Microbes with higher metabolic independence are enriched in human gut microbiomes under stress

(a) Description of studies/cohorts providing publicly available gut metagenomes from healthy people, non-inflammatory bowel disease (IBD) controls, and people with IBD. For each study, we note the sample groups it contributes metagenomes to; whether or not those samples were sufficiently deeply sequenced to be included in the main analyses; the country of origin of the samples; the sample type (fecal metagenome or ileal pouch luminal aspirate); the number of samples it contributes to each group before and after applying the sequencing depth threshold; and cohort details/exclusions as described within the study. (b) Description of 408 samples included in the primary analyses of this manuscript (i.e. those with sufficient sequencing depth of ≥ 25 million reads), including their associated diagnosis (ulcerative colitis (UC), Crohn’s disease (CD), non-IBD, healthy, colorectal cancer with adenoma (CRC_ADENOMA), or colorectal cancer with carcinoma (CRC_CARCINOMA)); study of origin; sample group; sequencing depth; and number of microbial populations estimated to be represented within the metagenome. (c) Description of all samples initially considered and their SRA accession numbers. (d) The number of gene calls and the number/proportion of annotations per gene call for KOfams, Clusters of Orthologous Groups (COGs), and Pfams in each sample. (e) The number of genes with at least one functional annotation and the number of tRNAs in each sample from the subset of deeply sequenced samples. (f) Description of the 57 antibiotic time-series gut metagenomes from Palleja et al., 2018 used for classifier testing, including SRA accession number; sampling day in the time series; sequencing depth; and estimated numbers of microbial populations represented in the sample.

(a) Description of the 33 KEGG modules enriched in inflammatory bowel disease (IBD) samples, including: module name, KEGG categorization, and definition; their median per-population copy numbers (PPCNs) in the healthy sample group and IBD sample group; the p-value, FDR-adjusted p-value, and W statistic from the per-module Wilcoxon rank-sum test used to determine enrichment in IBD; the difference between its median PPCN in IBD samples and median PPCN in healthy samples (‘effect size’); the fraction of samples in which the module occurs with nonzero copy number; whether the module is also enriched in the high metabolic independence (HMI) populations analyzed in Watson et al., 2023; the number of total enzymes in the module; the number of total compounds in the modules; and the numbers and proportions of shared enzymes or compounds between this module and the other IBD-enriched modules. (b) Description of all 179 KEGG modules with nonzero copy number in at least one metagenome. Most of the columns match the corresponding column in sheet (b) with the exception of the ‘enrichment status’ column, which indicates whether the module was found to be enriched in the IBD samples in this study (‘IBD_ENRICHED’), in the high-metabolic independence genomes in Watson et al., 2023 (‘HMI_ENRICHED’), in both (‘HMI_AND_IBD’), or in neither (‘OTHER’). (c) Matrix of stepwise copy number of each module in each deeply sequenced gut metagenome. (d) Per-population copy number of each module in each deeply sequenced gut metagenome in the IBD, non-IBD, and healthy sample groups. (e) Per-population copy number of each module in each antibiotic time-series sample from Palleja et al., 2018.

(a) List of 338 GTDB representative genomes identified as gut microbes, their taxonomy, metabolic independence score, classification as high metabolic independence (‘HMI’) or not (‘non-HMI’), genome length in base pairs, and number of gene calls. (b) Matrix of stepwise completeness of each module in each genome. (c) Matrix of genome detection in each deeply sequenced gut metagenome in the inflammatory bowel disease (IBD), non-IBD, and healthy sample groups. (d) Percent abundance of each genome in each deeply sequenced gut metagenome. (e) Per-genome proportion of samples from each sample group that the genome is detected in using a threshold of 50% (i.e. at least half of the genome sequence is covered by at least one sequencing read in a given sample). (f) Per-sample proportion of detected genomes that are classified as HMI. (g) Average completion of each IBD-enriched module within the HMI genome group and the non-HMI genome group, as well as the difference between these values. (h) Genome-level results when using different HMI score thresholds for determining HMI status. Each threshold is shown both as the average percent completeness required for the 33 IBD-enriched modules and as the HMI score above which a genome is considered HMI. Results for each threshold include the number of genomes assigned as HMI, the percent of genomes assigned as HMI (out of 338), the number and percent of Bacteroides genomes assigned as HMI, the mean genome size and mean number of gene calls for both HMI and non-HMI genomes, the p-values and W statistics of the Wilcoxon rank-sum tests comparing detection of HMI vs non-HMI genomes (1) in IBD samples and (2) in healthy samples, and the p-value of the Kruskal-Wallis rank-sum test comparing the fraction of genomes classified as HMI in IBD samples vs healthy/non-IBD samples.

(a) Details and performance of previously published classifiers for inflammatory bowel disease (IBD) and IBD subtypes. For each classifier, we summarize the cohort details as described by the study; the size of training datasets and validation datasets (if any); the type(s) of samples, data, and extracted features used for classification; the target classes (i.e. what the samples were being classified as); the classifier type and training/validation strategy; and the performance metrics as reported by the study. (b) Classification of each (Palleja et al., 2018) metagenome by our logistic regression model trained for distinguishing IBD vs healthy samples on the basis of PPCN data for IBD-enriched modules. This table describes whether the sample was classified as healthy (‘HEALTHY’) or stressed (‘IBD’, which we consider to be equivalent to an identification of gut stress), and also whether the sample had low sequencing depth (<25 million reads) or not. (c) Summary of the performance of our metagenome classifier across different training/validation strategies using the IBD and healthy metagenome samples. It also includes the details of our final classifier trained on all 330 samples, though performance data is not available for this model since there were no IBD/healthy samples left for validation – however, see manuscript for its performance on the (Palleja et al., 2018) antibiotic time-series dataset. The subsequent sheets include per-fold data and performance information for each train-test strategy: (d) random split cross-validation (25-fold) on PPCN data; (e) leave-two-studies-out cross-validation (24-fold); and (f) (10-fold) cross-validation leaving out samples from the two dominating studies in our dataset (Le Chatelier et al., 2013; Vineis et al., 2016).

For each tool (including the one published in this study), we summarize: the software category (based upon the tool’s architecture and mode of use); its metabolism reconstruction strategy (whether it is a pathway prediction tool or a modeling tool or both); the data source(s) it uses for enzyme and metabolic pathway information; how it calculates pathway completeness or generates models (depending on reconstruction strategy); what input and output types it accepts/generates; any additional capabilities as advertised by the tool’s publication; whether or not the tool is open-source; the program type; and what language(s) it is developed in (if known). The reference publication and code repository or webpage for each tool is also included.

(a) Distribution (mean and standard deviation) of PPCN and PPCN error (computed relative to either average genomic completeness or average genomic copy number) across each validation test case, as well as the proportion of correct, off-by-one, and off-by-two community size estimates in each test case. (b) Spearman’s correlation test results between sample parameters (genome size, community size, and diversity level) and important values computed in our approach (PPCN, PPCN accuracy metrics, and accuracy of community size estimates). Negative values are shown in red and nonsignificant p-values are highlighted in blue. (c) Normalized and rank-ordered relative abundance data from the top 20 most abundant microbial populations in healthy gut metagenomes that we used to recreate a ‘typical’ relative abundance curve for our simulated metagenomes, based upon data from Beghini et al., 2021. Each initial column provides the data from a single sample, and the final four columns describe: the average relative abundance at each rank order; the averages when scaled such that the minimum abundance is 1; the corresponding (integer) coverage values for each scaled average relative abundance value; and the coverage values when increased by 20× for sufficient sequencing depth for assembly.