Disruption of the spinal cord-gut axis alters microbial dynamics and carbohydrate cross-feeding in the gut Authors: Mohamed Mohssen, Ahmed A. Zayed, Kristina A. Kigerl, Jingjie Du, Garrett J. Smith, Jan M. Schwab, Matthew B. Sullivan, Phillip G. Popovich Published: 2026 ABSTRACT Spinal cord-gut communication regulates gut bacteria, yet the underlying mechanisms remain poorly understood. Previous studies relied primarily on gene markers with limited functional analysis or genome-resolved snapshots from small cohorts. Here, we assessed microbiome dynamics via genome-resolved metagenomics on 333 samples from male and female C57BL/6 mice collected before and up to six months after surgical disruption of the spinal cord-gut axis. This resulted in 6,635 microbial draft genomes as a foundation for a new “Mouse B6 Gut Catalog” that significantly expands species and strain representation for this widely used laboratory mouse strain. Sampling revealed that disrupted spinal cord-gut signaling causes persistent, lesion-severity-, sex-, and time-specific shifts in microbial community composition, with consistent depletion of Lactobacillus johnsonii. Feeding purified L. johnsonii to spinal cord-injured mice prevented metabolic defects and systemic inflammation caused by disruption of the spinal cord-gut axis. Analyses using genome-resolved and community-based metabolic profiling indicated altered carbohydrate sharing and utilization of gut microbes, potentially depleting L. johnsonii, providing a genome-inferred mechanism for future hypothesis testing. This study improves murine microbiome catalogs, illustrates how metagenome-informed microbial interventions can provide a mechanistic understanding to improve host health, and underscores the vital role of a healthy spinal cord in regulating gut ecosystem function. Introduction Mammals harbor a “second genome” composed of a complex microbial community residing in, on, and around their bodies. This microbial ecosystem, particularly prominent in the gut, assists the host in disease protection and influences the efficacy of drug therapies against illnesses like cancer1. These microbial beneficial roles depend on host control systems that monitor and regulate the microbiome2. One such system is the bi-directional signaling pathway connecting gut and brain functions3,4. Within this microbiota-gut-brain axis, the vagus (parasympathetic) nerve is often implicated as a solitary controller of host-microbiota mutualism5,6. However, the spinal sympathetic (fight-or-flight) nervous system also controls the gut, although its effects on the gut microbiota are rarely considered. The spinal cord is a critical “hub” integrating nerve impulses between the brain and body. Everyday activities, such as eating or filling of the bowel and bladder, stimulate spinal nerves, which in turn affect gut motility, mucus secretion, intestinal blood flow, and immune function7. These physiological processes affect the gut microbiota8,9. Even organ-specific signals directly transmitted to the brain via the vagus nerve, will directly10,11 and indirectly12,13 elicit spinal sympathetic reflexes that influence peripheral organs, including the gut14. Since spinal cord dysfunction is a complication of various neurological diseases (e.g., stroke, spinal injury, multiple sclerosis, Alzheimer’s, and other neurodegenerative diseases)15–21 and gut dysbiosis—imbalances in microbiome composition—has been linked to various health conditions22, a more comprehensive understanding of the “microbiota-gut-brain-spinal cord axis” is needed to achieve optimal microbiota modification. The functional importance of this axis is most clearly demonstrated when the ‘hub’ is disrupted. Evidence that the spinal cord is a pivotal node within the microbiota-gut-brain axis comes from preclinical and clinical traumatic spinal cord injury (SCI) research. Indeed, SCI consistently disrupts the gut microbiome, resulting in dysbiosis associated with impaired neurological recovery, increased intraspinal pathology, metabolic disturbances such as dyslipidemia, and behavioral changes, including heightened anxiety23–30. The extent of gut dysbiosis is modulated by factors such as injury severity and anatomical (spinal) level27–29, and pre-existing microbiome imbalances can exacerbate post-injury outcomes23,28,29. Notably, SCI-induced alterations in gut bacterial taxa are predictive of intestinal inflammation and increased gut permeability25,30, further linking microbiome composition to host health. Although research in this area is expanding, much of what we know still depends on gene marker methods, especially 16S rRNA gene sequencing31–33. These methods use primers to amplify conserved microbial genes, enabling taxonomic identification through comparison with reference genomes. However, gene-marker-based techniques are limited by primer bias, dependence on incomplete reference databases, and an inability to capture the functional diversity of niche-defining microbial traits, which can evolve more rapidly than core taxonomic features and fundamentally shape ecological interactions within the host34–36. To overcome these constraints, genome-resolved metagenomics offers improved taxonomic and functional resolution, allowing for the development of microbial genome catalogs tailored to specific ecosystems and facilitating direct analysis of metabolic capabilities37,38. So far, only one proof-of-concept study has used genome-resolved metagenomics to examine the spinal cord-gut axis in a small group of mice (n = 15, at a single time point, using only one sex)27. This current report marks substantial progress in the field by performing longitudinal sampling across both sexes and creating a high-resolution reference genome catalog of the murine microbiome. In addition, our method allows for a systematic investigation of how the spinal cord regulates the gut ecosystem. This work provids important new insights into how microbiota affect mammalian health, and it identifies sex- and time-dependent changes in microbial carbohydrate metabolism that are highly relevant for host physiology. Results and Discussion Expanding murine microbial reference genomes To expand on the only prior murine study that has used genome-resolved metagenomics to investigate the spinal cord-gut axis23, a more comprehensive murine microbial reference genomes catalog was developed using an expanded experimental design, summarized as follows (details under “Animals and spinal cord injury” in Methods). First, microbiomes were compared from three groups of mice: (i) control mice receiving a sham-lesion to the spinal cord (Laminectomy or Lam; surgical removal of vertebral body but no spinal lesion), (ii) mice that received a bilateral crush lesion in low-thoracic (T10) spinal cord, and (iii) mice that received a bilateral crush lesion in high-thoracic (T4) spinal cord. Severe lesions at these spinal levels partially (T10) or completely (T4) abolish supraspinal control over spinal sympathetic neurons that control the gut24,25, effectively titrating the magnitude of “spinal cord-gut axis” dysfunction. Second, to enhance the longitudinal resolution of spinal cord-dependent changes (i.e., acute and chronic) on gut microbiota dynamics, samples were collected across six time points for each mouse. Third, to improve translational relevance and statistical power, we included female and male mice, and for each sex and spinal lesion level, we doubled the number of mice sampled at each time point. Overall, these efforts resulted in 333 bulk metagenomes (294 fecal and 39 cecal) collected from 29 male and 30 female C57BL/6 mice at six time points spanning six months (Fig. 1A) – a considerable increase on our prior sampling23 of five female samples per experimental group (Lam, T4, and T10 spinal lesions) at a single time point. The overall physiological state of each mouse in the three groups was assessed after six months using a standardized scale of neurological recovery for mice (i.e., the Basso Mouse Scale (BMS) for Locomotion26; hereafter referred to as “BMS score” (Supplementary Fig. 1). This expanded experimental design was further augmented by enhancing metagenomic workflow at several key steps, including cross-assembly per mouse, ensemble binning, and bin refinement (see Methods and Supplementary Fig. 1A).Fig. 1Study design, characterization of recovered MAGs and comparative analysis of Mouse B6 Gut Catalog (MB6GC) against other published mouse gut microbiome catalogs.A Schematic illustrates the relative location of spinal lesions and depicts the spinal sympathetic and vagal parasympathetic innervation to the gut. Tables (middle) describe the number of samples in this study, categorized by sex, spinal lesion level (Lam, T10, T4), and time post-lesioning. Note that autonomic complications in mice with lesions at spinal level T4 significantly reduce lifespan, yielding fewer T4 SCI mice, regardless of sex, beyond 7 weeks. Each mouse was single-caged (number of cages=59). Cecal and fecal samples were used for building the reference genomes catalog whereas only fecal samples were used for all statistical analyses. Schematic created in BioRender. Kigerl, K. (2026) https://BioRender.com/5b3g9wd. B The completeness of the de-replicated MAGs (n = 263; de-replicated at 95% ANI from 6,635 MAGs) is shown together with the phyla and families they span. New representative taxa include 2 phyla (Verrucomicrobiota and Proteobacteria), 6 families (Akkermansiaceae, Burkholderiaceae, Atopobiaceae, CAG-552, Enterobacteriaceae, CAG-272), and 41 genera (see GTDB data under Zenodo DOI). The median completeness across all MAGs is shown with a vertical dotted red line at 94.35%. Contamination levels are <10% for all de-replicated MAGs. C Venn diagram comparing the number of de-replicated MAGs recovered in this study against those from our prior pilot study23. D The fraction of reads recruited to the de-replicated set of MAGs, where the mean (red dotted line) is ~80% across sexes and spinal injury levels (also see SingleM supporting data in Supplementary Fig. 1B). The mean for23 (blue dotted line) was only available for female mice. The mean number of reads mapped to MAGs, as evaluated by the Wilcoxon test, was significantly improved in the current study (p-value = 6.11e−11). E Venn diagram showing that at the MAG strain level (99% ANI dereplication), MAGs from MB6GC (this study; C57BL/6 mice) are largely mouse-strain specific (see Methods), with minimal overlap with the CBA/J MAG catalog. F A stacked bar chart shows that at the MAG species level (95% ANI dereplication), MAGs from MB6GC improve upon each of the previously published catalogs of mouse-gut reference genomes, including the Comprehensive Mouse Microbiota Genome Catalog (CMMG), the Mouse Gastrointestinal Bacteria Catalog (MGBC), and the Integrated Mouse Gut Metagenomic Catalog (iMGMC). G A taxonomically resolved heatmap shows that 53 MAGs from MB6GC were either novel (n = 15) or were the best representative genomes of their species (n = 38) compared to other catalogs. MB6GC’s improvements in genome completeness and contamination over all catalogs are shown. Study design, characterization of recovered MAGs and comparative analysis of Mouse B6 Gut Catalog (MB6GC) against other published mouse gut microbiome catalogs. A Schematic illustrates the relative location of spinal lesions and depicts the spinal sympathetic and vagal parasympathetic innervation to the gut. Tables (middle) describe the number of samples in this study, categorized by sex, spinal lesion level (Lam, T10, T4), and time post-lesioning. Note that autonomic complications in mice with lesions at spinal level T4 significantly reduce lifespan, yielding fewer T4 SCI mice, regardless of sex, beyond 7 weeks. Each mouse was single-caged (number of cages=59). Cecal and fecal samples were used for building the reference genomes catalog whereas only fecal samples were used for all statistical analyses. Schematic created in BioRender. Kigerl, K. (2026) https://BioRender.com/5b3g9wd. B The completeness of the de-replicated MAGs (n = 263; de-replicated at 95% ANI from 6,635 MAGs) is shown together with the phyla and families they span. New representative taxa include 2 phyla (Verrucomicrobiota and Proteobacteria), 6 families (Akkermansiaceae, Burkholderiaceae, Atopobiaceae, CAG-552, Enterobacteriaceae, CAG-272), and 41 genera (see GTDB data under Zenodo DOI). The median completeness across all MAGs is shown with a vertical dotted red line at 94.35%. Contamination levels are <10% for all de-replicated MAGs. C Venn diagram comparing the number of de-replicated MAGs recovered in this study against those from our prior pilot study23. D The fraction of reads recruited to the de-replicated set of MAGs, where the mean (red dotted line) is ~80% across sexes and spinal injury levels (also see SingleM supporting data in Supplementary Fig. 1B). The mean for23 (blue dotted line) was only available for female mice. The mean number of reads mapped to MAGs, as evaluated by the Wilcoxon test, was significantly improved in the current study (p-value = 6.11e−11). E Venn diagram showing that at the MAG strain level (99% ANI dereplication), MAGs from MB6GC (this study; C57BL/6 mice) are largely mouse-strain specific (see Methods), with minimal overlap with the CBA/J MAG catalog. F A stacked bar chart shows that at the MAG species level (95% ANI dereplication), MAGs from MB6GC improve upon each of the previously published catalogs of mouse-gut reference genomes, including the Comprehensive Mouse Microbiota Genome Catalog (CMMG), the Mouse Gastrointestinal Bacteria Catalog (MGBC), and the Integrated Mouse Gut Metagenomic Catalog (iMGMC). G A taxonomically resolved heatmap shows that 53 MAGs from MB6GC were either novel (n = 15) or were the best representative genomes of their species (n = 38) compared to other catalogs. MB6GC’s improvements in genome completeness and contamination over all catalogs are shown. Enhanced sampling and analytics drastically expanded microbiota representation. A total of 6,635 bacterial metagenome-assembled genomes (MAGs) meeting criteria of >50% completeness and <10% contamination were recovered. When combined with 112 previously generated MAGs23, this yielded 263 de-replicated MAGs (at >95% average nucleotide identity, or ANI) spanning 7 phyla and 31 families (Fig. 1B). These MAGs were used for subsequent ecological and metabolic analyses. Of the 263 de-replicated MAGs, 238 were either above-medium (n = 76; >70% complete, <10% contamination) or high (n = 162; >90% complete, <5% contamination) quality (see Methods) – more than doubling MAG-based genomic representation from our prior pilot study23 (Fig. 1C). Taxonomically, this expanded MAG collection captured additional representative taxa above the species level (Fig. 1B), with 20.5% (n = 54/263) of new MAGs likely representing novel species as they lack 95% Average Nucleotide Identity (ANI) matches, even against the ~62k bacterial species clusters from the many environments represented in the Genome Taxonomy Database (GTDB27; Release 207). Ecologically, this updated MAG collection greatly enhanced the representation of murine gut microbiota as the percentage of sequencing reads that align to de-replicated MAGs increased from a mean of 55%26 to ~80% (Fig. 1D). Similarly, assembly-independent read-based profiling showed that ~81% of microbial diversity was captured by the updated MAGs (Supplementary Fig. 1B – see Methods for details). To assess the taxonomic and ecological value of this new genomic resource, designated the “Mouse B6 Gut Catalog” (MB6GC), we compared MB6GC against other murine-relevant microbiome catalogs. At the strain level, MB6GC includes 480 medium- and high-quality MAGs (>50% completeness and <10% contamination) after de-replication at 99% ANI (representing strain-level – sensu28,29). This strain definition is common in the field of genome-resolved metagenomics28,29 but differs from other definitions used in pathogen surveillance and clinical epidemiology studies that delineate strains based on a few single-nucleotide polymorphisms30,31. Comparing these MAGs against a similar strain-resolved MAG database for CBA/J mice29 showed minimal overlap, with over 400 MAGs specific to C57BL/6 mice (Fig. 1E; see Methods). At the species-level (95%ANI – see ref. 32), MB6GC contributed 6-10% new or higher quality (based on improved completeness and reduced contamination) MAGs compared to larger murine gut metagenomic catalogs, despite those catalogs integrating 3-8x more metagenomic data from 36 to 74 projects across multiple mouse strains33–35 (Fig. 1F; see Methods). Collectively, across all catalogs, 53 MAGs from MB6GC were either novel (n = 15) or were the best representative genomes of their species (n = 38) (Fig. 1G). Overall, MB6GC is a new resource to support microbiome studies in the C57BL/6 mouse strain, which is the most widely used in biomedical research36. Also, for studies requiring strain-level (>99% ANI) taxonomic resolution to understand the (patho)physiological significance of microbiome-host interactions37,38, it is recommended to use mouse strain-specific MAG resources. Spinal cord integrity is essential for sustained regulation of gut microbiome composition across sexes Next, we leveraged MB6GC and fecal metagenomic data to determine how gut microbiome communities are influenced by spinal lesion level and sex, as a function of time post-lesion. After data curation (details under “Data curation” in Methods and Fig. 1A, Supplementary Fig. 1C, D), abundance-based beta-diversity comparisons (principal-coordinate analysis [PCoA] of Bray-Curtis dissimilarities) were made using the MAG data (MAGs were derived from all fecal and cecal metagenomes, whereas all ecological inferences were made solely from fecal metagenomic data). These comparisons revealed sex-specific microbial communities in mice before surgery (Fig. 2A, Supplementary Fig. 2A). Since sex-based differences persisted after surgery, regardless of the experimental group (Fig. 2B, Supplementary Fig. 2B), we analyzed the microbiome of male and female cohorts separately. Notably, in both male and female mice, partial (T10 lesion) or complete (T4 lesion) disruption of the spinal cord-gut axis significantly changed gut microbiome composition (Fig. 2C, D) relative to mice with intact spinal cords (Fig. 2E, F). Importantly, these changes appear to be permanent, with marked shifts in community composition (beta-diversity) evident for at least 6 months in all mice with spinal cord-gut axis dysfunction (Fig. 3A). Peak changes in gut microbial community composition varied by sex, occurring either three or five weeks after a partial or complete break of the spinal cord-gut axis in male or female mice, respectively (Fig. 3A). Like beta-diversity, disrupting spinal cord control over the gut increased alpha-diversity (estimated by Shannon’s index) in both sexes (Fig. 3B, Supplementary Fig. 2C) relative to mice with intact spinal cords. An increase in alpha diversity is associated with increased gut transit time8, a common comorbidity of spinal cord injury or dysfunction39. We confirmed this association in an independent mouse cohort using a red carmine dye assay40. The data show that compared to mice with intact spinal cords, total intestinal transit time increased 17-38% for at least 3 weeks in mice with a spinal lesion (Fig. 3C). However, because loss of autonomic control over the gut after SCI engages multiple convergent mechanisms that can shape microbial ecology and increase alpha diversity, changes in transit kinetics, and the attendant shifts in nutrient availability, should be viewed as one contributor among several, alongside intestinal inflammation, increased epithelial barrier permeability, and thinning/alteration of the mucus layer41,42.Fig. 2Community-level microbiome changes are spinal cord-dependent and vary as a function of sex and spinal lesion level.MAG abundance-based principal coordinate analysis (PCoA) of Bray-Curtis dissimilarities from metagenomes of 44 mice (see “Data Curation” in Methods and Supplementary Fig. 1C) show sex-based differences before (A) and after (B) spinal surgery. Consistent results arise from analyses before data curation (i.e., including outliers; see Supplementary Fig. 2A, B). C–F PCoAs constructed using the same methods as in (A, B) show microbiome differences only after a break in the spinal cord-gut axis (C, D). Before spinal lesioning (E, F) no group differences exist in male (n = 22) or female (n = 22) mice. See permanova p-values for assessment of significant differences in each panel. Small dots represent individual samples, while large dots represent the centroids per sex (A, B) or surgery (C–F).Fig. 3Dynamic changes in microbiome beta- and alpha-diversity are spinal cord-dependent and vary as a function of sex and spinal lesion level.A Boxplot injury-level- and time-resolved comparisons of Bray-Curtis dissimilarities (beta-diversity) relative to pre-surgery microbiome, shown for all data (n = 59 mice) as well as curated male (n = 22) and female (n = 22) mouse data (see “Data curation” in Methods and Supplementary Fig. 1C). Each small dot represents the microbiome change of an independent mouse relative to its baseline microbiome. Boxplot median values across mice per time-point are represented by large dots and are connected with thick lines. Peak changes are highlighted with gray boxes in male (3 weeks) and female (5 weeks) mice. B Boxplots showing spinal injury-level dependent microbiome alpha-diversity changes in curated male (n = 22) and female (n = 22) mice data. See baseline (Day 0) and sex- and time-resolved alpha diversity patterns in Supplementary Fig. 2C. C Complete spinal injury slows intestinal transit time (Red carmine dye assay) relative to sham-lesion (Lam) controls. For all boxplots, whiskers represent the 25th and 75th percentile ranges while p-values represent those from Wilcoxon tests, adjusted for multiple comparisons by the fdr (false discovery rate) method (see Methods). ****p <= 0.0001, ***: p <= 0.001, **p <= 0.01, *p <= 0.05, ns: p > 0.05. Community-level microbiome changes are spinal cord-dependent and vary as a function of sex and spinal lesion level. MAG abundance-based principal coordinate analysis (PCoA) of Bray-Curtis dissimilarities from metagenomes of 44 mice (see “Data Curation” in Methods and Supplementary Fig. 1C) show sex-based differences before (A) and after (B) spinal surgery. Consistent results arise from analyses before data curation (i.e., including outliers; see Supplementary Fig. 2A, B). C–F PCoAs constructed using the same methods as in (A, B) show microbiome differences only after a break in the spinal cord-gut axis (C, D). Before spinal lesioning (E, F) no group differences exist in male (n = 22) or female (n = 22) mice. See permanova p-values for assessment of significant differences in each panel. Small dots represent individual samples, while large dots represent the centroids per sex (A, B) or surgery (C–F). Dynamic changes in microbiome beta- and alpha-diversity are spinal cord-dependent and vary as a function of sex and spinal lesion level. A Boxplot injury-level- and time-resolved comparisons of Bray-Curtis dissimilarities (beta-diversity) relative to pre-surgery microbiome, shown for all data (n = 59 mice) as well as curated male (n = 22) and female (n = 22) mouse data (see “Data curation” in Methods and Supplementary Fig. 1C). Each small dot represents the microbiome change of an independent mouse relative to its baseline microbiome. Boxplot median values across mice per time-point are represented by large dots and are connected with thick lines. Peak changes are highlighted with gray boxes in male (3 weeks) and female (5 weeks) mice. B Boxplots showing spinal injury-level dependent microbiome alpha-diversity changes in curated male (n = 22) and female (n = 22) mice data. See baseline (Day 0) and sex- and time-resolved alpha diversity patterns in Supplementary Fig. 2C. C Complete spinal injury slows intestinal transit time (Red carmine dye assay) relative to sham-lesion (Lam) controls. For all boxplots, whiskers represent the 25th and 75th percentile ranges while p-values represent those from Wilcoxon tests, adjusted for multiple comparisons by the fdr (false discovery rate) method (see Methods). ****p <= 0.0001, ***: p <= 0.001, **p <= 0.01, *p <= 0.05, ns: p > 0.05. Unlike most studies that use static microbiome snapshots from different mice post-surgery, our within-mouse longitudinal analyses (i.e., before and periodically after injury; Fig. 3) show that spinal cord dysfunction leads to ongoing changes in gut microbiome composition. Moreover, our study confirms that sex is a critical biological variable when analyzing the microbiome (see also refs. 43,44) and that an intact spinal cord is necessary for regulating gut microbiome composition, regardless of sex. Notably, previous research in this area largely used only female mice (reviewed in ref. 45). Microbiome differences by sex are linked to the development of, or susceptibility to, autoimmune46–48, gastrointestinal49,50, metabolic51,52, neuropsychiatric and neurodegenerative conditions44,45,53,54. Collectively, our work extends existing microbiome datasets to include longitudinal spinal cord-dependent control over the gut ecosystem in both sexes, providing both a valuable new experimental resource and a more comprehensive understanding of CNS regulation of the microbiome. Microbes that persist after disrupting the spinal cord-gut axis include taxa previously linked to the host's physiological state Next, to determine which taxa respond to spinal cord-gut axis perturbation, microbial relative abundances were compared in mice with or without an intact spinal cord (see Methods for details on statistical analysis). Significant spinal cord-dependent effects are shown at an aggregated phylum-level (Fig. 4A, Supplementary Fig. 3A), and an individual MAG/species-level (Fig. 4B, C, Supplementary Fig. 3B). At the phylum level, a break in the spinal cord-gut axis decreased Bacteroidota and Proteobacteria, with a concurrent increase in Firmicutes_A in males and females (Fig. 4A, Supplementary Fig. 3A). Whereas the ratio between the phyla Firmicutes and Bacteroidetes (Bacteroidota here) was also reported to increase in our earlier work in mice with spinal cord injury23,55, others have reported that this same ratio decreases56,57. This discrepancy may be attributed to improvements in or differences between databases, as the GTDB uses genomic data that now distinguish the Firmicutes phyla into three separate phyla: Firmicutes, Firmicutes_A, and Firmicutes_B (see “GTDB vs NCBI Taxonomy” in Methods). In addition, other lower-resolution techniques (i.e., 16S rRNA gene amplicon sequencing) may not be able to distinguish between these three classically lumped lineages, partly due to relying on different taxonomy databases, a gap addressed in recent efforts58. Here, using genome-resolved metagenomics, Firmicutes_A was the only phylum that changed among the three. Note that taxonomic assignment was performed using GTDB release 207—the stable release available at the initiation of this study— and therefore does not reflect more recent taxonomic revisions, including the phylum-level reclassification of Firmicutes as Bacillota and of Proteobacteria as Pseudomonadota.Fig. 4Significant phylum- and MAG-level microbiome changes occur after disrupting the spinal cord-gut axis.A Violin plots of the most abundant (median relative abundance >0.25%) phyla show that Bacteriodota and Firmicutes_A significantly change only after disrupting the spinal cord-gut axis in male (right; n = 22) and female (left; n = 22) mice. Pre-surgery control plots and plots of the least abundant phyla (i.e., <0.25 median relative abundance) can be found in Supplementary Fig. 3A. B Boxplots of the most abundant (median relative abundance > 0.5%) significantly changing MAGs in male mice with spinal lesions. All 68 changing MAGs can be found in Supplementary Fig. 3B. C Boxplots of all significantly changing MAGs (n = 14) in female mice with spinal lesions. For violin plots and boxplots in all panels (A–C), the 25th and 75th percentile ranges are shown. p-values represent those from Wilcoxon tests, adjusted for multiple comparisons by the FDR (false discovery rate) method (see “Statistical methods for determining differentially abundant taxa” in Methods). ****p <= 0.0001, ***: p <= 0.001, **p <= 0.01, *p <= 0.05, ns: p > 0.05. Significant phylum- and MAG-level microbiome changes occur after disrupting the spinal cord-gut axis. A Violin plots of the most abundant (median relative abundance >0.25%) phyla show that Bacteriodota and Firmicutes_A significantly change only after disrupting the spinal cord-gut axis in male (right; n = 22) and female (left; n = 22) mice. Pre-surgery control plots and plots of the least abundant phyla (i.e., <0.25 median relative abundance) can be found in Supplementary Fig. 3A. B Boxplots of the most abundant (median relative abundance > 0.5%) significantly changing MAGs in male mice with spinal lesions. All 68 changing MAGs can be found in Supplementary Fig. 3B. C Boxplots of all significantly changing MAGs (n = 14) in female mice with spinal lesions. For violin plots and boxplots in all panels (A–C), the 25th and 75th percentile ranges are shown. p-values represent those from Wilcoxon tests, adjusted for multiple comparisons by the FDR (false discovery rate) method (see “Statistical methods for determining differentially abundant taxa” in Methods). ****p <= 0.0001, ***: p <= 0.001, **p <= 0.01, *p <= 0.05, ns: p > 0.05. At the species level, many differentially abundant MAGs were found in male (n = 68; Fig. 4B, Supplementary Fig. 3B) and female (n = 14; Fig. 4C) mice. The magnitude of change in relative abundance of several discrete microbial species (Fig. 4B, C, Supplementary Fig. 3B) was differentially affected by spinal lesion level, an observation that was also shown for species alpha diversity (Fig. 3B). Temporal changes in MAG abundances also varied by spinal lesion level and sex (Supplementary Fig. 4). However, time-resolved analysis of rare taxa shows persistent spinal lesion-dependent changes in Lactobacillus johnsonii (Fig. 5A) in both sexes. Changes in other rare taxa like Weissella cibaria (a significantly changing taxon in ref. 23) were time-dependent (Supplementary Fig. 4G), highlighting the importance of well-controlled temporal analyses to discern changes in the gut microbiome that may only be sporadically evident. Indeed, static, time-delimited sampling at any single time point would likely under- or overestimate change(s) for a given microbe or the effect that sex or the spinal cord has on these changes.Fig. 5The effects of feeding L. johnsonii to overcome the effects of a break in the spinal cord-gut axis.A A rapid and protracted reduction in L. johnsonii MAG abundance occurs in the gut of male (n = 22) and female (n = 22) mice after a break in the spinal cord-gut axis. B, C Fecal cultures grown on MRS agar show reduced L. johnsonii colonies in mice subjected to a clinically relevant model of spinal cord injury (SCI). SCI-dependent loss of L. johnsonii can be reversed by feeding mice 1 × 108cfu L. johnsonii daily after SCI (C left). 16S sequencing of fecal samples confirmed that this feeding regimen increased the abundance of L. johnsonii in the gut (C right). D Treatment with an L. johnsonii “probiotic” attenuates SCI-induced weight loss. After SCI, mesenteric lymph nodes (MLNs) showed an increased proportion of neutrophils (E) and a decreased proportion of lymphocytes (F), resulting in a higher neutrophil-to-lymphocyte ratio (NLR; G). These changes were normalized by treatment with L. johnsonii. p-values in (A) represent those from Wilcoxon tests, adjusted for multiple comparisons by the FDR (false discovery rate) method (see “Statistical methods for determining differentially abundant taxa” in Materials and Methods). B–G n = 5/group ****p <= 0.0001, ***p <= 0.001, **p <= 0.01, *p <= 0.05, ns: p > 0.05. p-values for (C–G) are from 2-way ANOVA with Šídák’s multiple comparisons test (C, E–G) or repeated measures 2-way ANOVA with Bonferroni’s multiple comparisons test (D). Note for E-G: Sensitivity analysis confirmed that the therapeutic effects of L. johnsonii were consistent across the group, with statistically significant results maintained after excluding the animal with a high neutrophil count (2-way ANOVA, p < 0.05). Formal outlier tests (Grubb’s and ROUT) indicated no statistical outliers. It is worth noting that high variability in immune cell counts is common in acute spinal cord injury models due to the heterogeneous nature of the systemic inflammatory response. The effects of feeding L. johnsonii to overcome the effects of a break in the spinal cord-gut axis. A A rapid and protracted reduction in L. johnsonii MAG abundance occurs in the gut of male (n = 22) and female (n = 22) mice after a break in the spinal cord-gut axis. B, C Fecal cultures grown on MRS agar show reduced L. johnsonii colonies in mice subjected to a clinically relevant model of spinal cord injury (SCI). SCI-dependent loss of L. johnsonii can be reversed by feeding mice 1 × 108cfu L. johnsonii daily after SCI (C left). 16S sequencing of fecal samples confirmed that this feeding regimen increased the abundance of L. johnsonii in the gut (C right). D Treatment with an L. johnsonii “probiotic” attenuates SCI-induced weight loss. After SCI, mesenteric lymph nodes (MLNs) showed an increased proportion of neutrophils (E) and a decreased proportion of lymphocytes (F), resulting in a higher neutrophil-to-lymphocyte ratio (NLR; G). These changes were normalized by treatment with L. johnsonii. p-values in (A) represent those from Wilcoxon tests, adjusted for multiple comparisons by the FDR (false discovery rate) method (see “Statistical methods for determining differentially abundant taxa” in Materials and Methods). B–G n = 5/group ****p <= 0.0001, ***p <= 0.001, **p <= 0.01, *p <= 0.05, ns: p > 0.05. p-values for (C–G) are from 2-way ANOVA with Šídák’s multiple comparisons test (C, E–G) or repeated measures 2-way ANOVA with Bonferroni’s multiple comparisons test (D). Note for E-G: Sensitivity analysis confirmed that the therapeutic effects of L. johnsonii were consistent across the group, with statistically significant results maintained after excluding the animal with a high neutrophil count (2-way ANOVA, p < 0.05). Formal outlier tests (Grubb’s and ROUT) indicated no statistical outliers. It is worth noting that high variability in immune cell counts is common in acute spinal cord injury models due to the heterogeneous nature of the systemic inflammatory response. Although abundant MAGs that decrease post-lesion in males and females were mostly Muribaculaceae family members (3 out of 4), most increasing MAGs (14 of 21) belong to the Lachnospiraceae family (Supplementary Table 1). A decrease in Muribaculaceae abundance with a concomitant increase in Lachnospiraceae has been linked to metabolic diseases, such as obesity, type 2 diabetes, insulin resistance, and non-alcoholic steatohepatitis59,60. Other rare taxa that are reduced in mice without normal spinal cord control of the gut, including L. johnsonii and Turicibacter sp002311155, may also play critical roles in influencing mammalian physiology and health61–63. L. johnsonii was largely depleted following spinal cord lesioning, irrespective of sex or lesion level (Fig. 5A), and this microbe has been previously reported to influence various murine metabolic pathways23. Based on these observations and our additional analyses64 that encompass both rare and abundant taxa, we evaluated the impact of reintroducing L. johnsonii into the gut of mice with spinal cord lesions. Lactobacillus johnsonii supplementation prevents metabolic and immune dysfunction caused by spinal injury L. johnsonii was isolated by culturing fecal samples from naïve C57BL/6 mice on a growth medium known to enrich for Lactobacilli (see Methods). Female C57BL/6 mice received complete mid-thoracic spinal lesions (injured) or were left uninjured (naive). Starting at 1 day post-lesion, both naive and injured cohorts received daily oral gavage of 1 × 108 cfu of L. johnsonii or vehicle (water). At nine days post-lesion, fecal samples from all mice were plated then the cfu/ml values were determined. A significant decrease in L. johnsonii abundance was noted in injured mice fed water only (Fig. 5B), confirming metagenomic data in Fig. 5A and our prior study23. Conversely, feeding L. johnsonii significantly increased its abundance in lesioned mice, which was visually evident in cultures, confirmed by quantifying cfu/ml and by 16S sequencing of fecal samples obtained from mice in each cohort (Fig. 5B, C). Importantly, rapid weight loss, a consistent phenomenon indicative of a changing metabolic state in mice with spinal injuries, was significantly reduced when mice were fed L. johnsonii (Fig. 5D). Therefore, we conclude that these results reflect a gut-microbiome-dependent “rescue” of metabolic perturbations caused by a break in the spinal cord-gut axis. For example, spinal cord injury causes rapid muscle wasting65, and other Lactobacillus species, but not necessarily L. johnsonii, can prevent sarcopenia by modulating inflammation, oxidative stress, and skeletal muscle metabolism66. Beyond the prevention of injury-induced weight loss, L. johnsonii treatment also preserved immune cell homeostasis in the intestine. At 9 days post-injury, mesenteric lymph nodes were collected and analyzed using a complete blood count with a 5-part differential to identify leukocyte subsets. Following injury, pronounced neutrophilia with concurrent lymphopenia develops in mesenteric lymph nodes (Fig. 5E, F), resulting in elevated neutrophil-to-lymphocyte ratios (NLRs; Fig. 5G). Conversely, feeding L. johnsonii restored the NLR to non-injury levels (Fig. 5G). Neutrophil migration to lymph nodes is a known response during infection and inflammation67,68, and high NLRs often indicate a heightened inflammatory state, which can be associated with disease progression and poorer patient outcomes. Indeed, the NLR serves as a useful biomarker for systemic inflammation and has been linked to increased risk of neurological deterioration and disability in conditions like stroke, multiple sclerosis, and traumatic injuries to the brain or spinal cord69–72. Importantly, spinal-injured mice treated with L. johnsonii had NLRs comparable to those observed in uninjured mice. While further research is needed to uncover the precise mechanisms underlying the rescue effects of L. johnsonii, this bacterium may improve intestinal barrier function, modulate systemic immunity, and prevent metabolic disease61,73–76. A more recent study77 has found body weight to be exclusively linked to an L. johnsonii-responsive host gene (Cyp1a1) that is critical for maintaining intestinal barrier integrity, providing a mechanism to explain L. johnsonii’s role in weight loss rescue. Notably, this observation was in a colitis model system where L. johnsonii was among the taxa that increased significantly upon increasing dietary fiber intake. We have independently observed the reduction of L. johnsonii in our spinal cord injury model system without any dietary modifications and experimentally shown its weight loss rescue effect, providing evidence for its potential role under an intact spinal cord-gut axis as well. Altered carbohydrate metabolism in spinal cord-dependent dysbiotic gut selects against rare beneficial taxa (e.g., L. johnsonii) Data in Fig. 5 indicate that the spinal cord regulates L. johnsonii homeostasis within the gut microbiome. Since spinal cord dysfunction occurs not only in traumatic spinal cord injury but also in various neurological diseases such as brain injury, stroke, Alzheimer’s disease, and multiple sclerosis, such regulation is relevant to a range of conditions. The consistent presence of spinal cord pathology and related autonomic dysregulation in these disorders may influence microbial homeostasis. Therefore, understanding how spinal cord dysfunction leads to sustained reductions in L. johnsonii may inform approaches for treating and managing neurological conditions. To clarify the mechanisms underlying this neuro-microbial connection, we used the entire MB6GC to infer the complex metabolic interactions that occur within the gut microbial community, focusing on the MAGs (n = 13) that were most abundant and changing after a break in the spinal cord-gut axis (Fig. 4B, C). Examining these complex interactions, as detailed below, resulted in a new conceptual model where L. johnsonii relies on the availability of ‘public goods’ as the bacterial community shifts after spinal cord dysfunction. All MB6GC MAGs were annotated using DRAM (Distilled and Refined Annotation of Metabolism78), a tool designed to functionally annotate microbial genes and categorize them across various expert-curated metabolic pathways. Interestingly, the four most abundant MAGs in non-lesioned mice (collectively ~28% and ~45% median relative abundance in female and male mice, respectively) belong to the Bacteroidales order (Supplementary Table 1), namely CAG-485 sp002362485, CAG-873 sp910577315, Duncaniella sp910578515, and Bacteroides thetaiotaomicron. Their proportionally large relative abundance can be attributed to these bacteria encoding a complete electron transport chain, including the high-oxygen affinity cytochrome bd (Supplementary Fig. 5), allowing them to survive and potentially thrive in the low oxygen environment of the gut79. Additionally, under homeostatic conditions, species from the Bacteroidales order can perform polysaccharide to oligo- and monosaccharide conversions that dictate cross-feeding relationships broadly across a gut microbial community38,59,80. The products of these carbohydrate conversions are primary energy sources for many gut bacteria, and carbohydrate metabolism by gut microbes is essential for producing various beneficial metabolites (e.g., short-chain fatty acids; SCFAs) that support host health38. However, as mechanistic genome-resolved approaches are only just emerging in studies of mammalian gut microbiome, community-contextualized microbial carbohydrate cross-feeding after dysbiosis has not been well-resolved in any system, especially involving L. johnsonii. Here, we examine state-dependent changes in systems used by bacteria to degrade and take up carbohydrates, namely glycoside hydrolases (GHs), sugar transporters, and polysaccharide utilization loci (PUL) systems. Based on the 13 most abundant changing MAGs, the capacity to degrade complex polysaccharides remained unchanged after disrupting the spinal cord-gut axis. This conclusion was drawn from analyzing the number and types of GHs, enzymes responsible for breaking down complex carbohydrates into simple sugars for microbial energy (Supplementary Fig. 6). When GHs were grouped into 14 substrate-based categories (sensu78), 11 of 13 MAGs encoded diverse GHs that act on all 14 carbohydrate substrates (Supplementary Fig. 6A). Of these 11 MAGs, three decreased and eight increased in mice with spinal cord lesions (Supplementary Fig. 6B). Unlike these abundant taxa, which can degrade these 14 carbohydrates, L. johnsonii scarcely encodes GHs (Supplementary Fig. 6A; see also Turicibacter sp002311155, another rare potentially beneficial taxon63,81,82 that is decreased in mice with spinal lesions). These data suggest that rare beneficial taxa depend on abundant community members for polysaccharide degradation and cross-feeding. Indeed, MAGs of these rare taxa encode many diverse types of sugar transporters that would enable them to scavenge simple sugars (Supplementary Fig. 7). This suggests that, under homeostasis, the final products of polysaccharide degradation carried out by abundant community members (e.g., Muribaculaceae) might be used by rare, beneficial taxa. After a significant perturbation, as Muribaculaceae decrease, the increase in other abundant bacteria is unlikely to support cross-feeding of rare taxa, a potential mechanism causing prolonged dysbiosis. For example, Lachnospiraceae family members encode GHs with broad polysaccharide substrate specificity (Supplementary Fig. 6), but unlike Muribaculaceae, Lachnospiraceae also encode numerous and diverse simple sugar transporters (Supplementary Fig. 7), so they are expected to consume the simple sugar products of polysaccharide degradation. Conversely, the four most abundant MAGs changing post-lesion (all Bacteroidales) scarcely encode these sugar transporters (Supplementary Fig. 7), and instead exclusively encode polysaccharide utilization loci (PULs), with multiple characteristic homologs of the starch utilization system (Supplementary Fig. 8). Bacteroidales use these systems to sense, bind, degrade, and import polysaccharides and oligosaccharides83. Of these four Bacteroidales MAGs, one increases after spinal lesioning (B. thetaiotaomicron) and three less-characterized Muribaculaceae decrease (Fig. 4 and Supplementary Fig. 4), with no evidence for such systems in any other taxa, consistent with previous reports83. PUL systems exhibit selfish and altruistic behaviors in complex trophic networks, like in the gut. Some Bacteriodales species encode “selfish” PUL systems that rapidly import broken-down oligosaccharides into the periplasm for further breakdown, conferring no direct benefit to neighboring species, while other Bacteroidales encode “leaky” PULs that release partial breakdown products (PBPs) for the benefit of the whole community84. We found evidence for a selfish PUL system (utilizing GH99 and GH76; reviewed in ref. 84) encoded only by B. thetaiotaomicron, potentially explaining the relative fitness of this taxon in the dysbiotic gut of mice with spinal lesions, where transporter-rich Lachnospiraceae compete for and scavenge monosaccharides. Other Bacteroidales species, notably those of the Muribaculaceae family, exhibit a seemingly “altruistic” behavior by selectively packaging GHs into outer membrane vesicles that are released into the extracellular environment, enabling the production of free saccharides for the greater gut community84–86. More broadly, the Muribaculaceae family has been credited with probiotic roles, including a protective oxygen-scavenging role59 – consistent with the cytochrome bd results above - see Supplementary Fig. 5) and cross-feeding with other microbial species (reviewed in ref. 59). Indeed, we find that some Muribaculaceae taxa (e.g., CAG-485 sp002362485) encode diverse GHs, potentially boosting the fitness of other Bacteroidales (e.g, CAG-873 sp910577315) and rare beneficial taxa (e.g., L. jonhsonii) that encode scarce GHs and instead depend on PBPs or monosaccharide products, respectively. A conceptual model that explains gut ecosystem function, both before and after a break in the spinal cord-gut axis, can be developed from our data (Fig. 6). Fundamental to this model is our data-driven hypothesis that reassembly of the microbial community post-injury, as orchestrated by the most abundant and dynamic taxa, may result in a unique carbohydrate metabolic fingerprint that disfavors rare beneficial taxa and cooperative trophic interactions. First, under homeostatic conditions, some abundant Bacteroidales members break down polysaccharides into oligo- and mono-saccharides, “public goods” that other Bacteroidales and potentially beneficial microbes, such as L. johnsonii and Turicibacter sp002311155, take up for energy using their sugar transporters. This is in line with a recent report demonstrating a beneficial cross-feeding relationship between L. johnsonii and Bacteroides uniformis, another Bacteroidales, for utilizing dietary fiber77. Second, after a break in the spinal cord-gut axis, the niche of Bacteroidales (Muribaculaceae members and B. thetaiotaomicron) is partially filled by other polysaccharide degraders, notably Lachnospiraceae family members. While Muribaculaceae MAGs significantly decrease, B. thetaiotaomicron seems to thrive, possibly because its PUL systems do not discard resources for other members and instead can quickly import oligosaccharides back to the periplasm. Unlike Bacteroidales, other polysaccharide degraders, mostly Lachnospiraceae family members, also encode large numbers of diverse sugar transporters and thus can fully utilize carbohydrates and consume the monosaccharides they produce, possibly starving rare beneficial taxa (e.g., L. johnsonii) and reducing their abundance in the community (Fig. 6).Fig. 6Data-driven model of gut microbial community carbohydrate metabolism explains L. johnsonii depletion in mice with spinal lesions.Metabolic cross-feeding model shows that a decrease in Muribaculaceae likely results in fewer public goods available to rare beneficial taxa like L. johnsonii. As Muribaculaceae decrease, their niche is partially replaced by other polysaccharide degraders that import most of the oligo- and mono-saccharides for their own use (e.g., Lachnospiraceae). This effectively starves rare beneficial taxa that cannot compete with taxa that encode “selfish” carbohydrate degradation and import systems. Created in BioRender. Kigerl, K. (2026) https://BioRender.com/1477l53. Data-driven model of gut microbial community carbohydrate metabolism explains L. johnsonii depletion in mice with spinal lesions. Metabolic cross-feeding model shows that a decrease in Muribaculaceae likely results in fewer public goods available to rare beneficial taxa like L. johnsonii. As Muribaculaceae decrease, their niche is partially replaced by other polysaccharide degraders that import most of the oligo- and mono-saccharides for their own use (e.g., Lachnospiraceae). This effectively starves rare beneficial taxa that cannot compete with taxa that encode “selfish” carbohydrate degradation and import systems. Created in BioRender. Kigerl, K. (2026) https://BioRender.com/1477l53. Because L. johnsonii thrives in the simple-sugar-rich forestomach and inhabits biofilms formed by Lactobacillus reuteri87, it is likely that increased intestinal transit time following disruption of the spinal cord–gut axis contributes to L. johnsonii depletion by reducing simple sugars (its primary food source). However, if sugar depletion alone explained this effect, other simple-sugar consumers (including various Lachnospiraceae taxa and coproplasma sp003979335, the latter also lacking polysaccharide-degradation capacity; see Supplementary Figs. 6, 7) would also be expected to decrease, which our data do not show. In the colon, where complex polysaccharides are abundant88, L. johnsonii’s (and other species such as Turicibacter sp002311155) persistence may instead depend on cross-feeding with specific Bacteroidales taxa (the Muribaculaceae taxa in our data), as described above. This model of gut ecosystem function provides a framework for predicting changes in gut microbe-host interactions beyond disruptions in the spinal cord-gut axis. For example, a study using paired metagenomics and metabolomics in insulin-resistant individuals (characterized by elevated sugar levels in blood, urine, and feces) found that Lachnospiraceae taxa positively correlate with monosaccharide concentrations in fecal matter, whereas Bacteroidales taxa show negative correlations with the same metabolites89. Our model suggests that high monosaccharide concentrations favor transporter-rich taxa like Lachnospiraceae while disfavoring transporter-scarce taxa like Bacteroidales. Thus, our proposed model provides a testable hypothesis for future investigations regarding gut microbe-host interactions in various disease states. Conclusions This work establishes a new conceptual framework for the microbiota-gut-brain-spinal cord axis, highlighting the intricate interplay between the host nervous system and microbial ecology. By improving study design, increasing sample diversity, and using advanced bioinformatic tools, we introduce new resources (e.g., the MB6GC mouse microbiome catalog) and identify microbial targets for better understanding disease mechanisms and potential interventions. For example, supplementing with L. johnsonii during recovery from spinal cord injury mitigates post-injury metabolic disturbances and pathological systemic inflammation (see Fig. 5). Other data show that loss of cooperative community carbohydrate cross-feeding networks in the gut of spinalized mice may render the gut microenvironment less permissive to L. johnsonii (and other potentially beneficial microbes; see Fig. 6). Collectively, these data underscore the need to understand how host perturbations affect gut microbes as part of an ecosystem, rather than isolated units. To decipher ecosystem-relevant changes in microbial functions without a priori knowledge, it will be important to apply ecological theory and systems biology approaches (sensu90) - to organize changing microbial genomes (about 70 MAGs in this study) into sub-communities that best predict the disease phenotype of interest and where a few key species can be identified as the primary drivers of these host-microbial subcommunity links. Methods Animals and creating spinal cord lesions All surgical and post-operative care procedures were performed in accordance with the Ohio State University Institutional Animal Care and Use Committee. We have complied with all relevant ethical regulations for animal use. Food (Teklad Global 14% Protein Rodent Maintenance Diet) and water were available ad libitum. For metagenomic sequencing, N = 29 female C57BL/6 mice and N = 30 male C57BL/6 mice from Jackson Laboratories (Bar Harbor, Maine) were used in this study. All mice were 8 weeks of age at the start of the study. To prevent gut microbial cross-contamination due to co-habitation, all mice were single-housed upon arrival at our animal facility and for the duration of the study. Mice were randomized to surgical groups using GraphPad QuickCalcs Random Numers (https://www.graphpad.com/quickcalcs), and groups were evenly spread among surgery days. Mice were anesthetized with an intraperitoneal cocktail of ketamine (80 mg/kg)/xylazine (10 mg/kg) after which a partial laminectomy was performed at the fourth thoracic spine (T4) or the tenth thoracic spine (T10). In mice modeling spinal cord injury (SCI), modified #5 Dumont forceps were inserted laterally and held closed for 3 seconds to produce a complete crush spinal cord injury. In surgical control animals (LAM), a partial laminectomy was performed, but no spinal crush was done. In total, 20 mice received a complete crush at T4, 20 mice received a complete crush at T10, and 19 mice received a laminectomy (Lam) only (Fig. 1A). Group sizes were based on previous published datasets23,55. Post-operatively, animals were hydrated with 2 mL Ringer’s solution (subcutaneous) for 5 days. Bladders were voided manually at least twice daily for the duration of the study. No prophylactic antibiotics were used during or after surgery. Mice were also monitored for symptoms of bacterial infection. This includes analysis of urine pH weekly to monitor for urinary tract infection, twice-per-day monitoring of the surgical incision for signs of infection (redness, edema, warmth), as well as monitoring for general sickness behaviors in the mice (including hunched posture, weight loss, ruffled fur). No mice exhibited symptoms of infection. Supportive care/treatment was provided to animals that exhibit >20% intractable weight loss, failure to eat and drink, signs of infection/abscess, or any other adverse effects. Animals that did not respond to care/treatment within 48 hours were euthanized, or earlier as directed by the ULAR veterinarian. This resulted in early removal of n = 18 T4 mice prior to the 6-month timepoint. Exact sample numbers at each time point are detailed in Fig. 1A. Fecal samples for metagenomic sequencing were collected pre-lesion, 1-week, 2-weeks, 3-weeks, 5-weeks, 9-weeks, and 6-months post-lesion. Experimenters were not blinded during fecal sample collection due to the apparent differences in paralysis that the various surgical groups display. Mice were removed from their home cage and placed into a ventilated, aseptic polystyrene compartment, and fresh fecal samples were collected from each mouse into sterile tubes and immediately frozen in liquid nitrogen. Mice were returned to their home cage after sample collection. At 6-months post-lesion, mice were anesthetized with ketamine/xylazine for tissue collection. Cecal contents were collected by removal of the cecum and extrusion of cecal contents into a sterile Eppendorf tube. All samples were immediately frozen in liquid nitrogen and stored at −80°C until further use. Analysis of neurological function Spinally-mediated neurological function was assessed using the Basso Mouse Scale for Locomotion (BMS score) at 6 months post-lesion. Two experimenters assessed BMS scores for mice during a 4-minute interval in the open field. Scores were tallied for each hind paw and then averaged to obtain a single score per animal. Lam mice have a score of 9, indicating normal locomotor function. Mice with spinal lesions have scores below 9, with scores varying as a function of lesion level, with higher lesion levels creating greater neurological impairment across the body and correspondingly lower BMS scores. Deceased mice were assigned a score of 0 in this study. Lactobacillus johnsonii isolation Fresh fecal samples were collected from naïve C57BL/6 mice (Jackson Laboratories) into sterile tubes and homogenized in 500 μL sterile 0.1 M PBS. 100ul of this homogenate was plated on agar plates made with De Man, Rogosa, and Sharpe agar (Lactobacilli MRS agar; Neogen culture media NCM0035A), a culture medium designed to enrich for many Lactobacilli species. These plates were cultured under anaerobic conditions using Mitsubishi™ AnaeroPack- Anaero Anaerobic Gas Generator (Thermo Scientific™ R681001) for 48 h, and numerous colonies were visible on the plates. To determine whether the resulting colonies were indeed L. johnsonii, we selected 10 isolated colonies for further characterization. These colonies were expanded by culturing in MRS broth (MRS Broth; Millipore 69966) under static anaerobic conditions for 48 h at 37 °C. After the incubation, the cultures were centrifuged to pellet the bacteria, and the cells were frozen in 40% glycerol-MRS broth. DNA was isolated from each culture using the Qiagen DNeasy PowerSoil Pro kit (Qiagen 47014). After DNA isolation, 16S PCR was run using universal 16S primers (GM3: AGA GTT TGA TCM TGG C; GM4: TAC CTT GTT ACG ACT T) following protocol (Natalie Solonenko 2016. 16S PCR Universal. protocols.io https://doi.org/10.17504/protocols.io.c5my45). These primers amplify the 16S gene region for all bacteria (they are not genus or species-specific). PCR products were then run on an agarose gel and gel-purified using Zymoclean Gel DNA Recovery Kit (Zymo Research; D4001). This gel-purified DNA was sent to the Genomics Core Lab at The Ohio State University for Sanger sequencing. The resulting sequences were then compared to all known DNA sequences using nucleotideBLAST (https://blast.ncbi.nlm.nih.gov/). The results of this query for isolate #10 are shown in Fig. 1D. All results returned were specific for strains of L. johnsonii, indicating that we have been successful in isolating and culturing this bacterium from the mouse gut microbiome. The same was true for all 10 isolates. Preparation of L. johnsonii glycerol stocks After identifying that bacterial isolates were L. johnsonii, cultures were prepared to generate glycerol stocks for storing and future use in experimental studies. Frozen glycerol stocks (see above) were thawed, and 1 mL was transferred to a 50 ml conical tube with 5 mL sterile 0.1 M PBS to remove glycerol. Cells were centrifuged at 4000xg for 10 minutes to pellet bacteria, and the supernatant was removed. Pellet was resuspended in 10 mL of sterile MRS broth (MRS Broth; Millipore 69966) and cultured for 24 h at 37 °C in anaerobic conditions (Mitsubishi™ AnaeroPack- Anaero Anaerobic Gas Generator; Thermo Scientific™ R681001). This 10 mL culture was then divided evenly (3.3 mL) into 3 new sterile 50 mL conical tubes containing a final volume of 25 mL. These cultures are incubated for 6–9 h at 37 °C in anaerobic chambers until absorbance/OD is between 0.35 and 0.55 when read at 600 nm. We compared the OD values to a growth curve chart to determine approximate cfu/mL values. Bacteria are pelleted as above and resuspended in 40% glycerol MRS broth (MRS Broth; Millipore 69966) at a final concentration of 1 ×109 cfu/ml in cryovials and stored at −80°C. To ensure accurate concentrations prior to use in animal studies, an aliquot if then thawed, washed, and plated on MRS agar (Lactobacilli MRS agar; Neogen culture media NCM0035A) plates using serial dilutions to determine exact cfu/mL. Oral delivery of L. johnsonii C57BL/6 female mice (Jackson Laboratories) received a mid-thoracic spinal cord lesion as described above. Mice were randomized to treatment groups using GraphPad QuickCalcs Random Numbers (https://www.graphpad.com/quickcalcs). Mice received 1 × 108 cfu/mL L. johnsonii from stored glycerol stocks, washed in sterile 0.1 M PBS, in 100 μL sterile saline by oral gavage each day starting on day 1 post-injury through day 9 post-injury. Fecal samples were collected before and again 9 days post-lesioning. Fecal samples were homogenized in 500 μL sterile 0.1 M PBS, and 100 μL was plated on MRS agar (Lactobacilli MRS agar; Neogen culture media NCM0035A) to enumerate Lactobacillus colonies. From the same fecal samples, DNA was isolated using Qiagen DNeasy PowerSoil Pro kit (Qiagen 47014) for 16S sequencing (Novogene). Body weights were measured before surgery and throughout the 9-day survival period. At 9 days post-injury, mesenteric lymph nodes were harvested, and a single cell suspension was made by homogenizing lymph nodes through a Falcon 40-µm mesh screen (BD) and washing with sterile PBS. 10 µl of cell suspension was analyzed on a hematology analyzer (Heska Element H5) to obtain a complete blood count (CBC) with 5-part differential. Intestinal transit time assay Total intestinal transit time was measured using a 6% Red Carmine dye assay following the protocol published in ref. 40. C57BL/6 female mice (Jackson Laboratories) were given either a T9 complete crush spinal cord injury or a laminectomy (sham) control surgery (see surgical details above). Starting at 7 days post-injury, and weekly thereafter, mice were given 150 μL of 6% red carmine (Sigma-Aldrich C1022) dissolved in a 0.5% methylcellulose (Sigma-Aldrich M0512) solution. After oral gavage of the 6% carmine solution, mice were fasted for the duration of the assay. After 180 minutes, each mouse was placed in a clean cage and monitored at 5 minutes intervals to assess fecal pellets for the passage of the red dye (fecal pellets are visibly red). The time of the first red fecal pellet is recorded, and the whole gut transit time is calculated as the time interval between the gavage and the first observed red fecal pellet. DNA extraction and metagenomic sequencing Experimenters were blinded to the surgical group during DNA extraction, sequencing, and initial sequencing analysis. From 59 mice, 333 bulk metagenomic samples were prepared for sequencing (batch #1). Both fecal and cecum samples were extracted using a DNeasy PowerSoil Pro Kit (# 47014, QIAGEN) following the manufacturer’s instructions. The samples were then normalized to 12 ng/uL and sent to the JP Sulzberger Columbia Genome Center. At the center, using a Riptide kit from iGenomX for library preparation, sequencing was performed using Illumina NovaSeq 6000 targeting 26 M 2x150bp reads per sample. Reads from bulk metagenomes ranged from 3.5 million paired-end reads to 64.5 million, with a median value of 25.1 million paired-end reads. For 11 samples with the lowest coverage, from 3.5 to 13.6 million reads, sequencing was redone at the same center, at 25 M 2x100bp reads per sample. Library prep was performed using the Nextera XT kit from Illumina (batch #2) at the Sullivan Lab before sending samples to the sequencing center. The prefix “Repeat_” was added to their sample labels in all metadata spreadsheets and codes. Reads ranged from 5.7 million paired-end reads to 26.8 million for these 11 samples, with a median value of 24.2 million paired-end reads. These samples, starting with “Repeat_” were used in addition to the original ones for binning bacterial genomes, but were discarded after calculations of abundances and further analyses due to having different read lengths (batch #2 having 2 x 100bp). Read quality control Adapter trimming: Reads were trimmed with bbduk91 (https://jgi.doe.gov/data-and-tools/bbtools/) from the right end (ktrim = r) to keep ones longer than 30 bp (minlength=30) by using a kmer of 23 (k = 23) and allowing for a mismatch of 1 (hdist=1 hdist2 = 1) and a lower kmer at the ends of the read (mink=11). In addition, reads mapping to adapters, artifacts, phix, and overrepresented sequences (i.e., low-complexity sequences) using “ref= adapters, artifacts,phix,” to remove all reads having a 23-mer match to these commonly found sequences in Illumina runs. Mouse reads removal: Reads mapping with a minimum identity of 95% to our mouse strain C57BL/6NJ (genome downloaded from the NCBI assembly database using the accession number “GCA_001632555.1”) were removed using bbmap91 (https://sourceforge.net/projects/bbmap). Additional settings were used in bbmap (maxindel = 3 bwr = 0.16, bw = 12, quickmatch fast, minhits = 2, overwrite = t, usemodulo). Quality trimming: Next, using bbduk, reads with an average quality of 10 or ones where base quality drops to 10 are trimmed from both sides, keeping reads with a minimum length of 30 basepairs (qtrim=rl maq=10 maxns=0 minlength=30 trimq=10). Read-based profiling Read-based profiling was performed using SingleM v 0.13.2 (https://github.com/wwood/singlem) using raw reads. First, the “pipe” command was run on raw reads and on the 263 set of MAGs. Operational taxonomic unit (OTU) tables resulting from each run (for paired reads of one sample) were then combined by sex. Finally, the “appraisal” method was used to assess how well these MAGs represented the whole community, resulting in Supplementary Fig. 1B. Co-assembly and decontamination of scaffolds for binning (see Supplementary Fig. 1A) After quality control, all trimmed paired-end reads per mouse were combined and assembled using SPAdes (v3.14.1) in the meta mode (i.e., metaSPAdes92; https://github.com/ablab/spades). All scaffolds below 500 bp were discarded. The scaffolds longer than 500 bp were blasted (-task megablast) against an in-house contaminants database (see DOI: 10.5281/zenodo.13871704 with details in the ReadMe file). This database included the mouse genome (strain C57BL/6NJ) and genomes of non-gut microbes and viruses that were sequenced together with samples from this study (batches #1 and #2). Alignments that have 95% nucleotide query-subject identity covering 80% of the scaffold were considered contaminants and were therefore discarded. After removing contaminants, trimmed reads from all bulk metagenomes were mapped to the set of co-assembled decontaminated scaffolds using CoverM v0.6.1 (https://github.com/wwood/CoverM) in the “contig” mode with the following parameters “--min-read-percent-identity. 95 --min-read-aligned-percent .75 --discard-unmapped”. The resulting BAM files were used as inputs for binning tools. Construction of MAGs, estimation of abundance, and taxonomic classification (see Supplementary Fig. 1A) Metabat293 and Maxbin294 (with the 107 marker gene set) were run with co-assemblies and per-sample bam files through UniteM v0.0.18 (https://github.com/dparks1134/UniteM) with a minimum contig length to be binned of 1500 bp. The resulting bins from each tool were refined per mouse with metaWRAP v1.2 (bin_refinement module with the --keep-ambiguous flag, keeping bins with more than 50% completion). The quality of all bins was assessed with CheckM95 v 1.0.18. All 6635 bins from all 59 mice were then pooled with the 112 bins from ref. 23. The 6149 bins that were estimated to be more than 60% complete and less than 10% contaminated were de-replicated via Galah v0.3.1 (https://github.com/wwood/galah), resulting in 263 bins that captured redundant sequence variation resulting from mouse-constrained co-assembly. These 263 bins were used for further analysis. Bins were taxonomically classified by the Genome Database Taxonomy Toolkit96 (GTDB-TK) v2.1.1 through the “classify_wf” module, which uses “release 207” of GTDB. The abundance of de-replicated MAGs was estimated using CoverM v0.6.1 (https://github.com/wwood/CoverM) in the “genome” mode with the trimmed mean method through the parameters “-m trimmed_mean --min-read-percent-identity 0.95 --min-read-aligned-percent 0.75 --discard-unmapped”. GTDB vs NCBI Taxonomy Differences between NCBI and GTDB (release 207) include the following: (a) NCBI’s Firmicutes is divided into Firmicutes (includes order Bacilli), Firmicutes_A (includes order Clostridia), and Firmicutes_B (includes order Dehalobacteriia); (b) Many members of the families Bacteroidaceae, Prevotellaceae, Muribaculaceae are reclassified as Muribaculaceae, belonging to the phylum Bacteroidota (i.e, Bacteroidetes in NCBI); (c) Some members of the Clostridiaceae family (p__Firmicutes; c__Clostridia; o__Eubacteriales) are classified as CAG-508 (p__Firmicutes_A; c__Clostridia; o__TANB77) and include mostly unclassified species; members classified as family CAG-552 (p__Firmicutes_A; c__Clostridia; o__Christensenellales) were largely unclassified firmicutes belonging to order Eubacteriales or novel orders; there have been also many events of reclassification in the class Clostridia from order Eubacteriales to Lachnospirales, Oscillospirales, Peptostreptococcales, and TANB77. Comparisons of genomic catalogs First, genomes from different catalogs were collected. These include the Comprehensive Mouse Microbiota Genome Catalog (CMMG)’s 1573 species-level representative genomes (95% ANI) (https://ezmeta.unige.ch/CMMG/), the Integrated Mouse Gut Metagenomic Catalog (iMGMC)’s 1296 representative genomes (https://github.com/tillrobin/iMGMC), the Mouse Gastrointestinal Bacteria Catalog (MGBC)’s 26,640 high-quality, non-redundant genomes (MGBC-hqnr_26640; https://github.com/BenBeresfordJones/MGBC), and CBAJDB’s 504 strain-level (99% ANI) medium- and high-quality genomes (CBAJDB_MQ-HQ; https://zenodo.org/records/8395759). Second, CheckM95 v1.0.18 was run on MAG catalogs to provide completeness and contamination information for each MAG – except for MGBC, for which MAG completeness and contamination data were extracted from genomes metadata (MGBC_md_26640). Third, for MGBC and CBAJDB (catalogs with no publicly available species-level representative genomes), Galah v0.3.1 (https://github.com/wwood/galah) was run (--ani 95 --precluster-ani 90) to get species-level representative genomes for MGBC (n = 1091) and CBAJDB (n = 113). Finally, at the strain level, MB6GC MAGs (n = 499) were clustered with CBAJDB (n = 504) using Galah (--ani 99 --precluster-ani 90) - Fig. 1E. At the species level, MB6GC MAGs (n = 263) were clustered with CMMG, iMGMC, and MGBC individually (Fig. 1F) using Galah (--ani 95 --precluster-ani 90). Additionally, MB6GC MAGs (n = 263) were also clustered with MAGs from all representative genomes of the 4 catalogs together (Fig. 1G) using Galah (--ani 95 --precluster-ani 90). Stacked bar charts were plotted using ggplot2 (https://ggplot2.tidyverse.org) while heatmaps (as in all heatmaps hereinafter unless otherwise stated) were made using the ComplexHeatmap package97 (https://github.com/jokergoo/ComplexHeatmap). All tables hereinafter were processed using the tidyverse package in R98. Alpha- and beta-diversity estimation Beta-diversity was assessed using principal coordinate analyses (PCoAs) with Bray-Curtis dissimilarity matrices. Bray-Curtis distances were calculated using the “vegdist” function from the R package “vegan” (https://github.com/vegandevs/vegan). Vegan’s “adonis2” function was used to perform PERMANOVA tests on Bray-Curtis distances with 9999 permutations. Longitudinal analyses were done by plotting Bray-Curtis distances to the mouse’s Day 0 sample to account for individual variations using ggplot2 (https://ggplot2.tidyverse.org). Statistical testing was performed via the o_test function from the “rstatix” package (https://CRAN.R-project.org/package=rstatix), with the following parameters: “paired=FALSE, ref.group = ‘Lam’, detailed=TRUE, alternative=’less’, p.adjust.method =’fdr’. Significance was shown using the “ggpubr” package’s stat_pvalue_manual function (https://rpkgs.datanovia.com/ggpubr). Alpha diversity was assessed by Shannon’s index using the “diversity” function from the vegan package. Shannon’s indices for samples were plotted by time point (i.e., samples at one time point) and as an aggregate (e.g., all samples post-injury). Longitudinal statistical comparisons shown in boxplots were made in a fashion similar to beta-diversity. Aggregate statistical comparisons shown in main figures were performed using ggpubr’s stat_compare_means, with the following parameters: method = “wilcox.test”, paired=FALSE, method.args = list(exact=FALSE), label = “p.signif”, hide.ns =TRUE, ref.group = “.all”. Data curation Upon observing beta-diversity in all samples after sham or spinal cord injury, samples were shown to cluster into 3 main groups. Since mice had their operations in 4 batches, at 4 different dates, we hypothesized that this might be one of the confounders. Colored by the day (i.e., date) on which the surgery was performed on mice, male mice having their operation performed at the first surgery day and female mice having theirs at the fourth day clustered closer to each other and differently from their respective larger male and female clusters (see Supplementary Fig. 1C). These mice were excluded from further analyses (e.g., alpha and beta diversity, etc.). To elaborate on the statistical methods, we performed an analysis of variance using distance matrices (the same that were used for the principal component plots, such as in Supplementary Fig. 1C). We used the function “adonis2” from the package “vegan” in R to partition the distance matrix and fit a model to it (see https://rdrr.io/cran/vegan/man/adonis.html) with 9999 permutations. When comparing male and female mice data at Day 0 separately, Male-Day1 and Female-Day4 were outliers from the male and female data from other surgery days, respectively, as indicated by <0.05 p-values adjusted for multiple comparisons. Further, upon comparing all male and female data together, Male-Day1 data were shown to be similar to Female-Day4 data, as indicated by an adjusted p-value of 1. Therefore, we decided to exclude these batches as they did not exhibit the same trends as other data from the other main male and female clusters. The two files (“Male_Day0_Date_Outlier_Test.csv” and “Female_Day0_Date_Outlier_Test.csv”) containing the results of the sex-stratified statistical tests have been uploaded to the accompanying Zenodo directory. The script and aggregate statistical tests are also available in the same directory under “Scripts/pcoas/sensitivity_analysis_figures”. Finally, repeated samples (batch #2), having a different read length from batch #1, were also excluded from these analyses. The main male and female clusters were treated separately and as different cohorts. Statistical methods for determining differentially abundant taxa (Fig. 4, Supplementary Fig. 3) The trimmed-mean coverage values from CoverM were converted to relative abundance per sample after excluding repeated samples. Relative abundances were summed for each taxon and used for further tests. Considering the complex experimental design and the large number of taxa to be investigated (Fig. 1), it was challenging to test significance per time point due to losing statistical power, especially for female mice. Therefore, two statistical approaches were used to examine the data. First, samples collected from every two consecutive timepoints (e.g., “Day0 and 1week”, “1week and 3weeks”, and so on) were combined and tested for omnibus changes between the three experimental groups; the relative abundance values were fed into a Kruskal-Wallis test (from the kruskal.test function in R) followed by a Benjamini-Hochberg correction for multiple testing to check if there is a difference between this taxon’s abundance in Lam, T10, and T4 mice collectively. To examine the specific changes (e.g., differences between the taxon’s abundances in Lam vs T10 or Lam vs T4), statistically significant taxa from Kruskal-Wallis tests (adjusted p-values of less than 0.05) were then tested by a Wilcoxon signed-rank test (from the pairwise.wilcox.test function in R) followed by a Benjamini-Hochberg correction. No taxa turned out to be differentially abundant between Lam and T10 or T4 from the first timepoint pair (i.e., “Day0 and 1week”). Taxa deemed differentially abundant between injured and control (i.e., Lam) mice by these tests were then collected and plotted in Fig. 4 and Supplementary Fig. 3. Taxa were plotted depending on their abundance as follows: if the median abundance of the taxon (in any of the three treatments) was above 20% for phyla or 0.5% for species, it was plotted in the high-relative abundance plots. Second, these taxa were compared post-surgery (> Day0) with a Wilcoxon signed-rank test followed by fdr correction. Hence, consensus taxa plotted in Fig. 4 and Supplementary Fig. 3 were both statistically significant upon testing the samples aggregated at the post-injury level (> Day 0) and at the two consecutive timepoints level. MAG functional annotation DRAM v1.3.6 (https://github.com/WrightonLabCSU/DRAM) was used to annotate MAGs against multiple databases, including the Carbohydrate-Active enZymes Database or CAZy99. DRAM’s distillate outputs, i.e., the “product.tsv” and the Carbon Utilization and Transporters tabs of the “metabolism_summary.xlsx,” were used to profile each MAG’s electron transport chain components (Supplementary Fig. 5), GHs (Supplementary Fig. 6), and sugar transporters (including PTS components; Supplementary Fig. 7), respectively, whereas each MAG’s “Annotations” table was used to profile its PUL systems (Supplementary Fig. 8). MAG-resolved heatmaps were generated for (i) the electron transport chain, (ii) GHs by counting the number of genes belonging to each “subheader” (representing a different carbohydrate substrate), including oligo- and backbone-cleavage enzymes, (iii) sugar transporters by counting the number of genes belonging to each “module” (representing a different transport system), and (iv) PUL systems by counting the number of genes whose annotations contain the strings “susA”, “susB”, “susC”,“susD”,“susE”, “susF”,“susG”, and “susR”. Temporal dynamics of select taxa Relative abundances of select taxa were plotted longitudinally using ggplot2 (https://ggplot2.tidyverse.org). Statistical testing was performed via the wilcox_test function from the “rstatix” package (https://CRAN.R-project.org/package=rstatix), with the following parameters: “paired=FALSE, ref.group = ‘Lam’, detailed=TRUE, alternative=’two.sided’, p.adjust.method =’fdr’. Significance was shown using the “ggpubr” package’s stat_pvalue_manual function (https://rpkgs.datanovia.com/ggpubr). Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file