Development of a (Poly)phenol Metabolic Signature for Assessing (Poly)phenol-Rich Dietary Patterns

The objective assessment of habitual (poly)phenol-rich diets in nutritional epidemiology studies remains challenging. This study developed and evaluated the metabolic signature of a (poly)phenol-rich dietary score (PPS) using a targeted metabolomics method comprising 105 representative (poly)phenol metabolites, analyzed in 24 h of urine samples collected from healthy volunteers. The metabolites that were significantly associated with PPS after adjusting for energy intake were selected to establish a metabolic signature using a combination of linear regression followed by ridge regression to estimate penalized weights for each metabolite. A metabolic signature comprising 51 metabolites was significantly associated with adherence to PPS in 24 h urine samples, as well as with (poly)phenol intake estimated from food frequency questionnaires and diaries. Internal and external data sets were used for validation, and plasma, spot urine, and 24 h urine samples were compared. The metabolic signature proposed here has the potential to accurately reflect adherence to (poly)phenol-rich diets, and may be used as an objective tool for the assessment of (poly)phenol intake.


■ INTRODUCTION
(Poly)phenols, as a large family of plant secondary metabolites, are distributed broadly in the plant kingdom and are present in almost all plant foods and beverages. 1−4 However, estimating (poly)phenol consumption accurately remains challenging, and this can be an important confounding factor when establishing health benefits, particularly in epidemiological studies. 5To overcome such issues, the use of dietary biomarkers as objective measurements of dietary intake has been proposed as an alternative to dietary assessment tools.Biomarkers have been used to estimate the consumption of specific foods, nutrients, or bioactive compounds, or to develop calibrated equations to correct intake estimated from food frequency questionnaires (FFQ) or food records. 6However, very few validated biomarkers exist, and it is very challenging to find suitable biomarkers for each individual food/component within the diet.
The field of metabolomics has emerged as a powerful tool in nutritional research in the last few decades to identify a group of biomarkers linked to specific dietary patterns.A metabolomic profiling approach can be employed to construct a metabolic signature for characterizing a certain diet and reflecting adherence to a specific dietary pattern. 7,8Although several studies have proposed signatures for dietary patterns, including the Mediterranean Diet Score (MDS) 7,9 and the Healthy Eating Index (HEI)-2020, 10 metabolic signature data of habitual diets in this field remains limited.Currently, the relationship between (poly)phenol metabolite levels and the consumption of (poly)phenol-rich diets is poorly understood, and little research has been conducted so far on whether (poly)phenol metabolites in biofluids can predict the intake of (poly)phenol-rich foods. 11Metabolomics studies with a combination of metabolites from a comprehensive profile of all (poly)phenol classes may improve the prediction of dietary (poly)phenol intake.
We have recently developed a (poly)phenol-rich diet score (PPS) to reflect adherence to a (poly)phenol-rich diet in freeliving individuals. 12In this work, we aim to establish a metabolic signature of a (poly)phenol-rich diet using the PPS score and a comprehensive high-throughput targeted metabolomics method for the quantification of a large number of (poly)phenol metabolites in urine and plasma 13 in healthy free-living individuals.
were provided on the initial pages of the 7DDs to help subjects record their diets in as much detail as possible.
Estimation of Dietary (Poly)phenol Intake.(Poly)phenol intake was calculated from food intake (g/d) estimated from both FFQs and 7DDs coupled with the corresponding content in foods (mg/100g, aglycones equivalent) derived from a (poly)phenol database (PPDB) developed at the Department of Nutritional Sciences of King's College London, which has been described in detail previously. 21The PPDB integrated (poly)phenol content data of 1260 raw and processed food items or dishes, which were obtained from multiple sources such as the Phenol-Explorer database, 22 USDA databases, 23−25 and published data.The (poly)phenol content for composition dishes was calculated based on recipes from McCane and Widdowson's (sixth edition) Supporting Information 20 and information on retailer websites.Food items such as animal products with little or no (poly)phenol content were removed from the calculation.The total subclass, class, and total (poly)phenol intake were calculated by summarizing the intake of all of the compounds within the group.
(Poly)phenol-Rich Dietary Score Assessment with FFQs.We calculated the (poly)phenol-rich diet score (PPS) to estimate adherence to a (poly)phenol-rich diet. 12The PPS was designed based on the relative intake levels of 20 (poly)phenol-rich food items from the EPIC-Norfolk FFQ, including tea, coffee, red wine, wholegrains, breakfast cereals, chocolate and cocoa products, berries, apples and apple juice, pears, grapes, plums, citrus fruits and citrus juice, potatoes and carrots, onions, peppers, garlic, green vegetables, pulses, soybeans and related products, nuts, and olive oil.
Participants were scored by quintiles of their intake of each food group (g/d) in the study population.The subjects in the highest quintile among the study population scored 5, and the subjects in the lowest quintile scored 1.The PPS was calculated as the total scores of all 20 food groups ranging from 20 to 100.All food groups were assigned the same level of weight. 12ample Collection and Metabolite Analysis.All plasma and urine (poly)phenol metabolite data presented here were analyzed using the same methodology.In the POLYNTAKE cohort, 24 h urine (n = 229) and fasting plasma samples (n = 204) were collected to analyze the metabolite levels.The collection, storage, processing, and UHPLC-MS analysis with a triple-quadruple mass spectrometer (SHIMADZU 8060, Shimadzu, Kyoto, Japan) of the 24 h urine and fasting plasma samples have been described in detail previously. 12,13A total of 105 urinary and 114 plasma metabolites were identified and quantified with authentic chemical standards.The spot urine samples in the TwinsUK cohort (n = 198) were processed and analyzed using the same method and device mentioned above. 12,17The urinary creatinine levels were analyzed with the Jaffe method by Affinity Biomarker Laboratories (London, U.K.), and the metabolite levels (nM) were adjusted by creatinine (mg/L) into mmol/g creatinine. 17tatistical Analysis.We used R software (version 4.1.2) for statistical analysis. 26Metabolite levels were log-transformed and adjusted for batch effects using the ComBat method (from the sva package in R). 27,28 We adjusted for energy intake estimated from FFQs in the linear regression model to explore the association between PPS and metabolites (lm.beta package in R).All analyses were adjusted for multiple testing (Benjamini and Hochberg false discovery rate (FDR) < 0.05, FDR suggestive significant 0.05− 0.10). 29Metabolites with significant associations with PPS were selected as the next step.
Metabolic signatures of PPS were constructed to represent adherence to (poly)phenol-rich diet based on selected individual significant metabolites, with penalized weights estimated through ridge regression 30 in the derivation data set (24 h urine samples from the POLYNTAKE study).To investigate whether it was possible to replicate the metabolic signatures of PPS in other studies, 3 validation data sets were used: the ABP study (internal subcohort of the POLYNTAKE study with 24 h urine samples), the plasma samples from the POLYNTAKE study, and the external TwinsUK data set with spot urine samples (Supporting Information, Table S1).
To explore the utility and robustness of the metabolite signatures, the correlation between PPS and its metabolic signature was tested by Spearman correlation analyses in both derivation and validation data sets.Spearman correlations were also employed to assess the association between the PPS signature and FFQ and 7DD estimated (poly)phenol intake in each data set.To further examine the agreement between PPS and metabolite signatures, alluvial plots were constructed to illustrate the cross-classification results between quartiles.A detailed flowchart of the analysis is exhibited in Figure 1.

■ RESULTS
Characteristics of Study Population.The characteristics of the participants from the POLYNTAKE cohort are listed in Table 1.The average age of the subjects was 52.3 (SD 17.6) years.The majority of subjects were from the white ethnic group (76.0%), and their average energy intake was 1586.9 kcal/d (SD 465.6).The 24 h urine samples were collected from 99 males and 130 females, whereas plasma samples were collected from 89 males and 115 females.The average PPS was 54.6 (SD 11.5), and total (poly)phenol intake estimated from FFQs and 7DDs were 1635.6 (899.1)mg/day, and 1707.0 (963.8)mg/day, respectively.
The characteristics of the participants in the ABP and TwinsUK cohorts are shown in Supporting Information, Table S2.The average age of subjects was 56.4 (SD 8.9) years in the ABP cohort and 62.0 (SD 9.9) years in the TwinsUK cohort.Most of the subjects were from the white ethnic group (ABP: 80.0% and TwinsUK: 99.0%), and their average energy intake was 1681.3 kcal/day (SD 490.7) and 1782.5 kcal/day (SD 554.3) in ABP and TwinsUK study, respectively.The 24 h urine samples in the ABP cohort were collected from 46 males and 49 females, and the spot urine samples (n = 198) in the TwinsUK cohort were all collected from females.
Correlation between PPS, FFQ, and 7DD Estimated (Poly)phenol Intake.The correlation between PPS, FFQ, and 7DD estimated (poly)phenol intake is shown in Supporting Information, Figure S1.The correlation coefficients ranged from 0.10 to 0.86, among which the strongest correlation was observed between phenolic acids estimated from FFQs and total (poly)phenol estimated from FFQs (r = 0.86), and the weakest correlation was found between phenolic acids estimated from FFQs and other (poly)phenols estimated from FFQs (r = 0.10).
The association between intake of classes of (poly)phenols estimated from FFQs and 7DDs and urinary metabolites was further explored, as shown in Supporting Information, Figure S2A,B.Phenolic acid intake was associated with the highest number of metabolites among all classes, with a total of 36 and 14 individual metabolites associated with phenolic acid intake estimated from FFQs and 7DDs, respectively (all FDRadjusted p < 0.05).The significant standardized regression coefficients (and 95% CI) between urinary metabolites and phenolic acids from both FFQs and 7DDs were all positive (n = 49), ranging from 0.16 (0.03, 0.30) for 3-hydroxybenzoic acid to 0.45 (0.34, 0.57) for 2-hydroxybenzene-1-glucuronide with phenolic acids from FFQs, except for stdBeta for 3,4dihydroxyphenylethanol, which was negative [−0.23 (−0.37, −0.10), FDR-adjusted p < 0.05].
As for other classes, lignans, flavonoids, other (poly)phenols, and stilbene intake estimated from FFQs were significantly associated with 13, 7, 6, and 9 urinary metabolites, respectively, Development and Validation of a Metabolic Signature for Estimating Adherence to a (Poly)phenol-Rich Diet.The 24 h urine sample was chosen to establish the metabolic signature, as it had the highest number of metabolites associated with PPS and had stronger correlations than plasma.Figure 3A shows the percentages of the classes and subclasses of metabolites included in the PPS metabolic signature.Hydroxycinnamic acids and hydroxybenzoic acids, among the subclasses of phenolic acids, accounted for 34.6 and 25.0% in the composition of PPS signature.Figure 3B shows the coefficients of the classes and subclasses for each selected metabolite in the metabolic signature.The scatter plot in Figure 3C identifies a positive visual relationship between the PPS in quintiles and the metabolic signature.In Figure 3D, a Spearman correlation was implemented to assess the statistical correlation between PPS and its metabolic signature in the derivation data set of the POLYNTAKE cohort using 24 h urine samples.In addition, Spearman correlations were also tested in the validation data sets with three sample types across three cohorts (the POLYNTAKE cohort with plasma samples, the ABP cohort with 24 h urine samples, and the TwinsUK cohort with spot urine samples).The strongest correlation was found between PPS and its metabolic signature in the derivation data set [0.33 (0.21, 0.44)], followed by the correlation in the ABP cohort [0.24 (0.04, 0.42)], TwinsUK cohort [0.19 (0.05, 0.32)], and POLYNTAKE cohort using plasma [0.18 (0.04, 0.31)] (FDR-adjusted p < 0.05).In the derivation cohort, the correlation between the PPS and metabolic signature was higher than that of the individual 51 significant metabolites (Supporting Information, Table S3).
Agreements between PPS Derived from FFQs and Metabolite Signatures.The agreement between PPS derived from FFQs and PPS metabolic signatures when ranking participants into quartiles in the derivation and validation data sets is shown in Figure 4.The two methods were comparable in differentiating participants into high and low adherence to the PPS dietary pattern in the derivation data set POLYNTAKE cohort (24 h urine), the validation data set POLYNTAKE cohort (plasma), ABP cohort (24 h urine), and TwinsUK cohort (spot urine) with 74.7, 70.6, 72.6, and 71.7% of participants ranked into the same quartile or adjacent quartile and 6.6, 8.8, 10.5, and 11.1% ranked into the opposite quartile (the first and fourth quartiles), respectively.
(Poly)phenol Metabolic Signatures and FFQ and 7DD Estimated (Poly)phenol Intake.The correlation between (poly)phenol metabolic signatures and FFQ and 7DD estimated (poly)phenol intake in the derivation and validation data sets are shown in Figure 5.The (poly)phenol signatures and FFQ and 7DD estimated (poly)phenol intake were positively correlated, and the highest number of significant correlations were observed in the derivation data set.(Poly)phenol intake subclasses estimated from FFQs were all significantly linked with the metabolic signature in this data set 0.18 (0.06, 0.31) for lignans to 0.30 (0.18, 0.41) for phenolic acids, FDR-adjusted (p < 0.05), except for flavonoids [0.07 (−0.06, 0.19), FDR-adjusted p = 0.32].For 7DDs, a trend for a significant correlation between total (poly)phenols and the metabolic signature was observed (FDR-adjusted p = 0.06).Lignans, other (poly)phenols, and stilbenes from 7DDs were positively linked with the signature [0.18 (0.03, 0.33) to 0.27 (0.12, 0.41)].
In the validation data set, the correlation between lignans from 7DDs and signature in the ABP cohort ranked highest [0.33 (0.14, 0.50), FDR-adjusted p < 0.01], followed by phenolic acids [0.

■ DISCUSSION
This is the first study to investigate and develop a metabolic signature to measure adherence to a (poly)phenol-rich diet based on targeted metabolomics data from multiple biofluids, including 24 h urine, plasma, and spot urine, in a UK population sample.The metabolic signature constituting 51 metabolites showed a stronger correlation with PPS than the individual biomarkers and was significantly correlated with PPS in the internal and external validation cohorts.The total (poly)phenol intake and each class of (poly)phenol intake assessed from FFQs or 7DDs were positively linked with metabolic signatures from at least one derivation or validation cohort.This result indicated the potential application of an array of metabolites as composite markers to assess adherence to a habitual (poly)phenol-rich diet in free-living populations.
More than 8000 different types of plant (poly)phenols have been identified, 31 which increases the complexity of estimating dietary (poly)phenol intake.The PPS dietary pattern was developed to assess adherence to habitual (poly)phenol-rich dietary consumption with a comprehensive list of predefined (poly)phenol-rich food groups, i.e., tea, coffee, red wine, wholegrains, chocolate and cocoa products, berries, grapes, plums, citrus fruits, green vegetables, soy and products, nuts, and olive oil. 12Coherently, the 51 metabolites from 5 major Red and blue indicate positive and negative effects, respectively, and the color intensity represents the degree of effect.7DDs were not collected in the TwinsUK study, so no correlations are shown between the (poly)phenol signature and 7DD estimated (poly)phenol intake in the TwinsUK cohort.The correlation with significance is listed for the coefficient (FDR-adjusted, p < 0.05).The correlation between the signature of PPS in the derivation cohort and total (poly)phenols from 7DDs is also listed for suggestive significance (FDR-adjusted, p = 0.06).PPS, (poly)phenol-rich dietary score; FFQ, food frequency questionnaire; and 7DD, 7-day food diary.
(poly)phenol classes were identified as positively associated with PPS, thus assembling a representative set of metabolites to establish a (poly)phenol-rich diet signature.
As the predominant component of the PPS metabolic signature, phenolic acids accounted for most metabolites associated with PPS, including 18 hydroxycinnamic acids, 15 hydroxybenzoic acids, and six phenylacetic acids.Phenolic acids are abundant in various dietary sources, including coffee, tea, red wine, vegetables, and fruits. 32In the EPIC study, phenolic acids were the main contributors to total (poly)phenol intake in non-Mediterranean (MED) countries, representing 57 and 53% of the total among males and females, respectively. 33As the most consumed (poly)phenols among all age groups in the UK National Diet and Nutrition Survey (NDNS), 34 hydroxycinnamic acids were also identified as having the highest number of associations with PPS among all subclasses in the present study, including 4-O-caffeoylquinic acid (cryptochlorogenic acid, abundant in coffee 35 ), 4′hydroxy-3′,5′-dimethoxycinnamic acid (sinapic acid, i.e., oranges, grapefruits, cranberries, and herbs 36 ), 3′,4′-dihydroxycinnamic acids and derivatives (caffeic acids and derivatives, i.e., tea, wine, and coffee 37 ), 4′/2′-hydroxycinnamic acid and derivatives (p/o-coumaric acid and derivatives, i.e., crops), and 3′/4′-methoxycinnamic acid and derivatives (iso/ferulic acid and derivatives, i.e., bread, fruits and cereals 33 ).Hydroxybenzoic acids are commonly present in glycosylated forms, with tea as the predominant food source in the UK NDNS, 34 compared to coffee as the major source of hydroxycinnamic acids among European regions. 33In our previous work with the same cohort (POLYNTAKE), 12 the (poly)phenol-rich food items included in the PPS contributed 99.7% to total (poly)phenol intake, with tea and coffee contributing the most (33.7 and 44.2% of the total (poly)phenol, respectively).Due to the high contribution of tea and coffee, hydroxybenzoic and hydroxycinnamic acids accounted for the highest number of metabolites that were positively associated with PPS in this study.Moreover, phenolic acids are also gut microbial metabolites of nearly all (poly)phenol classes, 38 which may also contribute to the strong association with PPS observed here.
Flavonoids represent one of the most diverse (poly)phenol groups, which cover more than half of the known (poly)phenol compounds 31 and are commonly found in the UK diet, including flavanones (i.e., citrus fruit and juice), flavonols (i.e., tea, apples, and onions), 1 and dihydrochalcones (i.e., apple and apple juice). 39These food groups were also included in the PPS due to their abundant (poly)phenol content.Here, several metabolites from these subclasses of flavonoids were positively linked with PPS, including naringenin-4′-glucuronide (flavanones), quercetin, quercetin-3-glucuronide, quercetin-7-glucuronide (flavonols), and phloretin (dihydrochalcones), indicating their potential to be biomarkers of a (poly)phenol-rich diet.Other abundant metabolites of flavonoids, e.g., epicatechin-3′sulfate (one of the main metabolites of flavan-3-ols) and (4R)-5-(3′,4′-dihydroxyphenyl)-γ-valerolactone-4′-sulfate (proposed biomarker of flavan-3-ol intake 40 ), were commonly found in tea drinkers in the UK population; 1 however, they were not significantly associated with PPS.The discrepancies might be related to the population of this study, which had a relatively low tea consumption in this cohort (1.5 ± 1.6 cups per day), which might weaken the association between PPS and flavan-3ol metabolites.
Metabolomics profiling, as a validated and high-throughput analytical methodology, 8 was employed in this study to evaluate more than 100 food and microbiota-derived metabolites, including a large number of phase II metabolites, in biofluid 24 h urine, fasting plasma, and spot urine, which provided a snapshot of the (poly)phenol metabolome in a freeliving UK population.To date, evidence has found a positive relationship between some of the metabolites discussed above, in particular hydroxybenzoic, hydroxycinnamic, phenylacetic, and hippuric acids and several plant-rich dietary patterns, such as Alternative HEI-2010 (AHEI-2010), 10 DASH, 45 PDI, 45 MIND, 45 and MDS. 45This overlap between PPS and other dietary patterns may be due to the shared plant-based food groups, such as fruits and vegetables, which highlight the importance of establishing multimetabolite panels of (poly)phenol biomarkers with a comprehensive metabolite profile across each class to be able to distinguish (poly)phenol-specific patterns from other plant-based diets.
The total (poly)phenol and each class of (poly)phenol intake estimated from FFQs and 7DDs in the derivation and three validation cohorts were further employed to test the application of the signature and presented positive correlations with the signature from at least one cohort, indicating that the PPS signature can serve as an assessment tool for (poly)phenol-rich diets.The plasma signature was different from the urine one, with very few metabolites correlating with PPS, which may be largely due to the diverse nature of these biofluids.Kinetic studies indicate that fasting plasma can only capture long half-life metabolites from common food sources, 46 and compounds like epicatechin-3′-sulfate with short half-lives require constant ingestion to maintain a high concentration in plasma for detection. 47On the contrary, urine samples, in particular 24 h urine collection, can capture most compounds excreted from (poly)phenol-rich food sources within several days postconsumption. 48In line with this, the 24 h urine sample metabolites exhibited more associations with total (poly)phenols estimated from FFQs and 7DD.More correlations were found in the signature from the 24 h and spot urine samples, which may be attributed to the similar biofluid nature of the derivation cohort.Since the primary signature was derived from urine samples, more correlations were identified between (poly)phenol intake and the signature from the urine sample.Compared with 24 h urine, spot urine is easier to collect and thus is widely applied in multiple nutritional epidemiological studies.Due to the variation in the time difference between sample collection and (poly)phenol consumption, metabolite estimation is still challenging in freeliving people.The capture of limited urinary excretion would restrict the application of spot urine, whereas 24 h urine would be a sufficient sample to reflect the most recent (poly)phenol consumption.Coherently, lignan intake estimated from FFQs only showed a significant correlation with 24 h urine samples in the POLYNTAKE cohort, but not in spot urine samples from the TwinsUK cohort.Furthermore, the diverse biofluid nature may also contribute to the coefficient of correlation between PPS and metabolic signature decline from the subcohort ABP (24 h urine), the TwinsUK cohort (spot urine), to the POLYNTAKE cohort (plasma), indicating the 24 h urine sample was the most suitable biofluid to reflect (poly)phenol consumption.The discrepancies may also partially be attributed to the difference in population, e.g., the TwinsUK cohort consists of middle-aged twin females. 17hen comparing the metabolic signature with commonly used dietary assessment methods in observational studies, we found a larger number of correlations with (poly)phenol intake estimated from FFQs than 7DDs.Stronger correlations with total (poly)phenol estimated from FFQs are shown compared to 7DDs.As for the class of (poly)phenols, flavonoids, other (poly)phenols, and stilbenes estimated from FFQs but not from 7DDs were significantly linked with the signature from the POLYNTAKE cohort (plasma), subcohort ABP (24 h urine), and POLYNTAKE cohort (24 h urine), respectively.Based on the large proportion (39 out of 51) of the selected individual (poly)phenols from the phenolic acid class, the most robust correlation was observed between the urinary metabolic signature from all cohorts and phenolic acid intake, with a stronger correlation with phenolic acid intake estimated from FFQs than 7DDs.In our previous study comparing the agreement between FFQs and 7DDs, 21 only total (poly)phenol and phenolic acid intake showed moderate agreement between tools, whereas poor agreement was exhibited among the rest of the classes, which is in line with the robust relationship between total (poly)phenol, phenolic acids, and the signature across cohorts with urine samples (24 h urine and spot urine).In addition, since the signature was established based on the PPS from FFQs, the same assessment tool might partially contribute to the stronger correlation with (poly)phenol intake estimated from FFQs.The EPIC FFQ collects habitual intake frequencies of a predefined food list but is prone to recall bias. 5he estimated food records by diaries require participants to provide detailed intake and portion sizes at the time of consumption, which eliminates recall bias but increases burdens on participants and researchers, including the high compliance to the protocol and the intensive work of converting food records, respectively. 49In previous EPIC study publications, the total (poly)phenol intake in the UK was estimated to be 1521 mg/day in healthy adults and around 1700 mg/day in the general population aged 35−74 years. 33onsistently, the total (poly)phenol intake assessed by FFQs and 7DDs in the present study ranged from 1620.2 ± 928.3 mg/day in men (FFQs) to 1750.1 ± 942.8 mg/day in women (7DDs).In addition, high positive correlations were found among the total and each class of (poly)phenols from FFQs and 7DDs, indicating an agreement between the tools.
This study is the first to develop a metabolic signature to evaluate adherence to a (poly)phenol-rich diet by differentiating participants into high-and low-intake levels using a large-scale metabolomics platform and dual dietary measurement tools (FFQs and 7DDs).The metabolomics profiling assembled by an array of (poly)phenol metabolites from urine/ plasma samples enabled us to capture the accurate (poly)phenol-rich food-related metabolome objectively. 13The signatures deriving from machine-learning models with multiple biofluids may allow a further application to assess the metabolic response and adherence to the (poly)phenol-rich diet in randomized controlled trials and prospective studies in other countries.
We acknowledge that quantifying 51 metabolites may not be very practical for routine research, in particular, in large epidemiological studies, and it would be more feasible to implement a simpler signature containing fewer metabolites.Although a few metabolites may not represent the wide range of (poly)phenol subclasses present in the diet, based on the strength of the associations and wide representation from different subclasses, a simplified signature may include cinnamic acid-4′-sulfate (p-coumaric acid-4′-sulfate), entero l a c t o n e -g l u c u r o n i d e , d i h y d r o r e s v e r a t r o l , 2 -( 4hydroxyphenyl)ethanol (tyrosol), phloretin, 2-hydroxybenzene-1-glucuronide (catechol-1-glucuronide), and quercetin.Further work is needed to determine whether this shorter panel of metabolites is a good signature of (poly) phenol-rich diets.
This study has several limitations.First, due to the vast diversity of thousands of (poly)phenol compounds present in the diet, increasing the number of compounds in the current targeted metabolomic profile may lead to a more accurate capture of habitual (poly)phenol consumption.Second, the a priori PPS dietary pattern shares several food items with other plant-rich dietary patterns, which may restrict the specificity of the signature in the evaluation of a (poly)phenol-rich diet.The selected 51 metabolites in the metabolic signature are only a fraction of the thousands of plant (poly)phenols identified so far and are likely predominant for people eating in the UK since the PPS was established based on the UK diet, which may restrict its further application to other countries.Third, the assessment of (poly)phenol intake using FFQs has many limitations, such as self-reported recall bias.In terms of (poly)phenol metabolite assessment, a limitation was that we did not monitor 24 h urine collection compliance.Further research needs to be conducted in the general UK population group with diverse dietary intakes.Interventional feeding studies are needed to validate this signature with actual (poly)phenol consumption from a (poly)phenol-rich diet.Future replication in more countries with larger study populations is also warranted.

Journal of Agricultural and Food Chemistry
To conclude, this study showed that a metabolic signature derived from targeted metabolomics in urine and plasma samples in combination with dietary assessment methods has the potential to reflect adherence to (poly)phenol-rich diets.

Figure 1 .
Figure 1.Flowchart of the analysis approach (generation and validation) to develop the metabolic signature to estimate adherence to (poly)phenol-rich dietary patterns.PPS, (poly)phenol-rich dietary score; FFQ, food frequency questionnaire; and 7DD, 7-day food diary.

Figure 2 .
Figure 2. Association between PPS, FFQ, and 7DD estimated total (poly)phenol intake and (A) urinary and (B) plasma metabolites.A heatmap was plotted according to the standardized regression coefficients (stdBeta).The color scale indicates the effect (stdBeta) of each metabolite on the PPS, FFQ, and 7DD estimated total (poly)phenol intake.Red and blue indicate positive and negative effects, and the color intensity represents the degree of effect.Asterisks indicate significance (*: FDR-adjusted p < 0.05).PPS, (poly)phenol-rich dietary score; FFQs, food frequency questionnaires; and 7DDs, 7-day food diaries.The associations were adjusted for energy intake.

Figure 3 .
Figure 3. Selected metabolites included in the PPS metabolic signature and the correlations between PPS and PPS metabolic signatures from the derivation and validation data set.(A) Percentage of the metabolites in the PPS metabolic signature from the derivation data set.The metabolic signature of PPS was derived based on selected metabolites that were significantly associated with PPS.(B) The coefficient of selected metabolites from the PPS in the derivation data set.Color intensity represents the degree of coefficient of each metabolite.(C) Mean and 95% CI of the derived metabolic signature (51 metabolites) by quintiles of PPS.(D) Forest plot of the correlation between PPS and PPS metabolic signatures from derivation and validation data sets.PPS, (poly)phenol-rich dietary score.

Figure 4 .
Figure 4. Agreements between (poly)phenol and the metabolic signature in ranking participants into quartiles.PPS, (poly)phenol-rich dietary score.

Figure 5 .
Figure 5. Correlation between PPS metabolic signature and FFQ and 7DD estimated (poly)phenol intake in derivation and validation data sets.Red and blue indicate positive and negative effects, respectively, and the color intensity represents the degree of effect.7DDs were not collected in the TwinsUK study, so no correlations are shown between the (poly)phenol signature and 7DD estimated (poly)phenol intake in the TwinsUK cohort.The correlation with significance is listed for the coefficient (FDR-adjusted, p < 0.05).The correlation between the signature of PPS in the derivation cohort and total (poly)phenols from 7DDs is also listed for suggestive significance (FDR-adjusted, p = 0.06).PPS, (poly)phenol-rich dietary score; FFQ, food frequency questionnaire; and 7DD, 7-day food diary.

Table 1 .
Characteristics of the Study Population of the POLYNTAKE Cohort a The number of missing plasma samples is 25.PPS, (poly)phenol-rich dietary score; FFQ, food frequency questionnaire; 7DD, 7-day food diary; and SD, standard deviation.