Meal-induced inflammation: postprandial insights from the Personalised REsponses to DIetary Composition Trial (PREDICT) study in 1000 participants

ABSTRACT Background Meal-induced metabolic changes trigger an acute inflammatory response, contributing to chronic inflammation and associated diseases. Objectives We aimed to characterize variability in postprandial inflammatory responses using traditional (IL-6) and novel [glycoprotein acetylation (GlycA)] biomarkers of inflammation and dissect their biological determinants with a focus on postprandial glycemia and lipemia. Methods Postprandial (0–6 h) glucose, triglyceride (TG), IL-6, and GlycA responses were measured at multiple intervals after sequential mixed-nutrient meals (0 h and 4 h) in 1002 healthy adults aged 18–65 y from the PREDICT (Personalised REsponses to DIetary Composition Trial) 1 study, a single-arm dietary intervention study. Measures of habitual diet, blood biochemistry, gut microbiome composition, and visceral fat mass (VFM) were also collected. Results The postprandial changes in GlycA and IL-6 concentrations were highly variable between individuals. Participants eliciting an increase in GlycA and IL-6 (60% and 94% of the total participants, respectively) had mean 6-h increases of 11% and 190%, respectively. Peak postprandial TG and glucose concentrations were significantly associated with 6-h GlycA (r = 0.83 and r = 0.24, respectively; both P < 0.001) but not with 6-h IL-6 (both P > 0.26). A random forest model revealed the maximum TG concentration was the strongest postprandial TG predictor of postprandial GlycA and structural equation modeling revealed that VFM and fasting TG were most strongly associated with fasting and postprandial GlycA. Network Mendelian randomization demonstrated a causal link between VFM and fasting GlycA, mediated (28%) by fasting TG. Individuals eliciting enhanced GlycA responses had higher predicted cardiovascular disease risk (using the atherosclerotic disease risk score) than the rest of the cohort. Conclusions The variable postprandial increases in GlycA and their associations with TG metabolism highlight the importance of modulating TG in concert with obesity to reduce GlycA and associated low-grade inflammation–related diseases. This trial was registered at clinicaltrials.gov as NCT03479866.


Introduction
Persistent low-grade inflammation is a common pathogenic feature of many chronic diseases, including cardiovascular disease (CVD), type 2 diabetes (T2D), and other chronic metabolic conditions. Although inflammation can be a consequence of the biological characteristics of an individual, it is chronically (1) and acutely (during the postprandial phase) (2, 3) affected by diet. However, current dietary approaches to lower cardiometabolic risk are not specifically aimed at reducing inflammation. Many of the observed chronic effects of dietary carbohydrates and fats on cardiometabolic disease (4) are underpinned by postprandial excursions in glucose and triglycerides (TGs). These include effects on oxidative stress [generation of reactive oxygen species (ROS)], hemostatic function, lipoprotein remodeling (5,6), and endotoxemia (7), which trigger an inflammatory response. Although an acute inflammatory response is a physiological defense mechanism, a continually activated response from abnormal postprandial metabolic excursions may result in persistent low-grade inflammation and increased risk of cardiometabolic diseases. Given the significant time spent in the postprandial state (up to ∼18 h/d), attenuating an individual's postprandial inflammatory response may provide a dietary target in cardiometabolic disease prevention.
Several small human studies (n = 6-86 participants) have evaluated the effects of food given as single meals (high-fat, highcarbohydrate, or mixed meals) on postprandial inflammatory responses (2,3). However, these reports are inconsistent, which may be attributable to the inflammatory mediators investigated [typically TNF-α, IL-8, IL-6, and C-reactive protein (CRP)]. IL-6 is the only inflammatory marker that has been shown to consistently change postprandially, which may be a consequence of its tissue synthesis and time scale of initiation compared with other markers (3). It is also unclear whether different features of the postprandial metabolic response (e.g., response duration compared with peak concentration) differentially affect postprandial inflammation. Moreover, the mechanisms linking metabolic responses and postprandial inflammation have not been explored in the context of wider biological determinants.
PREDICT (Personalised REsponses to DIetary Composition Trial) 1 (n = 1002) aimed to predict individual variations in postprandial TG and glucose responses to standardized meals in a tightly controlled setting and with multiple postprandial assessments. In this article, we characterize postprandial inflammatory responses using traditional (IL-6) and emerging (GlycA) biomarkers of inflammation and dissect their determinants with a focus on postprandial glycemia and lipemia.

The PREDICT 1 study
The PREDICT 1 clinical trial (NCT03479866) aimed to quantify and predict individual variations in metabolic responses to standardized and free-living meals according to the full protocol, published elsewhere (17). Briefly, the PREDICT 1 study was a single-arm, single-blinded dietary study conducted between June 2018 and May 2019 ( Figure 1); 1002 generally healthy adults, aged 18-65 y, from the United Kingdom {nontwins, and identical [monozygotic (MZ)] and nonidentical [dizygotic (DZ)] twins} were enrolled into the study (17) and completed baseline clinic measurements [see the Consolidated Standards of Reporting Trials (CONSORT) flow diagram in Supplemental Figure 1]. The study consisted of a 1-d clinical visit (day 1) at baseline followed by a 13-d at-home period. Primary outcomes are reported elsewhere (18,19) and include gut microbiome profile, blood lipids and glucose, sleep, physical activity, and hunger and appetite assessment. Data for the secondary outcomes of inflammation (IL-6 and GlycA) measured at the baseline visit only are reported in this article. At baseline (day 1), participants arrived fasted and were given a standardized metabolic challenge meal for breakfast (0 h; 86 g carbohydrate, 53 g fat, 16 g protein, as a muffin and milkshake) and a test lunch (4 h; 71 g carbohydrate, 22 g fat, 10 g protein, as a muffin). The fat was high-oleic sunflower oil: 85% oleic acid (18:1n-9) and 8% linoleic acid (18:2n-6). Fasting and postprandial (0-6 h) venous blood was collected to determine concentrations of glucose, TG, IL-6, and GlycA (NMR). Stool samples, anthropometry, and a questionnaire for habitual diet were also obtained. Habitual diet information was collected using the European Prospective Investigation into Cancer and Nutrition (EPIC) FFQ (20) to capture average intakes in the past year. The recruitment criteria, outcome variables, and sample collection and analysis procedures relevant to this article are described elsewhere (17). The trial was approved in the United Kingdom by the Research Ethics Committee and Integrated Research Application System (IRAS 236407) and was run in accordance with the Declaration of Helsinki and Good Clinical Practice.

Exposure measurements
The methods for biochemical and microbiome analyses are described in full elsewhere (17). Briefly, participants were cannulated, and venous blood was collected at fasting (before the test breakfast) and at 9 postprandial time points (15, 30,  180, 240, 270, 300, and 360 min). Plasma glucose and insulin were measured at all time points and serum TG was measured at hourly intervals. Fasting samples were analyzed for lipid profile, thyroid-stimulating hormone, alanine aminotransferase, liver function panel, and full blood count. Assays were performed by Affinity Biomarkers Labs.
Visceral fat mass (VFM) was measured using DXA-based visceral fat measurements. DNA for 16S ribosomal RNA (rRNA) sequencing was isolated by QIAGEN Genomic Services using DNeasy ® 96 PowerSoil ® Pro. Optical density measurement was done using spectrophotometer quantification (Tecan Infinite 200). For 16S sequencing, the V4 hyper-variable region of the 16S rRNA gene was then amplified at Genomescan. Libraries were sequenced for 300-bp paired-end reads using the Illumina NovaSeq6000 platform. In total, 9.6 Pbp were generated and raw reads were rarefied to 360k reads/sample.
Six features of the glycemic and lipemic postprandial response were analyzed to examine different traits of the postprandial response. Traditionally studies have relied on 0-2 h glucose incremental AUC (iAUC) or 0-6 h TG iAUC when examining glycemic and lipemic associations with disease; however, the importance of the shape of the curve and different features of response for different pathophysiologies is gaining recognition (21). We therefore selected time points to reflect the "total" postprandial response (iAUCs), the peak response (Cmax), the rate of absorption/digestion (Tmax), and the most dynamic part of the curves (for glucose the 30-and 60-min rise from fasting and for TG the 4-and 6-h rise from fasting). The following variables were therefore used in the analysis: for glycemia: 2-h iAUC (Glu 2hiAUC ), 1-h iAUC (Glu 1hiAUC ), fasting (Glu fasting ), 1-h rise from fasting (Glu 1h-rise ), 30-min rise from fasting (Glu 30min-rise ), and maximum concentration in the first peak at 0-2 h (after breakfast, 1 st Glu max ) and second peak at 4-6 h (after lunch, 2 nd Glu max ); for lipemia: 6-h iAUC (TG 6hiAUC ), fasting (TG fasting ), 6-h rise from fasting (TG 6h-rise ), 0-6 h maximum concentration (TG max ), 4-h concentration (TG 4h ), and 6-h concentration (TG 6h ).

Outcome measurements
GlycA and IL-6 concentrations were quantified at 3 time points from venous blood at fasting, 4 h postprandial, and 6 h postprandial. GlycA was measured using a high-throughput NMR metabolomics (Nightingale Health) 2016 panel, with a CV of 1.1% (22). Details of the experimentation and epidemiologic applications of the NMR metabolomics platform have been reviewed previously (22,23). IL-6 was measured by Affinity Biomarkers Lab using a Sandwich Immunoassay by Meso Scale Diagnostics, with an intra-assay CV of 4% within the same run.

Statistical analysis
Values are expressed as either mean ± SD or median [IQR]. The ln of IL-6 for all 3 time points (+1) was calculated to normalize data distributions. Pearson's r was used to determine the relation of different features of the glycemic and lipemic response with both GlycA and IL-6 at all time points (fasting, 4 h, and 6 h). Interindividual variability for each outcome (IL-6 and GlycA) was assessed using Levene's test of variance heterogeneity. TG max and Glu max were used as independent factors in multivariable linear regression models to predict GlycA at 6 h postprandial. Multicollinearity was assessed by evaluating variance inflation factors at each step (considered high when >10) (24). To determine which features of the glycemic and lipemic responses (separately, n = 6 for both; combined, n = 12) best predicted GlycA (fasting, 4 h, 6 h, 6-h rise), a random forest (RF) model was applied with cross-validation (25). This method fits a large number of classification trees to a data set, then combines the predictions from all trees to present a final predictive model that ranks variables by their predictive power. However, this model does not provide mechanistic insight and may mask variable interaction and nonlinearity. For the evaluation of our models we have used R 2 and Q 2 (an estimate of the predictive ability of the model calculated by cross-validation). A negative Q 2 means the model is not at all predictive.
Receiver operating characteristic (ROC) curves were constructed and the AUC and its 95% CI were calculated to assess the discriminatory power and incremental ability of 1 st Glu max compared with 2 nd Glu max , 1 st Glu max compared with TG max , and 2 nd Glu max compared with TG max to detect inflammatory risk (GlycA of 70% was applied as a cutoff). Values of AUC range from 0.5 to 1, with 0.5 indicating no discrimination and 1 indicating perfect discrimination (26). The postprandial increase in inflammatory markers was defined as the 6-h concentration minus the fasting concentration. The 6-h rise was selected owing to the kinetics of the inflammatory response previously reported (3) and the 6-h limit of blood sampling in the current study. The atherosclerotic cardiovascular disease (ASCVD) score was calculated as described previously (27). Repeated-measures ANOVA was used to derive the interaction term between our outcomes (GlycA and IL6) and age and sex. A P value < 0.05 was considered statistically significant. Statistical analysis was performed in R Environment for Statistical Computing version 3.5.1 (R Foundation for Statistical Computing; https://www.R-pr oject.org/).

Mendelian randomization
The Supplemental Methods describe the full methods for this section. Briefly, 2-sample Mendelian randomization (MR) was undertaken using summary statistics from published genomewide association studies (GWASs) to model the causal relations between visceral fat, fasting TG, and fasting glucose (exposures) and fasting GlycA (outcomes) according to published methods (28). The Supplemental Methods report specific information on the genetic instrumental variables [single nucleotide polymorphisms (SNPs)] used here, sensitivity analyses, and other statistical methods.

Participant characteristics
A total of 1002 generally healthy adults from the United Kingdom [including nontwins and identical (MZ; n = 183 pairs) and nonidentical (DZ; n = 47 pairs) twins] completed baseline clinical measurements, dietary assessment, and the sequential test meal challenge [CONSORT diagrams and eligibility criteria are described elsewhere (16)]. Table 1 summarizes descriptive characteristics of the study participants. Participants were aged between 18.5 and 65.9 y (mean: 45.5 ± 11.8 y) with a mean BMI of 25.6 ± 5.0 kg/m 2 .
There was a modest correlation between fasting IL-6 and fasting GlycA (r = 0.349, P < 0.001) and postprandial GlycA at 4 h (r = 0.309, P < 0.001) and 6 h (r = 0.275, P < 0.001). However, there was no significant correlation between postprandial IL-6 (4 h and 6 h) and GlycA (fasting or postprandial at 4 h and 6 h; all P > 0.326) (Supplemental Table 3), highlighting the different patterns of response and potentially divergent food-induced mechanisms of initiation for these 2 inflammatory factors.
Postprandial lipemic and glycemic responses measured during the corresponding postprandial phase [reported previously (18)] were also highly variable between individuals (Levene's test; P < 0.001 for all) ( Figure 2C

Relation between features of postprandial lipemia, postprandial glycemia, and inflammatory response
To investigate which traits of the lipemic and glycemic responses (e.g., iAUC or Cmax) were most closely correlated with the postprandial inflammatory response ( Figure 2E, F), we selected key postprandial features to reflect different pathophysiological parameters of the postprandial response, as outlined in the Methods. Weak correlations (unadjusted) were observed between fasting IL-6 and fasting TG (r = 0.155) and glucose (r = 0.160; both P < 0.001). However, postprandial IL-6 (4 h and 6 h) was not significantly correlated with any features of the postprandial glucose or TG responses (all P > 0.095). Fasting GlycA was correlated strongly with fasting TG (r = 0.751) and weakly with glucose (r = 0.293), as well as moderately correlated with postprandial TG at4h (r = 0.654) and weakly with 1 st Glu max (r = 0.217) measures (all P < 0.001). Postprandial GlycA (6 h) was also highly correlated with postprandial TG (TG max : r = 0.832; TG at6h : r = 0.816), and weakly correlated with glucose (1 st Glu max : r = 0.239; all P < 0.001). Further, the 6-h rise in GlycA was strongly correlated with postprandial TG (TG 6h rise : r = 0.884; TG max : r = 0.748; TG at6h : r = 0.830) and weakly correlated with glucose (1 st Glu max : r = 0.152; all P < 0.001).

Predicting postprandial inflammatory responses after a mixed meal
Owing to its strong association with multiple features of lipemia and glycemia (compared with IL-6), GlycA was used as the inflammatory biomarker for the remaining analyses. Accordingly, we used machine learning to assess which features of the lipemic and glycemic responses influenced the postprandial inflammatory GlycA response. Input features (n = 13) were, for glycemia (n = 7): Glu 2hiAUC , Glu 1hiAUC , Glu 1h rise , Glu 30min rise , 1 st Glu max , 2 nd Glu max , and Glu fasting ; and for lipemia (n = 6): TG 6hiAUC , TG 6h rise , TG max , TG fasting , TG at4h , and TG at6h . Machine learning (RF) revealed that lipemic features were stronger predictors of the postprandial GlycA response than glycemic features, and that GlycA concentration was mainly determined by the corresponding TG feature; e.g., fasting GlycA concentration was mainly determined by the fasting TG concentration (total model: R 2 = 0.56 and Q 2 = 0.54). Moreover, the same was found postprandially: TG at4h for GlycA at 4 h (total model: R 2 = 0.73 and Q 2 = 0.74); TG at6h for GlycA at 6 h (total model: R 2 = 0.80 and Q 2 = 0.80); and TG 6h rise for 6-h rise in GlycA (total model: R 2 = 0.72 and Q 2 = 0.76). The performance and generalizability of the machine learning model were higher for postprandial GlycA concentrations than for fasting GlycA, showing that we can predict the highly variable postprandial GlycA concentrations with greater accuracy than fasting values.

Independent predictors of the postprandial inflammatory response
The postprandial phase is highly dynamic, involving a complex interplay between simultaneous postprandial fluxes in glucose and TG from mixed meals. Therefore, to disentangle the effect of the glucose and TG responses, we evaluated the independent role of lipemia and glycemia [TG max , and glucose peaks (1 st Glu max compared with 2 nd Glu max )] in predicting 6-h GlycA concentration, using multiple regression analysis (Supplemental Figure 2A-C, Supplemental Table 4). Owing to the sequential meal study design and overlapping glucose and TG peaks (as occurs in real life), we also dissected the contribution of the first glucose peak (1 st Glu max ; between fasting and 2 h), second glucose peak (2 nd Glu max ; between 4 h and 6 h), and TG peak (TG max ; between 0 and 6 h) in independently predicting the 6-h GlycA concentrations. In a model including TG max and 1 st Glu max , both features were significantly and independently associated with 6-h GlycA (R 2 = 0.70). In a second model including only glucose (1 st Glu max and 2 nd Glu max ) (R 2 = 0.057), just the 1 st Glu max was a significant and independent predictor of 6-h GlycA, suggesting a delayed initiation of GlycA by glucose. In a third model with the coinciding 2 nd Glu max and TG max , both TG and glucose independently and significantly predicted 6-h GlycA (R 2 = 0.69). Finally, ROC-AUC analysis revealed that the TG max was more informative of postprandial 6-h GlycA than both the first and second glucose peaks (Supplemental Figure 2A-C).

MR
MR analysis was then undertaken to obtain unconfounded estimates of the causal association of genetically determined exposures (VFM, TG, and glucose) and the outcome (GlycA). Fasting GlycA, TG, and glucose were used owing to an absence of postprandial GWAS data; however, our SEM revealed that fasting GlycA was the strongest predictor of postprandial GlycA and we propose that the effect of fasting TG and fasting glucose on postprandial GlycA is mediated in part by fasting GlycA. All instrumental variables (SNPs) had F statistics >20 (31). Supplemental Table 7 presents the results, expressed as the β-coefficients for visceral fat and fasting TG and glucose concentrations per 1-SD increase in fasting GlycA. Individuals with genetically higher visceral fat [inverse-variance weighted (IVW): β = 0.217, P = 6.634e-09; Figure 3B] and fasting TG concentrations (IVW: β = 0.494, P = 6.915e-26; Figure 3C) had a significantly greater fasting GlycA, whereas fasting glucose had no significant causal effect on fasting GlycA (IVW: β = −0.132, P = 0.294; Figure 3D). Supplemental Table 7 also shows heterogeneity results and horizontal pleiotropy (where a variant has an effect on other traits outside of the pathway of the GlycA) bias. Estimations based on both MR Egger and IVW for visceral fat were >0.05, indicating a low chance of heterogeneity (IVW: P = 0.109; MR Egger: P = 0.108). Whereas, there was heterogeneity for the impact of fasting TG (IVW: P = 2.958e−05) and fasting glucose (IVW: P = 2.382e−09) on fasting GlycA. The results of the Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) did not indicate any outliers for all the estimates. The horizontal pleiotropy test, with negligible Egger regression intercept, also indicated a low likelihood of pleiotropy for all of our estimations (all P > 0.335). The results of the MR robust adjusted profile score (MR-RAPS) were identical with the IVW estimates, highlighting again a low likelihood of pleiotropy (Supplemental Table 7). The results of the leave-one-out method demonstrated that the links were not driven by single SNPs.

Network MR
To evaluate the extent to which fasting TG mediates the effect of visceral fat on GlycA variation we performed network MR ( Figure 3E). These analyses revealed that the effect of visceral fat on fasting TG variation was statistically significant (IVW: β = 0.127, SE = 0.03, P = 0.000024), whereas the effect of fasting TG on visceral fat variation was not (IVW: β = −0.052, SE = 0.02, P = 0.057), suggesting a predominantly unidirectional causal relation. The effect of a 1-SD increase in visceral fat on fasting GlycA was 0.22 (standardized-β). The effect of a 1-SD increase in visceral fat on fasting TG was 0.13 (standardized-β) and the effect of a 1-unit increase in fasting TG on fasting GlycA was 0.49 (standardized-β). Thus, the mediated effect of fasting TG was 0.13 × 0.50 = 0.065. The mediated proportion was 0.065/0.22 = 29%, suggesting therefore that ∼29% of the effect of visceral fat on inflammation is mediated by the metabolic processes that underlie fasting TG.

Discussion
The postprandial inflammatory response may affect the pathophysiology of many chronic diseases. In the largest postprandial inflammation study to date, we 1) evaluated the association of different features of the food-induced TG and glucose responses with postprandial inflammation, 2) determined the main predictors of postprandial inflammation, and 3) explored potential mechanisms underlying postprandial changes in inflammation. Postprandial lipemia was a stronger predictor of postprandial inflammation (measured by GlycA) than was postprandial glycemia. Visceral fat, partly mediated by fasting TG, was a key causal determinant of this postprandial inflammatory response, supporting current evidence that management of obesity and TG concentrations (via lifestyle or drugs) will reduce chronic inflammatory burden, a key factor in the pathogenesis of cardiometabolic diseases.
Several small-scale human studies have evaluated the effects of high-fat (3) or high-carbohydrate (32) meals on inflammatory responses. However, there has been no consensus regarding the timing, magnitude, and mechanism underpinning postprandially stimulated inflammation or the most relevant food-induced inflammatory biomarkers. Further, there is little agreement on the impact of postprandial glycemia on postprandial inflammation (33), despite a well-established relation between fasting hyperglycemia and low-grade inflammation (34). Although we observed a large increase (of 169%) in 6-h IL-6, there was no relation between postprandial IL-6 (4 h and 6 h) and any features of the postprandial glucose or TG response. One possible explanation for this is that postprandial increases in IL-6 may be a consequence of cannulation, which has been previously shown to elicit similar levels of increase in acute inflammation when no meal is consumed (35), rather than the direct effects of the meal.
We observed a modest correlation between fasting GlycA and IL-6, in accordance with previous studies (15,36), but no correlation between postprandial IL-6 and GlycA, highlighting the divergent mechanisms of postprandial initiation for these 2 inflammatory markers (or the aforementioned impact of cannulation). Indeed, unlike IL-6, GlycA concentrations reflect a composite measure of systemic inflammation (8,9), to which IL-6, CRP, fibrinogen, and cytokines contribute only negligibly, as to the GlycA signal from NMR (37).
Although the mean postprandial increase in plasma GlycA concentration (4.5%) was small in comparison with IL-6 (169%), when only the participants eliciting an increase in GlycA were examined (60% of the total) the increase was 10.5% (0.14 mmol/L). Given that even small changes in GlycA are associated with risk of disease and mortality (15,29,30), these findings highlight a potentially important postprandial inflammatory measure which may have clinical relevance. For example, risk (HR) of future cardiovascular events was 1.31 per 1-SD increment (0.24 mmol/L) in GlycA (after multiple adjustments) (38). In addition, in our trial, participants with the largest postprandial increase in GlycA (>90th percentile; 30% increase from fasting: 1.48 to 1.91 mmol/L), had a 2-fold greater ASCVD risk score (0.034) than the rest of the cohort (0.019; P = 0.012); suggesting that individuals that elicit higher postprandial inflammatory responses may be at greater CVD risk.
A strength of the current study is the use of a mixednutrient sequential test meal challenge, representing real-life eating patterns. Previous studies have typically focused on single components of the postprandial response-lipemia or glycemia-using supraphysiological and single challenge meals. However, we typically consume multiple mixed-nutrient meals throughout the day which elicit an interrelated lipemic and glycemic response. Indeed, previous studies have shown that the type of fat (39), macronutrient distribution (40,41), and overall nutrient density (42,43) of the meal can alter the postprandial inflammatory response.
Owing to the test meal design of our study, we demonstrated, to our knowledge for the first time, that postprandial glycemia and lipemia have an independent and cumulative association with GlycA. This may be mediated by the production of ROS, which may reach higher concentrations during co-occurrence of lipemia and glycemia (44). Our analysis (including multiple regression and ROC curves) suggests that during the co-occurrence of the second glucose peak and peak lipemia, lipemia is a stronger determinant of the inflammatory response than is glycemia. Owing to the 6-h duration of the study, we were unable to determine the kinetics of the lipemic initiation of inflammation; however, the strong correlation between 6-h TG and GlycA is suggestive of a rapid inflammatory response to dietary fat, as previously reported (45).
Although our results demonstrate the key role of lipemia and adiposity in food-induced inflammation, it is important to also target postprandial elevations in glucose in concert with TG, owing to the interacting and overlapping metabolic pathways regulating glucose and TG (46,47) and the association of fasting and postprandial TG with carbohydrate intake and fasting and postprandial circulating glucose (3,48). Potential strategies to attenuate fasting and postprandial TG include (49) consumption of low-glycemic foods, high-dose marine omega-3, fiber, and low alcohol intakes alongside lifestyle modifications including exercise in the 12 h preceding meals and consuming larger meals earlier in the day. Strategies to ameliorate the postprandially induced oxidative/inflammatory pathways, including coadministration of polyphenol-/antioxidant-rich foods (50), may also beneficially attenuate postprandial low-grade inflammation.
There is a growing awareness that single measures of postprandial responses typically used (e.g., 0-2 h glucose iAUC and 0-6 h TG iAUC) are an oversimplification and do not reflect different postprandial pathophysiologies. However, postprandial measurements are burdensome for investigators and participants. Therefore, an aim of the current study was to assess the contribution of the different features of the postprandial response in relation to food-induced inflammation to inform future studies and the potential clinical utility of fat-tolerance tests or oralglucose-tolerance tests to reduce testing burden and to enable the implementation of targeting strategies. Our results demonstrate that the 6-h concentration in GlycA and its 6-h rise from fasting are suitable target measures to determine postprandial inflammation and to discriminate an individual's inflammatory response after sequential meals.
To dissect the key determinants of an individual's postprandial inflammatory response beyond lipemia and glycemia, we applied machine learning, SEM, and MR, exploiting our dense phenotypic data. Participants with greater visceral fat had a bigger (causal) inflammatory response, partly mediated by fasting TG. In accordance with our results, obese individuals have been shown to elicit higher postprandial inflammatory responses (albeit in IL-6) than lean individuals (31).

Study limitations
The study duration (6) did not allow us to explore the full kinetics of the TG-induced inflammatory response; however, sequential blood sampling for >6 h is challenging for researchers and burdensome for participants. Future studies would benefit from assessing the impact of single compared with sequential test meals, different doses of fat and carbohydrates, as well as different fatty acids and dietary sources to dissect the effect of individual foods, with their inherent nutrient-matrix complexity, on food-induced inflammation. The adjacent GlycA and TG signals on the NMR platform may have affected our observed associations owing to the possibility of overlapping peaks. As previously discussed, the lack of GWAS data on postprandial GlycA, TG, and glucose is a limitation of our MR; however, given that the SEM revealed that fasting GlycA had a strong effect on postprandial GlycA, the results still highlight important relations between postprandial metabolism and postprandial GlycA. Our study also had a limited ethnic diversity, which may be relevant for inflammation (51,52) and warrants further investigation. In addition, future studies would benefit from measuring the full breadth of inflammatory biomarkers.

Conclusions
Postprandial inflammation is largely driven by acute elevations in circulating TG. We identified GlycA after mixed meals as a promising candidate biomarker for assessing the food-induced inflammatory response within typical dietary habits. The large interindividual variability in postprandial inflammation, partly mediated by adiposity, highlights the potential for personalized strategies to target obesity and postprandial metabolic responses to reduce low-grade inflammation in preventative health.
We thank Inbar Linenberg for feedback on the manuscript. We thank the staff of Zoe Global and the Department of Twin Research for their tireless work in contributing to the running of the study and data collection. We thank Peter Wurtz, Noora Kanerva, and Gergana Gateva from Nightingale Health for their advice on GlycA.
The authors' responsibilities were as follows-JW, GH, and TDS: obtained funding; SEB, AMV, JW, N Segata, PWF, and TDS: designed the study and developed the concept; SEB, N Segata, JW, and TDS: collected the data; SEB, N Segata, MM, AMV, RK, PWF, and JCP: analyzed the data; SEB, JW, N Segata, and TDS: coordinated the study; SEB, MM, WLH, N Sattar, AMV, JMO, TDS, and PWF: wrote the manuscript; and all authors: read, reviewed, revised, and approved the final manuscript. TDS, SEB, AMV, PWF, N Sattar, and JMO are consultants to Zoe Global Ltd ("Zoe"). JW, GH, and JCP are or have been employees of Zoe. All other authors report no conflicts of interest.

Data Availability
Data described in the article, code book, and analytic code are held with the Department of Twin Research at King's College London and will be made available using our normal procedures overseen by the Wellcome Trust and its guidelines as part of our core funding. The application is at: https://twinsuk.ac.uk/resour ces-for-researchers/access-our-data/.