CoChat

Gut microbiome stability and dynamics in healthy donors and patients with non-gastrointestinal cancers

Open access
Byrd, Allyson L., Liu, Menghan, Fujimura, Kei E., Lyalina, Svetlana, Nagarkar, Deepti R., Charbit, Bruno, Bergstedt, Jacob, Patin, Etienne, Harrison, Oliver J., Quintana-Murci, Lluís, Mellman, Ira, Duffy, Darragh, Albert, Matthew L., the Milieu Intérieur Consortium · 2021 · Journal of Experimental Medicine

Published version of a preprint

The DOI matches its Crossref record; 98 of the 98 references it cites were found in a registry or an index; 15 of the 47 sources its claims cite were checked for support.

Abstract

As microbial therapeutics are increasingly being tested in diverse patient populations, it is essential to understand the host and environmental factors influencing the microbiome. Through analysis of 1,359 gut microbiome samples from 946 healthy donors of the Milieu Intérieur cohort, we detail how microbiome composition is associated with host factors, lifestyle parameters, and disease states. Using a genome-based taxonomy, we found biological sex was the strongest driver of community composition. Additionally, bacterial populations shift across decades of life (age 20–69), with Bacteroidota species consistently increased with age while Actinobacteriota species, including Bifidobacterium, decreased. Longitudinal sampling revealed that short-term stability exceeds interindividual differences. By accounting for these factors, we defined global shifts in the microbiomes of patients with non-gastrointestinal tumors compared with healthy donors. Together, these results demonstrated that the microbiome displays predictable variations as a function of sex, age, and disease state. These variations must be considered when designing microbiome-targeted therapies or interpreting differences thought to be linked to pathophysiology or therapeutic response.

Overview

CoChat Summary

TL;DR

Byrd et al. profile the gut microbiome in 946 healthy French donors using 1,359 shotgun metagenomic samples and show that age and sex are major correlates of community composition. They also find that patients with non-gastrointestinal cancers have altered gut communities, including higher Bacteroidota/Firmicutes_A ratios and lower Shannon diversity than healthy donors.

Key findings

  • Among 154 metadata variables, 52 were associated with species-level Bray–Curtis composition (PERMANOVA FDR < 0.05), with age and sex the top contributors; together these factors explained <10% of population variability. Age and sex were themselves associated with 44 and 24 other metadata variables, respectively.
  • Sex differences included 71 differentially abundant species (FDR < 0.05), with 5 more abundant in females and 18 more abundant in males at log2 fold change >1; 11 of 32 prevalent *Prevotella* species were more abundant in males, and the creatinine degradation I pathway was most strongly enriched in men.
  • Age was associated with 192/485 abundant species (FDR < 0.05), including increased Bacteroidota and decreased Actinobacteriota, with 15 *Bifidobacterium* species declining with age; 108 prevalent pathways correlated with age (31 increased, 77 decreased), including decreased lactose/galactose degradation pathways.
  • In 413 donors sampled 17 ± 3.3 d apart, intraindividual differences were smaller than interindividual differences, but stability varied; baseline Shannon diversity was positively associated with stability (FDR = 0.014), while BMI and circulating triglycerides were negatively associated and sweet items and raw fruit positively associated (P < 0.05).
  • Across 375 samples from 283 patients with non-GI cancers in five cohorts, cancer patients had increased Bacteroidota/Firmicutes_A ratios and decreased Shannon diversity versus healthy donors; on checkpoint inhibitor treatment they were on average more stable than MI donors, and several Enterobacteriaceae species were more prevalent.

Methods

  • The Milieu Intérieur cohort included 946 healthy French donors (originally 500 women and 500 men, 20–69 yr) with 1,359 shotgun metagenomic samples; 413 donors provided longitudinal samples 17 ± 3.3 d apart. Metadata included 154 curated demographic, clinical, and dietary variables.
  • A custom Kraken2/Bracken pipeline was built using 23,505 RefSeq genomes with Genome Taxonomy Database (GTDB) taxonomy; HUMANn2 was used for MetaCyc pathway profiles. The pipeline was also applied to an Israeli cohort (1,159 samples/851 donors), five non-GI cancer cohorts (375 samples/283 patients), and IBD cohorts.
  • Analyses included PERMANOVA/adonis, DESeq2, Spearman correlations, generalized linear models, beta regression, Bray–Curtis and Jaccard distances, with Benjamini–Hochberg FDR correction; one baseline sample per donor was used except for longitudinal analyses.

Limitations

  • The paper states that the cross-sectional design makes it difficult to disentangle age associations from variables correlated with age, and that Bristol stool score and chromogranin A levels were absent. It also notes that cancer analyses lacked extensive metadata, had batch effects across cohorts, and could not separate cancer itself from lifestyle changes or therapies; attempts to associate PC1/PC2 with therapy response were unsuccessful and response definitions varied.
  • Evident from the text: the healthy cohort was restricted to self-reported metropolitan French origin, ages 20–69, BMI 18.5–32 kg/m², and stringent health exclusions, limiting generalizability. Cancer cohorts were heterogeneous across studies, with missing age/sex data for some patients and technical/geographic differences that could affect comparisons.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1084/jem.20200606

Cites

98 of 98 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Graphical abstract
Graphical abstract CC BY · PubMed Central
Figure S1. Quality control and Kraken analysis pipeline. Related to Fig. 1. (A) Pie chart indicating the distribution of genomes in RefSeq (as of June 25, 2019) to different species, based on GTDB taxonomy. 29 species represent >60% of the reference genomes. (B) Table of the 10 species with the greatest number of genomes in RefSeq. (C) Steps taken to quality control shotgun metagenomic reads. Additional details can be found in Materials and methods. (D) Bar plots show the number of paired-end reads remaining after each filtration step in C. Each point corresponds to one sample (n = 1,359). Colors correspond to steps in C. (E) Summary table for the number of paired-end reads in D. max, maximum; min, minimum. (F) Steps taken to run Kraken2 and Bracken2 on the quality-controlled reads, including removal of putative reagent contaminants. Additional details can be found in Materials and methods. See also Data S1, table 2.
Figure S1. Quality control and Kraken analysis pipeline. Related to Fig. 1. (A) Pie chart indicating the distribution of genomes in RefSeq (as of June 25, 2019) to different species, based on GTDB taxonomy. 29 species represent >60% of the reference genomes. (B) Table of the 10 species with the greatest number of genomes in RefSeq. (C) Steps taken to quality control shotgun metagenomic reads. Additional details can be found in Materials and methods. (D) Bar plots show the number of paired-end reads remaining after each filtration step in C. Each point corresponds to one sample (n = 1,359). Colors correspond to steps in C. (E) Summary table for the number of paired-end reads in D. max, maximum; min, minimum. (F) Steps taken to run Kraken2 and Bracken2 on the quality-controlled reads, including removal of putative reagent contaminants. Additional details can be found in Materials and methods. See also Data S1, table 2.CC BY · PubMed Central
Figure S2. Comparison of all program–database combinations tested. Related to Fig. 1. (A) Table indicates the richness across taxonomic levels of the different databases tested. The NCBI and GTDB databases used by Kraken2 include the same 23,505 RefSeq genomes with taxonomies assigned either by NCBI or the GTDB, respectively. GTDB taxonomic assignments often lead to collapsing or splitting of those taxa originally in NCBI; e.g., Firmicutes was split into Firmicutes, Firmicutes_A, Firmicutes_B, Firmicutes_C, etc. (B) Boxplots compare the percentage of shotgun metagenomic reads mapped by the different program–database combinations. Percentage of mapped reads for Metaphlan2 was based on its estimated number of reads from the clade; ****, P < 2.22 × 10−16 by Wilcoxon rank sum. (C) Boxplots compare the richness of taxa identified per sample with the different program–database combinations tested. All pairwise comparisons had P < 0.0001 by Wilcoxon rank sum. (D) Bars indicate the number of unique taxa present in the NCBI (purple) and GTDB (red) databases for each of the top phyla across taxonomic levels. (E) Relative abundance of the top phyla across different program–database combinations. Each bar corresponds to one sample. Samples are ordered based on the relative abundance of Firmicutes_A found by Kraken2-GTDB.
Figure S2. Comparison of all program–database combinations tested. Related to Fig. 1. (A) Table indicates the richness across taxonomic levels of the different databases tested. The NCBI and GTDB databases used by Kraken2 include the same 23,505 RefSeq genomes with taxonomies assigned either by NCBI or the GTDB, respectively. GTDB taxonomic assignments often lead to collapsing or splitting of those taxa originally in NCBI; e.g., Firmicutes was split into Firmicutes, Firmicutes_A, Firmicutes_B, Firmicutes_C, etc. (B) Boxplots compare the percentage of shotgun metagenomic reads mapped by the different program–database combinations. Percentage of mapped reads for Metaphlan2 was based on its estimated number of reads from the clade; ****, P < 2.22 × 10−16 by Wilcoxon rank sum. (C) Boxplots compare the richness of taxa identified per sample with the different program–database combinations tested. All pairwise comparisons had P < 0.0001 by Wilcoxon rank sum. (D) Bars indicate the number of unique taxa present in the NCBI (purple) and GTDB (red) databases for each of the top phyla across taxonomic levels. (E) Relative abundance of the top phyla across different program–database combinations. Each bar corresponds to one sample. Samples are ordered based on the relative abundance of Firmicutes_A found by Kraken2-GTDB.CC BY · PubMed Central
Figure 1. Interindividual variation of bacterial composition is associated with many factors. (A) Boxplots of the top eight phyla. Each dot corresponds to one donor. Firmicutes_A, Firmicutes_C, and Firmicutes were split into unique phyla by the GTDB. (B) MDS plots of BC distance of bacterial species composition. Ordination was primarily driven by the top two phyla Firmicutes_A and Bacteroidota. Each dot corresponds to one donor while color indicates relative abundance of each phyla. (C) In total, 52 factors (Benjamini–Hochberg FDR < 0.05) were associated with interindividual variation of the gut microbiome. The bar plots indicate the amount of interindividual variance explained by each factor for the species and genera level BC distance. Variables are ordered by the percentage variance explained in Kraken species. Colors of the bars correspond to the broad metadata category. The rectangles to the left indicate the statistical strength, as measured by FDR, of the association. For each variable, samples with NA values were excluded. For all these analyses, one sample per donor was used: when available, V1; if not V2. See also Fig. S1, Fig. S2, Fig. S3, and Data S1, tables 1–7. CMV, cytomegalovirus; CRP, C-reactive protein; EUR, Euro; HDL, high-density lipoprotein; MCV, mean corpuscular volume; MMRV, measles, mumps, rubella, and varicella vaccine.
Figure 1. Interindividual variation of bacterial composition is associated with many factors. (A) Boxplots of the top eight phyla. Each dot corresponds to one donor. Firmicutes_A, Firmicutes_C, and Firmicutes were split into unique phyla by the GTDB. (B) MDS plots of BC distance of bacterial species composition. Ordination was primarily driven by the top two phyla Firmicutes_A and Bacteroidota. Each dot corresponds to one donor while color indicates relative abundance of each phyla. (C) In total, 52 factors (Benjamini–Hochberg FDR < 0.05) were associated with interindividual variation of the gut microbiome. The bar plots indicate the amount of interindividual variance explained by each factor for the species and genera level BC distance. Variables are ordered by the percentage variance explained in Kraken species. Colors of the bars correspond to the broad metadata category. The rectangles to the left indicate the statistical strength, as measured by FDR, of the association. For each variable, samples with NA values were excluded. For all these analyses, one sample per donor was used: when available, V1; if not V2. See also Fig. S1, Fig. S2, Fig. S3, and Data S1, tables 1–7. CMV, cytomegalovirus; CRP, C-reactive protein; EUR, Euro; HDL, high-density lipoprotein; MCV, mean corpuscular volume; MMRV, measles, mumps, rubella, and varicella vaccine.CC BY · PubMed Central
Figure S3. 154 variables were associated with bacterial profiles. Related to Fig. 1. (A) Distribution of variables across broad categories. (B) Distribution of 64 continuous variables across donors. Colors correspond to those in A. (C) Percentage of donors for whom each of the 90 binary variables was true. Colors correspond to those in A. 2w, 2 wk; G/L, billion cells per liter; HBV, hepatitis B virus; MCHC, mean corpuscular hemoglobin concentration; mon, month; PC, principal component of genetics SNP array. (D) 24 variables were associated with sex with Spearman ρ > 0.2 or less than −0.2. Corr, correlation. (E) 44 variables were associated with age with Spearman ρ > 0.2 or less than −0.2. See also Data S1, tables 5 and 6. CMV, cytomegalovirus; CRP, C-reactive protein; EUR, Euro; HDL, high density lipoprotein; MCV, mean corpuscular volume; MMRV, measles, mumps, rubella, and varicella vaccine; PCA, principal component ancestry.
Figure S3. 154 variables were associated with bacterial profiles. Related to Fig. 1. (A) Distribution of variables across broad categories. (B) Distribution of 64 continuous variables across donors. Colors correspond to those in A. (C) Percentage of donors for whom each of the 90 binary variables was true. Colors correspond to those in A. 2w, 2 wk; G/L, billion cells per liter; HBV, hepatitis B virus; MCHC, mean corpuscular hemoglobin concentration; mon, month; PC, principal component of genetics SNP array. (D) 24 variables were associated with sex with Spearman ρ > 0.2 or less than −0.2. Corr, correlation. (E) 44 variables were associated with age with Spearman ρ > 0.2 or less than −0.2. See also Data S1, tables 5 and 6. CMV, cytomegalovirus; CRP, C-reactive protein; EUR, Euro; HDL, high density lipoprotein; MCV, mean corpuscular volume; MMRV, measles, mumps, rubella, and varicella vaccine; PCA, principal component ancestry.CC BY · PubMed Central
Figure 2. Microbial profiles of MI donors in comparison to those from Zeevi et al. (2015). (A) Bars indicate the number of females and males in both cohorts; P value by Fisher’s exact test; ***, P < 0.001. (B) Density plots show distribution of BMI and age in both cohorts; P value by Wilcoxon rank sum; *, P < 0.05; ***, P < 0.001. (C) Bar plots compare sequencing depth across cohorts. (D) Bar plots show percentage of reads unmapped after the Kraken2-GTDB pipeline. (E) Boxplots show richness across taxonomic levels. Each dot corresponds to one donor. (C–E) P values by Wilcoxon rank sum; ****, P < 0.0001. (F) Association of relative abundances of each taxon across taxonomic levels with Spearman correlation. Each dot corresponds to one taxon. Only taxa present in >5% of either cohort were considered. Log relative abundances are shown. (G) MDS ordination plot of all samples in both the MI and Zeevi cohorts. Each dot corresponds to one donor and is colored by the cohort. R2 indicates the amount of interindividual variation (calculated with BC) explained by the cohort and was calculated with the PERMANOVA test adonis. (H) MDS ordination plot of samples in both the MI and Zeevi cohorts. Each dot corresponds to one donor and is colored by the relative abundance of the phyla. (I) Bars indicate the amount of interindividual variation (calculated with BC) explained by each of the variables in each of the cohorts. See also Data S1, table 8.
Figure 2. Microbial profiles of MI donors in comparison to those from Zeevi et al. (2015). (A) Bars indicate the number of females and males in both cohorts; P value by Fisher’s exact test; ***, P < 0.001. (B) Density plots show distribution of BMI and age in both cohorts; P value by Wilcoxon rank sum; *, P < 0.05; ***, P < 0.001. (C) Bar plots compare sequencing depth across cohorts. (D) Bar plots show percentage of reads unmapped after the Kraken2-GTDB pipeline. (E) Boxplots show richness across taxonomic levels. Each dot corresponds to one donor. (C–E) P values by Wilcoxon rank sum; ****, P < 0.0001. (F) Association of relative abundances of each taxon across taxonomic levels with Spearman correlation. Each dot corresponds to one taxon. Only taxa present in >5% of either cohort were considered. Log relative abundances are shown. (G) MDS ordination plot of all samples in both the MI and Zeevi cohorts. Each dot corresponds to one donor and is colored by the cohort. R2 indicates the amount of interindividual variation (calculated with BC) explained by the cohort and was calculated with the PERMANOVA test adonis. (H) MDS ordination plot of samples in both the MI and Zeevi cohorts. Each dot corresponds to one donor and is colored by the relative abundance of the phyla. (I) Bars indicate the amount of interindividual variation (calculated with BC) explained by each of the variables in each of the cohorts. See also Data S1, table 8.CC BY · PubMed Central
Figure 3. Taxa, particularly Prevotella, were differentially present between males and females. (A) Volcano plot of 485 abundant species, of which 71 were differentially abundant between males and females based on DESeq2 (FDR < 0.05). Each species is colored by its taxonomic phyla. (B) 12 Prevotella species were more abundant in males consistently across cohorts. Left: Log2 fold change (log2FC) of species in males versus females. Middle panel indicates the FDR value. Notably, even when a species was significant in only one cohort, the direction was consistent in the other. Right: Prevalence of each species. Color indicates the sex, while shape indicates the cohort. (C) Boxplots show richness of Prevotella species in males and females. (D) Abundance measure in counts per million (CPM) and prevalence of the pathway CRNFORCAT-PWY: creatinine degradation I in males and females. (E) Blood creatinine levels in males and females. (C–E) P values by Wilcoxon rank sum; ***, P < 0.001, ****, P < 0.0001. (F) In the scatter plot, dots represent the 189 species that were significantly correlated with CRNFORCAT-PWY. Of those, species that were significantly differentially abundant between males and females are colored by their phyla label. Species not associated with sex are colored gray. Size of the point corresponds to the −log10(FDR) of the species by sex correlation. (G) Scatter plots show the correlation of Holdemanella biformis relative abundance and CPM of CRNFORCAT (left) or circulating creatinine levels (right). Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. See also Data S1, tables 9–11.
Figure 3. Taxa, particularly Prevotella, were differentially present between males and females. (A) Volcano plot of 485 abundant species, of which 71 were differentially abundant between males and females based on DESeq2 (FDR < 0.05). Each species is colored by its taxonomic phyla. (B) 12 Prevotella species were more abundant in males consistently across cohorts. Left: Log2 fold change (log2FC) of species in males versus females. Middle panel indicates the FDR value. Notably, even when a species was significant in only one cohort, the direction was consistent in the other. Right: Prevalence of each species. Color indicates the sex, while shape indicates the cohort. (C) Boxplots show richness of Prevotella species in males and females. (D) Abundance measure in counts per million (CPM) and prevalence of the pathway CRNFORCAT-PWY: creatinine degradation I in males and females. (E) Blood creatinine levels in males and females. (C–E) P values by Wilcoxon rank sum; ***, P < 0.001, ****, P < 0.0001. (F) In the scatter plot, dots represent the 189 species that were significantly correlated with CRNFORCAT-PWY. Of those, species that were significantly differentially abundant between males and females are colored by their phyla label. Species not associated with sex are colored gray. Size of the point corresponds to the −log10(FDR) of the species by sex correlation. (G) Scatter plots show the correlation of Holdemanella biformis relative abundance and CPM of CRNFORCAT (left) or circulating creatinine levels (right). Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. See also Data S1, tables 9–11.CC BY · PubMed Central
Figure 4. Bacterial profiles are dynamic across decades of life. (A) Abundance of the five most abundant phyla across decades of life. Curves show 95% confidence intervals and were modeled with LOESS regression. Stars in the figure legend indicate those phyla statistically associated with age. (B) GraPhlAn taxonomic tree of the 454 species in the top five phyla found at significant prevalence and abundance across all donors. Relative abundance of taxa in red were decreased with age, while those in blue increased. Association between taxa relative abundance and age was determined by Spearman correlation (FDR < 0.05). The heatmap on the outer ring indicates the strength of the correlation (Spearman ρ). Genera with at least four species associated with age are labeled. (C) Scatter plots compare the Spearman ρ values of bacteria ~ age for males and females, one point = one species. In the bottom plot, points are colored based on whether the correlation was significant (FDR < 0.05) in males, females, or both. In the top plot, points are colored based on the phyla designation of the species. (D) Venn diagram comparing the bacterial species statistically (FDR < 0.05) associated with age in the MI and Zeevi et al. (2015) cohorts. See also Data S1, tables 12 and 13.
Figure 4. Bacterial profiles are dynamic across decades of life. (A) Abundance of the five most abundant phyla across decades of life. Curves show 95% confidence intervals and were modeled with LOESS regression. Stars in the figure legend indicate those phyla statistically associated with age. (B) GraPhlAn taxonomic tree of the 454 species in the top five phyla found at significant prevalence and abundance across all donors. Relative abundance of taxa in red were decreased with age, while those in blue increased. Association between taxa relative abundance and age was determined by Spearman correlation (FDR < 0.05). The heatmap on the outer ring indicates the strength of the correlation (Spearman ρ). Genera with at least four species associated with age are labeled. (C) Scatter plots compare the Spearman ρ values of bacteria ~ age for males and females, one point = one species. In the bottom plot, points are colored based on whether the correlation was significant (FDR < 0.05) in males, females, or both. In the top plot, points are colored based on the phyla designation of the species. (D) Venn diagram comparing the bacterial species statistically (FDR < 0.05) associated with age in the MI and Zeevi et al. (2015) cohorts. See also Data S1, tables 12 and 13.CC BY · PubMed Central
Figure 5. Bifidobacterium species and Bifidobacterium-associated pathways decrease with age. (A) Lines indicate the percent prevalence of 15 Bifidobacterium species across the different decades. Only Bifidobacterium species significantly associated with age are shown. (B) Boxplots show richness of Bifidobacterium species in donors grouped by their age. (C) Correlation plot of Bifidobacteria species (prevalence >30%) and their cocorrelated Collinsella species (Spearman ρ > 0.4). (D) 15 Bifidobacterium species were associated with age across cohorts. Left: Correlation of species with age (Spearman ρ). Middle: FDR value. Notably, even when a species was significant in only one cohort, the trend was consistent in the other. Right: Prevalence of each species. Color indicates the cohort. (E) Abundance (counts per million [CPM]) of lactose/galactose degradation pathways decreased with age. Curves show 95% confidence intervals and were modeled with LOESS regression. (F) Abundance (CPM) of l-histidine degradation/synthesis pathways associated with age. Curves show 95% confidence intervals and were modeled with LOESS regression. (G) For the pathways correlated with age in E and F, in the scatter plot, dots represent the species that were significantly correlated with that pathway. Of those, species that were also significantly correlated with age are colored by their phyla label. Species not associated with age are colored gray. Size of the point corresponds to the −log10(FDR) of the species by age correlation. See also Data S1, tables 12–14.
Figure 5. Bifidobacterium species and Bifidobacterium-associated pathways decrease with age. (A) Lines indicate the percent prevalence of 15 Bifidobacterium species across the different decades. Only Bifidobacterium species significantly associated with age are shown. (B) Boxplots show richness of Bifidobacterium species in donors grouped by their age. (C) Correlation plot of Bifidobacteria species (prevalence >30%) and their cocorrelated Collinsella species (Spearman ρ > 0.4). (D) 15 Bifidobacterium species were associated with age across cohorts. Left: Correlation of species with age (Spearman ρ). Middle: FDR value. Notably, even when a species was significant in only one cohort, the trend was consistent in the other. Right: Prevalence of each species. Color indicates the cohort. (E) Abundance (counts per million [CPM]) of lactose/galactose degradation pathways decreased with age. Curves show 95% confidence intervals and were modeled with LOESS regression. (F) Abundance (CPM) of l-histidine degradation/synthesis pathways associated with age. Curves show 95% confidence intervals and were modeled with LOESS regression. (G) For the pathways correlated with age in E and F, in the scatter plot, dots represent the species that were significantly correlated with that pathway. Of those, species that were also significantly correlated with age are colored by their phyla label. Species not associated with age are colored gray. Size of the point corresponds to the −log10(FDR) of the species by age correlation. See also Data S1, tables 12–14.CC BY · PubMed Central
Figure 6. Bacterial profiles are stable over the short term; however, the degree of stability is variable across donors. (A) Boxplots of species BC distances (left), pathway BC distances (middle), and pathway binary Jaccard distances (right) between donors (n = 946, for each individual, the average distance from all other individuals at V1 was used) and within a donor over time (n = 413, time between samples = 17 ± 3.3 d); 1 = samples are completely different, 0 = samples are identical. ****, P < 2.2 × 10−16 by Wilcoxon rank sum. (B) Histogram of 413 donors’ species stability (1 − BC). (C) MDS plot of BC distance of bacterial species composition. Each dot corresponds to one donor who had a second sample. Color indicates longitudinal species stability of that donor (1 − within-sample BC). (D) Scatter plots show the top phyla associated with species stability (FDR < 0.05). (E) Scatter plots show the top phyla associated with pathway stability (FDR < 0.05). (D and E) Each point corresponds to one donor with a longitudinal sample (n = 413). Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. (F) Results of the series of GLM fits aimed at identifying factors associated with intraindividual species stability. Only factors with P value < 0.05 are shown. Std., standard. (G) Scatter plots show the stability of individual species (1 − BC) by their mean baseline relative abundance and prevalence. Each point corresponds to a bacterial species and is colored by the species Phyla. 533 species with mean prevalence >5% are shown. (H) Scatter plots show the stability of individual pathways (1 − BC) by their mean baseline relative abundance and prevalence. Each point corresponds to a pathway. 364 pathways with mean prevalence >5% are shown. (I) Boxplots compare stability of individual species and pathways (1 − BC). Points are colored by the mean prevalence of the species or pathway in both samples. ****, P < 2.2 × 10−16 by Wilcoxon rank sum. See also Data S1, tables 15–18.
Figure 6. Bacterial profiles are stable over the short term; however, the degree of stability is variable across donors. (A) Boxplots of species BC distances (left), pathway BC distances (middle), and pathway binary Jaccard distances (right) between donors (n = 946, for each individual, the average distance from all other individuals at V1 was used) and within a donor over time (n = 413, time between samples = 17 ± 3.3 d); 1 = samples are completely different, 0 = samples are identical. ****, P < 2.2 × 10−16 by Wilcoxon rank sum. (B) Histogram of 413 donors’ species stability (1 − BC). (C) MDS plot of BC distance of bacterial species composition. Each dot corresponds to one donor who had a second sample. Color indicates longitudinal species stability of that donor (1 − within-sample BC). (D) Scatter plots show the top phyla associated with species stability (FDR < 0.05). (E) Scatter plots show the top phyla associated with pathway stability (FDR < 0.05). (D and E) Each point corresponds to one donor with a longitudinal sample (n = 413). Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. (F) Results of the series of GLM fits aimed at identifying factors associated with intraindividual species stability. Only factors with P value < 0.05 are shown. Std., standard. (G) Scatter plots show the stability of individual species (1 − BC) by their mean baseline relative abundance and prevalence. Each point corresponds to a bacterial species and is colored by the species Phyla. 533 species with mean prevalence >5% are shown. (H) Scatter plots show the stability of individual pathways (1 − BC) by their mean baseline relative abundance and prevalence. Each point corresponds to a pathway. 364 pathways with mean prevalence >5% are shown. (I) Boxplots compare stability of individual species and pathways (1 − BC). Points are colored by the mean prevalence of the species or pathway in both samples. ****, P < 2.2 × 10−16 by Wilcoxon rank sum. See also Data S1, tables 15–18.CC BY · PubMed Central
Figure S4. Across studies and indications, cancer patients have altered gut bacterial profiles. Related to Fig. 7. (A) Table indicating the studies included in all cancer versus healthy analyses. For all but the longitudinal analysis, one sample per patient was included. When available, the baseline sample was prioritized. For cancer patients, V1 = baseline, pretreatment; V2-Frankel = within 1 mo of starting ICT; V2-Peters = week 6; V2-Routy = after second injection, ∼1 mo; V2-Peters = week 12; V3-Routy = after fourth injection, ∼2 mo. (B) Bars indicate percentage of males and females in each of the studies. Number indicates the actual number of males and females. Sex data were not provided by Matson et al. (2018). (C) Boxplots compare age of donors across the studies. Age data were not provided by Matson et al. (2018). (D–I) Boxplots compare values indicated on the y axis across the healthy and cancer cohorts. Color corresponds to the study. (J) Table indicating the number of donors in each of the IBD cohorts from the two studies. Patients in iHMP <20 yr old were grouped in the pediatric cohort. (K) PC plot of BC distance of bacterial species composition of donors in Fig. 7 A as well as donors in the IBD cohorts. For the HMP samples, multiple samples from the same donor were aggregated together, so there is only one dot per donor. Color corresponds to the health status. (L) Boxplots compare PC1, PC2, Shannon, and Bacteroidota (Bac)/Firmicutes_A (Firm) ratios of samples in K. On the x axis, “cancer” is all the samples in Fig. 7; the remaining are the IBD cohorts in J. CD, Crohn’s disease; ICT, immune checkpoint inhibitor; NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma; UC, ulcerative colitis.
Figure S4. Across studies and indications, cancer patients have altered gut bacterial profiles. Related to Fig. 7. (A) Table indicating the studies included in all cancer versus healthy analyses. For all but the longitudinal analysis, one sample per patient was included. When available, the baseline sample was prioritized. For cancer patients, V1 = baseline, pretreatment; V2-Frankel = within 1 mo of starting ICT; V2-Peters = week 6; V2-Routy = after second injection, ∼1 mo; V2-Peters = week 12; V3-Routy = after fourth injection, ∼2 mo. (B) Bars indicate percentage of males and females in each of the studies. Number indicates the actual number of males and females. Sex data were not provided by Matson et al. (2018). (C) Boxplots compare age of donors across the studies. Age data were not provided by Matson et al. (2018). (D–I) Boxplots compare values indicated on the y axis across the healthy and cancer cohorts. Color corresponds to the study. (J) Table indicating the number of donors in each of the IBD cohorts from the two studies. Patients in iHMP <20 yr old were grouped in the pediatric cohort. (K) PC plot of BC distance of bacterial species composition of donors in Fig. 7 A as well as donors in the IBD cohorts. For the HMP samples, multiple samples from the same donor were aggregated together, so there is only one dot per donor. Color corresponds to the health status. (L) Boxplots compare PC1, PC2, Shannon, and Bacteroidota (Bac)/Firmicutes_A (Firm) ratios of samples in K. On the x axis, “cancer” is all the samples in Fig. 7; the remaining are the IBD cohorts in J. CD, Crohn’s disease; ICT, immune checkpoint inhibitor; NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma; UC, ulcerative colitis.CC BY · PubMed Central
Figure 7. Cancer patients have altered gut bacterial profiles associated with higher Bacteroidota/Firmicutes_A ratios and lower Shannon diversity. (A) PC plot of BC distance of bacterial species composition of one sample per donor in the table in Fig. S4 A. When available, the baseline sample was prioritized. Each dot corresponds to one donor, while color indicates the study and indication. (B) Same PC analysis plot as A, with dots colored by health status of the donor. Red, cancer; blue, healthy. (C) Boxplots compare PC1 and PC2 values of samples from cancer patients and healthy donors stratified by age. (D) Same PC analysis plot as A with dots colored by relative phyla abundance. (E) Density plots of Bacteroidota and Firmicutes_A were generated using a moving average of the abundance of the phyla within the communities along PC2, with a scale from 0 to the maximum moving average. (F) Boxplots compare the log(Bacteroidota/Firmicutes_A) ratio of samples from cancer patients and healthy donors stratified by age. (G) Scatter plots show PC1 versus Shannon for cancer patients and healthy donors. Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. (H) Boxplots compare bacterial Shannon diversity of samples from cancer patients and healthy donors stratified by age. (I) Boxplots of species stability (1 − BC distance) within a donor over time points indicated in the x axis. For cancer patients, V1 = baseline, pretreatment; V2-Frankel = within 1 mo of starting ICT; V2-Peters = week 6; V2-Routy = after second injection, ∼1 mo; V2-Peters = week 12; V3-Routy = after fourth injection, ∼2 mo. (J) Scatter plots shows Bacteroidota abundance versus stability (1 − BC distance) for cancer and healthy patients. Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. For C, F, H, and I, ns, not significant (P > 0.05); *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001 by Wilcoxon rank sum with FDR correction. See also Fig. S4. ICT, immune checkpoint inhibitor; NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.
Figure 7. Cancer patients have altered gut bacterial profiles associated with higher Bacteroidota/Firmicutes_A ratios and lower Shannon diversity. (A) PC plot of BC distance of bacterial species composition of one sample per donor in the table in Fig. S4 A. When available, the baseline sample was prioritized. Each dot corresponds to one donor, while color indicates the study and indication. (B) Same PC analysis plot as A, with dots colored by health status of the donor. Red, cancer; blue, healthy. (C) Boxplots compare PC1 and PC2 values of samples from cancer patients and healthy donors stratified by age. (D) Same PC analysis plot as A with dots colored by relative phyla abundance. (E) Density plots of Bacteroidota and Firmicutes_A were generated using a moving average of the abundance of the phyla within the communities along PC2, with a scale from 0 to the maximum moving average. (F) Boxplots compare the log(Bacteroidota/Firmicutes_A) ratio of samples from cancer patients and healthy donors stratified by age. (G) Scatter plots show PC1 versus Shannon for cancer patients and healthy donors. Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. (H) Boxplots compare bacterial Shannon diversity of samples from cancer patients and healthy donors stratified by age. (I) Boxplots of species stability (1 − BC distance) within a donor over time points indicated in the x axis. For cancer patients, V1 = baseline, pretreatment; V2-Frankel = within 1 mo of starting ICT; V2-Peters = week 6; V2-Routy = after second injection, ∼1 mo; V2-Peters = week 12; V3-Routy = after fourth injection, ∼2 mo. (J) Scatter plots shows Bacteroidota abundance versus stability (1 − BC distance) for cancer and healthy patients. Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. For C, F, H, and I, ns, not significant (P > 0.05); *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001 by Wilcoxon rank sum with FDR correction. See also Fig. S4. ICT, immune checkpoint inhibitor; NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.CC BY · PubMed Central
Figure S5. Variables associated with cancer-like microbiome. Related to Fig. 8. (A) Boxplots compare PC2 values of samples from males and females stratified by study and age. *, P < 0.05; ***, P < 0.001; ****, P < 0.0001 by Wilcoxon rank sum with FDR correction. (B) Length and color of the bars indicates the Spearman ρ value of each of the variables by ratio of Bacteroidota/Firmicutes_A. Only those variables with an FDR < 0.05 are shown. Stars indicate the P value after correcting for sex and age in a linear model. *, P < 0.05; **, P < 0.01; ****, P < 0.0001. (C) Table shows DESeq2 results for species identified by Atarashi et al. (2013) and Narushima et al. (2014) as potent inducers of regulatory T cells. Only species assigned a name in Narushima et al. (2014) that could then be linked to a representative in our database were included. (D) For the most abundant species in C, bars indicate the mean relative abundance of the species in each study. Error bars indicate mean ± SE. (E) Boxplots compare PC1 and PC2 values of samples from responders and nonresponders across studies. P values by Wilcoxon rank sum with FDR correction; ns, not significant (P > 0.05). Definitions of responders and nonresponders by study: Frankel R, response, stable disease; Frankel NR, progressed; Matson R, complete response, partial response; Matson NR, stable disease, progressed; Routy R, complete response, partial response, or stable disease; Routy NR, progressed or died; Gopalakrishnan R, complete response, partial response, or stable disease for ≥6 mo; Gopalakrishnan NR, progressed or stable disease <6 mo; Peters R, did not progress; Peters NR, progressed. See also Data S1, tables 19–21. NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.
Figure S5. Variables associated with cancer-like microbiome. Related to Fig. 8. (A) Boxplots compare PC2 values of samples from males and females stratified by study and age. *, P < 0.05; ***, P < 0.001; ****, P < 0.0001 by Wilcoxon rank sum with FDR correction. (B) Length and color of the bars indicates the Spearman ρ value of each of the variables by ratio of Bacteroidota/Firmicutes_A. Only those variables with an FDR < 0.05 are shown. Stars indicate the P value after correcting for sex and age in a linear model. *, P < 0.05; **, P < 0.01; ****, P < 0.0001. (C) Table shows DESeq2 results for species identified by Atarashi et al. (2013) and Narushima et al. (2014) as potent inducers of regulatory T cells. Only species assigned a name in Narushima et al. (2014) that could then be linked to a representative in our database were included. (D) For the most abundant species in C, bars indicate the mean relative abundance of the species in each study. Error bars indicate mean ± SE. (E) Boxplots compare PC1 and PC2 values of samples from responders and nonresponders across studies. P values by Wilcoxon rank sum with FDR correction; ns, not significant (P > 0.05). Definitions of responders and nonresponders by study: Frankel R, response, stable disease; Frankel NR, progressed; Matson R, complete response, partial response; Matson NR, stable disease, progressed; Routy R, complete response, partial response, or stable disease; Routy NR, progressed or died; Gopalakrishnan R, complete response, partial response, or stable disease for ≥6 mo; Gopalakrishnan NR, progressed or stable disease <6 mo; Peters R, did not progress; Peters NR, progressed. See also Data S1, tables 19–21. NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.CC BY · PubMed Central
Figure 8. Variables associated with a more cancer-like microbiome. (A) Same PC analysis plot as Fig. 7 A. Each dot corresponds to one donor, while color indicates health status of the donor and their sex. (B) Boxplots compare PC2 values of samples from males and females stratified by study; ns, not significant (P > 0.05); *, P < 0.05; ****, P < 0.0001 by Wilcoxon rank sum with FDR correction. (C) Length and color of the bars indicates the Spearman ρ value of each of the variables by PC2. Only those variables with an FDR < 0.05 are shown. Stars indicate the P value after correcting for sex and age in a linear model. *, P < 0.05; **, P < 0.01; ***, P < 0.001. (D) Scatter plots show PC2 value versus variables from C. Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. (E) Volcano plot of species differentially abundant between cancer patients and healthy donors based on DESeq2. To ensure that results were not driven by rare species, particularly in the large healthy cohorts, we first removed any species not present in at least three of the cancer cohorts at >10% prevalence, and then kept only the 448 species with a mean relative abundance >0.01%. Each species is colored by its taxonomic phyla. (F) Boxplot of Bacteroidota species more abundant in cancer patients than healthy donors. Y axis is on log10 scale. (G) Prevalence of selected Enterobacteriaceae species across healthy and cancer cohorts. Color corresponds to study. See also Fig. S5 and Data S1, tables 19–21. NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.
Figure 8. Variables associated with a more cancer-like microbiome. (A) Same PC analysis plot as Fig. 7 A. Each dot corresponds to one donor, while color indicates health status of the donor and their sex. (B) Boxplots compare PC2 values of samples from males and females stratified by study; ns, not significant (P > 0.05); *, P < 0.05; ****, P < 0.0001 by Wilcoxon rank sum with FDR correction. (C) Length and color of the bars indicates the Spearman ρ value of each of the variables by PC2. Only those variables with an FDR < 0.05 are shown. Stars indicate the P value after correcting for sex and age in a linear model. *, P < 0.05; **, P < 0.01; ***, P < 0.001. (D) Scatter plots show PC2 value versus variables from C. Trend lines show 95% confidence intervals and were modeled with lm. Statistics based on Spearman correlation. (E) Volcano plot of species differentially abundant between cancer patients and healthy donors based on DESeq2. To ensure that results were not driven by rare species, particularly in the large healthy cohorts, we first removed any species not present in at least three of the cancer cohorts at >10% prevalence, and then kept only the 448 species with a mean relative abundance >0.01%. Each species is colored by its taxonomic phyla. (F) Boxplot of Bacteroidota species more abundant in cancer patients than healthy donors. Y axis is on log10 scale. (G) Prevalence of selected Enterobacteriaceae species across healthy and cancer cohorts. Color corresponds to study. See also Fig. S5 and Data S1, tables 19–21. NSCLC, non–small cell lung cancer; RCC, renal cell carcinoma.CC BY · PubMed Central

Verification

Claims the paper makes

9 Valid · 1 Not valid · 26 Not checked · 89 Own results and interpretations

60 more claims

Create a free account to see all 125 claims and to verify the 30 references still open.

See all claims — free account
  • Probably not a substantial claim. 35
  • It may state two separate claims as one. 32
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

4 of 4

Citations

Cites

Show all 98 references

Cited by

See all 63

Metadata

Authors
Byrd, A. L., Liu, M., Fujimura, K. E., Lyalina, S., Nagarkar, D. R., Charbit, B., Bergstedt, J., Patin, E., Harrison, O. J., Quintana-Murci, L., Mellman, I., Duffy, D., Albert, M. L., & the Milieu Intérieur Consortium
Affiliations
  • Genentech (US)
  • Institut Pasteur (FR)
  • Centre National de la Recherche Scientifique (FR)
  • Human Evolutionary Genetics Unit, Institut Pasteur, UMR 2000, Centre National de la Recherche Scientifique, Paris, France 5
  • Benaroya Research Institute (US)
  • insitro, inc. (United States) (US)
Corresponding author
  • Byrd, Allyson L.
  • Duffy, Darragh
Venue
Journal of Experimental Medicine
Published
2020-11-11
Volume
218
Issue
1
ISSN
  • 0022-1007 (ISSN-L)
  • 1540-9538
OpenAlex
W3098540694
Cited by
63
References
82
Citations by year
2026: 7 · 2025: 10 · 2024: 12 · 2023: 10 · 2022: 14 · 2021: 10
FWCI
2.76
Type
journal article
Language
en
Open access
Locations
  • The Journal of Experimental Medicine (publishedVersion, other-oa)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • HAL (submittedVersion, cc-by)
Funders
  • French government
  • Agence Nationale de la Recherche (ANR-10-LABX-69-01)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gut microbiota and health
  • Physical Sciences › Environmental Science › Ecology › Microbial Community Ecology and Physiology
  • Health Sciences › Dentistry › Periodontics › Oral microbiology and periodontitis research
Keywords
gut microbiota, microbiome stability, microbiome dynamics, sex differences, Bacteroidetes, Actinobacteriota, Bifidobacterium, host factors, lifestyle factors, microbiome-targeted therapy
MeSH terms
Gastrointestinal Microbiome, Adult, Aged, Bifidobacterium, Female, Humans, Male, Middle Aged, Neoplasms

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-06, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-06 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Gut microbiome stability and dynamics in healthy donors and patients with non-gastrointestinal cancers. https://app.cochat.ai/papers/10.1084/jem.20200606
The paper
Byrd, A. L., Liu, M., Fujimura, K. E., Lyalina, S., Nagarkar, D. R., Charbit, B., Bergstedt, J., Patin, E., Harrison, O. J., Quintana-Murci, L., Mellman, I., Duffy, D., Albert, M. L., & the Milieu Intérieur Consortium. (2021). Gut microbiome stability and dynamics in healthy donors and patients with non-gastrointestinal cancers. Journal of Experimental Medicine. https://doi.org/10.1084/jem.20200606

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.