Higher Dietary Polyphenol Intake Is Associated With Lower Blood Inflammatory Markers

Abstract

Background

Evidence suggests a link between polyphenol intake and reduced incidence of several chronic diseases. This could arise through associations between polyphenol intake and reduced systemic oxidative stress and subsequent inflammation. However, confirming this association is difficult, as few large cohorts allow for comprehensive assessments of both polyphenol intake and markers of systemic inflammation.

Objectives

To address this, polyphenol intake was assessed in the UK-based Airwave cohort using 7-d diet diaries and data from Phenol-Explorer to test for associations between polyphenol intake and blood biomarkers of inflammation.

Methods

Participants included 9008 males and females aged 17–74 y (median age: 42 y) whose data was included in a cross-sectional analysis. Phenol-Explorer was used to estimate individuals’ polyphenol intake from diet data describing the consumption of 4104 unique food items. C-reactive protein (CRP) and fibrinogen were used as blood biomarkers of inflammation.

Results

There were 448 polyphenols found in reported diet items. Median total polyphenol intake was 1536 mg/d (1058–2092 mg/d). Phenolic acids and flavonoids were the main types of polyphenols, and nonalcoholic beverages, vegetables, and fruit were the primary sources. Variation in energy-adjusted polyphenol intake was explained by age, sex, salary, body mass index, education level, smoking, and alcohol consumption. Linear regressions showed inverse associations between total daily intake and both CRP (β: −0.00702; P < 0.001) and fibrinogen (β: −0.00221; P = 0.038). Associations with specific polyphenol compound groups were also found. Logistic regressions using total polyphenol intake quartiles showed stepwise reductions in the odds of elevated CRP with higher intake (6%, 23%, and 24% compared with quartile 1; P = 0.003), alongside 3% and 7% lower odds per unit of polyphenol consumption equivalent to 1 cup of tea or coffee per day.

Conclusions

This study describes polyphenol intake in a large, contemporary UK cohort. We observed associations between higher intake and lower CRP and fibrinogen. This contributes to evidence supporting the health benefits of dietary polyphenols.

Introduction

Polyphenols are phytochemicals found in plant-derived foods including fruits, vegetables, and grains and form a significant proportion of phytonutrients in human diet; cumulative intake may be >100 times that of most vitamins [1]. Common polyphenol sources include coffee, tea, fruits, vegetables, juices, chocolate, and wine. Consumption has been associated with a reduced risk of cardiovascular diseases [2], cancers [3], metabolic disorders including diabetes [4,5], and late life general cognitive impairment and dementia [6,7].

Polyphenols include phenolic acids, flavonoids, stilbenes, and lignans and are present in plants in a range of forms including glycosides, esters, and to a lesser degree, aglycones (without bound groups) [1]. Compounds are often sorted into 3 groups according to structural characteristics: 1) phenolic acids; 2) flavonoids; and 3) aggregated nonflavonoids, or “other polyphenols” [8]. Factors of native polyphenol structure together with food preparation (including packaging and cooking) significantly impact polyphenol content, bioavailability, and metabolism [9,10]. It is therefore vital to account for the way that food is prepared when assessing polyphenol intake.

Polyphenol intake has been studied in populations in numerous regions including throughout Europe [[11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23]], Asia [[24], [25], [26], [27], [28]], the Americas [[29], [30], [31], [32], [33], [34]], and Australia [35,36]. Reported intake ranges widely according to location and population group, for example, with a lower reported mean intake being 173.31 ± 141.02 mg/d in Australian childhood cancer patients [36] and higher intakes including 2770.7 ± 1552.4 mg/d in elderly individuals in Klang Valley, Malaysia [37]. Reported European intakes typically range between ∼500 and 1800 mg/d with increases in more northern populations; the lowest reported intake was for males and females in Greece (774 and 584 mg/d, respectively) and the highest in Aarhus, Denmark (1786 and 1626 mg/d) [17]. However, many studies are limited by small sample size, have a narrow focus on select compounds or subgroups, or lack adjustments for the way that foods were processed.

Inflammation is closely linked to risk and prevalence of chronic health conditions including cardiovascular [38] and neurological diseases [39] and is often studied in relation to diet, including adherence to dietary patterns and intake of dietary components, with a majority of studies identifying an effect of diet on inflammatory markers in free-living adults [40]. Commonly used plasma markers of systemic inflammation include C-reactive protein (CRP), a protein produced in response to physiological stress [41], which independently predicts all-cause and cardiovascular mortality [42], and fibrinogen, a coagulation factor associated with increased vascular and nonvascular mortality [43]. Concentrations of both are associated inversely with diet quality scores [44], typically defined according to dietary indices such as the Healthy Eating Index [45] or Dietary Inflammatory Index [46] or patterns such as the Mediterranean diet [47], which favor polyphenol-rich foods such as fruit, vegetables, and whole grains over animal-derived and processed foods. Polyphenols have been shown to be anti-inflammatory in in vitro and in vivo models. As a first test of whether this mechanism could mediate the association of higher polyphenol intake with a lower incidence of chronic disease [48], intake would be expected to show inverse associations with inflammatory markers. Previous analyses have suggested this but have been limited in number; higher polyphenol-derived antioxidant content scores were associated with lower CRP [49], and an inverse association between consumption of grape products and CRP was also reported [50].

Here, we aim to comprehensively describe polyphenol intake for a large, mixed UK cohort by identifying the relative contributions of different foods, considering a large range of diet items and methods of food processing. We then investigate associations of intake with inflammatory markers. To our knowledge, this is the most thorough assessment of polyphenol intake in a large contemporary population. Our study provides a foundation for future assessments of polyphenol intake and its relationships with a broader range of health-related measures and outcomes.

Methods

Study population

The Airwave Health Monitoring Study is a longitudinal observational cohort study of UK police employees, originally devised to investigate the long-term effects of TETRA, a digital communication system, on health. Study design and rationale, as well as standard protocols followed for data collection, have been described previously [51]. This cohort provides data with extensive coverage including variables describing diet, lifestyle, blood and anthropomorphic measures, cognitive performance, and general health and disease outcomes. A participant flow chart describing sample sizes of eligible participants available for analyses is shown in Figure 1.

Figure 1.

Strobe diagram of participants available for analysis in the Airwaves Health Monitoring Study.

Dietary assessment

Dietary data was collected using a 7-d food diary (for which 89% of participants completed 7 full days). Full details of diet assessments have been described previously [52]. Diet items were disaggregated into groups describing their composition according to ingredients’ weight contributions to the total and assigned disaggregation factors ranging between 0 and 1 describing the proportion originating from each group (hereafter called food groups). The food groups considered in this study were fruit (excluding fruit juice), legumes (including peanuts), vegetables (excluding legumes), potatoes, cereals (whole grain), cereals (refined), nuts (excluding peanuts), seeds, olive oil, other fats and oils, milk products, herbs and spices, nonalcoholic beverages, alcoholic beverages, sugars/preserves, and sauces. Several groups were created specifically for this analysis: herbs and spices, nonalcoholic beverages, alcoholic beverages, other fats and oils, milk products, sugars/preserves, and sauces (breakdown summarized in Table 1). The content of each food diary was summed and divided by the number of days reported to create an average daily intake. Diet data for 9013 participants were available for this study; 5 were excluded due to extreme reported energy intakes, defined as <500 or >4000 kcal/d for females and <800 or >5000 kcal/d for males, leaving a total of 9008 participants.

TABLE 1.

Existing and added food groups, Airwave diet diary data

Fruits	Legumes	Vegetables	Potato
Cereals (whole)1	Cereals (refined)	Nuts	Seeds
Olive oil1	Herbs and spices1	Nonalcoholic beverages1	Alcoholic beverages1
Other oils/fats	Milk and milk products1	Sugars/preserves1	Sauces1

¹Manually added food groups.

Blood biomarkers

Data for several blood biomarkers were collected from participants. The biomarkers selected for use in this study are CRP (mg/L) and fibrinogen (g/L). Nonfasted venous blood samples were collected on site and processed at a designated study laboratory following transport in a thermoporter at 0°C–4°C. Serum was used for high-sensitivity CRP assays. For analyses requiring classification of normal or high (at risk) levels of blood biomarkers, the following criteria were used to define participant measures as high-risk: CRP ≥3 mg/L, fibrinogen >4 g/L. Blood biomarker data was available for 8966 of the 9008 participants in this study.

Estimation of polyphenol intake in the Airwave cohort

Introduction to Phenol-Explorer

Data describing the polyphenol content of items from diet diaries were obtained from Phenol-Explorer, a comprehensive online database aggregating >37,000 data points describing the polyphenol content of commonly consumed food items, collected from 638 peer-reviewed publications [53]. These data describe the composition of 502 distinct polyphenols in 452 plant foods or plant-derived products including fruits and vegetables, nuts, seeds, oils, cereals and cereal products, herbs and spices, alcoholic beverages such as beer and wine, and nonalcoholic beverages such as fruit juices, tea, and coffee. Phenol-Explorer details measures of 4 main polyphenol classes: flavonoids, phenolic acids, lignans, and stilbenes, as well as other compounds classified as “other polyphenols” including tyrosols, alkylphenols, and phenolic terpenes. Data were collected using various methods, including gas and normal phase HPLC chromatographic methods with or without a hydrolysis step, the Folin assay (total polyphenols), and the pH differential method (total anthocyanins). For this study, total polyphenol intake was defined as the sum of estimated polyphenol measures derived from chromatographic methods and normal phase HPLC; measures from the Folin assay and pH differential method were discarded as these both describe independent totals.

Correspondence between diet items and the Phenol-Explorer database

Participants collectively reported 4104 distinct food items, 2753 of which were identified as containing polyphenols. Animal-derived items such as meat, dairy, and eggs that contain no or only trace amounts of polyphenols were excluded. Where possible, items were matched directly to foods in the Phenol-Explorer database. For cases in which items did not have clear matches in Phenol-Explorer, or descriptions were too general for precise matching, items were considered individually and included via 1 of 3 procedures: substitution, generation of mean polyphenol profiles using available data from similar foods, or manual entry by adjusting relevant data using factors agreed with a nutritionist and research dietitian (Supplemental Figure 1, Supplemental Table 1, and Supplemental Excel sheets).

Application of retention factors

Phenol-Explorer also aggregates data describing the effects of cooking and food processing on polyphenolic content [54,55]. These processes modify polyphenol content and bioavailability and can subsequently affect the magnitude of their biological activities such as inhibition of inflammation-associated phenomena including protein denaturation, hemolysis denaturation, lipoxygenase activity, and proteinase activity [56]. These retention factors are the product of 2 measures: a yield factor describing the ratio of food weight after compared with before processing, and a factor describing the ratio of polyphenol concentration in processed compared with raw items. Retention factors are available for compounds from >100 foods sourced from 129 peer-reviewed publications and describe the relative concentrations of polyphenol compounds and groups, including total polyphenols measured using the Folin assay [54]. Where total polyphenol retention factors specific to a given food item were available, these factors were applied to all compounds. Measures of changes in total polyphenol content were chosen over compound-specific ones to preserve consistency in our approach of adjusting available and inferred polyphenol composition data and to avoid a data availability bias by applying sparse compound-specific data. For cases in which food-specific retention factors were unavailable, mean polyphenol profiles had been used, or processes not accounted for by Phenol-Explorer such as peeling and drying were reported, bespoke retention factors were generated and used to adjust measures of polyphenol content (full details provided in Supplemental Figure 1, Supplemental Table 1, and Supplemental Excel sheets).

Estimation of polyphenol intake

Total polyphenol content was defined as the sum of individual polyphenols measured using chromatography, chromatography after hydrolysis in foods in which polyphenols were linked to a food matrix and only solubilized after basic or acid hydrolysis (lignans in several foods, and for phenolic acids in cereals, legumes, walnuts, and olives), and normal phase HPLC (proanthocyanidins only). Polyphenol intake was also calculated as aglycone equivalents by removing the proportional contribution of conjugates such as glycosides and esters to the overall molecular weight of the molecule from the intake of a given compound. After these steps, polyphenol intakes were calculated for each participant as the sum of the product of the following components for a given food item (performed for each compound):

$$\frac{\text{Mean}\text{daily}\text{consumption}\left(\mathrm{g}\text{or}\text{ml}\right)\ast \text{Disaggregation}\text{factor}\ast \text{Polyphenol}\text{content}\left(\frac{\text{mg}}{100\mathrm{g}}\right)\ast \text{Retention}\text{factor}}{100}$$

Statistical analyses

For results of all analyses, unless otherwise stated, polyphenol intake is adjusted for energy intake using the residual method. Intakes for all individual polyphenols, subclasses (group, subgroup, and family), and total polyphenols were determined for the entire Airwave population, as well as according to sociodemographic characteristics including sex, age, BMI, education level, salary, physical activity, and smoking and alcohol consumption status. Nonparametric unpaired 2-samples Wilcoxon (2 categories) or Kruskal–Wallis (>2 categories) tests were carried out using median values to compare group differences. Within-text summary statistics are reported as median (lower quartile – upper quartile) for consistency. Categorical data are presented as frequency (percentage). Multiple linear regression analyses were used to investigate associations between polyphenol intake and blood biomarkers. For linear regressions, intake values for total, flavonoids, and other polyphenols were moderately positively skewed and adjusted using square root transformations. Intake values for phenolic acids were strongly positively skewed and adjusted using natural log (Ln) transformations. The same transformations were applied to intake values after adjustment for energy intake following confirmation using visual assessment. Measures of CRP were Ln-transformed. Logistic regression models were used to evaluate associations between quartiles of polyphenol intake and classification as “at risk” for elevated inflammatory markers (CRP and fibrinogen) and relationships between unit increases of polyphenol intake and risk of marker elevation. Wald tests were used to assess the significance of the effect of ascending quartile rank on outcome variables. All regression analyses were adjusted for age, sex, and salary category. Volunteers with missing blood biomarker data (n = 2 for CRP; n = 40 for fibrinogen) were not included in analyses of that specific parameter.

Complete summaries of polyphenol intake at the total, group, subgroup, family, and compound levels (for intact compounds and aglycones) were calculated alongside total and group-specific major food contributors (Supplemental Tables 2–13). Reported P values for nonparametric tests were based on 2-sided tests. Linear regression outputs are presented as β estimates and associated P values. Logistic regression outputs for quartile analyses are presented as odds ratios (ORs) with 95% confidence intervals (CIs) and P values for Wald tests describing the effect of ascending quartile rank on outcome variables. Logistic regression outputs for notional unit increases in tea and coffee intake, describing increases in the total mg of polyphenols in a given cup rather than tea- and coffee-specific profiles, are presented as ORs with 95% CIs and P values. All statistical tests used a 5% significance threshold. R version 4.1.1 (10 August, 2021) was used for all analyses. Results from all analyses including correlations, regressions performed with and without energy adjustment, and those adjusted for additional variables, are shown in Supplemental Tables 14–27.

Results

General characteristics of the study population are shown in Table 2. Of the 4104 food items reported in the Airwave cohort diet diaries, 2753 contained polyphenols. There were 448 different polyphenols found to be consumed across the cohort, of which 93 had a median intake of >1 mg/d; this is a summary statistic commonly given for comparison between cohorts.

TABLE 2.

General characteristics of the study population (n = 9008)

Characteristics	Median (LQ−UQ)	Minimum	Maximum
Age (y)	42 (35–47)	17	74
BMI (kg/m2)	26.6 (24.2–29.3)	14.4	58.0
Energy intake (kcal/d)	1892 (1587–2218)	593	4620

Abbreviations: BMI, body mass index; LQ, lower quartile; UQ, upper quartile. Data presented as median (LQ–UQ).

Polyphenol intake across sociodemographic characteristics

Polyphenol intake across sociodemographic factors in the Airwave cohort is shown in Table 3. Median total intake was 1536 mg/d (1058–2092 mg/d), while total intake of aglycones was 1038 mg/d (739–1403 mg/d). Unadjusted intake (not adjusted for total energy intake) was higher in males, and adjusted intake was higher in females. Polyphenol intake was higher with increasing age class, and higher intake was generally found with increasing BMI. Reduced intake was observed with increasing education level, and higher intake was seen with increasing salary. Smokers had higher adjusted polyphenol intake than nonsmokers, and participants who reported drinking alcohol also reported higher intake.

TABLE 3.

Total polyphenol intakes in the Airwave cohort across selected sociodemographics

Characteristics	n	Unadjusted intake (mg/d) Median (LQ–UQ)	P	Energy-adjusted intake (mg/d) Median (LQ–UQ)	P
Total population	9008	1536 (1058–2092)		1875 (1415–2411)
Sex			<0.001		<0.01
Male	5474	1581 (1095–2142)		1866 (1397–2411)
Female	3534	1458 (1000–2008)		1886 (1455–2412)
Age class			<0.001		<0.001
<35 y	2336	1116 (747–1595)		1477 (1118–1937)
35–44	3349	1556 (1102–2094)		1885 (1458–2406)
45–54	2721	1778 (1330–2349)		2117 (1664–2674)
≥55	602	1880 (1372–2502)		2230 (1730–2858)
BMI			<0.001		<0.001
Underweight	39	949 (677–1418)		1356 (1128–1825)
Healthy	2865	1442 (1006–1971)		1802 (1372–2322)
Overweight	4271	1586 (1092–2129)		1914 (1437–2439)
Obese	1833	1568 (1095–2163)		1915 (1445–2520)
Education level			<0.001		<0.001
Low	3037	1606 (1129–2132)		1948 (1498–2475)
Medium	3529	1498 (1020–2071)		1845 (1384–2372)
High	2442	1494 (1016–2070)		1827 (1365–2370)
Salary category			<0.001		<0.001
1	2428	1455 (988–2002)		1849 (1403–2375)
2	3799	1443 (993–1981)		1786 (1339–2301)
3	2577	1703 (1237–2287)		1994 (1533–2569)
4	204	1692 (1340–2405)		2054 (1674–2670)
Physical activity			0.022		0.185
Low	1413	1491 (1039–2052)		1856 (1402–2404)
Medium	2950	1535 (1046–2059)		1877 (1409–2372)
High	4645	1552 (1075–2128)		1879 (1428–2439)
Smoking status			0.094		<0.01
Yes	726	1592 (1076–2243)		1936 (1434–2587)
No	8282	1533 (1056–2085)		1868 (1413–2396)
Alcohol status			<0.001		<0.001
Yes	8523	1549 (1068–2099)		1883 (1423–2417)
No	485	1335 (809–1844)		1729 (1274–2253)

Abbreviations: BMI, body mass index; LQ, lower quartile; UQ, upper quartile.

Data presented as median (LQ–UQ) of polyphenol intake. Polyphenol intakes were adjusted for energy intake using the residual method. BMI: underweight, <18.5 kg/m²; healthy, 18.5 ≤ x < 24.9 kg/m²; overweight, 25 ≤ x < 29.9 kg/m²; obese, >30 kg/m². Education level: low, general certificate of secondary education or less; medium, advanced level, or vocational qualification; high, bachelor’s degree or postgraduate. Salary level: 1, <£25,999; 2, £26,000–£37,999; 3, £38,000–£59,999; 4, >£60,000. Physical activity: low, medium, and high – International Physical Activity Questionnaire score of 1, 2, or 3, respectively. Smoking and alcohol status based on self-reported responses unless responses conflicted with diet data. Comparisons between categories were performed using unpaired 2-samples Wilcoxon test (2 categories) or Kruskal–Wallis test (>2 categories).

We found significant variation in the intakes of different polyphenols. For example, we found higher unadjusted intakes of phenolic acids in males than females: 666 mg/d (343–1219 mg/d) compared with 530 mg/d (289–1014 mg/d) (P < 0.001), and higher intakes of flavonoids in females than males, for both unadjusted (632 mg/d [405–905 mg/d] compared with 616 mg/d [374–889 mg/d]; P = 0.008) and adjusted intakes (919 mg/d [698–1184 mg/d] compared with 844 mg/d [610–1109 mg/d]; P < 0.001).

Primary sources of polyphenol intake

Contributions to polyphenol intake from each food group were calculated alongside the primary sources for each group (full details provided in Supplemental Table 28). Measures of intake for polyphenol groups (phenolic acids, flavonoids, and aggregated other polyphenols including stilbenes and lignans) and aglycone intakes are also shown. Nonalcoholic beverages were the primary sources of polyphenols (955 mg/d [569–1438 mg/d], 665 mg/d [408–962 mg/d] as aglycones), followed by vegetables and fruit (222 mg/d [114–383 mg/d] and 87 mg/d [30–168 mg/d], respectively). Whole grain cereals, potatoes, and alcoholic beverages each contributed from 10 to 50 mg/d, while refined cereals, sugars/preserves (such as chocolate and jam), and legumes provided from 1 to 10 mg/d. Nuts, seeds, herbs and spices, olive oil, nonolive oils/fats (such as vegetable or sesame oil), milk products, and sauces each accounted for <1 mg/d. Within food groups, primary polyphenol sources included coffee and black tea for nonalcoholic beverages (51% and 36%, respectively), red wine for alcoholic beverages (80%), and bran cereals and whole grain bread for whole grain cereals (64% and 28%, respectively). Primary polyphenol sources for vegetables and fruit were food group-level polyphenol profiles (92% and 43%, respectively), partly due to the high frequency of ambiguous foods reported in these groups such as fruit salads and vegetable stir fries or casseroles. Subsequent primary sources for these food groups included yellow onions (vegetables, 2%) and apples and strawberries (fruit, 31% and 7%, respectively). Flavonoids and phenolic acids were the 2 major polyphenol groups in terms of intake (623 mg/d [389–896 mg/d] and 608 mg/d [317–1137 mg/d], respectively). The single largest contributing food group was nonalcoholic beverages (contributing 81% of flavonoids and 59% of phenolic acids).

Hydroxycinnamic acids and flavanols account for the greatest proportions of polyphenol consumption (subgroup and compound summary)

Intakes of polyphenol subgroups were also calculated for both intact compound totals and their related aglycones (full details provided in Supplemental Table 29). Hydroxycinnamic acids were the highest consumed subgroup when considered as intact compounds, comprising 42% of all polyphenols with a majority contribution from coffee of 79%. Flavanols were the second highest consumed in terms of intact compounds, and the highest for aglycones, comprising 25% of all polyphenols with a majority contribution from black tea of 65%. Intact flavonols and hydroxybenzoic acids comprised 8.4% and 7.9% of all polyphenols consumed, with majority contributions from vegetables and black tea (55% and 53%, respectively). When considered as aglycone equivalents, relative contributions from hydroxycinnamic acids to total intake decreased from 42% to 32%, while relative contributions from flavanols, flavonols, and hydroxybenzoic acids increased from 25% to 33%, 8.4% to 6.8%, and 7.9% to 12.7%, respectively. Relative contributions of other polyphenol subclasses remained similar.

The 30 compounds that together accounted for the largest daily intake of polyphenols all had median daily intakes of >8 mg/d (full details provided in Supplemental Table 30). Seventeen of these compounds were flavonoids, including 13 flavanols, mostly provided by black tea. Other flavonoid compounds included anthocyanins and flavonols and were mostly provided by vegetables. Of these, 9 were phenolic acids, including 7 hydroxycinnamic acids and 2 hydroxybenzoic acids, mostly provided by coffee and black tea, respectively. The remaining compounds were tyrosols and alkylphenols, which were primarily contributed by vegetables and bran breakfast cereals, respectively. Together, the 30 most-consumed compounds accounted for 73% of overall total polyphenol intake. Secondary analyses were also carried out to investigate the relationship between intake of these top 30 compounds, measures of inflammation, and risk of elevated inflammatory status (presented in full in Supplemental Tables 31–44). Although most relationships remained the same in terms of direction, strength, and significance for this analysis, some were exaggerated, including the inverse relationship between phenolic acid and flavonoid intakes and fibrinogen levels, which became stronger in magnitude and significant where previously they had not. This highlights the potential importance of differentiating compounds from one another in analyses and suggests a potentially nonlinear effect of increasing levels of intake, whereby effects may be greater when single compounds are consumed in greater amounts.

A full breakdown of contributing compounds to each group and their percentage contribution across the cohort is shown in Supplemental Table 7.

Intakes of all aglycones across the Airwave cohort were determined (Supplemental Tables 8–13); 208 aglycones were consumed, of which 42 were in amounts >1 mg/d. The aglycone accounting for the largest daily intake was caffeic acid (178 mg/d [62–437 mg/d]).

Higher polyphenol intake is inversely associated with inflammation

We tested for relationships between polyphenol intake (total and group-level) and inflammatory markers using linear and logistic regressions, adjusted for age, sex, and salary category. Outputs from all models, including initial correlations and regressions adjusted for additional variables, are presented in Supplemental Tables 14–27.

Linear regressions revealed small but significant inverse relationships between total polyphenol intake (P < 0.001), phenolic acids (P = 0.005), flavonoids (P < 0.001), and the aggregated total of other polyphenols (P < 0.001) and CRP, and between total polyphenol intake (P = 0.038) and aggregated total of other polyphenols (P = 0.013) and fibrinogen (Table 4).

TABLE 4.

Linear regressions between polyphenol intake and inflammatory markers

Biomarker	Polyphenol intake level	β	P
CRP (mg/L)	Total	−0.00702	<0.001
	Phenolic acids	−0.0584	0.005
	Flavonoids	−0.0109	<0.001
	Other polyphenols	−0.0172	<0.001
Fibrinogen (g/L)	Total	−0.00221	0.038
	Phenolic acids	−0.0247	0.141
	Flavonoids	−0.00224	0.128
	Other polyphenols	−0.00685	0.013

Abbreviation: CRP, C-reactive protein.

Data presented as linear regression coefficients and associated P values. Coefficients are presented to 3 significant digits. Bold indicates a significant relationship. “Other polyphenols” refers to aggregated compounds from stilbenes, lignans, and other polyphenols categories defined by Phenol-Explorer. Polyphenol intake measures are adjusted for energy intake and transformed using square root (total, flavonoids, and other polyphenols) or Ln (phenolic acids) transformations. CRP measures are Ln-transformed. Models were adjusted for age, sex, and salary category.

Logistic regression was then used to generate ORs to examine the observed relationships between higher polyphenol intake and lower plasma CRP and fibrinogen in terms of the odds of marker elevation. Odds of elevated CRP decreased in a stepwise trend with increasing quartiles of intake, with 6%, 23%, and 24% lower odds for quartiles 2, 3, and 4 relative to quartile 1 (Table 5). A similar trend was seen for fibrinogen, with quartiles 2, 3, and 4 showing ∼10% lower odds of elevation relative to quartile 1, but it was nonsignificant.

TABLE 5.

Odds ratios for elevated inflammatory markers by total polyphenol intake quartile

Biomarker	Quartile	OR (95% CI)	P-trend1
CRP (mg/L)	1	1.00	0.003
	2	0.94 (0.79, 1.10)
	3	0.77 (0.65, 0.91)
	4	0.76 (0.64, 0.91)
Fibrinogen (g/L)	1	1.00	0.200
	2	0.90 (0.79, 1.02)
	3	0.88 (0.77, 1.00)
	4	0.91 (0.80, 1.04)

Abbreviations: CI, confidence interval; CRP, C-reactive protein; OR, odds ratio.

Data presented as ORs (95% CIs) and associated P values for risk of categorization as “at risk” for inflammatory markers according to energy-adjusted polyphenol intake quartile. Bold indicates significant trend.

¹Trend across ranked quartiles detected using Wald test. Models were adjusted for age, sex, and salary category.

To further contextualize participants’ odds of inflammatory marker elevation with higher polyphenol consumption, polyphenol intake data was scaled to produce ORs describing lower odds according to notional unit increases of 1 additional cup of black tea or coffee per day (Table 6).Odds of elevated CRP were 3% lower per additional cup of tea per day and 7% lower per additional cup of coffee (in terms of total polyphenol milligram equivalent rather than beverage-specific profiles), with reductions in risk appearing to be driven by flavonoids (6% decrease in odds per cup). No significant odds of lower risk for elevated fibrinogen were observed. These alterations in risk were then considered in terms of actual tea and coffee intake across the cohort. Tea consumption across the cohort ranged from 0 to ∼13 cups/d, with a median intake of 1.3 cups/d (0.14–2.6 cups/d) across the entire cohort and 1.9 cups/d (0.99–3.0 cups/d) for those reporting any amount of black tea intake. Extrapolated across participants who reported drinking tea, these intakes correspond to a median inferred reduction in odds of elevated CRP of 5.2% (2.8–8.3%), ranging from 0.04% to 37%. Coffee consumption ranged from 0 to ∼13 cups/d, with a median intake of 0.49 cups/d (0–1.5 cups/d) across the entire cohort and 1.0 cups/d (0.42–2.0 cups/d) for those reporting any amount of coffee intake. Extrapolated across all participants who reported drinking coffee, these intakes correspond to a median inferred reduction of odds of elevated CRP of 6.8% (2.8–13%), ranging from 0.004% to 87%. Full model outputs for both energy-unadjusted and -adjusted polyphenol intake, alongside models with additional adjustment variables, are provided in Supplemental Tables 16–27.

TABLE 6.

Odds ratios for elevated inflammatory markers by unit increases in polyphenol intake

Biomarker	Polyphenol intake level	+1 Tea		+1 Coffee
		OR (95% CI)	P	OR (95% CI)	P
CRP (mg/L)	Total	0.97 (0.95, 0.99)	0.002	0.93 (0.89, 0.98)	0.002
	Phenolic acids	1.00 (0.99, 1.00)	0.362	0.98 (0.93, 1.03)	0.362
	Flavonoids	0.94 (0.92, 0.97)	<0.001	—	—
	Other polyphenols	1.00 (1.00, 1.00)	0.053	1.00 (0.99, 1.00)	0.053
Fibrinogen (g/L)	Total	0.99 (0.98, 1.00)	0.077	0.97 (0.94, 1.00)	0.077
	Phenolic acids	1.00 (1.00, 1.00)	0.333	0.98 (0.95, 1.02)	0.333
	Flavonoids	0.99 (0.97, 1.01)	0.234	—	—
	Other polyphenols	1.00 (1.00, 1.00)	0.008	1.00 (0.99, 1.00)	0.008

Abbreviations: CI, confidence interval; CRP, C-reactive protein; OR, odds ratio.

Data presented as ORs (95% CIs) and associated P values for risk of categorization as “at risk” for inflammatory markers according to unit increases of energy-adjusted polyphenol intake equivalent to 1 cup of tea or coffee (1 cup = 260 mL). Data for coffee polyphenol compound profile taken from caffeinated coffee. Data for tea polyphenol compound profile taken from black tea. Bold indicates a significant alteration in risk. Models were adjusted for age, sex, and salary category. Coffee does not contain flavonoids, so no data are presented for this category.

Discussion

This study provides a novel approach to estimating polyphenol intake that can be extended to other cohorts and is, to our knowledge, the most comprehensive combined assessment of polyphenol intake and its associations with inflammatory markers to date. There were 448 individual compounds consumed across the cohort, 93 of which were found to have a median intake in excess of 1 mg/d. Median polyphenol intake (1536 mg/d [1058–2092 mg/d]) was among the higher average measures reported across Europe (∼500–1800 mg/d) [17] and higher than other UK cohorts including one of UK females (1089 ± 814 mg/d) [16] and the UK “health-conscious” group from a European-wide assessment (median 1521 mg/d) [17]. Intake was also high compared with multiple cohorts in Brazil (300 and 1053 mg/d) [30,31], Mexico (694 mg/d) [33], Canada (974 mg/d) [32], and Australia (114 mg/d) [36], and approximately equivalent to 2 Iranian cohorts (1422 and 1552 mg/d) [27,28]. Our study and this earlier work describe similar major polyphenol sources and a consistent set of compounds with the greatest contribution to overall intake [12,17,18,20,31,33]. Variation in intake was seen across multiple sociodemographic factors (Table 3). Intake was higher in males than females, as reported elsewhere [12,18,31], though this reversed following energy adjustment [12]. Higher intake was seen with age [20,22,31] and in participants who reported smoking [12,17,20] or consuming alcohol [11,20]. Higher intake was also seen for increasing BMI, alongside increases in salary. Higher median vegetable and whole grain cereals consumption was seen in the highest relative to lowest salary groups (23% and 20% increase, respectively). A higher intake of alcoholic beverages was also found in the highest salary group (430% increase compared with the lowest group). For BMI, higher intakes were observed for most food groups between obese and underweight individuals (the highest and lowest BMI categories), most notably for legumes, and alcoholic and nonalcoholic beverages (280%, 222%, and 70% increases, respectively). The association between smoking and higher polyphenol intake may be a result of the increased odds of smokers consuming coffee (a rich polyphenol source and dominant contributor to overall intake), with smokers being 7.3% more likely to report drinking coffee than nonsmokers. Linear and logistic regression models with adjustment for additional variables including smoking status were carried out, and associations were found to remain robust with these adjustments (full details provided in Supplemental Tables 16–27). Other variables (including alcohol intake) were also included as adjustment variables in some earlier versions of regression models. However, we found that intake of dietary elements such as alcohol, fiber, vitamin C, and others were strongly correlated with polyphenol intake (virtually all alcohol-, fiber-, and vitamin-rich items are also potent polyphenol sources) and to further adjust for these variables would be confounding. Higher intake was not seen with increasing levels of education as is often, though not always [31], observed [12,17,18,22]. Higher median polyphenol intake between the highest and lowest education groups was seen for food groups including sugars/preserves (primarily chocolate), legumes, fruit, and vegetables (29%, 24%, 11%, and 7.5%, respectively), while nonalcoholic beverage consumption was 14% lower, a relationship reflected by 8.2% lower odds of highly educated participants to report consuming coffee than those in the lowest group.

We found associations between higher polyphenol intake and lower plasma concentrations of the inflammatory markers CRP and fibrinogen (TABLE 4, TABLE 5, TABLE 6). Numerous mechanisms by which polyphenols can reduce inflammation could explain these relationships. These include modulation of oxidative stress (a process intertwined with, and which can promote, inflammation) via mechanisms including the upregulation of antioxidant enzymes through stimulation of regulatory pathways, e.g., the Keap1-Nrf2 pathway [57]. Modulation might also occur through direct free radical scavenging capacities by virtue of their hydroxyl groups and double bond conjugations [58], though consensus in the literature suggests that these direct effects are unlikely to be of major importance [59,60]. Polyphenols can also modulate activities of enzymes including cyclooxygenase, lipoxygenase, and nitric oxide synthase to reduce the production of inflammatory mediators including prostaglandins, leukotrienes, and nitric oxide [48]. Several individual polyphenols have also demonstrated anti-inflammatory activities in animal models, including modulation of NLRP3 inflammasome activities by resveratrol in mice [61] and ellagic acid [62], quercetin [63] (in combination with allopurinol), and rutin [64] in rats, and stimulation of the AhR/Nrf2 pathway by green tea extract (of which 69.4% comprised polyphenols) in mice [65] and tangeretin in rats [66], in addition to various in vitro studies demonstrating these capacities [48]. To elucidate this further, future work could profile a wider range of markers indicative of such mechanisms or search for synergistic patterns of polyphenol intake that confer significant relationships with inflammatory markers and their risk of elevation. One previous study found negative associations between polyphenol intake and inflammation; however, only 7 polyphenol subclasses were used to construct an index of antioxidant content, rather than quantifying polyphenol intake itself [49]. Another study reported reduced risk of elevated CRP in breast cancer patients with high polyphenol intake [67]. Our work adds substantially to these studies by virtue of its sample size [67], depth of assessment of polyphenol intake [49,67], and direct focus on polyphenol intake rather than antioxidant content [49]. Other studies have typically examined the impact of single polyphenols or foods administered in randomized controlled trials, including quercetin [68], sea buckthorn berries [69], soy nuts [70], soy extract [71], olive leaf extract [72], and grape products (extract, juice, raisins, etc.) as part of a meta-analysis [50]. These approaches are comparatively strong in their control of confounding factors and potential for inferences regarding causation but are limited by small sample sizes and focus on polyphenols in relative isolation, ignoring the potential summative or synergistic benefits of a diverse polyphenol profile [73], and only the meta-analysis detected an inverse relationship between grape products and CRP levels [50].

Strengths of our study include the transformation of data from diet diaries into specific quantitative estimates of polyphenol intake using Phenol-Explorer, accounting for changes in polyphenol contents according to food processing. Past studies using Phenol-Explorer have often discarded items reported in diet data with no direct equivalent in the Phenol-Explorer dataset [15,18,30,31]. This is often justified by assuming a minimal contribution from items with missing data, with others imposing a minimum threshold of polyphenol contribution per food serving, for example 1 mg, and discarding foods below it [16]. Other studies reporting on polyphenol intake, for example those using the USDA flavonoid database as a primary data source, focus on a narrow band of polyphenols and hence only report on a limited number of compounds [13,29,35]. In addition, several publications omit in-depth explanations of how imprecise matches of food items were handled and may therefore share one or more of these limitations [21,22,67,74,75]. Other strengths include a large sample size and a detailed approach for adjusting polyphenol intake for food processing using bespoke retention factors. For example, in cases where food group-wide retention factors were required, these were calculated using all available total polyphenol retention factors for foods in a given group. Prior studies, when retention factor data was utilized at all, tended to only use available data, leaving a significant portion of diet data uncorrected for changes in polyphenol content during food processing [11,17,20,23,31,33,76,77], averaged all available factors regardless of food group specificity [19], or averaged retention factors for all compounds in a given food item [22], despite them covering a range of values several fold above and below 1. Additionally, the 7-d food diary method of diet data collection used in this study yields rich data and is considered the gold standard of diet data collection [78]. This method allowed the quantification of polyphenols consumed from 2753 unique diet items (to our knowledge, more than has been reported in previous studies). This approach yielded estimations of intake for 448 individual compounds. To our knowledge, this is the most in-depth characterization of relationships between polyphenol intake and variation in markers of systemic inflammation to date, both in terms of sample size and diversity of polyphenols assessed. Additionally, although the associations detected here are small, given that CRP and fibrinogen increase with age [[79], [80], [81]], confirmation of these associations in this relatively young cohort highlights the power of this large population analysis.

Limitations of our study include, first, that differences in estimated polyphenol intake compared to previous studies may be due to differing methodologies for estimating food intake. Methods for dietary data collection in other studies included diet diaries (as used here), food frequency questionnaires [12,14,[19], [20], [21],26,27,[32], [33], [34],37,67,74], and dietary recall (for example, over 24 h [15,17,22,23,30,35,36] or 48 h [13]). However, we believe that diet diaries may provide more reliable data because they are direct and contemporaneous records of foods consumed, rather than being derived or retrospective. However, given the use of diet diaries in this cohort covered only a single 1- to 7-d period, there is no information on whether participants’ diets were not or will not be different before or after the assessment period. We also recognize that, although it is a standard approach, the examination of the intake of different polyphenols after grouping them based on shared structural characteristics could mask differences between specific polyphenols. However, our study set out to define whether there were any effects of polyphenols apparent in a population of fixed size. Our work provides a basis for future validation efforts that could extend understanding of associations between specific types of polyphenols and systemic inflammatory markers. A third limitation is that, although most blood samples were collected to align with the period described by diet diaries, complete alignment was not always possible. Third, there may be some inaccuracies in our approach of estimating mean polyphenol profiles and of generalizing retention factors to food items other than those to which they specifically refer. Risks of this approach include inferring participants’ intake of compounds not actually included in their diet and inaccurate modification of their intake (usually by reduction) by applying retention factors not specific to a given item. However, we believe that any inaccuracies inherent to these inferential techniques are outweighed by the benefits they confer in providing the most complete possible description of polyphenol intake. Fourth, although our study was comparatively strong in terms of sample size and the number of polyphenols considered, we were unable to measure polyphenol concentrations in plasma and relate this to markers of inflammation, as done elsewhere, with associations having been found between concentrations of several polyphenol compounds and reduced odds of elevated CRP [82]. Finally, due to the observational and cross-sectional nature of our study design as a result of data availability, temporality and causality cannot be established between the phenotypes and markers being investigated.

In summary, the depth of data generated in this work is unique in the field of polyphenol research. Our study provides a valuable resource for future research, both regarding future explorations within this cohort and as a basis for estimating polyphenol intake in other samples. We have also strengthened existing evidence for the relationship between polyphenol intake and inflammation. More research is now needed to understand the mechanisms responsible for these associations.

Acknowledgments

We thank Professor Paul Elliot (Imperial College London) for making the Airwaves data available for this study, and Anwar Albaloul and Alexandra Kopytek, who together with JG, processed the raw dietary data. We also thank all participants in the Airwave Health Monitoring Study.

Author contributions

The authors’ responsibilities were as follows – EDB, GF, PMM: conceived the study; JG: advised and contributed to research design; EDB: performed statistical analyses and drafted the manuscript with GF and PMM; and all authors: read and approved the final manuscript.

Conflict of interest

PMM reports receipt of consultancy fees from Novartis and Biogen; speakers’ honoraria from Novartis, Biogen and Roche; and research or educational funds from Biogen, Novartis, and GlaxoSmithKline. GF reports involvement in educational grants funded in full or in part by Nestle, Heptares, and Quorn Foods; speaker’s fees from Nestle, Quorn, and Millbro; and consultancy from Unilever and Millbro. GF is also a director of a spin out company Melico and an NIHR senior investigator award. ERD-B is the recipient of the studentship. All other authors report no conflicts of interest.

Funding

Studentship name: MRC industrial Collaborative Awards in Science and Engineering (iCASE) studentships. MRC grant reference: MR/R015732/1. PMM acknowledges generous personal and research support from the Edmond J Safra Foundation and Lily Safra, a National Institute for Health Research (NIHR) Senior Investigator Award, the UK Dementia Research Institute, which receives its funding from UK DRI Ltd., funded by the UK Medical Research Council, Alzheimer’s Society and Alzheimer’s Research UK, and the NIHR Biomedical Research Center at Imperial College London. All of the authors are grateful for support from the Imperial College Healthcare Trust (ICHT) NIHR Biomedical Research Center.

Data availability

Data will be available through a request to the study team.

Appendix A. Supplementary data

The following is the Supplementary data to this article: