Metabolomic signature of ultraprocessed food consumption and microvascular complications among individuals with type 2 diabetes

Abstract

Ultra-processed food (UPF) consumption has been associated with higher risks of diabetic microvascular complications. However, whether a metabolite profile related to UPF consumption reflects these associations was unknown. Using data from the UK Biobank (a prospective cohort study), a metabolomic signature comprising 14 metabolites was derived and robustly correlated with UPF consumption. Compared with UPF consumption, its corresponding metabolomic signature was also associated with higher risks of composite microvascular complications and diabetic kidney disease, with an additionally identified significant association with diabetic retinopathy and diabetic neuropathy. The above associations between the metabolomic signature and diabetic microvascular complications remained significant after further adjustment for UPF consumption. Three metabolites from lipid components of VLDL and HDL, MUFA, albumin, and glycoprotein acetyls mediated the association between UPF intake and composite microvascular complications. These findings suggest that the metabolomic signature has the potential to stratify individuals with different dietary responses to UPFs and facilitate personalized nutrition for individuals with type 2 diabetes.

Introduction

Type 2 diabetes (T2D), contributing to nearly 90% of diabetes cases, is a major public health problem^1,2. The number of adults with diabetes was estimated to be 537 million in 2021 and is expected to increase to 783 million in 2045³. Diabetic microvascular complications, including diabetic retinopathy, diabetic kidney disease, and diabetic neuropathy, affect over 50% of individuals with T2D⁴. These complications can cause loss of vision, chronic pain, and end-stage renal disease, imposing a substantial socioeconomic burden and deteriorating quality of life^5–8.

Ultra-processed foods (UPFs) are industrial formulations made mostly from substances including colors, flavors, and other additives, with little intact whole food⁹. These foods are designed for longer shelf life and usually undergo biological, physical, and chemical processes, including extrusion, hydrogenation, and preprocessing by frying⁹. The category of UPFs encompasses a diverse range of foods such as reconstituted meat products, carbonated drinks, and breakfast cereals. Our previous study has reported that higher UPF intake was associated with higher risks of diabetic microvascular complications, and the association was partly mediated by traditional metabolic risk factors, including C-reactive protein, triglycerides, and body mass index (BMI)¹⁰. Given the above mediators provide the basic assumption that multiple pathways may link UPFs to diabetic microvascular complications, additional studies could further capture individuals’ metabolic responses to UPFs and elucidate the underlying mechanisms.

As a measurement of a broad spectrum of small-molecule metabolites in the biological system, metabolomics has emerged as a powerful tool in nutritional research. By objectively reflecting the body’s responses to diet intake, metabolomic profiling could complement traditional dietary assessment methods and offer insights into biological mechanisms underlying diet-disease relationships^11,12. Previous studies have constructed metabolomic features of UPF intake among general adult populations, which were further linked to liver disease, mental disorders, and mortality^13–15. However, compared with individuals without diabetes, those with diabetes have different levels of carbohydrate, lipid, and amino acid metabolites¹⁶. To our knowledge, little is known about the metabolomic feature of UPF intake among individuals with T2D and its association with risks of microvascular complications. In addition, it is unclear whether and the extent to which metabolites could relate UPFs to diabetic microvascular complications.

Therefore, using plasma metabolites among individuals with T2D in the UK Biobank, we aimed to identify the metabolomic signature of UPF consumption and prospectively investigate its associations with risks of microvascular complications. Additionally, we explored the potential individual metabolites linking UPFs to diabetic microvascular complications.

Results

3597 participants with T2D were included in the analysis to derive the metabolomic signature of UPF intake. The mean age of participants was 60.0 ± 6.7 y, and 2318 (64.4%) were male. There were 1848 (51.4%), 781 (21.7%), 543 (15.1%), 318 (8.8%), and 107 (3.0%) participants completing one to five dietary recalls within 36 months from the baseline assessment, respectively.

Baseline characteristics for each quartile of the proportion of UPFs in the diet are presented in Table 1. Participants with higher UPF consumption tended to be younger, Black or Black British participants, less educated, more deprived, never drinkers, never smokers, and physically inactive. With increasing UPF consumption, participants were more likely to have higher BMI, hypertension, and multiple medications for diabetes. Regarding diet factors, higher UPF consumers tended to have higher intakes of energy, free sugar, saturated fat, monounsaturated fat, and sodium, but lower intakes of vegetables, fruits, protein, and polyunsaturated fat. In addition, similar baseline characteristics were observed among participants with available dietary data and the subset with metabolomic data. In general, moderate or strong correlations existed among the lipoprotein subclasses (Supplementary Fig. 1).

Table 1.

Baseline characteristics among individuals with type 2 diabetes in the UK Biobank

	Quartiles of ultra-processed food consumptiona				Participants with both available dietary and metabolomic signature data (n = 3597)	Participants with available dietary data (n = 6737)
	Q1	Q2	Q3	Q4
Weight proportion (%), Median [IQR]	7.7 [5.6–9.5]	14.6 [13.0–16.2]	21.8 [19.7–24.0]	34.9 [29.9–41.8]	17.9 [11.3–26.9]	17.8 [11.2–26.7]
Number of subjects	898 (25.0)	900 (25.0)	900 (25.0)	899 (25.0)	–	–
Age(years)	60.2 ± 6.2	60.6 ± 6.3	60.3 ± 6.7	58.7 ± 7.2	60.0 ± 6.7	59.9 ± 6.8
Number of dietary recalls	1.8 ± 1.0	2.0 ± 1.2	2.0 ± 1.2	1.8 ± 1.1	1.9 ± 1.1	1.9 ± 1.1
Male	579 (64.5)	580 (64.4)	580 (64.4)	579 (64.4)	2318 (64.4)	4333 (64.3)
Race
White participants	804(89.9)	816 (91.3)	807 (90.6)	816 (91.2)	3243 (90.7)	5964 (89.0)
Mixed participants	12 (1.3)	17 (1.9)	26 (2.9)	13 (1.5)	68 (1.9)	142 (2.1)
Asian or Asian British participants	58 (6.5)	42 (4.7)	37 (4.2)	39 (4.4)	176 (4.9)	386 (5.8)
Black or Black British participants	20 (2.2)	19 (2.1)	21 (2.4)	27 (3.0)	87 (2.4)	206 (3.1)
College	317 (35.6)	291 (32.7)	287 (32.2)	241 (27.0)	1136 (31.9)	2178 (32.6)
Townsend Deprivation Index	−1.0 ± 3.1	−1.2 ± 3.0	−1.0 ± 3.1	−0.5 ± 3.3	−0.9 ± 3.1	−0.8 ± 3.2
Smoking status
Never	366 (40.9)	418 (46.8)	418 (46.6)	418 (46.7)	1620 (45.3)	3024 (45.1)
Past	441 (49.3)	405 (45.3)	422 (47.0)	408 (45.6)	1676 (46.8)	3129 (46.7)
Current	87 (9.7)	71 (7.9)	57 (6.4)	69 (7.7)	284 (7.9)	554 (8.3)
Drinking status
Never	43 (4.8)	43 (4.8)	59 (6.6)	59 (6.6)	204 (5.7)	409 (6.1)
Past	40 (4.5)	37 (4.1)	48 (5.4)	83 (9.3)	208 (5.8)	398 (6.0)
Current light to moderate	510 (57.0)	618 (69.1)	623 (70.0)	634 (71.2)	2385 (66.8)	4403 (65.8)
Current heavy	301 (33.7)	197 (22.0)	160 (18.0)	114 (12.8)	772 (21.6)	1478 (22.1)
Leisure-time physical activity (MET-min/week)	552.2 ± 823.3	535.5 ± 798.4	481.8 ± 713.4	470.3 ± 807.7	509.9 ± 787.2	524.9 ± 797.2
Body mass index (kg/m2)	30.5 ± 5.5	30.8 ± 5.8	31.4 ± 5.2	32.8 ± 6.4	31.4 ± 5.8	31.2 ± 5.8
Family history of CVD	566 (63.0)	585 (65.0)	586 (65.1)	545 (60.6)	2282 (63.4)	4232 (62.8)
History of hypertension	692 (77.1)	684 (76.0)	711 (79.0)	724 (80.5)	2811 (78.1)	5259 (78.1)
History of hyperlipidemia	737 (82.1)	728 (80.9)	730 (81.1)	748 (83.2)	2943 (81.8)	5444 (80.8)
History of CVD	160 (17.8)	196 (21.8)	171 (19.0)	200 (22.2)	727 (20.2)	1332 (19.8)
History of cancer	124 (13.8)	103 (11.4)	121 (13.4)	103 (11.5)	451 (12.5)	816 (12.1)
Duration of diabetes	6.4 ± 6.8	6.3 ± 6.3	6.1 ± 5.8	6.8 ± 6.4	6.4 ± 6.3	6.4 ± 6.5
Use of glucose-lowering medications
None	298 (33.2)	307 (34.1)	287 (31.9)	257 (28.6)	1149 (31.9)	2144 (31.8)
Only oral drugs	485 (54.0)	471 (52.3)	471 (52.3)	488 (54.3)	1915 (53.2)	3574 (53.1)
Insulin and others	115 (12.8)	122 (13.6)	142 (15.8)	154 (17.1)	533 (14.8)	1019 (15.1)
Dietary factors
Total energy intake (kcal/d)	1887.0 ± 576.4	2042.1 ± 581.4	2086.1 ± 570.2	2078.3 ± 607.7	2023.4 ± 589.4	2021.6 ± 583.6
Fruit (g/d)	242.6 ± 188.2	245.2 ± 195.7	229.1 ± 177.8	206.4 ± 181.0	230.8 ± 186.4	227.7 ± 181.5
Vegetable (g/d)	335.9 ± 252.2	323.9 ± 216.2	311.5 ± 220.7	249.0 ± 214.0	305.1 ± 228.7	305.3 ± 230.3
Protein (% total energy intake)	17.21 ± 4.09	16.77 ± 3.64	16.77 ± 3.54	16.16 ± 4.03	16.73 ± 3.85	16.73 ± 3.88
Saturated fat (% total energy intake)	10.74 ± 3.52	11.39 ± 3.43	11.70 ± 3.43	11.94 ± 3.55	11.44 ± 3.51	11.49 ± 3.54
Monounsaturated fat (% total energy intake)	11.11 ± 3.20	11.57 ± 2.91	11.66 ± 3.00	11.75 ± 3.02	11.52 ± 3.04	11.54 ± 3.11
Polyunsaturated fat (% total energy intake)	5.66 ± 2.06	5.78 ± 1.88	5.65 ± 1.70	5.59 ± 1.78	5.67 ± 1.86	5.69 ± 1.89
Trans fat (% total energy intake)	0.52 ± 0.30	0.52 ± 0.26	0.53 ± 0.28	0.51 ± 0.26	0.52 ± 0.27	0.52 ± 0.27
Free sugar (g/d)	32.5 ± 25.7	42.0 ± 24.8	52.3 ± 29.5	67.1 ± 42.9	48.5 ± 34.1	48.5 ± 34.0
Fiber (g/d)	17.6 ± 7.2	18.9 ± 6.9	18.6 ± 6.7	17.2 ± 6.8	18.1 ± 6.9	18.1 ± 6.9
Sodium (mg/d)	1794.5 ± 734.3	2063.5 ± 838.8	2141.0 ± 816.8	2172.4 ± 891.1	2043.0 ± 835.2	2047.0 ± 846.1

Values are n (%) for categorical variables and means ± standard deviations for continuous variables unless otherwise specified.

IQR interquartile range, MET metabolic equivalent, CVD cardiovascular disease.

^aQuartiles of the weight proportion of ultra-processed food intake in the total food consumed (%). Sex-specific cut-offs for quartiles of ultra-processed weight proportion were 11.2, 17.9, and 27.0 in males and 11.3, 17.9, and 26.6 in females.

Metabolite profiles related to UPF intake

Of the 169 metabolites, UPF intake was statistically significantly associated with 75 metabolites (false discovery rate [FDR] <0.05; Fig. 1 and Supplementary Table 1). Higher UPF intake was positively associated with the average diameter for very-low-density lipoprotein (VLDL) particles, concentrations of relatively larger sizes of VLDL, all categories of lipids (cholesterol, free cholesterol, cholesterol esters, triglycerides, phospholipids, and total lipids) from chylomicrons and extremely large VLDL and very large VLDL. It was also linked to higher triglyceride levels in all sizes of VLDL, medium and small HDL, intermediate-density lipoprotein (IDL), and all sizes of LDL, and elevated monounsaturated fatty acids (MUFA) and glycoprotein acetyls. On the contrary, higher UPF intake was inversely associated with the average diameter for HDL, LDL, and total lipoprotein particles, concentrations of relatively larger sizes of HDL particles, all categories of lipids except triglycerides from very large and large HDL, as well as sphingomyelins, apolipoprotein A1, degree of unsaturation, docosahexaenoic acid (DHA), histidine, and albumin.

Fig. 1. Associations between ultra-processed food intake (per 10%) and 169 metabolites (per SD) among individuals with type 2 diabetes in the UK Biobank.

Associations with ultra-processed food intake were conducted among 3597 participants with available baseline dietary and metabolomic data. β coefficients were adjusted for age (continuous), sex (male, female), total energy intake (continuous), the number of 24-h dietary recalls (continuous), race (White, Mixed, Asian, Black), education level (college/university degree, other degrees), smoking status (never, past, current), drinking status (never, past, current light to moderate, current heavy), leisure-time physical activity (continuous), Townsend Deprivation Index (continuous), history of hypertension (yes/no), history of hyperlipidemia (yes/no), history of cardiovascular disease (yes/no), history of diabetic microvascular complications (yes/no), history of cancer (yes/no), family history of cardiovascular disease (yes/no), duration of diabetes (continuous), and use of glucose-lowering medications (none, only oral medicine, insulin and others). ^*P < 0.05, ^**P < 0.01, ^***P < 0.001 (FDR-adjusted P values). Colors denote the association directions (red, positive; blue, inverse). The black solid lines indicate 95% confidence bands. VLDL very-low-density lipoprotein, IDL intermediate-density lipoprotein, LDL low-density lipoprotein, HDL high-density lipoprotein, P particle, XXL_VLDL chylomicrons and extremely large very-low-density lipoprotein, XL very large, L large, M medium, S small, XS very small, C cholesterol, FC free cholesterol, CE cholesterol esters, TG triglyceride, PL phospholipids, TL total lipids, ApoB apolipoprotein B, ApoA1 apolipoprotein A1, FA fatty acids, PUFA polyunsaturated fatty acids, MUFA monounsaturated fatty acids, SFA saturated fatty acids, LA linoleic acid, DHA docosahexaenoic acid, Corrected_Ala corrected alanine, Glu glutamine, Gly glycine, His histidine, Total_BCAA total concentration of branched-chain amino acids, Ile isoleucine, Leu leucine, Val valine, Phe phenylalanine, Tyr tyrosine, GlycA glycoprotein acetyls.

An elastic net model was then applied to the 75 significant metabolites to identify the metabolomic signature of UPF intake. The metabolomic signature consisting of 14 metabolites was significantly correlated with UPF intake (Spearman’s r = 0.18 in the training set and r = 0.19 in the testing set; both P < 0.001; Supplementary Fig. 2). The metabolomic signature explained 3.37% and 3.84% variance in UPF intake in the training and testing test, respectively. Among 14 selected metabolites, the positive weights were assigned to MUFA and glycoprotein acetyls, and the strongest negative weights were assigned to free cholesterol in chylomicrons and extremely large VLDL, phospholipids in HDL, and albumin. The detailed composition and weight of each metabolite within the metabolomic signature are shown in Supplementary Table 2.

Metabolomic signature and diabetic microvascular complications

Over a median follow-up period of 12.8 years, 517 cases of composite microvascular complications, 256 cases of diabetic retinopathy, 96 cases of diabetic neuropathy, and 262 cases of diabetic kidney disease were documented. In the multivariable-adjusted model, UPF intake was significantly associated with risks of composite microvascular complications (hazard ratio [HR] per SD increment = 1.14 [95% confidence interval, CI: 1.04, 1.24]) and diabetic kidney disease (HR = 1.17 [95% CI: 1.04, 1.32]), but not significantly associated with risks of diabetic retinopathy and diabetic neuropathy (Table 2). For the metabolomic signature, it was significantly associated with risks of composite microvascular complications (HR = 1.29 [95% CI: 1.18, 1.42]), diabetic retinopathy (HR = 1.17 [95% CI: 1.03, 1.33]), diabetic neuropathy (HR = 1.50 [95% CI: 1.22, 1.84]), and diabetic kidney disease (HR = 1.31 [95% CI: 1.15, 1.49]). Adding UFP intake or UPF-related traditional biomarkers (i.e. urea, HDL [high-density lipoprotein] cholesterol, triglycerides, apolipoprotein A, and C-reactive protein) to the basic model did not significantly improve the C statistic (Supplementary Table 3). In contrast, when the metabolomic signature was added to the basic model, the C statistic was significantly improved from 0.659 to 0.676. Furthermore, with the addition of the metabolomic signature, the continuous net reclassification improvement (NRI) and absolute integrated discrimination improvement (IDI) were 0.083 (0.022, 0.144) and 0.008 (0.002, 0.014), respectively.

Table 2.

Associations of ultra-processed food intake and the corresponding metabolomic signature with microvascular complications among individuals with type 2 diabetes in the UK Biobank

	Ultra-processed food intake		Metabolomic signature
	HR (95% CI)a	P value	HR (95% CI)a	P value
Composite microvascular complications
Model 1b	1.16 (1.13, 1.19)	<0.001	1.33 (1.30, 1.37)	<0.001
Model 2c	1.14 (1.04, 1.24)	0.003	1.29 (1.18, 1.42)	<0.001
Model 2c + mutual adjustment	1.10 (1.01, 1.20)	0.03	1.28 (1.16, 1.40)	<0.001
Diabetic retinopathy
Model 1b	1.09 (1.05, 1.13)	<0.001	1.20 (1.15, 1.25)	<0.001
Model 2c	1.05 (0.93, 1.19)	0.45	1.17 (1.03, 1.33)	0.02
Model 2c + mutual adjustment	1.03 (0.91, 1.17)	0.67	1.16 (1.02, 1.33)	0.02
Diabetic neuropathy
Model 1b	1.09 (1.03, 1.16)	0.004	1.49 (1.40, 1.58)	<0.001
Model 2c	1.10 (0.90, 1.34)	0.36	1.50 (1.22, 1.84)	<0.001
Model 2c + mutual adjustment	1.04 (0.84, 1.27)	0.74	1.49 (1.21, 1.84)	<0.001
Diabetic kidney disease
Model 1b	1.21 (1.17, 1.25)	<0.001	1.39 (1.34, 1.44)	<0.001
Model 2c	1.17 (1.04, 1.32)	0.009	1.31 (1.15, 1.49)	<0.001
Model 2c + mutual adjustment	1.13 (1.00, 1.27)	0.052	1.29 (1.13, 1.47)	<0.001

HR hazard ratio, CI confidence intervals.

^aHazard ratio per standard deviation increments in the proportion of ultra-processed food intake or the metabolomic signature.

^bModel 1: adjusted for age (continuous), sex (male, female), and total energy intake (continuous).

^cModel 2: Model 1 + the number of 24-hour dietary recalls (continuous), race (White, Mixed, Asian, Black), education level (college/university degree, other degrees), smoking status (never, past, current), drinking status (never, past, current light to moderate, current heavy), leisure-time physical activity (continuous), Townsend Deprivation Index (continuous), history of hypertension (yes/no), history of hyperlipidemia (yes/no), history of cancer (yes/no), family history of cardiovascular disease (yes/no), duration of diabetes (continuous), and use of glucose-lowering medications (none, only oral medicine, insulin and others).

Notably, the above associations for the metabolomic signature remained significant after adjusting for UPF intake, whereas the associations for UPF intake were attenuated or became non-significant after adjusting for the metabolomic signature (Table 2). In addition, the metabolomic signature mediated 26.2% and 23.8% of the associations of UPF intake with composite microvascular complications and diabetic kidney disease, respectively (Fig. 2).

Fig. 2. Mediators of associations of ultra-processed food intake with microvascular complications among individuals with type 2 diabetes in the UK Biobank.

Proportion of effect mediated by mediators with 95% CI and P value as produced by %MEDIATE macro in SAS are reported, in multivariable models adjusted for age (continuous), sex (male, female), total energy intake (continuous), the number of 24-h dietary recalls (continuous), race (White, Mixed, Asian, Black), education level (college/university degree, other degrees), smoking status (never, past, current), drinking status (never, past, current light to moderate, current heavy), leisure-time physical activity (continuous), Townsend Deprivation Index (continuous), history of hypertension (yes/no), history of hyperlipidemia (yes/no), history of cancer (yes/no), family history of cardiovascular disease (yes/no), duration of diabetes (continuous), and use of glucose-lowering medications (none, only oral medicine, insulin and others). Proportion refers to a 10% increment in the proportion of ultra-processed food. Mediation analyses were generated using the first imputed dataset. Other imputed datasets were omitted because they were similar to the first imputed dataset. Significant proportion mediated was defined based on a P value < 0.05. Error bars indicate 95% confidence intervals. The metabolomic signature included average diameter for VLDL and LDL particles, free cholesterol in chylomicrons and extremely large VLDL, total esterified cholesterol, cholesteryl esters in HDL and medium VLDL, phospholipids in VLDL and HDL, degree of unsaturation, monounsaturated fatty acids, docosahexaenoic acid, histidine, albumin, and glycoprotein acetyls. VLDL very-low-density lipoprotein, XXL_VLDL chylomicrons and extremely large very-low-density lipoprotein, HDL high-density lipoprotein, FC free cholesterol, CE cholesterol esters, PL phospholipids, MUFA monounsaturated fatty acids, GlycA glycoprotein acetyls.

Metabolites mediating the associations of UPF intake with diabetic microvascular complications

Six metabolites mediated the association of UPF intake with composite microvascular complications, including free cholesterol in chylomicrons and extremely large VLDL, phospholipids in VLDL, cholesteryl esters in HDL, MUFA, albumin, and glycoprotein acetyls (Fig. 2). The mediated proportion ranged between 4.2% and 9.6% among the six metabolites. Similarly, five metabolites were identified as potential mediators of the UPF intake-diabetic kidney disease association, including free cholesterol in chylomicrons and extremely large VLDL, phospholipids in VLDL, MUFA, albumin, and glycoprotein acetyls. The mediated proportion of the five metabolites ranged from 5.4% to 11.6%.

Subgroup and sensitivity analyses

The associations of UPF intake and the corresponding metabolomic signature with composite microvascular complications were largely consistent when stratified by age, BMI, sex, and the number of dietary recalls, with no significant interaction considering multiple testing (Supplementary Table 4). The results of sensitivity analyses did not substantially change when excluding participants who had diabetic microvascular complications within 3 y of follow-up, additionally adjusting for BMI or the healthy diet score, or removing from the multivariable model glucose-lowering medications (Supplementary Tables 5). The association between the metabolomic signature and diabetic kidney disease was attenuated but still significant when additionally adjusted for estimated glomerular filtration rate (eGFR). When using all 169 metabolites in the elastic net model, the metabolomic signature consisting of 18 metabolites (sharing 9 metabolites with the main analysis) showed similar associations with risks of diabetic microvascular complications (Supplementary Table 6). Among 6588 individuals with baseline metabolomic data but no valid baseline dietary data, the metabolomic signature was associated with all four outcomes of diabetic microvascular complications (Supplementary Table 7). When using 80% of the data for training and 20% for testing, the metabolomic signature consisted of 19 metabolites (including 14 metabolites in the main analysis) and showed essentially unchanged association with outcomes (Supplementary Table 8). In addition, the derived metabolomic signature of energy-based UPF proportion, which consisted of 7 metabolites and shared 5 metabolites with the main analysis, was also significantly associated with higher risks of composite microvascular complications, diabetic neuropathy, and diabetic kidney disease, supporting the robustness of our findings (Supplementary Table 9).

Discussion

Leveraging metabolomic data from this prospective cohort, we identified a metabolomic signature of UPF intake, which showed positive associations with higher risks of composite microvascular complications, diabetic retinopathy, diabetic neuropathy, and diabetic kidney disease, even after adjusting for UPF intake. Several potential metabolites including lipid components of VLDL and HDL subclasses, MUFA, albumin, and glycoprotein acetyls, partly mediated the association of UPF intake with the risk of composite microvascular complications.

Since UPFs are high in free sugar and saturated fat, and low in polyunsaturated fat, it is reasonable that a significant proportion of metabolites are engaged in lipid metabolic and fatty acid pathways. Previous studies have shown that UPF intake was related to lipoprotein profiles in the general adult population^{13–15,17,18}, but only part of the studies reported results on lipid composition within lipoprotein subclasses^13,14,17. In our study among individuals with T2D, with specific data on lipid composition within lipoprotein subclasses, we identified that the associations between UPF intake and lipoproteins were heterogeneous among subclasses; for example, the inverse associations between UPF intake and VLDL mainly occurred in lipid components among relatively larger VLDL. In addition, lipid composition within lipoprotein subclasses associated with UPF intake among individuals with T2D was concentrated, such as concentrated on lipid components among relatively larger VLDL and triglycerides among various lipoproteins. For fatty acids, our study results confirmed previous findings in the general adult population of positive associations of UPF intake with MUFA and most metabolites of MUFA and inverse association of UPF intake with DHA^{13–15,17,19,20}. High free sugar in UPF could be utilized for MUFA synthesis through the de novo lipogenesis (DNL) pathway^21,22. Apart from nutrient profiles, the associations with dyslipidemia could also be explained by chemical substances that are either formed during food processing (e.g. acrylamide), migrated from their packaging (e.g. bisphenol A), or added to UPFs (e.g. emulsifiers)^23–25. In addition, previous studies showed positive or negative associations between UPF intake and sphingomyelins^13,15,17,19. In our study, we observed that higher UPF intake was related to lower levels of plasma sphingomyelins. Our result is supported by the evidence that food processing of milk, which is an important source of dietary sphingomyelins, leads to a loss of sphingomyelin²⁶. For amino acids, consistent with our findings, previous studies observed that UPF intake was inversely associated with histidine or most metabolites of histidine^13,14,27. Given that histidine is an essential amino acid and mainly obtained from dietary proteins²⁸, the inverse association may be explained by low protein intake in high UPF diet. With respect to albumin, it is a biomarker of liver function. The decrease in albumin with high UPF intake is likely explained by its poor nutritional profile and the above-mentioned chemical substances, which might contribute to non-alcoholic fatty liver disease²⁹. Finally, the increase in glycoprotein acetyls with high UFP intake is likely linked to chemical substances added to UPFs or migrated from their packaging, which are shown to be associated with inflammation or immune-inflammatory response in the diabetic state^24,25.

In our previous findings, higher UPF intake was associated with higher risks of composite microvascular complications and diabetic kidney disease, but not significantly associated with diabetic retinopathy and diabetic neuropathy¹⁰. In comparison, the metabolomic signature of UPF intake was also positively related to risks of composite microvascular complications and diabetic kidney diseases in this study, with a newly identified significant association with diabetic retinopathy and diabetic neuropathy. Notably, the above associations of the metabolomic signature with diabetic microvascular complications were all stronger than UPF intake and remained significant even after adjusting for UPF intake. The outperformance of the metabolomic signature over UPF intake may be explained by not only reflecting metabolic changes linked to UPF intake but also integrating individual metabolic variability due to other factors that change dietary metabolism (e.g. genetic variabilities and microbiome)³⁰. Thus, the metabolomic signature could help to stratify individuals with different dietary responses to UPFs. For example, individuals reporting relatively low UPF intake but exhibiting high levels of the metabolomic signature may represent a subgroup particularly susceptible to adverse metabolic effects of UPFs, for whom stricter dietary advice and additional lifestyle interventions (e.g., higher levels of physical activity) could be prioritized. If our metabolomic signature is further validated, it holds promise in facilitating personalized nutrition in the dimension of food processing for individuals with T2D. UPF intake was modestly correlated with the metabolomic signature (r = 0.18–0.19) in our study. Although the metabolomic signature was derived by regressing UPF intake on metabolites, it more reflected an integrated view of the metabolic homeostasis resulting from UPF intake and individual metabolic responses to UPFs in the diabetic state. Consistent with our findings, previous studies among the general population also found relatively low correlation between UPF intake and the derived metabolomic signature, with the correlation coefficients ranging from 0.21 to 0.32^13,15. Even though the correlation was modest, considering the objectivity of the metabolomic signature, it may serve as a complementary strategy to evaluate diet. The metabolomic signature may be useful in validation studies or when self-reporting cannot be conducted. Future studies are needed to confirm and explore its use as a dietary biomarker. In addition, compared with the addition of UPF intake or UPF-related traditional biomarkers, the addition of the metabolomic signature significantly improved the C statistic, with an improvement in NRI and IDI. The results suggested that the metabolomic signature had the potential to improve the risk reclassification and discrimination of diabetic microvascular complications. The metabolomic signature is expected to be unaffected by reporting errors and captures more detailed metabolic information compared with traditional biomarkers³⁰, and these advantages may help to improve its prediction performance. Considering the high prevalence (over 50%) of microvascular complications among individuals with T2D⁴, an enhancement in risk prediction could yield significant public health benefits. Future studies are needed to verify the improvement in the prediction performance of the metabolomic signature.

Our analysis of plasma metabolites provided novel insights into understanding the associations between UPF intake and diabetic microvascular complications. The association between UPF intake and composite microvascular complications was partly mediated by BMI, triglycerides, and C-reactive protein in our previous study¹⁰. With metabolomic data, we identified glycoprotein acetyls as a mediator that supported the biological pathway of inflammation. Furthermore, compared with C-reactive protein, glycoprotein acetyls better reflects cumulative chronic inflammation³¹. We also identified phospholipids in VLDL and free cholesterol in chylomicrons and extremely large VLDL mediated the association, supporting previous findings that various VLDL lipid fractions are elevated among individuals with diabetic microvascular complications³². Another novel observation was that cholesteryl esters in HDL mediated the association between UPF intake and diabetic microvascular complications. Through the detailed metabolomic analysis, we were able to dissect HDL into finer lipid subfractions, which helps explain higher HDL cholesterol levels have been linked to a lower risk of diabetic kidney disease in previous findings^33–35. Previous studies showed that circulating MUFA is not strongly influenced by dietary MUFA and is mainly derived from the endogenous DNL pathway^36,37. Similarly, in our study, plasma MUFA was only weakly correlated to dietary MUFA (r = 0.03). Therefore, high plasma MUFA levels may hint at dysregulation of DNL and then contribute to many metabolic problems²². Some of these metabolic problems such as hyperglycemia and hyperlipidemia are risk factors for diabetic microvascular complications³⁸. In addition, we observed that albumin from plasma metabolomics mediated the association between UPF intake and the risk of composite microvascular complications. The result corroborates the findings that serum albumin is associated with diabetic microvascular complications³⁹.

The main strengths of this study include the prospective design, identifying a metabolomic signature reflecting metabolic response to UPF intake among individuals with T2D, and examining the association between the metabolomic signature and microvascular complications. In addition, we applied the elastic net regression to solve collinearity among metabolites, which performs well in high-dimensional data⁴⁰. Nevertheless, several limitations should be considered as well. First, as the Oxford WebQ used in the UK Biobank collected intake of pre-defined items representing major foods consumed in the UK and was not specifically designed to assess dietary intakes based on processing levels (e.g. not collecting the information on additives), it could not capture the full spectrum of UPFs. Although the Oxford WebQ has been validated for collecting nutrients and foods, it was not validated for collecting UPFs. All these factors may introduce misclassification of UPFs. However, such misclassification is most likely to be non-differential and thus result in possible underestimation of the magnitude of the observed associations. Second, the metabolomic platform used in the UK Biobank cannot measure some potential biomarkers of UPF intake, such as biomarkers of additives^15,27. More comprehensive detection methods could help draw a full picture of metabolites linking UPF intake and diabetic microvascular complications. Third, diet and metabolomic data reflected the baseline level, and potential changes over time may modify the strength of the results. Fourth, some participants did not have multiple dietary recalls, which may not reflect their usual dietary intake. However, we have adjusted the number of dietary recalls, and the results are similar when stratified by the number of dietary recalls. Fifth, our mediation analyses are hypothesis-generating. Therefore, we cannot confirm causal mediation relationships. Experimental studies are needed to further validate our findings. Sixth, the events of diabetic microvascular complications were mainly identified through hospital diagnoses. Hospital diagnoses are characterized by high specificity due to clinically verified conditions, though they may exhibit low sensitivity for the complications under investigation⁴¹. In addition, compared with diabetic kidney disease and diabetic retinopathy, which have gold standard diagnostic methods^7,42, diabetic neuropathy is more likely to be underdiagnosed, considering the lack of harmonized diagnostic criteria and heterogeneous clinical phenotypes⁴³. Thus, some outcome misclassification may have occurred. However, this misclassification should be unrelated to diet behavior and bias our results toward the null. Seventh, although we have excluded participants with prevalent cardiovascular disease (CVD) at baseline in the main analysis and excluded those who developed diabetic microvascular complications within 3 years of follow-up as a sensitivity analysis, the possibility of reverse causation could not be ruled out. For example, individuals experiencing early symptoms of diabetic microvascular complications may have already modified their diet before the baseline assessment. This situation may confound the observed associations of UPF intake and the corresponding metabolomic signature with outcomes. Eighth, residual confounding could not be excluded because of the observational study design. Ninth, participants in the UK Biobank mostly consist of White Europeans, which may limit the generalizability. The characteristics of UPF intake may differ by country; for example, UPF intake in Italians with T2D is lower than our study population and comprises mainly processed meat and crispbread rather than beverages in our study population^10,44. In addition, the frequency of subtypes of T2D exhibits variation among different populations¹. Unlike mild obesity-related diabetes and mild age-related diabetes clusters in European populations, the predominant cluster in Asian populations is severe insulin-deficient diabetes, which is associated with high risks of complications¹. Thus, replication of our findings is necessary in other large and independent populations. Finally, given that our sample size is relatively small when including participants with both dietary and metabolomic data, the statistical power may still not be enough to detect significant interactions in subgroup analyses.

In conclusion, our findings identified multi-metabolite profiles of UPF intake among individuals with T2D, which were positively associated with higher risks of composite microvascular complications, diabetic retinopathy, diabetic neuropathy, and diabetic kidney disease. The higher risk of composite microvascular complications possibly went through multiple metabolites including lipoprotein components of VLDL and HDL subclasses, MUFA, albumin, and glycoprotein acetyls. Future studies could further investigate these metabolites to provide mechanistic explanations.

Methods

Study population

The UK Biobank is a prospective study recruiting more than 500,000 participants aged 37 to 73 years in 22 sites across Wales, Scotland, and England from 2006 to 2010. At baseline, participants filled out a touchscreen questionnaire, participated in a face-to-face interview and physical measurements, and provided biological samples⁴⁵.

Participants with prevalent T2D at baseline were identified using electronic health records (ICD-10 code: E11) or an algorithm developed by the UK Biobank study. The algorithm identified T2D with 96% accuracy through medical and medication history⁴⁶. In the present study, after excluding participants without valid baseline dietary data, with extreme dietary intake (males with <800 kcal/day or >4200 kcal/day or females with <600 kcal/day or >3500 kcal/day), and without baseline plasma metabolome, a total of 3597 participants were included to derive the metabolomic signature correlated to UPF intake (Fig. 3). For the analysis of the metabolomic signature with diabetic microvascular complications, after a further exclusion of participants with prevalent microvascular complications and CVD, 2477 participants were included.

Fig. 3. Flow chart of the study design.

Participants with available baseline dietary and metabolomic data were randomly assigned to a training set (70%, n = 2517) and a testing set (30%, n = 1080). Elastic net regression, a regularized regression model that combines the Ridge and Lasso penalties, was used to construct a metabolomic signature related to UPF consumption. To make full use of the data in the training set, leave-one-out cross-validation was used to select optimal hyperparameters for elastic net regression. Based on the selected elastic net regression, a coefficient was assigned for each metabolite in the training set. The metabolomic signature was calculated as the weighted sum of metabolites with nonzero β coefficients in the training set. In the testing set, coefficients derived from the training set were applied, and the metabolomic signature was calculated in the same way.

The UK Biobank was approved by the Community Health Index Advisory Group in Scotland, the North West Multi-Centre Research Ethics Committee, and the National Information Governance Board for Health and Social Care in England and Wales in accordance with the Declaration of Helsinki. All participants provided written informed consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (Supplementary Checklist).

Dietary assessment

Self-reported dietary data were collected through the Oxford WebQ, a web-based 24-h dietary recall, which records 32 beverages and 206 foods in the previous 24 h. Participants were invited to fill out the Oxford WebQ on five separate occasions, including a baseline assessment between 2009 and 2010 and four online cycles between 2011 and 2012 via e-mail. The Oxford WebQ has been validated by an interviewer-administered 24-h dietary recall, demonstrating similar recordings of daily nutrient intake and estimated energy⁴⁷. In addition, the Oxford WebQ showed a high level of validity compared with the urinary biomarkers⁴⁸.

Considering the inconsistency in the time of the completion of 24-h recalls and the baseline assessment, dietary recalls within 36 months of baseline assessment were utilized to assess the baseline dietary intake level^10,49. Dietary intakes were average for those participants who had more than one 24-h dietary recall within 36 months. The quantity of each beverage or food consumed was computed by multiplying the serving consumed by the proportion size of each beverage or food⁵⁰. Additionally, nutrient and total energy intake were computed based on the calculated quantity.

Nova food classification was applied to classify each beverage and food into one of the four food groups based on the extent and purpose of food processing⁹: 1) unprocessed or minimally processed foods, e.g., milk and fresh vegetables; 2) processed culinary ingredients, e.g., salt and butter; 3) processed foods, e.g., canned fish and cheese; 4) UPFs, e.g., soft drinks and savory snacks. Detailed information about the Nova classification is available elsewhere⁹. Our study primarily focuses on the UPF group, and the specific details as well as several samples are shown in Supplementary Table 10. For each participant, a proportion of UPF was computed by dividing the sum weight of each food item among the UPF group (g/day) by the weight of total food consumption (g/day). Instead of energy ratio, we used weight proportion (%) because it can capture UPFs providing no or low energy (e.g., artificially sweetened beverages)⁵¹.

Metabolomic measurement

Metabolomics of ethylenediaminetetraacetic acid (EDTA) plasma samples were measured using high-throughput nuclear magnetic resonance (NMR) spectroscopy in Nightingale Health’s laboratories among random subsets of the full cohort. Phases 1 and 2 have been completed, and the current number of detected samples reached ~292,000 participants, of which 275,000 are from baseline recruitment. The samples were stored at −80 °C before preparation, were slowly thawed at +4 °C overnight, and then were centrifuged (3400 × g) at +4 °C for 3 min. Aliquots of each sample were transferred into NMR tubes and mixed with a phosphate buffer. The NMR spectra of each sample was recorded using a 500 MHz NMR spectrometer (Bruker AVANCE IIIHD), and the metabolomic biomarkers were quantified with Nightingale Health’s proprietary software. In each well plate, two blind duplicate samples and two control samples were included to monitor the consistency across multiple spectrometers. The coefficients of variation (CVs) were below 5% for most metabolomic biomarkers. Further details of the experimentation and the NMR platform could be found elsewhere⁵².

A total of 251 metabolomic measures including 170 absolute levels and 81 ratio measures, were quantified for each plasma sample, covering lipid concentrations and compositions of 14 lipoprotein subclasses, fatty acids, and various low-molecular-weight metabolites (e.g., ketones, amino acids, and glycolysis metabolites). Technical variation in the metabolomic data (i.e. sample preparation time, the position of samples in shipping plates, temporal drift within the spectrometer, and outlier shipping plates) was removed using the R package named “ukbnmr”⁵³. Specifically, the technical variation was removed by regressing on the time elapsed between sample preparation and sample measurement, plate row, plate column, and plates grouped by date within each of six spectrometers. Furthermore, outlier plates were systematically identified and removed. This strategy to remove technical variation was also applied in previous studies conducted in the UK Biobank^54,55. In the present study, we excluded ratio measures (e.g., Phospholipids to Total Lipids) and the original alanine which could be replaced by spectrometer-corrected alanine, leaving 169 metabolites in absolute levels for the final analysis. Detailed information on these metabolites is provided in Supplementary Table 11. Metabolite values flagged with “below the limit of quantification” for each metabolite were replaced by half of the minimum non-missing values, and then the other missing values were imputed using the random forest imputation since it was recommended and used in metabolomic analysis^56,57.

Outcome ascertainment

The diagnosis of incident diabetic microvascular complications was obtained through two sources of health records including the hospital inpatient records and death registries. The ICD10 code was employed to define diabetic neuropathy (E114, E144, G590, G629, G632, G990), diabetic retinopathy (E113, E143, H280, H360), and diabetic kidney disease (E112, E142, N180, N181, N182, N183, N184, N185, N188, N189)^58,59. Health records were available until 25 September 2021 for Scotland, 1 November 2022 for England, and 29 May 2021 for Wales.

Covariate assessment

Information about age, race, sex, education level, smoking status, leisure-time physical activity, drinking status, and family history of CVD was collected via a touch-screen questionnaire at baseline. Townsend Deprivation Index (TDI), which reflected socioeconomic status, was calculated based on national census data according to postcodes of residence⁶⁰. Leisure-time physical activity was assessed through the long-form version of the International Physical Activity Questionnaire, and weekly metabolic equivalent minutes (MET-min/week) were computed. Body weight and height were measured by trained nurses, and BMI was computed as body weight (kg) divided by the square of height (m²). A healthy diet score was calculated to evaluate the overall diet quality according to a previous UK Biobank study⁶¹, which considered adequate consumption of healthy food (whole grains, fruit, vegetables, seafood, vegetable oils, and dairy) and reduced consumption of unhealthy food (refined grains, sugar-sweetened beverages, processed meats, and unprocessed meats. Detailed scoring method for the healthy diet score can be found in Supplementary Table 12. The medical history of hyperlipidemia, hypertension, cancer, CVD, and duration of T2D was ascertained through verbal interviews, electronic health records, and questionnaires. The mediation for T2D was collected by verbal interviews and questionnaires.

Statistical analysis

The association between UPF intake and each metabolite (log1p-transformed and z score standardized) was assessed using the multivariable linear regression^62,63, with P values < 0.05 after Benjamini–Hochberg FDR correction considered as statistically significant. To identify the metabolomic signature of UPF intake, all the participants were randomly assigned to a training set or testing set in a ratio of 7:3 (Fig. 3). In the training set, we applied an elastic net model, which is a regularized regression model combining the Lasso and Ridge penalties, to derive a metabolomic signature reflecting UPF intake based on metabolites passing the threshold after FDR correction and to avoid collinearity among metabolites⁶⁴. Two tuning parameters for the elastic net model were determined, including α (balancing penalties of Lasso and Ridge) and λ (the penalty intensity parameter). To select core metabolites and achieve sparsity, α was chosen from 0.50, 0.75, and 1.00⁶². We then used a leave-one-out cross-validation framework to select optimal λ to achieve the minimum mean squared error¹⁵. In the final model, α = 0.50 and λ = 0.12298 were selected. The metabolomic signature was computed as the weighted sum of the chosen metabolites, with weights being the coefficients from the elastic net regression. In the testing set, the weights derived from the training set were applied to compute a metabolomic signature. Spearman correlation coefficients were calculated to assess the correlation between UPF intake and the metabolomic signature in the training and testing sets.

UPF intake and the metabolomic signature were converted to z scores (mean = 0, SD = 1) to ensure comparability. Cox proportional hazard regression models were applied to estimate the HRs and 95% CIs for associations of UPF intake and the metabolomic signature with risks of total and individual microvascular complications utilizing the combined data from the testing and training set (Fig. 3). Follow-up time was computed from the recruitment date until death, the diagnosis of microvascular complications, or the end of follow-up, whichever came first. We assessed the proportional hazards assumption by the product term of UPF intake or the metabolomic signature with follow-up time and found no significant violation.

Two models were built in this study. In Model 1, we adjusted for age (continuous, y), total energy intake (continuous, kcal/d), and sex (female, male). In Model 2, we additionally adjusted for race (White participants, Asian or Asian British participants, Mixed participants, and Black or Black British participants), smoking status (never, past, current), drinking status (never, past, current light to moderate [1–28 g ethanol per day for males or 1–14 g ethanol per day for females], current heavy [>28 ethanol per day for males or >14 g ethanol per day for females]), education level (college/university degree, other degrees), TDI (continuous), leisure-time physical activity (continuous, MET-min/wk), history of hyperlipidemia, hypertension, and cancer (yes, no), duration of T2D (continuous, y), family history of CVD (yes, no), use of glucose-lowering medications (none, only oral medicine, insulin, and others), and the number of 24-h dietary recalls (continuous). To maximize data availability, we imputed missing covariates by generating 10 imputed datasets using multiple imputations. To test whether associations between the metabolomic signature and risks of diabetic microvascular complications were attributable to its correlation with UPF intake, we further included both UPF intake and the metabolomic signature in Model 2. To assess the predictive performance of the metabolomic signature for composite microvascular complications, we calculated the C statistic, continuous NRI, and absolute IDI⁶⁵. The basic model was based on all covariates in Model 2. Other models were successively added with UPF intake, UPF-related traditional biomarkers, and the metabolomic signature. The traditional biomarkers included renal function (urate and urea), lipid profile (total cholesterol [TC], LDL [low-density lipoprotein] cholesterol, HDL cholesterol, triglycerides, apolipoprotein B, and apolipoprotein A), inflammation (white blood cell count and C-reactive protein), liver function (alkaline phosphatase [ALP], alanine aminotransferase [ALT], aspartate aminotransferase [AST], total bilirubin, total protein, and gamma glutamyltransferase [GGT]), blood pressure (systolic and diastolic blood pressure), and glycated hemoglobin A1c (HbA1c)^10,59. When UPF intake was significantly associated with a biomarker in the multivariable-adjusted linear regression model, it was selected as a UPF intake-related traditional biomarker (Supplementary Table 13).

Among the chosen metabolites in the metabolomic signature, potential mediating metabolites of associations between UPF intake and risks of diabetic microvascular complications were identified based on the mediation principle that mediators are associated with both the exposure and the outcome⁶⁶. These criteria were evaluated in the multivariable linear regression models (Fig. 1 and Supplementary Table 1) and multivariable Cox regression models (P values < 0.05 after FDR adjustment setting as significant; Supplementary Table 14). The %MEDIATE macro in SAS was used to compute the proportion of the diet-disease association that could be mediated by the metabolomic signature and individual metabolites.

Stratified analyses were conducted according to sex (female, male), age (≤60, >60 y), BMI (<30, ≥30 kg/m²), and the number of dietary recalls (1, ≥2). The interactions of stratified factors with UPF intake or the metabolomic signature on the risk of outcome were tested by the Wald test by adding product terms in Model 2. Several sensitivity analyses were conducted to test the robustness of these findings. First, we excluded participants who had diabetic microvascular complications within 3 years of follow-up. Second, considering BMI is an important mediator linking UPF intake and diabetic microvascular complications¹⁰, we did not adjust for it in the main analysis but further adjusted for it as a sensitivity analysis to verify the identified associations. Third, we further adjusted for the healthy diet score. Fourth, we investigated the association of UPF intake and the metabolomic signature with the risk of diabetic kidney disease after additionally adjusting for baseline eGFR. Fifth, we removed from the multivariable model glucose-lowering medications, which might be a confounder or mediator. Sixth, to test whether the pre-screening process impacts the results, we applied an elastic net model to all 169 metabolites instead of only those metabolites significantly associated with UPF intake. Seventh, to further validate the constructed metabolomic signature, we applied it to individuals with baseline metabolomic data but without valid baseline dietary data and assessed its associations with risks of diabetic microvascular complications. Eighth, we used the 80:20 split (80% of the data for training and 20% for testing) to derive a metabolomic signature⁶⁷. Finally, we repeated the analysis with UPF intake measured by the proportion of energy.

All analyses were performed using SAS version 9.4 (SAS Institute) and R software (version 4.4.3). Two-sided P < 0.05 unless specified using FDR correction was considered to be statistically significant.

Author contributions

Y.Li., Y.F.L., T.T.G., and A.P. conceived the study design. Y.Li. and D.X. conducted analyses. Y.Li. and X.Y.S. wrote the first draft of the paper. All authors contributed to the interpretation of the results and critical revision of the manuscript for important intellectual content. All authors approved the final version of the manuscript. Y.F.L., T.T.G., and A.P. are the guarantors of this work and, as such, have full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Data availability

The UK Biobank data are on application to the UK Biobank (https://www.ukbiobank.ac.uk/). This research was conducted using the UK Biobank Resource under Application Number 109546.

Code availability

The analytic code used in this study will be made available upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41538-025-00597-3.