Ultra-processed food intake and brain health in middle-aged and older adults

Graphical abstract

Abstract

Objective

The escalating burden of neurodegenerative diseases underscores the urgent need to identify modifiable dietary risks. Given the increasing prevalence of ultra-processed foods (UPFs) in modern diets and their links to various chronic diseases, this study investigates their associations with neurodegeneration and alterations in brain structure.

Design

A prospective cohort study.

Setting and participants

58,423 participants aged 40–70 years from the UK Biobank cohort (enrolled from 2006 to 2010) were included to examine the associations between UPF intake and neurodegenerative diseases, with a subsample of 5,462 participants with neuroimaging data used to investigate associations with regional gray matter morphology, including volume, thickness, and surface area.

Measurements

UPF intake was quantified as the weight-based proportion of ultra-processed foods, calculated by dividing the total daily weight (g/day) of UPFs by the total weight of all food consumed, based on the Oxford WebQ dietary questionnaire. Cox proportional hazards models, fully adjusted for potential confounders, were employed to assess the associations between UPF intake and neurodegenerative diseases. Polygenic risk scores derived from genome-wide summary statistics to reflect genetic predisposition to neurodegenerative diseases, were used to perform subgroup analyses. Multiple linear regression models were used to examine the associations between UPF intake and brain gray matter phenotypes.

Results

Our results indicated that high UPF intake was associated with increased risks of incident dementia (hazard ratio [95% confidence interval] = 1.37 [1.08, 1.74]), Parkinson’s disease (1.76 [1.22, 2.53]), and multiple sclerosis (2.38 [1.02, 5.55]), with stronger associations observed in participants with lower polygenic risk score. Moreover, high UPF intake corresponded to extensive gray matter compromise, including reduced subcortical volumes with right-hemispheric predominance, and widespread cortical deterioration in volume, thickness, and surface area.

Conclusions

These findings advance epidemiological evidence on the relationship between UPF intake and neurodegenerative outcomes, suggesting that dietary assessment may serve as a relevant consideration in population-level approaches to brain health promotion.

1. Introduction

Amid the rapid growth of aging populations worldwide, the rising prevalence of neurodegenerative diseases has become a critical public health challenge [1]. The limited efficacy of current pharmacological treatments in altering disease trajectories underscores the urgent need to identify modifiable risk factors and effective preventive strategies to delay the onset and progression of disease [2,3]. Among these, diet has been increasingly recognized as an important factor, with accumulating evidence linking nutritional patterns to the emergence of neurodegenerative conditions and alterations in brain structure [4].

However, ultra-processed foods (UPFs), characterized by industrial formulations rich in additives and depleted of whole food components, have rapidly become dominant in modern diets fueled by their palatability and affordability [5]. UPFs have been linked to escalating rates of several chronic diseases recent years, but their role in brain health remains relatively unclearly explored [[6], [7], [8]]. While emerging studies suggest associations of UPF intake with dementia risk, evidence regarding other major neurodegenerative diseases, such as Parkinson’s disease (PD) and multiple sclerosis (MS), is limited and inconsistent [9,10]. In addition, most research has focused on clinical endpoints, with little attention to subclinical brain morphology [11,12]. These limitations, coupled with modest sample sizes and potential residual confounding, highlight the need for large-scale investigations integrating both clinical and subclinical outcomes to comprehensively characterize the associations between UPF intake and brain health [13].

Leveraging a large prospective UK cohort with dietary assessments and neuroimaging data, this study aims to fill the gaps in understanding the associations of UPF intake with neurodegenerative diseases, including dementia, PD, and MS, as well as brain morphology across cortical and subcortical gray matter (GM). To better delineate context-specific associations, we performed stratified analyses across genetic susceptibility, sociodemographic characteristics, and lifestyle factors. We also incorporated hemisphere-specific assessments and a multidimensional set of neuroanatomical metrics—including volume, thickness, and surface area—to capture subtle patterns of structural brain vulnerability. By integrating these complementary approaches, our study aims to generate robust epidemiological evidence on the associations between UPF intake and brain health, thereby contributing to a better understanding of potential dietary correlates of neurodegeneration and informing strategies to promote healthy aging.

2. Material and methods

2.1. Study design and study population

This study utilized data from the UK Biobank, a large-scale, population-based prospective cohort. Between 2006 and 2010, approximately 500,000 individuals aged 37–73 years were recruited from 22 assessment centers across England, Scotland, and Wales, representing diverse sociodemographic backgrounds. Data collection at baseline encompassed multiple modalities, including web-based questionnaires, physical assessments, biospecimen collection, genetic analysis, and additional relevant evaluations. Longitudinal follow-up of health outcomes has been conducted through linkage with national health-related records. Ethical approval for the UK Biobank was obtained from the NHS North West Multi-Center Research Ethics Committee (Ref: 11/NW/0382), and all participants provided written informed consent prior to enrollment. Access to the data for the current study was granted under application number 99001 (valid from 31 January 2023 to 31 January 2026). The present analysis adhered to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

For the purposes of this study, individuals were excluded if they had missing data on exposure, outcome, relevant covariates, and genetic data; a prior diagnosis of specific neurodegenerative diseases; evidence of excessive heterozygosity; discordance between self-reported and genetically determined sex; or were lost to follow-up. After applying these exclusion criteria, a final analytic sample of 58,423 participants was retained, among whom 5,462 had available brain MRI data.

2.2. Assessing ultra-processed food intake

Dietary intake was evaluated using the Oxford WebQ, a web-based, self-completed 24-h dietary recall instrument. This tool captures detailed information on the consumption of up to 206 food items and 30 beverage types within the preceding 24 h. Within the UK Biobank framework, dietary data were collected through up to five WebQ administrations between 2009 and 2012. At baseline (2009–2010), 70,724 participants completed at least one WebQ questionnaire. Subsequently, between February 2011 and June 2012, participants who had provided email addresses were invited to complete four additional dietary recalls, each conducted at approximately 3- to 4-month intervals. These were scheduled across four distinct cycles: February-April 2011, June-September 2011, October-December 2011, and April-June 2012. The invitations were distributed across different days of the week to account for potential variation in weekday versus weekend dietary patterns [14]. The Oxford WebQ has demonstrated high validity for large-scale dietary assessment, producing intake estimates comparable to interviewer-led recalls and showing consistent agreement with biomarker-based evaluations [15,16]. To improve the reliability of dietary exposure estimation, we derived each participant’s baseline habitual intake by averaging values from a minimum of two available 24 -h dietary recalls.

In this study, UPF intake was identified from participants' dietary data based on the NOVA Food Classification System, which categorizes foods into four groups according to the degree and purpose of industrial processing, including unprocessed or minimally processed foods, processed culinary ingredients, processed foods, and UPFs [17]. Based on the characteristics of the UK Biobank dietary data, eight major UPF categories were defined for this analysis: (1) UP-beverages; (2) UP-dairy products; (3) UP-fruits and vegetables; (4) UP-meat, fish, and eggs; (5) UP-sauces and soups; (6) UP-starchy foods; (7) UP-sugary snacks; (8) UP-savory snacks (Supplementary Table S1). The total UPF consumption was calculated by summing the intakes of each UPF category above. To ensure comparability across individuals, the UPF intake was operationalized as a weight-based proportion, defined as the ratio of total weight of UPF intake (grams per day) to the overall weight of all foods intake (grams per day). Given the absence of universally accepted guidelines for UPF intake, participants were categorized into tertiles of UPF intake, including low, moderate, and high intake, with the lowest tertile serving as the reference group. To enhance the robustness of our results, UPF intake was also analyzed as a continuous variable, expressed as a median proportion of 15.6% (interquartile range: 9.3 %–20.8 %) of total daily food weight.

2.3. Assessment outcomes

The primary outcomes of this study encompassed a spectrum of neurodegenerative diseases, including dementia, PD and MS. Diagnoses were identified through hospital inpatient records available within the UK Biobank, obtained via systematic linkage with national health registries. Neurodegenerative disease cases were identified based on the International Classification of Diseases, Ninth and Tenth Revisions (ICD-9 and ICD-10), with diagnostic criteria aligned with UK Biobank definitions detailed in Supplementary Table S2. Follow-up extended from the baseline assessment date until the earliest of the following events: a confirmed diagnosis of neurodegenerative disease, death, the most recent data available from either primary care or hospital inpatient records, or December 19, 2022, whichever came first.

The secondary outcomes focused on the brain morphological characteristics of GM. Brain imaging acquisition commenced in 2014, utilizing Siemens Skyra 3 T scanners equipped with 32-channel head coils, at dedicated assessment centers. Structural imaging data were preprocessed centrally by the UK Biobank imaging team, yielding imaging-derived phenotypes (IDPs) for approved researchers. High-resolution T1-weighted MRI scans were analyzed using FreeSurfer (version 6.0.0) to derive gray matter phenotypes. The reliability and quality of FreeSurfer outputs were assessed using the Qoala-T framework, complemented by manual inspection for cases near quality thresholds. Data not meeting predefined quality standards were excluded from further analyses. The volumes of 7 subcortical regions were derived via FreeSurfer’s aseg tool (UK Biobank data field 1102), while volumes, thicknesses, and surface areas across 33 cortical regions were obtained using FreeSurfer’s surface-based atlases (UK Biobank data field 192). All continuous measures of brain structures were standardized to facilitate interpretation and cross-comparison. Additional details can be found in Supplementary Table S3-S5.

2.4. Covariates

To account for potential confounding factors influencing both the outcomes and UPF intake, a comprehensive set of covariates was incorporated into the statistical models, encompassing age, sex, ethnicity, income, education, socioeconomic status (SES), body mass index (BMI), waist-to-hip ratio (WHR), prevalent diseases (hypertension, diabetes, and stroke), healthy lifestyles (never smoking, moderate alcohol intake, healthy sleep pattern, healthy diet, and regular physical activity) and metabolic biomarkers (glucose, glycated hemoglobin [HbA1c], triglycerides [TG], low-density lipoprotein cholesterol [LDL], and total cholesterol).

The SES reflected participants' combined standing in household income, educational attainment, and employment status, derived using latent class analysis (LCA). Participants were classified into low, middle, and high SES categories, with higher SES representing more favorable socioeconomic conditions [18]. The BMI was calculated as weight divided by height squared (kg/m²) and categorized into four groups following standard classification including underweight (<18.5 kg/m²), normal weight (18.5−24.9 kg/m²), overweight (25.0−29.9 kg/m²), and obese (≥30.0 kg/m²) [19]. The WHR was computed using waist and hip circumferences measured at baseline and categorized as poor (≥0.9 for men, ≥0.85 for women) or ideal (<0.9 for men, <0.85 for women) [19]. Never smoking referred to participants reporting neither current nor past smoking at baseline [20]. Moderate alcohol intake was defined as an average consumption not exceeding 16 grams of pure alcohol daily (equivalent to 2 units) for both sexes [21]. Regular physical activity was self‑reported at baseline via the UK Biobank touchscreen questionnaire, which includes International Physical Activity Questionnaire (IPAQ)-derived items on frequency and duration of walking, moderate and vigorous activity and lifestyle physical behaviors [22]. Regular physical activity was defined following American Heart Association criteria as at least 150 min of moderate, 75 min of vigorous exercise per week, or an equivalent combination [23]. A healthy diet was defined as adherence to at least four of seven recommended food groups, including ≥3 servings/day of fruit, ≥3 servings/day of vegetables, ≥2 servings/week of fish, ≤1 serving/week of processed meats, ≤1.5 servings/week of unprocessed red meats, ≥3 servings/day of whole grains, and ≤1.5 servings/day of refined grains [24]. A healthy sleep pattern was characterized by meeting at least four of five indicators: an early chronotype, sleeping 7 h–8 h per day, experiencing insomnia rarely or never, not self-reporting snoring, and dozing off during the day rarely or never [25].

2.5. Polygenic risk score

To assess individual genetic predisposition to dementia, PD, and MS, polygenic risk scores (PRS) were constructed as aggregated measures of genetic susceptibility, based on sets of single nucleotide polymorphisms (SNPs) identified in large-scale genome-wide association studies (GWAS). Specifically, the PRS for dementia, PD, and MS were derived from 83, 44, and 155 SNPs, respectively (Supplementary Table S6-S8), all of which met stringent quality control standards to ensure data reliability and validity. PRS in this study were calculated using a weighted allele-counting approach, whereby each SNP was coded additively as 0, 1, or 2 according to the number of risk alleles present in an individual. These allele counts were then multiplied by the corresponding effect size (β coefficient) reported in previous GWAS meta-analyses for each target disease [[26], [27], [28]]. The resulting products were summed across all included SNPs to generate a composite score for each participant, using the formula as $PRS={\sum }_{i=1}^j{\beta }_i\times {SNP}_i$, where j denotes the total number of SNPs, β_i represents the effect size for the i-th SNP, and SNPi corresponds to the allele dosage. Thus, the PRS reflects the cumulative genetic risk contributed by multiple loci, with higher scores indicating a greater genetic predisposition to the corresponding neurodegenerative condition. For analytical purposes, the continuous PRS was subsequently stratified into three categories reflecting distinct levels of genetic risk: low (the lowest tertile), moderate (the middle tertile), and high (the highest tertile).

2.6. Statistical analysis

The baseline characteristics of study participants were presented as counts with corresponding percentages for categorical variables and as medians with interquartile ranges (IQRs) for continuous variables. Cox proportional hazards regression models were employed to evaluate the associations between UPF intake and the risks of incident neurodegenerative diseases, following confirmation that the proportional hazards assumption was satisfied using scaled Schoenfeld residuals. Hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) were estimated after full adjustment for potential confounders. To strengthen the robustness of our findings, UPF intake was also analyzed as a continuous variable in relation to neurodegenerative disease risks. Furthermore, stratified analyses were conducted to explore potential effect modifiers, including genetic risk (low/moderate/high), age (<60/≥60), sex (female/male), WHR (ideal/poor), and lifestyle factors not directly related to UPF intake including never smoking, healthy sleep pattern, and regular physical activity (each categorized as yes/no). Interaction tests were performed by introducing multiplicative interaction terms into the models to assess potential heterogeneity in the observed associations.

For subsequent neuroimaging analyses, multiple linear regression models with the same covariate adjustments as described above were applied to examine the associations between UPF intake and structural changes in brain gray matter, encompassing both cortical and subcortical regions, with neuroimaging data standardized prior to analysis. To explore potential lateralized effects, subcortical gray matter volumes were analyzed separately for the left and right hemispheres. Additionally, associations between UPF intake and cortical gray matter characteristics, including volume, mean thickness, and surface area, were systematically evaluated. Effect estimates were reported as standardized βs with corresponding 95% CIs. False discovery rate (FDR) correction was applied to all P values.

A series of sensitivity analyses were also conducted. First, the analyses were restricted to participants of White European ancestry to minimize potential bias arising from population stratification. Second, additional adjustments were made for a range of environmental factors, including proximity to roadways, noise pollution, nitrogen oxide, fine particulate matter (PM_2.5), and ambient benzene, to account for possible environmental confounders [29]. Third, to reduce the likelihood of reverse causation, participants who had been diagnosed with the targeted disease events within the first two years of follow-up were excluded. Fourth, considering the potential influence of the COVID-19 pandemic on disease incidence and related behaviors, the follow-up period was truncated on December 31, 2019. Fifth, instead of using the proportion of UPF intake, we re-estimated the models based on the absolute amount of UPF consumption (grams per day) to examine the consistency of associations [30]. Sixth, to address the issue of competing risk events that may influence the estimation of cause-specific hazards, Fine-Gray subdistribution hazard models were employed in place of the primary model. Seventh, we redefined UPF intake by excluding alcoholic beverages from the UPF group and instead included them as an adjustment covariate, to account for potential exposure misclassification due to limited information on beverage processing. Eighth, to assess the robustness of our findings to alternative definitions of exposure, we re-estimated UPF intake as the proportion of total energy derived from UPFs, based on standard portion sizes and UK-specific nutrient composition data [31]. The mean energy contribution of UPFs was 47.3% (standard deviation = 16.7). Participants were grouped into tertiles of energy-based UPF intake (low, moderate, and high), with the lowest tertile serving as the reference category. Finally, to assess the dependence of statistical significance on sample size, repeated subsampling (10%–100% in 10% increments, with 100 replicates each) was performed, and P values were modeled as an exponential function of sample size using Monte Carlo cross-validation [32]. Together, these sensitivity analyses comprehensively evaluated the stability of our results under varying assumptions and model specifications. R software (version 4.3.3) was used for all statistical analyses, with significance determined by a two-tailed P value of less than 0.05.

3. Results

In the full UK Biobank cohort (N = 502,369), the age range was 37–73 years (median: 58 years), with 54.4% female. After excluding participants with missing data on exposures, outcomes, or relevant covariates, as well as those with pre-existing neurodegenerative diseases, incomplete genetic information, excessive heterozygosity, sex mismatch, or loss to follow-up, a total of 58,423 participants (age range: 40–70 years; median: 56 years; 53.5% female) were included in our primary analysis. Table 1 shows that among the analytical sample, 58.3% were under the age of 60, and 47.4% had poor WHR. Participants with low, middle, and high socioeconomic status accounted for 24.6%, 54.7%, and 20.8% of the study population, respectively. Regarding lifestyle factors, 56.4% were never smokers, 67.3% reported moderate alcohol intake, 55.8% engaged in regular physical activity, 55.4% maintained a healthy sleep pattern, and 40.1% adhered to a healthy diet. Notably, individuals with higher UPF intake were more likely to exhibit poor WHR, higher BMI, and unfavorable lifestyle behaviors. For neuroimaging analyses, a subsample of 5,462 participants (age range: 40–70 years; median: 57 years; 52.1% female) was included. Detailed baseline characteristics of this subsample are provided in Supplementary Table S9. To assess the representativeness of this imaging subsample, baseline characteristics were compared between participants with and without neuroimaging data (Supplementary Table S10). Supplementary Table S11 presents the distribution of neurodegenerative disease incidence stratified by UPF intake, including total person-years and incidence rates.

Table 1.

Baseline characteristics of participants grouped by ultra-processed food intake level.

Characteristics	All participants (N = 58423)	Ultra-processed food intake level a			P value b
		Low (N = 19455)	Moderate (N = 19454)	High (N = 19514)
Age					<0.001
<60	34087 (58.3%)	10759 (55.3%)	10859 (55.8%)	12469 (63.9%)
≥60	24336 (41.7%)	8696 (44.7%)	8595 (44.2%)	7045 (36.1%)
Sex					<0.001
Female	31253 (53.5%)	10673 (54.9%)	10539 (54.2%)	10041 (51.5%)
Male	27170 (46.5%)	8782 (45.1%)	8915 (45.8%)	9473 (48.5%)
SESc					<0.001
Low	12950 (24.6%)	4938 (28.0%)	4250 (24.3%)	3762 (21.4%)
Middle	28822 (54.7%)	9462 (53.6%)	9637 (55.0%)	9723 (55.4%)
High	10951 (20.8%)	3242 (18.4%)	3629 (20.7%)	4080 (23.2%)
WHRd					<0.001
Ideal	30531 (52.3%)	10647 (54.7%)	10394 (53.4%)	9490 (48.7%)
Poor	27873 (47.7%)	8808 (45.3%)	9054 (46.6%)	10011 (51.3%)
BMI					<0.001
<18.5	314 (0.5%)	127 (0.7%)	119 (0.6%)	68 (0.3%)
18.5−24.9	20135 (34.5%)	7459 (38.3%)	6983 (35.9%)	5693 (29.2%)
25.0−29.9	24221 (41.5%)	7888 (40.5%)	8157 (41.9%)	8176 (41.9%)
≥30.0	13753 (23.5%)	3981 (20.5%)	4195 (21.6%)	5577 (28.6%)
Healthy lifestyle
Never smoking e	33027 (56.4%)	11342 (58.3%)	10933 (56.2%)	10752 (55.1%)	<0.001
Moderate alcohol intake f	39325 (67.3%)	11914 (61.3%)	13163 (67.7%)	14248 (73.1%)	<0.001
Regular physical activity g	32571 (55.8%)	11135 (57.2%)	10893 (56.0%)	10543 (54.0%)	<0.001
Healthy sleep pattern h	32358 (55.4%)	11082 (57.0%)	10929 (56.2%)	10347 (53.0%)	<0.001
Healthy diet i	23415 (40.1%)	9174 (47.2%)	7815 (40.2%)	6426 (32.9%)	<0.001
Prevalent disease
Hypertension	15492 (26.5%)	5149 (26.5%)	5041 (25.9%)	5302 (27.2%)	0.019
Diabetes	2769 (4.7%)	912 (4.7%)	796 (4.1%)	1061 (5.4%)	<0.001
Stroke	711 (1.2%)	213 (1.1%)	251 (1.3%)	247 (1.3%)	0.160
Metabolic biomarker
Glucose	5.0 (4.7, 5.4)	5.0 (4.7, 5.3)	5.0 (4.7, 5.3)	4.9 (4.6, 5.3)	<0.001
HbA1c	35.0 (32.6, 37.6)	35.0 (32.6, 37.5)	35.0 (32.7, 37.5)	35.1 (32.6, 37.8)	0.001
TG	1.4 (1.0, 2.0)	1.3 (0.9, 1.9)	1.4 (1.0, 2.0)	1.5 (1.0, 2.2)	<0.001
LDL	3.5 (2.9, 4.0)	3.5 (2.9, 4.0)	3.5 (2.9, 4.1)	3.5 (2.9, 4.0)	0.002
Total cholesterol	5.6 (4.9, 6.4)	5.6 (4.9, 6.4)	5.6 (4.9, 6.4)	5.5 (4.8, 6.3)	<0.001

Abbreviations: SES, socioeconomic status; WHR, waist-to-hip ratio; BMI, body mass index; HbA1c, glycated hemoglobin; TG, triglycerides; LDL, low-density lipoprotein cholesterol.

^aParticipants' ultra-processed food intake was categorized into tertiles based on the distribution, classified as low, moderate, and high.

^bGroup comparisons were conducted using analysis of variance (ANOVA) or the chi-square (χ²) test, as appropriate.

^cSES was classified into low, middle, and high categories based on household income, highest educational attainment, and employment status, with higher SES indicating superior socioeconomic standing.

^dThe WHR was calculated as waist circumference (cm) divided by hip circumference (cm) and classified as ideal (<0.9 for men and <0.85 for women) or poor (≥0.9 for men and ≥0.85 for women).

^eNever smoking was defined as the absence of both past and current smoking at baseline.

^fModerate alcohol intake was defined as an average daily intake of ≤16 g of pure alcohol (equivalent to ≤2 units) for both men and women.

^gRegular physical activity was defined as engagement in at least 150 min of moderate-intensity activity per week, at least 75 min of vigorous-intensity activity per week, or an equivalent combination of both.

^hA healthy sleep pattern was assessed based on five dimensions of sleep behavior: early chronotype, sleep duration of 7–8 h per day, absence of frequent insomnia (never/rarely or sometimes), no self-reported snoring, and absence of excessive daytime sleepiness (never/rarely or sometimes). Participants meeting at least four of these five criteria were classified as having a healthy sleep pattern.

ⁱA healthy diet was defined as the consumption of at least four out of seven key food groups prioritized for cardiometabolic health. The specific criteria for each component were as follows: ≥3 servings/day of fruits, ≥3 servings/day of vegetables, ≥2 servings/week of fish, ≤1 serving/week of processed meats, ≤1.5 servings/week of unprocessed red meats, ≥3 servings/day of whole grains, and ≤1.5 servings/day of refined grains.

Table 2 presents associations of UPF intake with risks of dementia, PD, and MS. After full adjustment for sociodemographic factors, anthropometric indices, lifestyle factors, prevalent diseases, and metabolic biomarkers, higher UPF intake was found significantly associated with higher risks of specific neurodegenerative diseases. Compared with participants in the low UPF intake group, those with the high UPF intake exhibited significantly increased risks of both dementia (HR [95% CI] = 1.37 [1.08, 1.74]), PD (1.76 [1.22, 2.53]) and MS (2.38 [1.02, 5.55]). Subgroup analyses stratified by genetic risk further revealed that the adverse associations of high UPF intake with both dementia and PD were more pronounced among participants with lower PRS. Specifically, for dementia, high UPF intake was associated with an increased risk of 53%, 33%, and 9% in the low, moderate, and high PRS groups, respectively. Similarly, for PD, the corresponding risk increases were 87%, 62%, and 46%, indicating that individuals with lower genetic susceptibility may be more vulnerable to the detrimental effects of UPF consumption. These associations remained robust when UPF intake was modeled as a continuous variable. Additional stratified analyses are presented in Table 3. Notably, among participants with high UPF intake, those aged 60 years and older had a significantly higher risk of developing dementia (P interaction = 0.037). Elevated risks of PD were also observed in female participants and individuals with poor WHR (P interaction = 0.019 for sex-stratified analysis, P interaction = 0.050 for WHR-stratified analysis). Associations of UPF intake with disease outcomes remained largely consistent across stratifications by lifestyle factors, as shown in Supplementary Table S12.

Table 2.

Associations of ultra-processed food intake and neurodegenerative diseases.

	Ultra-processed food intake level a				Continuous c
Disease	Low	Moderate	High	P trend b	HR (95% CI)	P value d
Dementia
All	Reference	1.15 (0.91, 1.47)	1.37 (1.08, 1.74)	<0.001	1.32 (1.17, 1.48)	<0.001
Low PRS e	Reference	1.29 (1.12, 1.63)	1.53 (1.28, 1.88)	<0.001	1.43 (1.24, 1.65)	<0.001
Moderate PRS	Reference	1.16 (0.98, 1.48)	1.33 (1.10, 1.56)	<0.001	1.27 (1.14, 1.42)	<0.001
High PRS	Reference	1.05 (0.86, 1.31)	1.09 (1.00, 1.16)	0.001	1.06 (0.93, 1.12)	0.058
PD
All	Reference	1.14 (0.77, 1.66)	1.76 (1.22, 2.53)	<0.001	1.38 (1.16, 1.63)	<0.001
Low PRS	Reference	1.19 (0.83, 1.84)	1.87 (1.40, 2.74)	0.001	1.57 (1.31, 1.76)	<0.001
Moderate PRS	Reference	1.14 (0.77, 1.66)	1.62 (1.20, 2.37)	0.001	1.34 (1.12, 1.59)	<0.001
High PRS	Reference	1.10 (0.75, 1.49)	1.46 (1.17, 1.92)	0.023	1.22 (1.05, 1.33)	0.041
MS
All	Reference	1.29 (0.51, 2.57)	2.38 (1.02, 5.55)	0.246	1.71 (0.87, 3.69)	0.560
Low PRS	Reference	1.35 (0.87, 2.80)	2.46 (1.12, 5.73)	0.392	1.93 (0.91, 3.77)	0.577
Moderate PRS	Reference	1.30 (0.56, 2.49)	2.31 (1.05, 4.87)	0.357	1.82 (0.83, 3.43)	0.450
High PRS	Reference	1.32 (0.54, 1.98)	2.30 (0.95, 4.34)	0.421	1.68 (0.74, 2.75)	0.468

Abbreviations: PD, Parkinson’s disease; MS, multiple sclerosis; PRS, polygenic risk score; HR, hazard ratio; CI, confidence interval.

^aParticipants' ultra-processed food intake was categorized into tertiles based on the distribution and classified as low, moderate, and high levels, with the low level serving as the reference group.

^bA low P trend (typically <0.05) indicates statistical significance, suggesting a trend across tertiles.

^cHRs with 95% CLs were calculated for each interquartile range increase in the proportion of ultra-processed food intake.

^dA low P value (typically <0.05) indicates statistical significance.

^eThe PRS score was classified into low, moderate, and high according to the distribution, with higher PRS representing the higher genetic susceptibility to specific disease.

Table 3.

Associations of ultra-processed food intake and neurodegenerative diseases by age, sex, and WHR.

	Dementia		PD		MS
Subgroup	HR (95% CI)	P interactiona	HR (95% CI)	P interaction	HR (95% CI)	P interaction
Age
<60	1.16 (1.03, 1.31)	0.037	1.64 (1.06, 1.82)	0.137	2.51 (1.05, 5.76)	0.864
≥60	1.64 (1.34, 2.06)		1.84 (1.28, 2.65)		2.12 (0.86, 5.07)
Sex
Female	1.33 (1.09, 1.67)	0.818	2.05 (1.37, 3.07)	0.019	2.68 (1.19, 5.81)	0.884
Male	1.47 (1.23, 2.54)		1.64 (1.02, 2.17)		2.03 (0.88, 4.87)
WHRb
Poor	1.20 (0.89, 1.63)	0.138	2.07 (1.34, 3.20)	0.050	2.45 (1.07, 6.08)	0.840
Ideal	1.49 (1.03, 2.15)		1.43 (1.06, 1.57)		1.96 (0.71, 5.15)

Abbreviations: PD, Parkinson’s disease; MS, multiple sclerosis; HR, hazard ratio; CI, confidence interval; WHR, waist-to-hip ratio; BMI, body mass index.

^aA low ‘P interaction’ (typically <0.05) indicates a significant interaction effect, suggesting that the association between ultra-processed food intake levels and disease incidence differs across subgroups.

^bThe WHR was calculated as waist circumference (cm) divided by hip circumference (cm) and classified as ideal (<0.9 for men and <0.85 for women) or poor (≥0.9 for men and ≥0.85 for women).

This study further evaluated the relationships of UPF intake with GM structures within 5,462 participants with available neuroimaging data. GM phenotypes were classified into subcortical structures, including mean, left, and right hemispheric volumes, and cortical structures, encompassing volume, thickness, and surface area (Supplementary Table S13-S14). In the analysis of subcortical GM volumes, 6 out of 7 regions demonstrated significant inverse associations with UPF intake after FDR correction (Fig. 1). Specifically, higher UPF intake was associated with reduced mean volumes of the caudate nucleus (β [95% CI] = −0.023 [−0.062, −0.016]), amygdala (−0.068 [−0.107, −0.029]), hippocampus (−0.057 [−0.096, −0.019]), globus pallidus (−0.046 [−0.084, −0.006]), putamen (−0.044 [−0.080, −0.009]), and thalamus (−0.038 [−0.074, −0.002]). Hemisphere-specific analyses revealed lateralized effects in several subcortical regions. For example, the amygdala exhibited significant negative associations with UPF intake in both hemispheres, with slightly stronger effects observed in the right hemisphere (right amygdala: β [95% CI] = −0.076 [−0.116, −0.036]; left amygdala: −0.057 [−0.096, −0.019]). Similar right-dominant reductions were noted in the other subcortical structures including globus pallidus (right globus pallidus: −0.060 [−0.098, −0.023]; left globus pallidus: −0.043 [−0.084, −0.008]), putamen (right putamen: −0.056 [−0.091, −0.021]; left putamen: −0.041 [−0.077, −0.004]), and others. The associations in the left hemisphere were generally weaker but still significant for several regions (Supplementary Table S15).

Fig. 1.

Associations of UPF intake with average and hemisphere-specific subcortical gray matter volumes. The associations between UPF intake and subcortical gray matter volumes were evaluated using multiple linear regression models among 5,462 UK Biobank participants with available brain MRI data. β coefficients and corresponding false discovery rate (FDR)-adjusted P values are presented. The upper panel depicts the associations with average subcortical gray matter volumes, while the lower panel illustrates hemisphere-specific results. Color gradients from light to dark reflects increasing degrees of volume reduction, with larger absolute β values indicating more pronounced atrophy. Asterisks denote statistically significant associations after FDR correction.

Analyses of high UPF intake in relation to cortical GM structures, including volume, thickness, and surface area, are presented in Fig. 2. Significant reductions across all three measures were observed in cortical regions spanning five major lobes. For instance, cortical volume reductions were identified in the frontal lobe (e.g., caudal middle frontal gyrus: β [95% CI] = −0.056 [−0.095, −0.016]; precentral gyrus: −0.064 [−0.102, −0.026]), parietal lobe (e.g., postcentral gyrus: −0.076 [−0.114, −0.037]; precuneus: −0.054 [−0.092, −0.017]), occipital lobe (e.g., lingual gyrus: −0.068 [−0.108, −0.029]; cuneus: −0.059 [−0.098, −0.020]), and others. Notably, UPF intake was also negatively associated with cortical thickness, particularly in regions such as the parahippocampal gyrus (−0.044 [−0.085, −0.004]), supramarginal gyrus (−0.048 [−0.086, −0.011]). In addition, significant inverse associations with cortical surface area were observed, including the fusiform gyrus (−0.042 [−0.077, −0.007]) and insula (−0.058 [−0.093, −0.023]), among others, as detailed in Supplementary Table S16. Together, these findings indicated that higher UPF intake is associated with widespread and multi-dimensional cortical GM structural alterations. The results of the sensitivity analyses suggested that the main findings were robust (Supplementary Table S17-S18).

Fig. 2.

Associations between UPF intake and cortical gray matter volume, thickness, and surface area. The relationships between UPF intake and cortical gray matter morphology were examined using multiple linear regression models among 5,462 UK Biobank participants with available brain MRI data. β coefficients and corresponding FDR-adjusted P values are presented. Cortical regions were categorized into five major lobes: frontal, temporal, parietal, occipital, and insular lobes. Color gradients from light to dark reflect increasing degrees of structural deterioration, with larger absolute β values indicating greater morphological changes across volume, thickness, and surface area metrics. Asterisks denote statistically significant associations after FDR correction.

4. Discussion

Drawing on a large UK population-based cohort, this study systematically examined the associations of UPF intake with neurodegenerative diseases and brain structural alterations. We observed that higher UPF intake was associated with increased incidence risks of dementia, PD, and MS, especially among individuals with lower genetic risk. Moreover, higher UPF intake correlated with reduction in subcortical gray matter volume, especially the hippocampus and amygdala, with stronger associations observed in the right hemisphere generally. At the cortical level, higher UPF intake was associated with significant declines in volume, thickness, and surface area across major lobes, notably the frontal, temporal, and occipital regions. Collectively, these associations highlight the relevance of UPF intake in neurodegenerative research and suggest exploring potential underlying pathways.

Our findings extended current knowledge on UPF intake and neurodegenerative diseases by integrating both clinical outcomes and brain structural alterations, a dual perspective largely absent in prior research. As an increasing component of modern dietary patterns, UPFs have been extensively implicated in adverse health outcomes, with robust evidence linking their consumption to obesity, cardiovascular diseases, and others [33,34]. However, despite the systemic effects, studies on the neurological consequences of UPF intake remain relatively underdeveloped [35]. While dementia has received more attention, previous epidemiological studies on its association with UPFs have been inconsistent, and studies examining other major neurodegenerative diseases, such as PD and MS, remain scarce [9,36]. Given the complex etiology of these disorders and their prolonged preclinical phases, existing research could be constrained by limited sample sizes and short follow-up periods, underscoring the need for large-scale longitudinal cohorts to strengthen the evidence base. Furthermore, emerging neuroimaging studies have linked dietary patterns such as high intakes of saturated fats and refined sugars to structural brain deterioration [37]. However, direct investigations into UPFs and brain structural alterations remain limited and fragmented. By jointly analyzing these outcomes, our study revealed both the heightened risk of neurodegenerative diseases and widespread gray matter deterioration associated with UPF intake, thus offering an integrated view that underscores the brain's structural vulnerability as a potential pathway linking diet to neurodegeneration.

Our findings provide compelling evidence that high UPF intake is associated with elevated risks of neurodegenerative diseases and widespread brain structural deterioration in middle-aged and older adults. Notably, the stronger associations observed among individuals with lower genetic susceptibility may reflect potential effect modification. Alternatively, this pattern may be attributable to differential measurement error—commonly encountered in nutritional epidemiology—which could disproportionately affect estimates in genetically low-risk populations [38,39]. In addition, the possibility of residual confounding by unmeasured behavioral or socioeconomic factors cannot be ruled out. Nonetheless, these findings underscore the relevance of dietary exposures as potentially modifiable contributors to brain health, irrespective of genetic background. Age-stratified analyses further revealed that the link between UPF intake and dementia was most pronounced in individuals aged ≥60 years, supporting the hypothesis that aging-related neuroinflammatory and oxidative stress processes heighten susceptibility to diet-induced neural damage [40]. Similarly, the amplified risk of PD in females and individuals with greater central adiposity highlights potential interactions between UPF consumption, neuroendocrine shifts, and metabolic dysfunction—factors that have been implicated in neurodegenerative processes [[41], [42], [43]]. Beyond clinical outcomes, the neuroimaging results revealed inverse associations between UPF intake and gray matter volume, thickness, and surface area—predominantly in regions characteristically vulnerable to age-related neurodegeneration. This spatially diffuse and right-lateralized pattern may signify structural alterations along neural circuits implicated in cognitive regulation and sensorimotor integration [[44], [45], [46]]. While the biological significance remains to be fully elucidated, the observed distribution aligns with mechanisms previously linked to neuroinflammatory processes, cerebrovascular compromise, and metabolic dysregulation in aging brains [47,48].

Several mechanisms may underlie the associations between UPF intake and neurodegenerative risks, involving both adverse nutritional composition and high degrees of industrial processing. Excessive refined sugars, unhealthy fats, and sodium in UPFs precipitate systemic inflammation, oxidative stress, and endothelial dysfunction—pathophysiological processes that collectively contribute to neurodegeneration [49,50]. Furthermore, extensive processing leads to the formation of neurotoxic compounds, such as advanced glycation end-products (AGEs), which impair neuronal integrity and exacerbate microvascular damage [51]. However, the pathways through which UPFs influence specific neurological outcomes may differ substantially. For dementia, particularly Alzheimer’s disease, chronic low-grade inflammation, cerebrovascular compromise, and disrupted insulin signaling are central to disease progression. UPF-induced metabolic dysfunction may exacerbate these processes, with possible implications for amyloid accumulation, tau pathology, and compromised brain integrity [12,52]. In contrast, Parkinson’s disease is characterized by dopaminergic neuronal loss and mitochondrial dysfunction. Dietary components of UPFs, including lipid peroxidation products and pro-oxidative additives, may aggravate oxidative stress within nigrostriatal circuits, potentially facilitating neurodegeneration [53]. Multiple sclerosis, as an immune-mediated demyelinating disease, may be uniquely vulnerable to diet-induced immune perturbations. High UPF consumption has been associated with gut microbiota dysbiosis and increased intestinal permeability, which can modulate systemic immune responses and potentially exacerbate autoimmune neuroinflammation [36]. Taken together, these differential mechanistic pathways highlight the necessity of disease-specific interpretations when evaluating dietary risk factors. While well-established dietary patterns such as the Mediterranean diet offer protective effects against cognitive decline, the NOVA classification system provides an essential framework for capturing the broader neurobiological consequences of food processing [54,55]. Integrating both compositional and processing dimensions into dietary assessments may enhance preventive strategies across heterogeneous neurodegenerative conditions.

Our study holds significant public health implications. First, it contributes robust epidemiological evidence on the associations between UPF intake and neurodegenerative outcomes, highlighting dietary composition as a potentially modifiable exposure relevant to brain health in contemporary populations [56]. Second, stratified analyses revealed specific subgroups with heightened vulnerability, offering valuable insights for the development of targeted preventive strategies. Third, by examining multidimensional brain structural parameters, the findings provide important neuroanatomical context that may support earlier identification of at-risk individuals and inform population-level monitoring frameworks for neurodegenerative conditions.

We should acknowledge several limitations in this study. First, exposure misclassification is possible due to the limited detail on food processing in the Oxford WebQ. Some items (e.g., yogurt, bread) exist in both minimally and ultra-processed forms, and classification relied on commonly consumed variants using established UK Biobank frameworks [57]. Second, weight-based and energy-based ratios capture distinct aspects of UPF intake, each with inherent methodological trade-offs. The weight-based ratio accounts for non-caloric yet heavily processed items and better reflects the structural modifications associated with processing, whereas the energy-based ratio aligns more closely with the caloric contribution of UPFs. Although we included the energy-based ratio in sensitivity analyses and observed generally consistent patterns, its estimation in the UK Biobank was limited by the absence of item-level calorie data, potentially compromising its accuracy [2,58]. Third, dietary data were collected using a 24 -h web-based recall, which is inherently prone to recall bias and random measurement error. Participants may misreport or omit certain foods due to memory limitations, social desirability, or misunderstanding of portion sizes, potentially attenuating true associations. While multiple recalls were averaged to better estimate habitual intake, such methods remain limited in fully capturing long-term dietary behavior. Additionally, reliance on an online platform may introduce selection bias, as individuals with limited digital access or literacy may be underrepresented [16]. Fourth, the identification of neurodegenerative diseases based on linked health records may result in under-ascertainment due to incomplete documentation, diagnostic delays, or potential misclassification, which could influence the observed associations. Fifth, although extensive covariate adjustments were performed, residual and unmeasured confounding cannot be entirely ruled out. Sixth, as with all observational studies, the associations identified are correlational and cannot establish causality. Seventh, the self-selected nature of the UK Biobank cohort may introduce healthy volunteer bias, as participants tend to be healthier and more health-conscious than the general population. Furthermore, our neuroimaging subsample appeared to be younger and exhibited overall healthier baseline characteristics compared with participants without imaging data, suggesting an additional layer of selection bias, which may lead to conservative effect estimates and limit the external validity of our findings [59]. Finally, the cohort consists predominantly of individuals of White European ancestry, which may constrain the generalizability of the results to more ethnically diverse populations, particularly in settings with differing genetic, cultural, or dietary backgrounds [60]. To enhance the evidence base, future research should prioritize longitudinal designs with multiple dietary assessments, refined UPF classification, and objective biomarkers to improve exposure accuracy. Additionally, more diverse populations and mechanistic investigations are needed to validate causal pathways and elucidate the neurobiological processes linking UPF intake to neurodegeneration.

5. Conclusions

In conclusion, our findings demonstrated that high UPF intake was associated with elevated risks of neurodegenerative diseases, including dementia, PD, and MS. Complementing these clinical associations, our neuroimaging analyses revealed that high UPF intake also correlated with lower GM characteristics, including lower subcortical volumes—particularly in the right hemisphere—as well as diminished cortical metrics across volume, thickness, and surface area. Taken together, our results suggested the potential role of UPFs in shaping long-term brain health and provided compelling support for public health strategies aimed at reducing UPF consumption. By targeting UPF as a modifiable dietary exposure, such robust epidemiological evidence may offer promising avenues to mitigate the growing burden of neurodegenerative diseases and enhance population-level brain well-being, thus fostering healthy aging.

CRediT authorship contribution statement

J.G.: formal analysis, methodology, validation, and writing—original draft; L.C.: visualization, writing—original draft, and funding acquisition; Y.B.: software and writing—review and editing; X.Y. and X.C.: data curation and software; Z.H., Y.Z. and X.D.: resources and investigation; Y.L.: conceptualization, writing—review and editing, supervision, project administration and funding acquisition; J.R.: conceptualization, writing—review and editing, supervision, project administration, and funding acquisition. All authors have read and agreed to the published version of the manuscript.

Ethics approval and consent to participate

The UK Biobank was designed to assess the relevance of numerous different exposures and to a range of health outcomes, which has been approved and supervised by the North West Multi-Center Research Ethics Committee, and written informed consent was obtained from all participants. We used data based on the UK Biobank with application number 99001, the approval date was from 2023-01-31 to 2026-01-31.

Funding

This work was supported by National Natural Science Foundation of China [grant number 82304102 (J. Ran)]; Natural Science Foundation of Shanghai pgrant number 23ZR1436200 (J. Ran)]; Scientific Research Fund Project of Yunnan Provincial Department of Education [grant number 2024J0208 (Y. Li)]; Minhang District Public Health Key Project [grant number MGWXK2023-11 (L. Chen)].

Availability of data and materials

All the data for this study will be made available upon reasonable request to the corresponding authors.

Declaration of competing interest

The authors declare no competing interests.

Appendix A. Supplementary data

The following are Supplementary data to this article: