상부 장을 따라 인간 대사체 변이

Nature Metabolism 5권, 777~788페이지(2023)이 기사 인용

13,000회 액세스

3 인용

230 알트메트릭

측정항목 세부정보

인간 식단의 대부분 처리는 소장에서 이루어집니다. 소장의 대사산물은 숙주 분비물과 섭취된 엑스포솜1 및 미생물 변형에서 비롯됩니다. 여기서 우리는 15명의 건강한 남성과 여성 참가자의 일상적인 소화 동안 상부 장 내강 내용물의 시공간적 변화를 조사합니다. 이를 위해 우리는 비침습적, 섭취 가능한 샘플링 장치를 사용하여 5개의 질량 분석 분석법2,3 및 16S rRNA 시퀀싱을 결합하여 274개의 장 샘플과 60개의 해당 대변 균질액을 수집하고 분석합니다. 우리는 술포노지질과 하이드록시지방산(FAHFA) 지질의 지방산 에스테르를 포함하여 1,909개의 대사산물을 식별합니다. 우리는 대변과 장의 대사체가 극적으로 다르다는 것을 관찰합니다. 식품 대사산물은 식이 바이오마커의 경향, 장관을 따라 디카르복실산의 예상치 못한 증가, 내강 케토산과 과일 섭취 사이의 긍정적인 연관성을 나타냅니다. 식이 유래 및 미생물에 의해 연결된 대사산물은 개인간 차이가 가장 큰 원인입니다. 특히, 샘플링 전 6개월 이내에 항생제를 복용한 2명의 개인은 생리활성 FAHFA, 설포노리피드 및 기타 미생물 관련 대사산물의 수준에 큰 차이가 있는 것으로 나타났습니다. 개인 간 변이로부터 우리는 Blautia 종을 FAHFA 대사에 관여할 후보로 식별합니다. 결론적으로, 생리학적 조건 하에서 인간 소장과 상행 결장에 대한 비침습적 생체 내 샘플링은 식이, 숙주 및 미생물 대사 사이의 연관성을 보여줍니다.

우리는 공간적, 시간적 변화의 정도를 더 잘 이해하고 대사체와 미생물군집 데이터를 통합할 가능성을 측정하기 위해 건강한 개인 15명의 상부 장관에서 채취한 관강 샘플 간의 대사체 차이를 종합적으로 연구하는 것을 목표로 했습니다. 관련 출판물4에서 우리는 이러한 장치를 사용하여 미생물총 구성, 프로파지 유도, 숙주 단백질체 및 담즙산의 미생물 변형의 장내 변화를 연구합니다. 자원봉사자들은 샘플링 시점당 4개의 샘플링 장치 세트를 삼켰습니다. 이러한 섭취 가능한 샘플링 장치는 pH 민감성 코팅이 된 캡슐에 일방향 밸브로 덮인 접힌 수집 주머니로 구성되었습니다. 네 가지 유형의 장치는 pH 5.5(유형 1), pH 6(유형 2) 및 pH 7.5(유형 3 및 4)에서 용해되는 장용 코팅에서만 달랐습니다(그림 1a). 코팅의 두께와 pH 반응성은 위를 비운 후 장관의 특정 위치에서 샘플링을 가능하게 했습니다. 장치에는 추적 목적을 위한 수동 무선 주파수 식별 칩 이상의 전자 장치가 포함되어 있지 않습니다. 코팅이 용해되면 탄성 수집 주머니가 확장되어 진공 흡입을 통해 최대 400μl의 관강 내용물을 수집합니다. 일방향 밸브는 샘플 손실과 하류 유체의 오염을 방지했습니다. 대변 샘플은 -20°C에서 냉동되었으며 분석 전에 모든 장치는 대변에서 회수되었습니다. 피하 주사바늘을 사용하여 장치에서 액체 내용물을 채취했습니다. 원시 샘플의 분취량은 16S 리보솜 RNA 미생물군유전체 분석에 사용되었으며 원심분리된 샘플의 상층액은 대사체학 연구에 사용되었습니다. 여기에서는 동일한 샘플의 대사체에 대한 세심한 분석을 수행하여 인간 샘플에서 이전에 검출되지 않은 대사 산물, 식이의 주요 바이오마커 및 참가자 전체 및 참가자 내 화학 프로필 비교를 보고합니다(보충 표 1 및 2).

a, 상부 장관 조사를 위한 연구 설계. 4가지 유형의 장 샘플링 장치를 사용하여 근위부에서 원위부 상부 장을 샘플링했습니다. 15명의 참가자가 1일차 초기 테스트 후 점심 식사 후와 저녁 식사 후에 2일 동안 최소 16개의 장치를 삼켰습니다. 장치는 표적 및 비표적 LC-MS/MS 및 GC-MS 방법으로 검색 및 분석되었습니다. b, 샘플 분석에 사용된 5가지 대사체 분석에서 확인된 대사산물. 화학물질 분류 부분은 자동화된 ClassyFire 화학물질 분류를 기반으로 포함됩니다. c, 상부 장관 영역 간의 차이의 유의성은 LMM을 사용하여 계산되었습니다. 가로 점선은 유의성 임계값 P < 0.05(n = 1,182 대사산물)를 나타냅니다. 원은 FDR 보정 후 유의성이 없음을 나타내고 다이아몬드 모양은 유의성(P < 0.05)을 나타냅니다. 이 분석에는 장 샘플의 50% 이상에서 검출된 대사산물만 포함되었습니다(n = 1,182). 효과 크기 계수는 LMM에 의해 추정된 기울기이며, 양수(음수) 계수는 근위 대장에 비해 원위에서 더 높은(낮은) 수준을 나타냅니다. 수직 점선은 ±0.2 효과 크기 계수입니다.

12,000 unknown chromatographic features were reliably detected above the level of method blanks (Supplementary Table 2). Using ClassyFire software7, structurally annotated metabolites fell into 61 chemical subclasses (Supplementary Table 1). Two untargeted high-resolution liquid chromatography (LC) MS/MS assays focusing on hydrophilic and lipophilic metabolites yielded most of the annotated compounds, with 1,612 identifications. Untargeted gas chromatography (GC)–MS added 119 primary metabolites, supplemented by targeting six short-chain fatty acids (SCFAs) and a targeted LC–MS/MS assay for 17 bile acids (Fig. 1b). QC analysis of total metabolic variance revealed separation of stool and intestinal samples, with strong clustering of pooled quality control samples (Extended Data Fig. 1b)./p>50% of device samples. Of these, 630 (54%) were significantly different in the proximal compared to distal upper intestine (false discovery rate (FDR) P < 0.05; LMM) (Fig. 1c and Supplementary Table 4), with 473 metabolites at higher levels in the proximal compared to distal upper intestine and 157 compounds at lower levels in the proximal compared to distal upper intestine (Fig. 1c). Known microbially generated chemicals including SCFAs8,9, secondary bile acids10 and some microbially conjugated bile acids11,12, increased from the proximal to distal upper intestine (Extended Data Table 1 and Fig. 1c). Of the 11 detected acetylated amino acids, 7 increased from the proximal to distal upper intestine (raw P < 0.05; LMM) (Extended Data Table 1 and Fig. 1c). We also examined the 12,346 chemically unannotated metabolite signals, restricting our attention to 9,317 signals that were detected in >50% of intestinal samples (Supplementary File 1). Overall, 3,594 (38%) features were significantly different between the proximal and distal upper intestine, with 1,937 features at higher levels in the proximal compared to distal upper intestine and 1,657 features at lower levels in the proximal compared to distal upper intestine (FDR P < 0.05; LMM) (Extended Data Fig. 4)./p>100 times more abundant on average in the intestine compared to stool. These metabolites consisted of glycinated lipids, sugars, plant natural products, carnitines, microbially conjugated bile acids and S-succinylcysteine (Supplementary Table 6). Peptides were also generally at much lower levels in stool samples compared to intestinal samples, especially when compared to the proximal intestine (Extended Data Fig. 2). We also identified >100 metabolites that were >100 times more abundant in stool compared to intestinal samples (Supplementary Table 6); these metabolites were mostly polar lipids such as phosphatidylethanolamines, phosphatidylinositols and phosphatidylglycerols, as well as specific FAHFAs. The high abundance of membrane lipids in stool samples is likely due to the high amount of bacterial cell material in stool compared to luminal samples from the upper intestine./p>50% of intestinal samples were included in this analysis. Effect size coefficient is the slope estimate calculated by LMM, with positive (negative) coefficient meaning the metabolite was higher (lower) after food consumption. Vertical dashed-dotted lines are ±0.2 effect size coefficient. c, Chemical enrichment statistics (ChemRICH) analysis revealed significant chemical classes after fruit consumption visualized by separating classes by chemical lipophilicity (logP) and chemical class significance level of −log10(P). Red circles indicate that the chemical class increased after fruit consumption and blue circle indicates that the chemical class decreased after fruit consumption. Circle size indicates the size of the chemical class. e, Theophylline and theobromine levels are strongly associated with caffeine levels. Circles represent measured levels in each sample for which both metabolites were detected. f, Chemical diagram of caffeine and known metabolic pathways with structures of detected metabolites and Spearman rank correlation coefficient (rs) for each structure (P < 1.0 × 10−13 for all metabolites; n = 1,182 metabolites)./p>70% of all significantly different metabolites in five participants and >40% for another seven participants (Extended Data Fig. 7a). For metabolites that differentiated sampling time points, sugars (organooxygen compounds) were enriched in 13 of 15 participants (Extended Data Fig. 7b). Similarly, more significantly different imidazopyrimidines, indoles and isoflavonoids were found to distinguish sampling time points than intestinal regions (Extended Data Fig. 7). These classes signify dietary metabolites that were different due to variation between food types ingested during different meals, but were not as useful for differentiating between intestinal regions./p>50% of annotated metabolites exhibited significantly different levels between proximal and distal locations. An important goal for future investigation is to characterize the effect of antibiotics on intestinal sulfonolipid-, stercobilin- and long-chain AAHFA-producing bacteria and the consequences of such disruptions on health and disease. The disruption of these bacteria by antibiotics may be linked to the incidence and etiology of inflammation, diabetes and inflammatory bowel disease55,60,63. Consequently, it will be important to uncover the dynamics and mechanisms of repopulation of antibiotic-treated individuals with these microbes./p>2,500) for genomic analysis. Every bowel movement during the study was immediately frozen by the participant at −20 °C. Participant 1 provided additional samples for assessment of replicability. A total of 333 intestinal and stool samples were analysed with metabolomics methods./p>

2,500 reads were retained for analyses./p>0.75 and Benjamini–Hochberg-corrected P < 0.1 were considered. ChemRICH75 was used to calculate enrichment statistics. Clustering was performed using the hclust function with the metabolite Spearman rank correlation matrix calculated using the cor function in R and Euclidean distance calculated with the as.dist function in R. PLS-DA and principal-component analysis (PCA) were performed with the ropls package in R76. PLS-DA models to distinguish participant and device type were assessed by sevenfold cross validation. Using 20–1,000 random permutations of class labels performed by the ropls R package to test for overfitting, models maintained Q2Y > 0.15 and P < 0.05 (ref. 77). Untargeted LC–MS/MS (HILIC and RP ESI+/−) features were normalized to the sum of internal standards for each platform, which has been shown to be more robust than normalizations to single compounds78. This normalization was performed by dividing each LC–MS feature by the sum of internal standard peak heights for that sample78,79. GC–MS data were normalized to the summed intensity of all annotated metabolites as extensively discussed in published protocols80. This method addresses differences specific to GC methods, recently called normalization to the total useful peak area81. Pooled QC data were found in a dense cluster when compared to CapScan and stool samples (Extended Data Fig. 1). During merging of datasets, metabolites detected by multiple assays were simplified to keep only data from one instrument, with preference for retaining data from the assay with lower technical variance (% coefficient of variance of pooled QC). Metabolites that were detected only in a single assay remained in the dataset, independent of the % coefficient of variance of pooled QC (Supplementary Table 1). Log10 transformation and zero-value imputation using one-tenth of the minimum reported peak height for non-detected features was performed for each metabolite before PCA and PLS-DA./p>

± 0.2. Only features detected in >50% of intestinal samples were included in this analysis (n = 9,317 features). Effect size coefficient is the slope estimated by the LMM, with positive (negative) coefficient representing a metabolite that is higher (lower) in the distal compared to proximal upper intestine. Vertical dashed lines are ±0.20 times the effect size coefficient./p>50% of intestinal samples were included in this analysis (n = 1182 metabolites). These results were visualized by separating classes by chemical lipophilicity (logP) and chemical class significance level of -log10(p-value). Red circles indicate that the chemical class was higher in the distal compared to proximal upper intestine, and blue indicates that the chemical class was lower in the distal compared to the proximal upper intestine. Purple indicates the chemical cluster has metabolites that are significantly higher as well as metabolites that are significantly lower in the distal compared to proximal upper intestine. Circle size represents the size of the chemical class./p>50% of samples for each subject were used for this analysis (n = 1182 metabolites). Non-FDR-corrected p < 0.05 was used as a significance threshold. b, Multivariate discriminant analysis (PLS-DA) was performed to identify metabolites that were most important for distinguishing between subjects, or between regions. The 100 metabolites most important for distinguishing these groups were ranked by variable importance in projection score (VIP) and are categorized by chemical subclass. Chemical subclasses with <3 metabolites are reported as ‘Other’./p>50% of samples for each subject were used for this analysis (n = 1182 metabolites). Non-FDR-corrected p < 0.05 was used as a significance value cutoff. a, Metabolites with significantly different abundance between intestinal regions for each subject, grouped by chemical class and the proportion of each chemical class. b, Metabolites with significantly different abundance between sampling time points, grouped by chemical class and the proportion of each chemical class./p>50% of all device samples. All device samples are shown, and are organized by subject. Within the top (FAHFA) and lower (fatty acid) sections, the metabolites are ordered based on hierarchical clustering. Color bar represents metabolite abundance (peak height) or concentration (ng/mL) for bile acids. Minimum and maximum values were used to set the color scale for each metabolite (each row)./p>