University of Groningen
Understanding the gut ecosystem: bugs, drugs & diseases
Vich Vila, Arnau
DOI:
10.33612/diss.102587978
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2019
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Vich Vila, A. (2019). Understanding the gut ecosystem: bugs, drugs & diseases. University of Groningen. https://doi.org/10.33612/diss.102587978
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Combining absolute quantification of faecal
bacteria with metagenomic sequencing
data improves characterization of the gut
microbiome in patients with Crohn’s disease
Arnau Vich Vila, Alexander Kurilshikov*, Anne Boddeke*, Valerie Collij, Paula Sureda Horrach, Julius Z. H. von Martels, Arno R. Bourgonje, Dianne B.H. Jansen, Marjolein A.Y. Klaaseen, Shixian Hu,
Klaas Nico Faber, Gerard Dijkstra, Jingyuan Fu, Alexandra Zhernakova, Hermie J.M. Harmsen, Rinse K. Weersma
*Authors contributed equally
Chapter 5
Most microbiome studies to date that focus on the gut ecosystem have relied on faecal material as a proxy for the gut composition. Characteristics of the faecal sample, such as appearance and moisture, have been associated to specific microbial signatures including changes in the number of bacteria in the samples. These factors are particularly relevant in the study of gastrointestinal disorders such as Crohn’s disease (CD), where the disease progression is directly implicated in changes in sample characteristics. Under the hypothesis that quantitative measurements of the microbial faecal content can improve microbiome characterization, we investigated the relationship between microbial loads and microbial composition and how this relates to the heterogeneous microbiome observed in patients with CD.
We combined absolute quantitative measures of bacterial cell counts, namely microbial loads, with shotgun metagenomic sequencing data in a cohort of 70 patients with CD. Bacterial counts were assessed by fluorescent in-situ hybridization (FISH). Participants with intestinal resections showed a lower cell density in stool. Conversely, we saw no influence of disease activity as indicated by the level of inflammation. Faecal samples with higher microbial loads were characterized by a higher abundance of F. prausnitzii and R. intestinalis, while lower microbial loads were associated with an expansion of potentially pathogenic bacteria like E. coli. Bacterial loads per sample proved to be an essential
confounding factor in the trait–bacteria association analysis of CD sub-phenotypes such as intestinal resection in the ileum. However, we also show that these effects can be partially corrected by estimating microbial densities using a combination of microbial richness and the concentration of extracted DNA per sample.
We find that patients with CD show a large variation in the number of bacterial cells in their faecal samples, and this variation is associated with specific microbial signatures and disease-associated phenotypes, including intestinal resection and ileum inflammation. Therefore, we suggest that differences in microbial loads should be considered when exploring the gut microbiota composition in gastrointestinal diseases.
The study of the gut microbial ecosystem using high throughput sequencing technologies has become a common practice. However, the lack of established standards for the data collection and analysis steps poses a challenge in microbiome research. Many technical and environmental factors have been shown to influence the microbial composition1–3. For example, the characteristics of the faecal samples, e.g. moisture or appearance (Bristol stool scale), have been shown to be strongly correlated with changes in the microbial composi-tion, suggesting that these characteristics can significantly confound the taxonomic relative abundances that are commonly used to represent the microbiome4,5. In addition, recent publications have found a strong relationship between the bacterial density in a sample and its microbial composition6,7, and microbial load (the bacterial cell density in a sample) itself is now being considered as a biologically relevant phenomenon due to its association with host phenotypes. This has led to the suggestion that combining standard composi-tional methods with quantitative microbial measurements can help improve the power and specificity of trait–microbiota associations8.
This is particularly important for gastrointestinal disorders, where microbial changes have been suggested to be both cause and consequence of host phenotypes9. Characteristics of the gut microbiota ecosystem in patients with Crohn’s disease (CD), an inflammatory and chronic disorder in the gut, have been extensively studied. The microbiota of patients with CD is characterized by a decrease in bacterial richness and an increase in aerotolerant bacteria which comes at the cost of anaerobic bacteria10–13. However, factors such as disease location, intestinal surgery or disease activity introduce considerable heterogeneity into the gut microbial community in CD patients, complicating the identification of microbiome features relevant to disease-onset and pathogenesis14, 15.
We hypothesize that combining sequencing data with absolute quantification of the gut microbiota (e.g. of the number of bacteria per gram of faeces) can provide better insights into disease heterogeneity and improve the accuracy of gut microbiome studies in diseases such as CD that have a substantial impact on the gut ecosystem. We also aimed to explore if the strong link between microbial load and microbial composition allows us to impute the microbial load from the compositional microbiome data as this would allow re-use of existing data from CD-related microbiome studies.
This study uses a cohort of 70 patients with CD. For each participant, faecal samples were taken at two time points three weeks apart. Whole genome metagenomic shotgun sequencing was performed to assess the relative composition of the microbiota. Microbial loads were measured using fluorescence in-situ hybridization (FISH) and represented as bacterial counts per gram of faeces (Methods). Four individuals had missing data for the second time point, and therefore only one sample was considered. Five individuals were excluded because they had a stoma (n=2) or colectomy (n=3). These patients presented a lower bacterial load in their samples (p=0.0005, Wilcoxon test, Supplementary Figure 1). In total, 127 samples belonging to 65 patients were used. The average age of partici-pants was 42 years old (SD=12.8), and they had an average body mass index (BMI) of 25 (SD=4.9). 68% of the participants were females. Clinically, 30 participants showed active disease (indicated by faecal calprotectin values >200) at least one of the two time points, while 83% had ileum involvement in their disease course and 40% had undergone an in-testinal resection. Overall, 77 host phenotypes were collected, including anthropometric traits (age, sex, BMI), clinical characteristics (disease activity, surgery, use of medication and biomarker measurements) and dietary information (Supplementary table S1).
The number of bacteria per sample was, on average, 1.20 x1010 cells/gram of faeces (range: 1.35 x109–5.45 x1010). No significant differences where observed when comparing the var-iation in bacterial loads between time points in the same individual and the varvar-iation between samples from random individuals (p-value = 0.65, Wilcoxon-test).
The amount of isolated DNA (rho=0.38, False Discovery Rate (FDR)=1.09 x 10-05, Spear-man correlation) and the microbial richness per sample (rho=0.55, p-value=1.59 x 10-11, Spearman correlation) were positively correlated with bacterial loads. Sequencing depth, however, showed a poor correlation with bacterial load (rho=0.17, p-value=0.05, Spearman correlation). Moreover, we observed a negative correlation with bacterial load in patients with a higher number of liquid stool samples in the day before sample collection (rho=-0.32, FDR=0.001, Spearman correlation) (Figure 1).
Faecal calprotectin, which is used as a biomarker for CD disease activity in clinical prac-tice, was not associated to changes in bacterial loads, not even when dividing samples into active disease (faecal calprotectin levels >200) and remission (faecal calprotectin levels <200) (FDR=0.62, Wilcoxon test). However, patients with an ileum resection and those who underwent multiple intestinal resections did show a significant decrease in bacterial load (FDR=0.0004 and FDR=0.001, respectively; Wilcoxon test; Supplementary Figure 3). Next we investigated if the frequency of resection or time elapsed since the last resection had an impact on bacterial loads. A subset of patients who had undergone an intestinal resection were selected from the original dataset (n=26, number of faecal samples=52). On average, these patients had undergone 1.5 resections (maximum 3), with the most recent having occurred, on average, 12 years before faecal sample collection (minimum 9
Results
Cohort description
Correlation between microbial cell loads and different phenotypes.
Values for panels a-d are -log10 transformed. “FISH” refers to the average number of cell counts (based on epifluorescence assay with bacterial oligonucleotide probe) per gram of stool sample. In panel e) the x-axis depicts the correlation between different microbial species abundances (invert rank transformed) and the microbial loads per sample and the y-axis the correlation between species and sequencing depth. Symbol size represents the predicted mean abundance of each species. Species names in blue are positive correlations. Species names in red are negative correlations.
months, maximum 24 years). The absence of an ileocecal valve did not have an impact on the number of bacterial cells in the sample (p-value=0.74, Wilcoxon test). Although the number of ileum resections showed a slight negative correlation (rho=-0.37, FDR=0.06), no significant correlations were observed between the number of resections or the time since last resection and the microbial loads in the faecal sample (Supplementary table 3).
Next, we explored the influence of each of the available phenotypes on the overall mi-crobial species composition (beta-diversity). The relationship between these factors and the microbial compositions were calculated as the Bray-Curtis dissimilarity metric and a PERMANOVA analysis was carried out for each phenotype.
In total, 44 phenotypes had a statistically significant impact on the overall composi-tion. Strikingly, bacterial load was the phenotype that explained the highest proportion of the microbiome variance, capturing 9% of the inter-individual variation (FDR=0.0003, PERMANOVA). After correcting for the other 43 significantly associated phenotypes, microbial load remained one of the factors that explained most of the variance (r2=0.013, FDR=0.001, PERMANOVA). As expected, phenotypes related to disease location and intestinal resection were also among the most relevant factors influencing microbiome composition (Supplementary Figure 2, Supplementary table 2).
To investigate whether differences in microbial loads were due to the potential overgrowth or depletion of specific bacterial taxa, or if they were random, relative abundances of each microbial species derived from the metagenomics data were correlated with the measured bacterial loads for each sample.
Bacterial loads were significantly correlated with the abundance of more than 50 mi-crobial species, with the relative abundances of Faecalibacterium prausnitzii, Ruminococcus
bromii, Alistipes onderdonkii and Eubacterium eligens showing the strongest positive
corre-lations (rho>0.5, FDR< 0.0001, Spearman correlation). On the other hand, the relative abundances of known opportunistic or potentially pathogenic bacteria such as Escherichia
coli (rho=-0.33, FDR=6.24 x 10-04, Spearman correlation) and Ruminococcus gnavus (rho=-0.40, FDR=2.11 x 10-05, Spearman correlation) were negatively correlated to bacterial load (Fig 2, Supplementary table 4). In contrast, the number of sequenced reads per sample showed almost no correlation with the bacterial abundances, with Pseudoflavonifrator
cap-illosus the only bacterial species that showed a slightly positive correlation with sequencing
depth (rho=0.32, FDR=0.03, Spearman correlation).
To validate that these effects were not due to the technical differences between metagen-omic sequencing and FISH quantifications, we calculated the relative abundances of three bacterial groups using the same quantification procedure. FISH counts were available for F. prausnitzii (probe Fprau 645), Enterobacteriaceae (probe Ec1531) and Clostridium
Differences in bacterial loads reflect different overall microbiome composition
Although microbial loads can differ significantly between individuals and between health conditions, estimation of microbial loads requires extra measurements that are not usually performed in large microbiome studies. We were therefore interested to see if any sample characteristics derived from the shotgun metagenomic sequencing procedures could be used as a proxy for microbial load. Considering the relationship of microbial richness, as measured by Shannon index, and DNA concentration with the total microbial loads, we estimated how much load variation could be explained by these factors. Here we found that bacterial richness could explain 31% of the variation and DNA concentration 17.8%, and the percentage of variance explained rose to 40% when combining both factors. The linear combination of bacterial richness and DNA concentration modelled in this cohort was then used to predict the bacterial loads in two independent cohorts: a cohort of patients16 and a Dutch population cohort17(Supplementary Table 6).
Adjusting for the differences in sequencing depth between samples is a common practice in the microbiome studies, yet differences in microbial loads are rarely considered. To ad-dress how correction for microbial loads would influence the discovery of microbe–trait associations, we selected four factors that have been reported to impact the microbiome composition. These were age, sex and two clinical characteristics: use of proton-pump in-hibitors (PPI) and, for patients with CD, resection of the ileum. We then tested the chang-es in the significance levels before and after inclusion of bacterial loads in the models and replicated the results using predicted microbial loads in two other metagenomic cohorts: a population cohort (LifeLines Deep, n=1040) and an inflammatory bowel disease (IBD) cohort (1000IBD, n=406).
The associations with age and sex did not change greatly after correcting for microbial load (Supplementary Table 7, Supplementary Figure 4). In our discovery cohort of patients with CD (n=70), the nominally significant age–taxa associations were highly concordant before and after correcting for microbial counts (p-value=1, McNemar’s chi-square test). This trend was also observed in both our additional cohorts using predicted microbial loads (p-valuepopulation cohort=0.28, p-valueIBD cohort=0.61, McNemar’s chi-square test).
Interestingly, although the use of PPIs has been reported to have a large impact on the gut ecosystem, this is not confounded by changes in the cell density of the sample. In line with previously reported data, Streptococcus and Rothia species abundances were positively
Microbial loads can be estimated using metagenomics sample characteristics
Statistical implications for discovery of microbial associations
coccoides-Eubacterium rectale (probe Erec482). Concordant with the results obtained using
metagenomic sequencing data, F. prausnitzii were positively correlated with total bacterial load (rho=0.38, p-value=8.56 x 10-06, Spearman correlation), while C.coccoides-E.rectale and Enterobacteriaceae abundances were negatively correlated (rho=-0.29, p-value=0.0008 and rho=-0.49, p-value=4.39 x 10-09, respectively, Spearman correlation) (Supplementary table 5).
correlated with PPI-use in all three cohorts (Supplementary table 7). In contrast, the asso-ciations between bacteria abundance and resections in the ileum changed after correcting for microbial loads (p-valuediscovery cohort=3.86 x 10-06, p-valueIBD cohort=0.0013, McNemar’s chi-square test). While decreases in the relative abundances of F. prausnitzii and
Subdol-igranulum spp. were strongly associated with having had an ileum resection, the associations
of other bacterial species such as Ruminococcus obeum, Eubacterium eligens or Coprobacillus
sp. were confounded by differences in microbial loads. In fact, when correcting for FISH
counts, the significant level of these bacteria did not meet the multiple testing adjusted significance threshold. In addition, when applying the same correction using predicted bacterial loads in the 1000IBD cohort, we could replicate the previously observed effect: of the 12 taxa previously correlated with the ileum resection phenotype, only two bacteria (F. prausnitzii and Subdoligranulum spp.) were still significant after the predicted cell count correction (Bonferroni p-value threshold <0.0003) (Figure 2, Supplementary table 7).
Overall these results suggest that correcting for microbial loads per sample has a small impact when investigating host phenotypes such as age, sex and PPI-use. In contrast, phe-notypes such as disruptions of the gastrointestinal tract, e.g. ileal-resections, tend to induce false positives in microbiome-wide association studies (MWAS) that are driven by the confounding effect of changes in microbial load. In these cases, correction for microbial loads is necessary in MWAS, and our results indicate that this confounding effect can be efficiently addressed by predicting microbial loads from metagenomic sequencing data.
In this study we have examined the implications of microbial load in the study of the gut microbial ecosystem. In agreement with previous research, patients with CD show a significant variation in faecal microbial cell densities6. We found that this variation was highly associated with previous intestinal surgery procedures: patients who had under-gone resection of the ileum showed a significant decrease in the microbial load in their faecal samples. However, we did not find any correlation between microbial load and the number of resections or the time since resection. Data indicates that surgical disruption of the intestinal environment has a long-term impact on the gut microbiota. Moreover, after adjusting for the differences in microbial loads, the association between ileal surgery and lower abundances of F. prausnitzii remained significant. Concordantly, lower proportions of F. prausnitzii in patients with CD were previously associated with ileum-disease re-currence18. Interestingly, we found no associations between inflammation biomarkers and bacterial loads, indicating that the changes in microbial loads may not be a consequence of intestinal disease flares.
We also observed that changes in microbial load are associated with changes in microbial composition. For example, species categorized as beneficial for the host, such as R. bromii and P. capillosus, showed a strong positive correlation with microbial load. Despite technical
Discussion
Partial correlation between species and phenotypes.
Panels a-d represent the p-values of the correlation between taxa abundances (invert rank transformed) and the use of Proton Pump Inhibitors (PPIs). Panels e-g represent the p-values of the correlation between taxa abundances (invert rank transformed) and resections in the ileum. X-axis indicates the -log10 p-val-ue of taxa–phenotype correlation and the y-axis the -log10 p-valp-val-ue of the partial correlation correcting for microbial loads per sample. In the study cohort (a,b,d and f), panels a and d represent the partial correlation considering the microbial loads measured by fl uorescence in-situ hybridisation (FISH), and panels b and f, when considering the microbial loads estimated from metagenomic data. Panels c, d and f represents the p-value relation in the replication cohorts. The dashed blue line indicates the Bonferroni corrected signifi cance level (-log10(0.05/131).
differences with our study, a Flemish study also identified a positive correlation between
Ruminococcus and the cell counts per sample in healthy controls6. We further observed that potentially pathogenic bacteria were more abundant in environments with lower microbial loads, suggesting opportunistic behaviour by these species. This co-occurrence is concord-ant with previous characterizations of the gut microbiota in patients with CD13, 19–21. Close and longitudinal monitoring of the gut microbiota is thus needed to assess if the expansion of these bacteria is cause or consequence in the loss of microbial biomass and richness.
Previous studies have reported changes in the gut microbiota composition associated with intestinal transit time and faecal moisture4, 17, 22. Both factors are known to be relat-ed, given that low water content indicates a longer colonic transit time for faecal mate-rial. Although these measurements were not available for our study cohort, we observed that the number of liquid stools before sample collection (a measure used to calculate the Harvey-Bradshaw index) showed a weak negative correlation to microbial loads. The Flemish study also found a relation between faecal moisture content and faecal cell count in two out of the three cohorts they investigated6. Therefore, changes in the microbial load seem to reflect several host characteristics that may shape the intestinal ecosystem and thereby exert selective pressure on the gut microbiota.
Due to the relation between microbiota composition and microbial densities, and the confounding factors affecting both composition and microbial load, it is important to incorporate absolute quantifications into microbiome studies. This measurement re-quires extra laboratory experiments and the availability of sufficient biological material, which is not always accessible to the researcher. As an alternative, we have shown that metagenomic measurements and technical sample characteristics, namely the microbial richness in the combination of DNA concentrations, partially capture the differences in microbial loads and can be used as a correction factor to improve the power and specificity of associations between microbial composition and host properties. In our study, changes in the microbial richness in the gut, combined with the concentration of DNA isolated from the faecal material, captured a reasonable percentage (40%) of microbial load variation, and we have shown that the combination of both factors can be used as a proxy for microbial loads in stool and applied as a correction factor in the microbiome analysis.
The incorporation of absolute quantification into microbiome studies also brings a resource that can be used to overcome limitations derived from the use of compositional data23, 24. Rarefying microbial profiles based on faecal bacterial loads and sequencing depth may help us better estimate microbial dynamics and causal relations in metagen-omic datasets6.
In summary, we have shown that patients with CD show a heterogeneous gut micro-biota composition that can be partially confounded by changes in the microbial load in stool samples as a consequence of diseases sub-phenotypes such as intestinal resections. We have also shown the importance of considering the variations in microbial loads per sample when exploring the gut microbiota of patients with gastrointestinal diseases, as well as demonstrating how differences in microbial loads can be partially captured by metagenomic measurements.
Samples were obtained from a previous study investigating the impact of vitamin B2 on the gut microbiota of CD patients25, which was approved by the Medical Ethical Committee of the UMCG (METc no. 2014/291) and registered on ClinicalTrials.gov (NCT02538354). In short, the cohort is composed of 79 patients with CD from the University Medical Center Groningen (UMCG). These patients underwent a 3-week vitamin B2 intervention, with blood and faecal samples collected at baseline and after intervention (day 0 and day 28). Samples were transported in dry ice and stored at -80˚C at the UMCG facilities.
Bacterial quantification was performed by FISH as described in Harmsen et al. 200226. EUB338 probe (Rhodamine) was used to estimate the total amount per bacteria. The quan-tification of other specific bacterial groups was also available using the following probes: Erec482 (FITC) for the Clostridium coccoides-Eubacterium rectale group, Ec1531 (CY3) for Enterobacteriaceae and Fprau645 (FITC) for F. prausnitzii. Fluorescent signals were counted manually, and 25 different fields were quantified for each well containing a sample.
The microbiome was characterized using frozen faecal samples, with two samples obtained for each participant: one taken before and one taken after the vitamin intervention. DNA extraction was done by combining mechanical lysis and the Qiagen Allprep DNA/RNA Mini Kit (cat #80204). Shotgun metagenomic sequencing of the DNA material was car-ried out in the Broad Institute in Cambridge (Massachusetts, USA) using the Illumina HiSeq platform. Low-quality reads were removed in the sequencing facility. Reads were trimmed, and short reads (<50% of the original length before trimming) or those aligned to the human genome were removed using kneaddata (v0.4.3). After quality control, micro-bial composition and relative abundances were predicted using MetaPhlan 2.0 (v2.14)27. A detailed description of the pipeline is available at:
https://github.com/WeersmaLabIBD/Microbiome/blob/master/Protocol_metagenomic_pipeline.md
Metagenomic sequencing data from two Dutch cohorts, 1000IBD and LifeLinesDeep cohort, were used for replication. Samples were selected based on the availability of in-formation about DNA concentrations before sequencing. In total, 1040 samples from the LifeLines cohort and of 406 from 1000IBD cohort were used. Sample collection and fur-ther processing of the metagenomic sequences was performed as in the discovery cohort.
All statistical analyses were performed in R (v.3.3.2). Microbial abundances were only con-sidered at the species level. Characteristics of the microbial ecosystem were calculated using vegan R package (v.2.5-3). Microbial diversity per sample was calculated using the Shannon Index implemented in the diversity function and the difference in the microbial composition
Methods
Cohorts and microbial load quantification
Metagenomic sequencing and characterization
between samples was calculated as Bray-Curtis dissimilarities using the vegdist function. The amount of variation in the microbial composition explained by each host phenotype was estimated using a PERMANOVA test with 10,000 permutations implemented in the adonis function.
Associations between microbial loads and host phenotypes or sample characteristics were tested using Spearman correlations for continuous data and two-paired Wilcoxon-test for categorical data. Significance levels were adjusted by multiple testing with the function p.ad-just and the Benjamini & Hochberg method and considered associated when FDR<0.05.
Microbial loads per sample were predicted using DNA concentrations and microbial di-versities. DNA concentration was measured after DNA extraction using nanodrop and ex-pressed as ng/µl. Microbial loads per sample were estimated in the study cohort by applying a linear regression model combining FISH measurements as an outcome and microbial diver-sity and DNA concentration as predictors. This model was then applied to both replication cohorts using the function predict in order to estimate microbial loads per sample.
Microbial taxa were normalized using an inverse rank transformation, and species were only considered for analysis if they were present in at least 10% of the samples in each cohort. The associations between phenotypes and taxa abundances were calculated using Spearman correlations. The package ppcor (v.1.1) was used to calculate partial correlations between each taxa and host phenotypes after correcting for the microbial loads differences per sample. Bonferroni multiple testing correlation was used as a significance threshold, and associations with a p-value <0.00038 (0.05/number of species tested (n=131) were considered significant.
Step by step description of the statistical tests can be found in:
Comparison of the microbial loads per sample between individuals with and without colectomy or sto-ma. Microbial loads are represented as the log10 FISH counts per gram of faecal material
Contribution of each phenotype to the inter-individual variation in the discovery cohort (n=70). The ex-plained variance is estimated for each phenotype individually by performing a PERMANOVA analysis on the Bray-Curtis dissimilarity metrics.
Correlation between the number of resections of the ileum per participant and the bacterial loads in the faecal sample.
Partial correlation between species and phenotypes.
Panels a-d represent the p-values of the correlation between taxa abundances (invert rank transformed) and the sex of the participant. Panels e-g represent the p-values of the correlation between taxa abun-dances (invert rank transformed) and age. X-axis indicates the -log10 p-value of taxa–phenotype cor-relation and the y-axis the -log10 p-value of the partial corcor-relation correcting for microbial loads per sample. In the study cohort (a,b,d and f), panels a and d represent the partial correlation considering the
microbial loads measured by fl uorescence in-situ hybridisation (FISH), and panels b and f, when con-sidering the microbial loads estimated from metagenomic data. Panels c, d and f represents the p-value relation in the replication cohorts. The dashed red line indicates the Bonferroni corrected signifi cance level (-log10(0.05/131).
https://github.com/ArnauVich/Supplementary_tables/blob/master/C5.zip
1. F. Teng et al., Impact of DNA extraction method and targeted 16S-rRNA hypervariable region on oral microbiota profiling. Sci. Rep. 8, 16321 (2018).
2. J. M. Choo, L. E. Leong, G. B. Rogers, Sample storage conditions significantly influence faecal microbiome profiles. Sci. Rep. 5, 16350 (2015). 3. J. Ferrand et al., Comparison of seven methods for extraction of bacterial DNA from fecal and cecal samples of mice. J. Microbiol. Methods. 105, 180–185 (2014).
4. D. Vandeputte et al., Stool consistency is strongly associated with gut microbiota richness and composition, enterotypes and bacterial growth rates. Gut. 65, 57–62 (2016).
5. E. F. Tigchelaar et al., Gut microbiota composition associated with stool consistency. Gut. 65, 540–542 (2016).
6. D. Vandeputte et al., Quantitative microbiome profiling links gut community variation to microbial load. Nature. 551, 507–511 (2017). 7. R. Props et al., Absolute quantification of microbial taxon abundances. ISME J. 11, 584–587 (2017).
8. S. Nayfach, K. S. Pollard, Toward Accurate and Quantitative Comparative Metagenomics. Cell. 166, 1103–1116 (2016).
9. S. Sanna et al., Causal relationships among the gut microbiome, short-chain fatty acids and metabolic diseases. Nat. Genet. 51, 600 (2019). 10. L. Jostins et al., Host-microbe interactions have shaped the genetic architecture of inflammatory bowel disease. Nature. 491, 119–124 (2012). 11. D. Gevers et al., The treatment-naive microbiome in new-onset Crohn’s disease. Cell Host Microbe. 15, 382–392 (2014).
12. M. Lopez-Siles et al., Mucosa-associated Faecalibacterium prausnitzii and Escherichia coli co-abundance can distinguish Irritable Bowel Syndrome and Inflammatory Bowel Disease phenotypes. Int. J. Med. Microbiol. 304, 464–475 (2014).
13. A. Vich Vila et al., Gut microbiota composition and functional changes in inflammatory bowel disease and irritable bowel syndrome. Sci.
Transl. Med. 10, eaap8914 (2018).
14. J. Halfvarson et al., Dynamics of the human gut microbiome in inflammatory bowel disease. Nat. Microbiol. 2, 17004 (2017).
15. F. Imhann et al., Interplay of host genetics and gut microbiota underlying the onset and clinical presentation of inflammatory bowel disease.
Gut. 67, 108–119 (2018).
16. F. Imhann et al., The 1000IBD project: multi-omics data of 1000 inflammatory bowel disease patients; data release 1. BMC Gastroenterol. 19, 5 (2019).
17. E. F. Tigchelaar et al., Cohort profile: LifeLines DEEP, a prospective, general population cohort study in the northern Netherlands: study design and baseline characteristics. BMJ Open. 5, e006772 (2015).
Supplementary tables and figures available at:
18. H. Sokol et al., Faecalibacterium prausnitzii is an anti-inflammatory commensal bacterium identified by gut microbiota analysis of Crohn disease patients. Proc. Natl. Acad. Sci. U. S. A. 105, 16731–16736 (2008).
19. X. C. Morgan et al., Dysfunction of the intestinal microbiome in inflammatory bowel disease and treatment. Genome Biol. 13, R79 (2012). 20. J. Lloyd-Price et al., Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases. Nature. 569, 655 (2019).
21. A. B. Hall et al., A novel Ruminococcus gnavus clade enriched in inflammatory bowel disease patients. Genome Med. 9 (2017), doi:10.1186/ s13073-017-0490-5.
22. H. M. Roager et al., Colonic transit time is related to bacterial metabolism and mucosal turnover in the gut. Nat. Microbiol. 1, 16093 (2016). 23. G. B. Gloor, J. R. Wu, V. Pawlowsky-Glahn, J. J. Egozcue, It’s all relative: analyzing microbiome data as compositions. Ann. Epidemiol. 26, 322–329 (2016).
24. G. B. Gloor, J. M. Macklaim, V. Pawlowsky-Glahn, J. J. Egozcue, Microbiome Datasets Are Compositional: And This Is Not Optional. Front.
Microbiol. 8 (2017), doi:10.3389/fmicb.2017.02224.
25. J. Z. H. von Martels et al., Riboflavin suppresses inflammation and attenuates Crohn’s disease symptoms (RISE-UP Study). (Submitted). 26. H. J. M. Harmsen, G. C. Raangs, T. He, J. E. Degener, G. W. Welling, Extensive Set of 16S rRNA-Based Probes for Detection of Bacteria in Human Feces. Appl. Environ. Microbiol. 68, 2982–2990 (2002).
