Diet is one of the most influential and accessible modulators of the gut microbiome, the human intestine’s vast and diverse microbial ecosystem increasingly recognised as a key player in the development of obesity, type 2 diabetes, CVD and cancer(Reference Cotillard, Kennedy and Kong1–Reference Daniel and McQuade6). The composition and collective function of the gut community affects how and what the human host is able to extract from the diet by providing the machinery that converts dietary content into biological signals with profound systemic effects on host health(Reference Flint, Scott, Louis and Duncan7).
Not unlike the complex and interactive nature of the microbiome, dietary habits are multidimensional with many interrelated components. Recently, Bowyer et al. and Maskarinec et al. demonstrated the utility of a priori dietary patterns or indices for capturing and controlling for variation in the gut microbiome due to the effects of participants’ diets(Reference Bowyer, Jackson and Pallister8–Reference De Filippis, Pellegrini and Vannini12). The Dietary Inflammatory Index (DII®) assesses the balance of pro- and anti-inflammatory dietary factors based on literature-derived associations between various dietary components and inflammatory biomarkers. This index, with its specific focus on inflammation, inherently differs from other index-based scores assessed in previous studies measuring adherence to established dietary guidelines or healthy eating patterns (e.g. the Healthy Eating Index or Mediterranean Diet Score)(Reference Shivappa, Steck and Hurley13).
Diet–microbiome interactions may be one of the most promising targets to reduce chronic inflammation, a key pathophysiological mechanism underlying diet’s influence on multiple chronic diseases through the common mechanism of the NF-κB pathway and a complex interaction of cellular, molecular and metabolic factors(Reference O’Keefe, Li and Lahti14–Reference Zitvogel, Pietrocola and Kroemer21). Given that diet can modulate inflammation through both pro-inflammatory and anti-inflammatory mechanisms, we hypothesised that the inflammatory potential of usual dietary habits would be associated with the overall composition and functional capacity of the gut microbiome. In the present study, we investigated total dietary inflammatory potential, as assessed by the literature-derived DII, in relation to gut microbiota diversity and composition. In a subset of individuals with fasting blood samples and metagenomic data, we further explored microbes and microbial gene pathways associated with DII score and circulating inflammatory markers.
Methods
Study population
The study population is composed of 101 cancer-free individuals, including patient volunteers initially approached in the colonoscopy clinic of MD Anderson Cancer Center (n 36) and community volunteers recruited from the medical centre community (n 65). Detailed study recruitment and eligibility criteria are provided in online Supplementary Fig. S1. Initially, from 2013 to 2016, a total of 132 patient volunteers with no history of cancer and 449 community volunteers expressed an interest in participating. Both subgroups were screened and interviewed by a clinical provider and/or trained study coordinator with regard to their medical history and medication use and asked to provide stool samples using the same collection protocol. Study ineligibility criteria included current smoker, antibiotic use within the past month, incident or prevalent cancer other than non-melanoma skin cancer, one or more chronic conditions that restricts dietary intake (e.g. coeliac disease), major intestinal surgery (e.g. gastric bypass), currently pregnant or lactating. BMI was calculated from measured weight (kg)/height (m)2 and categorised based on WHO criteria(Reference Seidell and Flegal22). Study subjects were included in the current analysis if they completed the dietary questionnaires, provided a stool sample and passed microbiome data quality filtering criteria. All procedures were reviewed and approved by The University of Texas MD Anderson Institutional Review Board. Informed consent was obtained from all research participants, and all methods were performed in accordance with relevant guidelines and regulations.
Dietary assessment
Participants completed dietary histories via one of two versions of the National Cancer Institute (NCI)-developed FFQ. For consistency with historic recruitment of patients in prior and ongoing MD Anderson studies, patient volunteers completed a modified version of the NCI Health Habits and History Questionnaire(Reference Melkonian, Daniel and Hildebrandt23,Reference Block, Coyle and Hartman24) , which queries the frequency of intake and portion size of 165 food and beverage items, including ethnic foods commonly consumed in the Texas region. Daily nutrient consumption was estimated using the US Department of Agriculture Food and Nutrient Database for Dietary Studies(Reference Venook, Papandreou and Furuse25). Community volunteers completed the NCI Dietary History Questionnaire II (DHQ II), a more recent web-based adaptation of the NCI Health Habits and History Questionnaire(Reference Venook, Papandreou and Furuse25), which queries 134 food items. NCI Dietary History Questionnaire II responses were processed using Diet*Calc software(Reference Subar, Midthune and Kulldorff26).
Energy-adjusted dietary inflammatory index score calculation
Food and nutrient intake derived from responses to FFQ was linked to the corresponding inflammatory effect scores designated in the DII to calculate the energy adjusted (E-DII) score for each participant(Reference Shivappa, Steck and Hurley13). The DII is a literature-derived, population-based dietary index designed to quantify the overall inflammatory potential of an individual’s entire diet. The details of the development and scoring algorithm are described elsewhere(Reference Shivappa, Steck and Hurley13). In short, approximately 2000 primary research articles published through 2010 which investigated the effects of forty-five different food parameters (mostly macronutrients, micronutrients, some bioactive components and individual food items such as garlic and tea) on six inflammatory markers (i.e. IL-1β, IL-4, IL-6, IL-10, TNF-α and C-reactive protein) were identified and scored to derive the component-specific inflammatory effect scores(Reference Shivappa, Steck and Hurley13). Thirty-one DII components (plus garlic in the NCI Health Habits and History Questionnaire) were used to calculate the instrument-specific, energy-adjusted E-DII score for analysis. Both FFQ lacked information on some less commonly consumed spices (ginger, saffron, turmeric, pepper, oregano, rosemary) and phenols/flavonoids (eugenol, flavan-3-ol, flavones, flavonols, flavonones, anthocyanidins, isoflavones). Food and nutrient consumption was first energy-adjusted per 4184 kJ (1000 kcal), and subsequently standardised for each component using mean and standard deviation data derived from a composite dietary database representing energy-adjusted intake from eleven populations around the world(Reference Shivappa, Steck and Hurley13). The energy-adjusted and standardised dietary intakes were then converted to centred proportions to account for skewness, multiplied by the inflammatory effect score for each available DII parameter and summed across all DII components to obtain the overall E-DII score(Reference Shivappa, Steck and Hurley13). Higher (i.e. more positive) E-DII scores represent more pro-inflammatory diets, while lower (i.e. more negative) E-DII scores indicate more anti-inflammatory diets. The E-DII score has been construct validated and consistently associated with elevated inflammatory biomarker levels such as IL-6(Reference Phillips, Shivappa and Hébert27), TNF-α (Reference Phillips, Shivappa and Hébert27) and C-reactive protein(Reference Shivappa, Wirth and Murphy28).
Stool and blood sample collection
All participants were provided the same stool sample collection kit with detailed instructions. Following defecation into a plastic ‘toilet hat’, gloved participants used a sterile BBL culture swab collection and transport system (Becton Dickinson) to collect a small portion of their sample. Samples were either Express (overnight or same day) shipped or brought to their next scheduled visit. All faecal samples were received within less than 48 h of collection, stored at –80°C and processed within 1 year of collection(Reference Sinha, Chen and Amir29). Residual fasting blood samples drawn at the colonoscopy clinic visit were available in a subset of thirty-four patient volunteers with FFQ and stool samples.
Microbiome characterisation
Total genomic DNA was extracted from faecal samples using the MoBio PowerSoil DNA Isolation Kit following the manufacturer’s instructions. The 16S rRNA gene was amplified using V4-targed primers (GGACTACHVGGGTWTCTAAT and GTGCCAGCMGCCGCGGTAA)(Reference Caporaso, Lauber and Walters30), and amplicons were sequenced using the MiSeq platform (Illumina). Raw FASTQ sequences were processed as previously described(Reference Hoffman, Hutchinson and Fowler31). Following quality filtering and chimera removal, sequence reads were clustered into operational taxonomic units using UPARSE(Reference Edgar32), with operational taxonomic units subsequently mapped to a V4-optimised version of the SILVA database (version 123) at 97 % similarity level. The remaining samples were rarefied to 2742 reads/sample, and relative abundances were calculated (online Supplemental Table S1). Basic Local Alignment Search Tool was used to identify the likely species represented by each operational taxonomic units (OTU) centroid sequence(Reference Zhang, Schwartz and Wagner33).
Whole-genome shotgun (WGS) sequencing was performed to comprehensively assess microbial genomic DNA of faecal samples in a subset of thirty-four patient volunteers who also had FFQ and residual fasting blood samples. WGS data provide species-level taxonomy of the gut microbiome, as well as the metabolic or functional gene content pathways represented within the microbial community. Individual libraries constructed from extracted total gDNA for each sample were pooled and sequenced via HiSeq 2000 (Illumina) using the 1 × 100 bp paired-end read protocol. Pooling resulted in a sequencing depth >3 Gb/sample. Quality filtering, trimming, demultiplexing and read mapping were carried out by an in-house pipeline described previously.(Reference Stewart, Ajami and O’Brien34) To determine metabolic pathway content of the entire metagenome, reads aligning to known orthologues in the Kyoto Encyclopedia of Genes and Genomes database were tabulated and pathways constructed by calculating the minimum set through MinPath45(Reference Kanehisa and Goto35).
Circulating inflammatory markers
Plasma leptin, TNF-α, IL-6, lipocalin 2, plasminogen activator inhibitor-1 (PAI-1), C-peptide and monocyte chemoattractant protein-1 were assessed via multiplex assay (Millipore). All samples were run in duplicate with internal standards (pooled cancer case plasma), healthy control samples (normal plasma) and kit quality controls to assure plate to plate consistency.
Statistical analyses
The E-DII score was categorised according to instrument-specific tertiles. Characteristics of study participants were described with medians and standard errors of the mean for continuous variables and frequencies and percentages for categorical variables. We examined the difference of categorical variables across E-DII tertiles by the χ 2 test and the difference of continuous variables by the Kruskal–Wallis non-parametric test. We also examined the associations of these covariates with bacterial α- and β-diversity to assess potential confounders for inclusion in the adjusted models of E-DII and microbiota associations.
α-Diversity was assessed by observed OTU, Shannon diversity index, Chao1 index and Simpson’s diversity index. Differences in α-diversity across E-DII tertiles among all subjects and by study subgroup were assessed via the Kruskal–Wallis non-parametric test. We also examined associations of the continuous E-DII score with α-diversity in each subgroup using linear regression models. None of the potential confounders examined, including age (continuous), sex (males and females), BMI (continuous), medication use (yes/no) and study subgroup (patient v. community volunteers), was associated with α-diversity or appreciably changed the model estimates and thus was not included in the final adjusted model.
Similarly, Bray–Curtis dissimilarity, Weighted Unifrac, Unweighted Unifrac and Jaccard (β-diversity metrics) between the highest v. the lowest E-DII tertile were assessed via permutational multivariate ANOVA among all the subjects and by study subgroup(Reference Morgan and Huttenhower36). While none of the measured covariates were clearly associated with β-diversity, we further examined whether BMI-status modified associations of β-diversity with E-DII in stratified analysis.
Differentially abundant bacterial taxa by E-DII tertiles were assessed using linear discriminant analysis effect size (LEfSe) under Galaxy environment(Reference Segata, Izard and Waldron37), applying the one-against-all strategy with a logarithmic linear discriminant analysis score threshold of 3 and α of 0·1 for factorial Kruskal–Wallis test among classes. Analysis was restricted to bacteria present in ≥20 % of the study population. We further assessed the associations between differentially abundant candidate taxa identified from LEfSe and potential confounders as mentioned above. Given some taxa were associated with BMI, we subsequently evaluated associations of LEfSe-identified taxa in association with E-DII tertiles via a negative binomial regression model adjusted for age and BMI.
Due to the large number of low abundance species in metagenomic data, we employed the least absolute shrinkage and selection operator (LASSO) method to identify bacterial species and microbial gene content pathways associated with E-DII among thirty-four patient volunteers with residual fasting blood samples using the glmnet package in R(Reference Friedman, Hastie and Tibshirani38). Spearman’s rank correlation method was subsequently used to estimate the correlations of selected species and functional pathways based on LASSO and 16S LEfSe analyses with seven circulating inflammatory biomarkers including leptin, TNF-α, IL-6, lipocalin 2, PAI-1, C-peptide and chemoattractant protein-1. The Benjamini–Hochberg (B-H) method was used to adjust P values in the multiple correlation analyses while controlling for the expected false discovery rate at 0·05.
Sensitivity analysis was performed by rerunning all the analyses described above after removing one subject identified as an outlier based on his/her Shannon index value, defined as 1·5 interquartile range below the 25th or above the 75th percentile of the population’s value. All analyses were performed in SAS, R, Galaxy(Reference Segata, Izard and Waldron37), Agile Toolkit for Incisive Microbial Analyses(Reference Paradis, Claude and Strimmer39). All tests were two-sided, with P values <0·05 considered statistically significant unless otherwise noted.
Results
Characteristics of participants
The median E-DII score was 1·79 (range –5·15 to –3·08) for patient volunteers and –0·85 (range –4·04 to 2·83) for community volunteers. Subjects with higher E-DII scores (i.e. more pro-inflammatory diets) tended to have higher BMI (Table 1). However, there was no statistically significant difference across E-DII levels in the distribution of age, sex, history of precancerous colorectal polyps and medication use. When we examined these factors in relation to overall microbial diversity, none of the factors were strongly associated with α-diversity (all P ≥ 0·15) or β-diversity (all rho ≤ 0·05; data not shown).
Table 1. Characteristics of participants by energy-adjusted dietary inflammatory index (E-DII) score (n 101)
(Medians and standard errors; numbers and percentages)
OTU, operational taxonomic units.
* E-DII values across the three levels were calculated from FFQ-specific E-DII scores and categorised into tertiles based on the distributions in each study subgroup.
† The difference of categorical variable across E-DII tertiles was tested by the χ 2 test, and difference of continuous variables was tested using the Kruskal–Wallis non-parametric test as they are not normally distributed.
Associations of energy-adjusted dietary inflammatory index score with microbial diversity
α-Diversity (within sample diversity), a measure of microbial richness and/or evenness, did not differ across E-DII levels among all participants in the crude analysis (Shannon index: P = 0·63 (Fig. 1(A)); other metrics (Table 1)). Null results were similarly consistent within study subgroup (Fig. 1(B) for Shannon Index; data not shown for other metrics). Null results also were observed when examining continuous associations of E-DII and α-diversity by subgroup (data not shown).
Fig. 1. Microbial α-diversity, as assessed by the Shannon index, across energy-adjusted dietary inflammatory index (E-DII) tertiles among (A) all study subjects (n 101) and (B) by study subgroup (n 65 community volunteers, n 36 patient volunteers). EDII level 
, a (most anti-inflammatory diet); 
, b; 
, c (most pro-inflammatory diet).
We examined several β-diversity (between sample diversity) metrics, representing different measures of dissimilarity or distance between groups by highest v. lowest E-DII level. We observed modest visual differences in overall community composition (Fig. 2 and online Supplementary Fig. S2). Although the results of Unweighted Unifrac reached statistical significance, E-DII clusters, comparing the highest (pro-inflammatory) v. lowest (anti-inflammatory) levels, were not overly distinct (P = 0·03, rho=0·02; online Supplementary Fig. S2). Similar results were observed when we examined E-DII and β-diversity associations stratified by BMI status.
Fig. 2. Microbial community differences, as assessed by Bray–Curtis dissimilarity, between individuals with the most anti-inflammatory v. pro-inflammatory diet among (A) all study subjects and (B) by study subgroup. Energy-adjusted dietary inflammatory index (E-DII) level 
, a (most anti-inflammatory diet); 
, c (most pro-inflammatory diet).
Differentially abundant taxa by energy-adjusted dietary inflammatory index score
We identified several differentially abundant taxa in the LEfSe analysis across the three levels of E-DII score (Fig. 3). Ruminococcus torques, Eubacterium nodatum, Acidaminococcus intestini and Clostridium leptum were more abundant in subjects consuming the most pro-inflammatory diets, while Akkermansia muciniphila was enriched in subjects with the most anti-inflammatory diets. With adjustment for age and BMI in the negative binomial model, E-DII associations with these five taxa did not appreciably change compared with crude model (Table 2). R. torques, E. nodatum and A. intestini remained significantly associated with a more pro-inflammatory diet.
Fig. 3. Differentially abundant taxa across energy-adjusted dietary inflammatory index (E-DII) tertiles using the linear discriminant analysis (LDA) effect sizes approach among 101 subjects. 
, E-DII group 1 (most anti-inflammatory diet); 
, E-DII group 2; 
, E-DII group 1 (most pro-inflammatory diet).
Table 2. Crude and age- and BMI-adjusted associations of energy-adjusted dietary inflammatory index (E-DII) with five candidate taxa* (n 101)
(e β and 95 % confidence intervals)
ref, Reference; LEfSe, linear discriminant analysis effect sizes.
* Computed for the association between E-DII and five LEfSe selected taxa; the significant P value indicates E-DII is significantly associated with abundance of the taxa.
† Negative binomial model with log link was used to estimate e β and 95 % CI. Magnitude indicates how much more abundant the taxa (treated as count), for example, three times more abundant in the highest, as compared with the lowest E-DII tertile.
Associations of energy-adjusted dietary inflammatory index with microbial species, pathways and circulating inflammatory markers
Although the E-DII has been construct validated in several large studies(Reference Tabung, Steck and Zhang40–Reference Shivappa, Steck and Hurley42), no significant associations were observed with inflammatory markers among the subset of subjects with fasting blood samples available, but all were in the expected direction (n 34; online Supplementary Table S2). In the correlation analysis of microbial species and functional pathways with inflammatory biomarkers (n 34), Luteimonas mephitis, a low abundance species identified via LASSO based on its non-zero negative correlation with E-DII, was found to be inversely related to PAI-1 (rho = –0·40, P = 0·02; B-H adjusted P = 0·24). A. intestini, the taxa associated with the most pro-inflammatory diet in the overall (16S) analysis, was positively related to PAI-1 (rho = 0·40, P = 0·02; B-H adjusted P = 0·24) in the WGS subset. Another LASSO-selected Lachnospiraceae bacterium strain associated with a more pro-inflammatory E-DII was also positively correlated with C-peptide (rho = 0·41, P = 0·02; B-H adjusted P = 0·24). No other taxa selected from LEfSe in the larger 16S analysis, or LASSO in the WGS subset analysis were associated with circulating markers (all P values > 0·05; Fig. 4). Polyketide sugar unit biosynthesis was significantly positively correlated with E-DII (rho = 0·32, P = 0·03; B-H adjusted P = 0·50). Secondary bile acid biosynthesis was positively correlated with IL-6 (rho = 0·35, P = 0·04; B-H adjusted P = 0·50), but appeared to be modestly inversely correlated with E-DII scores favouring a more anti-inflammatory diet. Mammalian AMP-activated protein kinase (AMPK) signalling and carbon metabolism pathways, which were each modestly correlated with E-DII scores favouring a more anti-inflammatory diet, were inversely correlated with C-peptide (rho = –0·36 for AMPK and rho = –0·40 for carbon metabolism, respectively; both P values <0·05; B-H adjusted P ≥ 0·50; Fig. 5). Of note, none of these associations were statistically significant following correction for multiple testing.
Fig. 4. Correlation heatmap of energy-adjusted dietary inflammatory index (E-DII)-associated species and circulating markers among thirty-four subjects with residual fasting blood samples and whole-genome shotgun (WGS) sequencing of the gut microbiome. Acidaminococcus intestini, Akkermansia muciniphila, Ruminococcus torques and Holdemanella biformis were selected as differentially abundant operational taxonomic units (OTU) in the 16S analyses, while other species were selected using the least absolute shrinkage and selection operator (LASSO) method based on non-zero estimates of correlation with E-DII. ** Statistically significant Spearman correlations (P < 0·05) before correction for multiple testing (none of the correlations was significant after Benjamini–Hochberg adjustment). MCP-1, monocyte chemoattractant protein-1; PAI-1, plasminogen activator inhibitor-1.
Fig. 5. Correlation heatmap of energy-adjusted dietary inflammatory index (E-DII)-associated pathways and circulating markers among thirty-four subjects with residual fasting blood samples and whole-genome shotgun (WGS) sequencing of the gut microbiome. A total of seven WGS characterised pathways were selected using the least absolute shrinkage and selection operator (LASSO) method based on non-zero estimates of correlation with E-DII. ** Statistically significant Spearman correlations (P < 0·05) before correction for multiple testing (none of the correlations was significant after Benjamini–Hochberg adjustment). MCP-1, monocyte chemoattractant protein-1; PAI-1, plasminogen activator inhibitor-1.
Discussion
Dietary inflammatory potential as measured by the E-DII score was associated with differential composition of specific microbes, but not overall diversity of the gut microbiome in this cross-sectional sample of 101 cancer-free individuals. In the overall analysis, R. torques, E. nodatum, A. intestini and C. leptum were more abundant in subjects consuming the most pro-inflammatory diets, while A. muciniphila was enriched in subjects with the most anti-inflammatory diets. In analysis adjusted for both age and BMI, R. torques, E. nodatum and A. intestini remained significantly associated with a more pro-inflammatory diet. In an exploratory subset of individuals with fasting blood samples and metagenomic characterisation of the gut microbiome, pathways reflecting the functional capacity of the gut microbiome to support AMPK signalling, carbon metabolism, polyketide and secondary bile acid biosynthesis were associated with the inflammatory potential of the diet and/or systemic inflammation in the host.
We found several differentially abundant microbes by E-DII level. Four bacteria: R. torques, E. nodatum, A. intestini and C. leptum were associated with the most pro-inflammatory diet, and A. muciniphila was enriched in participants with the most anti-inflammatory diet. A. mucinphila and R. torques have previously been highlighted in a randomised cross-over intervention trial of the fermentable oligosaccharides, disaccharides, monosaccharides and polyols (FODMAP) diet among twenty-seven irritable bowel syndrome patients and six healthy subjects(Reference Halmos, Christophersen and Bird43). Results of this trial showed that low compared with moderate FODMAP intake was associated with a 5-fold reduction in A. mucinphila and a 1·5-fold increase in R. torques (Reference Halmos, Christophersen and Bird43). FODMAP substrates are fermented by gut bacteria to SCFA with anti-carcinogenic and anti-inflammatory actions(Reference O’keefe16). A. muciniphila is a producer of SCFA, primarily acetate and propionate, through mucin degradation. Mouse studies have demonstrated that a high-fat diet (i.e. pro-inflammatory diet) decreases the abundance of this species and support a causative role for this species in lowering adipose tissue inflammation and improving insulin sensitivity and lipid metabolism(Reference Schneeberger, Everard and Gómez-Valadés44). However, the role of A. muciniphila in modulating inflammation in humans remains unclear. Two studies reported no association of A. muciniphila with systemic inflammatory markers, including high sensitivity-C-reactive protein, IL-6 and lipopolysaccharides(Reference Dao, Everard and Aron-Wisnewsky45,Reference de Moraes, Fernandes and da Silva46) . In a recently published randomised, double-blind, placebo-controlled pilot study of thirty-two overweight/obese insulin-resistant volunteers, a 3-month administration of pasteurised A. muciniphila was found to significantly decrease two enzyme activities that are thought to be involved in modulating inflammation, that is, dipeptidyl peptidase 4 and γ-glutamyltransferase(Reference Depommier, Everard and Druart47). Previous human studies that investigated other a priori dietary patterns or indices (e.g. Healthy Eating Index, Mediterranean Diet Score and Healthy Food Diversity index) also showed that greater adherence to healthy diets was associated with increased abundance of other SCFA-producing microbes, such as Faecalibacterium prausnitzii (Reference Bowyer, Jackson and Pallister8,Reference Tindall, Petersen and Kris-Etherton10) .
Less is known with regard to other potential inflammation-related taxa observed in our study and their interactions with the inflammatory potential of diet. E. nodatum is a periodontal pathogen(Reference Haffajee, Teles and Socransky48) that has been linked to active rheumatoid arthritis and psoriatic arthritis characterised by chronic inflammation affecting joints and connective tissues(Reference Moen, Brun and Valen49). C. leptum is a carbohydrate-fermenting bacteria associated with Treg responses known to modulate intestinal and systemic inflammation in experimental models and found to be lower in patients with inflammatory bowel disease(Reference Kabeerdoss, Sankaran and Pugazhendhi50–Reference Prosberg, Bendtsen and Vind52). Although in our study higher C. leptum was observed in healthy individuals with a more pro-inflammatory diet, this cross-sectional association was attenuated with adjustment for age and BMI. A. intestini, previously recovered from various human clinical samples of hospitalised patients(Reference Jumas-Bilak, Carlier and Jean-Pierre53), appears to be responsive to supplementation with quercetin and partially hydrolysed guar gum in human faecal samples(Reference Ghimire, Wongkuna and Sankaranarayanan54,Reference Inoue, Sakaue and Kawada55) and is positively associated with LPS-stimulated TNFα production(Reference Schirmer, Smeekens and Vlamakis56). This is generally consistent with our study finding of higher A. intestini in the most pro-inflammatory diet group and its positive correlation with circulating PAI-1, a pro-inflammatory marker closely associated with obesity, type-2 diabetes and CVD risk(Reference Yarmolinsky, Bordin Barbieri and Weinmann57). Conversely, L. mephitis, a species of Proteobacteria that reduces nitrite to nitrous oxide without production of nitrogen(Reference Finkmann, Altendorf and Stackebrandt58), was inversely correlated with E-DII and PAI-1 in our study.
Metagenomic pathways are expected to more closely reflect the functional nature of the gut microbiome as an interacting community of multiple microorganisms that supports (or hinders) the host. We found that the AMPK signalling pathway, a central regulator of cellular energy homeostasis and glucose and lipid metabolism(Reference Hardie, Ross and Hawley59), was inversely associated with E-DII and C-peptide. In both human and animal studies, activation of AMPK in adipose tissue has been linked to several anti-inflammatory dietary factors, such as n-3 PUFA, polyphenolic compounds and fibre(Reference Wang, Yu and Zhang60,Reference Gauthier, O’Brien and Bigornia61) . However, there is little evidence to support our findings for microbial AMPK and carbon metabolism pathways with C-peptide, an insulin resistance marker(Reference Haidet, Cifarelli and Trucco62). We also identified a positive association of secondary bile acids biosynthesis with IL-6. Primary bile acids are converted to secondary bile acids through microbial modifications in the gut and modulate signalling via the nuclear bile acid receptors, that is, farnesoid X receptor and G-protein-coupled bile acid receptor (TGR5). Farnesoid X receptor and TGR5 signalling influence many different metabolic processes in the host, including energy homeostasis, glucose homeostasis, obesity and inflammatory responses, involving IL-6 and TNF-α (Reference Wahlström, Sayin and Marschall63–Reference Kakiyama, Hylemon and Zhou65). The positive association between E-DII and polyketide biosynthesis is also interesting, as bacterially derived polyketides exhibit a number of bioactive properties that modulate antibacterial, antitumour and antiviral activities(Reference Lombó, Menéndez and Salas66–Reference Arthur, Perez-Chanona and Mühlbauer68). The direction of this relationship and some others (e.g. the suggestive inverse association of DII with secondary bile acids biosynthesis) are somewhat unexpected and may be chance findings due to very limited sample size (n 34). While these pathways reflect the presence of genes needed to perform a particular metabolic function, they do not necessarily reflect the gut-derived metabolites that may or may not be produced from interactions between microbes and host diet or other exposures (e.g. medications). Additionally, some pathological pathways may not be activated given the study exclusion criteria and focus on healthy individuals. Larger prospective studies are warranted to confirm or refute our findings, as this analysis was exploratory in nature and none of our findings were significant following adjustment for multiple testing.
The biological processes underlying diet–microbiome interactions that modulate inflammation are not fully known. With a diet enriched with more anti-inflammatory components such as fibre- and polyphenol-rich plant foods, saccharolytic fermentation of carbohydrates by gut microbes can produce SCFA (i.e. butyrate, acetate and propionate). SCFA promote anti-inflammatory responses in the host through a series of mechanisms, including intestinal homeostasis, genetic/epigenetic regulation and immunomodulatory signalling(Reference O’keefe16). Conversely, a pro-inflammatory diet (e.g. high in fat, simple carbohydrates and meat but low in fibre-rich plant foods) creates a pro-inflammatory milieu of protein catabolites and deconjugated bile acid residues, leading to increased inflammation-related phenotypes in the host(Reference O’keefe16). These include impairment of the mucosal barrier and altered gut permeability and immune responses(Reference Liu, Tabung and Zhang18). A local inflammatory environment further alters gut microbiota to affect systemic inflammation via adherence to the gut epithelium, passage through the gut barrier to enter systemic circulation, activation of an immune response through toll-like receptor binding and/or activation of regulatory T cells and through the synthesis and secretion of cytotoxic biomolecules or metabolites(Reference O’keefe16,Reference Liu, Tabung and Zhang18,Reference Candela, Turroni and Biagi69) . Additional human research is needed to elucidate the microbial pathways and metabolites that modulate diet-induced inflammation.
Interestingly, despite our findings for specific microbes and functional pathways, we did not observe significant or striking associations between gut microbial (α and β) diversity and E-DII score in either the crude or adjusted analyses. While this could be attributed to insufficient variation, sample size or residual confounding due to unmeasured factors, several studies comparing faecal microbiota diversity across participants following distinct diets (e.g. vegetarian, vegan and omnivore) also reported no, or only modest, differences in microbial diversity between diet groups(Reference Liszt, Zwielehner and Handschur70–Reference Wu, Compher and Chen74). Notably, the DII focuses on a balance of dietary components similar to other diet indices, but it may not be directly comparable to other studies due to its specific focus on inflammation. The Healthy Eating Index and Mediterranean Diet Score were significantly associated with various α-diversity metrics in an assessment of three different dietary indices based on FFQ data from 2070 members of the TwinsUK cohort(Reference Bowyer, Jackson and Pallister8). One(Reference Maskarinec, Hullar and Monroe9) of three additional observational studies focused on increased adherence to the Mediterranean Diet Score or other diet quality indices reported a concurrent increase or difference in microbial α-diversity(Reference Garcia-Mantrana, Selma-Royo and Alcantara11,Reference De Filippis, Pellegrini and Vannini12) . Similarly, a dietary intervention among overweight and obese subjects, as well as a companion observational study focused on healthy v. unhealthy dietary patterns, reported increased faecal microbial gene richness (total gene counts) among participants with healthier diets(Reference Cotillard, Kennedy and Kong1,Reference Kong, Holmes and Cotillard75) .
Our study is novel in its specific focus on diet-related inflammation and its interaction with microbiome, a hypothesis rooted in biologically plausible relationships that are highly relevant to host health(Reference O’Keefe, Li and Lahti14,Reference Liu, Ajami and El-Serag76) . Using a construct-validated tool which converts major inflammation-related dietary factors commonly consumed by Americans to an overall interpretable diet score, the E-DII provides a comprehensive summary of the dietary inflammatory potential of participants’ entire diets(Reference Wirth, Shivappa and Davis41–Reference Halmos, Christophersen and Bird43). Although not all of the DII components were queried in our study, the majority of the missing dietary components, like spices, are typically consumed in very low amounts in the USA. As previously reported, the range of DII scores may rely more on the amount of foods actually consumed rather than on the number of available DII components(Reference Tabung, Steck and Liese77). Furthermore, misclassification in dietary inflammatory potential due to the missing dietary factors would likely be non-differential, thus attenuating our results toward the null. Other limitations of nutritional epidemiological studies also apply here, as the FFQ is prone to response bias and measurement error. Importantly, other important factors or exposures that may affect the microbiota composition (e.g. medications) were carefully collected in personal interviews and medical charts. However, we are still learning the complexity and breadth of factors that affect the microbiome or mask observed diet–microbiome associations, and our sample size limited our ability to explore multiple factors simultaneously. At the time of recruitment, we excluded individuals who reported antibiotic use in the past month; however, this may not have been sufficient for some individual’s microbial communities to ‘return to normal’. Recent studies demonstrate that the recovery time may vary by the type, dose and duration of antibiotic use(Reference Suez, Zmora and Zilberman-Schapira78,Reference Modi, Collins and Relman79) . Our inclusion of cancer-free patient and community volunteers was designed to maximise the variation needed to identify associations of diet with the microbiome. However, homogeneity due to exclusion criteria coupled with limitations in sample size could explain the low variation observed in α-diversity, β-diversity and other associations. Also, it may have been easier to detect differences had the range in DII scores gone towards their theoretical extremes (i.e. –9 to + 8). We were particularly limited in our subset analysis of patients with residual fasting blood samples, for which we also conducted WGS sequencing of the gut microbiome. To address this, we used LASSO, which is an efficient method to select species and pathways given a small number of subjects and the large number of zero-inflated species revealed in WGS sequencing. Taken overall, the present study is of a preliminary and hypothesis-generating nature given the cross-sectional design with evaluation of diet, microbiome and blood markers at a single time point, as well as the lack of significant associations following adjustment for multiple comparisons, all of which prohibit causal and mechanistic inferences.
Dietary inflammatory potential was associated with differential composition of specific microbes but not overall gut microbiota diversity in this well-defined sample of 101 individuals. Our analysis highlighted several biologically plausible microbes potentially related to diet-induced inflammation. R. torques, E. nodatum, A. intestini and C. leptum were more abundant in subjects consuming the most pro-inflammatory diets, while A. muciniphila was enriched in subjects with the most anti-inflammatory diets. Correlations between E-DII, microbes, functional gene content pathways and inflammatory biomarkers in an exploratory subset further support the role of diet–microbiota interactions in modulating systemic inflammation in the host. Future prospective studies of dietary inflammatory potential and its interactions with the microbiome are warranted. It is increasingly important to understand how diet as a whole shapes the composition and function of the gut microbiome to modulate host inflammation, an important mechanism in the development of cancer and other major chronic diseases.
Acknowledgements
The authors would like to acknowledge the UT MD Anderson Center for Energy Balance in Cancer Prevention and Survivorship, the Center for Translational and Public Health Genomics, as well as Dr G. S. Raju, gastroenterologist at MD Anderson, and Jonathan Busquets, summer intern from Rice University.
This work was supported by a grant from The University of Texas MD Anderson Cancer Center Duncan Family Institute for Cancer Prevention and Risk Assessment (to C. R. D.). The authors were supported by grants from the American Cancer Society RSG-17-049-01-NEC (C. R. D.), the Cancer Prevention and Research Institute of Texas RP170259 Postdoctoral Training Fellowship (J. Z.), Chandler Cox Foundation (S. M. S.), the National Cancer Institute Cancer Prevention Research Training Program R25 CA056452 (A. S.) and R25 CA057730 (K. L. H.) and the National Cancer Institute Cancer Center Support Grant to MD Anderson CCSG5P30 CA016672-37 Risk, Detection and Outcomes Program (C. R. D., S. H.). The funders had no role in the design, analysis or writing of this article.
J. Z. and C. R. D. designed the research and drafted the manuscript; J. Z. conducted the analysis; J. Z., K. L. H., P. W., S. M. S. and C. R. D. contributed to the design of the analysis and interpretation of results. K. L. H., J.-S. C. and P. W. provided statistical support. C. R. D., S. M. S. and J. F. P. designed and carried out the initial studies; N. S. and J. R. H. contributed to the scoring of the E-DII and helped with E-DII score interpretation; D. D. D., G. J. B. and K. L. H. were involved in the collection and processing of the data. S. H. and J. F. P. provided essential reagents or materials. All authors contributed to the critical revision and approval of the manuscript. C. R. D. had primary responsibility for final content.
J. R. H. owns a controlling interest in Connecting Health Innovations LLC (CHI), a company has licensed the right to his invention of the dietary inflammatory index (DII®) from the University of South Carolina in order to develop computer and smartphone applications for patient counselling and dietary intervention in clinical settings. N. S. is an employee of CHI. The other authors have no conflicts of interest to disclose.
Supplementary material
For supplementary materials referred to in this article, please visit https://doi.org/10.1017/S0007114520001853
