adjustment for self-reported UPF intake, the empirical dietary inflammatory pattern (EDIP), or a Western dietary pattern attenuated the association. After further adjusting for fruit and vegetable intake, the association between the UPF metabolomic pattern and CRC slightly strengthened (1.71 [1.15 to 2.53], 0.002). Notably, the association between dietary UPF intake and CRC risk was weaker (1.54 [1.03to 2.30], 0.09). Results remained unchanged in sensitivity analyses. No substantial differences were observed according to sex, lifestyle factors, tumor subsite, or time from blood draw to cancer diagnosis (Supplementary Figure 5 and Supplementary Table 6). Correlations between the UPF-related metabolomic pattern and lifestyle risk factors ranged from −0.40 to 0.25.

Table 2.

Associations of the Metabolomic Pattern of Total UPF Intake and the Dietary Intake of Total UPFs with Colorectal Cancer Risk

Multivariable Odds Ratio (95% CI) of Colorectal Cancer According to Quintiles of UPF Pattern
Q1 Q2 Q3 Q4 Q5 P-trend Per 1 SD increment
Metabolomic Pattern of total UPF intake (servings/d), median (IQR) 5.46 (0.44) 5.90 (0.16) 6.20 (0.15) 6.49 (0.18) 6.91 (0.32)
Minimally adjusted model 11 Ref. 1.12 (0.79 to 1.59) 1.37 (0.98 to 1.94) 1.38 (0.99 to 1.93) 1.56 (1.10 to 2.22) 0.003 1.18 (1.06 to 1.32)
Multivariable (MV) adjusted model 22 Ref. 1.16 (0.81 to 1.66) 1.48 (1.03 to 2.13) 1.57 (1.10 to 2.23) 1.77 (1.21 to 2.58) 0.001 1.24 (1.10 to 1.40)
MV-adjusted model 2 + BMI Ref. 1.17 (0.82 to 1.69) 1.45 (1.01 to 2.08) 1.51 (1.05 to 2.16) 1.66 (1.14 to 2.43) 0.002 1.21 (1.07 to 1.37)
MV-adjusted model 2 + BMI + AHEI-2010 + Intakes of Fruits and Vegetables Ref. 1.19 (0.82 to 1.71) 1.46 (1.01 to 2.10) 1.52 (1.06 to 2.19) 1.71 (1.15 to 2.53) 0.002 1.23 (1.08 to 1.40)
Dietary Intake of total UPFs (servings/d), median (IQR) 3.59 (0.94) 4.90 (0.59) 6.06 (0.44) 7.26 (0.66) 9.61 (2.17)
Minimally adjusted model 11 Ref. 1.29 (0.92 to 1.81) 0.98 (0.70 to 1.37) 1.09 (0.77 to 1.55) 1.47 (1.02 to 2.13) 0.08 1.11 (0.99 to 1.25)
Multivariable (MV) adjusted model 22 Ref. 1.34 (0.95 to 1.90) 0.98 (0.70 to 1.38) 1.14 (0.80 to 1.63) 1.50 (1.03 to 2.20) 0.09 1.11 (0.98 to 1.25)
MV-adjusted model 2 + BMI Ref. 1.33 (0.94 to 1.90) 0.98 (0.69 to 1.38) 1.15 (0.80 to 1.64) 1.47 (1.00 to 2.16) 0.11 1.10 (0.98 to 1.25)
MV-adjusted model 2 + BMI + AHEI-2010 + Intakes of Fruits and Vegetables Ref. 1.32 (0.93 to 1.88) 0.97 (0.69 to 1.38) 1.14 (0.79 to 1.66) 1.54 (1.03 to 2.30) 0.09 1.12 (0.98 to 1.28)
Open in a new tab

Abbreviations: AHEI, Alternative Healthy Eating Index; BMI, Body Mass Index; MET, Metabolic Equivalent of Task; UPF, Ultra-processed Food

1.

Model 1 minimally adjusted for age in years at blood draw, sex (men vs. women), and race (non-Hispanic whites vs. other races).

2.

Model 2 is a multivariable model adjusted for age in years at blood draw, sex (men vs. women), race (non-Hispanic whites vs. other races), fasting status (fasting <8 hours vs. fasting ≥8 hours), total calorie intake (kcal per day), alcohol intake (servings per day), physical activity (MET-hours per week), smoking status (never smoke, former smoker, current smoker), pack-years of smoking, menopausal status (yes, no, and missing), menopausal hormone therapy, and aspirin use (yes vs. no).

For the 50 metabolites included in the UPF pattern, their correlations with total UPF intake were moderately correlated with their associations with CRC risk (Spearman rho = 0.50, p = 0.0002) (Figure 2). Certain individual metabolites showed an association with intakes of total UPF and UPF subgroups and CRC risk (Supplementary Tables 7 and 8). Of them, N2, N2-dimethylguanosine, a potential biomarker of meat/poultry intake, showed a particularly strong positive association with CRC risk (1.96 [1.27 to 3.03], 0.02), especially in men (4.04 [1.82 to 8.98], 0.001). In contrast, 21-deoxycortisol, related to cortisol biosynthesis, was negatively associated with CRC risk (0.59 [0.41 to 0.86], 0.10). Notably, this metabolite was positively but weakly correlated with self-reported UPF intake (rho=0.02, elastic net weight=0.01). When evaluating the association with CRC risk per 1-SD increment in metabolite levels, a borderline significant positive association was observed for N1-acetylspermidine (1.14 [1.01 to 1.29], 0.04; rho=−0.01, weight=−0.05) and C16:0 ceramide (1.13 [1.00 to 1.27], 0.048; rho=0.05, weight=0.007), with a stronger association observed among men compared to women. All p-values were greater than 0.05 after FDR adjustment.

Figure 2. Beta Coefficients for the Associations of Metabolites in the UPF-Related Metabolomic Pattern with CRC Risk and Their Correlations with Total UPF Intake.

Figure 2.

Open in a new tab

Metabolomic Patterns of UPF Subgroups and Their Associations with CRC Risk

We further derived the pattern for each of the eight UPF subgroups. The patterns derived for meat products, fats/sauces, sweets, and beverages showed relatively stronger correlations with dietary intake of these foods (Spearman rho range: 0.35 to 0.50), compared to mixed dishes (0.31), dairy foods (0.25), grain products (0.23), and savory snacks (0.10) (Supplementary Table 9). The metabolomic patterns of these subgroups were correlated with those of total UPFs (Supplementary Figure 6).

Among subgroups, the metabolomic pattern of sweets was associated with a higher risk of CRC (highest vs. lowest quintile: multivariable OR [95%CI] = 1.52 [1.03 to 2.22], p-trend = 0.01) (Table 3). No associations were observed for metabolomic patterns of the remaining UPF subgroups. The associations did not statistically differ by sex (p-interactions >0.05) (Supplementary Table 10).

Table 3.

Associations Between Metabolomic Patterns of Ultra-processed Food Subgroups and Colorectal Cancer Risk

Association with Colorectal Cancer Risk, Odds Ratio (95% CI)1
Metabolomic Patterns of UPF Subgroups (servings/d) Q1 Q2 Q3 Q4 Q5 P-trend Per 1 SD increment
Sweets, median (IQR) 0.91 (0.12) 1.04 (0.06) 1.14 (0.05) 1.24 (0.05) 1.38 (0.12)
MV-adjusted model Ref. 1.22 (0.85 to 1.73) 1.44 (1.00 to 2.07) 1.30 (0.90 to 1.89) 1.52 (1.03 to 2.22) 0.01 1.19 (1.05 to 1.35)
Savory snacks, median (IQR) 0.31 (0.02) 0.33 (0.01) 0.35 (0.01) 0.36 (0.01) 0.39 (0.02)
MV-adjusted model Ref. 0.90 (0.63 to 1.28) 0.94 (0.65 to 1.35) 1.01 (0.71 to 1.46) 1.04 (0.71 to 1.53) 0.03 1.14 (1.01 to 1.29)
Fats/sauces, median (IQR) 1.07 (0.19) 1.30 (0.07) 1.45 (0.07) 1.59 (0.07) 1.79 (0.15)
MV-adjusted model Ref. 1.50 (1.06 to 2.14) 1.24 (0.87 to 1.77) 1.36 (0.94 to 1.95) 1.34 (0.91 to 1.96) 0.06 1.12 (1.00 to 1.26)
Grain products, median (IQR) 1.64 (0.09) 1.76 (0.04) 1.83 (0.04) 1.90 (0.04) 2.01 (0.09)
MV-adjusted model Ref. 1.07 (0.74 to 1.53) 1.15 (0.80 to 1.66) 1.41 (0.96 to 2.05) 1.35 (0.92 to 1.99) 0.07 1.12 (0.99 to 1.26)
Meat products, median (IQR) 0.14 (0.05) 0.20 (0.02) 0.24 (0.02) 0.28 (0.02) 0.33 (0.04)
MV-adjusted model Ref. 0.92 (0.65 to 1.31) 0.97 (0.67 to 1.40) 1.08 (0.74 to 1.56) 1.08 (0.73 to 1.59) 0.37 1.06 (0.93 to 1.20)
Mixed dishes, median (IQR) 1.79 (0.02) 1.81 (0.01) 1.83 (0.01) 1.84 (0.01) 1.86 (0.02)
MV-adjusted model Ref. 0.93 (0.64 to 1.34) 1.00 (0.69 to 1.45) 1.10 (0.75 to 1.61) 0.94 (0.63 to 1.41) 0.55 1.04 (0.91 to 1.18)
Beverages, median (IQR) 0.44 (0.12) 0.61 (0.06) 0.72 (0.06) 0.83 (0.06) 1.00 (0.14)
MV-adjusted model Ref. 0.87 (0.61 to 1.24) 0.91 (0.63 to 1.32) 1.19 (0.82 to 1.73) 0.97 (0.66 to 1.42) 0.82 1.01 (0.90 to 1.15)
Dairy foods, median (IQR) 0.22 (0.02) 0.25 (0.01) 0.27 (0.01) 0.28 (0.01) 0.31 (0.02)
MV-adjusted model Ref. 0.97 (0.68 to 1.39) 0.94 (0.66 to 1.35) 0.89 (0.62 to 1.27) 0.78 (0.53 to 1.14) 0.33 0.94 (0.84 to 1.06)
Open in a new tab

Abbreviations: AHEI, Alternative Healthy Eating Index; MET, Metabolic Equivalent of Task; UPF, Ultra-processed Food

1.

Multivariable model adjusted for age in years at blood draw, sex (men vs. women), race (non-Hispanic whites vs. other races), fasting status (fasting <8 hours vs. fasting ≥8 hours), total calorie intake (kcal per day), diet quality (AHEI-2010 score), alcohol intake (servings per day), physical activity (MET-hours per week), smoking status (never smoke, former smoker, current smoker), pack-years of smoking, body mass index (underweight, normal weight, overweight, and obese), menopausal status (yes, no, and missing), menopausal hormone therapy, aspirin use (yes vs. no), fruits intake (servings per day), and vegetable intake (servings per day).

By combining all metabolites identified from UPF subgroups, we mapped out a list of 161 unique metabolites that were correlated with at least one of the UPF subgroups (Supplementary Figure 7). Among individual metabolites, C2 carnitine (1.74 [1.18 to 2.57], 0.02) showed a positive association with CRC, while a few showed negative associations, including C58:7 TAG (0.65 [0.45 to 0.95], 0.04), C36:5 PE plasmalogen (0.64 [0.44 to 0.92], 0.03), C58:9 TAG (0.68 [0.47 to 0.99], 0.03), choline phosphate (0.62 [0.41 to 0.93], 0.02), C60:12 TAG (0.68 [0.47 to 0.99], 0.01), and C58:11 TAG (0.63 [0.43 to 0.93], 0.01), although none of these associations were statistically significant after FDR adjustment (FDR adjusted p-trend > 0.20) (Supplementary Table 11).

DISCUSSION

Leveraging metabolomic profiling data and dietary information from two large prospective cohorts, we identified a metabolic pattern of total UPF intake comprising 50 metabolites. This pattern was associated with increased CRC risk, independent of diet quality (AHEI-2010), and did not differ across lifestyle and tumor subgroups. We further characterized metabolomic patterns of UPF subgroups and their constituent metabolites. Several metabolites, including N2, N2-dimethylguanosine and C16:0 ceramide, showed positive associations, while 21-deoxycortisol, C36:5 PE plasmalogen, C58:9 TAG, and C58:11 TAG were inversely associated with CRC risk. Our findings suggest candidate biomarkers of UPF intake and provide insight into potential metabolic pathways linking UPFs to CRC.

The metabolomic pattern captured plasma metabolome changes correlated with habitual UPF intake, potentially aiding dietary assessment, as it is associated with “internal” exposure, accounting for between-person variations in absorption and metabolism. Among identified metabolites, N2, N2-dimethylguanosine, a tRNA-derived nucleoside linked to cellular stress and aging, was proposed as a potential biomarker for poultry and meat consumption (Table 4).33,34 In our cohorts, however, it correlated with sweets and fats/sauces rather than meat, suggesting that it may reflect broader metabolic responses rather than direct meat intake. C16:0 ceramide, a food emulsifier, can be ingested from high-fat foods; C4-OH carnitine is strongly correlated with processed red meat intake.33,35,36 Several UPF-related glycerophospholipids, including phosphatidylethanolamine (PE) and phosphatidylcholine (PC) plasmalogen, have low unsaturation and contain animal fat-derived moieties.33,37 TAGs with <56 carbons and ≤4 double bonds (e.g., C52:4 TAG), positively correlated with UPF intake, are the main constituents of animal fats and very low-density lipoprotein cholesterol.33

Table 4.

Summary of Our Findings and Existing Evidence on Ultra-Processed Food-Related Individual Metabolites in Relation to Colorectal Cancer Risk

Metabolite Class/Subclass Selected in the Pattern Potential Mechanisms
Positively Correlated Negatively Correlated
N2,N2- dimethylguanosine Purine nucleosides/NA2 Total UPFs, Sweets, Fats/sauces • Positively correlated with inflammation and metabolic dysregulation41
• Product of transfer ribonucleic acid (tRNA) degradation and reflects oxidative stress34
• Inversely associated with high-density lipoprotein (HDL) and risk factor for cardiometabolic diseases and all-cause and cancer mortality45–47
• Potential biomarker of animal-based foods, including poultry and meat intake33
• Inversely associated with a healthy plant-based dietary pattern39
• Positively associated with an unhealthy plan-based diet and sugar-sweetened beverages35,54
C16:0 Ceramide Sphingolipids/Ceramides Meat products • Plays a role in the development of obesity and insulin resistance48–50
• Positively correlated with ultra-processed meat intake and high-fat foods36
21-Deoxycortisol1 Steroids and steroid derivatives/Pregnane steroids Fats/sauces Mixed dishes, Meat products • Slow cortisol synthesis, preventing GI inflammation52,53
C36:5 PE plasmalogen1 Glycerophospholipids/Glycerophosphoethanolamines Meat products Total UPFs, Sweets • May exhibit anti-oxidation and anti-inflammation effect56,57
• Correlated with chicken consumption37
C58:9 TAG Glycerolipids/Triradylcglycerols Mixed dishes, Sweets • Lipids with high carbon atoms and high double bonds were associated with normalized insulin function and signaling54,55
• Positively associated with a Prudent dietary pattern58
C58:11 TAG Glycerolipids/Triradylcglycerols Beverages • Lipids with high carbon atoms and high double bonds were associated with normalized insulin function and signaling54,55
• Negatively associated with a Western dietary pattern, but positively associated with a Prudent dietary pattern58
Open in a new tab
1.

Controlled-feeding validation should be prioritized for these metabolites to investigate the biological plausibility.

2.

No subclass was assigned to the metabolite.

Conversely, several metabolites inversely associated with UPF intake were highly correlated with minimally/unprocessed food intake. N-acetylornithine, cinnamoylglycine, proline betaine, pipecolic acid, and piperine are potential biomarkers of fruits, vegetables, and legumes.37–39 Trigonelline and hippuric acid are biomarkers of coffee, a minimally/unprocessed food.31 Cinnamoylglycine decreased after a two-week ultra-processed diet intervention in an RCT of 20 domiciled healthy adults.38 C38:6 PC was highly correlated with the Mediterranean diet, particularly with olive oil and seafood intake.40

The metabolomic pattern likely reflects dietary behaviors correlated with UPF intake, while also capturing biological variation beyond traditional nutrient-based diet scores. Adjustment for AHEI-2010 did not materially change the association with CRC risk, suggesting that the metabolomic pattern represents complementary information to established dietary indexes. The attenuation after additional adjustment for self-reported UPF intake or related dietary patterns (e.g., Western diet, EDIP) indicates partial overlap with other diet-derived exposures, which could reflect mediation or shared upstream behaviors. Although the nested study yielded slightly higher risk estimates for UPF intake than the full cohorts, these discrepancies likely reflect random variation and differences in case distribution or subsite classification. By contrast, the stronger association for the metabolomic pattern suggests that it may capture additional exposures or biological processes beyond UPF intake alone. A more detailed discussion is provided in the Supplementary Discussion.

Our findings suggest that UPF may increase CRC risk through inflammatory or metabolic pathways and/or by introducing carcinogens (Table 4). For example, N2, N2-dimethylguanosine, an important constituent of metabolomic patterns of inflammation and metabolic dysregulation,41 may promote inflammation leading to colorectal tumorigenesis. It was elevated in individuals with prevalent CRC and has been identified as one of the top metabolites associated with CRC risk in men.42–44 This metabolite was also inversely associated with high-density lipoprotein cholesterol and positively associated with cardiometabolic diseases and all-cause and cancer-specific mortality.45–47 C16:0 ceramide contributes to obesity and insulin resistance in animal models.48 It was also elevated in the adipose tissue of humans who were obese and identified as a principal mediator of obesity-related insulin resistance.49,50 Both obesity and insulin resistance are established risk factors for CRC.16,51

We also identified UPF-related metabolites inversely associated with CRC risk. The inverse association between 21-deoxycortisol and CRC may relate to its role in slowing cortisol synthesis, thereby maintaining immune response and preventing gastrointestinal inflammation.52,53 C36:5 PE plasmalogen, a membrane lipid with antioxidative and anti-inflammatory potential, and TAGs with longer chains and higher unsaturation (e.g., C58:9 and C58:11) have been linked to improved insulin signaling, partly explaining their inverse association with CRC.54–57 This is further supported by prior research linking C58:9 TAG to adherence to a Prudent dietary pattern associated with lower CRC risk.58,59 While further mechanistic studies are needed, our study highlights potential biological pathways linking UPF intake to CRC risk.

Notably, although the UPF-related metabolomic pattern was associated with CRC risk, individual metabolites showed varying concordance with UPF intake and CRC associations. For example, 21-deoxycortisol showed a weak positive correlation with UPF intake but was inversely associated with CRC risk. These findings suggest that the UPF-related metabolomic pattern captures a mixture of dietary and metabolic signals, not all of which reflect a direct UPF-CRC pathway. We interpret these findings cautiously and recommend targeted validation of individual metabolites in experimental studies.

Strengths and Limitations

Our study has several strengths. First, we integrated high-throughput metabolomic profiling and dietary intake data from the same participants with long-term, prospective follow-up. Second, the metabolomic patterns of total UPFs and UPF subgroups were rigorously derived and validated in separate data sets. Finally, the comprehensive epidemiologic data from the two well-established cohorts allowed for adjusting for a wide range of potential confounding factors when examining associations with CRC risk.

Several limitations to acknowledge. First, we only included named metabolites. We recognize that many diet-related compounds remain chemically uncharacterized. A prior study showed that some unknown metabolites exhibit higher correlations with meat intake, but it remains unclear whether these are derivatives of known metabolites or unrelated molecules.37 Future work is warranted to investigate unknown metabolites in relation to UPF intake and CRC risk. Second, our metabolomic pattern was developed according to self-reported UPF intake from FFQs, which are subject to measurement errors, due to the lack of brand-specific and ingredient-level information.60 While we employed a rigorous classification approach,31 including expert panel reviews, research dietitian input, cohort documentation, and supermarket scans, some misclassification is inevitable. For example, processed meat products are inconsistently categorized as UPFs across dietary assessment methods (e.g., FFQs vs. 24-hour recalls). In our study, items such as sausage, chicken nuggets, and hot dogs were classified as UPFs, generally aligning with 24-hour recall-based classification. Nova classifications were held constant across FFQ cycles to ensure consistency, while the FFQ content was periodically updated to reflect changes in the food supply. Only a small number of food items changed over time; therefore, any resulting nondifferential misclassification would likely attenuate, rather than exaggerate, observed associations and is unlikely to materially affect study conclusions. Additionally, FFQs are well-suited to assess long-term habitual dietary intake, relevant for chronic disease outcomes. Third, because our cohort participants are predominantly non-Hispanic whites and all health professionals, the generalizability of our findings might be limited. Lastly, as an observational study, we could not confirm causality.

CONCLUSION

This study identified a metabolomic pattern of UPF intake, which showed a prospective association with CRC risk, providing biological insights into the potential metabolic pathways linking UPFs to CRC.

Supplementary Material

Supp1
Supp2
Supp3
Supp4
Supp5
Supp6
Supp7
Supp8
Supp9

Key Messages.

What is already known on this topic

  • Growing evidence links high consumption of ultra-processed foods (UPFs) to increased colorectal cancer (CRC) risk, yet the underlying mechanisms remain largely unknown.

What this study adds

  • We developed and validated a metabolomic pattern of UPFs comprising primarily lipids and amino acids.

  • This pattern and several individual metabolites were associated with CRC risk.

How this study might affect research, practice or policy

  • Our study findings suggest that UPF may increase CRC risk through several inflammatory or metabolic pathways and/or by introducing carcinogens, providing biological insights into the pro-tumorigenic effect of these foods.

  • This metabolomic pattern reflects the profile of plasma metabolites correlated with habitual UPF intake, potentially aiding in UPF assessment as it is associated with the “internal” dietary exposure that accounts for between-person variations in the absorption and metabolism of food components.

Acknowledgments

We thank the participants and staff of the Nurses’ Health Study and the Health Professionals Follow-up Study for their valuable contributions as well as the following state cancer registries for their help: Alabama, Arizona, Arkansas, California, Colorado, Connecticut, Delaware, Florida, Georgia, Idaho, Illinois, Indiana, Iowa, Kentucky, Louisiana, Maine, Maryland, Massachusetts, Michigan, Nebraska, New Hampshire, New Jersey, New York, North Carolina, North Dakota, Ohio, Oklahoma, Oregon, Pennsylvania, Rhode Island, South Carolina, Tennessee, Texas, Virginia, Washington, and Wyoming.

Funding Statement

This work was supported by grants MRSG-17–220-01-NEC (MS) and CRP-24–1185864-01-PROF (SO) from the American Cancer Society, grants U01CA261961, R01CA263776, and R01CA285851 (MS), U2CDK129670 (FBH), R35CA253185 (ATC), K00CA274714 (MD), K01DK136968 (DEH), and cohort infrastructure grants (UM1CA186107, P01CA87969, and U01CA167552) from the National Institutes of Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication.

Footnotes

Conflict of interest statement

JAM serves as a consultant on the scientific advisory board for Merck. ATC serves as a consultant for Pfizer Inc and Boehringer Ingelheim. All of the above is outside the submitted work. No other relationships or activities could appear to have influenced the submitted work.

Data Sharing Statement

The data reported in this study cannot be deposited in a public repository due to participant confidentiality and privacy concerns. Therefore, data are available upon written request. According to standard controlled access procedure, applications to use NHS and HPFS resources will be reviewed by our External Collaborators Committee for scientific aims, evaluation of the fit of the data for the proposed methodology, and verification that the proposed use meets the guidelines of the Ethics and Governance Framework and the consent that was provided by the participants. Investigators wishing to use NHS and HPFS data are asked to submit a brief description of the proposed project (go to https://www.nurseshealthstudy.org/researchers (contact email: nhsaccess@channing.harvard.edu) and https://sites.sph.harvard.edu/hpfs/for-collaborators).

References

  • 1.Siegel RL, Kratzer TB, Giaquinto AN, et al. Cancer statistics, 2025. CA: A Cancer Journal for Clinicians 2025;75:10–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.World Cancer Research Fund/American Institute for Cancer Research. Diet, Nutrition, Physical Activity and Cancer: a Global Perspective. Continuous Update Project Expert Report 2018. Available at: https://www.wcrf.org/wp-content/uploads/2024/11/Summary-of-Third-Expert-Report-2018.pdf. [Google Scholar]
  • 3.Juul F, Parekh N, Martinez-Steele E, et al. Ultra-processed food consumption among US adults from 2001 to 2018. The American Journal of Clinical Nutrition 2022;115:211–221. [DOI] [PubMed] [Google Scholar]
  • 4.Lane MM, Gamage E, Du S, et al. Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses. BMJ 2024;384:e077310. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Meine GC, Picon RV, Espírito Santo PA, et al. Ultra-Processed Food Consumption and Gastrointestinal Cancer Risk: A Systematic Review and Meta-Analysis. Am J Gastroenterol 2024;119:1056–1065. [DOI] [PubMed] [Google Scholar]
  • 6.Wang L, Du M, Wang K, et al. Association of ultra-processed food consumption with colorectal cancer risk among men and women: results from three prospective US cohort studies. BMJ 2022;378:e068921. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hang D, Wang L, Fang Z, et al. Ultra-processed food consumption and risk of colorectal cancer precursors: results from 3 prospective cohorts. J Natl Cancer Inst 2023;115:155–164. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Hang D, Du M, Wang L, et al. Ultra-processed food consumption and mortality among patients with stages I-III colorectal cancer: a prospective cohort study. EClinicalMedicine 2024;71:102572. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Loftfield E, Moore SC, Mayne ST. The critical need for a robust research agenda on ultra-processed food consumption and cancer risk. PLoS Med 2024;21:e1004482. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.World Cancer Research Fund/American Institute for Cancer Research. Preservation and processing of foods and the risk of cancer. Available at: https://www.wcrf.org/wp-content/uploads/2024/10/Preservation-and-processing-of-foods.pdf. [Google Scholar]
  • 11.Gurjao C, Zhong R, Haruki K, et al. Discovery and Features of an Alkylating Signature in Colorectal Cancer. Cancer Discov 2021;11:2446–2455. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Bouvard V, Loomis D, Guyton KZ, et al. Carcinogenicity of consumption of red and processed meat. Lancet Oncol 2015;16:1599–1600. [DOI] [PubMed] [Google Scholar]
  • 13.Goldman R, Shields PG. Food mutagens. J Nutr 2003;133 Suppl 3:965S-973S. [DOI] [PubMed] [Google Scholar]
  • 14.Hall KD, Ayuketah A, Brychta R, et al. Ultra-Processed Diets Cause Excess Calorie Intake and Weight Gain: An Inpatient Randomized Controlled Trial of Ad Libitum Food Intake. Cell Metabolism 2019;30:67–77.e3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Juul F, Martinez-Steele E, Parekh N, et al. Ultra-processed food consumption and excess weight among US adults. Br J Nutr 2018;120:90–100. [DOI] [PubMed] [Google Scholar]
  • 16.Lauby-Secretan B, Scoccianti C, Loomis D, et al. Body Fatness and Cancer--Viewpoint of the IARC Working Group. N Engl J Med 2016;375:794–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Witkowski M, Nemet I, Alamri H, et al. The artificial sweetener erythritol and cardiovascular event risk. Nat Med 2023;29:710–718. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Debras C, Chazelas E, Sellem L, et al. Artificial sweeteners and risk of cardiovascular diseases: results from the prospective NutriNet-Santé cohort. BMJ 2022;378:e071204. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Sellem L, Srour B, Javaux G, et al. Food additive emulsifiers and cancer risk: Results from the French prospective NutriNet-Santé cohort. PLoS Med 2024;21:e1004338. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Sellem L, Srour B, Javaux G, et al. Food additive emulsifiers and risk of cardiovascular disease in the NutriNet-Santé cohort: prospective cohort study. BMJ 2023;382:e076058. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Whelan K, Bancil AS, Lindsay JO, et al. Ultra-processed foods and food additives in gut health and disease. Nat Rev Gastroenterol Hepatol 2024;21:406–427. [DOI] [PubMed] [Google Scholar]
  • 22.Johnson CH, Ivanisevic J, Siuzdak G. Metabolomics: beyond biomarkers and towards mechanisms. Nat Rev Mol Cell Biol 2016;17:451–459. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Rangel-Huerta OD, Pastor-Villaescusa B, Gil A. Are we close to defining a metabolomic signature of human obesity? A systematic review of metabolomics studies. Metabolomics 2019;15:93. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.A M, S R, D H. Microbiome risk profiles as biomarkers for inflammatory and metabolic disorders. Nature reviews Gastroenterology & hepatology 2022;19. Available at: https://pubmed.ncbi.nlm.nih.gov/35190727/ [Accessed November 6, 2025]. [DOI] [PubMed] [Google Scholar]
  • 25.Rimm EB, Giovannucci EL, Willett WC, et al. Prospective study of alcohol consumption and risk of coronary disease in men. Lancet 1991;338:464–468. [DOI] [PubMed] [Google Scholar]
  • 26.Bao Y, Bertoia ML, Lenart EB, et al. Origin, Methods, and Evolution of the Three Nurses’ Health Studies. Am J Public Health 2016;106:1573–1581. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Al-Shaar L, Yuan C, Rosner B, et al. Reproducibility and Validity of a Semiquantitative Food Frequency Questionnaire in Men Assessed by Multiple Methods. Am J Epidemiol 2021;190:1122–1132. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Yuan C, Spiegelman D, Rimm EB, et al. Relative Validity of Nutrient Intakes Assessed by Questionnaire, 24-Hour Recalls, and Diet Records as Compared With Urinary Recovery and Plasma Concentration Biomarkers: Findings for Women. Am J Epidemiol 2018;187:1051–1063. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Hunter DJ, Hankinson SE, Hough H, et al. A prospective study of NAT2 acetylation genotype, cigarette smoking, and risk of breast cancer. Carcinogenesis 1997;18:2127–2132. [DOI] [PubMed] [Google Scholar]
  • 30.Monteiro CA, Cannon G, Moubarac JC, et al. The UN Decade of Nutrition, the NOVA food classification and the trouble with ultra-processing. Public Health Nutr 2018;21:5–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Khandpur N, Rossato S, Drouin-Chartier J-P, et al. Categorising ultra-processed foods in large-scale cohort studies: evidence from the Nurses’ Health Studies, the Health Professionals Follow-up Study, and the Growing Up Today Study. J Nutr Sci 2021;10:e77. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Townsend MK, Clish CB, Kraft P, et al. Reproducibility of metabolomic profiles among men and women in two large cohort studies. Clin Chem 2013;59: 10.1373/clinchem.2012.199133. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Wishart DS, Guo A, Oler E, et al. HMDB 5.0: the Human Metabolome Database for 2022. Nucleic Acids Res 2022;50:D622–D631. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Su D, Chen J, Du S, et al. Metabolomic Markers of Ultra-Processed Food and Incident CKD. Clin J Am Soc Nephrol 2023;18:327–336. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Wedekind R, Kiss A, Keski-Rahkonen P, et al. A metabolomic study of red and processed meat intake and acylcarnitine concentrations in human urine and blood. Am J Clin Nutr 2020;112:381–388. [DOI] [PubMed] [Google Scholar]
  • 36.Mah MS, Cao E, Anderson D, et al. High-fat feeding drives the intestinal production and assembly of C16:0 ceramides in chylomicrons. Science Advances 2024;10:eadp2254. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Mazzilli KM, McClain KM, Lipworth L, et al. Identification of 102 Correlations between Serum Metabolites and Habitual Diet in a Metabolomics Study of the Prostate, Lung, Colorectal, and Ovarian Cancer Trial. J Nutr 2020;150:694–703. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.O’Connor LE, Hall KD, Herrick KA, et al. Metabolomic Profiling of an Ultraprocessed Dietary Pattern in a Domiciled Randomized Controlled Crossover Feeding Trial. J Nutr 2023;153:2181–2192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Wang F, Baden MY, Guasch-Ferré M, et al. Plasma metabolite profiles related to plant-based diets and the risk of type 2 diabetes. Diabetologia 2022;65:1119–1132. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Li J, Guasch-Ferré M, Chung W, et al. The Mediterranean diet, plasma metabolome, and cardiovascular disease risk. Eur Heart J 2020;41:2645–2656. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Bever AM, Hang D, Lee DH, et al. Metabolomic signatures of inflammation and metabolic dysregulation in relation to colorectal cancer risk. J Natl Cancer Inst 2024;116:1126–1136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Seidel A, Brunner S, Seidel P, et al. Modified nucleosides: an accurate tumour marker for clinical diagnosis of cancer, early detection and therapy control. Br J Cancer 2006;94:1726–1733. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Zhu J, Djukovic D, Deng L, et al. Targeted serum metabolite profiling and sequential metabolite ratio analysis for colorectal cancer progression monitoring. Anal Bioanal Chem 2015;407:7857–7863. [DOI] [PubMed] [Google Scholar]
  • 44.Downie JM, Joshi AD, Geraghty CM, et al. Novel metabolomic predictors of incident colorectal cancer in men and women. JNCI: Journal of the National Cancer Institute 2025;117:517–528. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Ottosson F, Smith E, Fernandez C, et al. Plasma Metabolites Associate with All-Cause Mortality in Individuals with Type 2 Diabetes. Metabolites 2020;10:315. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Wang F, Tessier A-J, Liang L, et al. Plasma metabolomic profiles associated with mortality and longevity in a prospective analysis of 13,512 individuals. Nat Commun 2023;14:5744. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Cirulli ET, Guo L, Leon Swisher C, et al. Profound Perturbation of the Metabolome in Obesity Is Associated with Health Risk. Cell Metab 2019;29:488–500.e2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Raichur S, Brunner B, Bielohuby M, et al. The role of C16:0 ceramide in the development of obesity and type 2 diabetes: CerS6 inhibition as a novel therapeutic approach. Mol Metab 2019;21:36–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Turpin SM, Nicholls HT, Willmes DM, et al. Obesity-Induced CerS6-Dependent C16:0 Ceramide Production Promotes Weight Gain and Glucose Intolerance. Cell Metabolism 2014;20:678–686. [DOI] [PubMed] [Google Scholar]
  • 50.Hla T, Kolesnick R. C16:0-Ceramide Signals Insulin Resistance. Cell Metabolism 2014;20:703–705. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Keum N, Giovannucci E. Global burden of colorectal cancer: emerging trends, risk factors and prevention strategies. Nature Reviews Gastroenterology & Hepatology 2019;16:713–732. [DOI] [PubMed] [Google Scholar]
  • 52.Yeager MP, Pioli PA, Guyre PM. Cortisol exerts bi-phasic regulation of inflammation in humans. Dose Response 2011;9:332–347. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Oligschlaeger Y, Yadati T, Houben T, et al. Inflammatory Bowel Disease: A Stressed GGut/Feeling. Cells 2019;8:659. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Rhee EP, Cheng S, Larson MG, et al. Lipid profiling identifies a triacylglycerol signature of insulin resistance and improves diabetes prediction in humans. J Clin Invest 2011;121:1402–1411. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Brantley KD, Zeleznik OA, Rosner B, et al. Plasma Metabolomics and Breast Cancer Risk over 20 Years of Follow-up among Postmenopausal Women in the Nurses’ Health Study. Cancer Epidemiol Biomarkers Prev 2022;31:839–850. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Hossain MS, Mawatari S, Fujino T. Biological Functions of Plasmalogens. Adv Exp Med Biol 2020;1299:171–193. [DOI] [PubMed] [Google Scholar]
  • 57.Udagawa J, Hino K. Plasmalogen in the brain: Effects on cognitive functions and behaviors attributable to its properties. Brain Research Bulletin 2022;188:197–202. [DOI] [PubMed] [Google Scholar]
  • 58.Chandler PD, Balasubramanian R, Paynter N, et al. Metabolic signatures associated with Western and Prudent dietary patterns in women. The American Journal of Clinical Nutrition 2020;112:268–283. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Mehta RS, Song M, Nishihara R, et al. Dietary Patterns and Risk of Colorectal Cancer: Analysis by Tumor Location and Molecular Subtypes. Gastroenterology 2017;152:1944–1953.e1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.O’Connor LE, Herrick KA, Papier K. Handle with care: challenges associated with ultra-processed foods research. Int J Epidemiol 2024;53:dyae106. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supp1
NIHMS2144676-supplement-Supp1.pdf (170.7KB, pdf)
Supp2
NIHMS2144676-supplement-Supp2.pdf (321.9KB, pdf)
Supp3
NIHMS2144676-supplement-Supp3.pdf (407.3KB, pdf)
Supp4
NIHMS2144676-supplement-Supp4.pdf (277.8KB, pdf)
Supp5
NIHMS2144676-supplement-Supp5.pdf (296.1KB, pdf)
Supp6
NIHMS2144676-supplement-Supp6.pdf (305KB, pdf)
Supp7
NIHMS2144676-supplement-Supp7.pdf (243.5KB, pdf)
Supp8
NIHMS2144676-supplement-Supp8.docx (165.8KB, docx)
Supp9
NIHMS2144676-supplement-Supp9.pptx (1.3MB, pptx)

Data Availability Statement

The data reported in this study cannot be deposited in a public repository due to participant confidentiality and privacy concerns. Therefore, data are available upon written request. According to standard controlled access procedure, applications to use NHS and HPFS resources will be reviewed by our External Collaborators Committee for scientific aims, evaluation of the fit of the data for the proposed methodology, and verification that the proposed use meets the guidelines of the Ethics and Governance Framework and the consent that was provided by the participants. Investigators wishing to use NHS and HPFS data are asked to submit a brief description of the proposed project (go to https://www.nurseshealthstudy.org/researchers (contact email: nhsaccess@channing.harvard.edu) and https://sites.sph.harvard.edu/hpfs/for-collaborators).

RESOURCES