Metagenomic editing of commensal bacteria in vivo using CRISPR-associated transposases

Abstract

Although metagenomic sequencing has revealed a rich microbial biodiversity in the mammalian gut, methods to genetically alter specific species in the microbiome are highly limited. Here, we introduce Metagenomic Editing (MetaEdit) as a platform technology for microbiome engineering that uses optimized CRISPR-associated transposases delivered by a broadly conjugative vector to directly modify diverse native commensal bacteria from mice and humans with new pathways at single-nucleotide genomic resolution. Using MetaEdit, we achieved in vivo genetic capture of native murine Bacteroides by integrating a metabolic payload that enables tunable growth control in the mammalian gut with dietary inulin. We further show in vivo editing of segmented filamentous bacteria, an immunomodulatory small-intestinal microbial species recalcitrant to cultivation. Collectively, this work provides a paradigm to precisely manipulate individual bacteria in native communities across gigabases of their metagenomic repertoire.

Fig. 1.. Development and optimization of MetaEdit for gut microbiome engineering.

(A) Schematic of MetaEdit where CAST machinery and megRNA coordinate the integration of a payload into target bacteria within a microbiome. (B) Diagram of eight pME vector designs with varying promoter sequence, transposon ends, CAST gene position and their corresponding integration efficiency and self-targeting frequency tested in B. vulgatus. (C) Circos genome plots for five human Bacteroidaceae isolates: P. vulgatus (Pv), B. stercoris (Bs), B. uniformis (Bu), B. caccae (Bc), and B. fragilis (Bf). Discrete genomic contig lengths are plotted as the inner ring. Three megRNA target sites (pink triangles) and their integration specificity (inner bar) and efficiency (outer bar) are shown. (D) Characterization of MetaEdit integration into a synthetic community of five Bacteroidaceae using distinct E. coli donors (Ec^recipient). Community compositions are shown as pie charts (left) and integration specificities across the concatenated recipient genomes are shown as bar plots (right) with intended target sites (pink triangles). (E) Experimental workflow for MetaEdit in gnotobiotic mice colonized with a 6-member human gut community. (F) Resulting MetaEdit integration efficiency in Pv and Bu target strains in the gut treated with donor Ec^Pv at Day 0 and Ec^Bu at Day 2 as measured by qPCR from fecal matter. (G) Day 3 integration specificities (%) of single Pv target cohort or dual Pv/Bu targets cohort compared to non-target (NT) control cohort across the community. Integration efficiency refers to the amount of detected wild-type that has been edited. Integration specificity refers to the amount of payload that is found at the on-target site compared to the rest of the (meta)genome. Integration efficiency data in (B–G) are shown as mean ± s.d. for n=3 independent biological or mouse replicates and deep sequencing specificity data are shown by a representative replicate.

Fig. 2:. Targeted in vivo engineering and tagging of native gut bacteria.

(A) MetaEdit experiment targeting a native murine microbe (inset). Phylogenetic tree and relative abundance (RA %) of all metagenome-assembled genomes (bottom) from a mouse microbiome overlayed with integration specificity (%) for targeted editing of the native B. thetaiotaomicron (top) across the 3.6 Gb concatenated metagenome. (B) Conjugation efficiency of MetaEdit vectors (left axis) and relative abundance of donor (right axis) over time within the murine gut. Donor relative abundance was calculated through metagenomics on fecal extracts and conjugation efficiency was calculated by platting fecal samples ± the pME backbone marker. (C) Integration efficiencies in Bt between mice cohorts treated with a targeted donor (Ec^Bt) or a non-targeting megRNA donor control by qPCR over time. (D) Schematic for payload-tagging strains within a microbiome for sequence-guided isolation (top), and 16S relative abundance plots on plated colonies isolated from fecal samples ± antibiotic payload selection (bottom). Plated fecal sample conditions are denoted as untreated (WT), non-targeted pME-treated (NT), and Bt-targeted pME-treated (MetaEdit w/ tetR) mice, respectively. Relative abundance only captures gut bacteria that can grow on plates. Conjugation efficiency and integration efficiency data in (B–C) are shown as mean ± s.d. for n=3 independent mouse replicates and deep sequencing specificity data are shown as replicates.

Fig. 3:. A polysaccharide utilization loci (PUL) payload enables controlled colonization of Bacteroides by dietary selections in mice.

(A) In vivo MetaEdit experiment targeting native Bt with pME vectors encoding PUL_inulin or tetR control with dietary perturbations. (B) Growth curve of PUL_inulin-integrated Bt in minimal media with increasing inulin concentrations. (C) Enrichment (%) of edited Bt relative to all native Bt with PUL_inulin (orange) or tetR (blue) payloads over time using intermittent inulin supplementation, quantified by qPCR on fecal matter. (D) Copy number of pME0024 (tetR, blue) and pME0027 (PUL_inulin) in Bt over time. (E) Principle Component Analysis (PCA) of 16S community composition of pME0027-treated (circles) and pME0024-treated (rectangles) mice during inulin supplementation over time. Fill color indicates different time points and dashed circles indicate grouping of cohorts. (F) Relative abundance of all Bt between PUL_inulin or tetR cohorts over time. Shaded orange boxes in (C, D, F) indicate time points with 5% dietary inulin supplement. Growth curves and qPCR data in (B–D) are shown as mean ± s.d. for n=3 independent biological or mouse replicates and 16S sequencing data are shown in replicates (n=4) (E) or as mean ± s.d. for n=4 independent mouse replicates in (F).

Fig. 4:. MetaEdit of Gram-positive Segmented Filamentous Bacteria (SFB).

(A) A schematic of MetaEdit vector and experimental design for SFB engineering. (B) Average 16S community composition of SFB-colonized mice treated with Ec^SFB donor over time. (C) FACS plots of mCherry and GFP channels from wild-type (no donor) or SFB-targeted mice (Ec^SFB donor). Inner boxes indicate gating for specific cell populations. The four gates represent I) unedited wild-type SFB, II) Ec^SFB donor with lost pME0029 vector, III) Ec^SFB donor with pME0029, and IV) edited SFB expressing GFP. (D) Integration efficiency of AmpR-GFP payload into SFB based on fecal analysis (magenta) and copy number of pME0029 vector in sorted SFB cells (blue) over time. (E) Genome-wide integration specificity quantified by deep sequencing in (gate I+IV)-sorted SFB cells from (C). (F) Bright field and fluorescence images of (gate I+IV)-sorted SFB populations, showing GFP⁺ SFB filaments in edited conditions. 16S and specificity data are shown by a representative replicate in (B, E) and qPCR data in (D) are shown as mean ± s.d. for n=3 independent mouse replicates.