CoChat
Sign Up
Cited by 286Gut microbiota and healthGenomics and Phylogenetic StudiesBacterial Identification and Susceptibility Testing

The DOI matches its Crossref record; 64 of the 64 references it cites were found in a registry or an index.

High-throughput microbial culturomics using automation and machine learning

Open access
Huang, Yiming, Sheth, Ravi U., Zhao, Shijie, Cohen, Lucas A., Dabaghi, Kendall, Moody, Thomas, Sun, Yiwei, Ricaurte, Deirdre, Richardson, Miles, Velez-Cortes, Florencia, Blazejewski, Tomasz, Kaufman, Andrew, Ronda, Carlotta, Wang, Harris H.·2023·Nature Biotechnology
DOI 10.1038/s41587-023-01674-2PMC PMC10567565
Open this paper in CoChat — free accountOpen PDF

Abstract

Pure bacterial cultures remain essential for detailed experimental and mechanistic studies in microbiome research, and traditional methods to isolate individual bacteria from complex microbial ecosystems are labor-intensive, difficult-to-scale and lack phenotype-genotype integration. Here we describe an open-source high-throughput robotic strain isolation platform for the rapid generation of isolates on demand. We develop a machine learning approach that leverages colony morphology and genomic data to maximize the diversity of microbes isolated and enable targeted picking of specific genera. Application of this platform on fecal samples from 20 humans yields personalized gut microbiome biobanks totaling 26,997 isolates that represented >80% of all abundant taxa. Spatial analysis on >100,000 visually captured colonies reveals cogrowth patterns between Ruminococcaceae, Bacteroidaceae, Coriobacteriaceae and Bifidobacteriaceae families that suggest important microbial interactions. Comparative analysis of 1,197 high-quality genomes from these biobanks shows interesting intra- and interpersonal strain evolution, selection and horizontal gene transfer. This culturomics framework should empower new research efforts to systematize the collection and quantitative analysis of imaging-based phenotypes with high-resolution genomics data for many emerging microbiome studies.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41587-023-01674-2

Cites

64 of 64 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 A data-driven microbial isolation strategy using phenotypic and morphologic features. a, Framework of phenotype and morphology-driven strain isolation and data collection of the human gut microbiome. Human fecal samples were plated and cultured under different antibiotics selection and morphologically diverse colonies were then isolated, biobanked and analyzed by downstream sequencing. b, Setup of the automated anaerobic microbial isolation and cultivation system CAMII. c, Illustration of morphology-guided colony isolation on CAMII. Colonies grown on plates are imaged under trans- and epi-illumination and subjected to contour segmentation and morphologic features extraction. Data are analyzed by PCA to identify the set of most morphologically diverse colonies that are then isolated by an integrated colony picker. d, Illustration of diverse colony morphology on plates. Colony size and shape features were extracted from transilluminated images, and colony color features were extracted from epi-illuminated images. e, Fecal sample H1t1 were cultured with seven different antibiotics to evaluate their capacity to yield the most unique and diverse bacteria by 16S analysis at the family level. Ciprofloxacin, trimethoprim and vancomycin were selected for subsequent colony isolations. f, Number of unique ASVs obtained from phenotype-guided isolation compared to random isolation of three human fecal samples H1t4, H5t1 and H6t1. Isolation was performed by CAMII; random isolation was performed on a random subset of all detected colonies on the plates, and phenotype-guided isolation was performed on morphology-selected colonies by the algorithm (Supplementary Fig. 1b). P value is calculated by a two-sided paired t-test on area under the curve. Ribbons on the curves represent the standard deviations of the number of obtained unique ASVs by the algorithm.
Fig. 1 A data-driven microbial isolation strategy using phenotypic and morphologic features. a, Framework of phenotype and morphology-driven strain isolation and data collection of the human gut microbiome. Human fecal samples were plated and cultured under different antibiotics selection and morphologically diverse colonies were then isolated, biobanked and analyzed by downstream sequencing. b, Setup of the automated anaerobic microbial isolation and cultivation system CAMII. c, Illustration of morphology-guided colony isolation on CAMII. Colonies grown on plates are imaged under trans- and epi-illumination and subjected to contour segmentation and morphologic features extraction. Data are analyzed by PCA to identify the set of most morphologically diverse colonies that are then isolated by an integrated colony picker. d, Illustration of diverse colony morphology on plates. Colony size and shape features were extracted from transilluminated images, and colony color features were extracted from epi-illuminated images. e, Fecal sample H1t1 were cultured with seven different antibiotics to evaluate their capacity to yield the most unique and diverse bacteria by 16S analysis at the family level. Ciprofloxacin, trimethoprim and vancomycin were selected for subsequent colony isolations. f, Number of unique ASVs obtained from phenotype-guided isolation compared to random isolation of three human fecal samples H1t4, H5t1 and H6t1. Isolation was performed by CAMII; random isolation was performed on a random subset of all detected colonies on the plates, and phenotype-guided isolation was performed on morphology-selected colonies by the algorithm (Supplementary Fig. 1b). P value is calculated by a two-sided paired t-test on area under the curve. Ribbons on the curves represent the standard deviations of the number of obtained unique ASVs by the algorithm.CC BY · PubMed Central
Fig. 2 Generation of personalized gut isolate biobanks for 20 individuals. a, Statistics of 20 personalized gut isolate biobanks. b, Phylogenetic tree of 394 ASVs covered by 26,997 gut microbiome isolates in this study. Neighbor-joining tree of phylogeny was constructed based on 16S V4 sequences. Branch color distinguishes bacterial phylum, and the outer circle shows the prevalence of isolated ASVs in the 20 biobanks. c, Number of isolates for top 20 family-level taxonomy. d, Accumulated relative abundance of the ASVs represented by isolates from personalized biobanks in original fecal samples. The bars show isolates from any individual in the entire collection and the red lines show isolates derived from the same person. e, Heatmaps for relative abundance of abundant ASVs in original fecal samples and presence or absence in the biobanks. ASVs with average relative abundance > 0.1% are shown and the side bar on the left represents their family-level taxonomy. ASVs found in personalized biobanks are shown as black bars in the right heatmap and uncultured ASVs not found in any biobank are highlighted. f, Correlation of average relative abundance in original feces sample and number of isolates in entire collection for ASVs. Highly abundant ASVs that are difficult to culture, that is, with fewer isolates, are highlighted. g, Average relative abundance of top abundant ASVs but with no more than 2 isolates in the entire collection. Color of bars represents family-level taxonomy.
Fig. 2 Generation of personalized gut isolate biobanks for 20 individuals. a, Statistics of 20 personalized gut isolate biobanks. b, Phylogenetic tree of 394 ASVs covered by 26,997 gut microbiome isolates in this study. Neighbor-joining tree of phylogeny was constructed based on 16S V4 sequences. Branch color distinguishes bacterial phylum, and the outer circle shows the prevalence of isolated ASVs in the 20 biobanks. c, Number of isolates for top 20 family-level taxonomy. d, Accumulated relative abundance of the ASVs represented by isolates from personalized biobanks in original fecal samples. The bars show isolates from any individual in the entire collection and the red lines show isolates derived from the same person. e, Heatmaps for relative abundance of abundant ASVs in original fecal samples and presence or absence in the biobanks. ASVs with average relative abundance > 0.1% are shown and the side bar on the left represents their family-level taxonomy. ASVs found in personalized biobanks are shown as black bars in the right heatmap and uncultured ASVs not found in any biobank are highlighted. f, Correlation of average relative abundance in original feces sample and number of isolates in entire collection for ASVs. Highly abundant ASVs that are difficult to culture, that is, with fewer isolates, are highlighted. g, Average relative abundance of top abundant ASVs but with no more than 2 isolates in the entire collection. Color of bars represents family-level taxonomy.CC BY · PubMed Central
Fig. 3 Using colony morphology to predict taxonomic identity enhances targeted isolation. a, Heatmap of average z scores of morphological features across different bacterial genera. Different genera exhibiting diverse morphological patterns were classified into different groups by hierarchical clustering and the colored dot on the right represents their class-level taxonomy. b, Examples of colony images. Transilluminated images are on the left side and epi-illuminated images are on the right side. c, PCA ordination of genera based on their colony morphological features. Colors indicate class-level taxonomy. d, Performance of bacterial genus prediction based on morphological features by a random forest classifier. The numbers in brackets represent the number of isolates for each genus. Model training and evaluation were bootstrapped 20 times and the box plots show the variance of performance (n = 20). Blue line represents the performance of null model. Definition of box-plot elements—center line, median; box limits, upper and lower 25th quartiles; whiskers, 1.5× interquartile range. e, Performance of model-based targeted isolation. Bars represent the mean of prediction precision by individual-specific models that were bootstrapped 20 times and error bars represent the standard deviations. P values were calculated by two-sided Student’s t-test on precisions from n = 20 randomly initialized model bootstrapping.
Fig. 3 Using colony morphology to predict taxonomic identity enhances targeted isolation. a, Heatmap of average z scores of morphological features across different bacterial genera. Different genera exhibiting diverse morphological patterns were classified into different groups by hierarchical clustering and the colored dot on the right represents their class-level taxonomy. b, Examples of colony images. Transilluminated images are on the left side and epi-illuminated images are on the right side. c, PCA ordination of genera based on their colony morphological features. Colors indicate class-level taxonomy. d, Performance of bacterial genus prediction based on morphological features by a random forest classifier. The numbers in brackets represent the number of isolates for each genus. Model training and evaluation were bootstrapped 20 times and the box plots show the variance of performance (n = 20). Blue line represents the performance of null model. Definition of box-plot elements—center line, median; box limits, upper and lower 25th quartiles; whiskers, 1.5× interquartile range. e, Performance of model-based targeted isolation. Bars represent the mean of prediction precision by individual-specific models that were bootstrapped 20 times and error bars represent the standard deviations. P values were calculated by two-sided Student’s t-test on precisions from n = 20 randomly initialized model bootstrapping.CC BY · PubMed Central
Fig. 4 Mapping interaction between gut microbiota by colony morphology analysis. a, Images of an example plate during 6 d of growth and colony identities on the plate by 16S sequencing. b, Proportion of detectable colonies at different time points compared to day 6 for each genus. Colors indicate the family-level taxonomy. c, Correlation of colony size and number of nearby colonies for two representative ASVs. A full list of correlations is provided in Supplementary Table 8. P values are calculated by one-sided Mann–Whitney U test on area of colonies with no more than one nearby colony or no less than four nearby colonies (n = 101 versus 82 for ASV-6 and 17 versus 9 for ASV-39). Definition of box-plot elements: center line: median; box limits: upper and lower 25th quartiles; whiskers: 1.5× interquartile range. d, Pairwise growth promoting and inhibiting networks between genera. Directional growth-promoting effects are shown in red sharp arrows and directional growth-inhibiting effects are shown in blue blunt arrows. Nodes represent bacterial genera and are colored by family. Node sizes are proportional to the number of isolates used in this analysis and edges widths are proportional to the significance of the interactions.
Fig. 4 Mapping interaction between gut microbiota by colony morphology analysis. a, Images of an example plate during 6 d of growth and colony identities on the plate by 16S sequencing. b, Proportion of detectable colonies at different time points compared to day 6 for each genus. Colors indicate the family-level taxonomy. c, Correlation of colony size and number of nearby colonies for two representative ASVs. A full list of correlations is provided in Supplementary Table 8. P values are calculated by one-sided Mann–Whitney U test on area of colonies with no more than one nearby colony or no less than four nearby colonies (n = 101 versus 82 for ASV-6 and 17 versus 9 for ASV-39). Definition of box-plot elements: center line: median; box limits: upper and lower 25th quartiles; whiskers: 1.5× interquartile range. d, Pairwise growth promoting and inhibiting networks between genera. Directional growth-promoting effects are shown in red sharp arrows and directional growth-inhibiting effects are shown in blue blunt arrows. Nodes represent bacterial genera and are colored by family. Node sizes are proportional to the number of isolates used in this analysis and edges widths are proportional to the significance of the interactions.CC BY · PubMed Central
Fig. 5 Strain-level genomic diversity of the gut microbiome within and between individuals. a, Number of genome-wide SNPs between isolates from the same or different individuals for 14 isolates-rich species. The numbers after the species name represent the number of inter-individual (red) and intra-individual (blue) pairs. Definition of box-plot elements: center line: median; box limits: upper and lower 25th quartiles; whiskers: 1.5× interquartile range. b, Correlation between number of genome-wide SNPs and relative abundance in original fecal sample for isolates-rich species in individual H1. The size of dot represents number of isolates used in this analysis and the color represents proportion of SNPs present in only one genotype. c, Network of 2 kb + HGT frequency based on 409 isolates from H1. Nodes represent bacterial species and are colored by family. Node sizes are proportional to the number of isolates in the H1 collection and edges opacity are proportional to the HGT frequency between the two connected species. Edge color represents different types of HGT, that is, interphyla, intraphyla and interfamily, or intrafamily HGT and node outline color represent Gram staining of the species.
Fig. 5 Strain-level genomic diversity of the gut microbiome within and between individuals. a, Number of genome-wide SNPs between isolates from the same or different individuals for 14 isolates-rich species. The numbers after the species name represent the number of inter-individual (red) and intra-individual (blue) pairs. Definition of box-plot elements: center line: median; box limits: upper and lower 25th quartiles; whiskers: 1.5× interquartile range. b, Correlation between number of genome-wide SNPs and relative abundance in original fecal sample for isolates-rich species in individual H1. The size of dot represents number of isolates used in this analysis and the color represents proportion of SNPs present in only one genotype. c, Network of 2 kb + HGT frequency based on 409 isolates from H1. Nodes represent bacterial species and are colored by family. Node sizes are proportional to the number of isolates in the H1 collection and edges opacity are proportional to the HGT frequency between the two connected species. Edge color represents different types of HGT, that is, interphyla, intraphyla and interfamily, or intrafamily HGT and node outline color represent Gram staining of the species.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

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.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 1

  1. Huang, Y. et al. High-throughput microbial culturomics using automation and machine learning. WGS of isolates collection. Sequence Read Archive. https://www.ncbi.nlm.nih.gov/bioproject/PRJNA745993 (2023).

Cited by

Sorted by citations

See all 286

Related

See all

Metadata

Authors
Huang, Y., Sheth, R. U., Zhao, S., Cohen, L. A., Dabaghi, K., Moody, T., Sun, Y., Ricaurte, D., Richardson, M., Velez-Cortes, F., Blazejewski, T., Kaufman, A., Ronda, C., & Wang, H. H.
Affiliations
Columbia University (US)
Corresponding author
Wang, Harris H.
ORCID
Venue
Nature Biotechnology
Publisher
Springer Science and Business Media LLC
Published
2023-02-20
Volume
41
Issue
10
Pages
1424-1433
ISSN
  • 0733-222X (ISSN-L)
  • 1087-0156
  • 1546-1696
DOI
10.1038/s41587-023-01674-2
PMC
PMC10567565
OpenAlex
W4321369770
Cited by
283
References
63
Citations by year
2026: 95 · 2025: 94 · 2024: 62 · 2023: 29
FWCI
24.73
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Biotechnology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • National Science Foundation (MCB-2025515, DGE-1644869)
  • U.S. Department of Health & Human Services | National Institutes of Health (1R01AI132403, 1R21AI146817, T32GM007367)
  • United States Department of Defense | United States Navy | Office of Naval Research (N00014-18-1-2237, N00014-17-1-2353)
  • Burroughs Wellcome Fund (1016691)
  • Irma T. Hirschl Trust (NA)
  • Schaefer Research Award
  • Hertz Foundation (NA)
  • Junior Fellows Scholarship from the Simons Society of Fellows
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gut microbiota and health
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
  • Health Sciences › Medicine › Microbiology (medical) › Bacterial Identification and Susceptibility Testing
Keywords
culturomics, colony morphology, fecal microbiota, microbial co-culture, Ruminococcaceae, Bacteroidaceae, Coriobacteriaceae, Bifidobacteriaceae, strain evolution, horizontal gene transfer
MeSH terms
Machine Learning, Automation, Bacteria, Humans, Genomics, Microbiota

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.

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