Re: Diet influences the functions of the human intestinal microbiome 2020 네

[문형철]

1. nature  
2. scientific reports  
3. articles  
4. article

Diet influences the functions of the human intestinal microbiome

Download PDF

• Article
• Open access
• Published: 06 March 2020

Diet influences the functions of the human intestinal microbiome

• Maria De Angelis, 
• Ilario Ferrocino, 
• Francesco Maria Calabrese, 
• Francesca De Filippis, 
• Noemi Cavallo, 
• Sonya Siragusa, 
• Simone Rampelli, 
• Raffaella Di Cagno, 
• Kalliopi Rantsiou, 
• Lucia Vannini, 
• Nicoletta Pellegrini, 
• Camilla Lazzi, 
• Silvia Turroni, 
• Nicola Lorusso, 
• Mario Ventura, 
• Marcello Chieppa, 
• Erasmo Neviani, 
• Patrizia Brigidi, 
• Paul W. O’Toole, 
• Danilo Ercolini, 
• Marco Gobbetti & 
• Luca Cocolin 

Scientific Reports volume 10, Article number: 4247 (2020) Cite this article

• 24k Accesses

• 154 Citations

• 66 Altmetric

• Metricsdetails

Abstract

Gut microbes programme their metabolism to suit intestinal conditions and convert dietary components into a panel of small molecules that ultimately affect host physiology. To unveil what is behind the effects of key dietary components on microbial functions and the way they modulate host–microbe interaction, we used for the first time a multi-omic approach that goes behind the mere gut phylogenetic composition and provides an overall picture of the functional repertoire in 27 fecal samples from omnivorous, vegan and vegetarian volunteers. Based on our data, vegan and vegetarian diets were associated to the highest abundance of microbial genes/proteins responsible for cell motility, carbohydrate- and protein-hydrolyzing enzymes, transport systems and the synthesis of essential amino acids and vitamins. A positive correlation was observed when intake of fiber and the relative fecal abundance of flagellin were compared. Microbial cells and flagellin extracted from fecal samples of 61 healthy donors modulated the viability of the human (HT29) colon carcinoma cells and the host response through the stimulation of the expression‎ of Toll-like receptor 5, lectin RegIIIα and three interleukins (IL-8, IL-22 and IL-23). Our findings concretize a further and relevant milestone on how the diet may prevent/mitigate disease risk.

초록

장내 미생물은

장 환경에 맞게 대사 과정을 조절하여 식이 성분을 다양한 소분자로 전환하며, 

이는 최종적으로 호스트의 생리학적 기능에 영향을 미칩니다. 

주요 식이 성분이 미생물 기능에 미치는 영향과 호스트-미생물 상호작용을 조절하는 메커니즘을 규명하기 위해,

 우리는 장 내 미생물 계통 구성 beyond을 넘어 기능적 레퍼토리를 종합적으로 파악할 수 있는 

다중 오믹스 접근법을 처음으로 적용했습니다. 

이 연구는 

잡식성, 채식주의자, 비건 식단을 섭취한 27명의 자원자로부터 수집된 분변 샘플을 분석했습니다. 

우리 데이터에 따르면, 

채식주의와 비건 식단은 세포 운동성, 탄수화물 및 단백질 분해 효소, 운반 시스템, 

필수 아미노산 및 비타민 합성에 관여하는 미생물 유전자/단백질의 

가장 높은 풍부도와 연관되었습니다. 

식이 섬유 섭취량과 분변 내 플래깅린 상대 풍부도 사이에는 양의 상관관계가 관찰되었습니다. 

61명의 건강한 기부자 분변 샘플에서 추출된 미생물 세포와 플래길린은 Toll-like 수용체 5, 레クチ닌 RegIIIα 및 세 가지 인터루킨(IL-8, IL-22, IL-23)의 발현을 자극함으로써 인간 대장 암세포(HT29)의 생존력과 호스트 반응을 조절했습니다. 

우리의 연구 결과는 

식단이 질병 위험을 예방하거나 완화하는 데 

어떻게 기여할 수 있는지에 대한 

추가적이고 중요한 단계를 구체화합니다.

Similar content being viewed by others

The interplay between diet and the gut microbiome: implications for health and disease

Article 15 July 2024

Monitoring the variation in the gut microbiota of captive woolly monkeys related to changes in diet during a reintroduction process

Article Open access22 March 2021

Gut microbiome signatures of vegan, vegetarian and omnivore diets and associated health outcomes across 21,561 individuals

Article Open access06 January 2025

Introduction

Comprising 1012-14 cells from 100–1000 species, the human intestinal microbiome is a complex entity living in our body. It affects human health, sustenance and well-being1. Functioning as an extra organ2, the intestinal microbiome uses nutrients from ingested foods, releases harmful or beneficial metabolites and regulates the immune system3,4. Dysbiosis of the intestinal microbiome has been associated to various gastrointestinal (GI) and non-GI diseases, such as obesity, heart-, kidney- and liver-related diseases, cancer and autism5,6,7, but the dilemma whether it acts as cause or consequence remains too far from the solution. Because the simple phylogenetic characterization does not provide deep information on functional repertoires8,9, a number of projects and studies have employed the whole-community shotgun sequencing to combine the composition and gene content of the human intestinal microbiome2,10. Current meta-genomic data show limited differences among individuals, which hypothesize the existence of a stable core microbiome and similar metabolic traits10. Notwithstanding the relevance of these findings, they do not necessarily imply similar in vivo microbial activities. Indeed, a panel of factors (e.g., diet, host genetics and pharmacological treatments) markedly affect such activities11. Dietary components have the capability to modulate the composition and mainly the function of the intestinal microbiome3,12,13. Correlations between entero-gradients/types and vegetable-rich (Prevotella and Lachnospira) or animal protein-rich diets (Bacteroides and Clostridia/Ruminococcus) have been already highlighted12,14, but the impact of dietary components still remains unclear. Previously, we have determined the compositional structure of the fecal microbiota and metabolome of 150 healthy omnivorous, vegan and vegetarian volunteers, demonstrating that vegetable-rich foods increased both the abundance of fiber-degrading bacteria and the synthesis fecal short-chain fatty acids (SCFAs)14. In contrast, omnivorous volunteers having low adherence to the Mediterranean diet (MD) had the highest levels of detrimental microbial metabolites, such as phenolic and indole derivatives, and trimethylamine N-oxide (TMAO). This emerging picture hypothesizes that diet modulates the functionality of the intestinal microbiome, which, in turn, affects the human metabolic status15,16.

In this study, we implemented a multi-omic approach (meta-genomic, -proteomic and -metabolomics) to thoroughly characterize fecal samples from omnivorous, vegan and vegetarian volunteers with the aim of showing the molecular relationship between diet and metabolic functions of the intestinal microbiome. We sought to identify the effects of key dietary components on microbial functions that modulate host-microbe interactions. This approach might lead to effective intervention strategies for maintaining human health via the diet-microbiome axis.

소개

인간 장내 미생물군은

100~1,000종에 달하는 1012~14개의 세포로 구성되어 있으며,

우리 몸 속에 존재하는 복잡한 생물체입니다.

이는 인간의 건강, 영양 섭취 및 웰빙에 영향을 미칩니다1.

장 미생물군은 추가적인 장기 역할을 하며2,

섭취한 음식에서 영양분을 흡수하고

유해하거나 유익한 대사물을 방출하며 면역 시스템을 조절합니다3,4.

장 미생물군의 불균형(dysbiosis)은

비만, 심장, 신장, 간 관련 질환, 암, 자폐증 등 다양한 위장관(GI) 및 비위장관 질환과 연관되어 있지만5,6,7,

그 원인과 결과의 관계는 여전히 명확히 규명되지 않았습니다.

단순한 계통학적 특성화는

기능적 레퍼토리에 대한 깊은 정보를 제공하지 못하기 때문에8,9,

여러 프로젝트와 연구는 인간 장내 미생물군의 구성과 유전자 내용을 결합하기 위해

전체 커뮤니티 샷건 시퀀싱을 활용해 왔습니다2,10.

현재 메타게놈 데이터는

개인 간 제한된 차이를 보여주며,

이는 안정적인 핵심 미생물군과 유사한 대사 특성의 존재를 가설로 제시합니다10.

이러한 발견의 중요성에도 불구하고,

이는 반드시 유사한 in vivo 미생물 활동을 의미하지는 않습니다.

실제로,

식이, 호스트 유전학 및 약물 치료와 같은 요인 패널은 이러한 활동에 크게 영향을 미칩니다11.

식이 성분은

장 미생물군의 구성과 주로 기능을 조절할 수 있습니다3,12,13.

장 내 미생물 분포/유형과 채소 풍부한 식이(Prevotella 및 Lachnospira) 또는

동물 단백질 풍부한 식이(Bacteroides 및 Clostridia/Ruminococcus) 간의 상관관계는

이미 강조되었습니다12,14,

그러나 식이 성분의 영향은 여전히 명확하지 않습니다. 이전 연구에서 우리는 150명의 건강한 잡식성, 채식주의자, 비건 자원자의 분변 미생물군집과 대사체 구성 구조를 분석하여 채소 풍부한 식품이 섬유 분해 세균의 풍부도와 분변 단쇄 지방산(SCFAs) 합성을 모두 증가시킨다는 것을 보여주었습니다14.

반면, 지중해 식이요법(MD) 준수도가 낮은 잡식성 자원자는 페놀 및 인돌 유도체와 트리메틸아민 N-옥사이드(TMAO)와 같은 유해 미생물 대사산물의 수준이 가장 높았습니다. 이 새로운 결과는 식이가 장내 미생물군의 기능성을 조절하며, 이는 다시 인간의 대사 상태에 영향을 미친다는 가설을 제기합니다15,16.

본 연구에서는

잡식성, 채식주의자, 비건주의자 자원자의 분변 샘플을 대상으로

메타게놈, 메타프로테옴, 메타메타볼로믹스(multi-omic) 접근법을 적용하여

식이와 장내 미생물군집의 대사 기능 간 분자적 관계를 규명했습니다.

우리는

호스트-미생물 상호작용을 조절하는 미생물 기능에 대한

주요 식이 성분의 영향을 식별하고자 했습니다.

이 접근 방식은 식이-미생물 축을 통해

인간 건강을 유지하기 위한 효과적인 개입 전략으로 이어질 수 있습니다.

Results

Fecal microbiome composition

Thirty volunteers were selected from a previous larger cohort14 where healthy adult non-smokers (15 men and 15 women) were enrolled, with an age of 25–55 years (36 ± 7.0), and a BMI > 18 (21.89 ± 2.20). All volunteers were on omnivorous, vegetarian or vegan diet for at least one year. Supplementary Table S1 in the online supplemental material describes the diet compositions together with the comparison of dietary intakes among the three considered diet groups (analysis of variance (ANOVA)). The degree of association between samples and dietary information as measured by Spearman’s correlations, (rows and columns are clustered by average-linkage clustering) allowed us to cluster omnivorous, vegetarian and vegan volunteers (Supplementary Fig. S1). Each individual collected fecal samples at three time-points (within one month), and the three samples were pooled. According to our previous findings14, the fecal microbiome of the individuals differed slightly (data not shown). Lachnospira was significantly (FDR < 0.05) associated to vegans and vegetarians, while Ruminococcaceae was the most abundant family for omnivores.

분변 미생물군 구성

이전 대규모 코호트 연구14에서 건강한 성인 비흡연자(남성 15명, 여성 15명)를 대상으로 25–55세(36 ± 7.0) 연령대와 BMI > 18(21.89 ± 2.20) 조건을 충족한 30명의 자원자를 선정했습니다. 모든 자원자는 최소 1년간 잡식성, 채식주의자 또는 비건 식단을 유지했습니다. 온라인 보충 자료의 보충 표 S1은 세 가지 식이 그룹 간의 식이 섭취량 비교와 함께 식이 구성물을 설명합니다(분산 분석(ANOVA)). 샘플과 식이 정보 간의 연관성 정도는 스피어맨 상관 분석으로 측정되었으며(행과 열은 평균 연결 클러스터링으로 그룹화됨), 이를 통해 잡식성, 채식주의자, 비건 참가자를 클러스터링할 수 있었습니다(보충 그림 S1). 각 개인은 3개의 시간점(1개월 이내)에 분변 샘플을 수집했으며, 3개의 샘플을 혼합했습니다. 이전 연구 결과14에 따르면 개인의 분변 미생물군은 약간 달랐습니다(데이터 미표시). Lachnospira는 채식주의자와 채식주의자와 유의미하게 연관되었으며(FDR < 0.05), Ruminococcaceae는 잡식성 그룹에서 가장 풍부한 가족이었습니다.

Fecal microbiome gene catalogue

To establish associations between microbiome genes and omnivorous, vegan or vegetarian diet, we developed a comprehensive meta-genome catalogue performing shotgun sequencing of the total DNA from fecal specimens. The poor DNA quality excluded three samples (two omnivores and one vegetarian). The remaining 27 samples gave 184 Gb sequences, with an average of 6.81 Gb per sample. Reads assembled into 1.68 M contigs longer than 500 bp. The total contig length per sample was 96 Mb with an N50 length of 2.3 kb (Supplementary Table S2).

The functional characterization and gene classification of the shotgun sequence reads was performed against the Integrated Gene Catalogue (IGC)17 database and the MetaHIT gene catalogue, inclusive of close-to-complete sets of genes for most gut microbes. Specifically, the MetaHIT project collected more than 1200 sequenced metagenomics samples of the Human Intestinal Tract (see materials and methods section). Based on this resource, we detected 3,644 KEGG Orthology (KO) genes, with 90% of samples sharing 2,227 of them. MetaPhlAn218 was used to determine the phylogenetic abundances; the related estimated α-diversity indices revealed how intestinal microbiomes were not discriminated by diet. No significant differences were detectable at the phylum level (data not shown), but several genera associated to the intake of specific dietary components. For instance, Lachnospira positively correlated to the intake of beta-carotene, vitamin E and vegetable fat but negatively with meat, proteins, cholesterol and total proteins (Supplementary Fig. S2).

분변 미생물군집 유전자 카탈로그

잡식성, 비건 또는 채식주의 식이와 미생물군집 유전자 간의 연관성을 확립하기 위해, 분변 시료의 전체 DNA에 대한 샷건 시퀀싱을 수행하여 포괄적인 메타게놈 카탈로그를 개발했습니다. DNA 품질이 낮은 3개 샘플(잡식성 2개, 채식주의자 1개)은 제외되었습니다. 남은 27개 샘플에서 184Gb의 시퀀스가 생성되었으며, 샘플당 평균 6.81Gb였습니다. 시퀀스 읽기는 500bp 이상의 길이를 가진 1.68M개의 컨티그로 조립되었습니다. 샘플당 총 컨티그 길이는 96Mb이며, N50 길이는 2.3kb입니다(보충 표 S2).

샷건 시퀀스 리드의 기능적 특성화 및 유전자 분류는 대부분의 장 미생물에 대한 거의 완전한 유전자 세트를 포함하는 Integrated Gene Catalogue (IGC)17 데이터베이스와 MetaHIT 유전자 카탈로그를 기반으로 수행되었습니다. 특히, MetaHIT 프로젝트는 인간 장 트랙의 시퀀싱된 메타게놈 샘플 1,200개 이상을 수집했습니다(재료 및 방법 섹션 참조). 이 자원을 기반으로 3,644개의 KEGG 정족군(KO) 유전자를 탐지했으며, 90%의 샘플이 이 중 2,227개를 공유했습니다. MetaPhlAn218을 사용하여 계통학적 풍부도를 결정했으며, 관련 추정 α-다양성 지수는 장 미생물군이 식이 요인에 의해 구분되지 않음을 보여주었습니다. 문장 수준에서는 유의미한 차이는 관찰되지 않았지만(데이터 미표시), 특정 식이 성분 섭취와 연관된 여러 속이 확인되었습니다. 예를 들어, Lachnospira는 베타카로틴, 비타민 E, 식물성 지방 섭취와 양의 상관관계를 보였지만, 육류, 단백질, 콜레스테롤, 총 단백질과는 음의 상관관계를 나타냈습니다(보충 그림 S2).

Meta-genomes associated to diets

Based on principal coordinate analysis (PCoA) using relative gene abundance, microbiomes grouped into three different clusters corresponding to diets (Fig. 1a). As defined by permutation multivariate analysis of variance (PERMANOVA) test, each cluster showed different microbiome layouts. Then, we applied Random Forests19 to functional data set. This allowed the identification of diet discriminatory KO genes having distinctive changes in abundance (Fig. 1b,c and Supplementary Table S3). Superimposing the biplot of KO gene coordinates on the PCoA plot, we identified those genes that differed among omnivores, vegans and vegetarians. Genes responsible for amino acid and carbohydrate metabolisms (Fig. 1c and Supplementary Table S3), two-component gene regulatory system (Ko02020), chemotaxis (Ko02030), and, especially, flagellar assembly (Ko02040) were associated to vegan and/or vegetarian diets (Fig. 1c,d and Supplementary Table S3). We used the Bioconductor package DESeq220 to investigate the relative abundance of specific genes depending on dietary habits. The DESeq2 comparison among the three diet groups evidenced statistically significant gene differences. The highest number of differentially abundant genes (adjusted p-values, calculated with Wald test in DESeq2 followed by Benjamini–Hochberg correction) was found comparing omnivores and vegans (Supplementary Table S4). Discriminatory genes were mainly responsible for cell mobility, environmental information, genetic information processing, and carbohydrate, amino acid, energy, nucleotide, cofactors and vitamins, lipid, and glycan biosynthesis metabolisms.

식단과 관련된 메타지놈

상대적 유전자 풍부도를 사용한 주좌표 분석(PCoA)에 따르면, 미생물 군집은 식단에 따라 세 가지 클러스터로 분류되었습니다(그림 1a). 순서 변경 다변량 분산 분석(PERMANOVA) 테스트에 의해 정의된 바와 같이, 각 클러스터는 서로 다른 미생물 군집 구성을 보였습니다. 그런 다음, 기능 데이터 세트에 랜덤 포레스트19를 적용했습니다. 이를 통해 풍부도에서 뚜렷한 변화를 보이는 식단 차별적 KO 유전자를 식별할 수 있었습니다(그림 1b, c 및 보충 표 S3). KO 유전자 좌표의 바이플롯을 PCoA 플롯에 중첩하여, 잡식성, 비건, 채식주의자에서 서로 다른 유전자를 식별했습니다. 아미노산 및 탄수화물 대사(그림 1c 및 보충 표 S3), 이중 구성 요소 유전자 조절 시스템(Ko02020), 화학유동(Ko02030), 특히 편모 조립(Ko02040)과 관련된 유전자들은 채식주의자 및/또는 채식주의 식이와 연관되었습니다(그림 1c,d 및 보충 표 S3). 우리는 Bioconductor 패키지 DESeq220을 사용하여 식습관에 따라 특정 유전자의 상대적 풍부도를 조사했습니다. 세 가지 식이 그룹 간의 DESeq2 비교는 통계적으로 유의미한 유전자 차이를 보여주었습니다. 차이 있는 유전자 수(Wald 검정 후 Benjamini–Hochberg 교정된 조정 p-값)가 가장 높은 그룹은 잡식성과 채식주의자 그룹 간 비교에서 관찰되었습니다(보조 표 S4). 차별화 유전자들은 주로 세포 이동성, 환경 정보, 유전 정보 처리, 탄수화물, 아미노산, 에너지, 핵산, 보조인자 및 비타민, 지질, 글리칸 생합성 대사 과정과 관련이 있었습니다.

Figure 1

Principal coordinate analysis (PCoA) of the KEGG Orthology (KO) genes that discriminated the fecal microbiomes of omnivorous (O), vegan (V) and vegetarian (VG) volunteers based on their diets. Panel a, Euclidean PCoA plot illustrating the observed diversity between samples. Panel b, the most abundant KO genes belonging to carbohydrate and amino acid metabolism. Panel c, the most abundant KO genes involved in flagellar assembly and bacterial chemotaxis. The spheres represent KO genes mapped onto the weighted average of the coordinates of all samples, where the weights are the relative abundances of the genes in the samples. The size of the spheres is proportional to the mean relative abundance of the corresponding genes across all samples. Purple spheres represent amino acid or carbohydrate metabolism; yellow spheres represent flagellar assembly, bacterial chemotaxis or two-component system genes. Panel d, heatmap showing the differentially (FDR < 0.05) detected genes involved in flagellar assembly and bacterial chemotaxis. The colors of the scale bar denote the abundance of the genes, with 1.15 indicating the highest abundance (red) and −1.15 indicating the lowest abundance (green) between diet groups.

Full size image

Meta-proteomes associated to diets

Thanks to a proteomic gel-free method based on ultra-high-pressure liquid chromatography tandem mass spectrometry (UHPLC-MS/MS) performed with Easy-nLC 1000 UHPLC system coupled to a quadrupole-Orbital mass spectrometer, we identified 5,760 proteins with EggNOG function, 2,950 of which associated to a KO code. On average, more than 90% of the proteins belonged to Firmicutes, Bacteroidetes, Actinobacteria and Proteobacteria (data not shown). We used the Bioconductor package DESeq2, PCoA and discriminant analysis of principal components (DAPC) to stratify samples and to identify proteins specific for dietary groups. In vegetable-rich diets, the intestinal microbiome showed statistically significant differences in the synthesis of 181 proteins (Supplementary Table S5), which were mainly responsible for carbohydrate, nitrogen and cofactor and vitamin metabolisms. Other differences related to environmental information processing and genetic information processing (replication and repair). A ribonuclease (K03601, E.C. 3.1.11.6) responsible for DNA replication and repair clearly associated with vegetable-rich diets. PCoA of the core protein relative abundance clustered samples (FDR = 0.040; permutational test with pseudo F-Ratio) depending on the diet (data not shown). DAPC group classification was consistent with the original clusters (prior provided clusters) of omnivores, vegans and vegetarians (Fig. 2a). Flagellin (K02406), cyclic pyranopterin monophosphate synthase, polygalacturonase and levansucrase proteins specifically discriminated vegans and vegetarians from omnivores (Fig. 2b). The omnivorous cluster had the highest relative amount of undecaprenyl diphosphatase. The vegetarian cluster distinguished because of argininosuccinate synthase, cyclic pyranopterin monophosphate synthase, peptidase T and methyl-accepting chemotaxis protein. UDP-N-acetylglucosamine 1-carboxyvinyl transferase, UDP-N-acetylmuramate dehydrogenase, starch synthase and xylulose 5-P-reductoisomerase specifically distinguished the vegan cluster. Based on the retained discriminant functions, the analysis derives probabilities for each individual of membership in each of the different groups. The posterior DAPC assignments were consistent with the original pre set diet clusters (Fig. 2c).

Figure 2

Multivariate statistical analyses based on the meta-proteomes of omnivores (O), vegans (V) or vegetarians (VG). Panel a, discriminant analysis of principal components (DAPC) score plot. Panel b, heatmap showing the differentially (FDR < 0.05) detected proteins in the sample meta-proteomes that mostly discriminated the diet groups. The colors of the scale bar denote the protein abundance with 1.15 indicating the highest abundance (red) and −1.15 indicating the lowest abundance (green) between diet groups. Panel c represents whether the individuals (rows) were correctly assigned (based on discriminant functions) to the genetic cluster where they were included a priori (columns) by K-means analyses used to infer the best-supported clustering solution. Colors represent membership probabilities to each cluster (red = 1, orange = 0.75, yellow = 0.25, white = 0) and blue crosses indicate the cluster where the individuals were originally assigned by K-means analyses. Sample label colours match the sample diet labels in the DAPC clusters (Panel a).

Full size image

Diet modulates carbohydrate metabolism and biosynthesis of SCFA of the intestinal microbiome

Biology Pathway Tools (PT) software and the MetaCyc multiorganism database21 were used to reconstruct metabolic pathways from meta-genomic and -proteomic data. This innovative approach enabled the reconstruction of the main metabolic pathways, which were active in fecal samples. We compared enzymes at two levels. At the first level, we compared the averaged values (relative amount) of the same gene detected in each diet cohort (omnivores vs. vegans, omnivores vs. vegetarians and vegans vs. vegetarians). At the second level, we compared the averaged values of the same protein detected in each diet cohort. Both meta-genomic and meta-proteomic data showed the synthesis of SCFAs through carbohydrates and free amino acids (Fig. 3a). Fecal samples from vegans and vegetarians harboured the highest levels of genes and/or proteins with statistically significant differences emerging by applying DESeq2 software (e.g., levansucrase, E.C. 2.4.1.10; pectate lyase, E.C. 4.2.2.2; and phosphoenolpyruvate carboxylase, E.C. 4.1.1.31) capable of hydrolyzing acetyl-CoA and succinate (Fig. 3a,b). Acetyl-CoA was used to synthesize acetate by phosphate acetyltransferase (E.C. 2.3.1.8), which was at a higher level in vegetarians and, especially, vegans. The main routes to synthesize butyrate and propionate were acetyl-CoA and succinate pathways, respectively. Vegans and vegetarians also had the highest level of the super-pathway of Clostridium acidogenic fermentation, which resulted in the synthesis of butyrate from pyruvate/acetyl-CoA. On the contrary, the metabolic pathway converting amino acids (β-lysine and glutarate) into butyrate mainly associated to omnivores. The levels of acetyl-CoA/propionyl-CoA carboxylase (E.C. 6.4.1.3) and acyl-coenzyme A synthetase (E.C. 6.2.1.1) from the succinate pathway were the highest in vegans and vegetarians. As determined by headspace solid-phase micro-extraction (SPME), coupled with gas chromatography mass spectrometry (GC-MS), a higher concentrations of acetate, butyrate and propionate was found in the fecal samples of vegans and vegetarians with respect to omnivores, which confirmed meta-genomic and -proteomic results (Fig. 3a).

Figure 3

Reconstruction of microbial pathways in the intestine involved in the biosynthesis of short-chain fatty acids (acetic acid, butanoate and propionate) (SCFAs) using DESeq statistically significant differences for genes and proteins identified from the multi-omics data sets belonging to omnivores (O), vegans (V) and vegetarians (VG). Panel a, schematic representation of the SCFA metabolic pathways. The blue numbers indicate enzymes that were differentially (FDR < 0.05) detected among diet groups. Principal metabolites are colored in green. The average concentrations (µM/g of feces) of acetate, butyrate and propionate found in the metabolome of omnivores, vegans and vegetarians are indicated in the histograms. Panel b, heatmap showing the differentially detected genes (red characters) and proteins (black characters) in the diet groups. The colors of the scale bar denote the abundance of the genes and proteins (indicated in blue characters in panel a), with 1.94 indicating the highest abundance (red) and −1.94 indicating the lowest abundance (green) between diet groups.

Full size image

Diet modulates the nitrogen metabolism of the intestinal microbiome

The metabolism of nitrogen relates to both proteolytic system and metabolism of amino acids, assuming a pivotal role for microbial growth and survival. The abundance of several peptidases (amidohydrolases, protease 4, M16 and M23 family peptidases, peptidase E and T, and methionine aminopeptidase) was higher (p-values generated using DESeq2, by applying a Wald test followed by Benjamini–Hochberg correction) in vegans and/or vegetarians (Supplementary Table S5) with respect to omnivores. No statistically significant differences occurred among dietary habits by using DNA-seq data and looking for gene levels for the same enzymes. Meta-genomic and -proteomic data reconstructed complete biosynthetic pathways for isoleucine (Ile) from threonine; L-lysine (Lys) and L-methionine (Met) from L-aspartate; L-valine (Val) from pyruvate; and L-tryptophan (Trp) from chorismate and L-glutamine (Fig. 4a). The level of several enzymes was lower in omnivores with respect to vegans and/or vegetarians (Fig. 4b). The relative amount of the specific methionine transport system (MetQ, E.C. 3.1.1.34) was the highest in vegetarians. Besides, the relative abundance of enzymes (e.g., E.C. 4.1.1.48; E.C. 4.2.1.20) responsible for Trp biosynthesis mainly associated to vegans and vegetarians. Compared to omnivores, vegans and, especially, vegetarians had a higher level of L-asparaginase (E.C. 3.5.1.1, K01424), which converts L-asparagine into L-aspartate.

Figure 4

Reconstruction of the microbial pathways in the intestine involved in the biosynthesis of L-methionine, L-lysine, L-isoleucine, L-valine and L-tryptophan using DESeq statistically significant differences for genes and proteins identified from multi-omics data sets belonging to omnivores (O), vegans (V) and vegetarians (VG). Panel a, schematic representation of metabolic pathways for the biosynthesis of amino acids. The blue numbers indicate enzymes that were differentially (FDR < 0.05) detected among diet groups. Principal metabolites are colored in green. Panel b, heatmap showing the differentially detected genes (red characters) and proteins (black characters) in the diet groups. The colors of the scale bar denote gene and protein abundance (indicated in blue characters in panel a), with 1.15 indicating the highest abundance (red) and −1.15 indicating the lowest abundance (green) between diet groups.

Full size image

Diet modulates the capacity of the intestinal microbiome to synthesize vitamins and cofactors

Meta-genomic and -proteomic data revealed the presence of enzymes responsible for the de novo synthesis of folate (Fig. 5a). Most of these significantly different enzymes (p-values generated using DESeq2, by applying a Wald test followed by Benjamini–Hochberg correction) were at higher levels in vegans and/or vegetarians with respect to omnivores (Fig. 5b). Biosynthetic genes for menaquinols (menaquinol-6 to menaquinol-13) were present in all the meta-genomes, but menaquinol-6 and −9 were mainly detectable in the vegan and vegetarian meta-proteomes. All meta-genomes and -proteomes shared genes and proteins for the biosynthesis of deoxyxylulose-5P (DXP) from pyruvate, glyceraldehyde-3P, L-cysteine or L-tyrosine. DXP is required for the synthesis of thiamine and pyridoxal (Fig. 5c). In particular, the level of thiG, encoding the key enzyme thiazole synthase (E.C. 2.8.1.10), was the lowest in omnivores (Fig. 5d). Indeed, the biosynthesis of thiamine from pyridoxal-P or 1-(5’-phospho-ribosyl)−5-aminoimidazole primarily related to vegans and vegetarians. The biosynthesis of pantothenic acid (vitamin B5) and CoA, starting from pyruvate and L-aspartate, was part of the core meta-genomes and -proteomes. Nevertheless, the higher relative abundance of proteins responsible for the biosynthesis of pantothenic acid and CoA was mostly associated to vegans and vegetarians. Besides, the higher abundance of proteins responsible for the biosynthesis of pyridoxal and pyridoxine also mostly related to the same dietary habits.

Figure 5

Reconstruction of the microbial intestinal pathways involved in the biosynthesis of folate (Panels a, b) and thiamine, pyridoxal, pantothenic acid and coenzyme A (Panels c, d) statistically significant differences for genes and proteins identified from multi-omics data sets by applying the DESeq2 R package and belonging to omnivores (O), vegans (V) and vegetarians (VG). Panels a and c, schematic representations of metabolic pathways for the biosynthesis of folate and thiamine, pyridoxal, pantothenic acid and coenzyme A. The blue numbers indicate enzymes that were differentially (FDR < 0.05) detected among diet groups by applying the DESeq2 package. Panels b and d, heatmap showing the differentially detected genes (red characters) and proteins (black characters) in the diet groups. The colors of the scale bar denote gene and protein abundance (indicated in blue characters in panel a), with 1.15 indicating the highest abundance (red) and −1.15 indicating the lowest abundance (green) between diet groups.

Full size image

Correlations between meta-proteome and diet

We further correlated meta-proteomic data to dietary habits. The R package psych was used to compute Pairwise Spearman’s correlations; adjusted p-values were generated by applying the Bonferroni correction (Supplementary Table S6). Fibers positively correlated with flagellin (R = 0.646, FDR = 0.003). The key enzyme phosphate acetyltransferase positively correlated with daily fiber (R = 0.617, FDR = 0.033) and legume (R = 0.698, FDR = 0.043) intakes, and negatively (FDR < 0.05) correlated with meat, preserved meat and fish intakes. Other enzymes responsible for propanoate and butanoate metabolisms (e.g., malate L-lactate dehydrogenase) positively correlated (FDR < 0.05) with daily intakes of fiber, vegetable oil, vegetable proteins and other components of the vegetable-based diet. On the contrary, the above enzymes negatively correlated with the intake of animal products.

Anti-proliferative effects of the intestinal microbiome in the HT29 human colon carcinoma cell line

The effects of SCFAs and flagellin on growth of the human HT29 adenocarcinoma cell line, together with their tumor-suppression and anti-inflammatory activities have been extensively studied in literature22,23. We here analysed HT29 colon carcinoma cell line in 61 volunteers in order to study the effect of flagellin on their viability. Microbiomes were from 22 omnivorous, 20 vegetarian and 19 vegan fecal samples (27 analyzed in this study plus 34 additional samples belonging to a previous larger cohort14). Microbial fecal cells (MFCs) (ca. 8 log cells/ml of Dulbecco’s modified Eagle’s medium [DMEM]) and the corresponding microbial protein cell extracts (MPCEs) (15 mg/ml) were used to perform growth inhibition assays. The ANOVA results (adjusted p-values corrected for multiple tests by applying the Tukey test) showed how growth of HT-29 cells was significantly inhibited by MFC and MPCE treatments (Fig. 6a,b,d,e). Hence, the strongest inhibition (P < 1e-05) was after 72 h and, especially, using MFCs from vegans and vegetarians. The highest ability of vegans and vegetarians to inhibit the HT29 cell growth could be due to the flagellin concentrations. As estimated by nanoHPLC coupled to nanoESI-MS/SM, the lower value of flagellin was found in MFCs of omnivores (ca. 0.001 µg/mg) than vegan and vegetarians (ca. 0.006 µg/mg) (P < 1e-05) MFCs from vegetarians vs. omnivores, P < 1e-05 MFCs from vegans vs. omnivores and P = 0.964 MFCs from vegetarians vs. vegans. To prove the effect of flagellin on the viability of HT29 cells, a pure preparation of flagellin was also tested at the same concentration found in 15 mg of MPCE from omnivores (e.g., 0.015 µg/ml of DMEM) or vegans and vegetarians (0.090 µg/ml). Flagellin inhibited the viability of HT29 cells in a dose- and time-dependent manner (Fig. 6c,f, Supplementary Table S7). Compared to control, the strongest inhibition by flagellin (47%, P < 0.001) was after 72 h (Fig. 6f).

Figure 6

The intestinal microbiota inhibits the proliferation of colon cancer cells. The capacity of microbial fecal cells (MFCs) extracted from 22 omnivores (O), 20 vegetarians (VG) and 19 vegans (V) and the corresponding microbial protein cell extracts (MPCEs) and flagellin (FlC) at two different concentrations (0.015 µg/ml of DMEM for O and 0.090 µg/ml for VG and V MPCE samples) to affect cell viability was assessed by the sulforhodamine B assay at different time-points in human HT29 colon cancer cells. Percentages of growth inhibition for colon cancer cells treated with MFCs, MPCEs or FlC compared to control (cells treated with PBS only), presented as the mean value ± s.d. from three replicates for each of the 61 samples. All data shown are representative of three independent experiments using fecal samples collected at three time-points in one month. Corrected P-values were obtained by using ANOVA test and Tukey test for multiple test correction.

Full size image

The intestinal microbiome increases the expression‎ of interleukins, Toll-like receptor 5 (TLR-5) and the lectin RegIIIα

We investigated whether MFCs and MPCEs affected the immune response in HT29 cells and in order to disclose significant differences between the samples, we computed an ANOVA test corrected for multiple comparisons by using the Tukey test (see material and method). Compared to PBS-exposed cells, exposure (6 h) of HT29 cells to omnivorous MFCs significantly increased the release of IL-8 (50 ± 2.14 and 15 ± 0.24 pg/ml, respectively, P < 1e-04; Fig. 7a, Supplementary Table S8). The release of IL-8 was higher when HT29 cells underwent treatments with vegetarian or vegan MFCs (95.9 ± 4.69 and 97.11 ± 4.87 pg/ml, respectively, P < 1e-04) (P < 1e-04 for vegetarians vs. omnivores, and P < 1e-04 for MFCs from vegans vs. omnivores). A prolonged exposure (24 h) to MFCs increased IL-8, but the trend did not vary.

Figure 7

The intestinal microbiota induces the expression‎ of interleukins, Toll-like receptor 5 and the lectin RegIIIα. The capacity of microbial fecal cells (MFCs) extracted from 22 omnivores (O), 20 vegetarians (VG) and 19 vegans (V) and the corresponding microbial protein cell extracts (MPCEs) and flagellin (FlC) at two different concentrations (0.015 µg/ml of DMEM for O and 0.090 µg/ml for VG and V MPCE samples) to affect the expression‎ of IL-8 (panels A–C), IL-22 (D–F) and IL-23 (G–I), Toll-like receptor 5 (J–L) and the lectin RegIIIα (M–O) was assessed at 6 and 24 h in human HT29 colon cancer cells. The levels of expression‎ of each gene in colon cancer cells treated with MFCs, MPCEs or FlC, compared to control (cells treated with PBS only), presented as the mean value ± s.d. from three replicates for each of the 61 samples. All data shown are representative of three independent experiments using fecal samples collected at three time-points in one month. Differences between control and treated cells were considered statistically significant when p < 0.05. A schematic representation of how interleukins, TLR-5 and RegIIIα work synergically is provided in panel P.

Full size image

To focus on the effect of microbial proteins without metabolite (e.g., butyrate) interferences, HT29 cells underwent exposure to MPCEs. As expected, the effect of MPCEs was lower than that of MFCs (P < 0.05; Figs. 7a,b, Supplementary Table S8). Compared to omnivorous MPCEs, exposure to vegetarian MPCEs increased the release of IL-8 (Fig. 7b, Supplementary Table S8). Vegan MPCEs behaved similarly. After 6 h of exposure, the release of IL-8 was similar between treatment with 0.090 µg/ml of flagellin and vegetarian or vegan MPCEs (Fig. 7c, Supplementary Table S8). At 24 h, the efficiency of flagellin was lower than MPCEs but significantly (P < 1e-05) higher than PBS.

Exposure of HT29 cells to MFCs, MPCEs or flagellin increased the expression‎ of the interleukins IL-22 (Fig. 7d–f) and IL-23 (Fig. 7g–i, Supplementary Table S8) and TLR-5 (Fig. 7j–l, Supplementary Table S8). Vegetarian and vegan MFCs or MPCEs promoted the highest (P < 0.05) induction of interleukins and TLR-5. Compared to PBS-exposed cells, exposure (6 h) of HT29 cells to omnivorous MFCs significantly increased the expression‎ of the bactericidal C-type lectin RegIIIα (fold change, 3.50 ± 0.05, P < 1e-04). The expression‎ of RegIIIα was the highest with vegetarian or vegan MFCs, both after 6 and 24 h (Fig. 7m, Supplementary Table S8). Compared to PBS-exposed cells, the exposure of HT29 cells to 0.090 µg/ml of flagellin increased the expression‎ of RegIIIα (Fig. 7o, Supplementary Table S8). Figure 7p reports a schematic representation of the synergistic activity of interleukins, TLR-5 and RegIIIα.

Discussion

Using a non-redundant catalogue of intestinal bacterial genes and proteins, we assessed how the intestinal microbiome responds and adapts to different dietary habits. Highlighting the relationships between diet and gut microbiome, and their repercussions in microbiome-host metabolic and immune interactions is one of the main current challenges in microbiology24,25,26. According to our data, the phylogenetic composition was not useful in discriminating omnivorous, vegetarian and vegan dietary habits. As previously reported2,11, the inter-individual variability was high. Few phylogenetic traits were associated to vegan and vegetarian (Lachnospira) or omnivorous (Ruminococcaceae) diets. Nevertheless, higher diversity at sub-genus level may exist as we recently showed for Prevotella and Bacteroides27,28. As previously described2, Gene Ontology (GO) categories based on meta-genome catalogue of the intestinal microbiome were shared in healthy individuals. However, vegans and vegetarians were associated with the highest abundance of genes responsible for flagellar assembly, chemotaxis and two-component systems. Vegans and vegetarians also harbored functional potential, related to carbohydrate, amino acid, cofactor and vitamin metabolisms. The meta-proteomic approach completed the estimation of the microbiome metabolic adaptation to dietary habits. The relationships between DNA and protein content in the complex microbiome ecology are poorly understood. Gene copy number and protein abundance eval‎uation is a critical step useful to eval‎uate how changes in DNA relate to changes in proteins. Some published data suggest that changes in the metagenome reflect changes in the metaproteome. We observed how, compared to meta-genomic, a higher variability of the GO categories occurred at proteomic level. Although the sample size in this study is relatively small, our results may reflect the environmental adaptation and meta-genomic plasticity of the intestinal microbiome under different dietary conditions. Vegetarians showed the highest relative amount of proteins for many GO categories. Proteins responsible for flagellar assembly (e.g., flagellin) were overrepresented in vegetarians and vegans. These findings, along with boosted chemotaxis functions, indicated that microbiomes of individuals with plant-based diet, which is rich of non-digestible carbohydrates, might had developed strategies to get physical access to nutrients. Overall, flagella correlated with cell adhesion and biofilm formation, and represented virulence factors that increase the host immune response29.

The intestinal microbiome showed numerous genes that are responsible for carbohydrate transport and metabolism. Proton-coupled active transport components (MFS, GPH, and the ATP binding cassette superfamily) and group trans-locators (PTS-GFL superfamily) were the most frequently identified. Depending on the type of available carbohydrate, bacteria modulate the synthesis of specific transporters. Compared to omnivorous, vegan and, especially, vegetarian fecal samples showed a higher relative number of carbohydrate-hydrolyzing enzymes and transport systems. This feature enhanced the carbohydrate metabolism, positively affected the levels of acetate, butyrate and propionate, which regulate glucose and energy homeostasis30, and maintain the integrity of the intestinal epithelial barrier31,32. The synthesis of propionate proceeds via succinic acid, with methyl-malonyl-CoA decarboxylase as the key enzyme33. Based on the core genes and proteins that we detected, methyl-malonyl-CoA decarboxylase is the main biosynthetic enzyme for propionate in human intestinal microbiomes. This pathway is connected, by pyruvate, to both C6 and C5 compounds. Interestingly, vegan and vegetarian diets induced some enzymes (e.g., malate-L-lactate dehydrogenase), which were positively correlated to vegetable-based foods. The acetyl-CoA-utilizing pathway is the main route to synthesize butyrate, followed by lysine, glutarate and 4-aminobutyrate34. Here, we show that vegan and vegetarian diets activated the super-pathway of Clostridium acidogenic fermentation, which produces butyrate and acetate from pyruvate/acetyl-CoA. This explains the strong correlation between fiber intake and fecal concentration of butyrate14,35 and strengthens the intake of such nutrients as pivotal for health-promoting metabolism. Acetyl-CoA and lysine pathways were detectable in the same fecal samples, suggesting the adaptation to a protein-rich diet34. Our data confirmed that alternative pathways (lysine and glutarate) for butyrate biosynthesis36 were present and mainly associated to omnivorous diet.

Vegans and even more vegetarians showed higher levels of protein-hydrolyzing enzymes than omnivores. This finding correlates to the high resistance of vegetable proteins to human-gut digestion and to the higher intake of legumes, as recorded in vegan and vegetarian questionnaires. The biosynthetic pathways for biogenic amines in vegans and vegetarians did not differ from those of the omnivores. The same was detectable for enzymes catalyzing the conversion of carnitine, choline or betaine into trimethylamine. This result was consistent with the absence of trimethylamine and TMAO in the fecal samples. Consequently, the high intake of carnitine, choline or betaine by omnivores with low adherence to MD might be responsible for the high level of TMAO detected in their urines14,15. We showed active biosynthetic pathways for Met, Lys, Ile, Val and Tyr, which represents an interesting feature of the functional meta-genome, since humans lack this biosynthetic ability. Biosynthetic pathways for folate, thiamine, pantothenic acid, pyridoxal and pyridoxine were also active. Several enzymes catalyzing these pathways had higher levels in vegans and/or vegetarians with respect to omnivores.

Overall, we found that omnivorous, vegetarian or vegan diets had an impact on the microbial synthesis of proteins (e.g., flagellin and L-asparaginase) and metabolites (e.g., butyrate and pyridoxal/pyridoxine) that are linked to mechanisms of oncogenesis and tumor suppression37. The microbial fecal cells, their protein extracts and flagellin showed anti-proliferative effects towards HT29 colon carcinoma cells. Although our model is oversimplified, it is worth noting that the strongest inhibitory effect was found by using high concentrations of flagellin and, especially, the intestinal microbiome from vegetarians and vegans. Previously published data demonstrated that HT29 cells express TLR-538 and that some probiotic bacteria inhibit cell growth and viability of colon carcinoma cells. Binding to TLR-5, flagellin induces the IL-22-dependent production of antibacterial lectins of the RegIII family39. Commonly, the induction of IL-22 is associated to the production of IL-23 from macrophages/dendritic cells. Nevertheless, colorectal carcinoma cells such as HT29 release IL-2340. Flagellin has attracted pharmacological interest to develop vaccines41 and because its capability to stimulate immune responses42. Thus, we hypothesized the existence of additional protection against gut-infecting bacteria and pathogens in vegans and vegetarians. Flagellin enhances tumor-specific CD8 + T cell immune responses43 and improves the defense towards genital cancer in mice44. A protective effect of flagellin against other types of cancer, as well as the capability to increase tumor necrosis, was described in animal models45. Currently, pathogen inhibition and SCFA production are the main colon cancer-preventing effects of fiber and vegetable-enriched diets. By correlating the fiber daily intake and the level of flagellin in fecal samples, we hypothesized a tangible role of fiber-rich diets in protecting against tumor occurrence. Nevertheless, flagellin alone showed weaker inductive effects on IL-8, IL-22, IL-23, TLR-5 and RegIIIα than vegetarian or vegan MPCEs, which indicated also a presumptive role of other proteins. We noticed that fiber- and vegetable-enriched diets exhibited higher levels of enzymes involved in tumor suppression (L-asparaginase and ribonuclease)46.

In conclusion, our multi-omics approach shed new light on complex diet-microbiome interactions. Intestinal microbial metabolism consists of many enzymatic reactions that function together in a synchronized manner. Complex regulatory mechanisms maintain the functional balance among individual pathways. Although the need of further investigations, our data demonstrated how responses from the intestinal microbiome to vegetable-rich diets primarily include: an increased cell motility to access nutrients, an increased catalytic activities for carbohydrates and food proteins, the synthesis/release of bioactive metabolites/proteins and potentially beneficial impacts on human health.

Methods

Participant recruitment

Volunteers were recruited through advertisements and using flyers, which were distributed in the areas surrounding 4 different Italian cities: Bologna, Parma, Torino and Bari. Thirty healthy adult volunteers (15 men and 15 women) were enrolled, with an age of 25–55 years (36 ± 7.0), and a BMI > 18 (21.89 ± 2.20). The volunteers had been following an omnivorous, an ovo-lacto-vegetarian or a vegan diet for at least one year. Omnivorous, vegetarian, or vegan diets were validated by one week FFQ (Food Frequency Questionnaire)47. The sample set included individuals who followed an omnivorous (total no = 10; 5 men and 5 women), an ovo-lacto-vegetarian (total no = 10; 4 men and 6 women) or a vegan (total no = 10; 5 men and 5 women) diet. Volunteer features, recruitment and exclusion criteria, dietary information, sample collection and storage procedures are reported in De Filippis et al., 201614. Prospective participants were excluded according to the following criteria: V, VG and O regime followed for less than one year, age under 18 or over 60, regular consumption of drug, regular supplementation with prebiotics or probiotics, consumption of antibiotics in the previous 3 months, evidence of intestinal pathologies (Crohn’s disease, chronic ulcerative colitis, bacterial overgrowth syndrome, constipation, celiac disease, Irritable Bowel Syndrome) and other pathologies (type I or type II diabetes, cardiovascular or cerebrovascular diseases, cancer, neurodegenerative diseases, rheumatoid arthritis, allergies), pregnancy and lactation. All participants were asked questions about consumption of animal product in order to understand if their dietary habits in the last year diverged from the self-declared diet type. The subjects were instructed on how to self-collect the samples; all materials were provided in a sterile convenient, refrigerated, specimen collection kit (VWR, Milan, Italy). Faecal samples were collected on the same day of three consecutive weeks, and the three samples were pooled before microbiome, metaproteome and metabolome analyses. Home collected samples were transferred to the sterile sampling containers using a polypropylene spoon and immediately stored at 4 °C by the volunteers. The specimens were transported to the laboratory within 12 hours of collection at a refrigerated temperature. Containers were immediately stored at −80 °C. Food and beverage intake was estimated by means of a 7-day weighed food diary, which was completed every day for a total of one week, to collect metadata and to confirm‎ the type of diet. The intake of macronutrients and micronutrients was calculated using the Microsoft Access application coupled to the European Institute of Oncology food database (European Institute of Oncology, 2008).

Ethical statement

The study protocol was approved by the Ethics Committee of: (a) Azienda Sanitaria Locale (Bari) (protocol N.1050), (b) Azienda Ospedaliera Universitaria of Bologna (protocol N.0018396), (c) Province of Parma (protocol N.22884) and (d) University of Torino (protocol N.1/2013/C) after having ascertained its compliance with the dictates of the Declaration of Helsinki (IV adaptation). All methods were performed in accordance with relevant guidelines and regulations. All patients provided written informed consent prior to participation in the study protocol. The study protocol was registered on ClinicalTrials.gov, with the identified number NCT02118857.

DNA extraction and sequencing

Triplicate fecal aliquots collected from each volunteer were pooled for DNA extraction. Ten grams of the pooled sample was aseptically homogenized with 90 ml of Ringer’s solution (Oxoid) for 2 min in a Stomacher. A 2-ml aliquot was collected and centrifuged at the maximum speed for 30 s; the supernatant was removed, and the DNA was extracted from the pellet using a Powersoil DNA kit (MO-BIO, Carlsbad, CA, USA) according to the manufacturer’s instructions. Single-end DNA library construction (one 151-bp) was performed by using the TruSeq DNA library preparation kit, and shotgun sequencing for the HiSeq. 1500 platform (Illumina, San Diego, CA, USA) was performed according to the manufacturer’s instructions (G4L Company, Salerno, Italy).

Functional meta-genomic annotation and statistical analysis

Raw sequencing reads were quality-trimmed (Phred score < 30), and reads shorter than 60 bp were discarded using the SolexaQA +  + (v3.1.7.1) software48. The remaining reads were aligned against the Integrated Gene Catalogue17 (IGC) of human gut developed within the MetaHit project using Bowtie2 (v2.3.5.1) software49 with the following parameters: -t -f -D 20; -R 3; -N 0; -L 20; -i S, 1, 0.50 – local. Reads that showed the best hit ( > 90% of identity over at least 30% of the query length) against the IGC were extracted using SAMtools (version 1.9) and normalized to the total read number mapped to the whole catalogue. An average value of 90% of reads were mapped against the IGC and only genes with KEGG ID were extracted and further used for downstream analysis (3,644 KEGG Orthology (KO) genes). Shotgun reads were also assembled with Velvet v1.2.10 with default parameters50. Reads that are human contaminants have been discarded by using the BMTagger software (ftp://ftp.ncbi.nlm.nih.gov/pub/agarwala/bmtagger/). Each contig was analyzed by using the automated gene prediction and annotation pipeline PROKKA51 v1.12. In order to reconstruct metabolic pathways the FASTA and genbank files relative to the set of annotated contigs were parsed then used as input for Pathway Tools v19.0.

Data normalization and the determination of differentially abundant genes, among the three dietary groups, were then conducted using the Bioconductor DESeq2 package20 in the statistical environment R with default parameters. P values were adjusted for multiple testing using the Benjamini-Hochberg procedure, which assesses the FDR.

PCoA was performed with R “adegenet” package (https://cran.r-project.org/web/packages/adegenet/adegenet.pdf) using the gene relative abundance based on Euclidean, Bray-Curtis and Jaccard distances. The Random Forests algorithm was used to discriminate genes among diet groups. The phylogenetic characterization of the shotgun sequences was assessed using MetaPhlAn218 software with default parameters. The resulting biological observation matrix (.biom files) was then imported into QIIME52 to produce an OTU table at the genus level. In order to find differences in microbiome composition among the samples as a function of diet the Wilcoxon test in R was used.

Alpha diversity indices were estimated by the R phyloseq package

Spearman’s non-parametric correlations through the psych package of R were used to study the relationships between the relative abundance of microbial taxa abundance and dietary variables. The correlation plots were visualized in R using the made4 package of R.

A succinct step-by-step workflow summarizes the analyses carried out both for the meta-genomic and the meta-proteomic counterparts (Supplementary Fig. S3).

Protein extraction, denaturation, digestion and desalting

Pooled fecal samples (2 g) were suspended in 18 ml of ice-cold Tris-buffered saline (TBS) buffer and homogenized using a lab Stomacher. The homogenate was passed through a 20-μm vacuum filter unit to remove larger fibrous material and human cells and centrifuged (1000 rpm for 5 min) to pellet bacterial cells. The pellet was collected, washed with 5 ml of Tris-HCl (50 mM pH 7.5) to remove attached human proteins and lysed via sonication. Proteins were precipitated with 20% TCA, digested by trypsin and analyzed by UHPLC-MS/MS.

Samples were prepared for digestion using the filter-assisted sample preparation (FASP) method53. Briefly, the samples were suspended in 1% SDC, 50 mM Tris-HCl, pH 7.6, and 3 mM DTT, sonicated briefly, and incubated in a Thermo-Mixer at 40 °C, 1000 rpm for 20 min. The samples were clarified by centrifugation, and the supernatant was transferred to a 30 kDa MWCO device (Millipore) and centrifuged at 13 kg for 30 min. The remaining sample was buffer exchanged with 1% SDC, 100 mM Tris-HCl, pH 7.6, then alkylated with 15 mM iodoacetamide. The SDC concentration was reduced to 0.1%. The samples were digested using trypsin at an enzyme-to-substrate ratio of 1:100 overnight at 37 °C in a Thermo-Mixer at 1000 rpm. Digested peptides were collected by centrifugation. A portion of the digested peptides, approximately 20 µg, were desalted using reversed phase stop-and-go extraction (STAGE) tips54. Peptides were eluted with 80% acetonitrile and 0.5% acetic acid and lyophilized to near dryness in a SpeedVac (Thermo Savant) for approximately 1 h.

Liquid chromatography-tandem mass spectrometry

Each digestion mixture was analyzed by liquid chromatography (LC) by using an Easy-nLC 1000 UHPLC system (Thermo Fisher). Mobile phase A was 97.5% MilliQ water, 2% acetonitrile, and 0.5% acetic acid. Mobile phase B was 99.5% acetonitrile and 0.5% acetic acid. The 240-min LC gradient ran from 0% B to 35% B over 210 min and then to 80% B for the remaining 30 min. Samples were loaded directly into the column. The column was 50 cm × 75 μm I.D. and packed with 2 micron C18 media (Thermo Easy Spray PepMap). The LC was interfaced to a quadrupole-Orbitrap mass spectrometer (Q-Exactive, Thermo Fisher) via nanoelectrospray ionization using a source with an integrated column heater (Thermo Easy Spray source). The column was heated to 50 °C. An electrospray voltage of 2.2 kV was applied. The mass spectrometer was programmed to acquire tandem mass spectra from the top 10 ions in the full scan from 400 to 1200 m/z by data-dependent acquisition. Dynamic exclusion was set to 15 s, singly charged ions were excluded, the isolation width was set to 1.6 Da, the full MS resolution was set to 70,000 and the MS/MS resolution was set to 17,500. Normalized collision energy was set to 25, max fill MS was set to 20 ms, max fill MS/MS was set to 60 ms and the underfill ratio was set to 0.1%. The mass spectrometer RAW data files were converted to mzML format using msconvert.

Functional meta-proteomic annotation and statistical analysis

Mascot Generic Format (MGF) files were generated from mzML using the Peak Picker HiRes tool, part of the OpenMS framework. All search instances were performed on an Amazon Web Services–based cluster through the Proteome Cluster interface. Proteome Cluster builds monthly species- and genus-specific protein sequence libraries from the most current UniProtKB distribution. The most recent protein sequence libraries available from UniProtKB were used. MGF files were searched using X!Tandem55, both with the native56 and k-score57 algorithms and using OMSSA58. XML output files were parsed and non-redundant protein sets were determined using the Proteome Cluster based on previously published rules59. MS1-based isotopic features were detected, and peptide peak areas were calculated using the Feature Finder Centroid tool from the OpenMS framework60. Data normalization and the determination of differentially abundant proteins, among the three dietary groups, were conducted using the Bioconductor DESeq2 package20 in the statistical environment R with default parameters. Wald test p-values were corrected for multiple testing by using the Benjamini-Hochberg post hoc procedure.

Looking for evidence of structure among the analysed diet groups, we filtered out non-informative non-core proteins, i.e. those proteins that occurred with a maximum of 15% in each diet group. We ran DAPC by using the adegenet R package. In this multivariate analysis we used the belonging to the sample diet group as the a priori clustering condition and we retained 4 principal components. We plotted the first two discriminant functions with the scatter function of the adegenet R package. In order to ascertain if DAPC classification is consistent with the original clusters (known from diet diaries), we used the “assignplot” R function to calculate the proportions of successful reassignments (based on the discriminant functions). This function is particularly useful when prior biological groups are used, as one may infer admixed or misclassified individuals.

Microbiome pathway reconstruction

PT software version 19.021 and the coupled MetaCyc multiorganism database were used to reconstruct metabolic pathways. For the meta-genomic counterpart, the assembled genbank and .fasta files were both used to generate the .pt (pathologic) format. For the proteomic data batch, the protein output was converted into the .pf format, miming the genbank entry fields. The .pf and .pt supported formats were then used, through the built-in Pathologic software, to obtain new PGDB databases.

The numbers of reactions (total reactions in the base pathways) and pathways (base pathways) where compared in each sample and used to generate 0/1 matrices. The PT software allowed us to infer the prediction of metabolic pathway hole in our meta-genomic and -proteomic samples. The REST-style version of the KEGG API utility (http://www.kegg.jp/kegg/rest/) was used to enrich the protein dataset in terms of KEGG codes and EC numbers.

Gas-chromatography mass spectrometry/solid-phase microextraction (GC-MS/SPME) analysis

According to the manufacturer’s instructions, the DVB/CAR/PDMS fibre (Supelco, Bellefonte, PA, USA) was exposed to headspace for 40 min to extract volatile organic compounds (VOCs) from fecal samples. VOCs were thermally desorbed by immediately transferring the fiber into the heated injection port (220 °C) of a Clarus 680 (Perkin Elmer, Beaconsfield UK) gas chromatography equipped with an Rtx-WAX column (30 m × 0.25 mm i.d., 0.25 μm film thickness) (Restek) and coupled to a Clarus SQ8MS (Perkin Elmer) with source and transfer line temperatures kept at 250 and 210 °C, respectively. Each chromatogram was analyzed for peak identification using the National Institute of Standard and Technology 2008 (NIST) library. Quantitative data of the identified compounds were obtained by interpolation of the relative areas versus the internal standard area.

Fecal microbes and preparation of protein extracts for HT29 cell line assays

Thirty fecal samples analyzed by a multi-omic approach, plus 31 additional samples belonging to the previous larger cohort (13), for a total of 22 omnivores, 20 vegetarians and 19 vegans, were used. MFCs and MPCEs were obtained using the protocols applied for meta-proteomic analysis. MFC samples were washed with sterile PBS and added to DMEM at a final cell density (O.D. 620 nm) of 0.65 UA, corresponding to ca. 8 log cells/ml. MPCEs were analyzed by the Bradford method to quantify the total protein concentrations. The flagellin content in the 61 MPCE samples was also purified by liquid chromatography and quantified by nano-HPLC coupled with nano-ESI-MS/MS. Each MPCE was added to DMEM at a final protein concentration of 15 mg/ml. Flagellin was also used at final concentrations of 0.015 and 0.090 µg/ml in DMEM.

Cell line

Based on the above results showing that diet modulates the microbial synthesis of molecules/proteins (e.g., SCFA and flagellin) involved in oncogenesis or tumor suppression, the microbiomes of 61 volunteers were tested using the human HT29 colon carcinoma cell line. HT29 cells were cultured in DMEM containing fetal bovine serum (10%, FBS, Life Technologies), 2 mM glutamine and 100 u/ml penicillin/100 μg/ml streptomycin (Life Technologies) at 37 °C in the presence of 5% CO2. For the co-incubation experiments with MFCs, MPCEs, fecal microbiomes or flagellin (InvivoGen, San Diego, CA, USA), the cells were maintained at 37 °C under CO2-independent conditions and cultured with the above-described standard DMEM supplemented with 25 mM HEPES.

HT29 cell viability assays

The cell viability of HT29 cells was assessed by the SRB assay61 using an initial cell density of 5,000 or 20,000 cells/well, respectively. The cells were incubated with MFCs, MPCEs or flagellin for 24, 48 and 72 h. After washing with PBS, the cells were fixed with 10% TCA. Staining of cells was performed using SRB for 30 min, and the cells were flushed repeatedly with 1% acetic acid. SRB was desorbed using 10 mM Trizma, and the plate was read at 492 nm using a microplate reader. Cells incubated in DMEM alone were used as controls.

Gene expression‎ analyses of HT29 cells

HT29 cells grown with DMEM or DMEM plus MFCs, MPCEs or commercial flagellin for 6 and 24 h were washed twice with PBS containing Pen-Strep and 50 µg/ml gentamicin and stored at −80 °C until use. Total RNA was extracted from the HT29 cells using a commercial kit (Ribospin Minikit-GeneAll, Seoul, Korea). cDNA was synthesized from 2 μg of template RNA in a 20-μl reaction volume using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Monza, Italy). Ten microliters of total RNA was added to the Master Mix and subjected to RT-PCR in a thermal cycler (Stratagene Mx3000P Real Time PCR System, Agilent Technologies Italia S.p.A., Milan, Italy). The cDNA was amplified and detected through Taqman primer-probe sets (Applied Biosystems) (IL8, Hs00174103_m1; IL22, Hs01574154_m1; IL23A, Hs00372324_m1; TLR5, Hs01920773_s1; and REG3A, Hs00170171_m1). Human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as the housekeeping gene and detected through Taqman primer-probe Hs999999_m1. The relative fold change in expression‎ was normalized to GAPDH expression‎. All procedures were performed according to the manufacturer’s instructions. TLR-5 was quantified by chromatin immunoprecipitation (ChIP) using EZChIPTM chromatin immunoprecipitation kit (Upstate) as described by Kumar Thakur et al.62. In details, HT29 cell pellets were resuspended in the lysis buffer of the kit, and the chromatin was precipitated overnight with 2 μg of rabbit antibodies against RNA polymerase II, Sp1, Sp3, acetyl-H3, acetyl-H4, p300, HDAC1 or IgG (negative control). At the end of incubation, samples were treated with Protein G agarose for 1 h. The immunoprecipitated complex was washed and subsequently extracted with elution buffer. DNA-protein complexes were reversed and DNA was purified by ethanol precipitation. The relative binding of proteins to the TLR-5 promoter was quantitatively analyzed by qPCR.

Enzyme-linked immunosorbent assay (ELISA)

Cell culture supernatants were analyzed for IL-8, IL-22 and IL-23 release in triplicate using an ELISA kit (Human IL-8/CXCL8; IL-22 and IL-23 DuoSet ELISA R&D Systems, Minneapolis, MN, USA CN: DY208, DY782 and DY1290 respectively).

Statistical analysis

All data coming from gas-chromatography mass spectrometry-solid-phase microextraction (GC-MS/SPME) were obtained at least in triplicates. The GC-MS/SPME analysis, was carried out on transformed data followed by separation of means with Tukey’s HSD, using a statistical software Statistica for Windows (Statistica 6.0 per Windows 1998, (StatSoft, Vigonza, Italia).

For cell line assay statistical analyses (data at least in triplicate), differences between groups were analyzed using the ANOVA test. The correction for multiple comparisons was performed using the Tukey test and the function glht (general linear hypothesis tests) in “multcomp” R package63.

Degree of association between genera and nutrients were assessed by Spearman correlation coefficients than clustered by Euclidean distance and Ward linkage hierarchical clustering. Correlations between enzymes abundances and dietary intake were assessed by using Spearman’s correlation coefficients (FDR < 0.05 and R > 0.6); the p-values were corrected for multiple testing by using the Bonferroni adjustment within the psych R package.

Data availability

Sequences filtered for human reads and trimmed of low-quality bases have been uploaded to the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA; SRP083099, Bioproject ID PRJNA340216).

References

1. Lozupone, C. et al. Identifying genomic and metabolic features that can underlie early successional and opportunistic lifestyles of human gut symbionts. Genome Res. 22, 1974–1984 (2012).

   Article CAS PubMed PubMed Central Google Scholar