Introduction

The portfolio diet, a cholesterol-lowering plant-based diet, is recommended by several clinical practice guidelines globally as a therapeutic diet to lower cholesterol and overall cardiovascular disease (CVD) risk [1,2,3,4,5,6]. The dietary pattern includes foods known to lower low-density lipoprotein cholesterol (LDL-C), including plant protein (soy, chickpeas, beans, lentils), nuts and seeds, viscous fiber sources (psyllium, oats, barley, eggplant, okra, berries, citrus fruit, etc.), and phytosterols (as supplements or fortified foods). A systematic review and meta-analysis of six randomized controlled trials (RCTs) showed that the portfolio diet significantly decreased LDL-C, non-high-density lipoprotein cholesterol (non-HDL-C), apolipoprotein B (apoB), total cholesterol, triglycerides, C-reactive protein (CRP), and systolic and diastolic blood pressure [7]. Higher adherence to the portfolio diet has also been associated with lower cardiometabolic risk factors and lower risk of CVD outcomes, type 2 diabetes, and total mortality in several prospective cohort studies [8,9,10,11,12].

One major challenge in dietary advice trials and epidemiological study designs is measuring adherence to dietary patterns of interest [13], which typically relies on self-reported data. Newly developed methods that show promise for identifying more objective biomarkers of food intake include metabolomics [14]. This approach uses the products of metabolism to develop a metabolite profile of foods and dietary patterns, which are not subject to the same measurement and reporting errors of traditional self-report dietary instruments [15, 16]. Controlled-feeding studies provide an exceptional opportunity to help discover objective markers of foods and dietary patterns as researchers design the menus, provide foods to the participants, and closely monitor intake [17]. Metabolomics analyses can also provide insight into biological mechanisms linking diet to cardiometabolic and other disease risk, and holds promise for opening avenues for precision nutrition approaches in the future by identifying individuals whose health and function may benefit the most from specific dietary patterns [18]. Metabolomics profiling has been conducted previously for dietary patterns in RCTs such as the Mediterranean diet [19], Dietary Approaches to Stop Hypertension (DASH) diet [20], a protein-rich dietary intervention trial (OmniHeart) [21], low glycemic index diet [22], and Nordic diet [23], all of which have highlighted significant changes that occur in the metabolome with these diets, however, no studies have explored the plasma metabolomic profile of the cholesterol-lowering plant-based portfolio diet.

The aim of the present study was therefore to conduct an exploratory metabolomics analysis of plasma samples from two randomized controlled feeding trials [24, 25] of the portfolio diet to identify metabolites that differed within and between the portfolio diet and the other interventions (which included a vegetarian National Cholesterol Education Program (NCEP) Step II diet [the control diet] and a statin intervention) that may be used as candidate biomarkers of dietary intake in future studies.

Methods

Study design

This study was a secondary analysis of two randomized controlled feeding trials, one cross-over and one parallel; the respective study designs and main results have been described previously [24, 25]. Both trials primary aim was to examine the effect of the portfolio diet on LDL-C. Briefly, study one was conducted between 2002-2003, and examined the effect of the portfolio diet compared to a control diet (NCEP Step II) and the control diet plus a statin (20 mg lovastatin) for 4 weeks each in a cross-over design (Fig. 1) in 34 participants, with a one-month washout period between interventions [24]. Study two was conducted in 2002 and examined the effect of the portfolio diet compared to the same control diet for 4 weeks in a parallel design (Fig. 1), in 25 participants [25]. We analyzed 381 citrate plasma samples at week 0, week 2, and week 4 (Fig. 1), that had been stored at -80 °C since collection and not previously thawed, to identify metabolites that differed within and between interventions. Both trials were approved by the research ethics board of the University of Toronto and St. Michael’s Hospital and all methods were performed in accordance with relevant guidelines and regulations. Informed consent was obtained from the participants.

Fig. 1: Study design of the two trials.
figure 1

a Study one, cross-over (Jenkins et al., JAMA, 2003, b Study two, parallel (Jenkins et al., Metabolism, 2003). Blood drops represent when plasma samples were collected.

Full size image

Participants

The eligibility criteria were identical for both trials, including men and postmenopausal women with high LDL-C (>4.1 mmol/L), not taking lipid-lowering medications or discontinued their use before study participation, and were free of chronic diseases such as CVD, diabetes, kidney disease, and liver disease. Table 1 includes baseline participant characteristics for both trials. The first RCT included 34 adults aged 58.4 ± 8.6 years, 82% of which were males, with a mean BMI of 27.3 ± 3.4 kg/m2, and LDL-C of 4.4 ± 0.8 mmol/L (Table 1). The second RCT included 25 adults aged 61.0 ± 9.9 years, 64% of which were males, with a mean BMI of 26.8 ± 2.9 kg/m2 and LDL-C of 4.5 ± 0.8 mmol/L (Table 1).

Table 1 Baseline characteristics of the two trials.
Full size table

Dietary interventions

The portfolio and control diet interventions were similar in both trials and were weight-maintaining diets. Participants were provided with study foods at weekly clinic visits, except for fruit and most non-starchy vegetables (okra and eggplant were provided in the portfolio groups). Participants purchased specific fruits and vegetables from their local grocery stores and were reimbursed thereafter. The aim of the portfolio diet was to provide less than 7% of total energy from saturated fat, and less than 200 mg of dietary cholesterol per day, as well as (on a 2000kcal basis) 42 g of legume protein (mainly soy), 24 g of almonds, 20 g of viscous fiber (from oats, barley, psyllium, okra, eggplant, berries, apples, citrus fruit, etc.), and 2 g of phytosterols from fortified margarine. Eggs (1 per week) and butter (9 g/day) were also provided on the portfolio diet to help match the saturated fat and dietary cholesterol content to the control diet.

The control diet was a vegetarian NCEP Step II diet (also less than 7% of total energy from saturated fat and 200 mg of dietary cholesterol per day) that lacked portfolio diet foods. The diet included low fat dairy and egg whites as main protein sources to maintain a low saturated fat intake. The high fiber intake was achieved using whole grain breakfast cereals, bread, and wheat bran. High monounsaturated sunflower oil (9 g/1000 kcal) and safflower oil (5 g/1000 kcal) were incorporated into the control diet to help match the fatty acid profile to the portfolio diet. Adherence to the dietary interventions, measured by food checklists and food diaries, ranged from 86-94% in the trials. Supplemental Table 1 includes sample menu plans for the portfolio and control diets. Table 2 includes the dietary intake among participants at week 4 in the two trials including the clinical portfolio diet score (c-PDS) [26] and individual dietary components and Supplemental Table 2 includes the dietary intake of the baseline diets in both trials.

Table 2 Mean dietary intake of the interventions in the two trials at week 4.
Full size table

Metabolomics analysis

Metabolomics profiling was performed on the 381 plasma samples (collected after fasting for >8 hours) at the Broad Institute of Harvard University and MIT, using high-throughput liquid chromatography–tandem mass spectrometry (HILIC-pos and C8-pos), as previously described [27]. Briefly, the hydrophilic interaction liquid chromatography (HILIC) analyses of water-soluble metabolites in the positive ionization mode (HILIC-pos) were conducted using an LC-MS system composed of a Shimadzu Nexera X2 U-HPLC (Shimadzu Corp) coupled to a Q Exactive hybrid quadrupole orbitrap mass spectrometer (Thermo Fisher 14 Scientific). Positive ion mode analyses (C8-pos), which measure polar and nonpolar plasma lipids, were conducted by using an LC-MS system composed of a Shimadzu Nexera X2 U20 HPLC (Shimadzu Corp) coupled to an Exactive Plus orbitrap mass spectrometer (Thermo Fisher Scientific) [27]. Samples were analyzed in random order and replicates were interspersed of common pooled plasma samples as controls to allow normalization. Quality control procedures were performed throughout the assays to ensure profiling quality and system performance [27].

A total of 504 metabolites of known identity were included in the analysis. Metabolites with missingness for >25% of the population were excluded (n = 8), for a total of 496 known metabolites. The 8 excluded metabolites included drug metabolites (metformin, atenolol, venlafaxine, quetiapine), hordenine, p-acetamidophenylglucuronide, solanidine, and (E,E,)-trichostachine. The 496 metabolites consisted of glycerophosphocholines (18%), triacylglycerols (16%), amino acids, amines, and peptides (15%), glycerophosphoethanolamines (9%), carnitines (7%), sphingomyelins (5%), sterol esters (3%), purines (3%), ceramides (3%), diracylglycerols (2%), and other metabolites. Metabolites with skewness above or below 2 were log-transformed [28], and missing metabolites were imputed using the random forest imputation method [29]. Metabolites were standardized to z-scores with a mean of 0 and standard deviation of 1.

Lipid and C-reactive protein analyses

In both trials, fasting serum was analyzed according to the Lipid Research Clinics protocol [30] for total cholesterol, triglycerides, and high-density lipoprotein cholesterol (HDL-C) after dextran sulphate–magnesium chloride precipitation [31]. LDL-C and non-HDL-C were calculated; LDL-C using the Friedewald equation [32]. Serum apoB was measured by nephelometry [33]. CRP was measured by endpoint nephelometry (Behring BN-100, N high-sensitivity C-reactive protein reagent, DadeBehring, Mississauga, Ontario).

Statistical analyses

Analyses were conducted separately in each trial. To evaluate the effect of the dietary interventions over all three time points (week 0, week, 2, and week 4 as in Fig. 1), we used linear mixed models to investigate which metabolites differed within and between each intervention to reflect the full intervention period in both trials. The models (per 1-SD increment in each metabolite) incorporated participant-specific intercepts as a random effect and the fixed effects were diet, time (week), and the interaction between the three groups term (diet x week). An additional covariate for study phase was included in the cross-over study (study one). For significant metabolites, we characterized the proportion in each biochemical class using the Broad Institute assigned “main class” and “sub class”. Significant metabolites that consistently changed with the portfolio diet intervention in both studies were identified from the linear mixed models. Partial-least squares discriminant analysis (PLS-DA) was used to identify the top 15 plasma metabolites which were influential in discriminating between the end of the portfolio diet intervention from baseline. This method was applied to identify the most statistically significant metabolites, reducing the list to those most likely to serve as potential biomarkers of the portfolio diet. The top 15 metabolites were selected using those with the highest variable importance in projection (VIP) scores. As a sensitivity analyses, we adjusted the linear mixed models of the portfolio diet intervention for LDL-C to further examine metabolites that may be related to the dietary intervention/foods consumed by controlling for the LDL-C lowering effect of the diet.

Lastly, we also examined the association between changes in blood lipids (LDL-C, apoB, non-HDL-C) and CRP over the three time points with change in metabolites using linear mixed models. The models included participant-specific intercepts as a random effect and fixed effects included week and intervention group. An additional covariate for study phase was included in the cross-over study (study one). Analyses were performed using R version 4.2.0. The R package “missRanger” was used for imputation of metabolites and “lme4” for linear mixed models. Statistical tests were two-sided and a Bonferroni-corrected p-value of <0.0001 (0.05 ÷ 496 metabolites) was considered significant.

Results

The portfolio diet nutritional profile was higher in plant protein, total fiber, and total fat intake (mainly due to increased monounsaturated fat) at week 4 compared to the control diets (Table 2). The cardiometabolic effects of the trials have been reported previously [24, 25]. Briefly, in study one, LDL-C significantly decreased by 29%, apoB by 23%, and CRP by 53% with the portfolio diet intervention. LDL-C decreased by 8% and 32%, apoB by 6% and 27%, and CRP by 21% and 40% in the control diet and statin plus control diet interventions, respectively [24]. In study two, LDL-C was reduced by 35% and 12% and apoB by 27% and 8% with the portfolio diet and control diet interventions, respectively [25]. CRP did not significantly decrease in study two. Body weight, HDL-C, triglycerides, and blood pressure did not significantly change in either study.

Study one

There were no interactions with time (week), and the phase variable was not significant for all interventions. After Bonferroni adjustment, 145 metabolites significantly changed after the portfolio diet intervention (Fig. 2 and Supplemental Table 3). Most of these metabolites were classified as glycerophosphocholines (32%, n = 47), triacylglycerols (20%, n = 29), glycerophosphoethanolamines (14%, n = 20), sphingomyelins (8%, n = 12), amino acids and peptides (8%, n = 11), among others (Fig. 2). Subclasses were mainly triacylglycerols (20%, n = 29), phosphatidylcholine (16%, n = 23), plasmalogen phosphatidylcholine (13%, n = 19), plasmalogen phosphatidylethanolamine (9%, n = 13), amino acids (7%, n = 10), cholesteryl esters (4%, n = 6), and acyl carnitines (3%, n = 5) (Supplemental Table 3). Most metabolites decreased after the portfolio diet intervention (n = 127), compared to increasing (n = 18) and the 3 metabolites that increased the most on the portfolio diet were N2-acetylornithine, O-ethylhomoserine, and L-pipecolic acid, and the 3 that decreased the most were CE (18:0), PC P-40:6 or PC O-40:7, and PC (40:6) (Fig. 2 and Supplemental Table 3).

Fig. 2: Significant metabolites that changed with the portfolio diet intervention from baseline in study one.
figure 2

a Metabolite classes of the 145 metabolites significantly changed after the portfolio diet intervention from a total of 496 metabolites assessed. b Volcano plot of mean difference in plasma metabolites over 4 weeks after consumption of the portfolio diet from baseline. Significant metabolites were estimated from 34 participants via linear mixed models and included a Bonferroni adjustment for multiple testing (p < 0.0001). The models (per 1-SD increment in each metabolite) incorporated participant-specific intercepts as a random effect and the fixed effects were diet, time (week), and the interaction between time and diet (diet × week), adjusted for phase. See Supplemental Table 3 for full list of the significant metabolites.

Full size image

For the control diet, 94 metabolites significantly changed after the control diet intervention, after Bonferroni adjustment (Supplemental Fig. 1 and Supplemental Table 4). Most of these metabolites were glycerophosphocholines (35%, n = 33), glycerophosphoethanolamines (19%, n = 18), triacylglycerols (12%, n = 11), sphingomyelins (10%, n = 9), among others (Supplemental Fig. 1). The 3 metabolites that increased the most on the control diet were TG(16:0_38:6), TG(16:0_38:5), and O-ethylhomoserine, and the 3 that decreased the most were methylimidazoleacetic acid, 1-methylhistidine, and PC(18:4/P-18:1) (Supplemental Table 4). Subclasses for the control diet are included in Supplemental Table 4.

After Bonferroni adjustment, 159 metabolites significantly changed after the statin and control diet intervention (Supplemental Fig. 2 and Supplemental Table 5). Most of these metabolites were also glycerophosphocholines (29%, n = 46), triacylglycerols (20%, n = 32), glycerophosphoethanolamines (15%, n = 24), sphingomyelins (8%, n = 13), among others (Supplemental Fig. 2). The 3 metabolites that increased the most with the statin intervention were PE P-44:12 or PE O-44:13, tryptophan, and TG(16:0_38:6), and the 3 that decreased the most were PC(18:4/P-18:1), PC(P-18:1/20:4), and PE(P-18:0/22:6) (Supplemental Table 5). Subclasses for the statin intervention are included in Supplemental Table 5.

Figure 3 includes the 16 metabolites that significantly differed between the portfolio diet and the other two interventions (control and statin groups) at the end of the study. Across both comparisons, N2-acetylornithine, L-pipecolic acid, trimethylamine N-oxide (TMAO), 3-hydroxynorvaline, ectoine, and lenticin were higher, and sarcosine lower, in participants following the portfolio diet intervention (Fig. 3). 7a-hydroxy-cholestene-3-one and TG(15:0/16:0/20:3) were also higher, and PC(18:0/20:4), tigylcarnitine, tyrosine, allantoin, tryptophan, and methionine were also lower in participants following the portfolio diet intervention compared to the statin intervention (Fig. 3). Only 4 metabolites differed between the control diet and statin plus control diet intervention, which included higher creatinine and tryptophan, and lower PC(16:0/18:2) and campesterol, in the statin group.

Fig. 3: Metabolites that significantly differed between the portfolio diet, the control diet and statin plus control diet interventions at the end of study one.
figure 3

Forest plot of mean difference in plasma metabolites over 4 weeks after consumption of the portfolio diet compared to (a) the control diet, and (b) the statin plus control diet. Significant metabolites were estimated from 34 participants via linear mixed models and included a Bonferroni adjustment for multiple testing (p < 0.0001). The models (per 1-SD increment in each metabolite) incorporated participant-specific intercepts as a random effect and the fixed effects were diet, time (week), and the interaction between time and diet (diet × week), adjusting for phase.

Full size image

Study two

There were no interactions with time (week). After Bonferroni adjustment, 63 metabolites were significantly changed after the portfolio diet intervention (Fig. 4 and Supplemental Table 6). These metabolites included glycerophosphocholines (48%, n = 30), glycerophosphoethanolamines (17%, n = 11), amino acids and peptides (8%, n = 5), among others (Fig. 4). Subclasses were mainly plasmalogen phosphatidylcholine (23% n = 15), plasmalogen phosphatidylethanolamine (14%, n = 9), phosphatidylcholine (13%, n = 8), cholesteryl esters (6%, n = 4), and acyl carnitines (6%, n = 4). Similar to study one, most metabolites decreased (n = 54) after the portfolio diet intervention (Fig. 4 and Supplemental Table 6). The 3 metabolites that increased the most with the portfolio diet were PE(15:0/22:4), cystine, and O-ethylhomoserine and the 3 that decreased the most were CE(18:0), PC(18:4/P-18:1), and PE P-39:6 or PE O-39:7 (Supplemental Table 6).

Fig. 4: Significant metabolites that changed with the portfolio diet intervention from baseline in study two.
figure 4

a Metabolite classes of the 63 metabolites significantly changed after the portfolio diet intervention from a total of 496 metabolites assessed. b Volcano plot of mean difference in plasma metabolites over 4 weeks after consumption of the portfolio diet from baseline. Significant metabolites were estimated from 25 participants via linear mixed models and included a Bonferroni adjustment for multiple testing (p < 0.0001). The models (per 1-SD increment in each metabolite) incorporated participant-specific intercepts as a random effect and the fixed effects were diet, time (week), and the interaction between time and diet (diet x week). See Supplemental Table 6 for full list of significant metabolites.

Full size image

For the control diet, only 11 metabolites significantly changed after Bonferroni adjustment (Supplemental Fig. 3 and Supplemental Table 7), most of which included glycerophosphocholines and triacylgycerols, with subclasses included in Supplemental Table 7. The 3 metabolites that increased the most with the control diet were cystine, TG(16:0_38:6) and TG(16:0_38:5) and the 3 that decreased the most were CE(20:5), PC(22:6/18:3), and PC(P-18:1/20:4) (Supplemental Table 7). Importantly, all participants in study two were encouraged to follow an NCEP Step II diet prior to randomization, therefore, less metabolite changes would be expected in this study. Only 3 metabolites differed between the portfolio diet and control diet in study two at the end of the study after Bonferroni adjustment. These metabolites included higher ectoine and lower allantoin and SM 18;1:O2/21:0 in the portfolio diet intervention.

Consistent and top metabolites changed with the portfolio diet intervention

Table 3 shows the 52 metabolites that consistently changed with the portfolio diet intervention in both trials after Bonferroni adjustment, all of which were altered in the same direction. The majority of these metabolites were glycerophosphocholines (n = 26), followed by glycerophosphoethanolamines (n = 10), sterol esters (n = 4), sphingomyelins (n = 3), carnitines (n = 3), amino acids and peptides (n = 2), sterols (n = 2), glycosphingolipids (n = 1), and imidazoles (n = 1). Figure 5 shows the top 15 metabolites discriminating the portfolio diet from baseline in both trials by their corresponding VIP score. For study one, these included increases in N2-acetylornithine, O-ethylhomoserine, 3-hydroxynorvaline, and lenticin, and decreases in PE(P-18:1/18:3), PC(P-16:0/18:4), CE(18:0), PE(P-18:0/20:5), PE(P-18:0/22:6), PC(40:6), PE P-39:6 or PE 0-39:7, PE(P-18:0/20:4), methylimidazoleacetic acid, and PC(15:0/18:0). In study two, these included decreases in PC(P-16:0/14:0), PC(P-18:1/20:4), 1-methylhistidine, PE(P-18:0/22:6), PC(P-16:0/22:6), CE(14:0), PE P-39:6 or PE 39:7, LysoPC(17:0/0:0), PC(40:6), PC(18:0/18:1), CE(18:0), PE(P-18:1/20:5), PS P-36:1 or PS O-36:2, LysoPE(22:0/0:0) and PC(38:6). In the sensitivity analyses where we further adjusted the portfolio diet intervention findings for LDL-C, many metabolites were no longer significant in both trials (full list of significant metabolites are included in Supplemental Tables 8 and 9). After Bonferroni correction, only changes in PC(P-18:1/22:5), PC(P-16:0/18:4), N2-acetylornithine, PC(40:6), PC(22:6/18:3), and PC(38:6) were significant in study one (Supplemental Table 8) and PC(22:6/18:3), O-ethylhomoserine, and PC(P-18:1/20:4) in study two (Supplemental Table 9).

Table 3 Metabolites that significantly differed within both portfolio diet intervention studies from baseline (n = 52).
Full size table
Fig. 5: Top 15 metabolites discriminating the portfolio diet from baseline in both intervention studies.
figure 5

a Study one, b study two. Metabolites represent those with the highest variable important in projection (VIP) score from baseline to week 4 with the portfolio diet intervention. VIP scores were calculated using partial least squares-discriminant analysis (PLS-DA).

Full size image

Blood lipids and inflammation

Supplemental Tables 10–16 show the metabolites significantly associated with changes in LDL-C, non-HDL-C, apoB, and CRP in the two trials, adjusting for intervention group, after Bonferroni correction. CRP findings from study two are not shown as no metabolites were significantly associated with CRP levels. Similar changes in metabolites were associated with LDL-C, non-HDL-C, and apoB within and between studies, ranging from 60 to 215 significant metabolites. The majority of these metabolites were lipids, including glycerophosphocholines, glycerophosphoethanolamines, triacylglycerols, and sterol esters, and for those that were also associated with the portfolio diet intervention, these metabolites were generally associated with the blood lipids in the opposite direction. Only 5 metabolites (xanthopterin, vitamin A, guanidoacetic acid, 1-methylnicotinamide, and LysoPC(18:2/0:0)) were significantly associated with changes in CRP in study one (Supplemental Table 13).

Discussion

In two randomized controlled feeding trials, we observed that following a plant-based cholesterol-lowering portfolio diet for 4 weeks had a significant impact on the plasma metabolome in participants with hypercholesterolemia. Many of the altered metabolites were lipids (i.e., phosphatidylcholines, plasmalogen phosphatidylcholines, plasmalogen

phosphatidylethanolamines, triacylglycerols), amino acids, cholesteryl esters, and acyl carnitines. Across both trials, 52 metabolites consistently changed in the same direction with the portfolio diet intervention, indicating a measurable metabolomic response to this dietary pattern. These changes are likely influenced by the diet’s lower saturated fat content, reduction in animal foods, high intake of plant foods and plant protein, and its LDL-C lowering effect. Metabolomics analyses may therefore be important in linking plasma metabolites to specific foods and dietary components.

Among lipids, several species containing stearic acid (C18:0) were reduced, likely due to the diet’s low saturated fat content (<7% of energy) and reduction in meat [34]. These lipid reductions were also observed in the control diet interventions, which were similarly low in saturated fat and also vegetarian diets. In addition, C18:1 species (oleic acid) also decreased, possibly reflecting the diet’s fat composition. Furthermore, stearic acid is known to lead to higher oleic acid, thus the observed decreases in C18:0 species may coincide with the decline in C18:1 fatty acids [35]. However, the precise relationship between these lipid changes and the fatty acid profile of the portfolio diet requires further investigation as even-chain saturated fatty acids can also be synthesized endogenously [36, 37]. The reduction in cholesterol metabolites observed aligns with the LDL-C-lowering effects of the portfolio diet. In addition, the opposing directions of association between these metabolites for the portfolio diet intervention and LDL-C further support their role in cholesterol reduction. Furthermore, we observed a decrease in metabolites associated with animal foods, including 1-methylhistidine, various acylcarnitines, and C22:6 and C20:5 lipids (n-3 DHA and EPA). These reductions likely reflect the plant-based nature of the portfolio diet, which excludes red meat, poultry, and fish [38,39,40].

Several metabolites linked to plant food consumption [41], including N2-acetylornithine, L-pipecolic acid, lenticin, and ectoine, increased in response to following the portfolio diet. N2-acetylornithine is often associated with legume consumption or polyphenol-rich foods like fruits and vegetables [40, 42]. Similarly, L-pipecolic acid may indicate dry bean intake, while lenticin, also known as hypaphorine or tryptophan betaine, is a known biomarker for legume consumption, particularly lentils and chickpeas [40]. Ectoine, a metabolite produced by certain bacterial genera [43], may reflect interactions between the gut microbiome and the portfolio diet. Ectoine has also been associated with olive oil consumption in the PREDIMED study [44]. Two metabolites, O-ethylhomoserine and 3-hydroxynorvaline, also increased with the portfolio diet (in study one after Bonferroni correction and study two before Bonferroni correction). However, these metabolites have limited information in the literature, and therefore, may represent possible novel metabolites related to the portfolio diet that require further investigation.

We observed that more metabolites significantly changed with the portfolio diet intervention in study one compared to study two, after applying a correction for multiple testing. However, study one had a larger sample size (34 participants vs. 25 participants), was a cross-over design compared to parallel, and in study two, the participants were advised to consume a NCEP Step II diet prior to randomization, therefore, the larger metabolite changes in study one would be expected. In addition, when comparing the portfolio diet with other interventions, fewer metabolite changes remained significant after correcting for multiple testing, potentially due to the vegetarian nature of the control diet, resulting in less opportunity for changes compared to the baseline diet. However, these metabolites likely represent those unique to the foods provided in the portfolio diet. The control diet in both studies was a low saturated fat vegetarian diet (without the portfolio diet foods), and the statin diet intervention resulted in similar LDL-C reductions as the portfolio diet. These comparisons allow us to control for the low saturated fat profile and LDL-C lowering effect of the portfolio diet. Many of the lipid changes were no longer significant, and the top significant metabolites included higher N2-acetylornithine, L-pipecolic acid, TMAO, 3-hydroxynorvaline, ectoine, and lenticin, and lower sarcosine, with the portfolio diet intervention. 7a-hydroxy-cholestene-3-one and TG(15:0/16:0/20:3) were also higher, and PC(18:0/20:4), tigylcarnitine, tyrosine, allantoin, tryptophan, and methionine were lower in the portfolio diet intervention compared to the statin intervention. Additionally, we adjusted the portfolio diet findings for subsequent changes in LDL-C levels. Many metabolites were no longer significant after Bonferroni correction, however, PC(P-18:1/22:5), PC(P-16:0/18:4), N2-acetylornithine, PC(40:6), PC(22:6/18:3), and PC(38:6) remained significant in study one and PC(22:6/18:3), O-ethylhomoserine, and PC(P-18:1/20:4) in study two. As N2-acetylornithine remained significant in several different analyses, these findings highlight this metabolite as an important biomarker for a plant-based diet high in legumes, nuts, viscous fiber, fruits, vegetables, and phytosterols, and based on previous studies, may be particularly related to the high legume intake in the portfolio diet [40, 42].

Several unexpected metabolites were identified. Notably, TMAO levels were elevated in the portfolio diet intervention compared to both the control diet and the statin intervention in study one. However, the increase in TMAO was not statistically significant when comparing the end of the portfolio diet intervention at 4 weeks to baseline levels. This elevated TMAO compared to a vegetarian control diet, which did not include the recommended foods of the portfolio diet, raises questions about the role of gut microbial metabolism in plant-based diets. Furthermore, any increase in the CVD risk must be weighed against the reduction in LDL-C consistently observed with the portfolio diet. Elevated TMAO levels have previously been associated with higher intakes of red meat and seafood [45]—both of which were absent from the portfolio diet, and have divergent associations with cardiometabolic disease risk [46,47,48]. However, previous research has also demonstrated that TMAO levels increased following the consumption of whole grains [49] and healthier dietary patterns compared to those high in ultra-processed foods [50]. Therefore, it is plausible that the portfolio diet’s high viscous fiber intake and choline intake, a dietary precursor to TMAO [51], from soy foods and the one egg (including the yolk) per week contributed to the observed rise in TMAO. Future research should explore whether these TMAO findings are clinically meaningful and such studies would benefit from the inclusion of plasma and stool samples to better understand the mechanisms involved. Comparing the portfolio diet to other dietary patterns that include meat and seafood, would help clarify whether these TMAO increases are relevant compared to other dietary patterns with foods that are known to more greatly impact TMAO levels.

We also observed that a pentadecanoic acid (C15:0) species increased with the portfolio diet intervention in both studies, which has been proposed as a biomarker of dairy intake [52], or may be synthesized endogenously from elongation of short-chain fatty acids in response to fiber intake [53, 54]. In these studies, 9 g/day of butter was provided in the portfolio diet menus and therefore may represent a marker of compliance to dairy or may be related to the high fiber intake in the portfolio diet interventions (>75 g/day). Lastly, campesterol levels decreased with the portfolio diet intervention, which was unexpected given 2 g/day of phytosterols were provided in both studies. Also given the high fiber content of the portfolio diet, plasma sterol absorption may have been reduced. These findings represent complex interactions in sterol metabolism that warrant further investigation.

While no prior studies have examined metabolomic changes in response to the portfolio diet, the metabolomic shifts we observed resemble some changes observed in other dietary patterns aimed at reducing cardiometabolic disease risk, such as the DASH and Mediterranean diets. For example, lenticin was also identified as a key metabolite in both the DASH diet and protein-rich diet (OmniHeart) trials [20, 21], where legume intake is also encouraged. Several lipid species, including cholesteryl esters like CE(18:0), were also similarly affected by the Mediterranean diet [19], underscoring the potential overlap in metabolic responses to heart-healthy dietary patterns. However, the top metabolites affected by the portfolio diet included fewer lipid species increasing (most decreased) than those observed with the Mediterranean diet [19], likely owing to the portfolio diet’s lower fat content, addition of butter to the diet to match the saturated fat content of the control diet, and limited olive oil consumption. More recent portfolio diet RCTs have included a fifth component to the diet, high monounsaturated fat oils such as olive oil, canola oil, and avocado [55,56,57]. Therefore, metabolomics analysis of these RCTs may result in more lipid species increasing. This overlap does, however, highlight the non-specificity of some metabolites given the differences in foods recommended as part of these dietary patterns, and the need to continue to measure diet using traditional methods to fully capture the breadth of dietary intake and specific foods consumed. The portfolio diet differs from other heart-healthy dietary patterns, such as the Mediterranean and DASH diets, by emphasizing viscous fiber sources, soy as a plant protein source, and phytosterol-fortified foods. While it is challenging to pinpoint specific metabolites linked to these unique dietary components within the broader dietary pattern, novel metabolites like O-ethylhomoserine and 3-hydroxynorvaline may be associated with these distinctive features. Additionally, compounds such as campesterol and TMAO could also reflect these unique dietary aspects. Furthermore, the portfolio diet may lead to greater reductions in LDL-C compared to other dietary patterns [58], and many of the affected metabolites, including cholesteryl esters and cholesterol itself, likely represent the diet’s LDL-C-lowering effect.

Our study findings do, however, address a pressing need in the field of nutrition for the discovery of objective biomarkers of dietary intake that can be used in combination with more traditional measures of diet (such as diet recalls and food frequency questionnaires), and our study highlights important metabolomic changes related to a plant-based diet rich in legumes, nuts, viscous fiber, and phytosterols. The portfolio diet has previously been shown to lower LDL-C (and non-HDL-C and apoB), and several of the metabolomic changes reported in this study are likely reflective of this cholesterol lowering effect. It is well established that these atherogenic blood lipids are causal in the development of atherosclerotic CVD [59]. However, the metabolomic changes that occurred with the portfolio diet highlight other potential biological mechanisms in which the diet may impact overall cardiometabolic disease risk. Several C18:0 lipid species decreased in response to the portfolio diet. Stearic acid (C18:0) has previously been linked to type 2 diabetes incidence [60] and was the saturated fatty acid most strongly positively related to coronary heart disease (CHD) in the EPIC-InterAct cohort studies [61]. However, a recent systematic review and meta-analysis indicated that the association with type 2 diabetes incidence appears to be more consistent than for CHD, as stearic acid was not associated with increased CVD risk [62], potentially due to its reported neutral effect on LDL-C and partial conversion to oleic acid [35, 63]; however, more research is needed to fully understand its long-term health effects. A recent combined analysis of three large cohorts and an updated meta-analysis of 49 studies demonstrated that polyunsaturated fatty acids (PUFAs) are associated with varying risks of CVD [64]. Specifically, higher intake of n-3 PUFAs was linked to a reduced risk of CHD, whereas total n-6 PUFAs were associated with a lower risk of stroke. Among PUFAs predominant in the portfolio diet, linoleic acid (LA) was inversely associated with CHD and stroke, while alpha-linolenic acid (ALA) showed no significant association with either condition. Additionally, plant-based monounsaturated fatty acids (MUFAs) have been associated with a lower risk of CHD compared to animal-based MUFAs [65]. These findings underscore the differential cardiovascular effects of fatty acids based on type and chain length, highlighting the need for further research into the fatty acid profile and the long-term health implications of the portfolio dietary pattern, including more recent iterations of the diet that add plant-based MUFAs. Acylcarnitines, of which several were lowered with the portfolio diet intervention, have also been associated with an increased risk of type 2 diabetes and may increase insulin resistance [66, 67], and have been associated with higher CVD risk [68]. Furthermore, N2-acetylornithine has been negatively associated with hypertension [69], and lenticin has been shown to improve insulin sensitivity in vitro [70]. In addition, a plasma metabolite profile related to healthy plant-based diets (that included several metabolites impacted by the portfolio diet including N2-acetylornithine and L-pipecolic acid) was associated with a lower risk of type 2 diabetes incidence in the Nurses’ Health Studies and Health Professionals Follow-up Study [71]. Overall, these metabolomic changes highlight additional biological mechanisms through which the portfolio diet may reduce CVD and type 2 diabetes risk, emphasizing its broader potential benefits for cardiometabolic health.

Strengths and limitations

Strengths of our study include that all foods were provided to participants in both trials, providing an ideal study design to measure metabolomic responses to this dietary pattern. We also examined metabolomic responses in two trials, and observed many similar metabolites across both studies, providing more confidence in these metabolites as potential biomarkers for the portfolio diet. Study one was also a cross-over design, in which participants acted as their own control. Furthermore, we examined repeated measures of samples, which is a strength as many metabolomics studies only include one or two samples per participant. Lastly, between intervention comparisons and adjustment for LDL-C allowed us to explore metabolites that may be more specific to the foods recommended in the portfolio diet. Our study does, however, have several limitations. First, the interventions were only 4 weeks in duration; therefore, we do not know if these metabolomic changes will be stable in longer-term studies, and they had small sample sizes. Second, the majority of the participants were older White men with hypercholesterolemia and at higher risk for CVD. Therefore, the generalizability to other populations needs to be confirmed, such as in lower CVD risk individuals. We were unable to assess dietary changes in n-3 and n-6 PUFAs, as well as other potential dietary factors influencing the metabolome, due to limited available data from the diet records. More comprehensive dietary data would allow for a better comparison of diet composition and its impact on metabolomic changes. Additionally, we recognize that individual metabolites may have only a modest influence on the cardiometabolic risk factors examined. Therefore, future studies should explore joint effect analyses to better capture the collective impact of multiple metabolites. We also only assessed known metabolites; therefore, future studies should examine untargeted metabolomics analysis or new known classes of metabolites not identified by the metabolomic platform used, such as polyphenol or phytochemical related metabolites. Finally, it is possible that additional biologically relevant metabolites associated with the portfolio diet were not captured in our analysis, particularly as we only examined plasma and no other biological samples.

Conclusions

Using plasma samples from two randomized controlled feeding trials, we identified many metabolites (most of which were lipids and amino acids) that significantly changed in response to the portfolio diet. Most of these metabolites reflect the plant-based nature and cholesterol-lowering effects of the dietary pattern. These metabolites may be used to assess adherence to a plant-based diet high in legumes, nuts, viscous fiber, and phytosterols in future nutrition research studies, however, further research is needed to validate these metabolites as biomarkers of a plant-based diet, particularly given the unique components of the portfolio diet, which may not be representative of all plant-based dietary patterns.